papi/SOURCES/papi-zen.patch

303 lines
14 KiB
Diff
Raw Normal View History

2022-06-28 11:00:00 +00:00
commit 20890adcb59a1c1648cb70be65332c03a3781e1a
Author: Anthony Castaldo <TonyCastaldo@icl.utk.edu>
Date: Thu Jan 16 16:43:51 2020 -0500
Added two machine types to papi_events.csv to be in line with
libpfm4 update to support amd64_fam17h_zen1 and zen2.
diff --git a/src/papi_events.csv b/src/papi_events.csv
index 97446ad2c..8e96adfbd 100644
--- a/src/papi_events.csv
+++ b/src/papi_events.csv
@@ -396,6 +396,8 @@ PRESET,PAPI_FSQ_INS,NOT_DERIVED,RETIRED_SSE_AVX_OPERATIONS:SINGLE_DIV_OPS:DOUBLE
#
#
CPU,amd64_fam17h
+CPU,amd64_fam17h_zen1
+CPU,amd64_fam17h_zen2
#
PRESET,PAPI_TOT_INS,NOT_DERIVED,RETIRED_INSTRUCTIONS
PRESET,PAPI_TOT_CYC,NOT_DERIVED,CYCLES_NOT_IN_HALT
commit ae449f73abd0849f05ab3e1f3a64bde0c670c645
Author: Anthony <adanalis@icl.utk.edu>
Date: Fri Jul 17 12:05:14 2020 -0400
Separated the cache preset events of AMD Zen1 and Zen2 and added some more.
diff --git a/src/papi_events.csv b/src/papi_events.csv
index 8e96adfbd..2325bd4dc 100644
--- a/src/papi_events.csv
+++ b/src/papi_events.csv
@@ -397,7 +397,6 @@ PRESET,PAPI_FSQ_INS,NOT_DERIVED,RETIRED_SSE_AVX_OPERATIONS:SINGLE_DIV_OPS:DOUBLE
#
CPU,amd64_fam17h
CPU,amd64_fam17h_zen1
-CPU,amd64_fam17h_zen2
#
PRESET,PAPI_TOT_INS,NOT_DERIVED,RETIRED_INSTRUCTIONS
PRESET,PAPI_TOT_CYC,NOT_DERIVED,CYCLES_NOT_IN_HALT
@@ -434,6 +433,27 @@ PRESET,PAPI_FML_INS,NOT_DERIVED,RETIRED_SSE_AVX_OPERATIONS:SP_MULT_FLOPS:DP_MULT
PRESET,PAPI_FAD_INS,NOT_DERIVED,RETIRED_SSE_AVX_OPERATIONS:SP_ADD_SUB_FLOPS:DP_ADD_SUB_FLOPS
PRESET,PAPI_FDV_INS,NOT_DERIVED,RETIRED_SSE_AVX_OPERATIONS:SP_DIV_FLOPS:DP_DIV_FLOPS,NOTE,"Counts both divide and square root instructions"
PRESET,PAPI_FSQ_INS,NOT_DERIVED,RETIRED_SSE_AVX_OPERATIONS:SP_DIV_FLOPS:DP_DIV_FLOPS,NOTE,"Counts both divide and square root instructions"
+# Events discovered via CAT
+PRESET,PAPI_L2_DCM,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:LS_RD_BLK_C
+PRESET,PAPI_L2_DCR,NOT_DERIVED,REQUESTS_TO_L2_GROUP1:RD_BLK_L
+PRESET,PAPI_L2_DCH,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:LS_RD_BLK_L_HIT_X
+#
+#
+CPU,amd64_fam17h_zen2
+# Events copied from zen1 that also exist on zen2
+PRESET,PAPI_TLB_DM,NOT_DERIVED,L1_DTLB_MISS:TLB_RELOAD_1G_L2_MISS:TLB_RELOAD_2M_L2_MISS:TLB_RELOAD_COALESCED_PAGE_MISS:TLB_RELOAD_4K_L2_MISS:TLB_RELOAD_1G_L2_HIT:TLB_RELOAD_2M_L2_HIT:TLB_RELOAD_COALESCED_PAGE_HIT:TLB_RELOAD_4K_L2_HIT
+PRESET,PAPI_TLB_IM,DERIVED_ADD,L1_ITLB_MISS_L2_ITLB_HIT,L1_ITLB_MISS_L2_ITLB_MISS:IF1G:IF2M:IF4K
+PRESET,PAPI_BR_TKN,NOT_DERIVED,RETIRED_TAKEN_BRANCH_INSTRUCTIONS
+PRESET,PAPI_BR_MSP,NOT_DERIVED,RETIRED_BRANCH_INSTRUCTIONS_MISPREDICTED
+PRESET,PAPI_TOT_INS,NOT_DERIVED,RETIRED_INSTRUCTIONS
+PRESET,PAPI_BR_INS,NOT_DERIVED,RETIRED_BRANCH_INSTRUCTIONS
+PRESET,PAPI_TOT_CYC,NOT_DERIVED,CYCLES_NOT_IN_HALT
+# Events discovered via CAT
+PRESET,PAPI_L1_DCA,NOT_DERIVED,perf::PERF_COUNT_HW_CACHE_L1D:ACCESS
+PRESET,PAPI_L2_DCM,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:LS_RD_BLK_C
+PRESET,PAPI_L2_DCR,NOT_DERIVED,REQUESTS_TO_L2_GROUP1:RD_BLK_L
+PRESET,PAPI_L2_DCH,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:LS_RD_BLK_L_HIT_X
+
#
#
CPU,Intel architectural PMU
commit ccc22b5dda46fea8933d99950c3e30b5298cdd1d
Author: Heike Jagode <jagode@icl.utk.edu>
Date: Thu Sep 24 13:33:38 2020 -0400
Added presets for floating-point operations (FP_OPS, DP_OPS, SP_OPS)
for AMD zen2.
