[PATCH] D76795: [HIP] Change default --gpu-max-threads-per-block value to 1024

Yaxun Liu via Phabricator via cfe-commits Wed, 25 Mar 2020 11:57:38 -0700

yaxunl created this revision.
yaxunl added reviewers: b-sumner, tra.
Herald added subscribers: kerbowa, nhaehnle, jvesely.


This better matches CUDA behavior.


https://reviews.llvm.org/D76795

Files:
  clang/include/clang/Basic/LangOptions.def
  clang/lib/CodeGen/TargetInfo.cpp
  clang/test/CodeGenCUDA/amdgpu-kernel-attrs.cu
  clang/test/CodeGenCUDA/kernel-amdgcn.cu


Index: clang/test/CodeGenCUDA/kernel-amdgcn.cu
===================================================================
--- clang/test/CodeGenCUDA/kernel-amdgcn.cu
+++ clang/test/CodeGenCUDA/kernel-amdgcn.cu
@@ -39,4 +39,4 @@
   launch((void*)D.Empty());
   return 0;
 }
-// CHECK: attributes #[[ATTR]] = {{.*}}"amdgpu-flat-work-group-size"="1,256"
+// CHECK: attributes #[[ATTR]] = {{.*}}"amdgpu-flat-work-group-size"="1,1024"
Index: clang/test/CodeGenCUDA/amdgpu-kernel-attrs.cu
===================================================================
--- clang/test/CodeGenCUDA/amdgpu-kernel-attrs.cu
+++ clang/test/CodeGenCUDA/amdgpu-kernel-attrs.cu
@@ -39,7 +39,7 @@
 // NAMD-NOT: "amdgpu-num-vgpr"
 // NAMD-NOT: "amdgpu-num-sgpr"
 
-// DEFAULT-DAG: attributes [[FLAT_WORK_GROUP_SIZE_DEFAULT]] = 
{{.*}}"amdgpu-flat-work-group-size"="1,256"{{.*}}"uniform-work-group-size"="true"
+// DEFAULT-DAG: attributes [[FLAT_WORK_GROUP_SIZE_DEFAULT]] = 
{{.*}}"amdgpu-flat-work-group-size"="1,1024"{{.*}}"uniform-work-group-size"="true"
 // MAX1024-DAG: attributes [[FLAT_WORK_GROUP_SIZE_DEFAULT]] = 
{{.*}}"amdgpu-flat-work-group-size"="1,1024"
 // CHECK-DAG: attributes [[FLAT_WORK_GROUP_SIZE_32_64]] = 
{{.*}}"amdgpu-flat-work-group-size"="32,64"
 // CHECK-DAG: attributes [[WAVES_PER_EU_2]] = {{.*}}"amdgpu-waves-per-eu"="2"
Index: clang/lib/CodeGen/TargetInfo.cpp
===================================================================
--- clang/lib/CodeGen/TargetInfo.cpp
+++ clang/lib/CodeGen/TargetInfo.cpp
@@ -8119,9 +8119,12 @@
       assert(Max == 0 && "Max must be zero");
   } else if (IsOpenCLKernel || IsHIPKernel) {
     // By default, restrict the maximum size to a value specified by
-    // --gpu-max-threads-per-block=n or its default value.
-    std::string AttrVal =
-        std::string("1,") + 
llvm::utostr(M.getLangOpts().GPUMaxThreadsPerBlock);
+    // --gpu-max-threads-per-block=n or its default value for HIP.
+    const unsigned OpenCLMaxWorkGroupSize = 256;
+    const unsigned MaxWorkGroupSize =
+        IsOpenCLKernel ? OpenCLMaxWorkGroupSize
+                       : M.getLangOpts().GPUMaxThreadsPerBlock;
+    std::string AttrVal = std::string("1,") + llvm::utostr(MaxWorkGroupSize);
     F->addFnAttr("amdgpu-flat-work-group-size", AttrVal);
   }
 
Index: clang/include/clang/Basic/LangOptions.def
===================================================================
--- clang/include/clang/Basic/LangOptions.def
+++ clang/include/clang/Basic/LangOptions.def
@@ -231,7 +231,7 @@
 LANGOPT(CUDADeviceApproxTranscendentals, 1, 0, "using approximate 
transcendental functions")
 LANGOPT(GPURelocatableDeviceCode, 1, 0, "generate relocatable device code")
 LANGOPT(GPUAllowDeviceInit, 1, 0, "allowing device side global init functions 
for HIP")
-LANGOPT(GPUMaxThreadsPerBlock, 32, 256, "default max threads per block for 
kernel launch bounds for HIP")
+LANGOPT(GPUMaxThreadsPerBlock, 32, 1024, "default max threads per block for 
kernel launch bounds for HIP")
 
