openblas/SOURCES/openblas-0.3.15-asmflags.patch

diff --git a/kernel/Makefile b/kernel/Makefile
index 1a6c9413..afd54abc 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -132,25 +132,32 @@ endif
 all : libs
 
 scabs1.$(SUFFIX): $(KERNELDIR)/$(SCABS_KERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE -DF_INTERFACE $< -o $(@F)
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE -DF_INTERFACE $< -o $(@F)
 
 dcabs1.$(SUFFIX): $(KERNELDIR)/$(DCABS_KERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -DF_INTERFACE $< -o $(@F)
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE -DF_INTERFACE $< -o $(@F)
 
 qcabs1.$(SUFFIX): $(KERNELDIR)/$(QCABS_KERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DF_INTERFACE $< -o $(@F)
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -DF_INTERFACE $< -o $(@F)
 
 qconjg.$(SUFFIX): $(KERNELDIR)/qconjg.S
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DF_INTERFACE $< -o $(@F)
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -DF_INTERFACE $< -o $(@F)
 
 lsame.$(SUFFIX): $(KERNELDIR)/$(LSAME_KERNEL)
-	$(CC) -c $(CFLAGS) -DF_INTERFACE $< -o $(@F)
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DF_INTERFACE $< -o $(@F)
 
 setparam$(TSUFFIX).$(SUFFIX): setparam$(TSUFFIX).c kernel$(TSUFFIX).h
 ifeq ($(USE_GEMM3M), 1)
-	$(CC) -c $(CFLAGS) -DUSE_GEMM3M $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DUSE_GEMM3M $< -o $@
 else
-	$(CC) -c $(CFLAGS) $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $< -o $@
 endif
 
 setparam$(TSUFFIX).c : setparam-ref.c
@@ -161,25 +168,32 @@ kernel$(TSUFFIX).h : $(KERNEL_INTERFACE)
 
 
 cpuid.$(SUFFIX): $(KERNELDIR)/cpuid.S
-	$(CC) -c $(CFLAGS) $< -o $(@F)
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $< -o $(@F)
 
 scabs1.$(PSUFFIX): $(KERNELDIR)/$(SCABS_KERNEL)
-	$(CC) -c $(PFLAGS) -DCOMPLEX -UDOUBLE -DF_INTERFACE $< -o $(@F)
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DCOMPLEX -UDOUBLE -DF_INTERFACE $< -o $(@F)
 
 dcabs1.$(PSUFFIX): $(KERNELDIR)/$(DCABS_KERNEL)
-	$(CC) -c $(PFLAGS) -DCOMPLEX -DDOUBLE -DF_INTERFACE $< -o $(@F)
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DCOMPLEX -DDOUBLE -DF_INTERFACE $< -o $(@F)
 
 qcabs1.$(PSUFFIX): $(KERNELDIR)/$(QCABS_KERNEL)
-	$(CC) -c $(PFLAGS) -DCOMPLEX -DXDOUBLE -DF_INTERFACE $< -o $(@F)
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DCOMPLEX -DXDOUBLE -DF_INTERFACE $< -o $(@F)
 
 qconjg.$(PSUFFIX): $(KERNELDIR)/qconjg.S
-	$(CC) -c $(PFLAGS) -DCOMPLEX -DXDOUBLE -DF_INTERFACE $< -o $(@F)
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DCOMPLEX -DXDOUBLE -DF_INTERFACE $< -o $(@F)
 
 lsame.$(PSUFFIX): $(KERNELDIR)/$(LSAME_KERNEL)
-	$(CC) -c $(PFLAGS) -DF_INTERFACE $< -o $(@F)
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DF_INTERFACE $< -o $(@F)
 
 cpuid.$(PSUFFIX): $(KERNELDIR)/cpuid.S
-	$(CC) -c $(PFLAGS) $< -o $(@F)
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $< -o $(@F)
 
 #ifdef DYNAMIC_ARCH
 clean ::
diff --git a/kernel/Makefile.L1 b/kernel/Makefile.L1
index 09337363..a368277e 100644
--- a/kernel/Makefile.L1
+++ b/kernel/Makefile.L1
@@ -553,217 +553,282 @@ endif
 
 
 $(KDIR)samax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)samax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SAMAXKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE  -DUSE_ABS -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE  -DUSE_ABS -UUSE_MIN $< -o $@
 
 $(KDIR)damax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)damax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DAMAXKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE  -DUSE_ABS -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE  -DUSE_ABS -UUSE_MIN $< -o $@
 
 $(KDIR)qamax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)qamax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QAMAXKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE -DUSE_ABS -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE -DUSE_ABS -UUSE_MIN $< -o $@
 
 $(KDIR)camax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)camax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CAMAXKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE  -DUSE_ABS -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE  -DUSE_ABS -UUSE_MIN $< -o $@
 
 $(KDIR)zamax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zamax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZAMAXKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE  -DUSE_ABS -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE  -DUSE_ABS -UUSE_MIN $< -o $@
 
 $(KDIR)xamax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xamax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XAMAXKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DUSE_ABS -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -DUSE_ABS -UUSE_MIN $< -o $@
 
 ### AMIN ###
 
 $(KDIR)samin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)samin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SAMINKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
 
 $(KDIR)damin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)damin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DAMINKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
 
 $(KDIR)qamin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)qamin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QAMINKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE -DUSE_ABS -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE -DUSE_ABS -DUSE_MIN $< -o $@
 
 $(KDIR)camin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)camin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CAMINKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
 
 $(KDIR)zamin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zamin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZAMINKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
 
 $(KDIR)xamin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xamin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XAMINKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DUSE_ABS -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -DUSE_ABS -DUSE_MIN $< -o $@
 
 ### MAX ###
 
 $(KDIR)smax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)smax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SMAXKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -UUSE_ABS  -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE -UUSE_ABS  -UUSE_MIN $< -o $@
 
 $(KDIR)dmax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)dmax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DMAXKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE -UUSE_ABS  -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE -UUSE_ABS  -UUSE_MIN $< -o $@
 
 $(KDIR)qmax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)qmax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QMAXKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE -UUSE_ABS -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE -UUSE_ABS -UUSE_MIN $< -o $@
 
 ### MIN ###
 
 $(KDIR)smin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)smin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SMINKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -UUSE_ABS  -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE -UUSE_ABS  -DUSE_MIN $< -o $@
 
 $(KDIR)dmin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)dmin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DMINKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE -UUSE_ABS  -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE -UUSE_ABS  -DUSE_MIN $< -o $@
 
 $(KDIR)qmin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)qmin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QMINKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE -UUSE_ABS -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE -UUSE_ABS -DUSE_MIN $< -o $@
 
 
 ### IAMAX ###
 
 $(KDIR)isamax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)isamax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ISAMAXKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DUSE_ABS  -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE -DUSE_ABS  -UUSE_MIN $< -o $@
 
 $(KDIR)idamax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)idamax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(IDAMAXKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE -DUSE_ABS  -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE -DUSE_ABS  -UUSE_MIN $< -o $@
 
 $(KDIR)iqamax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)iqamax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(IQAMAXKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE -DUSE_ABS -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE -DUSE_ABS -UUSE_MIN $< -o $@
 
 $(KDIR)icamax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)icamax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ICAMAXKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE -DUSE_ABS  -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE -DUSE_ABS  -UUSE_MIN $< -o $@
 
 $(KDIR)izamax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)izamax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(IZAMAXKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -DUSE_ABS  -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE -DUSE_ABS  -UUSE_MIN $< -o $@
 
 $(KDIR)ixamax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)ixamax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(IXAMAXKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DUSE_ABS -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -DUSE_ABS -UUSE_MIN $< -o $@
 
 ### IAMIN ###
 
 $(KDIR)isamin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)isamin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ISAMINKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
 
 $(KDIR)idamin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)idamin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(IDAMINKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
 
 $(KDIR)iqamin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)iqamin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(IQAMINKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE -DUSE_ABS -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE -DUSE_ABS -DUSE_MIN $< -o $@
 
 $(KDIR)icamin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)icamin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ICAMINKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
 
 $(KDIR)izamin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)izamin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(IZAMINKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE -DUSE_ABS  -DUSE_MIN $< -o $@
 
 $(KDIR)ixamin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)ixamin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(IXAMINKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DUSE_ABS -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -DUSE_ABS -DUSE_MIN $< -o $@
 
 ### IMAX ###
 
 $(KDIR)ismax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)ismax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ISMAXKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -UUSE_ABS  -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE -UUSE_ABS  -UUSE_MIN $< -o $@
 
 $(KDIR)idmax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)idmax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(IDMAXKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE -UUSE_ABS  -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE -UUSE_ABS  -UUSE_MIN $< -o $@
 
 $(KDIR)iqmax_k$(TSUFFIX).$(SUFFIX)  $(KDIR)iqmax_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(IQMAXKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE -UUSE_ABS -UUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE -UUSE_ABS -UUSE_MIN $< -o $@
 
 ### IMIN ###
 
 $(KDIR)ismin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)ismin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ISMINKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -UUSE_ABS  -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE -UUSE_ABS  -DUSE_MIN $< -o $@
 
 $(KDIR)idmin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)idmin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(IDMINKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE -UUSE_ABS  -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE -UUSE_ABS  -DUSE_MIN $< -o $@
 
 $(KDIR)iqmin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)iqmin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(IQMINKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE -UUSE_ABS -DUSE_MIN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE -UUSE_ABS -DUSE_MIN $< -o $@
 
 ### ASUM ###
 $(KDIR)sasum_k$(TSUFFIX).$(SUFFIX)   $(KDIR)sasum_k$(TPSUFFIX).$(PSUFFIX)   : $(KERNELDIR)/$(SASUMKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE $< -o $@
 
 $(KDIR)dasum_k$(TSUFFIX).$(SUFFIX)   $(KDIR)dasum_k$(TPSUFFIX).$(PSUFFIX)   : $(KERNELDIR)/$(DASUMKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE $< -o $@
 
 $(KDIR)qasum_k$(TSUFFIX).$(SUFFIX)   $(KDIR)qasum_k$(TPSUFFIX).$(PSUFFIX)   : $(KERNELDIR)/$(QASUMKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE $< -o $@
 
 $(KDIR)casum_k$(TSUFFIX).$(SUFFIX)  $(KDIR)casum_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CASUMKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE $< -o $@
 
 $(KDIR)zasum_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zasum_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZASUMKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE $< -o $@
 
 $(KDIR)xasum_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xasum_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XASUMKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE $< -o $@
 
 ### SUM ###
 $(KDIR)ssum_k$(TSUFFIX).$(SUFFIX)   $(KDIR)ssum_k$(TPSUFFIX).$(PSUFFIX)   : $(KERNELDIR)/$(SSUMKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE $< -o $@
 
 $(KDIR)dsum_k$(TSUFFIX).$(SUFFIX)   $(KDIR)dsum_k$(TPSUFFIX).$(PSUFFIX)   : $(KERNELDIR)/$(DSUMKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE $< -o $@
 
 $(KDIR)qsum_k$(TSUFFIX).$(SUFFIX)   $(KDIR)qsum_k$(TPSUFFIX).$(PSUFFIX)   : $(KERNELDIR)/$(QSUMKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE $< -o $@
 
 $(KDIR)csum_k$(TSUFFIX).$(SUFFIX)  $(KDIR)csum_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CSUMKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE $< -o $@
 
 $(KDIR)zsum_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zsum_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZSUMKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE $< -o $@
 
 $(KDIR)xsum_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xsum_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XSUMKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE $< -o $@
 
 ### AXPY ###
 $(KDIR)saxpy_k$(TSUFFIX).$(SUFFIX)  $(KDIR)saxpy_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SAXPYKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE $< -o $@
 
 $(KDIR)daxpy_k$(TSUFFIX).$(SUFFIX)  $(KDIR)daxpy_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DAXPYKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE $< -o $@
 
 $(KDIR)qaxpy_k$(TSUFFIX).$(SUFFIX)  $(KDIR)qaxpy_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QAXPYKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE $< -o $@
 
 $(KDIR)caxpy_k$(TSUFFIX).$(SUFFIX)  $(KDIR)caxpy_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CAXPYKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DCOMPLEX -UCONJ -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DCOMPLEX -UCONJ -UDOUBLE $< -o $@
 
 $(KDIR)zaxpy_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zaxpy_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZAXPYKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UCONJ -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UCONJ -DDOUBLE $< -o $@
 
 $(KDIR)xaxpy_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xaxpy_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XAXPYKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UCONJ -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UCONJ -DXDOUBLE $< -o $@
 
 $(KDIR)caxpyc_k$(TSUFFIX).$(SUFFIX)  $(KDIR)caxpyc_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CAXPYKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DCONJ -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DCONJ -UDOUBLE $< -o $@
 
 $(KDIR)zaxpyc_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zaxpyc_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZAXPYKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DCONJ -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DCONJ -DDOUBLE $< -o $@
 
 $(KDIR)xaxpyc_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xaxpyc_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XAXPYKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DCONJ -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DCONJ -DXDOUBLE $< -o $@
 
 $(KDIR)scopy_k$(TSUFFIX).$(SUFFIX) $(KDIR)scopy_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SCOPYKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DC_INTERFACE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE -DC_INTERFACE $< -o $@
 
 $(KDIR)dcopy_k$(TSUFFIX).$(SUFFIX) $(KDIR)dcopy_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DCOPYKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE -DC_INTERFACE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE -DC_INTERFACE $< -o $@
 
 $(KDIR)qcopy_k$(TSUFFIX).$(SUFFIX) $(KDIR)qcopy_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QCOPYKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE -DC_INTERFACE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE -DC_INTERFACE $< -o $@
 
 $(KDIR)ccopy_k$(TSUFFIX).$(SUFFIX) $(KDIR)ccopy_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CCOPYKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE -DC_INTERFACE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE -DC_INTERFACE $< -o $@
 
 $(KDIR)zcopy_k$(TSUFFIX).$(SUFFIX) $(KDIR)zcopy_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZCOPYKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -DC_INTERFACE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE -DC_INTERFACE $< -o $@
 
 $(KDIR)xcopy_k$(TSUFFIX).$(SUFFIX) $(KDIR)xcopy_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XCOPYKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DC_INTERFACE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -DC_INTERFACE $< -o $@
 
 $(KDIR)ddot_k$(TSUFFIX).$(SUFFIX) $(KDIR)ddot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DDOTKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE $< -o $@
 
 $(KDIR)qdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)qdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QDOTKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE $< -o $@
 
 ifeq ($(BUILD_BFLOAT16),1)
 $(KDIR)sbdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)sbdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SBDOTKERNEL)
@@ -779,126 +844,165 @@ $(KDIR)dbf16tod_k$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(BF16TOKERNEL)
 endif
 
 $(KDIR)sdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)sdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SDOTKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE $< -o $@
 
 ifdef DSDOTKERNEL
 
 $(KDIR)dsdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)dsdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DSDOTKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DDSDOT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE -DDSDOT $< -o $@
 
 $(KDIR)sdsdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)sdsdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DSDOTKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DDSDOT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE -DDSDOT $< -o $@
 
 else
 
 $(KDIR)dsdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)dsdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SDOTKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DDSDOT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE -DDSDOT $< -o $@
 
 $(KDIR)sdsdot_k$(TSUFFIX).$(SUFFIX) $(KDIR)sdsdot_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SDOTKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DDSDOT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE -DDSDOT $< -o $@
 
 endif
 
 $(KDIR)zdotu_k$(TSUFFIX).$(SUFFIX) $(KDIR)zdotu_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZDOTKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE -UCONJ $< -o $@
 
 $(KDIR)zdotc_k$(TSUFFIX).$(SUFFIX) $(KDIR)zdotc_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZDOTKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE -DCONJ $< -o $@
 
 $(KDIR)xdotu_k$(TSUFFIX).$(SUFFIX) $(KDIR)xdotu_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XDOTKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -UCONJ $< -o $@
 
 $(KDIR)xdotc_k$(TSUFFIX).$(SUFFIX) $(KDIR)xdotc_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XDOTKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -DCONJ $< -o $@
 
 $(KDIR)cdotu_k$(TSUFFIX).$(SUFFIX) $(KDIR)cdotu_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CDOTKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE -UCONJ $< -o $@
 
 $(KDIR)cdotc_k$(TSUFFIX).$(SUFFIX) $(KDIR)cdotc_k$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CDOTKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE -DCONJ $< -o $@
 
 $(KDIR)snrm2_k$(TSUFFIX).$(SUFFIX)  $(KDIR)snrm2_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SNRM2KERNEL)
-	$(CC) $(CFLAGS) -UCOMPLEX -c -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -UCOMPLEX -c -UDOUBLE $< -o $@
 
 $(KDIR)dnrm2_k$(TSUFFIX).$(SUFFIX)  $(KDIR)dnrm2_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DNRM2KERNEL)
-	$(CC) $(CFLAGS) -UCOMPLEX -c -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -UCOMPLEX -c -DDOUBLE $< -o $@
 
 $(KDIR)qnrm2_k$(TSUFFIX).$(SUFFIX)  $(KDIR)qnrm2_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QNRM2KERNEL)
-	$(CC) $(CFLAGS) -UCOMPLEX -c -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -UCOMPLEX -c -DXDOUBLE $< -o $@
 
