[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Gheorghe-Teodor Bercea via Phabricator via cfe-commits]

doru1004 closed this revision.
doru1004 added a comment.

Commit: 01910787d386584ea5a3d5dc317a908423ba39ed 



CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

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

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Joseph Huber via Phabricator via cfe-commits]

jhuber6 accepted this revision.
jhuber6 added a comment.
This revision is now accepted and ready to land.

LGTM unless anyone else has any concerns.




Comment at: clang/test/OpenMP/target_team_variable_codegen.cpp:33
+//.
+// CHECK-NVIDIA: @local_a = internal addrspace(3) global [10 x i32] 
zeroinitializer, align 4
+//.

doru1004 wrote:
> jhuber6 wrote:
> > jdoerfert wrote:
> > > doru1004 wrote:
> > > > jhuber6 wrote:
> > > > > Shouldn't the Nvidia version also be undefined? Not sure why this 
> > > > > should vary depending on the target.
> > > > Perhaps NVIDIA code path can tolerate a zeroinitializer? I don't want 
> > > > to change it if it's not needed. I am basing this check on the code 
> > > > path for AMD GPUs and the initial bug that was reported.
> > > for AS 3 we should make it always poison.
> > We should probably change this in `HeadToShared` in `OpenMPOpt` as well.
> Happy to remove the guard and have it always use poison for both NVIDIA and 
> AMD.
These should be a single check line now.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

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

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Gheorghe-Teodor Bercea via Phabricator via cfe-commits]

doru1004 added a comment.

ping


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

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

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Gheorghe-Teodor Bercea via Phabricator via cfe-commits]

doru1004 updated this revision to Diff 511444.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

Files:
  clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
  clang/test/OpenMP/target_team_variable_codegen.cpp


Index: clang/test/OpenMP/target_team_variable_codegen.cpp
===
--- /dev/null
+++ clang/test/OpenMP/target_team_variable_codegen.cpp
@@ -0,0 +1,57 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py 
UTC_ARGS: --check-globals --prefix-filecheck-ir-name _ --global-value-regex 
"llvm.compiler.used" "_[0-9a-zA-Z]+A[0-9a-zA-Z]+pi[0-9a-zA-Z]+" 
"_[0-9a-zA-Z]+anotherPi" --version 2
+// REQUIRES: amdgpu-registered-target
+
+
+// Test target codegen - host bc file has to be created first.
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown 
-fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm-bc %s -o %t-ppc-host-amd.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple amdgcn-amd-amdhsa 
-fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm %s -fopenmp-target-debug 
-fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-amd.bc -o - | 
FileCheck %s --check-prefix=CHECK-AMD
+
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown 
-fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host-nvidia.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown 
-fopenmp-targets=nvptx64-unknown-unknown -emit-llvm %s -fopenmp-target-debug 
-fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-nvidia.bc -o - | 
FileCheck %s --check-prefix=CHECK-NVIDIA
+
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+typedef enum omp_allocator_handle_t {
+  omp_null_allocator = 0,
+  omp_default_mem_alloc = 1,
+  omp_large_cap_mem_alloc = 2,
+  omp_const_mem_alloc = 3,
+  omp_high_bw_mem_alloc = 4,
+  omp_low_lat_mem_alloc = 5,
+  omp_cgroup_mem_alloc = 6,
+  omp_pteam_mem_alloc = 7,
+  omp_thread_mem_alloc = 8,
+  KMP_ALLOCATOR_MAX_HANDLE = __UINTPTR_MAX__
+} omp_allocator_handle_t;
+
+//.
+// CHECK-AMD: @local_a = internal addrspace(3) global [10 x i32] poison, align 
4
+//.
+// CHECK-NVIDIA: @local_a = internal addrspace(3) global [10 x i32] poison, 
align 4
+//.
+int main()
+{
+   int N = 1;
+   int *a = new int[N];
+#pragma omp target data map(tofrom:a[:N])
+   {
+#pragma omp target teams distribute parallel for
+for(int i = 0; i < N; i++)
+{
+  int local_a[10];
+#pragma omp allocate(local_a) allocator(omp_pteam_mem_alloc)
+  for(int j = 0; j < 10; j++)
+   local_a[j] = a[(i + j) % N];
+  a[i] = local_a[0];
+}
+   }
+  return a[17];
+}
+
+#endif
+ NOTE: These prefixes are unused and the list is autogenerated. Do not add 
tests below this line:
+// CHECK-AMD: {{.*}}
+// CHECK-NVIDIA: {{.*}}
Index: clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
===
--- clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
+++ clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
@@ -3351,7 +3351,7 @@
 llvm::Type *VarTy = CGF.ConvertTypeForMem(VD->getType());
 auto *GV = new llvm::GlobalVariable(
 CGM.getModule(), VarTy, /*isConstant=*/false,
-llvm::GlobalValue::InternalLinkage, 
llvm::Constant::getNullValue(VarTy),
+llvm::GlobalValue::InternalLinkage, llvm::PoisonValue::get(VarTy),
 VD->getName(),
 /*InsertBefore=*/nullptr, llvm::GlobalValue::NotThreadLocal,
 CGM.getContext().getTargetAddressSpace(AS));


