[HLSL][SPIR-V] Implement vk::push_constant #166793
Open
+559
−24
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
s-perron
reviewed
Nov 10, 2025
Keenuts
force-pushed
the
push-constants
branch
2 times, most recently
from
d326ece to
fb3dec4
Compare
llvmbot
added
clang
Member
|
@llvm/pr-subscribers-clang-codegen @llvm/pr-subscribers-clang Author: Nathan Gauër (Keenuts) ChangesImplements initial support for vk::push_constant. The old fix would be to use target specific types, but this is Patch is 27.93 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/166793.diff 33 Files Affected:
diff --git a/clang/include/clang/Basic/AddressSpaces.h b/clang/include/clang/Basic/AddressSpaces.h
index 48e4a1c61fe02..7280b8fc923d2 100644
--- a/clang/include/clang/Basic/AddressSpaces.h
+++ b/clang/include/clang/Basic/AddressSpaces.h
@@ -62,6 +62,7 @@ enum class LangAS : unsigned {
hlsl_private,
hlsl_device,
hlsl_input,
+ hlsl_push_constant,
// Wasm specific address spaces.
wasm_funcref,
diff --git a/clang/include/clang/Basic/Attr.td b/clang/include/clang/Basic/Attr.td
index 8dfe4bc08c48e..e00765a57cb23 100644
--- a/clang/include/clang/Basic/Attr.td
+++ b/clang/include/clang/Basic/Attr.td
@@ -5146,6 +5146,14 @@ def HLSLVkExtBuiltinInput : InheritableAttr {
let Documentation = [HLSLVkExtBuiltinInputDocs];
}
+def HLSLVkPushConstant : InheritableAttr {
+ let Spellings = [CXX11<"vk", "push_constant">];
+ let Args = [];
+ let Subjects = SubjectList<[GlobalVar], ErrorDiag>;
+ let LangOpts = [HLSL];
+ let Documentation = [HLSLVkPushConstantDocs];
+}
+
def HLSLVkConstantId : InheritableAttr {
let Spellings = [CXX11<"vk", "constant_id">];
let Args = [IntArgument<"Id">];
diff --git a/clang/include/clang/Basic/AttrDocs.td b/clang/include/clang/Basic/AttrDocs.td
index 4813191d2d602..3938c624c9d0c 100644
--- a/clang/include/clang/Basic/AttrDocs.td
+++ b/clang/include/clang/Basic/AttrDocs.td
@@ -8777,6 +8777,11 @@ https://github.com/microsoft/hlsl-specs/blob/main/proposals/0011-inline-spirv.md
}];
}
+def HLSLVkPushConstantDocs : Documentation {
+ let Category = DocCatVariable;
+ let Content = [{ FIXME }];
+}
+
def AnnotateTypeDocs : Documentation {
let Category = DocCatType;
let Heading = "annotate_type";
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index a6e60fe4692ee..c9d128d77c90f 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -13183,6 +13183,9 @@ def err_hlsl_attr_invalid_type : Error<
"attribute %0 only applies to a field or parameter of type '%1'">;
def err_hlsl_attr_invalid_ast_node : Error<
"attribute %0 only applies to %1">;
+def err_hlsl_attr_incompatible
+ : Error<"%0 attribute is not compatible with %1 attribute">;
+
def err_hlsl_entry_shader_attr_mismatch : Error<
"%0 attribute on entry function does not match the target profile">;
def err_hlsl_numthreads_argument_oor : Error<"argument '%select{X|Y|Z}0' to numthreads attribute cannot exceed %1">;
@@ -13294,6 +13297,9 @@ def err_hlsl_incomplete_resource_array_in_function_param: Error<
def err_hlsl_assign_to_global_resource: Error<
"assignment to global resource variable %0 is not allowed">;
+def err_hlsl_push_constant_unique
+ : Error<"cannot have more than one push constant block">;
+
// Layout randomization diagnostics.