PPR (under section 2.1.15.3. -- https://www.amd.com/system/files/TechDocs/54945_3.03_ppr_ZP_B2_pub.zip)
explains that FLOP events require MergeEvent support, which was included
in the 5.6 kernel.
===>>> Hence, a kernel version 5.6 or greater is required.
NOTE: without the MergeEvent support in the kernel,
there is no guarantee that the SSE/AVX FLOP
events produce any useful data whatsoever.
These events have been tested and verified for
scalar flops, SSE, AVX, and FMA:
(1) for one AVX instruction (e.g. _mm256_add_pd()),
the RETIRED_SSE_AVX_FLOPS:ADD_SUB_FLOPS event returns
a count of 4 (in the case of double precision), and
a count of 8 (in the case of single precision).
(2) for one AVX FMA instruction (e.g. _mm256_macc_pd()),
the RETIRED_SSE_AVX_FLOPS:MAC_FLOPS event returns
a count of 8 (in the case of double precision), and
a count of 16 (in the case of single precision).
(3) for one SSE instruction (e.g. _mm_mul_pd()),
the RETIRED_SSE_AVX_FLOPS:MULT_FLOPS event returns
a count of 2 (in the case of double precision), and
a count of 4 (in the case of single precision).
diff --git a/src/papi_events.csv b/src/papi_events.csv
index 2325bd4dc..2ff3e4d16 100644
--- a/src/papi_events.csv
+++ b/src/papi_events.csv
@@ -454,8 +454,19 @@ PRESET,PAPI_L2_DCM,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:LS_RD_
PRESET,PAPI_L2_DCR,NOT_DERIVED,REQUESTS_TO_L2_GROUP1:RD_BLK_L
PRESET,PAPI_L2_DCH,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:LS_RD_BLK_L_HIT_X
-#
-#
+# New FLOP event on zen2
+# PPR (under section 2.1.15.3. --
+# https://www.amd.com/system/files/TechDocs/54945_3.03_ppr_ZP_B2_pub.zip)
+# explains that FLOP events require MergeEvent support, which was included
+# in the 5.6 kernel.
+# Hence, a kernel version 5.6 or greater is required.
+# NOTE: without the MergeEvent support in the kernel, there is no guarantee
+# that this SSE/AVX FLOP event produces any useful data whatsoever.
+PRESET,PAPI_FP_OPS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
+PRESET,PAPI_DP_OPS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
+PRESET,PAPI_SP_OPS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
+
+
CPU,Intel architectural PMU
CPU,ix86arch
#
commit 35f93252a6e222299c03f2c94912334488e76b02
Author: Heike Jagode <jagode@icl.utk.edu>
Date: Thu Sep 24 18:40:59 2020 -0400
Added presets for floating-point instructions (FP_INS, VEC_DP, VEC_SP)
for AMD zen2.
For unoptimized code (like native MMM), these events may include
non-numeric floating-point instructions, e.g. MOVSD: move or merge
scalar double-precision floating-point value instructions.
Tested with:
1) SSE double: _mm_mul_pd / _mm_add_pd
2) SSE single: _mm_mul_ps / _mm_add_ps
3) AVX double: _mm256_mul_pd / _mm256_add_pd
4) AVX single: _mm256_mul_ps / _mm256_add_ps
5) FMA double: _mm256_macc_pd
6) FMA single: _mm256_macc_pd
diff --git a/src/papi_events.csv b/src/papi_events.csv
index 2ff3e4d16..60a64564d 100644
--- a/src/papi_events.csv
+++ b/src/papi_events.csv
@@ -465,6 +465,11 @@ PRESET,PAPI_L2_DCH,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:LS_RD_
PRESET,PAPI_FP_OPS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
PRESET,PAPI_DP_OPS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
PRESET,PAPI_SP_OPS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
+# Floating-point instructions (including non-numeric floating-point instructions,
+# e.g. Move or Merge Scalar Double-Precision Floating-Point values)
+PAPI_FP_INS,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
+PAPI_VEC_DP,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
+PAPI_VEC_SP,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
CPU,Intel architectural PMU
commit 344f6493425d865577508ff32b6f65516b1b4394
Author: Heike Jagode <jagode@icl.utk.edu>
Date: Thu Sep 24 19:03:31 2020 -0400
Added missing 'PRESET' to csv file.
