rpms/llvm
7
0
Fork 0
Code Issues Pull Requests Releases Activity

Update to LLVM 19.1.1

Browse Source 
This is an import from Fedora rawhide with the following changes:
 * Replace with RHEL changelog
 * Enable bundle_compat_lib

Resolves: RHEL-57456
This commit is contained in:
Nikita Popov 2024-10-02 16:47:26 +02:00
parent f9583268af
commit 13ecd6f381
14 changed files with 2193 additions and 1211 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

41
0001-19-Remove-myst_parser-dependency-for-RHEL.patch Normal file
View File

@ -0,0 +1,41 @@
From d8742e9b361e5fd6fee2298b8ea0aeb4671ec05a Mon Sep 17 00:00:00 2001
From: Nikita Popov <npopov@redhat.com>
Date: Wed, 22 May 2024 09:39:26 +0200
Subject: [PATCH] Remove myst_parser dependency for RHEL
---
clang/docs/conf.py | 3 +--
llvm/docs/conf.py | 3 +--
2 files changed, 2 insertions(+), 4 deletions(-)
diff --git a/clang/docs/conf.py b/clang/docs/conf.py
index 4cee382a718f..d2e2198e05d4 100644
--- a/clang/docs/conf.py
+++ b/clang/docs/conf.py
@@ -43,8 +43,7 @@ try:
extensions.append("myst_parser")
except ImportError:
- if not tags.has("builder-man"):
- raise
+ pass
# The encoding of source files.
diff --git a/llvm/docs/conf.py b/llvm/docs/conf.py
index 7f2ed5309606..354a41f11280 100644
--- a/llvm/docs/conf.py
+++ b/llvm/docs/conf.py
@@ -36,8 +36,7 @@ try:
extensions.append("myst_parser")
except ImportError:
- if not tags.has("builder-man"):
- raise
+ pass
# Automatic anchors for markdown titles
from llvm_slug import make_slug
--
2.44.0

43
0001-20-Remove-myst_parser-dependency-for-RHEL.patch Normal file
View File

@ -0,0 +1,43 @@
From 50cd36c2156d375a6d50f661908b460fbbd22e78 Mon Sep 17 00:00:00 2001
From: Nikita Popov <npopov@redhat.com>
Date: Wed, 22 May 2024 09:39:26 +0200
Subject: [PATCH] Remove myst_parser dependency for RHEL
---
clang/docs/conf.py | 3 +--
llvm/docs/conf.py | 5 +----
2 files changed, 2 insertions(+), 6 deletions(-)
diff --git a/clang/docs/conf.py b/clang/docs/conf.py
index 4cee382a718f..d2e2198e05d4 100644
--- a/clang/docs/conf.py
+++ b/clang/docs/conf.py
@@ -43,8 +43,7 @@ try:
extensions.append("myst_parser")
except ImportError:
- if not tags.has("builder-man"):
- raise
+ pass
# The encoding of source files.
diff --git a/llvm/docs/conf.py b/llvm/docs/conf.py
index d9fa6961032b..e38c009a457d 100644
--- a/llvm/docs/conf.py
+++ b/llvm/docs/conf.py
@@ -36,10 +36,7 @@ try:
extensions.append("myst_parser")
except ImportError:
- if not tags.has("builder-man"):
- raise
-else:
- myst_enable_extensions = ["substitution"]
+ pass
# Automatic anchors for markdown titles
myst_heading_anchors = 6
--
2.46.0

30
0001-Always-build-shared-libs-for-LLD.patch Normal file
View File

@ -0,0 +1,30 @@
From 69faadbc396000bfa60c722f6fb9c0fc3fb2daf0 Mon Sep 17 00:00:00 2001
From: Nikita Popov <npopov@redhat.com>
Date: Wed, 8 May 2024 12:30:36 +0900
Subject: [PATCH] Always build shared libs for LLD
We don't want to enable BUILD_SHARED_LIBS for the whole build,
but we do want to build lld libraries.
---
lld/cmake/modules/AddLLD.cmake | 5 ++---
1 file changed, 2 insertions(+), 3 deletions(-)
diff --git a/lld/cmake/modules/AddLLD.cmake b/lld/cmake/modules/AddLLD.cmake
index 9f2684b6f933..743ec87814a2 100644
--- a/lld/cmake/modules/AddLLD.cmake
+++ b/lld/cmake/modules/AddLLD.cmake
@@ -7,9 +7,8 @@ macro(add_lld_library name)
""
""
${ARGN})
- if(ARG_SHARED)
- set(ARG_ENABLE_SHARED SHARED)
- endif()
+ # Always build shared libs for LLD.
+ set(ARG_ENABLE_SHARED SHARED)
llvm_add_library(${name} ${ARG_ENABLE_SHARED} ${ARG_UNPARSED_ARGUMENTS})
if (NOT LLVM_INSTALL_TOOLCHAIN_ONLY)
--
2.45.1

25
0001-Fix-page-size-constant-on-aarch64-and-ppc64le.patch Normal file
View File

@ -0,0 +1,25 @@
From 5f73befe5a0df82e455f4b1052e62f34009e98bb Mon Sep 17 00:00:00 2001
From: Tom Stellard <tstellar@redhat.com>
Date: Tue, 23 Apr 2024 15:08:34 -0700
Subject: [PATCH] Fix page size constant on aarch64 and ppc64le
---
compiler-rt/lib/cfi/cfi.cpp | 2 ++
1 file changed, 2 insertions(+)
diff --git a/compiler-rt/lib/cfi/cfi.cpp b/compiler-rt/lib/cfi/cfi.cpp
index ad1c91623514..e7e86e5807a8 100644
--- a/compiler-rt/lib/cfi/cfi.cpp
+++ b/compiler-rt/lib/cfi/cfi.cpp
@@ -53,6 +53,8 @@ namespace __cfi {
#if SANITIZER_LOONGARCH64
#define kCfiShadowLimitsStorageSize 16384 // 16KiB on loongarch64 per page
+#elif defined(__aarch64__) || defined(__powerpc64__)
+#define kCfiShadowLimitsStorageSize 65536 // 1 page
#else
#define kCfiShadowLimitsStorageSize 4096 // 1 page
#endif
--
2.40.1

