Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

llpc: GFX11 flat/custom parameter loads need to be strict WQM #2817

Merged
merged 1 commit into from
Nov 16, 2023
Merged

llpc: GFX11 flat/custom parameter loads need to be strict WQM #2817

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 4 additions & 3 deletions lgc/patch/PatchInOutImportExport.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1754,9 +1754,10 @@ Value *PatchInOutImportExport::performFsParameterLoad(BuilderBase &builder, Valu
compValue = builder.CreateIntrinsic(Intrinsic::amdgcn_mov_dpp, builder.getInt32Ty(),
{compValue, builder.getInt32(static_cast<unsigned>(dppCtrl)),
builder.getInt32(15), builder.getInt32(15), builder.getTrue()});
// NOTE: Make mov_dpp and its source instructions run in WQM to make sure the mov_dpp could fetch
// correct data from possible inactive lanes.
compValue = builder.CreateIntrinsic(Intrinsic::amdgcn_wqm, builder.getInt32Ty(), compValue);
// NOTE: mov_dpp must run in strict WQM to access lanes potentially inactive with normal exec/WQM.
// lds_param_load always runs in strict WQM, but exec/WQM may not match this due to discards or divergence.
// Ideally we would use the FI bit on the mov_dpp, but there is currently no backend support.
compValue = builder.CreateIntrinsic(Intrinsic::amdgcn_strict_wqm, builder.getInt32Ty(), compValue);
compValue = builder.CreateBitCast(compValue, builder.getFloatTy());
} else {
Value *args[] = {
Expand Down
34 changes: 34 additions & 0 deletions llpc/test/shaderdb/gfx11/FlatParamDpp.frag
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,34 @@
// Check that flat parameter load uses DPP in strict WQM

// RUN: amdllpc %gfxip --v %s |\
// RUN: FileCheck %s --check-prefix=CHECK
//
// CHECK-LABEL: {{^}}// LLPC pipeline patching results
// CHECK: call void @llvm.amdgcn.kill(i1 false)
// CHECK: [[P0:%.*]] = call float @llvm.amdgcn.lds.param.load(i32 immarg 2, i32 immarg 2, i32 %PrimMask)
// CHECK: [[P1:%.*]] = bitcast float [[P0]] to i32
// CHECK: [[P2:%.*]] = call i32 @llvm.amdgcn.mov.dpp.i32(i32 [[P1]], i32 0, i32 15, i32 15, i1 true)
// CHECK: [[P3:%.*]] = call i32 @llvm.amdgcn.strict.wqm.i32(i32 [[P2]])
// CHECK-LABEL: {{^}}===== AMDLLPC SUCCESS =====

#version 450

layout (location = 0) in vec3 texCoordIn;
layout (location = 1) in float discardPixel;
layout (location = 2) flat in vec4 p0;

layout (binding = 0) uniform sampler2D image1;
layout (binding = 1) uniform sampler s0;
layout (binding = 2) uniform textureCube t0;

layout (location = 0) out vec4 fragColor;

void main() {
fragColor = texture(image1, texCoordIn.xy);

if (discardPixel > 0.0)
discard;

float lod = p0.z;
fragColor += textureLod(samplerCube(t0, s0), texCoordIn, lod);
}
Loading