def err_non_designated_init_used : Error<
"a randomized struct can only be initialized with a designated initializer">;
diff --git a/clang/include/clang/Basic/HLSLRuntime.h b/clang/include/clang/Basic/HLSLRuntime.h
index 03166805daa6a..f6a1cf9636467 100644
--- a/clang/include/clang/Basic/HLSLRuntime.h
+++ b/clang/include/clang/Basic/HLSLRuntime.h
@@ -14,6 +14,7 @@
#ifndef CLANG_BASIC_HLSLRUNTIME_H
#define CLANG_BASIC_HLSLRUNTIME_H
+#include "clang/Basic/AddressSpaces.h"
#include "clang/Basic/LangOptions.h"
#include <cstdint>
@@ -30,6 +31,10 @@ getStageFromEnvironment(const llvm::Triple::EnvironmentType &E) {
return static_cast<ShaderStage>(Pipeline);
}
+constexpr bool isInitializedByPipeline(LangAS AS) {
+ return AS == LangAS::hlsl_input || AS == LangAS::hlsl_push_constant;
+}
+
#define ENUM_COMPARE_ASSERT(Value) \
static_assert( \
getStageFromEnvironment(llvm::Triple::Value) == ShaderStage::Value, \
diff --git a/clang/include/clang/Sema/SemaHLSL.h b/clang/include/clang/Sema/SemaHLSL.h
index 86da323892f98..2fcac237eba1c 100644
--- a/clang/include/clang/Sema/SemaHLSL.h
+++ b/clang/include/clang/Sema/SemaHLSL.h
@@ -190,6 +190,7 @@ class SemaHLSL : public SemaBase {
void handleSemanticAttr(Decl *D, const ParsedAttr &AL);
void handleVkExtBuiltinInputAttr(Decl *D, const ParsedAttr &AL);
+ void handleVkPushConstantAttr(Decl *D, const ParsedAttr &AL);
bool CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall);
QualType ProcessResourceTypeAttributes(QualType Wrapped);
@@ -239,6 +240,8 @@ class SemaHLSL : public SemaBase {
IdentifierInfo *RootSigOverrideIdent = nullptr;
+ bool HasDeclaredAPushConstant = false;
+
struct SemanticInfo {
HLSLParsedSemanticAttr *Semantic;
std::optional<uint32_t> Index;
diff --git a/clang/lib/AST/Type.cpp b/clang/lib/AST/Type.cpp
index 4548af17e37f2..53082bcf78f6a 100644
--- a/clang/lib/AST/Type.cpp
+++ b/clang/lib/AST/Type.cpp
@@ -101,6 +101,7 @@ bool Qualifiers::isTargetAddressSpaceSupersetOf(LangAS A, LangAS B,
(A == LangAS::Default && B == LangAS::hlsl_private) ||
(A == LangAS::Default && B == LangAS::hlsl_device) ||
(A == LangAS::Default && B == LangAS::hlsl_input) ||
+ (A == LangAS::Default && B == LangAS::hlsl_push_constant) ||
// Conversions from target specific address spaces may be legal
// depending on the target information.
Ctx.getTargetInfo().isAddressSpaceSupersetOf(A, B);
diff --git a/clang/lib/AST/TypePrinter.cpp b/clang/lib/AST/TypePrinter.cpp
index c18b2eafc722c..8448dd3748e28 100644
--- a/clang/lib/AST/TypePrinter.cpp
+++ b/clang/lib/AST/TypePrinter.cpp
@@ -2749,6 +2749,8 @@ std::string Qualifiers::getAddrSpaceAsString(LangAS AS) {
return "hlsl_device";
case LangAS::hlsl_input:
return "hlsl_input";
+ case LangAS::hlsl_push_constant:
+ return "hlsl_push_constant";
case LangAS::wasm_funcref:
return "__funcref";
default:
diff --git a/clang/lib/Basic/TargetInfo.cpp b/clang/lib/Basic/TargetInfo.cpp
index ffaf98bf9c366..92ca7a66a9593 100644
--- a/clang/lib/Basic/TargetInfo.cpp
+++ b/clang/lib/Basic/TargetInfo.cpp
@@ -52,6 +52,7 @@ static const LangASMap FakeAddrSpaceMap = {
15, // hlsl_private
16, // hlsl_device
17, // hlsl_input
+ 18, // hlsl_push_constant
20, // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/AArch64.h b/clang/lib/Basic/Targets/AArch64.h
index 1a7aa658e9d87..8e8d8e6ae86b5 100644
--- a/clang/lib/Basic/Targets/AArch64.h
+++ b/clang/lib/Basic/Targets/AArch64.h
@@ -48,6 +48,7 @@ static const unsigned ARM64AddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/AMDGPU.cpp b/clang/lib/Basic/Targets/AMDGPU.cpp
index d4d696b8456b6..993a73a89c9e9 100644
--- a/clang/lib/Basic/Targets/AMDGPU.cpp
+++ b/clang/lib/Basic/Targets/AMDGPU.cpp
@@ -63,6 +63,7 @@ const LangASMap AMDGPUTargetInfo::AMDGPUDefIsGenMap = {
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_private
llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_device
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_input
+ llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_push_constant
};
const LangASMap AMDGPUTargetInfo::AMDGPUDefIsPrivMap = {
