[PATCH] D57977: [HIP] compile option code-object-v3 propagate to llc

2019-02-19 Thread Aaron Enye Shi via Phabricator via cfe-commits 
ashi1 added a comment.

This was submitted again with the fix. Thanks!


Repository:
  rC Clang

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57977/new/

https://reviews.llvm.org/D57977



___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[PATCH] D57977: [HIP] compile option code-object-v3 propagate to llc

2019-02-12 Thread Reid Kleckner via Phabricator via cfe-commits 
rnk added a comment.

I reverted this in rC353893  because it was 
still causing test failures.


Repository:
  rC Clang

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57977/new/

https://reviews.llvm.org/D57977



___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[PATCH] D57977: [HIP] compile option code-object-v3 propagate to llc

2019-02-12 Thread Matt Arsenault via Phabricator via cfe-commits 
arsenm added a comment.

I think this breaks the attributes already listed in the functions on the IR 
since -mattr overrides those. I think we need to stop using subtarget features 
to communicate this


Repository:
  rC Clang

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57977/new/

https://reviews.llvm.org/D57977



___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[PATCH] D57977: [HIP] compile option code-object-v3 propagate to llc

2019-02-12 Thread Phabricator via Phabricator via cfe-commits 
This revision was automatically updated to reflect the committed changes.
Closed by commit rC353880: [HIP] Handle compile -m options and propagate into 
LLC (authored by aaronenyeshi, committed by ).

Changed prior to commit:
  https://reviews.llvm.org/D57977?vs=186528&id=186534#toc

Repository:
  rC Clang

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57977/new/

https://reviews.llvm.org/D57977

Files:
  lib/Driver/ToolChains/HIP.cpp
  test/Driver/hip-toolchain-features.hip


Index: test/Driver/hip-toolchain-features.hip
===
--- test/Driver/hip-toolchain-features.hip
+++ test/Driver/hip-toolchain-features.hip
@@ -0,0 +1,49 @@
+// REQUIRES: clang-driver
+// REQUIRES: x86-registered-target
+// REQUIRES: amdgpu-registered-target
+
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mcode-object-v3 2>&1 | FileCheck %s -check-prefix=COV3
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mno-code-object-v3 2>&1 | FileCheck %s -check-prefix=NOCOV3
+
+// COV3: {{.*}}clang{{.*}}"-target-feature" "+code-object-v3"
+// NOCOV3: {{.*}}clang{{.*}}"-target-feature" "-code-object-v3"
+
+
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mxnack 2>&1 | FileCheck %s -check-prefix=XNACK
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mno-xnack 2>&1 | FileCheck %s -check-prefix=NOXNACK
+
+// XNACK: {{.*}}clang{{.*}}"-target-feature" "+xnack"
+// NOXNACK: {{.*}}clang{{.*}}"-target-feature" "-xnack"
+
+
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -msram-ecc 2>&1 | FileCheck %s -check-prefix=SRAM
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mno-sram-ecc 2>&1 | FileCheck %s -check-prefix=NOSRAM
+
+// SRAM: {{.*}}clang{{.*}}"-target-feature" "+sram-ecc"
+// NOSRAM: {{.*}}clang{{.*}}"-target-feature" "-sram-ecc"
+
+
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mcode-object-v3 -mxnack -msram-ecc \
+// RUN:   2>&1 | FileCheck %s -check-prefix=ALL3
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mno-code-object-v3 -mno-xnack -mno-sram-ecc \
+// RUN:   2>&1 | FileCheck %s -check-prefix=NOALL3
+
+// ALL3: {{.*}}clang{{.*}}"-target-feature" "+code-object-v3" 
"-target-feature" "+xnack" "-target-feature" "+sram-ecc"
+// NOALL3: {{.*}}clang{{.*}}"-target-feature" "-code-object-v3" 
"-target-feature" "-xnack" "-target-feature" "-sram-ecc"
+
Index: lib/Driver/ToolChains/HIP.cpp
===
--- lib/Driver/ToolChains/HIP.cpp
+++ lib/Driver/ToolChains/HIP.cpp
@@ -159,8 +159,25 @@
 llvm::StringRef OutputFilePrefix, const char *InputFileName) const {
   // Construct llc command.
   ArgStringList LlcArgs{InputFileName, "-mtriple=amdgcn-amd-amdhsa",
-"-filetype=obj", "-mattr=-code-object-v3",
-Args.MakeArgString("-mcpu=" + SubArchName), "-o"};
+"-filetype=obj",
+Args.MakeArgString("-mcpu=" + SubArchName)};
+
+  // Extract all the -m options
+  std::vector Features;
+  handleTargetFeaturesGroup(
+Args, Features, options::OPT_m_amdgpu_Features_Group);
+
+  // Add features to mattr such as code-object-v3 and xnack
+  std::string MAttrString = "-mattr=";
+  for(auto OneFeature : Features) {
+MAttrString.append(Args.MakeArgString(OneFeature));
+if (OneFeature != Features.back())
+  MAttrString.append(",");
+  }
+  LlcArgs.push_back(Args.MakeArgString(MAttrString));
+
+  // Add output filename
+  LlcArgs.push_back("-o");
   std::string LlcOutputFileName =
   C.getDriver().GetTemporaryPath(OutputFilePrefix, "o");
   const char *LlcOutputFile =


