[DirectX] Implement Max64UAVs shader flag analysis #136229
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-backend-directx Author: Deric C. (Icohedron) ChangesFixes #114553 This implementation replicates the behavior of DXC in setting the The behavior of how UAV (resource) arrays are counted differs based on Shader Model version:
I initially thought the complete implementation of this analysis may be blocked by the resource arrays implementation, but it seems that it is not the case, as the Full diff: https://github.com/llvm/llvm-project/pull/136229.diff 4 Files Affected:
diff --git a/llvm/lib/Target/DirectX/DXILShaderFlags.cpp b/llvm/lib/Target/DirectX/DXILShaderFlags.cpp
index 1331b0d1d852b..282b9dcf6de2b 100644
--- a/llvm/lib/Target/DirectX/DXILShaderFlags.cpp
+++ b/llvm/lib/Target/DirectX/DXILShaderFlags.cpp
@@ -255,6 +255,16 @@ void ModuleShaderFlags::initialize(Module &M, DXILResourceTypeMap &DRTM,
EntryFunProps.Entry->getContext().diagnose(DiagnosticInfoUnsupported(
*(EntryFunProps.Entry), "Inconsistent optnone attribute "));
}
+
+ // Set the Max64UAVs flag if the number of UAVs is > 8
+ uint32_t NumUAVs = 0;
+ for (auto &UAV : DRM.uavs())
+ if (MMDI.DXILVersion < VersionTuple(1, 6))
+ NumUAVs++;
+ else // MMDI.DXILVersion >= VersionTuple(1, 6)
+ NumUAVs += UAV.getBinding().Size;
+ if (NumUAVs > 8)
+ CombinedSFMask.Max64UAVs = true;
}
void ComputedShaderFlags::print(raw_ostream &OS) const {
diff --git a/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs-array-sm6_5.ll b/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs-array-sm6_5.ll
new file mode 100644
index 0000000000000..5b978d67866be
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs-array-sm6_5.ll
@@ -0,0 +1,33 @@
+; RUN: opt -S --passes="print-dx-shader-flags" 2>&1 %s | FileCheck %s
+
+; This test makes sure that resource arrays only add 1 to the count of the
+; number of UAVs for setting the shader flag '64 UAV slots' when the shader
+; model version is < 6.6
+
+; Note: there is no feature flag here (only a module flag), so we don't have an
+; object test.
+
+target triple = "dxil-pc-shadermodel6.5-library"
+
+; CHECK: Combined Shader Flags for Module
+; CHECK-NEXT: Shader Flags Value: 0x00000000
+
+; CHECK-NOT: Note: shader requires additional functionality:
+; CHECK-NOT: 64 UAV slots
+
+; CHECK: Function test : 0x00000000
+define void @test() "hlsl.export" {
+ ; RWBuffer<float> Buf : register(u0, space0)
+ %buf0 = call target("dx.TypedBuffer", float, 1, 0, 1)
+ @llvm.dx.resource.handlefrombinding.tdx.TypedBuffer_f32_1_0t(
+ i32 0, i32 0, i32 1, i32 0, i1 false)
+
+ ; RWBuffer<float> Buf[8] : register(u1, space0)
+ %buf1 = call target("dx.TypedBuffer", float, 1, 0, 1)
+ @llvm.dx.resource.handlefrombinding.tdx.TypedBuffer_f32_1_0t(
+ i32 0, i32 1, i32 8, i32 0, i1 false)
+ ret void
+}
+
+!llvm.module.flags = !{!0}
+!0 = !{i32 1, !"dx.resmayalias", i32 1}
diff --git a/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs-array-sm6_6.ll b/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs-array-sm6_6.ll
new file mode 100644
index 0000000000000..4b901a78e6ea4
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs-array-sm6_6.ll
@@ -0,0 +1,33 @@
+; RUN: opt -S --passes="print-dx-shader-flags" 2>&1 %s | FileCheck %s
+
+; This test makes sure that resource arrays sizes are accounted for when
+; counting the number of UAVs for setting the shader flag '64 UAV slots' when
+; the shader model version is >= 6.6
+
+; Note: there is no feature flag here (only a module flag), so we don't have an
+; object test.
+
+target triple = "dxil-pc-shadermodel6.6-library"
+
+; CHECK: Combined Shader Flags for Module
+; CHECK-NEXT: Shader Flags Value: 0x00008000
+
+; CHECK: Note: shader requires additional functionality:
+; CHECK: 64 UAV slots
+
+; CHECK: Function test : 0x00000000
+define void @test() "hlsl.export" {
+ ; RWBuffer<float> Buf : register(u0, space0)
+ %buf0 = call target("dx.TypedBuffer", float, 1, 0, 1)
+ @llvm.dx.resource.handlefrombinding.tdx.TypedBuffer_f32_1_0t(
+ i32 0, i32 0, i32 1, i32 0, i1 false)
+
+ ; RWBuffer<float> Buf[8] : register(u1, space0)
+ %buf1 = call target("dx.TypedBuffer", float, 1, 0, 1)
+ @llvm.dx.resource.handlefrombinding.tdx.TypedBuffer_f32_1_0t(
+ i32 0, i32 1, i32 8, i32 0, i1 false)
+ ret void
+}
+
+!llvm.module.flags = !{!0}
+!0 = !{i32 1, !"dx.resmayalias", i32 1}
diff --git a/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs.ll b/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs.ll
new file mode 100644
index 0000000000000..c002ff2851452
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs.ll
@@ -0,0 +1,60 @@
+; RUN: opt -S --passes="print-dx-shader-flags" 2>&1 %s | FileCheck %s
+
+; This test makes sure that the shader flag '64 UAV slots' is set when there are
+; more than 8 UAVs in the module.
+
+; Note: there is no feature flag here (only a module flag), so we don't have an
+; object test.
+
+target triple = "dxil-pc-shadermodel6.7-library"
+
+; CHECK: Combined Shader Flags for Module
+; CHECK-NEXT: Shader Flags Value: 0x00008000
+
+; CHECK: Note: shader requires additional functionality:
+; CHECK: 64 UAV slots
+
+; Note: 64 UAV slots does not get set per-function
+; CHECK: Function test : 0x00000000
+define void @test() "hlsl.export" {
+ ; RWBuffer<float> Buf : register(u0, space0)
+ %buf0 = call target("dx.TypedBuffer", float, 1, 0, 1)
+ @llvm.dx.resource.handlefrombinding.tdx.TypedBuffer_f32_1_0t(
+ i32 0, i32 0, i32 1, i32 0, i1 false)
+ ; RWBuffer<float> Buf : register(u1, space0)
+ %buf1 = call target("dx.TypedBuffer", float, 1, 0, 1)
+ @llvm.dx.resource.handlefrombinding.tdx.TypedBuffer_f32_1_0t(
+ i32 0, i32 1, i32 1, i32 0, i1 false)
+ ; RWBuffer<float> Buf : register(u2, space0)
+ %buf2 = call target("dx.TypedBuffer", float, 1, 0, 1)
+ @llvm.dx.resource.handlefrombinding.tdx.TypedBuffer_f32_1_0t(
+ i32 0, i32 2, i32 1, i32 0, i1 false)
+ ; RWBuffer<float> Buf : register(u3, space0)
+ %buf3 = call target("dx.TypedBuffer", float, 1, 0, 1)
+ @llvm.dx.resource.handlefrombinding.tdx.TypedBuffer_f32_1_0t(
+ i32 0, i32 3, i32 1, i32 0, i1 false)
+ ; RWBuffer<float> Buf : register(u4, space0)
+ %buf4 = call target("dx.TypedBuffer", float, 1, 0, 1)
+ @llvm.dx.resource.handlefrombinding.tdx.TypedBuffer_f32_1_0t(
+ i32 0, i32 4, i32 1, i32 0, i1 false)
+ ; RWBuffer<float> Buf : register(u5, space0)
+ %buf5 = call target("dx.TypedBuffer", float, 1, 0, 1)
+ @llvm.dx.resource.handlefrombinding.tdx.TypedBuffer_f32_1_0t(
+ i32 0, i32 5, i32 1, i32 0, i1 false)
+ ; RWBuffer<float> Buf : register(u6, space0)
+ %buf6 = call target("dx.TypedBuffer", float, 1, 0, 1)
+ @llvm.dx.resource.handlefrombinding.tdx.TypedBuffer_f32_1_0t(
+ i32 0, i32 6, i32 1, i32 0, i1 false)
+ ; RWBuffer<float> Buf : register(u7, space0)
+ %buf7 = call target("dx.TypedBuffer", float, 1, 0, 1)
+ @llvm.dx.resource.handlefrombinding.tdx.TypedBuffer_f32_1_0t(
+ i32 0, i32 7, i32 1, i32 0, i1 false)
+ ; RWBuffer<float> Buf : register(u8, space0)
+ %buf8 = call target("dx.TypedBuffer", float, 1, 0, 1)
+ @llvm.dx.resource.handlefrombinding.tdx.TypedBuffer_f32_1_0t(
+ i32 0, i32 8, i32 1, i32 0, i1 false)
+ ret void
+}
+
+!llvm.module.flags = !{!0}
+!0 = !{i32 1, !"dx.resmayalias", i32 1}
|
farzonl
approved these changes
Apr 18, 2025
IanWood1
pushed a commit
to IanWood1/llvm-project
that referenced
this pull request
May 6, 2025
Fixes [llvm#114553](llvm#114553) This implementation replicates the behavior of DXC in setting the `m_b64UAVs` flag: the `Max64UAVs` DXIL module flag is set in the presence of more than 8 UAVs in a DXIL module. The behavior of how UAV (resource) arrays are counted differs based on Shader Model version: - If Shader Model < 6.6, then a UAV array counts as a single UAV regardless of its range size - if Shader Model >= 6.6, then a UAV array contributes its range size to the total number of UAVs I initially thought the complete implementation of this analysis may be blocked by the resource arrays implementation, but it seems that it is not the case, as the `@llvm.dx.resource.handle*` already includes a range size argument.
IanWood1
pushed a commit
to IanWood1/llvm-project
that referenced
this pull request
May 6, 2025
Fixes [llvm#114553](llvm#114553) This implementation replicates the behavior of DXC in setting the `m_b64UAVs` flag: the `Max64UAVs` DXIL module flag is set in the presence of more than 8 UAVs in a DXIL module. The behavior of how UAV (resource) arrays are counted differs based on Shader Model version: - If Shader Model < 6.6, then a UAV array counts as a single UAV regardless of its range size - if Shader Model >= 6.6, then a UAV array contributes its range size to the total number of UAVs I initially thought the complete implementation of this analysis may be blocked by the resource arrays implementation, but it seems that it is not the case, as the `@llvm.dx.resource.handle*` already includes a range size argument.
IanWood1
pushed a commit
to IanWood1/llvm-project
that referenced
this pull request
May 6, 2025
Fixes [llvm#114553](llvm#114553) This implementation replicates the behavior of DXC in setting the `m_b64UAVs` flag: the `Max64UAVs` DXIL module flag is set in the presence of more than 8 UAVs in a DXIL module. The behavior of how UAV (resource) arrays are counted differs based on Shader Model version: - If Shader Model < 6.6, then a UAV array counts as a single UAV regardless of its range size - if Shader Model >= 6.6, then a UAV array contributes its range size to the total number of UAVs I initially thought the complete implementation of this analysis may be blocked by the resource arrays implementation, but it seems that it is not the case, as the `@llvm.dx.resource.handle*` already includes a range size argument.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes #114553
This implementation replicates the behavior of DXC in setting the
m_b64UAVsflag: theMax64UAVsDXIL module flag is set in the presence of more than 8 UAVs in a DXIL module.The behavior of how UAV (resource) arrays are counted differs based on Shader Model version:
I initially thought the complete implementation of this analysis may be blocked by the resource arrays implementation, but it seems that it is not the case, as the
@llvm.dx.resource.handle*already includes a range size argument.