 $(KDIR)cnrm2_k$(TSUFFIX).$(SUFFIX)  $(KDIR)cnrm2_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CNRM2KERNEL)
-	$(CC) $(CFLAGS) -DCOMPLEX -c -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -DCOMPLEX -c -UDOUBLE $< -o $@
 
 $(KDIR)znrm2_k$(TSUFFIX).$(SUFFIX)  $(KDIR)znrm2_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZNRM2KERNEL)
-	$(CC) $(CFLAGS) -DCOMPLEX -c -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -DCOMPLEX -c -DDOUBLE $< -o $@
 
 $(KDIR)xnrm2_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xnrm2_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XNRM2KERNEL)
-	$(CC) $(CFLAGS) -DCOMPLEX -c -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -DCOMPLEX -c -DXDOUBLE $< -o $@
 
 $(KDIR)srot_k$(TSUFFIX).$(SUFFIX)  $(KDIR)srot_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SROTKERNEL)
-	$(CC) -c $(CFLAGS) $(FMAFLAG) -UCOMPLEX -UCOMPLEX -UDOUBLE  $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(FMAFLAG) -UCOMPLEX -UCOMPLEX -UDOUBLE  $< -o $@
 
 $(KDIR)drot_k$(TSUFFIX).$(SUFFIX)  $(KDIR)drot_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DROTKERNEL)
-	$(CC) -c $(CFLAGS) $(FMAFLAG) -UCOMPLEX -UCOMPLEX -DDOUBLE  $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(FMAFLAG) -UCOMPLEX -UCOMPLEX -DDOUBLE  $< -o $@
 
 $(KDIR)qrot_k$(TSUFFIX).$(SUFFIX)  $(KDIR)qrot_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QROTKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UCOMPLEX -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UCOMPLEX -DXDOUBLE $< -o $@
 
 $(KDIR)csrot_k$(TSUFFIX).$(SUFFIX)  $(KDIR)csrot_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CROTKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DCOMPLEX -UDOUBLE  $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DCOMPLEX -UDOUBLE  $< -o $@
 
 $(KDIR)zdrot_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zdrot_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZROTKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DCOMPLEX -DDOUBLE  $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DCOMPLEX -DDOUBLE  $< -o $@
 
 $(KDIR)xqrot_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xqrot_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XROTKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DCOMPLEX -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DCOMPLEX -DXDOUBLE $< -o $@
 
 $(KDIR)sscal_k$(TSUFFIX).$(SUFFIX)  $(KDIR)sscal_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SSCALKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE $< -o $@
 
 $(KDIR)dscal_k$(TSUFFIX).$(SUFFIX)  $(KDIR)dscal_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DSCALKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE $< -o $@
 
 $(KDIR)qscal_k$(TSUFFIX).$(SUFFIX)  $(KDIR)qscal_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QSCALKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE $< -o $@
 
 $(KDIR)cscal_k$(TSUFFIX).$(SUFFIX)  $(KDIR)cscal_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CSCALKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE $< -o $@
 
 $(KDIR)zscal_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zscal_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZSCALKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE $< -o $@
 
 $(KDIR)xscal_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xscal_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XSCALKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE $< -o $@
 
 $(KDIR)sswap_k$(TSUFFIX).$(SUFFIX)  $(KDIR)sswap_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SSWAPKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE $< -o $@
 
 $(KDIR)dswap_k$(TSUFFIX).$(SUFFIX)  $(KDIR)dswap_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DSWAPKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE $< -o $@
 
 $(KDIR)qswap_k$(TSUFFIX).$(SUFFIX)  $(KDIR)qswap_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QSWAPKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE $< -o $@
 
 $(KDIR)cswap_k$(TSUFFIX).$(SUFFIX)  $(KDIR)cswap_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CSWAPKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE $< -o $@
 
 $(KDIR)zswap_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zswap_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZSWAPKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE $< -o $@
 
 $(KDIR)xswap_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xswap_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XSWAPKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE $< -o $@
 
 $(KDIR)saxpby_k$(TSUFFIX).$(SUFFIX)  $(KDIR)saxpby_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SAXPBYKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE $< -o $@
 
 $(KDIR)daxpby_k$(TSUFFIX).$(SUFFIX)  $(KDIR)daxpby_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DAXPBYKERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE $< -o $@
 
 $(KDIR)caxpby_k$(TSUFFIX).$(SUFFIX)  $(KDIR)caxpby_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CAXPBYKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UCONJ -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UCONJ -UDOUBLE $< -o $@
 
 $(KDIR)zaxpby_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zaxpby_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZAXPBYKERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UCONJ -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UCONJ -DDOUBLE $< -o $@
 
 
diff --git a/kernel/Makefile.L2 b/kernel/Makefile.L2
index 888a9b95..16e94c6a 100644
--- a/kernel/Makefile.L2
+++ b/kernel/Makefile.L2
@@ -252,252 +252,321 @@ endif
 
 ifneq "$(or $(BUILD_SINGLE), $(BUILD_DOUBLE), $(BUILD_COMPLEX))" ""
 $(KDIR)sgemv_n$(TSUFFIX).$(SUFFIX)  $(KDIR)sgemv_n$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SGEMVNKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -UDOUBLE -UCOMPLEX  -UTRANS $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE -UCOMPLEX  -UTRANS $< -o $@
 
 $(KDIR)sgemv_t$(TSUFFIX).$(SUFFIX)  $(KDIR)sgemv_t$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SGEMVTKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -UDOUBLE -UCOMPLEX  -DTRANS  $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE -UCOMPLEX  -DTRANS  $< -o $@
 endif
 
 ifneq "$(or $(BUILD_DOUBLE),$(BUILD_COMPLEX16))" ""
 $(KDIR)dgemv_n$(TSUFFIX).$(SUFFIX)  $(KDIR)dgemv_n$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DGEMVNKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -DDOUBLE -UCOMPLEX  -UTRANS $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE -UCOMPLEX  -UTRANS $< -o $@
 
 $(KDIR)dgemv_t$(TSUFFIX).$(SUFFIX)  $(KDIR)dgemv_t$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DGEMVTKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -DDOUBLE -UCOMPLEX  -DTRANS $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE -UCOMPLEX  -DTRANS $< -o $@
 endif
 
 $(KDIR)qgemv_n$(TSUFFIX).$(SUFFIX)  $(KDIR)qgemv_n$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QGEMVNKERNEL)
-	$(CC) -c $(CFLAGS) -DXDOUBLE -UCOMPLEX -UTRANS $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE -UCOMPLEX -UTRANS $< -o $@
 
 $(KDIR)qgemv_t$(TSUFFIX).$(SUFFIX)  $(KDIR)qgemv_t$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QGEMVTKERNEL)
-	$(CC) -c $(CFLAGS) -DXDOUBLE -UCOMPLEX -DTRANS $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE -UCOMPLEX -DTRANS $< -o $@
 
 
 ifneq "$(or $(BUILD_COMPLEX),$(BUILD_COMPLEX16))" ""
 $(KDIR)cgemv_n$(TSUFFIX).$(SUFFIX) $(KDIR)cgemv_n$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMVNKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -UDOUBLE -DCOMPLEX -UTRANS -UCONJ -UXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE -DCOMPLEX -UTRANS -UCONJ -UXCONJ $< -o $@
 
 $(KDIR)cgemv_t$(TSUFFIX).$(SUFFIX) $(KDIR)cgemv_t$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMVTKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -UDOUBLE -DCOMPLEX -DTRANS -UCONJ -UXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE -DCOMPLEX -DTRANS -UCONJ -UXCONJ $< -o $@
 
 $(KDIR)cgemv_r$(TSUFFIX).$(SUFFIX) $(KDIR)cgemv_r$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMVNKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -UDOUBLE -DCOMPLEX -UTRANS -DCONJ -UXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE -DCOMPLEX -UTRANS -DCONJ -UXCONJ $< -o $@
 
 $(KDIR)cgemv_c$(TSUFFIX).$(SUFFIX) $(KDIR)cgemv_c$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMVTKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -UDOUBLE -DCOMPLEX -DTRANS -DCONJ -UXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE -DCOMPLEX -DTRANS -DCONJ -UXCONJ $< -o $@
 
 $(KDIR)cgemv_o$(TSUFFIX).$(SUFFIX) $(KDIR)cgemv_o$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMVNKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -UDOUBLE -DCOMPLEX -UTRANS -UCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE -DCOMPLEX -UTRANS -UCONJ -DXCONJ $< -o $@
 
 $(KDIR)cgemv_u$(TSUFFIX).$(SUFFIX) $(KDIR)cgemv_u$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMVTKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -UDOUBLE -DCOMPLEX -DTRANS -UCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE -DCOMPLEX -DTRANS -UCONJ -DXCONJ $< -o $@
 
 $(KDIR)cgemv_s$(TSUFFIX).$(SUFFIX) $(KDIR)cgemv_s$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMVNKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -UDOUBLE -DCOMPLEX -UTRANS -DCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE -DCOMPLEX -UTRANS -DCONJ -DXCONJ $< -o $@
 
 $(KDIR)cgemv_d$(TSUFFIX).$(SUFFIX) $(KDIR)cgemv_d$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMVTKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -UDOUBLE -DCOMPLEX -DTRANS -DCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE -DCOMPLEX -DTRANS -DCONJ -DXCONJ $< -o $@
 endif
 
 
 ifeq ($(BUILD_COMPLEX16),1)
 
 $(KDIR)zgemv_n$(TSUFFIX).$(SUFFIX) $(KDIR)zgemv_n$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMVNKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -DDOUBLE -DCOMPLEX -UTRANS -UCONJ -UXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE -DCOMPLEX -UTRANS -UCONJ -UXCONJ $< -o $@
 
 $(KDIR)zgemv_t$(TSUFFIX).$(SUFFIX) $(KDIR)zgemv_t$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMVTKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -DDOUBLE -DCOMPLEX -DTRANS -UCONJ -UXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE -DCOMPLEX -DTRANS -UCONJ -UXCONJ $< -o $@
 
 $(KDIR)zgemv_r$(TSUFFIX).$(SUFFIX) $(KDIR)zgemv_r$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMVNKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -DDOUBLE -DCOMPLEX -UTRANS -DCONJ -UXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE -DCOMPLEX -UTRANS -DCONJ -UXCONJ $< -o $@
 
 $(KDIR)zgemv_c$(TSUFFIX).$(SUFFIX) $(KDIR)zgemv_c$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMVTKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -DDOUBLE -DCOMPLEX -DTRANS -DCONJ -UXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE -DCOMPLEX -DTRANS -DCONJ -UXCONJ $< -o $@
 
 $(KDIR)zgemv_o$(TSUFFIX).$(SUFFIX) $(KDIR)zgemv_o$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMVNKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -DDOUBLE -DCOMPLEX -UTRANS -UCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE -DCOMPLEX -UTRANS -UCONJ -DXCONJ $< -o $@
 
 $(KDIR)zgemv_u$(TSUFFIX).$(SUFFIX) $(KDIR)zgemv_u$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMVTKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -DDOUBLE -DCOMPLEX -DTRANS -UCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE -DCOMPLEX -DTRANS -UCONJ -DXCONJ $< -o $@
 
 $(KDIR)zgemv_s$(TSUFFIX).$(SUFFIX) $(KDIR)zgemv_s$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMVNKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -DDOUBLE -DCOMPLEX -UTRANS -DCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE -DCOMPLEX -UTRANS -DCONJ -DXCONJ $< -o $@
 
 $(KDIR)zgemv_d$(TSUFFIX).$(SUFFIX) $(KDIR)zgemv_d$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMVTKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -DDOUBLE -DCOMPLEX -DTRANS -DCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE -DCOMPLEX -DTRANS -DCONJ -DXCONJ $< -o $@
 endif
 
 $(KDIR)xgemv_n$(TSUFFIX).$(SUFFIX) $(KDIR)xgemv_n$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMVNKERNEL)
-	$(CC) -c $(CFLAGS) -DXDOUBLE -DCOMPLEX -UTRANS -UCONJ -UXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE -DCOMPLEX -UTRANS -UCONJ -UXCONJ $< -o $@
 
 $(KDIR)xgemv_t$(TSUFFIX).$(SUFFIX) $(KDIR)xgemv_t$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMVTKERNEL)
-	$(CC) -c $(CFLAGS) -DXDOUBLE -DCOMPLEX -DTRANS -UCONJ -UXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE -DCOMPLEX -DTRANS -UCONJ -UXCONJ $< -o $@
 
 $(KDIR)xgemv_r$(TSUFFIX).$(SUFFIX) $(KDIR)xgemv_r$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMVNKERNEL)
-	$(CC) -c $(CFLAGS) -DXDOUBLE -DCOMPLEX -UTRANS -DCONJ -UXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE -DCOMPLEX -UTRANS -DCONJ -UXCONJ $< -o $@
 
 $(KDIR)xgemv_c$(TSUFFIX).$(SUFFIX) $(KDIR)xgemv_c$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMVTKERNEL)
-	$(CC) -c $(CFLAGS) -DXDOUBLE -DCOMPLEX -DTRANS -DCONJ -UXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE -DCOMPLEX -DTRANS -DCONJ -UXCONJ $< -o $@
 
 $(KDIR)xgemv_o$(TSUFFIX).$(SUFFIX) $(KDIR)xgemv_o$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMVNKERNEL)
-	$(CC) -c $(CFLAGS) -DXDOUBLE -DCOMPLEX -UTRANS -UCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE -DCOMPLEX -UTRANS -UCONJ -DXCONJ $< -o $@
 
 $(KDIR)xgemv_u$(TSUFFIX).$(SUFFIX) $(KDIR)xgemv_u$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMVTKERNEL)
-	$(CC) -c $(CFLAGS) -DXDOUBLE -DCOMPLEX -DTRANS -UCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE -DCOMPLEX -DTRANS -UCONJ -DXCONJ $< -o $@
 
 $(KDIR)xgemv_s$(TSUFFIX).$(SUFFIX) $(KDIR)xgemv_s$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMVNKERNEL)
-	$(CC) -c $(CFLAGS) -DXDOUBLE -DCOMPLEX -UTRANS -DCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE -DCOMPLEX -UTRANS -DCONJ -DXCONJ $< -o $@
 
 $(KDIR)xgemv_d$(TSUFFIX).$(SUFFIX) $(KDIR)xgemv_d$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMVTKERNEL)
-	$(CC) -c $(CFLAGS) -DXDOUBLE -DCOMPLEX -DTRANS -DCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE -DCOMPLEX -DTRANS -DCONJ -DXCONJ $< -o $@
 
 
 ifeq ($(BUILD_SINGLE),1)
 
 $(KDIR)ssymv_U$(TSUFFIX).$(SUFFIX)  $(KDIR)ssymv_U$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SSYMV_U_KERNEL) $(SSYMV_U_PARAM)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE -ULOWER $< -o $@
 
 $(KDIR)ssymv_L$(TSUFFIX).$(SUFFIX)  $(KDIR)ssymv_L$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SSYMV_L_KERNEL)  $(SSYMV_L_PARAM)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -UDOUBLE -DLOWER $< -o $@
 endif
 
 
 ifeq ($(BUILD_DOUBLE),1)
 
 $(KDIR)dsymv_U$(TSUFFIX).$(SUFFIX)  $(KDIR)dsymv_U$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DSYMV_U_KERNEL)  $(DSYMV_U_PARAM)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE -ULOWER $< -o $@
 
 $(KDIR)dsymv_L$(TSUFFIX).$(SUFFIX)  $(KDIR)dsymv_L$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DSYMV_L_KERNEL)  $(DSYMV_L_PARAM)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DDOUBLE -DLOWER $< -o $@
 endif
 
 $(KDIR)qsymv_U$(TSUFFIX).$(SUFFIX)  $(KDIR)qsymv_U$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QSYMV_U_KERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE -ULOWER $< -o $@
 
 $(KDIR)qsymv_L$(TSUFFIX).$(SUFFIX)  $(KDIR)qsymv_L$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QSYMV_L_KERNEL)
-	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UCOMPLEX -DXDOUBLE -DLOWER $< -o $@
 
 ifeq ($(BUILD_COMPLEX),1)
 
 $(KDIR)csymv_U$(TSUFFIX).$(SUFFIX)  $(KDIR)csymv_U$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CSYMV_U_KERNEL)   $(CSYMV_U_PARAM)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE -ULOWER $< -o $@
 
 $(KDIR)csymv_L$(TSUFFIX).$(SUFFIX)  $(KDIR)csymv_L$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CSYMV_L_KERNEL)   $(CSYMV_L_PARAM)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE -DLOWER $< -o $@
 endif
 
 ifeq ($(BUILD_COMPLEX16),1)
 
 $(KDIR)zsymv_U$(TSUFFIX).$(SUFFIX)  $(KDIR)zsymv_U$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZSYMV_U_KERNEL)   $(ZSYMV_U_PARAM)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE -ULOWER $< -o $@
 
 $(KDIR)zsymv_L$(TSUFFIX).$(SUFFIX)  $(KDIR)zsymv_L$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZSYMV_L_KERNEL)   $(ZSYMV_L_PARAM)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE -DLOWER $< -o $@
 endif
 
 $(KDIR)xsymv_U$(TSUFFIX).$(SUFFIX)  $(KDIR)xsymv_U$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XSYMV_U_KERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -ULOWER $< -o $@
 
