AMDGPU: Mark features as read-only only when target supports them

  "image-insts", "gws", and "vmem-to-lds-load-insts" are these features
that are potentially marked as read-only.

Author: changpeng

clang/lib/Basic/Targets/AMDGPU.cpp

@@ -266,8 +266,13 @@ AMDGPUTargetInfo::AMDGPUTargetInfo(const llvm::Triple &Triple,
 
   MaxAtomicPromoteWidth = MaxAtomicInlineWidth = 64;
   CUMode = !(GPUFeatures & llvm::AMDGPU::FEATURE_WGP);
-  for (auto F : {"image-insts", "vmem-to-lds-load-insts"})
-    ReadOnlyFeatures.insert(F);
+  // Mark these features as read-only when the target supports them.
+  llvm::StringMap<bool> Features;
+  llvm::AMDGPU::fillAMDGPUFeatureMap(Opts.CPU, Triple, Features);
+  for (auto F : {"image-insts", "gws", "vmem-to-lds-load-insts"}) {
+    if (Features[F] == true)
+      ReadOnlyFeatures.insert(F);
+  }
   HalfArgsAndReturns = true;
 }
 

clang/test/CodeGen/link-builtin-bitcode.c

@@ -43,7 +43,7 @@ int bar() { return no_attr() + attr_in_target() + attr_not_in_target() + attr_in
 // CHECK-LABEL: @attr_incompatible
 // CHECK-SAME: () #[[ATTR_INCOMPATIBLE:[0-9]+]] {
 
-// CHECK: attributes #[[ATTR_BAR]] = { {{.*}} "target-cpu"="gfx90a" "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-fadd-rtn-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gws,+mai-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64" }
+// CHECK: attributes #[[ATTR_BAR]] = { {{.*}} "target-cpu"="gfx90a" "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-fadd-rtn-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64" }
 // CHECK: attributes #[[ATTR_COMPATIBLE]] = { {{.*}} "target-cpu"="gfx90a" "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-fadd-rtn-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gws,+image-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+vmem-to-lds-load-insts,+wavefrontsize64" }
 // CHECK: attributes #[[ATTR_EXTEND]] = { {{.*}} "target-cpu"="gfx90a" "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-fadd-rtn-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+extended-image-insts,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gws,+image-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+vmem-to-lds-load-insts,+wavefrontsize64" }
 // CHECK: attributes #[[ATTR_INCOMPATIBLE]] = { {{.*}} "target-cpu"="gfx90a" "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-fadd-rtn-insts,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+gfx8-insts,+gfx90a-insts,+gws,+image-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+vmem-to-lds-load-insts,+wavefrontsize64,-gfx9-insts" }

clang/test/CodeGenOpenCL/amdgpu-enqueue-kernel.cl

@@ -816,12 +816,12 @@ kernel void test_target_features_kernel(global int *i) {
 // NOCPU: attributes #[[ATTR10]] = { convergent nounwind }
 //.
 // GFX900: attributes #[[ATTR0:[0-9]+]] = { "objc_arc_inert" }
-// GFX900: attributes #[[ATTR1]] = { convergent norecurse nounwind "denormal-fp-math-f32"="preserve-sign,preserve-sign" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="gfx900" "target-features"="+16-bit-insts,+ci-insts,+dpp,+gfx8-insts,+gfx9-insts,+gws,+s-memrealtime,+s-memtime-inst,+wavefrontsize64,-sram-ecc" }
-// GFX900: attributes #[[ATTR2]] = { convergent norecurse nounwind "amdgpu-flat-work-group-size"="1,256" "denormal-fp-math-f32"="preserve-sign,preserve-sign" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="gfx900" "target-features"="+16-bit-insts,+ci-insts,+dpp,+gfx8-insts,+gfx9-insts,+gws,+s-memrealtime,+s-memtime-inst,+wavefrontsize64,-sram-ecc" "uniform-work-group-size"="false" }
-// GFX900: attributes #[[ATTR3]] = { alwaysinline convergent norecurse nounwind "amdgpu-flat-work-group-size"="1,256" "denormal-fp-math-f32"="preserve-sign,preserve-sign" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="gfx900" "target-features"="+16-bit-insts,+ci-insts,+dpp,+gfx8-insts,+gfx9-insts,+gws,+s-memrealtime,+s-memtime-inst,+wavefrontsize64,-sram-ecc" }
+// GFX900: attributes #[[ATTR1]] = { convergent norecurse nounwind "denormal-fp-math-f32"="preserve-sign,preserve-sign" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="gfx900" "target-features"="+16-bit-insts,+ci-insts,+dpp,+gfx8-insts,+gfx9-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64,-sram-ecc" }
+// GFX900: attributes #[[ATTR2]] = { convergent norecurse nounwind "amdgpu-flat-work-group-size"="1,256" "denormal-fp-math-f32"="preserve-sign,preserve-sign" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="gfx900" "target-features"="+16-bit-insts,+ci-insts,+dpp,+gfx8-insts,+gfx9-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64,-sram-ecc" "uniform-work-group-size"="false" }
+// GFX900: attributes #[[ATTR3]] = { alwaysinline convergent norecurse nounwind "amdgpu-flat-work-group-size"="1,256" "denormal-fp-math-f32"="preserve-sign,preserve-sign" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="gfx900" "target-features"="+16-bit-insts,+ci-insts,+dpp,+gfx8-insts,+gfx9-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64,-sram-ecc" }
 // GFX900: attributes #[[ATTR4:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
 // GFX900: attributes #[[ATTR5:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
-// GFX900: attributes #[[ATTR6]] = { convergent nounwind "denormal-fp-math-f32"="preserve-sign,preserve-sign" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="gfx900" "target-features"="+16-bit-insts,+ci-insts,+dpp,+gfx8-insts,+gfx9-insts,+gws,+s-memrealtime,+s-memtime-inst,+wavefrontsize64,-sram-ecc" }
+// GFX900: attributes #[[ATTR6]] = { convergent nounwind "denormal-fp-math-f32"="preserve-sign,preserve-sign" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="gfx900" "target-features"="+16-bit-insts,+ci-insts,+dpp,+gfx8-insts,+gfx9-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64,-sram-ecc" }
 // GFX900: attributes #[[ATTR7:[0-9]+]] = { nocallback nofree nosync nounwind willreturn }
 // GFX900: attributes #[[ATTR8]] = { convergent nounwind }
 // GFX900: attributes #[[ATTR9]] = { nounwind }

clang/test/CodeGenOpenCL/amdgpu-features-readonly.cl

@@ -1,6 +1,6 @@
-// RUN: %clang_cc1 -triple amdgcn -target-feature +vmem-to-lds-load-insts -o /dev/null %s 2>&1 \
-// RUN:   | FileCheck --check-prefix=READONLY %s
+// RUN: %clang_cc1 -triple amdgcn -target-cpu gfx942 -target-feature +gws -o /dev/null %s 2>&1 \
+// RUN:   | FileCheck --check-prefix=GWS %s
 
-// READONLY: warning: feature flag '+vmem-to-lds-load-insts' is ignored since the feature is read only [-Winvalid-command-line-argument]
+// GWS: warning: feature flag '+gws' is ignored since the feature is read only [-Winvalid-command-line-argument]
 
 kernel void test() {}