diff --git a/src/papi_events.csv b/src/papi_events.csv
index 60a64564d..724d520f0 100644
--- a/src/papi_events.csv
+++ b/src/papi_events.csv
@@ -467,9 +467,9 @@ PRESET,PAPI_DP_OPS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
PRESET,PAPI_SP_OPS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
# Floating-point instructions (including non-numeric floating-point instructions,
# e.g. Move or Merge Scalar Double-Precision Floating-Point values)
-PAPI_FP_INS,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
-PAPI_VEC_DP,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
-PAPI_VEC_SP,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
+PRESET,PAPI_FP_INS,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
+PRESET,PAPI_VEC_DP,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
+PRESET,PAPI_VEC_SP,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
CPU,Intel architectural PMU
commit 4616aa717c5301a9a478876661eb8ac1f18c0333
Author: Heike Jagode <jagode@icl.utk.edu>
Date: Thu Oct 8 11:36:23 2020 -0400
For zen2, since FP_OPS counts both single- and double-prec operations
correctly, we don't need to confuse the user with additional
DP_OPS and SP_OPS events. So, I'm taking them out.
Same applies for events counting FP instructions.
diff --git a/src/papi_events.csv b/src/papi_events.csv
index 724d520f0..9ebf557e1 100644
--- a/src/papi_events.csv
+++ b/src/papi_events.csv
@@ -463,13 +463,20 @@ PRESET,PAPI_L2_DCH,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:LS_RD_
# NOTE: without the MergeEvent support in the kernel, there is no guarantee
# that this SSE/AVX FLOP event produces any useful data whatsoever.
PRESET,PAPI_FP_OPS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
-PRESET,PAPI_DP_OPS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
-PRESET,PAPI_SP_OPS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
+# Since FP_OPS counts both single- and double-prec operations
+# correctly, we don't need to confuse the user with additional
+# DP_OPS and SP_OPS events. So, I'm taking them out.
+#PRESET,PAPI_DP_OPS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
+#PRESET,PAPI_SP_OPS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
+#
# Floating-point instructions (including non-numeric floating-point instructions,
# e.g. Move or Merge Scalar Double-Precision Floating-Point values)
PRESET,PAPI_FP_INS,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
-PRESET,PAPI_VEC_DP,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
-PRESET,PAPI_VEC_SP,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
+# Since FP_INS counts both single- and double-prec instuctions
+# correctly, we don't need to confuse the user with additional
+# VEC_DP and VEC_SP events. So, I'm taking them out.
+#PRESET,PAPI_VEC_DP,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
+#PRESET,PAPI_VEC_SP,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
CPU,Intel architectural PMU
commit 274219e85ba8adcd2e9c78507adf7edb05b71daa
Author: Sebastian Mobo <smobo@vols.utk.edu>
Date: Thu Oct 8 13:40:21 2020 -0400
Added instruction-cache preset events for the Zen2.
Signed-off-by: Anthony <adanalis@icl.utk.edu>
diff --git a/src/papi_events.csv b/src/papi_events.csv
index 9ebf557e1..fd75f9371 100644
--- a/src/papi_events.csv
+++ b/src/papi_events.csv
@@ -453,7 +453,12 @@ PRESET,PAPI_L1_DCA,NOT_DERIVED,perf::PERF_COUNT_HW_CACHE_L1D:ACCESS
PRESET,PAPI_L2_DCM,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:LS_RD_BLK_C
PRESET,PAPI_L2_DCR,NOT_DERIVED,REQUESTS_TO_L2_GROUP1:RD_BLK_L
PRESET,PAPI_L2_DCH,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:LS_RD_BLK_L_HIT_X
-
+#
+PRESET,PAPI_L1_ICM,NOT_DERIVED,REQUESTS_TO_L2_GROUP1:CACHEABLE_IC_READ
+#
+PRESET,PAPI_L2_ICR,NOT_DERIVED,REQUESTS_TO_L2_GROUP1:CACHEABLE_IC_READ
+PRESET,PAPI_L2_ICM,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:IC_FILL_MISS
+PRESET,PAPI_L2_ICH,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:IC_FILL_HIT_X:IC_FILL_HIT_S
# New FLOP event on zen2
# PPR (under section 2.1.15.3. --
# https://www.amd.com/system/files/TechDocs/54945_3.03_ppr_ZP_B2_pub.zip)
commit 02f34baafb868d183f21bebfd3c46574847b9929
Author: Swarup Sahoo <swarup-chandra.sahoo@amd.com>
Date: Tue May 18 02:51:56 2021 +0530
Added AMD Zen3 preset events. Refer section 2.1.17.2 of PPR for AMD family 19h model 01h, https://www.amd.com/system/files/TechDocs/55898_pub.zip
Signed-off-by: Swarup Sahoo <swarup-chandra.sahoo@amd.com>
diff --git a/src/papi_events.csv b/src/papi_events.csv
index 4ef647959..d9e9da8a3 100644
--- a/src/papi_events.csv
+++ b/src/papi_events.csv
@@ -482,6 +482,33 @@ PRESET,PAPI_FP_INS,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X
# VEC_DP and VEC_SP events. So, I'm taking them out.
#PRESET,PAPI_VEC_DP,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
#PRESET,PAPI_VEC_SP,NOT_DERIVED,RETIRED_MMX_FP_INSTRUCTIONS:SSE_INSTR:MMX_INSTR:X87_INSTR
+#
+#
+CPU,amd64_fam19h_zen3
+PRESET,PAPI_TOT_INS,NOT_DERIVED,RETIRED_INSTRUCTIONS
+PRESET,PAPI_TOT_CYC,NOT_DERIVED,CYCLES_NOT_IN_HALT
+PRESET,PAPI_BR_INS,NOT_DERIVED,RETIRED_BRANCH_INSTRUCTIONS
+PRESET,PAPI_BR_TKN,NOT_DERIVED,RETIRED_TAKEN_BRANCH_INSTRUCTIONS
+PRESET,PAPI_BR_MSP,NOT_DERIVED,RETIRED_BRANCH_INSTRUCTIONS_MISPREDICTED
+PRESET,PAPI_TLB_DM,NOT_DERIVED, L1_DTLB_MISS:TLB_RELOAD_1G_L2_MISS:TLB_RELOAD_2M_L2_MISS:TLB_RELOAD_COALESCED_PAGE_MISS:TLB_RELOAD_4K_L2_MISS:TLB_RELOAD_1G_L2_HIT:TLB_RELOAD_2M_L2_HIT:TLB_RELOAD_COALESCED_PAGE_HIT:TLB_RELOAD_4K_L2_HIT
+PRESET,PAPI_TLB_IM,DERIVED_ADD,L1_ITLB_MISS_L2_ITLB_HIT,L1_ITLB_MISS_L2_ITLB_MISS:COALESCED4K:IF1G:IF2M:IF4K
+PRESET,PAPI_L1_DCA,NOT_DERIVED,LS_DISPATCH:LD_ST_DISPATCH:STORE_DISPATCH:LD_DISPATCH
+PRESET,PAPI_L1_DCM,NOT_DERIVED,REQUESTS_TO_L2_GROUP1:RD_BLK_L:RD_BLK_X:LS_RD_BLK_C_S:CHANGE_TO_X
+PRESET,PAPI_L2_DCM,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:LS_RD_BLK_C
+PRESET,PAPI_L2_DCR,NOT_DERIVED,REQUESTS_TO_L2_GROUP1:RD_BLK_L:RD_BLK_X:LS_RD_BLK_C_S:CHANGE_TO_X
+PRESET,PAPI_L2_DCH,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:LS_RD_BLK_C_S:LS_RD_BLK_L_HIT_X:LS_RD_BLK_L_HIT_S:LS_RD_BLK_X
+PRESET,PAPI_L2_ICR,NOT_DERIVED,REQUESTS_TO_L2_GROUP1:CACHEABLE_IC_READ
+PRESET,PAPI_L2_ICA,NOT_DERIVED,REQUESTS_TO_L2_GROUP1:CACHEABLE_IC_READ
+PRESET,PAPI_L2_ICM,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:IC_FILL_MISS
+PRESET,PAPI_L2_ICH,NOT_DERIVED,CORE_TO_L2_CACHEABLE_REQUEST_ACCESS_STATUS:IC_FILL_HIT_X:IC_FILL_HIT_S
+# RETIRED_SSE_AVX_FLOPS requires MergeEvent support.
+PRESET,PAPI_VEC_INS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
+PRESET,PAPI_FP_INS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
+PRESET,PAPI_FP_OPS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ANY
+PRESET,PAPI_FML_INS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:MULT_FLOPS
+PRESET,PAPI_FAD_INS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:ADD_SUB_FLOPS
+PRESET,PAPI_FDV_INS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:DIV_FLOPS
+PRESET,PAPI_FSQ_INS,NOT_DERIVED,RETIRED_SSE_AVX_FLOPS:DIV_FLOPS
CPU,Intel architectural PMU