Index: clang/test/OpenMP/target_team_variable_codegen.cpp
===
--- /dev/null
+++ clang/test/OpenMP/target_team_variable_codegen.cpp
@@ -0,0 +1,57 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --check-globals --prefix-filecheck-ir-name _ --global-value-regex "llvm.compiler.used" "_[0-9a-zA-Z]+A[0-9a-zA-Z]+pi[0-9a-zA-Z]+" "_[0-9a-zA-Z]+anotherPi" --version 2
+// REQUIRES: amdgpu-registered-target
+
+
+// Test target codegen - host bc file has to be created first.
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm-bc %s -o %t-ppc-host-amd.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple amdgcn-amd-amdhsa -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm %s -fopenmp-target-debug -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-amd.bc -o - | FileCheck %s --check-prefix=CHECK-AMD
+
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host-nvidia.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-unknown-unknown -emit-llvm %s -fopenmp-target-debug -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-nvidia.bc -o - | FileCheck %s 

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Gheorghe-Teodor Bercea via Phabricator via cfe-commits]

doru1004 updated this revision to Diff 511436.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

Files:
  clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
  clang/test/OpenMP/target_team_variable_codegen.cpp


Index: clang/test/OpenMP/target_team_variable_codegen.cpp
===
--- /dev/null
+++ clang/test/OpenMP/target_team_variable_codegen.cpp
@@ -0,0 +1,57 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py 
UTC_ARGS: --check-globals --prefix-filecheck-ir-name _ --global-value-regex 
"llvm.compiler.used" "_[0-9a-zA-Z]+A[0-9a-zA-Z]+pi[0-9a-zA-Z]+" 
"_[0-9a-zA-Z]+anotherPi" --version 2
+// REQUIRES: amdgpu-registered-target
+
+
+// Test target codegen - host bc file has to be created first.
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown 
-fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm-bc %s -o %t-ppc-host-amd.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple amdgcn-amd-amdhsa 
-fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm %s -fopenmp-target-debug 
-fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-amd.bc -o - | 
FileCheck %s --check-prefix=CHECK-AMD
+
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown 
-fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host-nvidia.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown 
-fopenmp-targets=nvptx64-unknown-unknown -emit-llvm %s -fopenmp-target-debug 
-fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-nvidia.bc -o - | 
FileCheck %s --check-prefix=CHECK-NVIDIA
+
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+typedef enum omp_allocator_handle_t {
+  omp_null_allocator = 0,
+  omp_default_mem_alloc = 1,
+  omp_large_cap_mem_alloc = 2,
+  omp_const_mem_alloc = 3,
+  omp_high_bw_mem_alloc = 4,
+  omp_low_lat_mem_alloc = 5,
+  omp_cgroup_mem_alloc = 6,
+  omp_pteam_mem_alloc = 7,
+  omp_thread_mem_alloc = 8,
+  KMP_ALLOCATOR_MAX_HANDLE = __UINTPTR_MAX__
+} omp_allocator_handle_t;
+
+//.
+// CHECK-AMD: @local_a = internal addrspace(3) global [10 x i32] poison, align 
4
+//.
+// CHECK-NVIDIA: @local_a = internal addrspace(3) global [10 x i32] 
zeroinitializer, align 4
+//.
+int main()
+{
+   int N = 1;
+   int *a = new int[N];