@@ -91,6 +92,7 @@ const LangASMap AMDGPUTargetInfo::AMDGPUDefIsPrivMap = {
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_private
llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_device
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_input
+ llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_push_constant
};
} // namespace targets
} // namespace clang
diff --git a/clang/lib/Basic/Targets/DirectX.h b/clang/lib/Basic/Targets/DirectX.h
index a21a593365773..c0799a6f7610f 100644
--- a/clang/lib/Basic/Targets/DirectX.h
+++ b/clang/lib/Basic/Targets/DirectX.h
@@ -46,6 +46,7 @@ static const unsigned DirectXAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/NVPTX.h b/clang/lib/Basic/Targets/NVPTX.h
index f5c8396f398aa..6338a4f2f9036 100644
--- a/clang/lib/Basic/Targets/NVPTX.h
+++ b/clang/lib/Basic/Targets/NVPTX.h
@@ -50,6 +50,7 @@ static const unsigned NVPTXAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/SPIR.h b/clang/lib/Basic/Targets/SPIR.h
index 22b2799518dd0..94449231efb94 100644
--- a/clang/lib/Basic/Targets/SPIR.h
+++ b/clang/lib/Basic/Targets/SPIR.h
@@ -51,6 +51,7 @@ static const unsigned SPIRDefIsPrivMap[] = {
10, // hlsl_private
11, // hlsl_device
7, // hlsl_input
+ 13, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
@@ -87,6 +88,7 @@ static const unsigned SPIRDefIsGenMap[] = {
10, // hlsl_private
11, // hlsl_device
7, // hlsl_input
+ 13, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/SystemZ.h b/clang/lib/Basic/Targets/SystemZ.h
index 4e15d5af1cde6..4ce515b31a001 100644
--- a/clang/lib/Basic/Targets/SystemZ.h
+++ b/clang/lib/Basic/Targets/SystemZ.h
@@ -46,6 +46,7 @@ static const unsigned ZOSAddressMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
0 // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/TCE.h b/clang/lib/Basic/Targets/TCE.h
index 005cab9819472..161025378c471 100644
--- a/clang/lib/Basic/Targets/TCE.h
+++ b/clang/lib/Basic/Targets/TCE.h
@@ -55,6 +55,7 @@ static const unsigned TCEOpenCLAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/WebAssembly.h b/clang/lib/Basic/Targets/WebAssembly.h
index 4de6ce6bb5a21..c8065843aeb42 100644
--- a/clang/lib/Basic/Targets/WebAssembly.h
+++ b/clang/lib/Basic/Targets/WebAssembly.h
@@ -46,6 +46,7 @@ static const unsigned WebAssemblyAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
20, // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/X86.h b/clang/lib/Basic/Targets/X86.h
index e7da2622e78b5..7b88ac70e234f 100644
--- a/clang/lib/Basic/Targets/X86.h
+++ b/clang/lib/Basic/Targets/X86.h
@@ -50,6 +50,7 @@ static const unsigned X86AddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 3eeb1718e455a..c3b536c7a267f 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -6049,9 +6049,11 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D,
getCUDARuntime().handleVarRegistration(D, *GV);
}
- if (LangOpts.HLSL && GetGlobalVarAddressSpace(D) == LangAS::hlsl_input) {
+ if (LangOpts.HLSL &&
+ hlsl::isInitializedByPipeline(GetGlobalVarAddressSpace(D))) {
// HLSL Input variables are considered to be set by the driver/pipeline, but
- // only visible to a single thread/wave.
+ // only visible to a single thread/wave. Push constants are also externally
+ // initialized, but constant, hence cross-wave visibility is not relevant.
GV->setExternallyInitialized(true);
} else {
GV->setInitializer(Init);
@@ -6102,10 +6104,11 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D,
!D->hasAttr<ConstInitAttr>())
Linkage = llvm::GlobalValue::InternalLinkage;
- // HLSL variables in the input address space maps like memory-mapped
- // variables. Even if they are 'static', they are externally initialized and
- // read/write by the hardware/driver/pipeline.
- if (LangOpts.HLSL && GetGlobalVarAddressSpace(D) == LangAS::hlsl_input)
+ // HLSL variables in the input or push-constant address space maps are like
+ // memory-mapped variables. Even if they are 'static', they are externally
+ // initialized and read/write by the hardware/driver/pipeline.