 LANGOPT(SYCL              , 1, 0, "SYCL")
 LANGOPT(SYCLIsDevice      , 1, 0, "Generate code for SYCL device")

Index: clang/test/CodeGenCUDA/kernel-amdgcn.cu
===================================================================
--- clang/test/CodeGenCUDA/kernel-amdgcn.cu
+++ clang/test/CodeGenCUDA/kernel-amdgcn.cu
@@ -39,4 +39,4 @@
   launch((void*)D.Empty());
   return 0;
 }
-// CHECK: attributes #[[ATTR]] = {{.*}}"amdgpu-flat-work-group-size"="1,256"
+// CHECK: attributes #[[ATTR]] = {{.*}}"amdgpu-flat-work-group-size"="1,1024"
Index: clang/test/CodeGenCUDA/amdgpu-kernel-attrs.cu
===================================================================
--- clang/test/CodeGenCUDA/amdgpu-kernel-attrs.cu
+++ clang/test/CodeGenCUDA/amdgpu-kernel-attrs.cu
@@ -39,7 +39,7 @@
 // NAMD-NOT: "amdgpu-num-vgpr"
 // NAMD-NOT: "amdgpu-num-sgpr"
 
-// DEFAULT-DAG: attributes [[FLAT_WORK_GROUP_SIZE_DEFAULT]] = {{.*}}"amdgpu-flat-work-group-size"="1,256"{{.*}}"uniform-work-group-size"="true"
+// DEFAULT-DAG: attributes [[FLAT_WORK_GROUP_SIZE_DEFAULT]] = {{.*}}"amdgpu-flat-work-group-size"="1,1024"{{.*}}"uniform-work-group-size"="true"
 // MAX1024-DAG: attributes [[FLAT_WORK_GROUP_SIZE_DEFAULT]] = {{.*}}"amdgpu-flat-work-group-size"="1,1024"
 // CHECK-DAG: attributes [[FLAT_WORK_GROUP_SIZE_32_64]] = {{.*}}"amdgpu-flat-work-group-size"="32,64"
 // CHECK-DAG: attributes [[WAVES_PER_EU_2]] = {{.*}}"amdgpu-waves-per-eu"="2"
Index: clang/lib/CodeGen/TargetInfo.cpp
===================================================================
--- clang/lib/CodeGen/TargetInfo.cpp
+++ clang/lib/CodeGen/TargetInfo.cpp
@@ -8119,9 +8119,12 @@
       assert(Max == 0 && "Max must be zero");
   } else if (IsOpenCLKernel || IsHIPKernel) {
     // By default, restrict the maximum size to a value specified by
-    // --gpu-max-threads-per-block=n or its default value.
-    std::string AttrVal =
-        std::string("1,") + llvm::utostr(M.getLangOpts().GPUMaxThreadsPerBlock);
+    // --gpu-max-threads-per-block=n or its default value for HIP.
+    const unsigned OpenCLMaxWorkGroupSize = 256;
+    const unsigned MaxWorkGroupSize =
+        IsOpenCLKernel ? OpenCLMaxWorkGroupSize
+                       : M.getLangOpts().GPUMaxThreadsPerBlock;
+    std::string AttrVal = std::string("1,") + llvm::utostr(MaxWorkGroupSize);
     F->addFnAttr("amdgpu-flat-work-group-size", AttrVal);
   }
 
Index: clang/include/clang/Basic/LangOptions.def
===================================================================
--- clang/include/clang/Basic/LangOptions.def
+++ clang/include/clang/Basic/LangOptions.def
@@ -231,7 +231,7 @@
 LANGOPT(CUDADeviceApproxTranscendentals, 1, 0, "using approximate transcendental functions")
 LANGOPT(GPURelocatableDeviceCode, 1, 0, "generate relocatable device code")
 LANGOPT(GPUAllowDeviceInit, 1, 0, "allowing device side global init functions for HIP")
-LANGOPT(GPUMaxThreadsPerBlock, 32, 256, "default max threads per block for kernel launch bounds for HIP")
+LANGOPT(GPUMaxThreadsPerBlock, 32, 1024, "default max threads per block for kernel launch bounds for HIP")
 
 LANGOPT(SYCL              , 1, 0, "SYCL")
 LANGOPT(SYCLIsDevice      , 1, 0, "Generate code for SYCL device")

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

[PATCH] D76795: [HIP] Change default --gpu-max-threads-per-block value to 1024

Reply via email to