Index: test/Driver/hip-toolchain-features.hip
===
--- test/Driver/hip-toolchain-features.hip
+++ test/Driver/hip-toolchain-features.hip
@@ -0,0 +1,49 @@
+// REQUIRES: clang-driver
+// REQUIRES: x86-registered-target
+// REQUIRES: amdgpu-registered-target
+
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mcode-object-v3 2>&1 | FileCheck %s -check-prefix=COV3
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx9

[PATCH] D57977: [HIP] compile option code-object-v3 propagate to llc

2019-02-12 Thread Konstantin Zhuravlyov via Phabricator via cfe-commits 
kzhuravl accepted this revision.
kzhuravl added a comment.
This revision is now accepted and ready to land.

LGTM. Please make sure to update your commit message as your patch processes 
all m options now.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57977/new/

https://reviews.llvm.org/D57977



___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[PATCH] D57977: [HIP] compile option code-object-v3 propagate to llc

2019-02-12 Thread Aaron Enye Shi via Phabricator via cfe-commits 
ashi1 added a comment.

Yes, but it will need this PR: 
https://github.com/ROCm-Developer-Tools/HIP/pull/910
So that the default is set.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57977/new/

https://reviews.llvm.org/D57977



___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[PATCH] D57977: [HIP] compile option code-object-v3 propagate to llc

2019-02-12 Thread Konstantin Zhuravlyov via Phabricator via cfe-commits 
kzhuravl added a comment.

Is it still code object v2 by default?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57977/new/

https://reviews.llvm.org/D57977



___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[PATCH] D57977: [HIP] compile option code-object-v3 propagate to llc

2019-02-12 Thread Aaron Enye Shi via Phabricator via cfe-commits 
ashi1 updated this revision to Diff 186528.
ashi1 added a comment.

I've added a lit test to check for options such as -mxnack/-mno-xnack, 
-msram-ecc/-mno-sram-ecc, -mcode-object-v3/-mno-code-object-v3