+ if (LangOpts.HLSL &&
+ hlsl::isInitializedByPipeline(GetGlobalVarAddressSpace(D)))
Linkage = llvm::GlobalValue::ExternalLinkage;
GV->setLinkage(Linkage);
diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
index 25b89d65847ad..5c29bf5e77414 100644
--- a/clang/lib/Sema/SemaDecl.cpp
+++ b/clang/lib/Sema/SemaDecl.cpp
@@ -30,6 +30,7 @@
#include "clang/AST/Type.h"
#include "clang/Basic/Builtins.h"
#include "clang/Basic/DiagnosticComment.h"
+#include "clang/Basic/HLSLRuntime.h"
#include "clang/Basic/PartialDiagnostic.h"
#include "clang/Basic/SourceManager.h"
#include "clang/Basic/TargetInfo.h"
@@ -14559,10 +14560,10 @@ void Sema::ActOnUninitializedDecl(Decl *RealDecl) {
if (getLangOpts().HLSL && HLSL().ActOnUninitializedVarDecl(Var))
return;
- // HLSL input variables are expected to be externally initialized, even
- // when marked `static`.
+ // HLSL input & push-constant variables are expected to be externally
+ // initialized, even when marked `static`.
if (getLangOpts().HLSL &&
- Var->getType().getAddressSpace() == LangAS::hlsl_input)
+ hlsl::isInitializedByPipeline(Var->getType().getAddressSpace()))
return;
// C++03 [dcl.init]p9:
diff --git a/clang/lib/Sema/SemaDeclAttr.cpp b/clang/lib/Sema/SemaDeclAttr.cpp
index a9e7b44ac9d73..0396155bd6a9d 100644
--- a/clang/lib/Sema/SemaDeclAttr.cpp
+++ b/clang/lib/Sema/SemaDeclAttr.cpp
@@ -7614,6 +7614,9 @@ ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D, const ParsedAttr &AL,
case ParsedAttr::AT_HLSLVkExtBuiltinInput:
S.HLSL().handleVkExtBuiltinInputAttr(D, AL);
break;
+ case ParsedAttr::AT_HLSLVkPushConstant:
+ S.HLSL().handleVkPushConstantAttr(D, AL);
+ break;
case ParsedAttr::AT_HLSLVkConstantId:
S.HLSL().handleVkConstantIdAttr(D, AL);
break;
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 2b9b3abbd5360..1831584c88697 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -1667,6 +1667,11 @@ void SemaHLSL::handleVkExtBuiltinInputAttr(Decl *D, const ParsedAttr &AL) {
HLSLVkExtBuiltinInputAttr(getASTContext(), AL, ID));
}
+void SemaHLSL::handleVkPushConstantAttr(Decl *D, const ParsedAttr &AL) {
+ D->addAttr(::new (getASTContext())
+ HLSLVkPushConstantAttr(getASTContext(), AL));
+}
+
void SemaHLSL::handleVkConstantIdAttr(Decl *D, const ParsedAttr &AL) {
uint32_t Id;
if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(0), Id))
@@ -3837,12 +3842,15 @@ QualType SemaHLSL::getInoutParameterType(QualType Ty) {
return Ty;
}
-static bool IsDefaultBufferConstantDecl(VarDecl *VD) {
+static bool IsDefaultBufferConstantDecl(const ASTContext &Ctx, VarDecl *VD) {
+ bool IsVulkan =
+ Ctx.getTargetInfo().getTriple().getOS() == llvm::Triple::Vulkan;
+ bool IsVKPushConstant = IsVulkan && VD->hasAttr<HLSLVkPushConstantAttr>();
QualType QT = VD->getType();
return VD->getDeclContext()->isTranslationUnit() &&
QT.getAddressSpace() == LangAS::Default &&
VD->getStorageClass() != SC_Static &&
- !VD->hasAttr<HLSLVkConstantIdAttr>() &&
+ !VD->hasAttr<HLSLVkConstantIdAttr>() && !IsVKPushConstant &&
!isInvalidConstantBufferLeafElementType(QT.getTypePtr());
}
@@ -3863,6 +3871,19 @@ void SemaHLSL::deduceAddressSpace(VarDecl *Decl) {
return;
}
+ bool IsVulkan = getASTContext().getTargetInfo().getTriple().getOS() ==
+ llvm::Triple::Vulkan;
+ if (IsVulkan && Decl->hasAttr<HLSLVkPushConstantAttr>()) {
+ if (HasDeclaredAPushConstant)
+ SemaRef.Diag(Decl->getLocation(), diag::err_hlsl_push_constant_unique);
+
+ LangAS ImplAS = LangAS::hlsl_push_constant;
+ Type = SemaRef.getASTContext().getAddrSpaceQualType(Type, ImplAS);
+ Decl->setType(Type);
+ HasDeclaredAPushConstant = true;
+ return;
+ }
+
if (Type->isSamplerT() || Type->isVoidType())
return;
@@ -3895,7 +3916,7 @@ void SemaHLSL::ActOnVariableDeclarator(VarDecl *VD) {
// Global variables outside a cbuffer block that are not a resource, static,
// groupshared, or an empty array or struct belong to the default constant
// buffer $Globals (to be created at the end of the translation unit).
