Skip to content

[DirectX] Calculate resource binding offsets using the lower bound #117303

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

Merged
merged 2 commits into from
Nov 25, 2024
Merged

[DirectX] Calculate resource binding offsets using the lower bound #117303

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
05f412f
[HLSL] Get the index for resource bindings from the slot
bogner Nov 22, 2024
2360f3a
Move the fix to the DirectX backend
bogner Nov 22, 2024
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
2 changes: 1 addition & 1 deletion clang/lib/CodeGen/CGHLSLRuntime.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -588,7 +588,7 @@ llvm::Function *CGHLSLRuntime::createResourceBindingInitFn() {
auto *Slot = llvm::ConstantInt::get(CGM.IntTy, RBA->getSlotNumber());
// FIXME: resource arrays are not yet implemented
auto *Range = llvm::ConstantInt::get(CGM.IntTy, 1);
auto *Index = llvm::ConstantInt::get(CGM.IntTy, 0);
auto *Index = llvm::ConstantInt::get(CGM.IntTy, RBA->getSlotNumber());
Copy link
Contributor

@s-perron s-perron Nov 22, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is not the way I implemented this for SPIR-V. This assumes the Slot and Index as interwined, which they are in DXIL, but not in SPIR-V.

In SPIR-V, the Slot is the binding number, and all elements of the resource array are accessed through the same binding number. The element does not get their own binding number. The binding and index are independent.

Would it be possible to keep this as is, and have the DXIL backend add the slot number to the index?

Copy link
Collaborator

@llvm-beanz llvm-beanz Nov 22, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It might be better to just diverge this by the target. That way the intermediate IR reflects the index correctly.

Copy link
Contributor

@s-perron s-perron Nov 22, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Diverging is fine. We will have to diverge when identifying the slots anyway. Because we keep samplers and textures in the same "space", we will need a more complicated way of determining the right space and slot. Plus we will eventually add the vk::binding attribute which will be another consideration.

Copy link
Contributor Author

@bogner bogner Nov 22, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we opt to diverge this by target we should probably not use the matching intrinsic names between dxil and spirv in this case, as that could end up confusing. It might be best to defer that until we're dealing with textures or the binding attribute so we can make sure we're making the right decisions on how to split it up.

I think it would be fine to just do this indexing in the backend, since any index that's less than the slot should be invalid anyway. I'll try that out and report back.

Copy link
Contributor Author

@bogner bogner Nov 22, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I've changed this to do the translation in the backend.

// FIXME: NonUniformResourceIndex bit is not yet implemented
auto *NonUniform = llvm::ConstantInt::get(Int1Ty, false);
llvm::Value *Args[] = {Space, Slot, Range, Index, NonUniform};
Expand Down
2 changes: 1 addition & 1 deletion clang/test/CodeGenHLSL/builtins/RWBuffer-constructor.hlsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,7 @@ RWBuffer<float> Buf : register(u5, space3);

// CHECK: define internal void @_init_resource_bindings() {
// CHECK-NEXT: entry:
// CHECK-DXIL-NEXT: %Buf_h = call target("dx.TypedBuffer", float, 1, 0, 0) @llvm.dx.handle.fromBinding.tdx.TypedBuffer_f32_1_0_0t(i32 3, i32 5, i32 1, i32 0, i1 false)
// CHECK-DXIL-NEXT: %Buf_h = call target("dx.TypedBuffer", float, 1, 0, 0) @llvm.dx.handle.fromBinding.tdx.TypedBuffer_f32_1_0_0t(i32 3, i32 5, i32 1, i32 5, i1 false)
// CHECK-DXIL-NEXT: store target("dx.TypedBuffer", float, 1, 0, 0) %Buf_h, ptr @Buf, align 4
// CHECK-SPIRV-NEXT: %Buf_h = call target("dx.TypedBuffer", float, 1, 0, 0) @llvm.spv.handle.fromBinding.tdx.TypedBuffer_f32_1_0_0t(i32 3, i32 5, i32 1, i32 0, i1 false)
// CHECK-SPIRV-NEXT: store target("dx.TypedBuffer", float, 1, 0, 0) %Buf_h, ptr @Buf, align 4
20 changes: 10 additions & 10 deletions clang/test/CodeGenHLSL/builtins/StructuredBuffers-constructors.hlsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,24 +37,24 @@ RasterizerOrderedStructuredBuffer<float> Buf5 : register(u1, space2);

// CHECK: define internal void @_init_resource_bindings() {
// CHECK-NEXT: entry:
// CHECK-DXIL-NEXT: %Buf_h = call target("dx.RawBuffer", float, 0, 0) @llvm.dx.handle.fromBinding.tdx.RawBuffer_f32_0_0t(i32 0, i32 10, i32 1, i32 0, i1 false)
// CHECK-DXIL-NEXT: %Buf_h = call target("dx.RawBuffer", float, 0, 0) @llvm.dx.handle.fromBinding.tdx.RawBuffer_f32_0_0t(i32 0, i32 10, i32 1, i32 10, i1 false)
// CHECK-DXIL-NEXT: store target("dx.RawBuffer", float, 0, 0) %Buf_h, ptr @Buf, align 4
// CHECK-DXIL-NEXT: %Buf2_h = call target("dx.RawBuffer", float, 1, 0) @llvm.dx.handle.fromBinding.tdx.RawBuffer_f32_1_0t(i32 1, i32 5, i32 1, i32 0, i1 false)
// CHECK-DXIL-NEXT: %Buf2_h = call target("dx.RawBuffer", float, 1, 0) @llvm.dx.handle.fromBinding.tdx.RawBuffer_f32_1_0t(i32 1, i32 5, i32 1, i32 5, i1 false)
// CHECK-DXIL-NEXT: store target("dx.RawBuffer", float, 1, 0) %Buf2_h, ptr @Buf2, align 4
// CHECK-DXIL-NEXT: %Buf3_h = call target("dx.RawBuffer", float, 1, 0) @llvm.dx.handle.fromBinding.tdx.RawBuffer_f32_1_0t(i32 0, i32 3, i32 1, i32 0, i1 false)
// CHECK-DXIL-NEXT: %Buf3_h = call target("dx.RawBuffer", float, 1, 0) @llvm.dx.handle.fromBinding.tdx.RawBuffer_f32_1_0t(i32 0, i32 3, i32 1, i32 3, i1 false)
// CHECK-DXIL-NEXT: store target("dx.RawBuffer", float, 1, 0) %Buf3_h, ptr @Buf3, align 4
// CHECK-DXIL-NEXT: %Buf4_h = call target("dx.RawBuffer", float, 1, 0) @llvm.dx.handle.fromBinding.tdx.RawBuffer_f32_1_0t(i32 0, i32 4, i32 1, i32 0, i1 false)
// CHECK-DXIL-NEXT: %Buf4_h = call target("dx.RawBuffer", float, 1, 0) @llvm.dx.handle.fromBinding.tdx.RawBuffer_f32_1_0t(i32 0, i32 4, i32 1, i32 4, i1 false)
// CHECK-DXIL-NEXT: store target("dx.RawBuffer", float, 1, 0) %Buf4_h, ptr @Buf4, align 4
// CHECK-DXIL-NEXT: %Buf5_h = call target("dx.RawBuffer", float, 1, 1) @llvm.dx.handle.fromBinding.tdx.RawBuffer_f32_1_1t(i32 2, i32 1, i32 1, i32 0, i1 false)
// CHECK-DXIL-NEXT: %Buf5_h = call target("dx.RawBuffer", float, 1, 1) @llvm.dx.handle.fromBinding.tdx.RawBuffer_f32_1_1t(i32 2, i32 1, i32 1, i32 1, i1 false)
// CHECK-DXIL-NEXT: store target("dx.RawBuffer", float, 1, 1) %Buf5_h, ptr @Buf5, align 4

// CHECK-SPIRV-NEXT: %Buf_h = call target("dx.RawBuffer", float, 0, 0) @llvm.spv.handle.fromBinding.tdx.RawBuffer_f32_0_0t(i32 0, i32 10, i32 1, i32 0, i1 false)
// CHECK-SPIRV-NEXT: %Buf_h = call target("dx.RawBuffer", float, 0, 0) @llvm.spv.handle.fromBinding.tdx.RawBuffer_f32_0_0t(i32 0, i32 10, i32 1, i32 10, i1 false)
// CHECK-SPIRV-NEXT: store target("dx.RawBuffer", float, 0, 0) %Buf_h, ptr @Buf", align 4
// CHECK-SPIRV-NEXT: %Buf2_h = call target("dx.RawBuffer", float, 1, 0) @llvm.spv.handle.fromBinding.tdx.RawBuffer_f32_1_0t(i32 1, i32 5, i32 1, i32 0, i1 false)
// CHECK-SPIRV-NEXT: %Buf2_h = call target("dx.RawBuffer", float, 1, 0) @llvm.spv.handle.fromBinding.tdx.RawBuffer_f32_1_0t(i32 1, i32 5, i32 1, i32 5, i1 false)
// CHECK-SPIRV-NEXT: store target("dx.RawBuffer", float, 1, 0) %Buf2_h, ptr @Buf2", align 4
// CHECK-SPIRV-NEXT: %Buf3_h = call target("dx.RawBuffer", float, 0, 0) @llvm.spv.handle.fromBinding.tdx.RawBuffer_f32_1_0t(i32 0, i32 3, i32 1, i32 0, i1 false)
// CHECK-SPIRV-NEXT: %Buf3_h = call target("dx.RawBuffer", float, 0, 0) @llvm.spv.handle.fromBinding.tdx.RawBuffer_f32_1_0t(i32 0, i32 3, i32 1, i32 3, i1 false)
// CHECK-SPIRV-NEXT: store target("dx.RawBuffer", float, 0, 0) %Buf3_h, ptr @Buf3, align 4
// CHECK-SPIRV-NEXT: %Buf4_h = call target("dx.RawBuffer", float, 1, 0) @llvm.spv.handle.fromBinding.tdx.RawBuffer_f32_1_0t(i32 0, i32 4, i32 1, i32 0, i1 false)
// CHECK-SPIRV-NEXT: %Buf4_h = call target("dx.RawBuffer", float, 1, 0) @llvm.spv.handle.fromBinding.tdx.RawBuffer_f32_1_0t(i32 0, i32 4, i32 1, i32 4, i1 false)
// CHECK-SPIRV-NEXT: store target("dx.RawBuffer", float, 1, 0) %Buf4_h, ptr @Buf4, align 4
// CHECK-SPIRV-NEXT: %Buf5_h = call target("dx.RawBuffer", float, 1, 1) @llvm.spv.handle.fromBinding.tdx.RawBuffer_f32_1_1t(i32 2, i32 1, i32 1, i32 0, i1 false)
// CHECK-SPIRV-NEXT: %Buf5_h = call target("dx.RawBuffer", float, 1, 1) @llvm.spv.handle.fromBinding.tdx.RawBuffer_f32_1_1t(i32 2, i32 1, i32 1, i32 1, i1 false)
// CHECK-SPIRV-NEXT: store target("dx.RawBuffer", float, 1, 1) %Buf5_h, ptr @Buf5, align 4
19 changes: 19 additions & 0 deletions clang/test/CodeGenHLSL/resource-bindings.hlsl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,19 @@
// RUN: %clang_cc1 -triple dxil--shadermodel6.6-compute -x hlsl -finclude-default-header -emit-llvm -o - %s | FileCheck %s

// CHECK: define internal void @_init_resource_bindings() {

// CHECK: %U0S0_h = call target("dx.TypedBuffer", <4 x float>, 1, 0, 0) @llvm.dx.handle.fromBinding.tdx.TypedBuffer_v4f32_1_0_0t(i32 0, i32 0, i32 1, i32 0, i1 false)
RWBuffer<float4> U0S0 : register(u0);

// CHECK: %U5S3_h = call target("dx.TypedBuffer", float, 1, 0, 0) @llvm.dx.handle.fromBinding.tdx.TypedBuffer_f32_1_0_0t(i32 3, i32 5, i32 1, i32 5, i1 false)
RWBuffer<float> U5S3 : register(u5, space3);

// CHECK: %T2S2_h = call target("dx.RawBuffer", i32, 0, 0) @llvm.dx.handle.fromBinding.tdx.RawBuffer_i32_0_0t(i32 2, i32 2, i32 1, i32 2, i1 false)
StructuredBuffer<int> T2S2 : register(t2, space2);
struct S {
float4 f;
int i;
};

// CHECK: %T3S0_h = call target("dx.RawBuffer", %struct.S = type { <4 x float>, i32, [12 x i8] }, 0, 0) @llvm.dx.handle.fromBinding.tdx.RawBuffer_s_struct.Ss_0_0t(i32 0, i32 3, i32 1, i32 3, i1 false)
StructuredBuffer<S> T3S0 : register(t3);
Loading