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57977/new/

https://reviews.llvm.org/D57977

Files:
  lib/Driver/ToolChains/HIP.cpp
  test/Driver/hip-toolchain-features.hip


Index: test/Driver/hip-toolchain-features.hip
===
--- /dev/null
+++ test/Driver/hip-toolchain-features.hip
@@ -0,0 +1,48 @@
+// REQUIRES: clang-driver
+// REQUIRES: x86-registered-target
+// REQUIRES: amdgpu-registered-target
+
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mcode-object-v3 2>&1 | FileCheck %s -check-prefix=COV3
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mno-code-object-v3 2>&1 | FileCheck %s -check-prefix=NOCOV3
+
+// COV3: {{.*}}clang{{.*}}"-target-feature" "+code-object-v3"
+// NOCOV3: {{.*}}clang{{.*}}"-target-feature" "-code-object-v3"
+
+
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mxnack 2>&1 | FileCheck %s -check-prefix=XNACK
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mno-xnack 2>&1 | FileCheck %s -check-prefix=NOXNACK
+
+// XNACK: {{.*}}clang{{.*}}"-target-feature" "+xnack"
+// NOXNACK: {{.*}}clang{{.*}}"-target-feature" "-xnack"
+
+
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -msram-ecc 2>&1 | FileCheck %s -check-prefix=SRAM
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mno-sram-ecc 2>&1 | FileCheck %s -check-prefix=NOSRAM
+
+// SRAM: {{.*}}clang{{.*}}"-target-feature" "+sram-ecc"
+// NOSRAM: {{.*}}clang{{.*}}"-target-feature" "-sram-ecc"
+
+
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mcode-object-v3 -mxnack -msram-ecc \
+// RUN:   2>&1 | FileCheck %s -check-prefix=ALL3
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mno-code-object-v3 -mno-xnack -mno-sram-ecc \
+// RUN:   2>&1 | FileCheck %s -check-prefix=NOALL3
+
+// ALL3: {{.*}}clang{{.*}}"-target-feature" "+code-object-v3" 
"-target-feature" "+xnack" "-target-feature" "+sram-ecc"
+// NOALL3: {{.*}}clang{{.*}}"-target-feature" "-code-object-v3" 
"-target-feature" "-xnack" "-target-feature" "-sram-ecc"
Index: lib/Driver/ToolChains/HIP.cpp
===
--- lib/Driver/ToolChains/HIP.cpp
+++ lib/Driver/ToolChains/HIP.cpp
@@ -162,9 +162,26 @@
   // AMDGPUPromoteAlloca pass which cause invalid memory access in PyTorch.
   // Remove this once the issue is fixed.
   ArgStringList LlcArgs{InputFileName, "-mtriple=amdgcn-amd-amdhsa",
-"-filetype=obj", "-mattr=-code-object-v3",
+"-filetype=obj",
 "-disable-promote-alloca-to-lds",
-Args.MakeArgString("-mcpu=" + SubArchName), "-o"};
+Args.MakeArgString("-mcpu=" + SubArchName)};
+
+  // Extract all the -m options
+  std::vector Features;
+  handleTargetFeaturesGroup(
+Args, Features, options::OPT_m_amdgpu_Features_Group);
+
+  // Add features to mattr such as code-object-v3 and xnack
+  std::string MAttrString = "-mattr=";
+  for(auto OneFeature : Features) {
+MAttrString.append(Args.MakeArgString(OneFeature));
+if (OneFeature != Features.back())
+  MAttrString.append(",");
+  }
+  LlcArgs.push_back(Args.MakeArgString(MAttrString));
+
+  // Add output filename
+  LlcArgs.push_back("-o");
   std::string LlcOutputFileName =
   C.getDriver().GetTemporaryPath(OutputFilePrefix, "o");
   const char *LlcOutputFile =


Index: test/Driver/hip-toolchain-features.hip
===
--- /dev/null
+++ test/Driver/hip-toolchain-features.hip
@@ -0,0 +1,48 @@
+// REQUIRES: clang-driver
+// REQUIRES: x86-registered-target
+// REQUIRES: amdgpu-registered-target
+
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mcode-object-v3 2>&1 | FileCheck %s -check-prefix=COV3
+// RUN: %clang -### -c -target x86_64-linux-gnu -fgpu-rdc \
+// RUN:   -x hip --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s \
+// RUN:   -mno-code-object-v3 2>&1 | FileCheck %s -check-prefix=NOCOV3
+
+// CO

[PATCH] D57977: [HIP] compile option code-object-v3 propagate to llc

2019-02-12 Thread Konstantin Zhuravlyov via Phabricator via cfe-commits 
kzhuravl added a comment.