+#pragma omp target data map(tofrom:a[:N])
+   {
+#pragma omp target teams distribute parallel for
+for(int i = 0; i < N; i++)
+{
+  int local_a[10];
+#pragma omp allocate(local_a) allocator(omp_pteam_mem_alloc)
+  for(int j = 0; j < 10; j++)
+   local_a[j] = a[(i + j) % N];
+  a[i] = local_a[0];
+}
+   }
+  return a[17];
+}
+
+#endif
+ NOTE: These prefixes are unused and the list is autogenerated. Do not add 
tests below this line:
+// CHECK-AMD: {{.*}}
+// CHECK-NVIDIA: {{.*}}
Index: clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
===
--- clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
+++ clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
@@ -3351,7 +3351,7 @@
 llvm::Type *VarTy = CGF.ConvertTypeForMem(VD->getType());
 auto *GV = new llvm::GlobalVariable(
 CGM.getModule(), VarTy, /*isConstant=*/false,
-llvm::GlobalValue::InternalLinkage, 
llvm::Constant::getNullValue(VarTy),
+llvm::GlobalValue::InternalLinkage, llvm::PoisonValue::get(VarTy),
 VD->getName(),
 /*InsertBefore=*/nullptr, llvm::GlobalValue::NotThreadLocal,
 CGM.getContext().getTargetAddressSpace(AS));


Index: clang/test/OpenMP/target_team_variable_codegen.cpp
===
--- /dev/null
+++ clang/test/OpenMP/target_team_variable_codegen.cpp
@@ -0,0 +1,57 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --check-globals --prefix-filecheck-ir-name _ --global-value-regex "llvm.compiler.used" "_[0-9a-zA-Z]+A[0-9a-zA-Z]+pi[0-9a-zA-Z]+" "_[0-9a-zA-Z]+anotherPi" --version 2
+// REQUIRES: amdgpu-registered-target
+
+
+// Test target codegen - host bc file has to be created first.
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm-bc %s -o %t-ppc-host-amd.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple amdgcn-amd-amdhsa -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm %s -fopenmp-target-debug -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-amd.bc -o - | FileCheck %s --check-prefix=CHECK-AMD
+
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host-nvidia.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-unknown-unknown -emit-llvm %s -fopenmp-target-debug -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-nvidia.bc -o - | FileCheck %s 

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Gheorghe-Teodor Bercea via Phabricator via cfe-commits]

doru1004 added inline comments.



Comment at: clang/test/OpenMP/target_team_variable_codegen.cpp:33
+//.
+// CHECK-NVIDIA: @local_a = internal addrspace(3) global [10 x i32] 
zeroinitializer, align 4
+//.

jhuber6 wrote:
> jdoerfert wrote:
> > doru1004 wrote:
> > > jhuber6 wrote:
> > > > Shouldn't the Nvidia version also be undefined? Not sure why this 
> > > > should vary depending on the target.
> > > Perhaps NVIDIA code path can tolerate a zeroinitializer? I don't want to 
> > > change it if it's not needed. I am basing this check on the code path for 
> > > AMD GPUs and the initial bug that was reported.
> > for AS 3 we should make it always poison.
> We should probably change this in `HeadToShared` in `OpenMPOpt` as well.
Happy to remove the guard and have it always use poison for both NVIDIA and AMD.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

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

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Joseph Huber via Phabricator via cfe-commits]

jhuber6 added inline comments.