- if (IsDefaultBufferConstantDecl(VD)) {
+ if (IsDefaultBufferConstantDecl(getASTContext(), VD)) {
// update address space to hlsl_constant
QualType NewTy = getASTContext().getAddrSpaceQualType(
VD->getType(), LangAS::hlsl_constant);
@@ -4196,8 +4217,11 @@ void SemaHLSL::processExplicitBindingsOnDecl(VarDecl *VD) {
bool HasBinding = false;
for (Attr *A : VD->attrs()) {
- if (isa<HLSLVkBindingAttr>(A))
+ if (isa<HLSLVkBindingAttr>(A)) {
HasBinding = true;
+ if (auto PA = VD->getAttr<HLSLVkPushConstantAttr>())
+ Diag(PA->getLoc(), diag::err_hlsl_attr_incompatible) << A << PA;
+ }
HLSLResourceBindingAttr *RBA = dyn_cast<HLSLResourceBindingAttr>(A);
if (!RBA || !RBA->hasRegisterSlot())
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl
new file mode 100644
index 0000000000000..412ec4dffc572
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl
@@ -0,0 +1,20 @@
+// RUN: %clang_cc1 -triple spirv-pc-vulkan-compute -x hlsl -emit-llvm -finclude-default-header -disable-llvm-passes -o - %s | FileCheck %s
+
+struct S {
+ uint32_t a : 1;
+ uint32_t b : 1;
+};
+// CHECK: %struct.S = type { i8 }
+
+[[vk::push_constant]] S buffer;
+// CHECK: @buffer = external hidden addrspace(13) externally_initialized global %struct.S, align 1
+
+[numthreads(1, 1, 1)]
+void main() {
+ uint32_t v = buffer.b;
+// CHECK: %bf.load = load i8, ptr addrspace(13) @buffer, align 1
+// CHECK: %bf.lshr = lshr i8 %bf.load, 1
+// CHECK: %bf.clear = and i8 %bf.lshr, 1
+// CHECK: %bf.cast = zext i8 %bf.clear to i32
+// CHECK: store i32 %bf.cast
+}
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl
new file mode 100644
index 0000000000000..2b2e9d09c7ab0
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl
@@ -0,0 +1,17 @@
+// RUN: %clang_cc1 -triple spirv-pc-vulkan-compute -x hlsl -emit-llvm -finclude-default-header -disable-llvm-passes -o - %s | FileCheck %s
+
+[[vk::push_constant]]
+struct {
+ int a;
+ float b;
+ float3 c;
+}
+PushConstants;
+
+// CHECK: %struct.anon = type <{ i32, float, <3 x float> }>
+// CHECK: @PushConstants = external hidden addrspace(13) externally_initialized global %struct.anon, align 1
+
+[numthreads(1, 1, 1)]
+void main() {
+ float tmp = PushConstants.b;
+}
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl
new file mode 100644
index 0000000000000..6b58decfa5188
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl
@@ -0,0 +1,13 @@
+// RUN: %clang_cc1 -triple spirv-unknown-vulkan-compute -x hlsl -emit-llvm -disable-llvm-passes -o - -hlsl-entry main %s -verify
+
+struct S {
+ float f;
+};
+
+// expected-error@+1 {{'vk::binding' attribute is not compatible with 'vk::push_constant' attribute}}
+[[vk::push_constant, vk::binding(5)]]
+S pcs;
+
+[numthreads(1, 1, 1)]
+void main() {
+}
diff...