Can you add some tests? Preferably with -mxnack/-mno-xnack, 
-msram-ecc/-mno-sram-ecc, -mcode-object-v3/-mno-code-object-v3


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57977/new/

https://reviews.llvm.org/D57977



___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[PATCH] D57977: [HIP] compile option code-object-v3 propagate to llc

2019-02-12 Thread Aaron Enye Shi via Phabricator via cfe-commits 
ashi1 updated this revision to Diff 186519.
ashi1 added a comment.
Herald added a subscriber: jdoerfert.

Changed patch to us handleTargetFeatures, to consume all -m options passed into 
clang.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57977/new/

https://reviews.llvm.org/D57977

Files:
  lib/Driver/ToolChains/HIP.cpp


Index: lib/Driver/ToolChains/HIP.cpp
===
--- lib/Driver/ToolChains/HIP.cpp
+++ lib/Driver/ToolChains/HIP.cpp
@@ -162,9 +162,26 @@
   // AMDGPUPromoteAlloca pass which cause invalid memory access in PyTorch.
   // Remove this once the issue is fixed.
   ArgStringList LlcArgs{InputFileName, "-mtriple=amdgcn-amd-amdhsa",
-"-filetype=obj", "-mattr=-code-object-v3",
+"-filetype=obj",
 "-disable-promote-alloca-to-lds",
-Args.MakeArgString("-mcpu=" + SubArchName), "-o"};
+Args.MakeArgString("-mcpu=" + SubArchName)};
+
+  // Extract all the -m options
+  std::vector Features;
+  handleTargetFeaturesGroup(
+Args, Features, options::OPT_m_amdgpu_Features_Group);
+
+  // Add features to mattr such as code-object-v3 and xnack
+  std::string MAttrString = "-mattr=";
+  for(auto OneFeature : Features) {
+MAttrString.append(Args.MakeArgString(OneFeature));
+if (OneFeature != Features.back())
+  MAttrString.append(",");
+  }
+  LlcArgs.push_back(Args.MakeArgString(MAttrString));
+
+  // Add output filename
+  LlcArgs.push_back("-o");
   std::string LlcOutputFileName =
   C.getDriver().GetTemporaryPath(OutputFilePrefix, "o");
   const char *LlcOutputFile =


Index: lib/Driver/ToolChains/HIP.cpp
===
--- lib/Driver/ToolChains/HIP.cpp
+++ lib/Driver/ToolChains/HIP.cpp
@@ -162,9 +162,26 @@
   // AMDGPUPromoteAlloca pass which cause invalid memory access in PyTorch.
   // Remove this once the issue is fixed.
   ArgStringList LlcArgs{InputFileName, "-mtriple=amdgcn-amd-amdhsa",
-"-filetype=obj", "-mattr=-code-object-v3",
+"-filetype=obj",
 "-disable-promote-alloca-to-lds",
-Args.MakeArgString("-mcpu=" + SubArchName), "-o"};
+Args.MakeArgString("-mcpu=" + SubArchName)};
+
+  // Extract all the -m options
+  std::vector Features;
+  handleTargetFeaturesGroup(
+Args, Features, options::OPT_m_amdgpu_Features_Group);
+
+  // Add features to mattr such as code-object-v3 and xnack
+  std::string MAttrString = "-mattr=";
+  for(auto OneFeature : Features) {
+MAttrString.append(Args.MakeArgString(OneFeature));
+if (OneFeature != Features.back())
+  MAttrString.append(",");
+  }
+  LlcArgs.push_back(Args.MakeArgString(MAttrString));
+
+  // Add output filename
+  LlcArgs.push_back("-o");
   std::string LlcOutputFileName =
   C.getDriver().GetTemporaryPath(OutputFilePrefix, "o");
   const char *LlcOutputFile =
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[PATCH] D57977: [HIP] compile option code-object-v3 propagate to llc

2019-02-08 Thread Konstantin Zhuravlyov via Phabricator via cfe-commits 
kzhuravl added a comment.

https://github.com/llvm-mirror/clang/blob/master/lib/Driver/ToolChains/AMDGPU.cpp#L52


Repository:
  rC Clang

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57977/new/

https://reviews.llvm.org/D57977



___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[PATCH] D57977: [HIP] compile option code-object-v3 propagate to llc

2019-02-08 Thread Konstantin Zhuravlyov via Phabricator via cfe-commits 
kzhuravl added a comment.