Comment at: clang/test/OpenMP/target_team_variable_codegen.cpp:33
+//.
+// CHECK-NVIDIA: @local_a = internal addrspace(3) global [10 x i32] 
zeroinitializer, align 4
+//.

jdoerfert wrote:
> doru1004 wrote:
> > jhuber6 wrote:
> > > Shouldn't the Nvidia version also be undefined? Not sure why this should 
> > > vary depending on the target.
> > Perhaps NVIDIA code path can tolerate a zeroinitializer? I don't want to 
> > change it if it's not needed. I am basing this check on the code path for 
> > AMD GPUs and the initial bug that was reported.
> for AS 3 we should make it always poison.
We should probably change this in `HeadToShared` in `OpenMPOpt` as well.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

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

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Johannes Doerfert via Phabricator via cfe-commits]

jdoerfert added inline comments.



Comment at: clang/test/OpenMP/target_team_variable_codegen.cpp:33
+//.
+// CHECK-NVIDIA: @local_a = internal addrspace(3) global [10 x i32] 
zeroinitializer, align 4
+//.

doru1004 wrote:
> jhuber6 wrote:
> > Shouldn't the Nvidia version also be undefined? Not sure why this should 
> > vary depending on the target.
> Perhaps NVIDIA code path can tolerate a zeroinitializer? I don't want to 
> change it if it's not needed. I am basing this check on the code path for AMD 
> GPUs and the initial bug that was reported.
for AS 3 we should make it always poison.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

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

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Gheorghe-Teodor Bercea via Phabricator via cfe-commits]

doru1004 added inline comments.



Comment at: clang/test/OpenMP/target_team_variable_codegen.cpp:33
+//.
+// CHECK-NVIDIA: @local_a = internal addrspace(3) global [10 x i32] 
zeroinitializer, align 4
+//.

jhuber6 wrote:
> Shouldn't the Nvidia version also be undefined? Not sure why this should vary 
> depending on the target.
Perhaps NVIDIA code path can tolerate a zeroinitializer? I don't want to change 
it if it's not needed. I am basing this check on the code path for AMD GPUs and 
the initial bug that was reported.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

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

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Joseph Huber via Phabricator via cfe-commits]

jhuber6 added inline comments.



Comment at: clang/test/OpenMP/target_team_variable_codegen.cpp:33
+//.
+// CHECK-NVIDIA: @local_a = internal addrspace(3) global [10 x i32] 
zeroinitializer, align 4
+//.

Shouldn't the Nvidia version also be undefined? Not sure why this should vary 
depending on the target.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

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

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Gheorghe-Teodor Bercea via Phabricator via cfe-commits]

doru1004 updated this revision to Diff 510943.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

Files:
  clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
  clang/test/OpenMP/target_team_variable_codegen.cpp


Index: clang/test/OpenMP/target_team_variable_codegen.cpp
===
--- /dev/null
+++ clang/test/OpenMP/target_team_variable_codegen.cpp
@@ -0,0 +1,57 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py 
UTC_ARGS: --check-globals --prefix-filecheck-ir-name _ --global-value-regex 
"llvm.compiler.used" "_[0-9a-zA-Z]+A[0-9a-zA-Z]+pi[0-9a-zA-Z]+" 
"_[0-9a-zA-Z]+anotherPi" --version 2
+// REQUIRES: amdgpu-registered-target
+
+
+// Test target codegen - host bc file has to be created first.
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown 
-fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm-bc %s -o %t-ppc-host-amd.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple amdgcn-amd-amdhsa 
-fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm %s -fopenmp-target-debug 
-fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-amd.bc -o - | 
FileCheck %s --check-prefix=CHECK-AMD
+
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown 
-fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host-nvidia.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown 
-fopenmp-targets=nvptx64-unknown-unknown -emit-llvm %s -fopenmp-target-debug 
-fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-nvidia.bc -o - | 
FileCheck %s --check-prefix=CHECK-NVIDIA
+
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+typedef enum omp_allocator_handle_t {
+  omp_null_allocator = 0,
+  omp_default_mem_alloc = 1,
+  omp_large_cap_mem_alloc = 2,
+  omp_const_mem_alloc = 3,
+  omp_high_bw_mem_alloc = 4,
+  omp_low_lat_mem_alloc = 5,
+  omp_cgroup_mem_alloc = 6,
+  omp_pteam_mem_alloc = 7,
+  omp_thread_mem_alloc = 8,
+  KMP_ALLOCATOR_MAX_HANDLE = __UINTPTR_MAX__
+} omp_allocator_handle_t;
+
+//.
+// CHECK-AMD: @local_a = internal addrspace(3) global [10 x i32] poison, align 
4
+//.
+// CHECK-NVIDIA: @local_a = internal addrspace(3) global [10 x i32] 
zeroinitializer, align 4
+//.
+int main()
+{
+   int N = 1;
+   int *a = new int[N];
+#pragma omp target data map(tofrom:a[:N])
+   {
+#pragma omp target teams distribute parallel for
+for(int i = 0; i < N; i++)
+{
+  int local_a[10];
+#pragma omp allocate(local_a) allocator(omp_pteam_mem_alloc)
+  for(int j = 0; j < 10; j++)
+   local_a[j] = a[(i + j) % N];
+  a[i] = local_a[0];
+}
+   }
+  return a[17];
+}
+
+#endif
+ NOTE: These prefixes are unused and the list is autogenerated. Do not add 
tests below this line:
+// CHECK-AMD: {{.*}}
+// CHECK-NVIDIA: {{.*}}
Index: clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
===
--- clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
+++ clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
@@ -3351,7 +3351,9 @@
 llvm::Type *VarTy = CGF.ConvertTypeForMem(VD->getType());
 auto *GV = new llvm::GlobalVariable(
 CGM.getModule(), VarTy, /*isConstant=*/false,
-llvm::GlobalValue::InternalLinkage, 
llvm::Constant::getNullValue(VarTy),
+llvm::GlobalValue::InternalLinkage,
+CGM.getTriple().isAMDGCN() ? llvm::PoisonValue::get(VarTy)
+   : llvm::Constant::getNullValue(VarTy),
 VD->getName(),
 /*InsertBefore=*/nullptr, llvm::GlobalValue::NotThreadLocal,
 CGM.getContext().getTargetAddressSpace(AS));


Index: clang/test/OpenMP/target_team_variable_codegen.cpp
===
--- /dev/null
+++ clang/test/OpenMP/target_team_variable_codegen.cpp
@@ -0,0 +1,57 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --check-globals --prefix-filecheck-ir-name _ --global-value-regex "llvm.compiler.used" "_[0-9a-zA-Z]+A[0-9a-zA-Z]+pi[0-9a-zA-Z]+" "_[0-9a-zA-Z]+anotherPi" --version 2
+// REQUIRES: amdgpu-registered-target
+
+
+// Test target codegen - host bc file has to be created first.
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm-bc %s -o %t-ppc-host-amd.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple amdgcn-amd-amdhsa -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm %s -fopenmp-target-debug -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-amd.bc -o - | FileCheck %s --check-prefix=CHECK-AMD
+
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host-nvidia.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown 

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Gheorghe-Teodor Bercea via Phabricator via cfe-commits]

doru1004 added inline comments.



