llpc: disable spilling of SGPR spill VGPRs

Add function attribute amdgpu-prealloc-sgpr-spill-vgprs to prevent spilling of SGPR spill VGPRs. This occurs in shaders with extremely high VGPR pressure, e.g. large RT shaders, and can yield unpredictable results. This is an interim workaround and should be removed when backend support for these spills is fully implemented.
GPUOpen-Drivers · Nov 13, 2023 · bfd2d2c · bfd2d2c
1 parent 4320b9c
commit bfd2d2c
Show file tree

Hide file tree

Showing 4 changed files with 9 additions and 4 deletions.
diff --git a/lgc/patch/PatchEntryPointMutate.cpp b/lgc/patch/PatchEntryPointMutate.cpp
@@ -1449,6 +1449,10 @@ void PatchEntryPointMutate::setFuncAttrs(Function *entryPoint) {
   builder.addAttribute("amdgpu-memory-bound", shaderOptions->favorLatencyHiding ? "true" : "false");
   builder.addAttribute("amdgpu-wave-limiter", "false");
 
+  // Prevent spilling of VGPRs holding SGPR spills as this can have undefined behaviour in callee functions.
+  // Note: this is an intermediate workaround and should be removed when backend support is complete.
+  builder.addAttribute("amdgpu-prealloc-sgpr-spill-vgprs");
+
   entryPoint->addFnAttrs(builder);
 
   // NOTE: Remove "readnone" attribute for entry-point. If GS is empty, this attribute will allow

diff --git a/lgc/test/Transforms/CpsLowering/cps-stack-lowering.lgc b/lgc/test/Transforms/CpsLowering/cps-stack-lowering.lgc
@@ -237,7 +237,7 @@ define void @test.2({ ptr addrspace(32) } %state) !lgc.cps !{i32 1} !lgc.shaders
 ;
 ;.
 ; CHECK: attributes #[[ATTR0:[0-9]+]] = { noreturn }
-; CHECK: attributes #[[ATTR1]] = { memory(readwrite) "amdgpu-memory-bound"="false" "amdgpu-unroll-threshold"="700" "amdgpu-wave-limiter"="false" }
+; CHECK: attributes #[[ATTR1]] = { memory(readwrite) "amdgpu-memory-bound"="false" "amdgpu-prealloc-sgpr-spill-vgprs" "amdgpu-unroll-threshold"="700" "amdgpu-wave-limiter"="false" }
 ; CHECK: attributes #[[ATTR2:[0-9]+]] = { convergent nocallback nofree nounwind willreturn memory(none) }
 ; CHECK: attributes #[[ATTR3:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
 ; CHECK: attributes #[[ATTR4:[0-9]+]] = { convergent nocallback nofree nounwind speculatable willreturn memory(none) }

diff --git a/llpc/test/shaderdb/general/PipelineCs_LdsSpillLimitDwordsOption.pipe b/llpc/test/shaderdb/general/PipelineCs_LdsSpillLimitDwordsOption.pipe
@@ -19,5 +19,5 @@ options.ldsSpillLimitDwords = 1024
 ; CHECK-NEXT:    ret void
 ;
 ;.
-; CHECK: attributes #[[ATTR0:[0-9]+]] = { nounwind memory(readwrite) "amdgpu-flat-work-group-size"="66,66" "amdgpu-lds-spill-limit-dwords"="1024" "amdgpu-memory-bound"="false" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-unroll-threshold"="700" "amdgpu-wave-limiter"="false" "amdgpu-work-group-info-arg-no"="1" "denormal-fp-math-f32"="preserve-sign" "target-features"=",+wavefrontsize64,+cumode,+enable-flat-scratch" }
+; CHECK: attributes #[[ATTR0:[0-9]+]] = { nounwind memory(readwrite) "amdgpu-flat-work-group-size"="66,66" "amdgpu-lds-spill-limit-dwords"="1024" "amdgpu-memory-bound"="false" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-prealloc-sgpr-spill-vgprs" "amdgpu-unroll-threshold"="700" "amdgpu-wave-limiter"="false" "amdgpu-work-group-info-arg-no"="1" "denormal-fp-math-f32"="preserve-sign" "target-features"=",+wavefrontsize64,+cumode,+enable-flat-scratch" }
 ;.
diff --git a/llpc/test/shaderdb/general/TestWorkgroupIdOpt.comp b/llpc/test/shaderdb/general/TestWorkgroupIdOpt.comp
@@ -28,7 +28,8 @@ void main()
 // CHECK-NEXT:    ret void
 //
 //.
-// CHECK: attributes #[[ATTR0]] = { nounwind memory(readwrite) "amdgpu-flat-work-group-size"="256,256" "amdgpu-memory-bound"="false" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-unroll-threshold"="700" "amdgpu-wave-limiter"="false" "amdgpu-work-group-info-arg-no"="3" "denormal-fp-math-f32"="preserve-sign" "target-features"=",+wavefrontsize64,+cumode,+enable-flat-scratch" }
-// CHECK: attributes #[[ATTR1:[0-9]+]] = { nounwind willreturn memory(none) }
+// CHECK: attributes #[[ATTR0]] = { nounwind memory(readwrite) "amdgpu-flat-work-group-size"="256,256" "amdgpu-memory-bound"="false" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-prealloc-sgpr-spill-vgprs" "amdgpu-unroll-threshold"="700" "amdgpu-wave-limiter"="false" "amdgpu-work-group-info-arg-no"="3" "denormal-fp-math-f32"="preserve-sign" "target-features"=",+wavefrontsize64,+cumode,+enable-flat-scratch" }
+// CHECK: attributes #[[ATTR1:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
+// CHECK: attributes #[[ATTR2:[0-9]+]] = { nounwind willreturn memory(none) }
 // CHECK: attributes #[[ATTR3:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(write) }
 //.