45
0001-Fix-python3-clang.patch Normal file
View File

@ -0,0 +1,45 @@
From 1c8a88c870a00eea6c80109cc682e0276ff7888d Mon Sep 17 00:00:00 2001
From: Nikita Popov <npopov@redhat.com>
Date: Mon, 8 Jul 2024 12:32:57 +0200
Subject: [PATCH] Fix python3-clang
Drop confusing `cd ..` so we stay in the root of the LLVM sources
and the install command succeeds.
---
install.spec.inc | 6 ++----
1 file changed, 2 insertions(+), 4 deletions(-)
diff --git a/install.spec.inc b/install.spec.inc
index 0fc1424..5f4a43b 100644
--- a/install.spec.inc
+++ b/install.spec.inc
@@ -98,8 +98,6 @@ touch %{buildroot}%{_bindir}/llvm-config%{exec_suffix}
mkdir -p %{buildroot}%{pkg_datadir}/llvm/cmake
cp -Rv cmake/* %{buildroot}%{pkg_datadir}/llvm/cmake
-cd ..
-
#endregion
#region CLANG installation
@@ -134,7 +132,7 @@ mkdir -p %{buildroot}%{python3_sitelib}/clang/
# install: omitting directory 'bindings/python/clang/__pycache__'
# NOTE: this only happens if we include the gdb plugin of libomp.
# Remove the plugin with command and we're good: rm -rf %{buildroot}/%{_datarootdir}/gdb
-install -p -m644 clang/bindings/python/clang/* %{buildroot}%{python3_sitelib}/clang/ || true
+install -p -m644 clang/bindings/python/clang/* %{buildroot}%{python3_sitelib}/clang/
%py_byte_compile %{__python3} %{buildroot}%{python3_sitelib}/clang
# install scanbuild-py to python sitelib.
@@ -268,7 +266,7 @@ done
# https://docs.fedoraproject.org/en-US/packaging-guidelines/Alternatives/
touch %{buildroot}%{_bindir}/ld
-install -D -m 644 -t %{buildroot}%{_mandir}/man1/ %{src_tarball_dir}/lld/docs/ld.lld.1
+install -D -m 644 -t %{buildroot}%{_mandir}/man1/ lld/docs/ld.lld.1
%post -n %{pkg_name_lld}
%{_sbindir}/update-alternatives --install %{_bindir}/ld ld %{_bindir}/ld.lld 1
--
2.45.2

27
0001-PATCH-clang-Make-funwind-tables-the-default-on-all-a.patch Normal file
View File

@ -0,0 +1,27 @@
From 49f827b09db549de62dcaf8b90b3fcb3e08c0ee5 Mon Sep 17 00:00:00 2001
From: Serge Guelton <sguelton@redhat.com>
Date: Mon, 6 Mar 2023 12:37:48 +0100
Subject: [PATCH] Make -funwind-tables the default on all archs
---
clang/lib/Driver/ToolChains/Gnu.cpp | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/clang/lib/Driver/ToolChains/Gnu.cpp b/clang/lib/Driver/ToolChains/Gnu.cpp
index 24fbdcffc07b..8fed46b49515 100644
--- a/clang/lib/Driver/ToolChains/Gnu.cpp
+++ b/clang/lib/Driver/ToolChains/Gnu.cpp
@@ -2904,6 +2904,10 @@ Generic_GCC::getDefaultUnwindTableLevel(const ArgList &Args) const {
case llvm::Triple::riscv64:
case llvm::Triple::x86:
case llvm::Triple::x86_64:
+ // Enable -funwind-tables on all architectures supported by Fedora:
+ // rhbz#1655546
+ case llvm::Triple::systemz:
+ case llvm::Triple::arm:
return UnwindTableLevel::Asynchronous;
default:
return UnwindTableLevel::None;
--
2.39.1

74
0001-PEI-Don-t-zero-out-noreg-operands.patch
View File

@ -1,74 +0,0 @@
From 9d1f05a7b8537deb5f626cd1b7b26ef2678f4c8e Mon Sep 17 00:00:00 2001
From: Arthur Eubanks <aeubanks@google.com>
Date: Thu, 27 Jul 2023 13:27:58 -0700
Subject: [PATCH] [PEI] Don't zero out noreg operands
A tail call may have $noreg operands.
Fixes a crash.
Reviewed By: xgupta
Differential Revision: https://reviews.llvm.org/D156485
(cherry picked from commit f800c1f3b207e7bcdc8b4c7192928d9a078242a0)
---
llvm/lib/CodeGen/PrologEpilogInserter.cpp | 9 +++++++--
llvm/test/CodeGen/X86/zero-call-used-regs.ll | 14 ++++++++++++++
2 files changed, 21 insertions(+), 2 deletions(-)
diff --git a/llvm/lib/CodeGen/PrologEpilogInserter.cpp b/llvm/lib/CodeGen/PrologEpilogInserter.cpp
index e323aaaeefaf..49047719fdaa 100644
--- a/llvm/lib/CodeGen/PrologEpilogInserter.cpp
+++ b/llvm/lib/CodeGen/PrologEpilogInserter.cpp
@@ -1285,6 +1285,8 @@ void PEI::insertZeroCallUsedRegs(MachineFunction &MF) {
continue;
MCRegister Reg = MO.getReg();
+ if (!Reg)
+ continue;
// This picks up sibling registers (e.q. %al -> %ah).
for (MCRegUnit Unit : TRI.regunits(Reg))
@@ -1308,8 +1310,11 @@ void PEI::insertZeroCallUsedRegs(MachineFunction &MF) {
if (!MO.isReg())
continue;
- for (const MCPhysReg &Reg :
- TRI.sub_and_superregs_inclusive(MO.getReg()))
+ MCRegister Reg = MO.getReg();
+ if (!Reg)
+ continue;
+
+ for (const MCPhysReg Reg : TRI.sub_and_superregs_inclusive(Reg))
RegsToZero.reset(Reg);
}
}
diff --git a/llvm/test/CodeGen/X86/zero-call-used-regs.ll b/llvm/test/CodeGen/X86/zero-call-used-regs.ll
index 63d51c916bb9..97ad5ce9c8cb 100644
--- a/llvm/test/CodeGen/X86/zero-call-used-regs.ll
+++ b/llvm/test/CodeGen/X86/zero-call-used-regs.ll
@@ -241,6 +241,20 @@ entry:
ret i32 %x
}
+define dso_local void @tailcall(ptr %p) local_unnamed_addr #0 "zero-call-used-regs"="used-gpr" {
+; I386-LABEL: tailcall:
+; I386: # %bb.0:
+; I386-NEXT: movl {{[0-9]+}}(%esp), %eax
+; I386-NEXT: jmpl *(%eax) # TAILCALL
+;
+; X86-64-LABEL: tailcall:
+; X86-64: # %bb.0:
+; X86-64-NEXT: jmpq *(%rdi) # TAILCALL
+ %c = load ptr, ptr %p
+ tail call void %c()
+ ret void
+}
+
; Don't emit zeroing registers in "main" function.
define dso_local i32 @main() local_unnamed_addr #1 {
; I386-LABEL: main:
--
2.43.0