Comment at: llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp:303
+if (GV->hasInitializer() && !(isa(GV->getInitializer()) ||
+  isa(GV->getInitializer( {
   OutContext.reportError({},

arsenm wrote:
> Isa covers PoisonValue already 
Perfect! I'll revert this part.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

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

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Matt Arsenault via Phabricator via cfe-commits]

arsenm added inline comments.



Comment at: llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp:303
+if (GV->hasInitializer() && !(isa(GV->getInitializer()) ||
+  isa(GV->getInitializer( {
   OutContext.reportError({},

Isa covers PoisonValue already 


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

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

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Gheorghe-Teodor Bercea via Phabricator via cfe-commits]

doru1004 marked an inline comment as done.
doru1004 added inline comments.



Comment at: clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp:3355
+llvm::GlobalValue::InternalLinkage,
+CGM.getTriple().isAMDGCN() ? llvm::UndefValue::get(VarTy)
+   : llvm::Constant::getNullValue(VarTy),

nlopes wrote:
> Please use poison instead of undef wherever possible as we are tying to 
> remove undef. The replacement is usually safe when you just need a 
> placeholder.
> Thank you!
I've made the change as requested, this also means that I had to add another 
check in `AMDGPUAsmPrinter.cpp`.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

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

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Gheorghe-Teodor Bercea via Phabricator via cfe-commits]

doru1004 updated this revision to Diff 510934.
Herald added subscribers: llvm-commits, kosarev, foad, kerbowa, hiraditya, 
jvesely, arsenm.
Herald added a project: LLVM.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

Files:
  clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
  clang/test/OpenMP/target_team_variable_codegen.cpp
  llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp


Index: llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
===
--- llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
+++ llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
@@ -299,7 +299,8 @@
 
 void AMDGPUAsmPrinter::emitGlobalVariable(const GlobalVariable *GV) {
   if (GV->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS) {
-if (GV->hasInitializer() && !isa(GV->getInitializer())) {
+if (GV->hasInitializer() && !(isa(GV->getInitializer()) ||
+  isa(GV->getInitializer( {
   OutContext.reportError({},
  Twine(GV->getName()) +
  ": unsupported initializer for address 
space");
Index: clang/test/OpenMP/target_team_variable_codegen.cpp
===
--- /dev/null
+++ clang/test/OpenMP/target_team_variable_codegen.cpp
@@ -0,0 +1,57 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py 
UTC_ARGS: --check-globals --prefix-filecheck-ir-name _ --global-value-regex 
"llvm.compiler.used" "_[0-9a-zA-Z]+A[0-9a-zA-Z]+pi[0-9a-zA-Z]+" 
"_[0-9a-zA-Z]+anotherPi" --version 2
+// REQUIRES: amdgpu-registered-target
+
+
+// Test target codegen - host bc file has to be created first.
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown 
-fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm-bc %s -o %t-ppc-host-amd.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple amdgcn-amd-amdhsa 
-fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm %s -fopenmp-target-debug 
-fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-amd.bc -o - | 
FileCheck %s --check-prefix=CHECK-AMD
+
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown 
-fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host-nvidia.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown 
-fopenmp-targets=nvptx64-unknown-unknown -emit-llvm %s -fopenmp-target-debug 
-fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-nvidia.bc -o - | 
FileCheck %s --check-prefix=CHECK-NVIDIA
+
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+typedef enum omp_allocator_handle_t {
+  omp_null_allocator = 0,
+  omp_default_mem_alloc = 1,
+  omp_large_cap_mem_alloc = 2,
+  omp_const_mem_alloc = 3,
+  omp_high_bw_mem_alloc = 4,
+  omp_low_lat_mem_alloc = 5,
+  omp_cgroup_mem_alloc = 6,
+  omp_pteam_mem_alloc = 7,
+  omp_thread_mem_alloc = 8,
+  KMP_ALLOCATOR_MAX_HANDLE = __UINTPTR_MAX__
+} omp_allocator_handle_t;
+
+//.
+// CHECK-AMD: @local_a = internal addrspace(3) global [10 x i32] poison, align 
4
+//.
+// CHECK-NVIDIA: @local_a = internal addrspace(3) global [10 x i32] 
zeroinitializer, align 4
+//.
+int main()
+{
+   int N = 1;
+   int *a = new int[N];
+#pragma omp target data map(tofrom:a[:N])
+   {
+#pragma omp target teams distribute parallel for
+for(int i = 0; i < N; i++)
+{
+  int local_a[10];
+#pragma omp allocate(local_a) allocator(omp_pteam_mem_alloc)
+  for(int j = 0; j < 10; j++)
+   local_a[j] = a[(i + j) % N];
+  a[i] = local_a[0];
+}
+   }
+  return a[17];
+}
+
+#endif
+ NOTE: These prefixes are unused and the list is autogenerated. Do not add 
tests below this line:
+// CHECK-AMD: {{.*}}
+// CHECK-NVIDIA: {{.*}}
Index: clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
===
--- clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
+++ clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
@@ -3351,7 +3351,9 @@
 llvm::Type *VarTy = CGF.ConvertTypeForMem(VD->getType());
 auto *GV = new llvm::GlobalVariable(
 CGM.getModule(), VarTy, /*isConstant=*/false,
-llvm::GlobalValue::InternalLinkage, 
llvm::Constant::getNullValue(VarTy),
+llvm::GlobalValue::InternalLinkage,
+CGM.getTriple().isAMDGCN() ? llvm::PoisonValue::get(VarTy)
+   : llvm::Constant::getNullValue(VarTy),
 VD->getName(),
 /*InsertBefore=*/nullptr, llvm::GlobalValue::NotThreadLocal,
 CGM.getContext().getTargetAddressSpace(AS));


Index: llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
===
--- llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
+++ llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
@@ -299,7 +299,8 @@
 
 void AMDGPUAsmPrinter::emitGlobalVariable(const GlobalVariable *GV) {
   if (GV->getAddressSpace() 

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Nuno Lopes via Phabricator via cfe-commits]

nlopes added inline comments.



Comment at: clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp:3355
+llvm::GlobalValue::InternalLinkage,
+CGM.getTriple().isAMDGCN() ? llvm::UndefValue::get(VarTy)
+   : llvm::Constant::getNullValue(VarTy),

Please use poison instead of undef wherever possible as we are tying to remove 
undef. The replacement is usually safe when you just need a placeholder.
Thank you!


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147572/new/

https://reviews.llvm.org/D147572

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

[PATCH] D147572: [Clang][OpenMP] Fix failure with team-wide allocated variable

[Gheorghe-Teodor Bercea via Phabricator via cfe-commits]

doru1004 created this revision.
doru1004 added reviewers: ronl, carlo.bertolli, jhuber6, jdoerfert, dhruvachak, 
gregrodgers.
doru1004 added a project: OpenMP.
Herald added subscribers: sunshaoce, nlopes, guansong, arichardson, yaxunl.
Herald added a project: All.
doru1004 requested review of this revision.
Herald added subscribers: cfe-commits, jplehr, sstefan1.
Herald added a project: clang.

This patch aims to resolve issue: 
https://github.com/llvm/llvm-project/issues/60345

The following code:

  #include 
  #include 
  #include 
  
  
  int main()
  {
int N =1<<30;
int *a = new int[N];
  #pragma omp target data map(tofrom:a[:N])
{
 #pragma omp target teams distribute parallel for
  for(int i = 0; i < N; i++)
  {
   int local_a[10];
 #pragma omp allocate(local_a) allocator(omp_pteam_mem_alloc)
for(int j = 0; j < 10; j++)
local_a[j] = a[(i+j)%N];
a[i] = local_a[0];
}
}
  std::cout << a[0] << "\n";
  }

Fails with the following linker errors:

  clang-linker-wrapper: error: :0: local_a: unsupported initializer 
for address space
  
  clang-linker-wrapper: error: Errors encountered inside the LTO pipeline.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D147572

Files:
  clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
  clang/test/OpenMP/target_team_variable_codegen.cpp


Index: clang/test/OpenMP/target_team_variable_codegen.cpp
===
--- /dev/null
+++ clang/test/OpenMP/target_team_variable_codegen.cpp
@@ -0,0 +1,57 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py 
UTC_ARGS: --check-globals --prefix-filecheck-ir-name _ --global-value-regex 
"llvm.compiler.used" "_[0-9a-zA-Z]+A[0-9a-zA-Z]+pi[0-9a-zA-Z]+" 
"_[0-9a-zA-Z]+anotherPi" --version 2
+// REQUIRES: amdgpu-registered-target
+
+
+// Test target codegen - host bc file has to be created first.
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown 
-fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm-bc %s -o %t-ppc-host-amd.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple amdgcn-amd-amdhsa 
-fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm %s -fopenmp-target-debug 
-fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-amd.bc -o - | 
FileCheck %s --check-prefix=CHECK-AMD
+
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown 
-fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host-nvidia.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown 
-fopenmp-targets=nvptx64-unknown-unknown -emit-llvm %s -fopenmp-target-debug 
-fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-nvidia.bc -o - | 
FileCheck %s --check-prefix=CHECK-NVIDIA
+
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+typedef enum omp_allocator_handle_t {
+  omp_null_allocator = 0,
+  omp_default_mem_alloc = 1,
+  omp_large_cap_mem_alloc = 2,
+  omp_const_mem_alloc = 3,
+  omp_high_bw_mem_alloc = 4,
+  omp_low_lat_mem_alloc = 5,
+  omp_cgroup_mem_alloc = 6,
+  omp_pteam_mem_alloc = 7,
+  omp_thread_mem_alloc = 8,
+  KMP_ALLOCATOR_MAX_HANDLE = __UINTPTR_MAX__
+} omp_allocator_handle_t;
+
+//.
+// CHECK-AMD: @local_a = internal addrspace(3) global [10 x i32] undef, align 4
+//.
+// CHECK-NVIDIA: @local_a = internal addrspace(3) global [10 x i32] 
zeroinitializer, align 4
+//.
+int main()
+{
+   int N = 1;
+   int *a = new int[N];
+#pragma omp target data map(tofrom:a[:N])
+   {
+#pragma omp target teams distribute parallel for
+for(int i = 0; i < N; i++)
+{
+  int local_a[10];
+#pragma omp allocate(local_a) allocator(omp_pteam_mem_alloc)
+  for(int j = 0; j < 10; j++)
+   local_a[j] = a[(i + j) % N];
+  a[i] = local_a[0];
+}
+   }
+  return a[17];
+}
+
+#endif
+ NOTE: These prefixes are unused and the list is autogenerated. Do not add 
tests below this line:
+// CHECK-AMD: {{.*}}
+// CHECK-NVIDIA: {{.*}}
Index: clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
===
--- clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
+++ clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
@@ -3351,7 +3351,9 @@
 llvm::Type *VarTy = CGF.ConvertTypeForMem(VD->getType());
 auto *GV = new llvm::GlobalVariable(
 CGM.getModule(), VarTy, /*isConstant=*/false,
-llvm::GlobalValue::InternalLinkage, 
llvm::Constant::getNullValue(VarTy),
+llvm::GlobalValue::InternalLinkage,
+CGM.getTriple().isAMDGCN() ? llvm::UndefValue::get(VarTy)
+   : llvm::Constant::getNullValue(VarTy),
 VD->getName(),
 /*InsertBefore=*/nullptr, llvm::GlobalValue::NotThreadLocal,
 CGM.getContext().getTargetAddressSpace(AS));


Index: clang/test/OpenMP/target_team_variable_codegen.cpp