[truncated]
|
Member
|
@llvm/pr-subscribers-hlsl Author: Nathan Gauër (Keenuts) ChangesImplements initial support for vk::push_constant. The old fix would be to use target specific types, but this is Patch is 27.93 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/166793.diff 33 Files Affected:
diff --git a/clang/include/clang/Basic/AddressSpaces.h b/clang/include/clang/Basic/AddressSpaces.h
index 48e4a1c61fe02..7280b8fc923d2 100644
--- a/clang/include/clang/Basic/AddressSpaces.h
+++ b/clang/include/clang/Basic/AddressSpaces.h
@@ -62,6 +62,7 @@ enum class LangAS : unsigned {
hlsl_private,
hlsl_device,
hlsl_input,
+ hlsl_push_constant,
// Wasm specific address spaces.
wasm_funcref,
diff --git a/clang/include/clang/Basic/Attr.td b/clang/include/clang/Basic/Attr.td
index 8dfe4bc08c48e..e00765a57cb23 100644
--- a/clang/include/clang/Basic/Attr.td
+++ b/clang/include/clang/Basic/Attr.td
@@ -5146,6 +5146,14 @@ def HLSLVkExtBuiltinInput : InheritableAttr {
let Documentation = [HLSLVkExtBuiltinInputDocs];
}
+def HLSLVkPushConstant : InheritableAttr {
+ let Spellings = [CXX11<"vk", "push_constant">];
+ let Args = [];
+ let Subjects = SubjectList<[GlobalVar], ErrorDiag>;
+ let LangOpts = [HLSL];
+ let Documentation = [HLSLVkPushConstantDocs];
+}
+
def HLSLVkConstantId : InheritableAttr {
let Spellings = [CXX11<"vk", "constant_id">];
let Args = [IntArgument<"Id">];
diff --git a/clang/include/clang/Basic/AttrDocs.td b/clang/include/clang/Basic/AttrDocs.td
index 4813191d2d602..3938c624c9d0c 100644
--- a/clang/include/clang/Basic/AttrDocs.td
+++ b/clang/include/clang/Basic/AttrDocs.td
@@ -8777,6 +8777,11 @@ https://github.com/microsoft/hlsl-specs/blob/main/proposals/0011-inline-spirv.md
}];
}
+def HLSLVkPushConstantDocs : Documentation {
+ let Category = DocCatVariable;
+ let Content = [{ FIXME }];
+}
+
def AnnotateTypeDocs : Documentation {
let Category = DocCatType;
let Heading = "annotate_type";
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index a6e60fe4692ee..c9d128d77c90f 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -13183,6 +13183,9 @@ def err_hlsl_attr_invalid_type : Error<
"attribute %0 only applies to a field or parameter of type '%1'">;
def err_hlsl_attr_invalid_ast_node : Error<
"attribute %0 only applies to %1">;
+def err_hlsl_attr_incompatible
+ : Error<"%0 attribute is not compatible with %1 attribute">;
+
def err_hlsl_entry_shader_attr_mismatch : Error<
"%0 attribute on entry function does not match the target profile">;
def err_hlsl_numthreads_argument_oor : Error<"argument '%select{X|Y|Z}0' to numthreads attribute cannot exceed %1">;
@@ -13294,6 +13297,9 @@ def err_hlsl_incomplete_resource_array_in_function_param: Error<
def err_hlsl_assign_to_global_resource: Error<
"assignment to global resource variable %0 is not allowed">;
+def err_hlsl_push_constant_unique
+ : Error<"cannot have more than one push constant block">;
+
// Layout randomization diagnostics.