30
0001-Workaround-a-bug-in-ORC-on-ppc64le.patch Normal file
View File

@ -0,0 +1,30 @@
From a2449cee8c995b56f1892502aab3dfad3d6f3ca1 Mon Sep 17 00:00:00 2001
From: Tulio Magno Quites Machado Filho <tuliom@redhat.com>
Date: Fri, 8 Sep 2023 11:45:34 -0300
Subject: [PATCH] Workaround a bug in ORC on ppc64le
The Jit code appears to be returning the wrong printf symbol on ppc64le
after the transition of the default long double to IEEE 128-bit floating
point.
---
clang/unittests/Interpreter/InterpreterTest.cpp | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/clang/unittests/Interpreter/InterpreterTest.cpp b/clang/unittests/Interpreter/InterpreterTest.cpp
index abb8e6377aab..7b6697ebc6ed 100644
--- a/clang/unittests/Interpreter/InterpreterTest.cpp
+++ b/clang/unittests/Interpreter/InterpreterTest.cpp
@@ -243,7 +243,9 @@ TEST(IncrementalProcessing, FindMangledNameSymbol) {
EXPECT_FALSE(!Addr);
// FIXME: Re-enable when we investigate the way we handle dllimports on Win.
-#ifndef _WIN32
+ // FIXME: The printf symbol returned from the Jit may not be correct on
+ // ppc64le when the default long double is IEEE 128-bit fp.
+#if !defined _WIN32 && !(defined __PPC64__ && defined __LITTLE_ENDIAN__)
EXPECT_EQ((uintptr_t)&printf, Addr->getValue());
#endif // _WIN32
}
--
2.41.0

25
0003-PATCH-clang-Don-t-install-static-libraries.patch Normal file
View File

@ -0,0 +1,25 @@
From 88704fc2eabb9dd19a9c3eb81a9b3dc37d95651c Mon Sep 17 00:00:00 2001
From: Tom Stellard <tstellar@redhat.com>
Date: Fri, 31 Jan 2020 11:04:57 -0800
Subject: [PATCH][clang] Don't install static libraries
---
clang/cmake/modules/AddClang.cmake | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/clang/cmake/modules/AddClang.cmake b/clang/cmake/modules/AddClang.cmake
index 5752f4277444..0f52822d91f0 100644
--- a/clang/cmake/modules/AddClang.cmake
+++ b/clang/cmake/modules/AddClang.cmake
@@ -113,7 +113,7 @@ macro(add_clang_library name)
if(TARGET ${lib})
target_link_libraries(${lib} INTERFACE ${LLVM_COMMON_LIBS})
- if (NOT LLVM_INSTALL_TOOLCHAIN_ONLY OR ARG_INSTALL_WITH_TOOLCHAIN)
+ if (ARG_SHARED AND (NOT LLVM_INSTALL_TOOLCHAIN_ONLY OR ARG_INSTALL_WITH_TOOLCHAIN))
get_target_export_arg(${name} Clang export_to_clangtargets UMBRELLA clang-libraries)
install(TARGETS ${lib}
COMPONENT ${lib}
--
2.30.2

13
0101-Deactivate-markdown-doc.patch
View File

@ -1,13 +0,0 @@
diff --git a/llvm/docs/conf.py b/llvm/docs/conf.py
index cf8a75980b53..b208ad138e89 100644
--- a/llvm/docs/conf.py
+++ b/llvm/docs/conf.py
@@ -26,7 +26,7 @@ from datetime import date
# Add any Sphinx extension module names here, as strings. They can be extensions
# coming with Sphinx (named 'sphinx.ext.*') or your custom ones.
-extensions = ["myst_parser", "sphinx.ext.intersphinx", "sphinx.ext.todo"]
+extensions = ["sphinx.ext.intersphinx", "sphinx.ext.todo"]
# Automatic anchors for markdown titles
from llvm_slug import make_slug

893
99273.patch
View File

@ -1,893 +0,0 @@
From 91052169960477fbc39169c10f9fae3bec732510 Mon Sep 17 00:00:00 2001
From: Carl Ritson <carl.ritson@amd.com>
Date: Wed, 17 Jul 2024 15:07:42 +0900
Subject: [PATCH 1/3] [AMDGPU] Implement workaround for GFX11.5 export priority
On GFX11.5 shaders having completed exports need to execute/wait
at a lower priority than shaders still executing exports.
Add code to maintain normal priority of 2 for shaders that export
and drop to priority 0 after exports.
---
llvm/lib/Target/AMDGPU/AMDGPU.td | 15 +-
.../lib/Target/AMDGPU/GCNHazardRecognizer.cpp | 112 ++++++
llvm/lib/Target/AMDGPU/GCNHazardRecognizer.h | 1 +
llvm/lib/Target/AMDGPU/GCNSubtarget.h | 3 +
.../AMDGPU/required-export-priority.ll | 344 ++++++++++++++++++
.../AMDGPU/required-export-priority.mir | 293 +++++++++++++++
6 files changed, 765 insertions(+), 3 deletions(-)
create mode 100644 llvm/test/CodeGen/AMDGPU/required-export-priority.ll
create mode 100644 llvm/test/CodeGen/AMDGPU/required-export-priority.mir
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.td b/llvm/lib/Target/AMDGPU/AMDGPU.td
index dfc8eaea66f7b..14fcf6a210a78 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.td
@@ -947,6 +947,12 @@ def FeatureHasRestrictedSOffset : SubtargetFeature<"restricted-soffset",
"Has restricted SOffset (immediate not supported)."
>;
+def FeatureRequiredExportPriority : SubtargetFeature<"required-export-priority",
+ "HasRequiredExportPriority",
+ "true",
+ "Export priority must be explicitly manipulated on GFX11.5"
+>;
+
//===------------------------------------------------------------===//
// Subtarget Features (options and debugging)
//===------------------------------------------------------------===//
@@ -1597,14 +1603,16 @@ def FeatureISAVersion11_5_0 : FeatureSet<
!listconcat(FeatureISAVersion11_Common.Features,
[FeatureSALUFloatInsts,
FeatureDPPSrc1SGPR,
- FeatureVGPRSingleUseHintInsts])>;
+ FeatureVGPRSingleUseHintInsts,
+ FeatureRequiredExportPriority])>;
def FeatureISAVersion11_5_1 : FeatureSet<
!listconcat(FeatureISAVersion11_Common.Features,
[FeatureSALUFloatInsts,
FeatureDPPSrc1SGPR,
FeatureVGPRSingleUseHintInsts,
- FeatureGFX11FullVGPRs])>;
+ FeatureGFX11FullVGPRs,
+ FeatureRequiredExportPriority])>;
def FeatureISAVersion12 : FeatureSet<
[FeatureGFX12,
diff --git a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
index a402fc6d7e611..a8b171aa82840 100644
--- a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
+++ b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
@@ -14,6 +14,7 @@
#include "GCNSubtarget.h"
#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
#include "SIMachineFunctionInfo.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/ScheduleDAG.h"
#include "llvm/TargetParser/TargetParser.h"
@@ -1104,6 +1105,7 @@ void GCNHazardRecognizer::fixHazards(MachineInstr *MI) {
fixWMMAHazards(MI);
fixShift64HighRegBug(MI);
fixVALUMaskWriteHazard(MI);
+ fixRequiredExportPriority(MI);
}
bool GCNHazardRecognizer::fixVcmpxPermlaneHazards(MachineInstr *MI) {
@@ -2895,3 +2897,113 @@ bool GCNHazardRecognizer::fixVALUMaskWriteHazard(MachineInstr *MI) {
return true;
}
+
+static bool ensureEntrySetPrio(MachineFunction *MF, int Priority,
+ const SIInstrInfo &TII) {
+ MachineBasicBlock &EntryMBB = MF->front();
+ if (EntryMBB.begin() != EntryMBB.end()) {
+ auto &EntryMI = *EntryMBB.begin();
+ if (EntryMI.getOpcode() == AMDGPU::S_SETPRIO &&
+ EntryMI.getOperand(0).getImm() >= Priority)
+ return false;
+ }
+
+ BuildMI(EntryMBB, EntryMBB.begin(), DebugLoc(), TII.get(AMDGPU::S_SETPRIO))
+ .addImm(Priority);
+ return true;
+}
+
+bool GCNHazardRecognizer::fixRequiredExportPriority(MachineInstr *MI) {
+ if (!ST.hasRequiredExportPriority())
+ return false;
+
+ // Assume the following shader types will never have exports,
+ // and avoid adding or adjusting S_SETPRIO.
+ MachineBasicBlock *MBB = MI->getParent();
+ MachineFunction *MF = MBB->getParent();
+ auto CC = MF->getFunction().getCallingConv();
+ switch (CC) {
+ case CallingConv::AMDGPU_CS:
+ case CallingConv::AMDGPU_CS_Chain:
+ case CallingConv::AMDGPU_CS_ChainPreserve:
+ case CallingConv::AMDGPU_KERNEL:
+ return false;
+ default:
+ break;
+ }
+
+ const int MaxPriority = 3;
+ const int NormalPriority = 2;
+ const int PostExportPriority = 0;
+
+ auto It = MI->getIterator();
+ switch (MI->getOpcode()) {
+ case AMDGPU::S_ENDPGM:
+ case AMDGPU::S_ENDPGM_SAVED:
+ case AMDGPU::S_ENDPGM_ORDERED_PS_DONE:
+ case AMDGPU::SI_RETURN_TO_EPILOG:
+ // Ensure shader with calls raises priority at entry.
+ // This ensures correct priority if exports exist in callee.
+ if (MF->getFrameInfo().hasCalls())
+ return ensureEntrySetPrio(MF, NormalPriority, TII);
+ return false;
+ case AMDGPU::S_SETPRIO: {
+ // Raise minimum priority unless in workaround.
+ auto &PrioOp = MI->getOperand(0);
+ int Prio = PrioOp.getImm();
+ bool InWA = (Prio == PostExportPriority) &&
+ (It != MBB->begin() && TII.isEXP(*std::prev(It)));
+ if (InWA || Prio >= NormalPriority)
+ return false;
+ PrioOp.setImm(std::min(Prio + NormalPriority, MaxPriority));
+ return true;
+ }
+ default:
+ if (!TII.isEXP(*MI))
+ return false;
+ break;
+ }
+
+ // Check entry priority at each export (as there will only be a few).
+ // Note: amdgpu_gfx can only be a callee, so defer to caller setprio.
+ bool Changed = false;
+ if (CC != CallingConv::AMDGPU_Gfx)
+ Changed = ensureEntrySetPrio(MF, NormalPriority, TII);
+
+ auto NextMI = std::next(It);
+ bool EndOfShader = false;
+ if (NextMI != MBB->end()) {
+ // Only need WA at end of sequence of exports.
+ if (TII.isEXP(*NextMI))
+ return Changed;
+ // Assume appropriate S_SETPRIO after export means WA already applied.
+ if (NextMI->getOpcode() == AMDGPU::S_SETPRIO &&
+ NextMI->getOperand(0).getImm() == PostExportPriority)
+ return Changed;
+ EndOfShader = NextMI->getOpcode() == AMDGPU::S_ENDPGM;
+ }
+
+ const DebugLoc &DL = MI->getDebugLoc();
+
+ // Lower priority.
+ BuildMI(*MBB, NextMI, DL, TII.get(AMDGPU::S_SETPRIO))
+ .addImm(PostExportPriority);
+
+ if (!EndOfShader) {
+ // Wait for exports to complete.
+ BuildMI(*MBB, NextMI, DL, TII.get(AMDGPU::S_WAITCNT_EXPCNT))
+ .addReg(AMDGPU::SGPR_NULL)
+ .addImm(0);
+ }
+
+ BuildMI(*MBB, NextMI, DL, TII.get(AMDGPU::S_NOP)).addImm(0);
+ BuildMI(*MBB, NextMI, DL, TII.get(AMDGPU::S_NOP)).addImm(0);
+
+ if (!EndOfShader) {
+ // Return to normal (higher) priority.
+ BuildMI(*MBB, NextMI, DL, TII.get(AMDGPU::S_SETPRIO))
+ .addImm(NormalPriority);
+ }
+
+ return true;
+}
diff --git a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.h b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.h
index 3ccca527c626b..f2a64ab48e180 100644
--- a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.h
+++ b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.h
@@ -107,6 +107,7 @@ class GCNHazardRecognizer final : public ScheduleHazardRecognizer {
bool fixWMMAHazards(MachineInstr *MI);
bool fixShift64HighRegBug(MachineInstr *MI);
bool fixVALUMaskWriteHazard(MachineInstr *MI);
+ bool fixRequiredExportPriority(MachineInstr *MI);
int checkMAIHazards(MachineInstr *MI);
int checkMAIHazards908(MachineInstr *MI);
diff --git a/llvm/lib/Target/AMDGPU/GCNSubtarget.h b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
index e5817594a4521..def89c785b855 100644
--- a/llvm/lib/Target/AMDGPU/GCNSubtarget.h
+++ b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
@@ -238,6 +238,7 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
bool HasVOPDInsts = false;
bool HasVALUTransUseHazard = false;
bool HasForceStoreSC0SC1 = false;
+ bool HasRequiredExportPriority = false;
// Dummy feature to use for assembler in tablegen.
bool FeatureDisable = false;
@@ -1282,6 +1283,8 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
bool hasRestrictedSOffset() const { return HasRestrictedSOffset; }
+ bool hasRequiredExportPriority() const { return HasRequiredExportPriority; }
+
/// \returns true if the target uses LOADcnt/SAMPLEcnt/BVHcnt, DScnt/KMcnt
/// and STOREcnt rather than VMcnt, LGKMcnt and VScnt respectively.
bool hasExtendedWaitCounts() const { return getGeneration() >= GFX12; }
diff --git a/llvm/test/CodeGen/AMDGPU/required-export-priority.ll b/llvm/test/CodeGen/AMDGPU/required-export-priority.ll
new file mode 100644
index 0000000000000..377902f3f0d1a
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/required-export-priority.ll
@@ -0,0 +1,344 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1150 -amdgpu-enable-vopd=0 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+
+define amdgpu_ps void @test_export_zeroes_f32() #0 {
+; GCN-LABEL: test_export_zeroes_f32:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: exp mrt0 off, off, off, off
+; GCN-NEXT: exp mrt0 off, off, off, off done
+; GCN-NEXT: s_setprio 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_endpgm
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 0, float 0.0, float 0.0, float 0.0, float 0.0, i1 false, i1 false)
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 0, float 0.0, float 0.0, float 0.0, float 0.0, i1 true, i1 false)
+ ret void
+}
+
+define amdgpu_ps void @test_export_en_src0_f32() #0 {
+; GCN-LABEL: test_export_en_src0_f32:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: v_mov_b32_e32 v0, 4.0
+; GCN-NEXT: v_mov_b32_e32 v1, 0.5
+; GCN-NEXT: v_mov_b32_e32 v2, 2.0
+; GCN-NEXT: v_mov_b32_e32 v3, 1.0
+; GCN-NEXT: exp mrt0 v3, off, off, off done
+; GCN-NEXT: s_setprio 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_endpgm
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 1, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+ ret void
+}
+
+define amdgpu_gs void @test_export_gs() #0 {
+; GCN-LABEL: test_export_gs:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: v_mov_b32_e32 v0, 4.0
+; GCN-NEXT: v_mov_b32_e32 v1, 0.5
+; GCN-NEXT: v_mov_b32_e32 v2, 2.0
+; GCN-NEXT: v_mov_b32_e32 v3, 1.0
+; GCN-NEXT: exp mrt0 off, v2, off, off done
+; GCN-NEXT: s_setprio 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_endpgm
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+ ret void
+}
+
+define amdgpu_hs void @test_export_hs() #0 {
+; GCN-LABEL: test_export_hs:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: v_mov_b32_e32 v0, 4.0
+; GCN-NEXT: v_mov_b32_e32 v1, 0.5
+; GCN-NEXT: v_mov_b32_e32 v2, 2.0
+; GCN-NEXT: v_mov_b32_e32 v3, 1.0
+; GCN-NEXT: exp mrt0 off, v2, off, off done
+; GCN-NEXT: s_setprio 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_endpgm
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+ ret void
+}
+
+define amdgpu_gfx void @test_export_gfx(float %v) #0 {
+; GCN-LABEL: test_export_gfx:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: v_mov_b32_e32 v1, 4.0
+; GCN-NEXT: v_mov_b32_e32 v2, 0.5
+; GCN-NEXT: v_mov_b32_e32 v3, 2.0
+; GCN-NEXT: exp mrt0 off, v3, off, off done
+; GCN-NEXT: s_setprio 0
+; GCN-NEXT: s_waitcnt_expcnt null, 0x0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: s_waitcnt expcnt(0)
+; GCN-NEXT: s_setpc_b64 s[30:31]
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float %v, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+ ret void
+}
+
+define amdgpu_cs void @test_export_cs() #0 {
+; GCN-LABEL: test_export_cs:
+; GCN: ; %bb.0:
+; GCN-NEXT: v_mov_b32_e32 v0, 4.0
+; GCN-NEXT: v_mov_b32_e32 v1, 0.5
+; GCN-NEXT: v_mov_b32_e32 v2, 2.0
+; GCN-NEXT: v_mov_b32_e32 v3, 1.0
+; GCN-NEXT: exp mrt0 off, v2, off, off done
+; GCN-NEXT: s_endpgm
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+ ret void
+}
+
+define amdgpu_kernel void @test_export_kernel() #0 {
+; GCN-LABEL: test_export_kernel:
+; GCN: ; %bb.0:
+; GCN-NEXT: v_mov_b32_e32 v0, 4.0
+; GCN-NEXT: v_mov_b32_e32 v1, 0.5
+; GCN-NEXT: v_mov_b32_e32 v2, 2.0
+; GCN-NEXT: v_mov_b32_e32 v3, 1.0
+; GCN-NEXT: exp mrt0 off, v2, off, off done
+; GCN-NEXT: s_endpgm
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
+ ret void
+}
+
+define amdgpu_gfx void @test_no_export_gfx(float %v) #0 {
+; GCN-LABEL: test_no_export_gfx:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_setpc_b64 s[30:31]
+ ret void
+}
+
+define amdgpu_ps void @test_no_export_ps(float %v) #0 {
+; GCN-LABEL: test_no_export_ps:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_endpgm
+ ret void
+}
+
+define amdgpu_ps void @test_if_export_f32(i32 %flag, float %x, float %y, float %z, float %w) #0 {
+; GCN-LABEL: test_if_export_f32:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: s_mov_b32 s0, exec_lo
+; GCN-NEXT: v_cmpx_ne_u32_e32 0, v0
+; GCN-NEXT: s_cbranch_execz .LBB9_2
+; GCN-NEXT: ; %bb.1: ; %exp
+; GCN-NEXT: exp mrt0 v1, v2, v3, v4
+; GCN-NEXT: s_setprio 0
+; GCN-NEXT: s_waitcnt_expcnt null, 0x0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: .LBB9_2: ; %end
+; GCN-NEXT: s_endpgm
+ %cc = icmp eq i32 %flag, 0
+ br i1 %cc, label %end, label %exp
+
+exp:
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %x, float %y, float %z, float %w, i1 false, i1 false)
+ br label %end
+
+end:
+ ret void
+}
+
+define amdgpu_ps void @test_if_export_vm_f32(i32 %flag, float %x, float %y, float %z, float %w) #0 {
+; GCN-LABEL: test_if_export_vm_f32:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: s_mov_b32 s0, exec_lo
+; GCN-NEXT: v_cmpx_ne_u32_e32 0, v0
+; GCN-NEXT: s_cbranch_execz .LBB10_2
+; GCN-NEXT: ; %bb.1: ; %exp
+; GCN-NEXT: exp mrt0 v1, v2, v3, v4
+; GCN-NEXT: s_setprio 0
+; GCN-NEXT: s_waitcnt_expcnt null, 0x0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: .LBB10_2: ; %end
+; GCN-NEXT: s_endpgm
+ %cc = icmp eq i32 %flag, 0
+ br i1 %cc, label %end, label %exp
+
+exp:
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %x, float %y, float %z, float %w, i1 false, i1 true)
+ br label %end
+
+end:
+ ret void
+}
+
+define amdgpu_ps void @test_if_export_done_f32(i32 %flag, float %x, float %y, float %z, float %w) #0 {
+; GCN-LABEL: test_if_export_done_f32:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: s_mov_b32 s0, exec_lo
+; GCN-NEXT: v_cmpx_ne_u32_e32 0, v0
+; GCN-NEXT: s_cbranch_execz .LBB11_2
+; GCN-NEXT: ; %bb.1: ; %exp
+; GCN-NEXT: exp mrt0 v1, v2, v3, v4 done
+; GCN-NEXT: s_setprio 0
+; GCN-NEXT: s_waitcnt_expcnt null, 0x0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: .LBB11_2: ; %end
+; GCN-NEXT: s_endpgm
+ %cc = icmp eq i32 %flag, 0
+ br i1 %cc, label %end, label %exp
+
+exp:
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %x, float %y, float %z, float %w, i1 true, i1 false)
+ br label %end
+
+end:
+ ret void
+}
+
+define amdgpu_ps void @test_if_export_vm_done_f32(i32 %flag, float %x, float %y, float %z, float %w) #0 {
+; GCN-LABEL: test_if_export_vm_done_f32:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: s_mov_b32 s0, exec_lo
+; GCN-NEXT: v_cmpx_ne_u32_e32 0, v0
+; GCN-NEXT: s_cbranch_execz .LBB12_2
+; GCN-NEXT: ; %bb.1: ; %exp
+; GCN-NEXT: exp mrt0 v1, v2, v3, v4 done
+; GCN-NEXT: s_setprio 0
+; GCN-NEXT: s_waitcnt_expcnt null, 0x0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: .LBB12_2: ; %end
+; GCN-NEXT: s_endpgm
+ %cc = icmp eq i32 %flag, 0
+ br i1 %cc, label %end, label %exp
+
+exp:
+ call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %x, float %y, float %z, float %w, i1 true, i1 true)
+ br label %end
+
+end:
+ ret void
+}
+
+define amdgpu_ps void @test_export_pos_before_param_across_load(i32 %idx) #0 {
+; GCN-LABEL: test_export_pos_before_param_across_load:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: buffer_load_b32 v0, v0, s[0:3], 0 offen
+; GCN-NEXT: v_mov_b32_e32 v1, 0
+; GCN-NEXT: v_mov_b32_e32 v2, 1.0
+; GCN-NEXT: v_mov_b32_e32 v3, 0.5
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: exp pos0 v1, v1, v1, v0 done
+; GCN-NEXT: exp invalid_target_32 v2, v2, v2, v2
+; GCN-NEXT: exp invalid_target_33 v2, v2, v2, v3
+; GCN-NEXT: s_setprio 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_endpgm
+ call void @llvm.amdgcn.exp.f32(i32 32, i32 15, float 1.0, float 1.0, float 1.0, float 1.0, i1 false, i1 false)
+ call void @llvm.amdgcn.exp.f32(i32 33, i32 15, float 1.0, float 1.0, float 1.0, float 0.5, i1 false, i1 false)
+ %load = call float @llvm.amdgcn.raw.ptr.buffer.load.f32(ptr addrspace(8) undef, i32 %idx, i32 0, i32 0)
+ call void @llvm.amdgcn.exp.f32(i32 12, i32 15, float 0.0, float 0.0, float 0.0, float %load, i1 true, i1 false)
+ ret void
+}
+
+define amdgpu_ps void @test_export_across_store_load(i32 %idx, float %v) #0 {
+; GCN-LABEL: test_export_across_store_load:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: v_mov_b32_e32 v2, 24
+; GCN-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
+; GCN-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GCN-NEXT: v_cndmask_b32_e64 v0, v2, 8, vcc_lo
+; GCN-NEXT: v_mov_b32_e32 v2, 0
+; GCN-NEXT: scratch_store_b32 v0, v1, off
+; GCN-NEXT: scratch_load_b32 v0, off, off
+; GCN-NEXT: v_mov_b32_e32 v1, 1.0
+; GCN-NEXT: exp pos0 v2, v2, v2, v1 done
+; GCN-NEXT: s_setprio 0
+; GCN-NEXT: s_waitcnt_expcnt null, 0x0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: exp invalid_target_32 v0, v2, v1, v2
+; GCN-NEXT: exp invalid_target_33 v0, v2, v1, v2
+; GCN-NEXT: s_setprio 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_nop 0
+; GCN-NEXT: s_endpgm
+ %data0 = alloca <4 x float>, align 8, addrspace(5)
+ %data1 = alloca <4 x float>, align 8, addrspace(5)
+ %cmp = icmp eq i32 %idx, 1
+ %data = select i1 %cmp, ptr addrspace(5) %data0, ptr addrspace(5) %data1
+ store float %v, ptr addrspace(5) %data, align 8
+ call void @llvm.amdgcn.exp.f32(i32 12, i32 15, float 0.0, float 0.0, float 0.0, float 1.0, i1 true, i1 false)
+ %load0 = load float, ptr addrspace(5) %data0, align 8
+ call void @llvm.amdgcn.exp.f32(i32 32, i32 15, float %load0, float 0.0, float 1.0, float 0.0, i1 false, i1 false)
+ call void @llvm.amdgcn.exp.f32(i32 33, i32 15, float %load0, float 0.0, float 1.0, float 0.0, i1 false, i1 false)
+ ret void
+}
+
+define amdgpu_ps void @test_export_in_callee(float %v) #0 {
+; GCN-LABEL: test_export_in_callee:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: s_getpc_b64 s[0:1]
+; GCN-NEXT: s_add_u32 s0, s0, test_export_gfx@gotpcrel32@lo+4
+; GCN-NEXT: s_addc_u32 s1, s1, test_export_gfx@gotpcrel32@hi+12
+; GCN-NEXT: v_add_f32_e32 v0, 1.0, v0
+; GCN-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GCN-NEXT: s_mov_b32 s32, 0
+; GCN-NEXT: s_waitcnt lgkmcnt(0)
+; GCN-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GCN-NEXT: s_endpgm
+ %x = fadd float %v, 1.0
+ call void @test_export_gfx(float %x)
+ ret void
+}
+
+define amdgpu_ps void @test_export_in_callee_prio(float %v) #0 {
+; GCN-LABEL: test_export_in_callee_prio:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: s_mov_b32 s32, 0
+; GCN-NEXT: v_add_f32_e32 v0, 1.0, v0
+; GCN-NEXT: s_setprio 2
+; GCN-NEXT: s_getpc_b64 s[0:1]
+; GCN-NEXT: s_add_u32 s0, s0, test_export_gfx@gotpcrel32@lo+4
+; GCN-NEXT: s_addc_u32 s1, s1, test_export_gfx@gotpcrel32@hi+12
+; GCN-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GCN-NEXT: s_waitcnt lgkmcnt(0)
+; GCN-NEXT: s_swappc_b64 s[30:31], s[0:1]
+; GCN-NEXT: s_endpgm
+ %x = fadd float %v, 1.0
+ call void @llvm.amdgcn.s.setprio(i16 0)
+ call void @test_export_gfx(float %x)
+ ret void
+}
+
+declare void @llvm.amdgcn.exp.f32(i32, i32, float, float, float, float, i1, i1) #1
+declare void @llvm.amdgcn.exp.i32(i32, i32, i32, i32, i32, i32, i1, i1) #1
+declare float @llvm.amdgcn.raw.ptr.buffer.load.f32(ptr addrspace(8), i32, i32, i32) #2
+declare void @llvm.amdgcn.s.setprio(i16)
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind inaccessiblememonly }
+attributes #2 = { nounwind readnone }
diff --git a/llvm/test/CodeGen/AMDGPU/required-export-priority.mir b/llvm/test/CodeGen/AMDGPU/required-export-priority.mir
new file mode 100644
index 0000000000000..eee04468036e5
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/required-export-priority.mir
@@ -0,0 +1,293 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+# RUN: llc -mtriple=amdgcn -mcpu=gfx1150 -mattr=-wavefrontsize32,+wavefrontsize64 -run-pass=post-RA-hazard-rec -verify-machineinstrs %s -o - | FileCheck -check-prefixes=GFX1150 %s
+
+--- |
+ define amdgpu_ps void @end_of_shader() {
+ ret void
+ }
+ define amdgpu_ps void @end_of_shader_return_to_epilogue() {
+ ret void
+ }
+ define amdgpu_ps void @end_of_block() {
+ ret void
+ }
+ define amdgpu_ps void @start_of_block() {
+ ret void
+ }
+ define amdgpu_ps void @block_of_exports() {
+ ret void
+ }
+ define amdgpu_ps void @sparse_exports() {
+ ret void
+ }
+ define amdgpu_ps void @existing_setprio_1() {
+ ret void
+ }
+ define amdgpu_ps void @existing_setprio_2() {
+ ret void
+ }
+...
+
+---
+name: end_of_shader
+tracksRegLiveness: true
+liveins:
+ - { reg: '$vgpr0' }
+body: |
+ bb.0:
+ liveins: $vgpr0
+ ; GFX1150-LABEL: name: end_of_shader
+ ; GFX1150: liveins: $vgpr0
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ ; GFX1150-NEXT: S_SETPRIO 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_ENDPGM 0
+ EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ S_ENDPGM 0
+...
+
+---
+name: end_of_shader_return_to_epilogue
+tracksRegLiveness: true
+liveins:
+ - { reg: '$vgpr0' }
+body: |
+ bb.0:
+ liveins: $vgpr0
+ ; GFX1150-LABEL: name: end_of_shader_return_to_epilogue
+ ; GFX1150: liveins: $vgpr0
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ ; GFX1150-NEXT: S_SETPRIO 0
+ ; GFX1150-NEXT: S_WAITCNT_EXPCNT $sgpr_null, 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: SI_RETURN_TO_EPILOG $vgpr0
+ EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ SI_RETURN_TO_EPILOG $vgpr0
+...
+
+---
+name: end_of_block
+tracksRegLiveness: true
+liveins:
+ - { reg: '$vgpr0' }
+body: |
+ ; GFX1150-LABEL: name: end_of_block
+ ; GFX1150: bb.0:
+ ; GFX1150-NEXT: successors: %bb.1(0x80000000)
+ ; GFX1150-NEXT: liveins: $vgpr0
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ ; GFX1150-NEXT: S_SETPRIO 0
+ ; GFX1150-NEXT: S_WAITCNT_EXPCNT $sgpr_null, 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: bb.1:
+ ; GFX1150-NEXT: S_ENDPGM 0
+ bb.0:
+ liveins: $vgpr0
+ EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+
+ bb.1:
+ S_ENDPGM 0
+...
+
+---
+name: start_of_block
+tracksRegLiveness: true
+liveins:
+ - { reg: '$vgpr0' }
+body: |
+ ; GFX1150-LABEL: name: start_of_block
+ ; GFX1150: bb.0:
+ ; GFX1150-NEXT: successors: %bb.1(0x80000000)
+ ; GFX1150-NEXT: liveins: $vgpr0
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: bb.1:
+ ; GFX1150-NEXT: successors: %bb.2(0x80000000)
+ ; GFX1150-NEXT: liveins: $vgpr0
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ ; GFX1150-NEXT: S_SETPRIO 0
+ ; GFX1150-NEXT: S_WAITCNT_EXPCNT $sgpr_null, 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: bb.2:
+ ; GFX1150-NEXT: S_ENDPGM 0
+ bb.0:
+ liveins: $vgpr0
+
+ bb.1:
+ liveins: $vgpr0
+ EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+
+ bb.2:
+ S_ENDPGM 0
+...
+
+---
+name: block_of_exports
+tracksRegLiveness: true
+liveins:
+ - { reg: '$vgpr0' }
+body: |
+ bb.0:
+ liveins: $vgpr0
+ ; GFX1150-LABEL: name: block_of_exports
+ ; GFX1150: liveins: $vgpr0
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: EXP 2, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ ; GFX1150-NEXT: EXP 1, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ ; GFX1150-NEXT: EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ ; GFX1150-NEXT: S_SETPRIO 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_ENDPGM 0
+ EXP 2, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ EXP 1, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ S_ENDPGM 0
+...
+
+---
+name: sparse_exports
+tracksRegLiveness: true
+liveins:
+ - { reg: '$vgpr0' }
+body: |
+ bb.0:
+ liveins: $vgpr0
+ ; GFX1150-LABEL: name: sparse_exports
+ ; GFX1150: liveins: $vgpr0
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: EXP 2, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ ; GFX1150-NEXT: S_SETPRIO 0
+ ; GFX1150-NEXT: S_WAITCNT_EXPCNT $sgpr_null, 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: $vgpr0 = V_AND_B32_e32 1, $vgpr0, implicit $exec
+ ; GFX1150-NEXT: EXP 1, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ ; GFX1150-NEXT: S_SETPRIO 0
+ ; GFX1150-NEXT: S_WAITCNT_EXPCNT $sgpr_null, 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: $vgpr0 = V_OR_B32_e32 2, $vgpr0, implicit $exec
+ ; GFX1150-NEXT: EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ ; GFX1150-NEXT: S_SETPRIO 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_ENDPGM 0
+ EXP 2, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ $vgpr0 = V_AND_B32_e32 1, $vgpr0, implicit $exec
+ EXP 1, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ $vgpr0 = V_OR_B32_e32 2, $vgpr0, implicit $exec
+ EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ S_ENDPGM 0
+...
+
+---
+name: existing_setprio_1
+tracksRegLiveness: true
+liveins:
+ - { reg: '$vgpr0' }
+body: |
+ ; GFX1150-LABEL: name: existing_setprio_1
+ ; GFX1150: bb.0:
+ ; GFX1150-NEXT: successors: %bb.1(0x80000000)
+ ; GFX1150-NEXT: liveins: $vgpr0
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: $vgpr0 = V_AND_B32_e32 1, $vgpr0, implicit $exec
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: bb.1:
+ ; GFX1150-NEXT: successors: %bb.2(0x80000000)
+ ; GFX1150-NEXT: liveins: $vgpr0
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: S_SETPRIO 3
+ ; GFX1150-NEXT: $vgpr0 = V_OR_B32_e32 2, $vgpr0, implicit $exec
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: bb.2:
+ ; GFX1150-NEXT: successors: %bb.3(0x80000000)
+ ; GFX1150-NEXT: liveins: $vgpr0
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: S_SETPRIO 3
+ ; GFX1150-NEXT: $vgpr0 = V_OR_B32_e32 3, $vgpr0, implicit $exec
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: bb.3:
+ ; GFX1150-NEXT: liveins: $vgpr0
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: EXP 1, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ ; GFX1150-NEXT: EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ ; GFX1150-NEXT: S_SETPRIO 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_ENDPGM 0
+ bb.0:
+ liveins: $vgpr0
+ $vgpr0 = V_AND_B32_e32 1, $vgpr0, implicit $exec
+
+ bb.1:
+ liveins: $vgpr0
+ S_SETPRIO 3
+ $vgpr0 = V_OR_B32_e32 2, $vgpr0, implicit $exec
+ S_SETPRIO 0
+
+ bb.2:
+ liveins: $vgpr0
+ S_SETPRIO 1
+ $vgpr0 = V_OR_B32_e32 3, $vgpr0, implicit $exec
+ S_SETPRIO 0
+
+ bb.3:
+ liveins: $vgpr0
+ EXP 1, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ S_ENDPGM 0
+...
+
+---
+name: existing_setprio_2
+tracksRegLiveness: true
+liveins:
+ - { reg: '$vgpr0' }
+body: |
+ bb.0:
+ liveins: $vgpr0
+ ; GFX1150-LABEL: name: existing_setprio_2
+ ; GFX1150: liveins: $vgpr0
+ ; GFX1150-NEXT: {{ $}}
+ ; GFX1150-NEXT: S_SETPRIO 3
+ ; GFX1150-NEXT: EXP 1, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ ; GFX1150-NEXT: EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ ; GFX1150-NEXT: S_SETPRIO 0
+ ; GFX1150-NEXT: S_WAITCNT_EXPCNT $sgpr_null, 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_NOP 0
+ ; GFX1150-NEXT: S_SETPRIO 2
+ ; GFX1150-NEXT: S_SETPRIO 3
+ ; GFX1150-NEXT: S_ENDPGM 0
+ S_SETPRIO 3
+ EXP 1, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+ S_SETPRIO 3
+ S_ENDPGM 0
+...
From 8ea44e65f2c19facff751aeb2ac960f907fb210f Mon Sep 17 00:00:00 2001
From: Carl Ritson <carl.ritson@amd.com>
Date: Wed, 17 Jul 2024 16:18:02 +0900
Subject: [PATCH 2/3] Remove -verify-machineinstrs from test.
---
llvm/test/CodeGen/AMDGPU/required-export-priority.ll | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/llvm/test/CodeGen/AMDGPU/required-export-priority.ll b/llvm/test/CodeGen/AMDGPU/required-export-priority.ll
index 377902f3f0d1a..ebc209bd4d451 100644
--- a/llvm/test/CodeGen/AMDGPU/required-export-priority.ll
+++ b/llvm/test/CodeGen/AMDGPU/required-export-priority.ll
@@ -1,5 +1,5 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
-; RUN: llc -mtriple=amdgcn -mcpu=gfx1150 -amdgpu-enable-vopd=0 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1150 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefix=GCN %s
define amdgpu_ps void @test_export_zeroes_f32() #0 {
; GCN-LABEL: test_export_zeroes_f32:

2137
llvm.spec
View File

File diff suppressed because it is too large Load Diff

11
macros.clang Normal file
View File

@ -0,0 +1,11 @@
%clang_major_version @@CLANG_MAJOR_VERSION@@
%clang_minor_version @@CLANG_MINOR_VERSION@@
%clang_patch_version @@CLANG_PATCH_VERSION@@
%clang_version %{clang_major_version}.%{clang_minor_version}.%{clang_patch_version}
# This is the path to the clang resource directory that has clang's internal
# headers and libraries. This path should be used by packages that need to
# install files into this directory. This macro's value changes every time
# clang's version changes.
%clang_resource_dir %{_prefix}/lib/clang/%{clang_major_version}

10
sources
View File

@ -1,6 +1,4 @@
SHA512 (llvm-18.1.8.src.tar.xz) = 930814730bb2d80cf7f7b2968f0f1f1442009ca62a7ca29992b69d63823270584b059d16aa845bb381411da566e7e4f255fcfbc38acbdf865eb0419b4dfd7459 SHA512 (llvm-project-19.1.1.src.tar.xz) = 84adab40ffb9ec236dbf203d86c08a0c2c651f98278a9d0936490c7901159eb26eabd3db9316013886b549426d4acb43b75d866f7dc670ab299bf93ba35b1891
SHA512 (llvm-18.1.8.src.tar.xz.sig) = aab7cb61a6b5dd3776a9b306d91d08763710725b72ba6a4263d3cca5ae5959e3b073b27dbfd95f9a53a78600c6f414e2fd1cc0dbe3176d7cf142996f7af700ca SHA512 (llvm-project-19.1.1.src.tar.xz.sig) = 07bb7bffb2b035417d702ca47be9d5759250f1a2cd57606855027d458ceb972a293b45d3d93bcda195588986acbb5eace60524f4aecdc0da7aeb3a8414c37c31
SHA512 (cmake-18.1.8.src.tar.xz) = e02243b491f9e688db28d7b53270fcf87debf09d3c95b136a7c7b96e26890de68712c60a1e85f5a448a95ad8c81f2d8ae77047780822443bbe39f1a9e6211007 SHA512 (llvm-project-18.1.8.src.tar.xz) = 25eeee9984c8b4d0fbc240df90f33cbb000d3b0414baff5c8982beafcc5e59e7ef18f6f85d95b3a5f60cb3d4cd4f877c80487b5768bc21bc833f107698ad93db
SHA512 (cmake-18.1.8.src.tar.xz.sig) = 99191e95130fe4363a8db8f411a0e61af0549ad182a1280f99f0dd3ee679a321b993d103c6915d535a55d9f8a4d7fea86b7fdcc77605e02150e8edf1e18dee57 SHA512 (llvm-project-18.1.8.src.tar.xz.sig) = ddfd1e8a06756759af6cbe488c82a6d6a62ba91f3e8a0eb4cece561321824f5d165b08ed91010588790b76e19790931d2651b24dba8567e3b151d3cb43bec25b
SHA512 (third-party-18.1.8.src.tar.xz) = bedaa5d29ebeaf0ee1c700eb8492d0fef185e7c16528202927c81117d94fadd568829aa0e1873e1217e8e72866f3876a9681bbdb2a6a0a5466fc911f7b3620d4
SHA512 (third-party-18.1.8.src.tar.xz.sig) = 32c4d779a56a3908b291a4f0cf1df72ccb86b55439ad66f9cbad1b48a77cb92b129b131806d2914d0e63cb319cde3181a2c03b75856ec36cee5f88120bb58214