 $(KDIR)xsymv_L$(TSUFFIX).$(SUFFIX)  $(KDIR)xsymv_L$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XSYMV_L_KERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -DLOWER $< -o $@
 
 ifeq ($(BUILD_SINGLE),1)
 
 $(KDIR)sger_k$(TSUFFIX).$(SUFFIX)  $(KDIR)sger_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SGERKERNEL) $(SGERPARAM)
-	$(CC) -c $(CFLAGS) -UDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE $< -o $@
 endif
 
 ifeq ($(BUILD_DOUBLE),1)
 
 $(KDIR)dger_k$(TSUFFIX).$(SUFFIX)  $(KDIR)dger_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DGERKERNEL) $(DGERPARAM)
-	$(CC) -c $(CFLAGS) -DDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE $< -o $@
 endif
 
 $(KDIR)qger_k$(TSUFFIX).$(SUFFIX)  $(KDIR)qger_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QGERKERNEL) $(QGERPARAM)
-	$(CC) -c $(CFLAGS) -DXDOUBLE $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE $< -o $@
 
 ifeq ($(BUILD_COMPLEX),1)
 
 $(KDIR)cgeru_k$(TSUFFIX).$(SUFFIX)  $(KDIR)cgeru_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CGERUKERNEL) $(CGERPARAM)
-	$(CC) -c $(CFLAGS) -UDOUBLE -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE -UCONJ $< -o $@
 
 $(KDIR)cgerc_k$(TSUFFIX).$(SUFFIX)  $(KDIR)cgerc_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CGERCKERNEL) $(CGERPARAM)
-	$(CC) -c $(CFLAGS) -UDOUBLE -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE -DCONJ $< -o $@
 
 $(KDIR)cgerv_k$(TSUFFIX).$(SUFFIX)  $(KDIR)cgerv_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CGERUKERNEL) $(CGERPARAM)
-	$(CC) -c $(CFLAGS) -UDOUBLE -UCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE -UCONJ -DXCONJ $< -o $@
 
 $(KDIR)cgerd_k$(TSUFFIX).$(SUFFIX)  $(KDIR)cgerd_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CGERCKERNEL) $(CGERPARAM)
-	$(CC) -c $(CFLAGS) -UDOUBLE -DCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -UDOUBLE -DCONJ -DXCONJ $< -o $@
 endif
 
 ifeq ($(BUILD_COMPLEX16),1)
 
 $(KDIR)zgeru_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zgeru_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZGERUKERNEL) $(ZGERPARAM)
-	$(CC) -c $(CFLAGS) -DDOUBLE -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE -UCONJ $< -o $@
 
 $(KDIR)zgerc_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zgerc_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZGERCKERNEL) $(ZGERPARAM)
-	$(CC) -c $(CFLAGS) -DDOUBLE -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE -DCONJ $< -o $@
 
 $(KDIR)zgerv_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zgerv_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZGERUKERNEL) $(ZGERPARAM)
-	$(CC) -c $(CFLAGS) -DDOUBLE -UCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE -UCONJ -DXCONJ $< -o $@
 
 $(KDIR)zgerd_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zgerd_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZGERCKERNEL) $(ZGERPARAM)
-	$(CC) -c $(CFLAGS) -DDOUBLE -DCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DDOUBLE -DCONJ -DXCONJ $< -o $@
 endif
 
 $(KDIR)xgeru_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xgeru_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XGERUKERNEL) $(XGERPARAM)
-	$(CC) -c $(CFLAGS) -DXDOUBLE -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE -UCONJ $< -o $@
 
 $(KDIR)xgerc_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xgerc_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XGERCKERNEL) $(XGERPARAM)
-	$(CC) -c $(CFLAGS) -DXDOUBLE -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE -DCONJ $< -o $@
 
 $(KDIR)xgerv_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xgerv_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XGERUKERNEL) $(XGERPARAM)
-	$(CC) -c $(CFLAGS) -DXDOUBLE -UCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE -UCONJ -DXCONJ $< -o $@
 
 $(KDIR)xgerd_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xgerd_k$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XGERCKERNEL) $(XGERPARAM)
-	$(CC) -c $(CFLAGS) -DXDOUBLE -DCONJ -DXCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DXDOUBLE -DCONJ -DXCONJ $< -o $@
 
 ifeq ($(BUILD_COMPLEX),1)
 
 $(KDIR)chemv_U$(TSUFFIX).$(SUFFIX)  $(KDIR)chemv_U$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CHEMV_U_KERNEL) $(CHEMV_U_PARAM)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE -ULOWER -DHEMV $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE -ULOWER -DHEMV $< -o $@
 
 $(KDIR)chemv_L$(TSUFFIX).$(SUFFIX)  $(KDIR)chemv_L$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CHEMV_L_KERNEL) $(CHEMV_L_PARAM)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE -DLOWER -DHEMV $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE -DLOWER -DHEMV $< -o $@
 
 $(KDIR)chemv_V$(TSUFFIX).$(SUFFIX)  $(KDIR)chemv_V$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CHEMV_V_KERNEL) $(CHEMV_U_PARAM) ../symcopy.h
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE -ULOWER -DHEMV -DHEMVREV $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE -ULOWER -DHEMV -DHEMVREV $< -o $@
 
 $(KDIR)chemv_M$(TSUFFIX).$(SUFFIX)  $(KDIR)chemv_M$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CHEMV_M_KERNEL) $(CHEMV_L_PARAM) ../symcopy.h
-	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE -DLOWER -DHEMV -DHEMVREV $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -UDOUBLE -DLOWER -DHEMV -DHEMVREV $< -o $@
 endif
 
 ifeq ($(BUILD_COMPLEX16),1)
 
 $(KDIR)zhemv_U$(TSUFFIX).$(SUFFIX)  $(KDIR)zhemv_U$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZHEMV_U_KERNEL) $(ZHEMV_U_PARAM)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -ULOWER -DHEMV $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE -ULOWER -DHEMV $< -o $@
 
 $(KDIR)zhemv_L$(TSUFFIX).$(SUFFIX)  $(KDIR)zhemv_L$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZHEMV_L_KERNEL) $(ZHEMV_L_PARAM)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -DLOWER -DHEMV $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE -DLOWER -DHEMV $< -o $@
 
 $(KDIR)zhemv_V$(TSUFFIX).$(SUFFIX)  $(KDIR)zhemv_V$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZHEMV_V_KERNEL) $(ZHEMV_U_PARAM) ../symcopy.h
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -ULOWER -DHEMV -DHEMVREV $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE -ULOWER -DHEMV -DHEMVREV $< -o $@
 
 $(KDIR)zhemv_M$(TSUFFIX).$(SUFFIX)  $(KDIR)zhemv_M$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZHEMV_M_KERNEL) $(ZHEMV_L_PARAM) ../symcopy.h
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE -DLOWER -DHEMV -DHEMVREV $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DDOUBLE -DLOWER -DHEMV -DHEMVREV $< -o $@
 endif
 $(KDIR)xhemv_U$(TSUFFIX).$(SUFFIX)  $(KDIR)xhemv_U$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XHEMV_U_KERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -ULOWER -DHEMV $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -ULOWER -DHEMV $< -o $@
 
 $(KDIR)xhemv_L$(TSUFFIX).$(SUFFIX)  $(KDIR)xhemv_L$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XHEMV_L_KERNEL)
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DLOWER -DHEMV $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -DLOWER -DHEMV $< -o $@
 
 $(KDIR)xhemv_V$(TSUFFIX).$(SUFFIX)  $(KDIR)xhemv_V$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XHEMV_V_KERNEL)  ../symcopy.h
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -ULOWER -DHEMV -DHEMVREV $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -ULOWER -DHEMV -DHEMVREV $< -o $@
 
 $(KDIR)xhemv_M$(TSUFFIX).$(SUFFIX)  $(KDIR)xhemv_M$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XHEMV_M_KERNEL)  ../symcopy.h
-	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DLOWER -DHEMV -DHEMVREV $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DCOMPLEX -DXDOUBLE -DLOWER -DHEMV -DHEMVREV $< -o $@
 
 ifeq ($(BUILD_BFLOAT16),1)
 $(KDIR)sbgemv_n$(TSUFFIX).$(SUFFIX) $(KDIR)sbgemv_n$(TPSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SBGEMVNKERNEL)
diff --git a/kernel/Makefile.L3 b/kernel/Makefile.L3
index d8d73996..60ca4c86 100644
--- a/kernel/Makefile.L3
+++ b/kernel/Makefile.L3
@@ -526,119 +526,150 @@ XGEMMOTCOPYOBJ_P = $(XGEMMOTCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
 
 ifeq ($(BUILD_BFLOAT16),1)
 $(KDIR)sbgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SBGEMM_BETA)
-	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
 endif
 
 $(KDIR)sgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_BETA)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)dgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMM_BETA)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)qgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMM_BETA)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)cgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM_BETA)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX $< -o $@
 
 $(KDIR)zgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_BETA)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX $< -o $@
 
 $(KDIR)xgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMM_BETA)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX $< -o $@
 
 
 ifeq ($(BUILD_BFLOAT16), 1)
 
 $(KDIR)$(SBGEMMONCOPYOBJ) : $(KERNELDIR)/$(SBGEMMONCOPY)
-	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(SBGEMMOTCOPYOBJ) : $(KERNELDIR)/$(SBGEMMOTCOPY)
 
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o - > sbgemmotcopy.s
-	m4 sbgemmotcopy.s > sbgemmotcopy_nomacros.s
-	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX sbgemmotcopy_nomacros.s -o $@
-	rm sbgemmotcopy.s sbgemmotcopy_nomacros.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o shgemmotcopy.s
+	m4 shgemmotcopy.s > shgemmotcopy_nomacros.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DBFLOAT16 -UDOUBLE -UCOMPLEX shgemmotcopy_nomacros.s -o $@
+	rm shgemmotcopy.s shgemmotcopy_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
 endif
 
 ifneq ($(SBGEMM_UNROLL_M), $(SBGEMM_UNROLL_N))
 
 $(KDIR)$(SBGEMMINCOPYOBJ) : $(KERNELDIR)/$(SBGEMMINCOPY)
-	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(SBGEMMITCOPYOBJ) : $(KERNELDIR)/$(SBGEMMITCOPY)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o - > sbgemmitcopy.s
-	m4 sbgemmitcopy.s > sbgemmitcopy_nomacros.s
-	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX sbgemmitcopy_nomacros.s -o $@
-	rm sbgemmitcopy.s sbgemmitcopy_nomacros.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o shgemmitcopy.s
+	m4 shgemmitcopy.s > shgemmitcopy_nomacros.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DBFLOAT16 -UDOUBLE -UCOMPLEX shgemmitcopy_nomacros.s -o $@
+	rm shgemmitcopy.s shgemmitcopy_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
 endif
 
 endif
 endif
 
 $(KDIR)$(SGEMMONCOPYOBJ) : $(KERNELDIR)/$(SGEMMONCOPY)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(SGEMMOTCOPYOBJ) : $(KERNELDIR)/$(SGEMMOTCOPY)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -UDOUBLE -UCOMPLEX $< -o - > sgemmotcopy.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -UDOUBLE -UCOMPLEX $< -o - > sgemmotcopy.s
 	m4 sgemmotcopy.s > sgemmotcopy_nomacros.s
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX sgemmotcopy_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX sgemmotcopy_nomacros.s -o $@
 	rm sgemmotcopy.s sgemmotcopy_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 endif	
 
 
 ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N))
 
 $(KDIR)$(SGEMMINCOPYOBJ) : $(KERNELDIR)/$(SGEMMINCOPY)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(SGEMMITCOPYOBJ) : $(KERNELDIR)/$(SGEMMITCOPY)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -UDOUBLE -UCOMPLEX $< -o - > sgemmitcopy.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -UDOUBLE -UCOMPLEX $< -o - > sgemmitcopy.s
 	m4 sgemmitcopy.s > sgemmitcopy_nomacros.s
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX sgemmitcopy_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX sgemmitcopy_nomacros.s -o $@
 	rm sgemmitcopy.s sgemmitcopy_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@ 
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@ 
 endif
 	
 endif
 
 $(KDIR)$(DGEMMONCOPYOBJ) : $(KERNELDIR)/$(DGEMMONCOPY)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DDOUBLE -UCOMPLEX $< -o - > dgemm_ncopy.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DDOUBLE -UCOMPLEX $< -o - > dgemm_ncopy.s
 	m4 dgemm_ncopy.s > dgemm_ncopy_nomacros.s
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX dgemm_ncopy_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX dgemm_ncopy_nomacros.s -o $@
 	rm dgemm_ncopy.s dgemm_ncopy_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 endif
 
 $(KDIR)$(DGEMMOTCOPYOBJ) : $(KERNELDIR)/$(DGEMMOTCOPY)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 ifneq ($(DGEMM_UNROLL_M), $(DGEMM_UNROLL_N))
 
 $(KDIR)$(DGEMMINCOPYOBJ) : $(KERNELDIR)/$(DGEMMINCOPY)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(DGEMMITCOPYOBJ) : $(KERNELDIR)/$(DGEMMITCOPY)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DDOUBLE -UCOMPLEX $< -o - > dgemm_itcopy.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DDOUBLE -UCOMPLEX $< -o - > dgemm_itcopy.s
 	m4 dgemm_itcopy.s > dgemm_itcopy_nomacros.s
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX dgemm_itcopy_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX dgemm_itcopy_nomacros.s -o $@
 	rm dgemm_itcopy.s dgemm_itcopy_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 endif
 
 endif
@@ -646,65 +677,81 @@ endif
 ifdef EXPRECISION
 
 $(KDIR)$(QGEMMONCOPYOBJ) : $(KERNELDIR)/$(QGEMMONCOPY)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(QGEMMOTCOPYOBJ) : $(KERNELDIR)/$(QGEMMOTCOPY)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 ifneq ($(QGEMM_UNROLL_M), $(QGEMM_UNROLL_N))
 
 $(KDIR)$(QGEMMINCOPYOBJ) : $(KERNELDIR)/$(QGEMMINCOPY)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(QGEMMITCOPYOBJ) : $(KERNELDIR)/$(QGEMMITCOPY)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 endif
 
 endif
 
 $(KDIR)$(CGEMMONCOPYOBJ) : $(KERNELDIR)/$(CGEMMONCOPY)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(CGEMMOTCOPYOBJ) : $(KERNELDIR)/$(CGEMMOTCOPY)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 ifneq ($(CGEMM_UNROLL_M), $(CGEMM_UNROLL_N))
 
 $(KDIR)$(CGEMMINCOPYOBJ) : $(KERNELDIR)/$(CGEMMINCOPY)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(CGEMMITCOPYOBJ) : $(KERNELDIR)/$(CGEMMITCOPY)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -UDOUBLE -UCOMPLEX -S $< -o - > cgemm_itcopy.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -UDOUBLE -UCOMPLEX -E $< -o - > cgemm_itcopy.s
 	m4 cgemm_itcopy.s > cgemm_itcopy_nomacros.s
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX cgemm_itcopy_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX cgemm_itcopy_nomacros.s -o $@
 	rm cgemm_itcopy.s cgemm_itcopy_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 endif
 
 endif
 
 $(KDIR)$(ZGEMMONCOPYOBJ) : $(KERNELDIR)/$(ZGEMMONCOPY)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(ZGEMMOTCOPYOBJ) : $(KERNELDIR)/$(ZGEMMOTCOPY)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 ifneq ($(ZGEMM_UNROLL_M), $(ZGEMM_UNROLL_N))
 
 $(KDIR)$(ZGEMMINCOPYOBJ) : $(KERNELDIR)/$(ZGEMMINCOPY)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(ZGEMMITCOPYOBJ) : $(KERNELDIR)/$(ZGEMMITCOPY)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DDOUBLE -UCOMPLEX $< -o - > zgemm_itcopy.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DDOUBLE -UCOMPLEX $< -o - > zgemm_itcopy.s
 	m4 zgemm_itcopy.s > zgemm_itcopy_nomacros.s
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX zgemm_itcopy_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX zgemm_itcopy_nomacros.s -o $@
 	rm zgemm_itcopy.s zgemm_itcopy_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 endif
 
 endif
@@ -712,18 +759,22 @@ endif
 ifdef EXPRECISION
 
 $(KDIR)$(XGEMMONCOPYOBJ) : $(KERNELDIR)/$(XGEMMONCOPY)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(XGEMMOTCOPYOBJ) : $(KERNELDIR)/$(XGEMMOTCOPY)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 ifneq ($(XGEMM_UNROLL_M), $(XGEMM_UNROLL_N))
 
 $(KDIR)$(XGEMMINCOPYOBJ) : $(KERNELDIR)/$(XGEMMINCOPY)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)$(XGEMMITCOPYOBJ) : $(KERNELDIR)/$(XGEMMITCOPY)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 endif
 
@@ -731,12 +782,15 @@ endif
 
 $(KDIR)sgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL) $(SGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -UDOUBLE -UCOMPLEX  $< -o - > sgemm_kernel$(TSUFFIX).s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -UDOUBLE -UCOMPLEX  $< -o - > sgemm_kernel$(TSUFFIX).s
 	m4 sgemm_kernel$(TSUFFIX).s > sgemm_kernel$(TSUFFIX)_nomacros.s
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX sgemm_kernel$(TSUFFIX)_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX sgemm_kernel$(TSUFFIX)_nomacros.s -o $@
 	rm sgemm_kernel$(TSUFFIX).s sgemm_kernel$(TSUFFIX)_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 endif
 
 ifdef USE_DIRECT_SGEMM
@@ -750,1768 +804,2329 @@ ifeq ($(BUILD_BFLOAT16), 1)
 
 $(KDIR)sbgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SBGEMMKERNEL) $(SBGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DBFLOAT16 -UDOUBLE -UCOMPLEX  $< -o - > sbgemm_kernel$(TSUFFIX).s
-	m4 sbgemm_kernel$(TSUFFIX).s > sbgemm_kernel$(TSUFFIX)_nomacros.s
-	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX sbgemm_kernel$(TSUFFIX)_nomacros.s -o $@
-	rm sbgemm_kernel$(TSUFFIX).s sbgemm_kernel$(TSUFFIX)_nomacros.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DHALF -UDOUBLE -UCOMPLEX  $< -o - > shgemm_kernel$(TSUFFIX).s
+	m4 shgemm_kernel$(TSUFFIX).s > shgemm_kernel$(TSUFFIX)_nomacros.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DHALF -UDOUBLE -UCOMPLEX shgemm_kernel$(TSUFFIX)_nomacros.s -o $@
+	rm shgemm_kernel$(TSUFFIX).s shgemm_kernel$(TSUFFIX)_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
 endif
 endif
 
 $(KDIR)dgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL) $(DGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DDOUBLE -UCOMPLEX $< -o - > dgemm_kernel$(TSUFFIX).s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DDOUBLE -UCOMPLEX $< -o - > dgemm_kernel$(TSUFFIX).s
 	m4 dgemm_kernel$(TSUFFIX).s > dgemm_kernel$(TSUFFIX)_nomacros.s
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX dgemm_kernel$(TSUFFIX)_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX dgemm_kernel$(TSUFFIX)_nomacros.s -o $@
 	rm dgemm_kernel$(TSUFFIX).s dgemm_kernel$(TSUFFIX)_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 endif
 
 $(KDIR)qgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL) $(QGEMMDEPEND)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)cgemm_kernel_n$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -UDOUBLE -DCOMPLEX -DNN $< -o - > cgemm_kernel_n.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -UDOUBLE -DCOMPLEX -DNN $< -o - > cgemm_kernel_n.s
 	m4 cgemm_kernel_n.s > cgemm_kernel_n_nomacros.s
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNN cgemm_kernel_n_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DNN cgemm_kernel_n_nomacros.s -o $@
 	rm cgemm_kernel_n.s cgemm_kernel_n_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DNN $< -o $@
 endif
 
 $(KDIR)cgemm_kernel_l$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -UDOUBLE -DCOMPLEX -DCN $< -o - > cgemm_kernel_l.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -UDOUBLE -DCOMPLEX -DCN $< -o - > cgemm_kernel_l.s
 	m4 cgemm_kernel_l.s > cgemm_kernel_l_nomacros.s
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCN cgemm_kernel_l_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DCN cgemm_kernel_l_nomacros.s -o $@
 	rm cgemm_kernel_l.s cgemm_kernel_l_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DCN $< -o $@
 endif
 
 $(KDIR)cgemm_kernel_r$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -UDOUBLE -DCOMPLEX -DNC  $< -o - > cgemm_kernel_r.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -UDOUBLE -DCOMPLEX -DNC  $< -o - > cgemm_kernel_r.s
 	m4 cgemm_kernel_r.s > cgemm_kernel_r_nomacros.s
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNC cgemm_kernel_r_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DNC cgemm_kernel_r_nomacros.s -o $@
 	rm cgemm_kernel_r.s cgemm_kernel_r_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DNC $< -o $@
 endif
 
 $(KDIR)cgemm_kernel_b$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -UDOUBLE -DCOMPLEX -DCC $< -o - > cgemm_kernel_b.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -UDOUBLE -DCOMPLEX -DCC $< -o - > cgemm_kernel_b.s
 	m4 cgemm_kernel_b.s > cgemm_kernel_b_nomacros.s
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCC cgemm_kernel_b_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DCC cgemm_kernel_b_nomacros.s -o $@
 	rm cgemm_kernel_b.s cgemm_kernel_b_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DCC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DCC $< -o $@
 endif
 
 $(KDIR)zgemm_kernel_n$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DDOUBLE -DCOMPLEX -DNN $< -o - > zgemm_kernel_n.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DDOUBLE -DCOMPLEX -DNN $< -o - > zgemm_kernel_n.s
 	m4 zgemm_kernel_n.s > zgemm_kernel_n_nomacros.s
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNN zgemm_kernel_n_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DNN zgemm_kernel_n_nomacros.s -o $@
 	rm zgemm_kernel_n.s zgemm_kernel_n_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DNN $< -o $@
 endif
 
 $(KDIR)zgemm_kernel_l$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DDOUBLE -DCOMPLEX -DCN $< -o - > zgemm_kernel_l.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DDOUBLE -DCOMPLEX -DCN $< -o - > zgemm_kernel_l.s
 	m4 zgemm_kernel_l.s > zgemm_kernel_l_nomacros.s
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCN zgemm_kernel_l_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DCN zgemm_kernel_l_nomacros.s -o $@
 	rm zgemm_kernel_l.s zgemm_kernel_l_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DCN $< -o $@
 endif
 
 $(KDIR)zgemm_kernel_r$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DDOUBLE -DCOMPLEX -DNC $< -o - > zgemm_kernel_r.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DDOUBLE -DCOMPLEX -DNC $< -o - > zgemm_kernel_r.s
 	m4 zgemm_kernel_r.s > zgemm_kernel_r_nomacros.s
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNC zgemm_kernel_r_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DNC zgemm_kernel_r_nomacros.s -o $@
 	rm zgemm_kernel_r.s zgemm_kernel_r_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DNC $< -o $@
 endif
 
 $(KDIR)zgemm_kernel_b$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DDOUBLE -DCOMPLEX -DCC $< -o - > zgemm_kernel_b.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DDOUBLE -DCOMPLEX -DCC $< -o - > zgemm_kernel_b.s
 	m4 zgemm_kernel_b.s > zgemm_kernel_b_nomacros.s
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCC zgemm_kernel_b_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DCC zgemm_kernel_b_nomacros.s -o $@
 	rm zgemm_kernel_b.s zgemm_kernel_b_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DCC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DCC $< -o $@
 endif
 
 $(KDIR)xgemm_kernel_n$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL) $(XGEMMDEPEND)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DNN $< -o $@
 
 $(KDIR)xgemm_kernel_l$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL) $(XGEMMDEPEND)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DCN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DCN $< -o $@
 
 $(KDIR)xgemm_kernel_r$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL) $(XGEMMDEPEND)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DNC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DNC $< -o $@
 
 $(KDIR)xgemm_kernel_b$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL) $(XGEMMDEPEND)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DCC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DCC $< -o $@
 
 
 ifdef USE_TRMM
 $(KDIR)strmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o - > strmmkernel_ln.s	
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o - > strmmkernel_ln.s	
 	m4 strmmkernel_ln.s > strmmkernel_ln_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA strmmkernel_ln_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA strmmkernel_ln_nomacros.s -o $@
 	rm strmmkernel_ln.s strmmkernel_ln_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
 endif
 
 $(KDIR)strmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o - > strmmkernel_lt.s	
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o - > strmmkernel_lt.s	
 	m4 strmmkernel_lt.s > strmmkernel_lt_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA strmmkernel_lt_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA strmmkernel_lt_nomacros.s -o $@
 	rm strmmkernel_lt.s strmmkernel_lt_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
 endif
 
 $(KDIR)strmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o - > strmmkernel_rn.s	
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o - > strmmkernel_rn.s	
 	m4 strmmkernel_rn.s > strmmkernel_rn_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA strmmkernel_rn_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA strmmkernel_rn_nomacros.s -o $@
 	rm strmmkernel_rn.s strmmkernel_rn_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
 endif
 
 $(KDIR)strmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > strmm_kernel_rt.s	
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > strmm_kernel_rt.s	
 	m4 strmm_kernel_rt.s > strmm_kernel_rt_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA strmm_kernel_rt_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA strmm_kernel_rt_nomacros.s -o $@
 	rm strmm_kernel_rt.s strmm_kernel_rt_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
 endif
 
 $(KDIR)dtrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o - > dtrmm_kernel_ln.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o - > dtrmm_kernel_ln.s
 	m4 dtrmm_kernel_ln.s > dtrmm_kernel_ln_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA dtrmm_kernel_ln_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA dtrmm_kernel_ln_nomacros.s -o $@
 	rm dtrmm_kernel_ln.s dtrmm_kernel_ln_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
 endif
 
 $(KDIR)dtrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o - > dtrmm_kernel_lt.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o - > dtrmm_kernel_lt.s
 	m4 dtrmm_kernel_lt.s > dtrmm_kernel_lt_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA dtrmm_kernel_lt_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA dtrmm_kernel_lt_nomacros.s -o $@
 	rm dtrmm_kernel_lt.s dtrmm_kernel_lt_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
 endif
 
 $(KDIR)dtrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o - > dtrmm_kernel_rn.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o - > dtrmm_kernel_rn.s
 	m4 dtrmm_kernel_rn.s > dtrmm_kernel_rn_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA dtrmm_kernel_rn_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA dtrmm_kernel_rn_nomacros.s -o $@
 	rm dtrmm_kernel_rn.s dtrmm_kernel_rn_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
 endif
 
 $(KDIR)dtrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > dtrmm_kernel_rt.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > dtrmm_kernel_rt.s
 	m4 dtrmm_kernel_rt.s > dtrmm_kernel_rt_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA dtrmm_kernel_rt_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA dtrmm_kernel_rt_nomacros.s -o $@
 	rm dtrmm_kernel_rt.s dtrmm_kernel_rt_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
 endif
 
 $(KDIR)qtrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
 
 $(KDIR)qtrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
 
 $(KDIR)qtrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
 
 $(KDIR)qtrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
 
 $(KDIR)ctrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN  $< -o - > ctrmm_kernel_ln.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN  $< -o - > ctrmm_kernel_ln.s
 	m4 ctrmm_kernel_ln.s > ctrmm_kernel_ln_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN ctrmm_kernel_ln_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN ctrmm_kernel_ln_nomacros.s -o $@
 	rm ctrmm_kernel_ln.s ctrmm_kernel_ln_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
 endif
 
 $(KDIR)ctrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o - > ctrmm_kernel_lt.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o - > ctrmm_kernel_lt.s
 	m4 ctrmm_kernel_lt.s > ctrmm_kernel_lt_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN ctrmm_kernel_lt_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN ctrmm_kernel_lt_nomacros.s -o $@
 	rm ctrmm_kernel_lt.s ctrmm_kernel_lt_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
 endif
 
 $(KDIR)ctrmm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o - > ctrmm_kernel_lr.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o - > ctrmm_kernel_lr.s
 	m4 ctrmm_kernel_lr.s > ctrmm_kernel_lr_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN  ctrmm_kernel_lr_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN  ctrmm_kernel_lr_nomacros.s -o $@
 	rm ctrmm_kernel_lr.s ctrmm_kernel_lr_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
 endif
 
 $(KDIR)ctrmm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o - > ctrmm_kernel_lc.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o - > ctrmm_kernel_lc.s
 	m4 ctrmm_kernel_lc.s > ctrmm_kernel_lc_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN ctrmm_kernel_lc_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN ctrmm_kernel_lc_nomacros.s -o $@
 	rm ctrmm_kernel_lc_nomacros.s ctrmm_kernel_lc.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
 endif
 
 $(KDIR)ctrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o - > ctrmm_kernel_rn.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o - > ctrmm_kernel_rn.s
 	m4 ctrmm_kernel_rn.s > ctrmm_kernel_rn_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN ctrmm_kernel_rn_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN ctrmm_kernel_rn_nomacros.s -o $@
 	rm ctrmm_kernel_rn.s ctrmm_kernel_rn_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
 endif
 
 $(KDIR)ctrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o - > ctrmm_kernel_rt.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o - > ctrmm_kernel_rt.s
 	m4 ctrmm_kernel_rt.s > ctrmm_kernel_rt_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN ctrmm_kernel_rt_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN ctrmm_kernel_rt_nomacros.s -o $@
 	rm ctrmm_kernel_rt.s ctrmm_kernel_rt_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
 endif
 
 $(KDIR)ctrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o - > ctrmm_kernel_rr.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o - > ctrmm_kernel_rr.s
 	m4 ctrmm_kernel_rr.s > ctrmm_kernel_rr_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC ctrmm_kernel_rr_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC ctrmm_kernel_rr_nomacros.s -o $@
 	rm ctrmm_kernel_rr.s ctrmm_kernel_rr_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
 endif
 
 $(KDIR)ctrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o - > ctrmm_kernel_RC.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o - > ctrmm_kernel_RC.s
 	m4 ctrmm_kernel_RC.s > ctrmm_kernel_RC_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC ctrmm_kernel_RC_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC ctrmm_kernel_RC_nomacros.s -o $@
 	rm ctrmm_kernel_RC.s ctrmm_kernel_RC_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
 endif
 
 $(KDIR)ztrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_ln.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_ln.s
 	m4 ztrmm_kernel_ln.s > ztrmm_kernel_ln_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN ztrmm_kernel_ln_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN ztrmm_kernel_ln_nomacros.s -o $@
 	rm ztrmm_kernel_ln.s ztrmm_kernel_ln_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
 endif
 
 $(KDIR)ztrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_lt.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_lt.s
 	m4 ztrmm_kernel_lt.s > ztrmm_kernel_lt_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN ztrmm_kernel_lt_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN ztrmm_kernel_lt_nomacros.s -o $@
 	rm ztrmm_kernel_lt.s ztrmm_kernel_lt_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
 endif
 
 $(KDIR)ztrmm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o - > ztrmm_kernel_lr.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o - > ztrmm_kernel_lr.s
 	m4 ztrmm_kernel_lr.s > ztrmm_kernel_lr_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN ztrmm_kernel_lr_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN ztrmm_kernel_lr_nomacros.s -o $@
 	rm ztrmm_kernel_lr.s ztrmm_kernel_lr_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
 endif
 
 $(KDIR)ztrmm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o - > ztrmm_kernel_lc.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o - > ztrmm_kernel_lc.s
 	m4 ztrmm_kernel_lc.s >ztrmm_kernel_lc_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN ztrmm_kernel_lc_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN ztrmm_kernel_lc_nomacros.s -o $@
 	rm ztrmm_kernel_lc.s ztrmm_kernel_lc_nomacros.s 
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
 endif
 
 $(KDIR)ztrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_rn.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_rn.s
 	m4 ztrmm_kernel_rn.s > ztrmm_kernel_rn_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN ztrmm_kernel_rn_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN ztrmm_kernel_rn_nomacros.s -o $@
 	rm ztrmm_kernel_rn.s ztrmm_kernel_rn_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
 endif
 
 $(KDIR)ztrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_rt.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o - > ztrmm_kernel_rt.s
 	m4 ztrmm_kernel_rt.s > ztrmm_kernel_rt_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN ztrmm_kernel_rt_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN ztrmm_kernel_rt_nomacros.s -o $@
 	rm ztrmm_kernel_rt.s ztrmm_kernel_rt_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
 endif
 
 $(KDIR)ztrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o - > ztrmm_kernel_rr.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o - > ztrmm_kernel_rr.s
 	m4 ztrmm_kernel_rr.s > ztrmm_kernel_rr_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC ztrmm_kernel_rr_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC ztrmm_kernel_rr_nomacros.s -o $@
 	rm ztrmm_kernel_rr.s ztrmm_kernel_rr_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
 endif
 
 $(KDIR)ztrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o - > ztrmm_kernel_rc.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o - > ztrmm_kernel_rc.s
 	m4 ztrmm_kernel_rc.s > ztrmm_kernel_rc_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC ztrmm_kernel_rc_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC ztrmm_kernel_rc_nomacros.s -o $@
 	rm ztrmm_kernel_rc.s ztrmm_kernel_rc_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
 endif
 
 else
 $(KDIR)strmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
 
 $(KDIR)strmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
 
 $(KDIR)strmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
 
 $(KDIR)strmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > strmm_kernel_rt.s	
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > strmm_kernel_rt.s	
 	m4 strmm_kernel_rt.s > strmm_kernel_rt_nomacros.s
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA strmm_kernel_rt_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA strmm_kernel_rt_nomacros.s -o $@
 	rm strmm_kernel_rt.s strmm_kernel_rt_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
 endif
 
 $(KDIR)dtrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
 
 $(KDIR)dtrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
 
 $(KDIR)dtrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
 
 $(KDIR)dtrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
 
 $(KDIR)qtrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
 
 $(KDIR)qtrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
 
 $(KDIR)qtrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
 
 $(KDIR)qtrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
 
 $(KDIR)ctrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ctrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ctrmm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
 
 $(KDIR)ctrmm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
 
 $(KDIR)ctrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ctrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ctrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
 
 $(KDIR)ctrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
 
 $(KDIR)ztrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ztrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ztrmm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
 
 $(KDIR)ztrmm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
 
 $(KDIR)ztrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ztrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ztrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
 
 $(KDIR)ztrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
 endif
 
 
 
 
 $(KDIR)xtrmm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)xtrmm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)xtrmm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
 
 $(KDIR)xtrmm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
 
 $(KDIR)xtrmm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)xtrmm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)xtrmm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
 
 $(KDIR)xtrmm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
 
 $(KDIR)cgemm3m_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEMM3MKERNEL)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DNN $< -o $@
 
 $(KDIR)zgemm3m_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM3MKERNEL)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DNN $< -o $@
 
 $(KDIR)xgemm3m_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMM3MKERNEL)
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DNN $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DNN $< -o $@
 
 $(KDIR)strsm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRSMKERNEL_LN) $(STRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -UDOUBLE -DUPPER -DLN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -UCOMPLEX -UDOUBLE -DUPPER -DLN -UCONJ $< -o $@
 
 $(KDIR)strsm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRSMKERNEL_LT) $(STRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -UDOUBLE -UUPPER -DLT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -UCOMPLEX -UDOUBLE -UUPPER -DLT -UCONJ $< -o $@
 
 $(KDIR)strsm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRSMKERNEL_RN) $(STRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -UDOUBLE -DUPPER -DRN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -UCOMPLEX -UDOUBLE -DUPPER -DRN -UCONJ $< -o $@
 
 $(KDIR)strsm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(STRSMKERNEL_RT) $(STRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -UDOUBLE -UUPPER -DRT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -UCOMPLEX -UDOUBLE -UUPPER -DRT -UCONJ $< -o $@
 
 $(KDIR)dtrsm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRSMKERNEL_LN) $(DTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -DDOUBLE -DUPPER -DLN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -UCOMPLEX -DDOUBLE -DUPPER -DLN -UCONJ $< -o $@
 
 $(KDIR)dtrsm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRSMKERNEL_LT) $(DTRSMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ $< -o - > dtrsm_kernel_lt.s
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ $< -o - > dtrsm_kernel_lt.s
 	m4 dtrsm_kernel_lt.s > dtrsm_kernel_lt_nomacros.s
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ dtrsm_kernel_lt_nomacros.s -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ dtrsm_kernel_lt_nomacros.s -o $@
 	rm dtrsm_kernel_lt.s dtrsm_kernel_lt_nomacros.s
 else
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ $< -o $@
 endif
 
 $(KDIR)dtrsm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRSMKERNEL_RN) $(DTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -DDOUBLE -DUPPER -DRN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -UCOMPLEX -DDOUBLE -DUPPER -DRN -UCONJ $< -o $@
 
 $(KDIR)dtrsm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DTRSMKERNEL_RT) $(DTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DRT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DRT -UCONJ $< -o $@
 
 $(KDIR)qtrsm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QTRSMKERNEL_LN) $(QTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -DUPPER -DLN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -DUPPER -DLN -UCONJ $< -o $@
 
 $(KDIR)qtrsm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QTRSMKERNEL_LT) $(QTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -UUPPER -DLT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -UUPPER -DLT -UCONJ $< -o $@
 
 $(KDIR)qtrsm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QTRSMKERNEL_RN) $(QTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -DUPPER -DRN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -DUPPER -DRN -UCONJ $< -o $@
 
 $(KDIR)qtrsm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(QTRSMKERNEL_RT) $(QTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -UUPPER -DRT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -UUPPER -DRT -UCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_LN) $(CTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DLN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DLN -UCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_LT) $(CTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DLT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DLT -UCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_LN) $(CTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DLN -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DLN -DCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_LT) $(CTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DLT -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DLT -DCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_RN) $(CTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DRN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DRN -UCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_RT) $(CTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DRT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DRT -UCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_RN) $(CTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DRN -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DRN -DCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_RT) $(CTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DRT -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DRT -DCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_LN) $(ZTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DLN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DLN -UCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_LT) $(ZTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_LN) $(ZTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DLN -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DLN -DCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_LT) $(ZTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DLT -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DLT -DCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_RN) $(ZTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DRN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DRN -UCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_RT) $(ZTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DRT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DRT -UCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_RN) $(ZTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DRN -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DRN -DCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_RT) $(ZTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DRT -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DRT -DCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_LN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_LN) $(XTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DLN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DLN -UCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_LT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_LT) $(XTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DLT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DLT -UCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_LR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_LN) $(XTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DLN -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DLN -DCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_LC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_LT) $(XTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DLT -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DLT -DCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_RN$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_RN) $(XTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DRN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DRN -UCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_RT$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_RT) $(XTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DRT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DRT -UCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_RR$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_RN) $(XTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DRN -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DRN -DCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_RC$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_RT) $(XTRSMDEPEND)
-	$(CC) -c $(CFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DRT -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DRT -DCONJ $< -o $@
 
 
 $(KDIR)strmm_iunucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_uncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_iunncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_uncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strmm_ilnucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_lncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_ilnncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_lncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)strmm_iutucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_utcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_iutncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_utcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strmm_iltucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_ltcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_iltncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_ltcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)strmm_ounucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_uncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_ounncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_uncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strmm_olnucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_lncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_olnncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_lncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)strmm_outucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_utcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_outncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_utcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strmm_oltucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_ltcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_oltncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_ltcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_iunucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_uncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_iunncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_uncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_ilnucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_lncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_ilnncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_lncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_iutucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_utcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_iutncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_utcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_iltucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_ltcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_iltncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_ltcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_ounucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_uncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_ounncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_uncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_olnucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_lncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_olnncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_lncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_outucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_utcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_outncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_utcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_oltucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_ltcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_oltncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_ltcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_iunucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_uncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_iunncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_uncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_ilnucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_lncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_ilnncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_lncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_iutucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_utcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_iutncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_utcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_iltucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_ltcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_iltncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_ltcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_ounucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_uncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_ounncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_uncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_olnucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_lncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_olnncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_lncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_outucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_utcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_outncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_utcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_oltucopy$(TSUFFIX).$(SUFFIX) : generic/trmm_ltcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_oltncopy$(TSUFFIX).$(SUFFIX) : generic/trmm_ltcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_iunucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_uncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_iunncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_uncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_ilnucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_lncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_ilnncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_lncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_iutucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_utcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_iutncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_utcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_iltucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_ltcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_iltncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_ltcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_ounucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_uncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_ounncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_uncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_olnucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_lncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_olnncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_lncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_outucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_utcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_outncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_utcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_oltucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_ltcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_oltncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_ltcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_iunucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_uncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_iunncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_uncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_ilnucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_lncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_ilnncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_lncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_iutucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_utcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_iutncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_utcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_iltucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_ltcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_iltncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_ltcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_ounucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_uncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_ounncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_uncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_olnucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_lncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_olnncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_lncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_outucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_utcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_outncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_utcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_oltucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_ltcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_oltncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_ltcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_iunucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_uncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_iunncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_uncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_ilnucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_lncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_ilnncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_lncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_iutucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_utcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_iutncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_utcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_iltucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_ltcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_iltncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_ltcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_ounucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_uncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_ounncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_uncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_olnucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_lncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_olnncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_lncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_outucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_utcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_outncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_utcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_oltucopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_ltcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_oltncopy$(TSUFFIX).$(SUFFIX) : generic/ztrmm_ltcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ssymm_outcopy$(TSUFFIX).$(SUFFIX) : generic/symm_ucopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER $< -o $@
 
 $(KDIR)ssymm_oltcopy$(TSUFFIX).$(SUFFIX) : generic/symm_lcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER $< -o $@
 
 $(KDIR)ssymm_iutcopy$(TSUFFIX).$(SUFFIX) : generic/symm_ucopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER $< -o $@
 
 $(KDIR)ssymm_iltcopy$(TSUFFIX).$(SUFFIX) : generic/symm_lcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER $< -o $@
 
 $(KDIR)dsymm_outcopy$(TSUFFIX).$(SUFFIX) : generic/symm_ucopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER $< -o $@
 
 $(KDIR)dsymm_oltcopy$(TSUFFIX).$(SUFFIX) : generic/symm_lcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER $< -o $@
 
 $(KDIR)dsymm_iutcopy$(TSUFFIX).$(SUFFIX) : generic/symm_ucopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER $< -o $@
 
 $(KDIR)dsymm_iltcopy$(TSUFFIX).$(SUFFIX) : generic/symm_lcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER $< -o $@
 
 $(KDIR)qsymm_outcopy$(TSUFFIX).$(SUFFIX) : generic/symm_ucopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER $< -o $@
 
 $(KDIR)qsymm_oltcopy$(TSUFFIX).$(SUFFIX) : generic/symm_lcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER $< -o $@
 
 $(KDIR)qsymm_iutcopy$(TSUFFIX).$(SUFFIX) : generic/symm_ucopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER $< -o $@
 
 $(KDIR)qsymm_iltcopy$(TSUFFIX).$(SUFFIX) : generic/symm_lcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER $< -o $@
 
 $(KDIR)csymm_outcopy$(TSUFFIX).$(SUFFIX) : generic/zsymm_ucopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER $< -o $@
 
 $(KDIR)csymm_oltcopy$(TSUFFIX).$(SUFFIX) : generic/zsymm_lcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER $< -o $@
 
 $(KDIR)csymm_iutcopy$(TSUFFIX).$(SUFFIX) : generic/zsymm_ucopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER $< -o $@
 
 $(KDIR)csymm_iltcopy$(TSUFFIX).$(SUFFIX) : generic/zsymm_lcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER $< -o $@
 
 $(KDIR)zsymm_outcopy$(TSUFFIX).$(SUFFIX) : generic/zsymm_ucopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER $< -o $@
 
 $(KDIR)zsymm_oltcopy$(TSUFFIX).$(SUFFIX) : generic/zsymm_lcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER $< -o $@
 
 $(KDIR)zsymm_iutcopy$(TSUFFIX).$(SUFFIX) : generic/zsymm_ucopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER $< -o $@
 
 $(KDIR)zsymm_iltcopy$(TSUFFIX).$(SUFFIX) : generic/zsymm_lcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER $< -o $@
 
 $(KDIR)xsymm_outcopy$(TSUFFIX).$(SUFFIX) : generic/zsymm_ucopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER $< -o $@
 
 $(KDIR)xsymm_oltcopy$(TSUFFIX).$(SUFFIX) : generic/zsymm_lcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER $< -o $@
 
 $(KDIR)xsymm_iutcopy$(TSUFFIX).$(SUFFIX) : generic/zsymm_ucopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER $< -o $@
 
 $(KDIR)xsymm_iltcopy$(TSUFFIX).$(SUFFIX) : generic/zsymm_lcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER $< -o $@
 
 $(KDIR)chemm_outcopy$(TSUFFIX).$(SUFFIX) : generic/zhemm_utcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER $< -ULOWER -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER $< -ULOWER -o $@
 
 $(KDIR)chemm_oltcopy$(TSUFFIX).$(SUFFIX) : generic/zhemm_ltcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER $< -DLOWER -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER $< -DLOWER -o $@
 
 $(KDIR)chemm_iutcopy$(TSUFFIX).$(SUFFIX) : generic/zhemm_utcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER $< -ULOWER -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER $< -ULOWER -o $@
 
 $(KDIR)chemm_iltcopy$(TSUFFIX).$(SUFFIX) : generic/zhemm_ltcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER $< -DLOWER -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER $< -DLOWER -o $@
 
 $(KDIR)zhemm_outcopy$(TSUFFIX).$(SUFFIX) : generic/zhemm_utcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER $< -ULOWER -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER $< -ULOWER -o $@
 
 $(KDIR)zhemm_oltcopy$(TSUFFIX).$(SUFFIX) : generic/zhemm_ltcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER $< -DLOWER -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER $< -DLOWER -o $@
 
 $(KDIR)zhemm_iutcopy$(TSUFFIX).$(SUFFIX) : generic/zhemm_utcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER $< -ULOWER -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER $< -ULOWER -o $@
 
 $(KDIR)zhemm_iltcopy$(TSUFFIX).$(SUFFIX) : generic/zhemm_ltcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER $< -DLOWER -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER $< -DLOWER -o $@
 
 $(KDIR)xhemm_outcopy$(TSUFFIX).$(SUFFIX) : generic/zhemm_utcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER $< -ULOWER -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER $< -ULOWER -o $@
 
 $(KDIR)xhemm_oltcopy$(TSUFFIX).$(SUFFIX) : generic/zhemm_ltcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER $< -DLOWER -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER $< -DLOWER -o $@
 
 $(KDIR)xhemm_iutcopy$(TSUFFIX).$(SUFFIX) : generic/zhemm_utcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER $< -ULOWER -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER $< -ULOWER -o $@
 
 $(KDIR)xhemm_iltcopy$(TSUFFIX).$(SUFFIX) : generic/zhemm_ltcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER $< -DLOWER -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER $< -DLOWER -o $@
 
 $(KDIR)cgemm3m_oncopyb$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)cgemm3m_oncopyr$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)cgemm3m_oncopyi$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)cgemm3m_otcopyb$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)cgemm3m_otcopyr$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)cgemm3m_otcopyi$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)cgemm3m_incopyb$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
 
 $(KDIR)cgemm3m_incopyr$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)cgemm3m_incopyi$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)cgemm3m_itcopyb$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
 
 $(KDIR)cgemm3m_itcopyr$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)cgemm3m_itcopyi$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zgemm3m_oncopyb$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)zgemm3m_oncopyr$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zgemm3m_oncopyi$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zgemm3m_otcopyb$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)zgemm3m_otcopyr$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zgemm3m_otcopyi$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zgemm3m_incopyb$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
 
 $(KDIR)zgemm3m_incopyr$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zgemm3m_incopyi$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zgemm3m_itcopyb$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
 
 $(KDIR)zgemm3m_itcopyr$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zgemm3m_itcopyi$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xgemm3m_oncopyb$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)xgemm3m_oncopyr$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xgemm3m_oncopyi$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xgemm3m_otcopyb$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)xgemm3m_otcopyr$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xgemm3m_otcopyi$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xgemm3m_incopyb$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
 
 $(KDIR)xgemm3m_incopyr$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xgemm3m_incopyi$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_ncopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xgemm3m_itcopyb$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
 
 $(KDIR)xgemm3m_itcopyr$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xgemm3m_itcopyi$(TSUFFIX).$(SUFFIX) : generic/zgemm3m_tcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)csymm3m_oucopyb$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)csymm3m_olcopyb$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)csymm3m_oucopyr$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)csymm3m_olcopyr$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)csymm3m_oucopyi$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)csymm3m_olcopyi$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)csymm3m_iucopyb$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)csymm3m_ilcopyb$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)csymm3m_iucopyr$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)csymm3m_ilcopyr$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)csymm3m_iucopyi$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)csymm3m_ilcopyi$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zsymm3m_oucopyb$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)zsymm3m_olcopyb$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)zsymm3m_oucopyr$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zsymm3m_olcopyr$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zsymm3m_oucopyi$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zsymm3m_olcopyi$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zsymm3m_iucopyb$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)zsymm3m_ilcopyb$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)zsymm3m_iucopyr$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zsymm3m_ilcopyr$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zsymm3m_iucopyi$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zsymm3m_ilcopyi$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xsymm3m_oucopyb$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)xsymm3m_olcopyb$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)xsymm3m_oucopyr$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xsymm3m_olcopyr$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xsymm3m_oucopyi$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xsymm3m_olcopyi$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xsymm3m_iucopyb$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)xsymm3m_ilcopyb$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)xsymm3m_iucopyr$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xsymm3m_ilcopyr$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xsymm3m_iucopyi$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_ucopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xsymm3m_ilcopyi$(TSUFFIX).$(SUFFIX) : generic/zsymm3m_lcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)chemm3m_oucopyb$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)chemm3m_olcopyb$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)chemm3m_oucopyr$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)chemm3m_olcopyr$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)chemm3m_oucopyi$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)chemm3m_olcopyi$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)chemm3m_iucopyb$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)chemm3m_ilcopyb$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)chemm3m_iucopyr$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)chemm3m_ilcopyr$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)chemm3m_iucopyi$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)chemm3m_ilcopyi$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zhemm3m_oucopyb$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)zhemm3m_olcopyb$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)zhemm3m_oucopyr$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zhemm3m_olcopyr$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zhemm3m_oucopyi$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zhemm3m_olcopyi$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zhemm3m_iucopyb$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)zhemm3m_ilcopyb$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)zhemm3m_iucopyr$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zhemm3m_ilcopyr$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zhemm3m_iucopyi$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zhemm3m_ilcopyi$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xhemm3m_oucopyb$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)xhemm3m_olcopyb$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)xhemm3m_oucopyr$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xhemm3m_olcopyr$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xhemm3m_oucopyi$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xhemm3m_olcopyi$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xhemm3m_iucopyb$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)xhemm3m_ilcopyb$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)xhemm3m_iucopyr$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xhemm3m_ilcopyr$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xhemm3m_iucopyi$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_ucopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xhemm3m_ilcopyi$(TSUFFIX).$(SUFFIX) : generic/zhemm3m_lcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(CFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)strsm_iunucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_uncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_iunncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_uncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strsm_ilnucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_lncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_ilnncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_lncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)strsm_iutucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_utcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_iutncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_utcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strsm_iltucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_ltcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_iltncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_ltcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)strsm_ounucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_uncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_ounncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_uncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strsm_olnucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_lncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_olnncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_lncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)strsm_outucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_utcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_outncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_utcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strsm_oltucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_ltcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_oltncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_ltcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_iunucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_uncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_iunncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_uncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_ilnucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_lncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_ilnncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_lncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_iutucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_utcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_iutncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_utcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_iltucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_ltcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_iltncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_ltcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_ounucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_uncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_ounncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_uncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_olnucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_lncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_olnncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_lncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_outucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_utcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_outncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_utcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_oltucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_ltcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_oltncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_ltcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_iunucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_uncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_iunncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_uncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_ilnucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_lncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_ilnncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_lncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_iutucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_utcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_iutncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_utcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_iltucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_ltcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_iltncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_ltcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_ounucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_uncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_ounncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_uncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_olnucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_lncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_olnncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_lncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_outucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_utcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_outncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_utcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_oltucopy$(TSUFFIX).$(SUFFIX) : generic/trsm_ltcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_oltncopy$(TSUFFIX).$(SUFFIX) : generic/trsm_ltcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_iunucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_uncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_iunncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_uncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_ilnucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_lncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_ilnncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_lncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_iutucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_utcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_iutncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_utcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_iltucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_ltcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_iltncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_ltcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_ounucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_uncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_ounncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_uncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_olnucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_lncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_olnncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_lncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_outucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_utcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_outncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_utcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_oltucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_ltcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_oltncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_ltcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_iunucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_uncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_iunncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_uncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_ilnucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_lncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_ilnncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_lncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_iutucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_utcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_iutncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_utcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_iltucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_ltcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_iltncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_ltcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_ounucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_uncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_ounncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_uncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_olnucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_lncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_olnncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_lncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_outucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_utcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_outncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_utcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_oltucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_ltcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_oltncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_ltcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_iunucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_uncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_iunncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_uncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_ilnucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_lncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_ilnncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_lncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_iutucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_utcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_iutncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_utcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_iltucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_ltcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_iltncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_ltcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_ounucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_uncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_ounncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_uncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_olnucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_lncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_olnncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_lncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_outucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_utcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_outncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_utcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_oltucopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_ltcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_oltncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_ltcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 
 $(KDIR)sgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SGEMM_BETA)
-	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 ifeq ($(BUILD_BFLOAT16),1)
 $(KDIR)sbgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SBGEMM_BETA)
-	$(CC) $(PFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
 endif
 
 $(KDIR)dgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DGEMM_BETA)
-	$(CC) $(PFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)qgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QGEMM_BETA)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)cgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMM_BETA)
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX $< -o $@
 
 $(KDIR)zgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMM_BETA)
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX $< -o $@
 
 $(KDIR)xgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMM_BETA)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX $< -o $@
 
 
 ifeq ($(BUILD_BFLOAT16), 1)
 $(SBGEMMONCOPYOBJ_P) : $(KERNELDIR)/$(SBGEMMONCOPY)
-	$(CC) $(PFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
 
 $(SBGEMMOTCOPYOBJ_P) : $(KERNELDIR)/$(SBGEMMOTCOPY)
-	$(CC) $(PFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
 
 ifneq ($(SBGEMM_UNROLL_M), $(SBGEMM_UNROLL_N))
 $(SBGEMMINCOPYOBJ_P) : $(KERNELDIR)/$(SBGEMMINCOPY)
-	$(CC) $(PFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
 
 $(SBGEMMITCOPYOBJ_P) : $(KERNELDIR)/$(SBGEMMITCOPY)
-	$(CC) $(PFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
 
 endif
 endif
 
 $(SGEMMONCOPYOBJ_P) : $(KERNELDIR)/$(SGEMMONCOPY)
-	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 $(SGEMMOTCOPYOBJ_P) : $(KERNELDIR)/$(SGEMMOTCOPY)
-	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N))
 
 $(SGEMMINCOPYOBJ_P) : $(KERNELDIR)/$(SGEMMINCOPY)
-	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 $(SGEMMITCOPYOBJ_P) : $(KERNELDIR)/$(SGEMMITCOPY)
-	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 endif
 
 $(DGEMMONCOPYOBJ_P) : $(KERNELDIR)/$(DGEMMONCOPY)
-	$(CC) $(PFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 $(DGEMMOTCOPYOBJ_P) : $(KERNELDIR)/$(DGEMMOTCOPY)
-	$(CC) $(PFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 ifneq ($(DGEMM_UNROLL_M), $(DGEMM_UNROLL_N))
 
 $(DGEMMINCOPYOBJ_P) : $(KERNELDIR)/$(DGEMMINCOPY)
-	$(CC) $(PFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 $(DGEMMITCOPYOBJ_P) : $(KERNELDIR)/$(DGEMMITCOPY)
-	$(CC) $(PFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 endif
 
 ifdef EXPRECISION
 
 $(QGEMMONCOPYOBJ_P) : $(KERNELDIR)/$(QGEMMONCOPY)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 $(QGEMMOTCOPYOBJ_P) : $(KERNELDIR)/$(QGEMMOTCOPY)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 ifneq ($(QGEMM_UNROLL_M), $(QGEMM_UNROLL_N))
 
 $(QGEMMINCOPYOBJ_P) : $(KERNELDIR)/$(QGEMMINCOPY)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 $(QGEMMITCOPYOBJ_P) : $(KERNELDIR)/$(QGEMMITCOPY)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 endif
 
 endif
 
 $(CGEMMONCOPYOBJ_P) : $(KERNELDIR)/$(CGEMMONCOPY)
-	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 $(CGEMMOTCOPYOBJ_P) : $(KERNELDIR)/$(CGEMMOTCOPY)
-	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 ifneq ($(CGEMM_UNROLL_M), $(CGEMM_UNROLL_N))
 
 $(CGEMMINCOPYOBJ_P) : $(KERNELDIR)/$(CGEMMINCOPY)
-	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 $(CGEMMITCOPYOBJ_P) : $(KERNELDIR)/$(CGEMMITCOPY)
-	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 endif
 
 $(ZGEMMONCOPYOBJ_P) : $(KERNELDIR)/$(ZGEMMONCOPY)
-	$(CC) $(PFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 $(ZGEMMOTCOPYOBJ_P) : $(KERNELDIR)/$(ZGEMMOTCOPY)
-	$(CC) $(PFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 ifneq ($(ZGEMM_UNROLL_M), $(ZGEMM_UNROLL_N))
 
 $(ZGEMMINCOPYOBJ_P) : $(KERNELDIR)/$(ZGEMMINCOPY)
-	$(CC) $(PFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 $(ZGEMMITCOPYOBJ_P) : $(KERNELDIR)/$(ZGEMMITCOPY)
-	$(CC) $(PFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 endif
 
 ifdef EXPRECISION
 
 $(XGEMMONCOPYOBJ_P) : $(KERNELDIR)/$(XGEMMONCOPY)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 $(XGEMMOTCOPYOBJ_P) : $(KERNELDIR)/$(XGEMMOTCOPY)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 ifneq ($(XGEMM_UNROLL_M), $(XGEMM_UNROLL_N))
 
 $(XGEMMINCOPYOBJ_P) : $(KERNELDIR)/$(XGEMMINCOPY)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 $(XGEMMITCOPYOBJ_P) : $(KERNELDIR)/$(XGEMMITCOPY)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 endif
 
@@ -2520,1301 +3135,1732 @@ endif
 
 ifeq ($(BUILD_BFLOAT16), 1)
 $(KDIR)sbgemm_kernel$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SBGEMMKERNEL) $(SBGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
 endif
 
 $(KDIR)sgemm_kernel$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL) $(SGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)dgemm_kernel$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL) $(DGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)qgemm_kernel$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL) $(QGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -UCOMPLEX $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -UCOMPLEX $< -o $@
 
 $(KDIR)cgemm_kernel_n$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DNN $< -o $@
 
 $(KDIR)cgemm_kernel_l$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DCN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DCN $< -o $@
 
 $(KDIR)cgemm_kernel_r$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
 ifeq ($(OS), AIX)
-	$(CC) $(PFLAGS) -S -UDOUBLE -DCOMPLEX -DNC $< -o - > cgemm_kernel_r.s
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -E -UDOUBLE -DCOMPLEX -DNC $< -o - > cgemm_kernel_r.s
 	m4 cgemm_kernel_r.s > cgemm_kernel_r_nomacros.s
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DNC cgemm_kernel_r_nomacros.s -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DNC cgemm_kernel_r_nomacros.s -o $@
 	rm cgemm_kernel_r.s cgemm_kernel_r_nomacros.s 
 else
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DNC $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DNC $< -o $@
 endif
 
 $(KDIR)cgemm_kernel_b$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL) $(CGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DCC $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DCC $< -o $@
 
 $(KDIR)zgemm_kernel_n$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DNN $< -o $@
 
 $(KDIR)zgemm_kernel_l$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DCN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DCN $< -o $@
 
 $(KDIR)zgemm_kernel_r$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DNC $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DNC $< -o $@
 
 $(KDIR)zgemm_kernel_b$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL) $(ZGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DCC $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DCC $< -o $@
 
 $(KDIR)xgemm_kernel_n$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL) $(XGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DNN $< -o $@
 
 $(KDIR)xgemm_kernel_l$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL) $(XGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DCN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DCN $< -o $@
 
 $(KDIR)xgemm_kernel_r$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL) $(XGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DNC $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DNC $< -o $@
 
 $(KDIR)xgemm_kernel_b$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL) $(XGEMMDEPEND)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DCC $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DCC $< -o $@
 
 $(KDIR)strmm_kernel_LN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
 
 $(KDIR)strmm_kernel_LT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
 
 $(KDIR)strmm_kernel_RN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
 
 $(KDIR)strmm_kernel_RT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL)
 ifeq ($(OS), AIX)
-	$(CC) $(CFLAGS) -S -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > strmm_kernel_rt.s	
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -E -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o - > strmm_kernel_rt.s	
 	m4 strmmkernel_rn.s > strmm_kernel_rt_nomacros.s
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA strmm_kernel_rt_nomacros.s -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA strmm_kernel_rt_nomacros.s -o $@
 	rm strmm_kernel_rt.s strmm_kernel_rt_nomacros.s
 else
-	$(CC) $(CFLAGS) -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DTRMMKERNEL -UDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
 endif
 
 $(KDIR)dtrmm_kernel_LN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
 
 $(KDIR)dtrmm_kernel_LT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
 
 $(KDIR)dtrmm_kernel_RN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
 
 $(KDIR)dtrmm_kernel_RT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
 
 $(KDIR)qtrmm_kernel_LN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -UTRANSA $< -o $@
 
 $(KDIR)qtrmm_kernel_LT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -DLEFT -DTRANSA $< -o $@
 
 $(KDIR)qtrmm_kernel_RN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -UTRANSA $< -o $@
 
 $(KDIR)qtrmm_kernel_RT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DXDOUBLE -UCOMPLEX -ULEFT -DTRANSA $< -o $@
 
 $(KDIR)ctrmm_kernel_LN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ctrmm_kernel_LT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ctrmm_kernel_LR$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
 
 $(KDIR)ctrmm_kernel_LC$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
 
 $(KDIR)ctrmm_kernel_RN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ctrmm_kernel_RT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ctrmm_kernel_RR$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
 
 $(KDIR)ctrmm_kernel_RC$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -UDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
 
 $(KDIR)ztrmm_kernel_LN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ztrmm_kernel_LT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ztrmm_kernel_LR$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
 
 $(KDIR)ztrmm_kernel_LC$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
 
 $(KDIR)ztrmm_kernel_RN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ztrmm_kernel_RT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)ztrmm_kernel_RR$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
 
 $(KDIR)ztrmm_kernel_RC$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
 
 $(KDIR)xtrmm_kernel_LN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -UTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)xtrmm_kernel_LT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -DTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)xtrmm_kernel_LR$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -UTRANSA -DCONJ -DCN $< -o $@
 
 $(KDIR)xtrmm_kernel_LC$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -DLEFT -DTRANSA -DCONJ -DCN $< -o $@
 
 $(KDIR)xtrmm_kernel_RN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -UTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)xtrmm_kernel_RT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -DTRANSA -UCONJ -DNN $< -o $@
 
 $(KDIR)xtrmm_kernel_RR$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -UTRANSA -DCONJ -DNC $< -o $@
 
 $(KDIR)xtrmm_kernel_RC$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMMKERNEL)
-	$(CC) $(PFLAGS) -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DTRMMKERNEL -DXDOUBLE -DCOMPLEX -ULEFT -DTRANSA -DCONJ -DNC $< -o $@
 
 $(KDIR)cgemm3m_kernel$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CGEMM3MKERNEL)
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DNN $< -o $@
 
 $(KDIR)zgemm3m_kernel$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMM3MKERNEL)
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DNN $< -o $@
 
 $(KDIR)xgemm3m_kernel$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMM3MKERNEL)
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DNN $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DNN $< -o $@
 
 $(KDIR)strsm_kernel_LN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(STRSMKERNEL_LN) $(STRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -UCOMPLEX -UDOUBLE -DUPPER -DLN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -UCOMPLEX -UDOUBLE -DUPPER -DLN -UCONJ $< -o $@
 
 $(KDIR)strsm_kernel_LT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(STRSMKERNEL_LT) $(STRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -UCOMPLEX -UDOUBLE -UUPPER -DLT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -UCOMPLEX -UDOUBLE -UUPPER -DLT -UCONJ $< -o $@
 
 $(KDIR)strsm_kernel_RN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(STRSMKERNEL_RN) $(STRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -UCOMPLEX -UDOUBLE -DUPPER -DRN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -UCOMPLEX -UDOUBLE -DUPPER -DRN -UCONJ $< -o $@
 
 $(KDIR)strsm_kernel_RT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(STRSMKERNEL_RT) $(STRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -UCOMPLEX -UDOUBLE -UUPPER -DRT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -UCOMPLEX -UDOUBLE -UUPPER -DRT -UCONJ $< -o $@
 
 $(KDIR)dtrsm_kernel_LN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DTRSMKERNEL_LN) $(DTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -UCOMPLEX -DDOUBLE -DUPPER -DLN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -UCOMPLEX -DDOUBLE -DUPPER -DLN -UCONJ $< -o $@
 
 $(KDIR)dtrsm_kernel_LT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DTRSMKERNEL_LT) $(DTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ $< -o $@
 
 $(KDIR)dtrsm_kernel_RN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DTRSMKERNEL_RN) $(DTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -UCOMPLEX -DDOUBLE -DUPPER -DRN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -UCOMPLEX -DDOUBLE -DUPPER -DRN -UCONJ $< -o $@
 
 $(KDIR)dtrsm_kernel_RT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DTRSMKERNEL_RT) $(DTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DRT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -UCOMPLEX -DDOUBLE -UUPPER -DRT -UCONJ $< -o $@
 
 $(KDIR)qtrsm_kernel_LN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QTRSMKERNEL_LN) $(QTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -DUPPER -DLN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -DUPPER -DLN -UCONJ $< -o $@
 
 $(KDIR)qtrsm_kernel_LT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QTRSMKERNEL_LT) $(QTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -UUPPER -DLT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -UUPPER -DLT -UCONJ $< -o $@
 
 $(KDIR)qtrsm_kernel_RN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QTRSMKERNEL_RN) $(QTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -DUPPER -DRN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -DUPPER -DRN -UCONJ $< -o $@
 
 $(KDIR)qtrsm_kernel_RT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(QTRSMKERNEL_RT) $(QTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -UUPPER -DRT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -UCOMPLEX -DXDOUBLE -UUPPER -DRT -UCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_LN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_LN) $(CTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DLN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DLN -UCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_LT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_LT) $(CTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DLT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DLT -UCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_LR$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_LN) $(CTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DLN -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DLN -DCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_LC$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_LT) $(CTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DLT -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DLT -DCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_RN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_RN) $(CTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DRN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DRN -UCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_RT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_RT) $(CTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DRT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DRT -UCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_RR$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_RN) $(CTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DRN -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -DUPPER -DRN -DCONJ $< -o $@
 
 $(KDIR)ctrsm_kernel_RC$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(CTRSMKERNEL_RT) $(CTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DRT -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -UDOUBLE -UUPPER -DRT -DCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_LN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_LN) $(ZTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DLN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DLN -UCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_LT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_LT) $(ZTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DLT -UCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_LR$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_LN) $(ZTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DLN -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DLN -DCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_LC$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_LT) $(ZTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DLT -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DLT -DCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_RN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_RN) $(ZTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DRN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DRN -UCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_RT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_RT) $(ZTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DRT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DRT -UCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_RR$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_RN) $(ZTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DRN -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -DUPPER -DRN -DCONJ $< -o $@
 
 $(KDIR)ztrsm_kernel_RC$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZTRSMKERNEL_RT) $(ZTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DRT -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DDOUBLE -UUPPER -DRT -DCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_LN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_LN) $(XTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DLN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DLN -UCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_LT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_LT) $(XTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DLT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DLT -UCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_LR$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_LN) $(XTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DLN -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DLN -DCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_LC$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_LT) $(XTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DLT -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DLT -DCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_RN$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_RN) $(XTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DRN -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DRN -UCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_RT$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_RT) $(XTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DRT -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DRT -UCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_RR$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_RN) $(XTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DRN -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -DUPPER -DRN -DCONJ $< -o $@
 
 $(KDIR)xtrsm_kernel_RC$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XTRSMKERNEL_RT) $(XTRSMDEPEND)
-	$(CC) -c $(PFLAGS) -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DRT -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS -DTRSMKERNEL -DCOMPLEX -DXDOUBLE -UUPPER -DRT -DCONJ $< -o $@
 
 
 $(KDIR)strmm_iunucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_uncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_iunncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_uncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strmm_ilnucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_lncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_ilnncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_lncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)strmm_iutucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_utcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_iutncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_utcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strmm_iltucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_ltcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_iltncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_ltcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)strmm_ounucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_uncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_ounncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_uncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strmm_olnucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_lncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_olnncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_lncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)strmm_outucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_utcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_outncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_utcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strmm_oltucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_ltcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strmm_oltncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_ltcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_iunucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_uncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_iunncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_uncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_ilnucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_lncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_ilnncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_lncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_iutucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_utcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_iutncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_utcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_iltucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_ltcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_iltncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_ltcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_ounucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_uncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_ounncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_uncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_olnucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_lncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_olnncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_lncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_outucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_utcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_outncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_utcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrmm_oltucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_ltcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrmm_oltncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_ltcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_iunucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_uncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_iunncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_uncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_ilnucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_lncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_ilnncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_lncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_iutucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_utcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_iutncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_utcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_iltucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_ltcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_iltncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_ltcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_ounucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_uncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_ounncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_uncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_olnucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_lncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_olnncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_lncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_outucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_utcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_outncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_utcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrmm_oltucopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_ltcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrmm_oltncopy$(TSUFFIX).$(PSUFFIX) : generic/trmm_ltcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_iunucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_uncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_iunncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_uncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_ilnucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_lncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_ilnncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_lncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_iutucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_utcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_iutncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_utcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_iltucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_ltcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_iltncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_ltcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_ounucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_uncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_ounncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_uncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_olnucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_lncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_olnncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_lncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_outucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_utcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_outncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_utcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrmm_oltucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_ltcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrmm_oltncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_ltcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_iunucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_uncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_iunncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_uncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_ilnucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_lncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_ilnncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_lncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_iutucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_utcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_iutncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_utcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_iltucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_ltcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_iltncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_ltcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_ounucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_uncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_ounncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_uncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_olnucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_lncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_olnncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_lncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_outucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_utcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_outncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_utcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrmm_oltucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_ltcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrmm_oltncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_ltcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_iunucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_uncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_iunncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_uncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_ilnucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_lncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_ilnncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_lncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_iutucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_utcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_iutncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_utcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_iltucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_ltcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_iltncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_ltcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_ounucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_uncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_ounncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_uncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_olnucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_lncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_olnncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_lncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_outucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_utcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_outncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_utcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrmm_oltucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_ltcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrmm_oltncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrmm_ltcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ssymm_outcopy$(TSUFFIX).$(PSUFFIX) : generic/symm_ucopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER $< -o $@
 
 $(KDIR)ssymm_oltcopy$(TSUFFIX).$(PSUFFIX) : generic/symm_lcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER $< -o $@
 
 $(KDIR)ssymm_iutcopy$(TSUFFIX).$(PSUFFIX) : generic/symm_ucopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER $< -o $@
 
 $(KDIR)ssymm_iltcopy$(TSUFFIX).$(PSUFFIX) : generic/symm_lcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER $< -o $@
 
 $(KDIR)dsymm_outcopy$(TSUFFIX).$(PSUFFIX) : generic/symm_ucopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER $< -o $@
 
 $(KDIR)dsymm_oltcopy$(TSUFFIX).$(PSUFFIX) : generic/symm_lcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER $< -o $@
 
 $(KDIR)dsymm_iutcopy$(TSUFFIX).$(PSUFFIX) : generic/symm_ucopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER $< -o $@
 
 $(KDIR)dsymm_iltcopy$(TSUFFIX).$(PSUFFIX) : generic/symm_lcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER $< -o $@
 
 $(KDIR)qsymm_outcopy$(TSUFFIX).$(PSUFFIX) : generic/symm_ucopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER $< -o $@
 
 $(KDIR)qsymm_oltcopy$(TSUFFIX).$(PSUFFIX) : generic/symm_lcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER $< -o $@
 
 $(KDIR)qsymm_iutcopy$(TSUFFIX).$(PSUFFIX) : generic/symm_ucopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER $< -o $@
 
 $(KDIR)qsymm_iltcopy$(TSUFFIX).$(PSUFFIX) : generic/symm_lcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER $< -o $@
 
 $(KDIR)csymm_outcopy$(TSUFFIX).$(PSUFFIX) : generic/zsymm_ucopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER $< -o $@
 
 $(KDIR)csymm_oltcopy$(TSUFFIX).$(PSUFFIX) : generic/zsymm_lcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER $< -o $@
 
 $(KDIR)csymm_iutcopy$(TSUFFIX).$(PSUFFIX) : generic/zsymm_ucopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER $< -o $@
 
 $(KDIR)csymm_iltcopy$(TSUFFIX).$(PSUFFIX) : generic/zsymm_lcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER $< -o $@
 
 $(KDIR)zsymm_outcopy$(TSUFFIX).$(PSUFFIX) : generic/zsymm_ucopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER $< -o $@
 
 $(KDIR)zsymm_oltcopy$(TSUFFIX).$(PSUFFIX) : generic/zsymm_lcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER $< -o $@
 
 $(KDIR)zsymm_iutcopy$(TSUFFIX).$(PSUFFIX) : generic/zsymm_ucopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER $< -o $@
 
 $(KDIR)zsymm_iltcopy$(TSUFFIX).$(PSUFFIX) : generic/zsymm_lcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER $< -o $@
 
 $(KDIR)xsymm_outcopy$(TSUFFIX).$(PSUFFIX) : generic/zsymm_ucopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER $< -o $@
 
 $(KDIR)xsymm_oltcopy$(TSUFFIX).$(PSUFFIX) : generic/zsymm_lcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER $< -o $@
 
 $(KDIR)xsymm_iutcopy$(TSUFFIX).$(PSUFFIX) : generic/zsymm_ucopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER $< -o $@
 
 $(KDIR)xsymm_iltcopy$(TSUFFIX).$(PSUFFIX) : generic/zsymm_lcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER $< -o $@
 
 $(KDIR)chemm_outcopy$(TSUFFIX).$(PSUFFIX) : generic/zhemm_utcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER $< -ULOWER -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER $< -ULOWER -o $@
 
 $(KDIR)chemm_oltcopy$(TSUFFIX).$(PSUFFIX) : generic/zhemm_ltcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER $< -DLOWER -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER $< -DLOWER -o $@
 
 $(KDIR)chemm_iutcopy$(TSUFFIX).$(PSUFFIX) : generic/zhemm_utcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER $< -ULOWER -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER $< -ULOWER -o $@
 
 $(KDIR)chemm_iltcopy$(TSUFFIX).$(PSUFFIX) : generic/zhemm_ltcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER $< -DLOWER -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER $< -DLOWER -o $@
 
 $(KDIR)zhemm_outcopy$(TSUFFIX).$(PSUFFIX) : generic/zhemm_utcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER $< -ULOWER -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER $< -ULOWER -o $@
 
 $(KDIR)zhemm_oltcopy$(TSUFFIX).$(PSUFFIX) : generic/zhemm_ltcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER $< -DLOWER -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER $< -DLOWER -o $@
 
 $(KDIR)zhemm_iutcopy$(TSUFFIX).$(PSUFFIX) : generic/zhemm_utcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER $< -ULOWER -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER $< -ULOWER -o $@
 
 $(KDIR)zhemm_iltcopy$(TSUFFIX).$(PSUFFIX) : generic/zhemm_ltcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER $< -DLOWER -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER $< -DLOWER -o $@
 
 $(KDIR)xhemm_outcopy$(TSUFFIX).$(PSUFFIX) : generic/zhemm_utcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER $< -ULOWER -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER $< -ULOWER -o $@
 
 $(KDIR)xhemm_oltcopy$(TSUFFIX).$(PSUFFIX) : generic/zhemm_ltcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER $< -DLOWER -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER $< -DLOWER -o $@
 
 $(KDIR)xhemm_iutcopy$(TSUFFIX).$(PSUFFIX) : generic/zhemm_utcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER $< -ULOWER -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER $< -ULOWER -o $@
 
 $(KDIR)xhemm_iltcopy$(TSUFFIX).$(PSUFFIX) : generic/zhemm_ltcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER $< -DLOWER -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER $< -DLOWER -o $@
 
 $(KDIR)cgemm3m_oncopyb$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)cgemm3m_oncopyr$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)cgemm3m_oncopyi$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)cgemm3m_otcopyb$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)cgemm3m_otcopyr$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)cgemm3m_otcopyi$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)cgemm3m_incopyb$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
 
 $(KDIR)cgemm3m_incopyr$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)cgemm3m_incopyi$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)cgemm3m_itcopyb$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
 
 $(KDIR)cgemm3m_itcopyr$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)cgemm3m_itcopyi$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -UDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zgemm3m_oncopyb$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)zgemm3m_oncopyr$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zgemm3m_oncopyi$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zgemm3m_otcopyb$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)zgemm3m_otcopyr$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zgemm3m_otcopyi$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zgemm3m_incopyb$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
 
 $(KDIR)zgemm3m_incopyr$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zgemm3m_incopyi$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zgemm3m_itcopyb$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
 
 $(KDIR)zgemm3m_itcopyr$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zgemm3m_itcopyi$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xgemm3m_oncopyb$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)xgemm3m_oncopyr$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xgemm3m_oncopyi$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xgemm3m_otcopyb$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)xgemm3m_otcopyr$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xgemm3m_otcopyi$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xgemm3m_incopyb$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
 
 $(KDIR)xgemm3m_incopyr$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xgemm3m_incopyi$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_ncopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xgemm3m_itcopyb$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA $< -o $@
 
 $(KDIR)xgemm3m_itcopyr$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xgemm3m_itcopyi$(TSUFFIX).$(PSUFFIX) : generic/zgemm3m_tcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS -c -DXDOUBLE -DCOMPLEX -DICOPY -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)csymm3m_oucopyb$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)csymm3m_olcopyb$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)csymm3m_oucopyr$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)csymm3m_olcopyr$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)csymm3m_oucopyi$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)csymm3m_olcopyi$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)csymm3m_iucopyb$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)csymm3m_ilcopyb$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)csymm3m_iucopyr$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)csymm3m_ilcopyr$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)csymm3m_iucopyi$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)csymm3m_ilcopyi$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zsymm3m_oucopyb$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)zsymm3m_olcopyb$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)zsymm3m_oucopyr$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zsymm3m_olcopyr$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zsymm3m_oucopyi$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zsymm3m_olcopyi$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zsymm3m_iucopyb$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)zsymm3m_ilcopyb$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)zsymm3m_iucopyr$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zsymm3m_ilcopyr$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zsymm3m_iucopyi$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zsymm3m_ilcopyi$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xsymm3m_oucopyb$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)xsymm3m_olcopyb$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)xsymm3m_oucopyr$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xsymm3m_olcopyr$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xsymm3m_oucopyi$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xsymm3m_olcopyi$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xsymm3m_iucopyb$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)xsymm3m_ilcopyb$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)xsymm3m_iucopyr$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xsymm3m_ilcopyr$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xsymm3m_iucopyi$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_ucopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xsymm3m_ilcopyi$(TSUFFIX).$(PSUFFIX) : generic/zsymm3m_lcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)chemm3m_oucopyb$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)chemm3m_olcopyb$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)chemm3m_oucopyr$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)chemm3m_olcopyr$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)chemm3m_oucopyi$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)chemm3m_olcopyi$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(CGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)chemm3m_iucopyb$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)chemm3m_ilcopyb$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)chemm3m_iucopyr$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)chemm3m_ilcopyr$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)chemm3m_iucopyi$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)chemm3m_ilcopyi$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(CGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -UDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zhemm3m_oucopyb$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)zhemm3m_olcopyb$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)zhemm3m_oucopyr$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zhemm3m_olcopyr$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zhemm3m_oucopyi$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zhemm3m_olcopyi$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(ZGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zhemm3m_iucopyb$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)zhemm3m_ilcopyb$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)zhemm3m_iucopyr$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zhemm3m_ilcopyr$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)zhemm3m_iucopyi$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)zhemm3m_ilcopyi$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(ZGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xhemm3m_oucopyb$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)xhemm3m_olcopyb$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA $< -o $@
 
 $(KDIR)xhemm3m_oucopyr$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xhemm3m_olcopyr$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xhemm3m_oucopyi$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xhemm3m_olcopyi$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(XGEMM3M_UNROLL_N).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -DUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xhemm3m_iucopyb$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)xhemm3m_ilcopyb$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA $< -o $@
 
 $(KDIR)xhemm3m_iucopyr$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xhemm3m_ilcopyr$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DREAL_ONLY $< -o $@
 
 $(KDIR)xhemm3m_iucopyi$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_ucopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)xhemm3m_ilcopyi$(TSUFFIX).$(PSUFFIX) : generic/zhemm3m_lcopy_$(XGEMM3M_UNROLL_M).c
-	$(CC) $(PFLAGS) $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) $$PFLAGS $(NO_UNINITIALIZED_WARN) -c -DXDOUBLE -DCOMPLEX -UUSE_ALPHA -DIMAGE_ONLY $< -o $@
 
 $(KDIR)strsm_iunucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_uncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_iunncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_uncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strsm_ilnucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_lncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_ilnncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_lncopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)strsm_iutucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_utcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_iutncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_utcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strsm_iltucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_ltcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_iltncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_ltcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)strsm_ounucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_uncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_ounncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_uncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strsm_olnucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_lncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_olnncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_lncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)strsm_outucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_utcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_outncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_utcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)strsm_oltucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_ltcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)strsm_oltncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_ltcopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_iunucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_uncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_iunncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_uncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_ilnucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_lncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_ilnncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_lncopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_iutucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_utcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_iutncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_utcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_iltucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_ltcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_iltncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_ltcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_ounucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_uncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_ounncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_uncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_olnucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_lncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_olnncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_lncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_outucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_utcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_outncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_utcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)dtrsm_oltucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_ltcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)dtrsm_oltncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_ltcopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_iunucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_uncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_iunncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_uncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_ilnucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_lncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_ilnncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_lncopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_iutucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_utcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_iutncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_utcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_iltucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_ltcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_iltncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_ltcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_ounucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_uncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_ounncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_uncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_olnucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_lncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_olnncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_lncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_outucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_utcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_outncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_utcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)qtrsm_oltucopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_ltcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)qtrsm_oltncopy$(TSUFFIX).$(PSUFFIX) : generic/trsm_ltcopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -UCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_iunucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_uncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_iunncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_uncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_ilnucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_lncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_ilnncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_lncopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_iutucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_utcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_iutncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_utcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_iltucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_ltcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_iltncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_ltcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_ounucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_uncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_ounncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_uncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_olnucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_lncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_olnncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_lncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_outucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_utcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_outncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_utcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ctrsm_oltucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_ltcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ctrsm_oltncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_ltcopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -UDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_iunucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_uncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_iunncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_uncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_ilnucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_lncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_ilnncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_lncopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_iutucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_utcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_iutncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_utcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_iltucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_ltcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_iltncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_ltcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_ounucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_uncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_ounncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_uncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_olnucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_lncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_olnncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_lncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_outucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_utcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_outncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_utcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)ztrsm_oltucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_ltcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)ztrsm_oltncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_ltcopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_iunucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_uncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_iunncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_uncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_ilnucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_lncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_ilnncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_lncopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_iutucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_utcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_iutncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_utcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_iltucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_ltcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_iltncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_ltcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -UOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_ounucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_uncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_ounncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_uncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_olnucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_lncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_olnncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_lncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_outucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_utcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_outncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_utcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -ULOWER -UUNIT $< -o $@
 
 $(KDIR)xtrsm_oltucopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_ltcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -DUNIT $< -o $@
 
 $(KDIR)xtrsm_oltncopy$(TSUFFIX).$(PSUFFIX) : generic/ztrsm_ltcopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(PFLAGS) $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
+	[[ "$<" == *.S ]] && PFLAGS="$(PFLAGS) $(ASMFLAGS)" || PFLAGS="$(PFLAGS)"; \
+	$(CC) -c $$PFLAGS $(NO_UNINITIALIZED_WARN) -DXDOUBLE -DCOMPLEX -DOUTER -DLOWER -UUNIT $< -o $@
 
 
 #####  BLAS extensions ######
@@ -3824,112 +4870,128 @@ DOMATCOPY_CN = ../arm/omatcopy_cn.c
 endif
 
 $(KDIR)domatcopy_k_cn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DOMATCOPY_CN)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX -UROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX -UROWM $< -o $@
 
 ifndef DOMATCOPY_RN
 DOMATCOPY_RN = ../arm/omatcopy_rn.c
 endif
 
 $(KDIR)domatcopy_k_rn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DOMATCOPY_RN)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX -DROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX -DROWM $< -o $@
 
 ifndef DOMATCOPY_CT
 DOMATCOPY_CT = ../arm/omatcopy_ct.c
 endif
 
 $(KDIR)domatcopy_k_ct$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DOMATCOPY_CT)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX -UROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX -UROWM $< -o $@
 
 ifndef DOMATCOPY_RT
 DOMATCOPY_RT = ../arm/omatcopy_rt.c
 endif
 
 $(KDIR)domatcopy_k_rt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DOMATCOPY_RT)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX -DROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX -DROWM $< -o $@
 
 ifndef DIMATCOPY_CN
 DIMATCOPY_CN = ../generic/imatcopy_cn.c
 endif
 
 $(KDIR)dimatcopy_k_cn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DIMATCOPY_CN)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX -UROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX -UROWM $< -o $@
 
 ifndef DIMATCOPY_RN
 DIMATCOPY_RN = ../generic/imatcopy_rn.c
 endif
 
 $(KDIR)dimatcopy_k_rn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DIMATCOPY_RN)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX -DROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX -DROWM $< -o $@
 
 ifndef DIMATCOPY_CT
 DIMATCOPY_CT = ../generic/imatcopy_ct.c
 endif
 
 $(KDIR)dimatcopy_k_ct$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DIMATCOPY_CT)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX -UROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX -UROWM $< -o $@
 
 ifndef DIMATCOPY_RT
 DIMATCOPY_RT = ../generic/imatcopy_rt.c
 endif
 
 $(KDIR)dimatcopy_k_rt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DIMATCOPY_RT)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX -DROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX -DROWM $< -o $@
 
 ifndef SOMATCOPY_CN
 SOMATCOPY_CN = ../arm/omatcopy_cn.c
 endif
 
 $(KDIR)somatcopy_k_cn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SOMATCOPY_CN)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX -UROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX -UROWM $< -o $@
 
 ifndef SOMATCOPY_RN
 SOMATCOPY_RN = ../arm/omatcopy_rn.c
 endif
 
 $(KDIR)somatcopy_k_rn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SOMATCOPY_RN)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX -DROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX -DROWM $< -o $@
 
 ifndef SOMATCOPY_CT
 SOMATCOPY_CT = ../arm/omatcopy_ct.c
 endif
 
 $(KDIR)somatcopy_k_ct$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SOMATCOPY_CT)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX -UROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX -UROWM $< -o $@
 
 ifndef SOMATCOPY_RT
 SOMATCOPY_RT = ../arm/omatcopy_rt.c
 endif
 
 $(KDIR)somatcopy_k_rt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SOMATCOPY_RT)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX -DROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX -DROWM $< -o $@
 
 ifndef SIMATCOPY_CN
 SIMATCOPY_CN = ../generic/imatcopy_cn.c
 endif
 
 $(KDIR)simatcopy_k_cn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SIMATCOPY_CN)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX -UROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX -UROWM $< -o $@
 
 ifndef SIMATCOPY_RN
 SIMATCOPY_RN = ../generic/imatcopy_rn.c
 endif
 
 $(KDIR)simatcopy_k_rn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SIMATCOPY_RN)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX -DROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX -DROWM $< -o $@
 
 ifndef SIMATCOPY_CT
 SIMATCOPY_CT = ../generic/imatcopy_ct.c
 endif
 
 $(KDIR)simatcopy_k_ct$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SIMATCOPY_CT)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX -UROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX -UROWM $< -o $@
 
 ifndef SIMATCOPY_RT
 SIMATCOPY_RT = ../generic/imatcopy_rt.c
 endif
 
 $(KDIR)simatcopy_k_rt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SIMATCOPY_RT)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX -DROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX -DROWM $< -o $@
 
 
 ifndef COMATCOPY_CN
@@ -3937,112 +4999,128 @@ COMATCOPY_CN = ../arm/zomatcopy_cn.c
 endif
 
 $(KDIR)comatcopy_k_cn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(COMATCOPY_CN)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
 
 ifndef COMATCOPY_RN
 COMATCOPY_RN = ../arm/zomatcopy_rn.c
 endif
 
 $(KDIR)comatcopy_k_rn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(COMATCOPY_RN)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
 
 ifndef COMATCOPY_CT
 COMATCOPY_CT = ../arm/zomatcopy_ct.c
 endif
 
 $(KDIR)comatcopy_k_ct$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(COMATCOPY_CT)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
 
 ifndef COMATCOPY_RT
 COMATCOPY_RT = ../arm/zomatcopy_rt.c
 endif
 
 $(KDIR)comatcopy_k_rt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(COMATCOPY_RT)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
 
 ifndef COMATCOPY_CNC
 COMATCOPY_CNC = ../arm/zomatcopy_cnc.c
 endif
 
 $(KDIR)comatcopy_k_cnc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(COMATCOPY_CNC)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
 
 ifndef COMATCOPY_RNC
 COMATCOPY_RNC = ../arm/zomatcopy_rnc.c
 endif
 
 $(KDIR)comatcopy_k_rnc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(COMATCOPY_RNC)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
 
 ifndef COMATCOPY_CTC
 COMATCOPY_CTC = ../arm/zomatcopy_ctc.c
 endif
 
 $(KDIR)comatcopy_k_ctc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(COMATCOPY_CTC)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
 
 ifndef COMATCOPY_RTC
 COMATCOPY_RTC = ../arm/zomatcopy_rtc.c
 endif
 
 $(KDIR)comatcopy_k_rtc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(COMATCOPY_RTC)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
 
 ifndef CIMATCOPY_CN
 CIMATCOPY_CN = ../generic/zimatcopy_cn.c
 endif
 
 $(KDIR)cimatcopy_k_cn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CIMATCOPY_CN)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
 
 ifndef CIMATCOPY_RN
 CIMATCOPY_RN = ../generic/zimatcopy_rn.c
 endif
 
 $(KDIR)cimatcopy_k_rn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CIMATCOPY_RN)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
 
 ifndef CIMATCOPY_CT
 CIMATCOPY_CT = ../generic/zimatcopy_ct.c
 endif
 
 $(KDIR)cimatcopy_k_ct$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CIMATCOPY_CT)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
 
 ifndef CIMATCOPY_RT
 CIMATCOPY_RT = ../generic/zimatcopy_rt.c
 endif
 
 $(KDIR)cimatcopy_k_rt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CIMATCOPY_RT)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
 
 ifndef CIMATCOPY_CNC
 CIMATCOPY_CNC = ../generic/zimatcopy_cnc.c
 endif
 
 $(KDIR)cimatcopy_k_cnc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CIMATCOPY_CNC)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
 
 ifndef CIMATCOPY_RNC
 CIMATCOPY_RNC = ../generic/zimatcopy_rnc.c
 endif
 
 $(KDIR)cimatcopy_k_rnc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CIMATCOPY_RNC)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
 
 ifndef CIMATCOPY_CTC
 CIMATCOPY_CTC = ../generic/zimatcopy_ctc.c
 endif
 
 $(KDIR)cimatcopy_k_ctc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CIMATCOPY_CTC)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
 
 ifndef CIMATCOPY_RTC
 CIMATCOPY_RTC = ../generic/zimatcopy_rtc.c
 endif
 
 $(KDIR)cimatcopy_k_rtc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CIMATCOPY_RTC)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
 
 
 
@@ -4051,112 +5129,128 @@ ZOMATCOPY_CN = ../arm/zomatcopy_cn.c
 endif
 
 $(KDIR)zomatcopy_k_cn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZOMATCOPY_CN)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
 
 ifndef ZOMATCOPY_RN
 ZOMATCOPY_RN = ../arm/zomatcopy_rn.c
 endif
 
 $(KDIR)zomatcopy_k_rn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZOMATCOPY_RN)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
 
 ifndef ZOMATCOPY_CT
 ZOMATCOPY_CT = ../arm/zomatcopy_ct.c
 endif
 
 $(KDIR)zomatcopy_k_ct$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZOMATCOPY_CT)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
 
 ifndef ZOMATCOPY_RT
 ZOMATCOPY_RT = ../arm/zomatcopy_rt.c
 endif
 
 $(KDIR)zomatcopy_k_rt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZOMATCOPY_RT)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
 
 ifndef ZOMATCOPY_CNC
 ZOMATCOPY_CNC = ../arm/zomatcopy_cnc.c
 endif
 
 $(KDIR)zomatcopy_k_cnc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZOMATCOPY_CNC)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
 
 ifndef ZOMATCOPY_RNC
 ZOMATCOPY_RNC = ../arm/zomatcopy_rnc.c
 endif
 
 $(KDIR)zomatcopy_k_rnc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZOMATCOPY_RNC)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
 
 ifndef ZOMATCOPY_CTC
 ZOMATCOPY_CTC = ../arm/zomatcopy_ctc.c
 endif
 
 $(KDIR)zomatcopy_k_ctc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZOMATCOPY_CTC)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
 
 ifndef ZOMATCOPY_RTC
 ZOMATCOPY_RTC = ../arm/zomatcopy_rtc.c
 endif
 
 $(KDIR)zomatcopy_k_rtc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZOMATCOPY_RTC)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
 
 ifndef ZIMATCOPY_CN
 ZIMATCOPY_CN = ../generic/zimatcopy_cn.c
 endif
 
 $(KDIR)zimatcopy_k_cn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZIMATCOPY_CN)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
 
 ifndef ZIMATCOPY_RN
 ZIMATCOPY_RN = ../generic/zimatcopy_rn.c
 endif
 
 $(KDIR)zimatcopy_k_rn$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZIMATCOPY_RN)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
 
 ifndef ZIMATCOPY_CT
 ZIMATCOPY_CT = ../generic/zimatcopy_ct.c
 endif
 
 $(KDIR)zimatcopy_k_ct$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZIMATCOPY_CT)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -UROWM -UCONJ $< -o $@
 
 ifndef ZIMATCOPY_RT
 ZIMATCOPY_RT = ../generic/zimatcopy_rt.c
 endif
 
 $(KDIR)zimatcopy_k_rt$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZIMATCOPY_RT)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DROWM -UCONJ $< -o $@
 
 ifndef ZIMATCOPY_CNC
 ZIMATCOPY_CNC = ../generic/zimatcopy_cnc.c
 endif
 
 $(KDIR)zimatcopy_k_cnc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZIMATCOPY_CNC)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
 
 ifndef ZIMATCOPY_RNC
 ZIMATCOPY_RNC = ../generic/zimatcopy_rnc.c
 endif
 
 $(KDIR)zimatcopy_k_rnc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZIMATCOPY_RNC)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
 
 ifndef ZIMATCOPY_CTC
 ZIMATCOPY_CTC = ../generic/zimatcopy_ctc.c
 endif
 
 $(KDIR)zimatcopy_k_ctc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZIMATCOPY_CTC)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -UROWM -DCONJ $< -o $@
 
 ifndef ZIMATCOPY_RTC
 ZIMATCOPY_RTC = ../generic/zimatcopy_rtc.c
 endif
 
 $(KDIR)zimatcopy_k_rtc$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZIMATCOPY_RTC)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -DROWM -DCONJ $< -o $@
 
 
 ifndef SGEADD_K
@@ -4164,26 +5258,30 @@ SGEADD_K = ../generic/geadd.c
 endif
 
 $(KDIR)sgeadd_k$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEADD_K)
-	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX -UROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -UCOMPLEX -UROWM $< -o $@
 
 ifndef DGEADD_K
 DGEADD_K = ../generic/geadd.c
 endif
 
 $(KDIR)dgeadd_k$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEADD_K)
-	$(CC) $(CFLAGS) -c -DDOUBLE -UCOMPLEX -UROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -UCOMPLEX -UROWM $< -o $@
 
 ifndef CGEADD_K
 CGEADD_K = ../generic/zgeadd.c
 endif
 
 $(KDIR)cgeadd_k$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(CGEADD_K)
-	$(CC) $(CFLAGS) -c -UDOUBLE -DCOMPLEX -UROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -UDOUBLE -DCOMPLEX -UROWM $< -o $@
 
 ifndef ZGEADD_K
 ZGEADD_K = ../generic/zgeadd.c
 endif
 
 $(KDIR)zgeadd_k$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEADD_K)
-	$(CC) $(CFLAGS) -c -DDOUBLE -DCOMPLEX -UROWM $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) $$CFLAGS -c -DDOUBLE -DCOMPLEX -UROWM $< -o $@
 
diff --git a/kernel/Makefile.LA b/kernel/Makefile.LA
index 88342718..d9c67390 100644
--- a/kernel/Makefile.LA
+++ b/kernel/Makefile.LA
@@ -14,38 +14,50 @@ XBLASOBJS	+= xneg_tcopy$(TSUFFIX).$(SUFFIX) xlaswp_ncopy$(TSUFFIX).$(SUFFIX)
 endif
 
 $(KDIR)sneg_tcopy$(TSUFFIX).$(SUFFIX) $(KDIR)sneg_tcopy$(TSUFFIX).$(PSUFFIX) : generic/neg_tcopy_$(SGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $< -o $@
 
 $(KDIR)dneg_tcopy$(TSUFFIX).$(SUFFIX) $(KDIR)dneg_tcopy$(TSUFFIX).$(PSUFFIX) : generic/neg_tcopy_$(DGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $< -o $@
 
 $(KDIR)qneg_tcopy$(TSUFFIX).$(SUFFIX) $(KDIR)qneg_tcopy$(TSUFFIX).$(PSUFFIX) : generic/neg_tcopy_$(QGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $< -o $@
 
 $(KDIR)cneg_tcopy$(TSUFFIX).$(SUFFIX) $(KDIR)cneg_tcopy$(TSUFFIX).$(PSUFFIX) : generic/zneg_tcopy_$(CGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $< -o $@
 
 $(KDIR)zneg_tcopy$(TSUFFIX).$(SUFFIX) $(KDIR)zneg_tcopy$(TSUFFIX).$(PSUFFIX) : generic/zneg_tcopy_$(ZGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $< -o $@
 
 $(KDIR)xneg_tcopy$(TSUFFIX).$(SUFFIX) $(KDIR)xneg_tcopy$(TSUFFIX).$(PSUFFIX) : generic/zneg_tcopy_$(XGEMM_UNROLL_M).c
-	$(CC) -c $(CFLAGS) $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $< -o $@
 
 $(KDIR)slaswp_ncopy$(TSUFFIX).$(SUFFIX) $(KDIR)slaswp_ncopy$(TSUFFIX).$(PSUFFIX) : generic/laswp_ncopy_$(SGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $< -o $@
 
 $(KDIR)dlaswp_ncopy$(TSUFFIX).$(SUFFIX) $(KDIR)dlaswp_ncopy$(TSUFFIX).$(PSUFFIX) : generic/laswp_ncopy_$(DGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $< -o $@
 
 $(KDIR)qlaswp_ncopy$(TSUFFIX).$(SUFFIX) $(KDIR)qlaswp_ncopy$(TSUFFIX).$(PSUFFIX) : generic/laswp_ncopy_$(QGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $< -o $@
 
 $(KDIR)claswp_ncopy$(TSUFFIX).$(SUFFIX) $(KDIR)claswp_ncopy$(TSUFFIX).$(PSUFFIX) : generic/zlaswp_ncopy_$(CGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $< -o $@
 
 $(KDIR)zlaswp_ncopy$(TSUFFIX).$(SUFFIX) $(KDIR)zlaswp_ncopy$(TSUFFIX).$(PSUFFIX) : generic/zlaswp_ncopy_$(ZGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $< -o $@
 
 $(KDIR)xlaswp_ncopy$(TSUFFIX).$(SUFFIX) $(KDIR)xlaswp_ncopy$(TSUFFIX).$(PSUFFIX) : generic/zlaswp_ncopy_$(XGEMM_UNROLL_N).c
-	$(CC) -c $(CFLAGS) $< -o $@
+	[[ "$<" == *.S ]] && CFLAGS="$(CFLAGS) $(ASMFLAGS)" || CFLAGS="$(CFLAGS)"; \
+	$(CC) -c $$CFLAGS $< -o $@