Can you handle all features as follows instead of checking a particular option?

‘handleTargetFeaturesGroup(

  Args, Features, options::OPT_m_amdgpu_Features_Group);’ ?

There is code in AMDGPU tool chain you can use as a reference.


Repository:
  rC Clang

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57977/new/

https://reviews.llvm.org/D57977



___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[PATCH] D57977: [HIP] compile option code-object-v3 propagate to llc

2019-02-08 Thread Aaron Enye Shi via Phabricator via cfe-commits 
ashi1 created this revision.
ashi1 added reviewers: yaxunl, kzhuravl.
Herald added a project: clang.
Herald added a subscriber: cfe-commits.

This patch allows the compile option -mcode-object-v3 to be set by default, 
unless the -mno-code-object-v3 is added to the compile options.


Repository:
  rC Clang

https://reviews.llvm.org/D57977

Files:
  lib/Driver/ToolChains/HIP.cpp


Index: lib/Driver/ToolChains/HIP.cpp
===
--- lib/Driver/ToolChains/HIP.cpp
+++ lib/Driver/ToolChains/HIP.cpp
@@ -157,12 +157,20 @@
 Compilation &C, const JobAction &JA, const InputInfoList &Inputs,
 const llvm::opt::ArgList &Args, llvm::StringRef SubArchName,
 llvm::StringRef OutputFilePrefix, const char *InputFileName) const {
+
+  llvm::StringRef CodeObjectVersion;
+  // Add Code Object Version
+  if (Args.hasArg(options::OPT_mno_code_object_v3))
+CodeObjectVersion = "-mattr=-code-object-v3";
+  else
+CodeObjectVersion = "-mattr=+code-object-v3";
+
   // Construct llc command.
   // FIXME: -disable-promote-alloca-to-lds is a workaround for issues in
   // AMDGPUPromoteAlloca pass which cause invalid memory access in PyTorch.
   // Remove this once the issue is fixed.
   ArgStringList LlcArgs{InputFileName, "-mtriple=amdgcn-amd-amdhsa",
-"-filetype=obj", "-mattr=-code-object-v3",
+"-filetype=obj", Args.MakeArgString(CodeObjectVersion),
 "-disable-promote-alloca-to-lds",
 Args.MakeArgString("-mcpu=" + SubArchName), "-o"};
   std::string LlcOutputFileName =


Index: lib/Driver/ToolChains/HIP.cpp
===
--- lib/Driver/ToolChains/HIP.cpp
+++ lib/Driver/ToolChains/HIP.cpp
@@ -157,12 +157,20 @@
 Compilation &C, const JobAction &JA, const InputInfoList &Inputs,
 const llvm::opt::ArgList &Args, llvm::StringRef SubArchName,
 llvm::StringRef OutputFilePrefix, const char *InputFileName) const {
+
+  llvm::StringRef CodeObjectVersion;
+  // Add Code Object Version
+  if (Args.hasArg(options::OPT_mno_code_object_v3))
+CodeObjectVersion = "-mattr=-code-object-v3";
+  else
+CodeObjectVersion = "-mattr=+code-object-v3";
+
   // Construct llc command.
   // FIXME: -disable-promote-alloca-to-lds is a workaround for issues in
   // AMDGPUPromoteAlloca pass which cause invalid memory access in PyTorch.
   // Remove this once the issue is fixed.
   ArgStringList LlcArgs{InputFileName, "-mtriple=amdgcn-amd-amdhsa",
-"-filetype=obj", "-mattr=-code-object-v3",
+"-filetype=obj", Args.MakeArgString(CodeObjectVersion),
 "-disable-promote-alloca-to-lds",
 Args.MakeArgString("-mcpu=" + SubArchName), "-o"};
   std::string LlcOutputFileName =
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits