[DirectX] Set whole-module flags prior to evaluating per-function flags #139967
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 #139024 and #139954
Full diff: https://github.com/llvm/llvm-project/pull/139967.diff 11 Files Affected:
diff --git a/llvm/lib/Target/DirectX/DXILShaderFlags.cpp b/llvm/lib/Target/DirectX/DXILShaderFlags.cpp
index 8bdaf68e18e70..18fd56e76c531 100644
--- a/llvm/lib/Target/DirectX/DXILShaderFlags.cpp
+++ b/llvm/lib/Target/DirectX/DXILShaderFlags.cpp
@@ -32,7 +32,7 @@
using namespace llvm;
using namespace llvm::dxil;
-static bool hasUAVsAtEveryStage(DXILResourceMap &DRM,
+static bool hasUAVsAtEveryStage(const DXILResourceMap &DRM,
const ModuleMetadataInfo &MMDI) {
if (DRM.uavs().empty())
return false;
@@ -143,7 +143,7 @@ void ModuleShaderFlags::updateFunctionFlags(ComputedShaderFlags &CSF,
}
if (CSF.LowPrecisionPresent) {
- if (CanSetNativeLowPrecisionMode)
+ if (CSF.NativeLowPrecisionMode)
CSF.NativeLowPrecision = true;
else
CSF.MinimumPrecision = true;
@@ -207,26 +207,71 @@ void ModuleShaderFlags::updateFunctionFlags(ComputedShaderFlags &CSF,
}
}
-/// Construct ModuleShaderFlags for module Module M
-void ModuleShaderFlags::initialize(Module &M, DXILResourceTypeMap &DRTM,
- DXILResourceMap &DRM,
- const ModuleMetadataInfo &MMDI) {
+/// Set shader flags that apply to all functions within the module
+void ModuleShaderFlags::gatherGlobalModuleFlags(
+ ComputedShaderFlags &CSF, const Module &M, const DXILResourceMap &DRM,
+ const ModuleMetadataInfo &MMDI) {
- CanSetResMayNotAlias = MMDI.DXILVersion >= VersionTuple(1, 7);
- // The command line option -res-may-alias will set the dx.resmayalias module
- // flag to 1, thereby disabling the ability to set the ResMayNotAlias flag
- if (auto *ResMayAlias = mdconst::extract_or_null<ConstantInt>(
- M.getModuleFlag("dx.resmayalias")))
- CanSetResMayNotAlias = !ResMayAlias->getValue().getBoolValue();
+ // Set DisableOptimizations flag based on the presence of OptimizeNone
+ // attribute of entry functions.
+ if (MMDI.EntryPropertyVec.size() > 0) {
+ CSF.DisableOptimizations =
+ MMDI.EntryPropertyVec[0].Entry->hasFnAttribute(
+ llvm::Attribute::OptimizeNone);
+ // Ensure all entry functions have the same optimization attribute
+ for (const auto &EntryFunProps : MMDI.EntryPropertyVec)
+ if (CSF.DisableOptimizations !=
+ EntryFunProps.Entry->hasFnAttribute(llvm::Attribute::OptimizeNone))
+ EntryFunProps.Entry->getContext().diagnose(DiagnosticInfoUnsupported(
+ *(EntryFunProps.Entry), "Inconsistent optnone attribute "));
+ }
+ CSF.UAVsAtEveryStage = hasUAVsAtEveryStage(DRM, MMDI);
+
+ // Set the Max64UAVs flag if the number of UAVs is > 8
+ uint32_t NumUAVs = 0;
+ for (auto &UAV : DRM.uavs())
+ if (MMDI.ValidatorVersion < VersionTuple(1, 6))
+ NumUAVs++;
+ else // MMDI.ValidatorVersion >= VersionTuple(1, 6)
+ NumUAVs += UAV.getBinding().Size;
+ if (NumUAVs > 8)
+ CSF.Max64UAVs = true;
+
+ // Set the module flag that enables native low-precision execution mode.
// NativeLowPrecisionMode can only be set when the command line option
// -enable-16bit-types is provided. This is indicated by the dx.nativelowprec
// module flag being set
- CanSetNativeLowPrecisionMode = false;
+ // This flag is needed even if the module does not use 16-bit types because a
+ // corresponding debug module may include 16-bit types, and tools that use the
+ // debug module may expect it to have the same flags as the original
if (auto *NativeLowPrec = mdconst::extract_or_null<ConstantInt>(
M.getModuleFlag("dx.nativelowprec")))
if (MMDI.ShaderModelVersion >= VersionTuple(6, 2))
- CanSetNativeLowPrecisionMode = NativeLowPrec->getValue().getBoolValue();
+ CSF.NativeLowPrecisionMode = NativeLowPrec->getValue().getBoolValue();
+
+ CanSetResMayNotAlias = MMDI.DXILVersion >= VersionTuple(1, 7);
+ // The command line option -res-may-alias will set the dx.resmayalias module
+ // flag to 1, thereby disabling the ability to set the ResMayNotAlias flag
+ if (auto *ResMayAlias = mdconst::extract_or_null<ConstantInt>(
+ M.getModuleFlag("dx.resmayalias")))
+ if (ResMayAlias->getValue().getBoolValue())
+ CanSetResMayNotAlias = false;
+
+ // Set ResMayNotAlias to true if DXIL validator version < 1.8 and there
+ // are UAVs present globally.
+ if (CanSetResMayNotAlias && MMDI.ValidatorVersion < VersionTuple(1, 8))
+ CSF.ResMayNotAlias = !DRM.uavs().empty();
+ }
+
+/// Construct ModuleShaderFlags for module Module M
+void ModuleShaderFlags::initialize(Module &M, DXILResourceTypeMap &DRTM,
+ const DXILResourceMap &DRM,
+ const ModuleMetadataInfo &MMDI) {
+
+ ComputedShaderFlags GlobalSFMask;
+ gatherGlobalModuleFlags(GlobalSFMask, M, DRM, MMDI);
+ CombinedSFMask.merge(GlobalSFMask);
CallGraph CG(M);
@@ -252,7 +297,7 @@ void ModuleShaderFlags::initialize(Module &M, DXILResourceTypeMap &DRTM,
continue;
}
- ComputedShaderFlags CSF;
+ ComputedShaderFlags CSF = GlobalSFMask;
for (const auto &BB : *F)
for (const auto &I : BB)
updateFunctionFlags(CSF, I, DRTM, MMDI);
@@ -273,43 +318,6 @@ void ModuleShaderFlags::initialize(Module &M, DXILResourceTypeMap &DRTM,
// Merge SCCSF with that of F
FunctionFlags[F].merge(SCCSF);
}
-
- // Set DisableOptimizations flag based on the presence of OptimizeNone
- // attribute of entry functions.
- if (MMDI.EntryPropertyVec.size() > 0) {
- CombinedSFMask.DisableOptimizations =
- MMDI.EntryPropertyVec[0].Entry->hasFnAttribute(
- llvm::Attribute::OptimizeNone);
- // Ensure all entry functions have the same optimization attribute
- for (const auto &EntryFunProps : MMDI.EntryPropertyVec)
- if (CombinedSFMask.DisableOptimizations !=
- EntryFunProps.Entry->hasFnAttribute(llvm::Attribute::OptimizeNone))
- EntryFunProps.Entry->getContext().diagnose(DiagnosticInfoUnsupported(
- *(EntryFunProps.Entry), "Inconsistent optnone attribute "));
- }
-
- // Set ResMayNotAlias to true if DXIL validator version < 1.8 and there
- // are UAVs present globally.
- if (CanSetResMayNotAlias && MMDI.ValidatorVersion < VersionTuple(1, 8))
- CombinedSFMask.ResMayNotAlias = !DRM.uavs().empty();
-
- // Set the module flag that enables native low-precision execution mode. This
- // is needed even if the module does not use 16-bit types because a
- // corresponding debug module may include 16-bit types, and tools that use the
- // debug module may expect it to have the same flags as the original
- CombinedSFMask.NativeLowPrecisionMode = CanSetNativeLowPrecisionMode;
-
- // Set the Max64UAVs flag if the number of UAVs is > 8
- uint32_t NumUAVs = 0;
- for (auto &UAV : DRM.uavs())
- if (MMDI.ValidatorVersion < VersionTuple(1, 6))
- NumUAVs++;
- else // MMDI.ValidatorVersion >= VersionTuple(1, 6)
- NumUAVs += UAV.getBinding().Size;
- if (NumUAVs > 8)
- CombinedSFMask.Max64UAVs = true;
-
- CombinedSFMask.UAVsAtEveryStage = hasUAVsAtEveryStage(DRM, MMDI);
}
void ComputedShaderFlags::print(raw_ostream &OS) const {
diff --git a/llvm/lib/Target/DirectX/DXILShaderFlags.h b/llvm/lib/Target/DirectX/DXILShaderFlags.h
index c4eef4e708cfd..37b822cf5b36e 100644
--- a/llvm/lib/Target/DirectX/DXILShaderFlags.h
+++ b/llvm/lib/Target/DirectX/DXILShaderFlags.h
@@ -85,15 +85,15 @@ struct ComputedShaderFlags {
};
struct ModuleShaderFlags {
- void initialize(Module &, DXILResourceTypeMap &DRTM, DXILResourceMap &DRM,
- const ModuleMetadataInfo &MMDI);
+ void initialize(Module &, DXILResourceTypeMap &DRTM,
+ const DXILResourceMap &DRM, const ModuleMetadataInfo &MMDI);
const ComputedShaderFlags &getFunctionFlags(const Function *) const;
const ComputedShaderFlags &getCombinedFlags() const { return CombinedSFMask; }
private:
- // Booleans set by module flags
- bool CanSetResMayNotAlias; // dx.resmayalias
- bool CanSetNativeLowPrecisionMode; // dx.nativelowprec
+ // This boolean is inversely set by the LLVM module flag dx.resmayalias to
+ // determine whether or not the ResMayNotAlias DXIL module flag can be set
+ bool CanSetResMayNotAlias;
/// Map of Function-Shader Flag Mask pairs representing properties of each of
/// the functions in the module. Shader Flags of each function represent both
@@ -101,6 +101,8 @@ struct ModuleShaderFlags {
DenseMap<const Function *, ComputedShaderFlags> FunctionFlags;
/// Combined Shader Flag Mask of all functions of the module
ComputedShaderFlags CombinedSFMask{};
+ void gatherGlobalModuleFlags(ComputedShaderFlags &, const Module &M, const DXILResourceMap &,
+ const ModuleMetadataInfo &);
void updateFunctionFlags(ComputedShaderFlags &, const Instruction &,
DXILResourceTypeMap &, const ModuleMetadataInfo &);
};
diff --git a/llvm/test/CodeGen/DirectX/ShaderFlags/disable-opt-cs.ll b/llvm/test/CodeGen/DirectX/ShaderFlags/disable-opt-cs.ll
index 8bf242fdbec67..3f2ec9a85b2a0 100644
--- a/llvm/test/CodeGen/DirectX/ShaderFlags/disable-opt-cs.ll
+++ b/llvm/test/CodeGen/DirectX/ShaderFlags/disable-opt-cs.ll
@@ -8,7 +8,7 @@
; CHECK-NEXT: ; Disable shader optimizations
; CHECK: ; Shader Flags for Module Functions
-; CHECK: ; Function main : 0x00000000
+; CHECK: ; Function main : 0x00000001
; The test source in this file generated from the following command:
; clang -cc1 -triple dxil-pc-shadermodel6.0-compute -x hlsl -emit-llvm -O0 -o - <<EOF
; [numthreads(1,1,1)]
diff --git a/llvm/test/CodeGen/DirectX/ShaderFlags/disable-opt-lib.ll b/llvm/test/CodeGen/DirectX/ShaderFlags/disable-opt-lib.ll
index 3bf582cc2e255..c6e3cc9e96478 100644
--- a/llvm/test/CodeGen/DirectX/ShaderFlags/disable-opt-lib.ll
+++ b/llvm/test/CodeGen/DirectX/ShaderFlags/disable-opt-lib.ll
@@ -8,7 +8,7 @@
; CHECK-NEXT: ; Disable shader optimizations
; CHECK: ; Shader Flags for Module Functions
-; CHECK: ; Function main : 0x00000000
+; CHECK: ; Function main : 0x00000001
; The test source in this file generated from the following command:
; clang -cc1 -triple dxil-pc-shadermodel6.3-library -x hlsl -emit-llvm -O0 -o - <<EOF
diff --git a/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs-array-valver1.5.ll b/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs-array-valver1.5.ll
index af6001be1f610..6079071919dbc 100644
--- a/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs-array-valver1.5.ll
+++ b/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs-array-valver1.5.ll
@@ -13,7 +13,7 @@ target triple = "dxil-pc-shadermodel6.7-library"
; CHECK: UAVs at every shader stage
; CHECK-NOT: 64 UAV slots
-; CHECK: Function test : 0x00000000
+; CHECK: Function test : 0x00010000
define void @test() "hlsl.export" {
; RWBuffer<float> Buf : register(u0, space0)
%buf0 = call target("dx.TypedBuffer", float, 1, 0, 1)
diff --git a/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs-array-valver1.6.ll b/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs-array-valver1.6.ll
index 7e1d73b31b35b..4f1a1e7b3da53 100644
--- a/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs-array-valver1.6.ll
+++ b/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs-array-valver1.6.ll
@@ -14,7 +14,7 @@ target triple = "dxil-pc-shadermodel6.7-library"
; CHECK: UAVs at every shader stage
; CHECK: 64 UAV slots
-; CHECK: Function test : 0x00000000
+; CHECK: Function test : 0x00018000
define void @test() "hlsl.export" {
; RWBuffer<float> Buf : register(u0, space0)
%buf0 = call target("dx.TypedBuffer", float, 1, 0, 1)
diff --git a/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs.ll b/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs.ll
index a97fe5d45d00a..e0d4ac737704e 100644
--- a/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs.ll
+++ b/llvm/test/CodeGen/DirectX/ShaderFlags/max-64-uavs.ll
@@ -13,7 +13,7 @@ target triple = "dxil-pc-shadermodel6.7-library"
; CHECK: 64 UAV slots
; Note: 64 UAV slots does not get set per-function
-; CHECK: Function test : 0x00000000
+; CHECK: Function test : 0x00008000
define void @test() "hlsl.export" {
; RWBuffer<float> Buf : register(u0, space0)
%buf0 = call target("dx.TypedBuffer", float, 1, 0, 1)
diff --git a/llvm/test/CodeGen/DirectX/ShaderFlags/res-may-not-alias-sm6.7.ll b/llvm/test/CodeGen/DirectX/ShaderFlags/res-may-not-alias-sm6.7.ll
index fc6560e321b4b..c0ea6b4c1e8f3 100644
--- a/llvm/test/CodeGen/DirectX/ShaderFlags/res-may-not-alias-sm6.7.ll
+++ b/llvm/test/CodeGen/DirectX/ShaderFlags/res-may-not-alias-sm6.7.ll
@@ -19,7 +19,7 @@ target triple = "dxil-pc-shadermodel6.7-library"
; CHECK: Any UAV may not alias any other UAV
;
-; CHECK: Function loadUAV : 0x00000000
+; CHECK: Function loadUAV : 0x200010000
define float @loadUAV() #0 {
%res = call target("dx.TypedBuffer", float, 1, 0, 0)
@llvm.dx.resource.handlefrombinding(i32 0, i32 0, i32 1, i32 0, i1 false)
@@ -29,7 +29,7 @@ define float @loadUAV() #0 {
ret float %val
}
-; CHECK: Function loadSRV : 0x00000010
+; CHECK: Function loadSRV : 0x200010010
define float @loadSRV() #0 {
%res = tail call target("dx.RawBuffer", float, 0, 0)
@llvm.dx.resource.handlefrombinding(i32 0, i32 0, i32 1, i32 0, i1 false)
diff --git a/llvm/test/CodeGen/DirectX/ShaderFlags/uavs-at-every-stage-lib-valver1.7.ll b/llvm/test/CodeGen/DirectX/ShaderFlags/uavs-at-every-stage-lib-valver1.7.ll
index 552f513095fa5..bd1de58732183 100644
--- a/llvm/test/CodeGen/DirectX/ShaderFlags/uavs-at-every-stage-lib-valver1.7.ll
+++ b/llvm/test/CodeGen/DirectX/ShaderFlags/uavs-at-every-stage-lib-valver1.7.ll
@@ -12,7 +12,7 @@ target triple = "dxil-pc-shadermodel6.5-library"
; CHECK: Note: shader requires additional functionality:
; CHECK: UAVs at every shader stage
-; CHECK: Function test : 0x00000000
+; CHECK: Function test : 0x00010000
define void @test() "hlsl.export" {
; RWBuffer<float> Buf : register(u0, space0)
%buf0 = call target("dx.TypedBuffer", float, 1, 0, 1)
diff --git a/llvm/test/CodeGen/DirectX/ShaderFlags/uavs-at-every-stage-vs.ll b/llvm/test/CodeGen/DirectX/ShaderFlags/uavs-at-every-stage-vs.ll
index d3f556b62ed0c..bdb07b41d1559 100644
--- a/llvm/test/CodeGen/DirectX/ShaderFlags/uavs-at-every-stage-vs.ll
+++ b/llvm/test/CodeGen/DirectX/ShaderFlags/uavs-at-every-stage-vs.ll
@@ -13,7 +13,7 @@ target triple = "dxil-pc-shadermodel6.5-vertex"
; CHECK: Note: shader requires additional functionality:
; CHECK: UAVs at every shader stage
-; CHECK: Function VSMain : 0x00000000
+; CHECK: Function VSMain : 0x00010000
define void @VSMain() {
; RWBuffer<float> Buf : register(u0, space0)
%buf0 = call target("dx.TypedBuffer", float, 1, 0, 1)
diff --git a/llvm/test/CodeGen/DirectX/ShaderFlags/use-native-low-precision-1.ll b/llvm/test/CodeGen/DirectX/ShaderFlags/use-native-low-precision-1.ll
index cb3b486cebce5..ba844b182fd11 100644
--- a/llvm/test/CodeGen/DirectX/ShaderFlags/use-native-low-precision-1.ll
+++ b/llvm/test/CodeGen/DirectX/ShaderFlags/use-native-low-precision-1.ll
@@ -18,19 +18,19 @@ target triple = "dxil-pc-shadermodel6.7-library"
;CHECK-NEXT: ;
;CHECK-NEXT: ; Shader Flags for Module Functions
-;CHECK-LABEL: ; Function add_i16 : 0x00000020
+;CHECK-LABEL: ; Function add_i16 : 0x00800020
define i16 @add_i16(i16 %a, i16 %b) "hlsl.export" {
%sum = add i16 %a, %b
ret i16 %sum
}
-;CHECK-LABEL: ; Function add_i32 : 0x00000000
+;CHECK-LABEL: ; Function add_i32 : 0x00800000
define i32 @add_i32(i32 %a, i32 %b) "hlsl.export" {
%sum = add i32 %a, %b
ret i32 %sum
}
-;CHECK-LABEL: ; Function add_half : 0x00000020
+;CHECK-LABEL: ; Function add_half : 0x00800020
define half @add_half(half %a, half %b) "hlsl.export" {
%sum = fadd half %a, %b
ret half %sum
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
bogner
reviewed
May 14, 2025
|
I think the bugs you are fixing were mostly found because of the validator? Should we be doing some testing that invokes the validator something similaor to what |
farzonl
reviewed
May 15, 2025
farzonl
reviewed
May 15, 2025
bogner
approved these changes
May 15, 2025
Yes, the bugs were found in the validator. Or, the DirectX tests could have RUN lines conditioned on having dxv in the path, and runs the validator if it is in the path; similar to spirv-val I saw in SPIRV tests. Either way, I think this topic should be relegated to a separate issue and PR. |
|
farzonl
approved these changes
May 15, 2025
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 #139024 and #139954