def err_non_designated_init_used : Error<
"a randomized struct can only be initialized with a designated initializer">;
diff --git a/clang/include/clang/Basic/HLSLRuntime.h b/clang/include/clang/Basic/HLSLRuntime.h
index 03166805daa6a..f6a1cf9636467 100644
--- a/clang/include/clang/Basic/HLSLRuntime.h
+++ b/clang/include/clang/Basic/HLSLRuntime.h
@@ -14,6 +14,7 @@
#ifndef CLANG_BASIC_HLSLRUNTIME_H
#define CLANG_BASIC_HLSLRUNTIME_H
+#include "clang/Basic/AddressSpaces.h"
#include "clang/Basic/LangOptions.h"
#include <cstdint>
@@ -30,6 +31,10 @@ getStageFromEnvironment(const llvm::Triple::EnvironmentType &E) {
return static_cast<ShaderStage>(Pipeline);
}
+constexpr bool isInitializedByPipeline(LangAS AS) {
+ return AS == LangAS::hlsl_input || AS == LangAS::hlsl_push_constant;
+}
+
#define ENUM_COMPARE_ASSERT(Value) \
static_assert( \
getStageFromEnvironment(llvm::Triple::Value) == ShaderStage::Value, \
diff --git a/clang/include/clang/Sema/SemaHLSL.h b/clang/include/clang/Sema/SemaHLSL.h
index 86da323892f98..2fcac237eba1c 100644
--- a/clang/include/clang/Sema/SemaHLSL.h
+++ b/clang/include/clang/Sema/SemaHLSL.h
@@ -190,6 +190,7 @@ class SemaHLSL : public SemaBase {
void handleSemanticAttr(Decl *D, const ParsedAttr &AL);
void handleVkExtBuiltinInputAttr(Decl *D, const ParsedAttr &AL);
+ void handleVkPushConstantAttr(Decl *D, const ParsedAttr &AL);
bool CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall);
QualType ProcessResourceTypeAttributes(QualType Wrapped);
@@ -239,6 +240,8 @@ class SemaHLSL : public SemaBase {
IdentifierInfo *RootSigOverrideIdent = nullptr;
+ bool HasDeclaredAPushConstant = false;
+
struct SemanticInfo {
HLSLParsedSemanticAttr *Semantic;
std::optional<uint32_t> Index;
diff --git a/clang/lib/AST/Type.cpp b/clang/lib/AST/Type.cpp
index 4548af17e37f2..53082bcf78f6a 100644
--- a/clang/lib/AST/Type.cpp
+++ b/clang/lib/AST/Type.cpp
@@ -101,6 +101,7 @@ bool Qualifiers::isTargetAddressSpaceSupersetOf(LangAS A, LangAS B,
(A == LangAS::Default && B == LangAS::hlsl_private) ||
(A == LangAS::Default && B == LangAS::hlsl_device) ||
(A == LangAS::Default && B == LangAS::hlsl_input) ||
+ (A == LangAS::Default && B == LangAS::hlsl_push_constant) ||
// Conversions from target specific address spaces may be legal
// depending on the target information.
Ctx.getTargetInfo().isAddressSpaceSupersetOf(A, B);
diff --git a/clang/lib/AST/TypePrinter.cpp b/clang/lib/AST/TypePrinter.cpp
index c18b2eafc722c..8448dd3748e28 100644
--- a/clang/lib/AST/TypePrinter.cpp
+++ b/clang/lib/AST/TypePrinter.cpp
@@ -2749,6 +2749,8 @@ std::string Qualifiers::getAddrSpaceAsString(LangAS AS) {
return "hlsl_device";
case LangAS::hlsl_input:
return "hlsl_input";
+ case LangAS::hlsl_push_constant:
+ return "hlsl_push_constant";
case LangAS::wasm_funcref:
return "__funcref";
default:
diff --git a/clang/lib/Basic/TargetInfo.cpp b/clang/lib/Basic/TargetInfo.cpp
index ffaf98bf9c366..92ca7a66a9593 100644
--- a/clang/lib/Basic/TargetInfo.cpp
+++ b/clang/lib/Basic/TargetInfo.cpp
@@ -52,6 +52,7 @@ static const LangASMap FakeAddrSpaceMap = {
15, // hlsl_private
16, // hlsl_device
17, // hlsl_input
+ 18, // hlsl_push_constant
20, // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/AArch64.h b/clang/lib/Basic/Targets/AArch64.h
index 1a7aa658e9d87..8e8d8e6ae86b5 100644
--- a/clang/lib/Basic/Targets/AArch64.h
+++ b/clang/lib/Basic/Targets/AArch64.h
@@ -48,6 +48,7 @@ static const unsigned ARM64AddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/AMDGPU.cpp b/clang/lib/Basic/Targets/AMDGPU.cpp
index d4d696b8456b6..993a73a89c9e9 100644
--- a/clang/lib/Basic/Targets/AMDGPU.cpp
+++ b/clang/lib/Basic/Targets/AMDGPU.cpp
@@ -63,6 +63,7 @@ const LangASMap AMDGPUTargetInfo::AMDGPUDefIsGenMap = {
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_private
llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_device
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_input
+ llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_push_constant
};
const LangASMap AMDGPUTargetInfo::AMDGPUDefIsPrivMap = {
@@ -91,6 +92,7 @@ const LangASMap AMDGPUTargetInfo::AMDGPUDefIsPrivMap = {
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_private
llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_device
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_input
+ llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_push_constant
};
} // namespace targets
} // namespace clang
diff --git a/clang/lib/Basic/Targets/DirectX.h b/clang/lib/Basic/Targets/DirectX.h
index a21a593365773..c0799a6f7610f 100644
--- a/clang/lib/Basic/Targets/DirectX.h
+++ b/clang/lib/Basic/Targets/DirectX.h
@@ -46,6 +46,7 @@ static const unsigned DirectXAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/NVPTX.h b/clang/lib/Basic/Targets/NVPTX.h
index f5c8396f398aa..6338a4f2f9036 100644
--- a/clang/lib/Basic/Targets/NVPTX.h
+++ b/clang/lib/Basic/Targets/NVPTX.h
@@ -50,6 +50,7 @@ static const unsigned NVPTXAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/SPIR.h b/clang/lib/Basic/Targets/SPIR.h
index 22b2799518dd0..94449231efb94 100644
--- a/clang/lib/Basic/Targets/SPIR.h
+++ b/clang/lib/Basic/Targets/SPIR.h
@@ -51,6 +51,7 @@ static const unsigned SPIRDefIsPrivMap[] = {
10, // hlsl_private
11, // hlsl_device
7, // hlsl_input
+ 13, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
@@ -87,6 +88,7 @@ static const unsigned SPIRDefIsGenMap[] = {
10, // hlsl_private
11, // hlsl_device
7, // hlsl_input
+ 13, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/SystemZ.h b/clang/lib/Basic/Targets/SystemZ.h
index 4e15d5af1cde6..4ce515b31a001 100644
--- a/clang/lib/Basic/Targets/SystemZ.h
+++ b/clang/lib/Basic/Targets/SystemZ.h
@@ -46,6 +46,7 @@ static const unsigned ZOSAddressMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
0 // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/TCE.h b/clang/lib/Basic/Targets/TCE.h
index 005cab9819472..161025378c471 100644
--- a/clang/lib/Basic/Targets/TCE.h
+++ b/clang/lib/Basic/Targets/TCE.h
@@ -55,6 +55,7 @@ static const unsigned TCEOpenCLAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/WebAssembly.h b/clang/lib/Basic/Targets/WebAssembly.h
index 4de6ce6bb5a21..c8065843aeb42 100644
--- a/clang/lib/Basic/Targets/WebAssembly.h
+++ b/clang/lib/Basic/Targets/WebAssembly.h
@@ -46,6 +46,7 @@ static const unsigned WebAssemblyAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
20, // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/X86.h b/clang/lib/Basic/Targets/X86.h
index e7da2622e78b5..7b88ac70e234f 100644
--- a/clang/lib/Basic/Targets/X86.h
+++ b/clang/lib/Basic/Targets/X86.h
@@ -50,6 +50,7 @@ static const unsigned X86AddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 3eeb1718e455a..c3b536c7a267f 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -6049,9 +6049,11 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D,
getCUDARuntime().handleVarRegistration(D, *GV);
}
- if (LangOpts.HLSL && GetGlobalVarAddressSpace(D) == LangAS::hlsl_input) {
+ if (LangOpts.HLSL &&
+ hlsl::isInitializedByPipeline(GetGlobalVarAddressSpace(D))) {
// HLSL Input variables are considered to be set by the driver/pipeline, but
- // only visible to a single thread/wave.
+ // only visible to a single thread/wave. Push constants are also externally
+ // initialized, but constant, hence cross-wave visibility is not relevant.
GV->setExternallyInitialized(true);
} else {
GV->setInitializer(Init);
@@ -6102,10 +6104,11 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D,
!D->hasAttr<ConstInitAttr>())
Linkage = llvm::GlobalValue::InternalLinkage;
- // HLSL variables in the input address space maps like memory-mapped
- // variables. Even if they are 'static', they are externally initialized and
- // read/write by the hardware/driver/pipeline.
- if (LangOpts.HLSL && GetGlobalVarAddressSpace(D) == LangAS::hlsl_input)
+ // HLSL variables in the input or push-constant address space maps are like
+ // memory-mapped variables. Even if they are 'static', they are externally
+ // initialized and read/write by the hardware/driver/pipeline.
+ if (LangOpts.HLSL &&
+ hlsl::isInitializedByPipeline(GetGlobalVarAddressSpace(D)))
Linkage = llvm::GlobalValue::ExternalLinkage;
GV->setLinkage(Linkage);
diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
index 25b89d65847ad..5c29bf5e77414 100644
--- a/clang/lib/Sema/SemaDecl.cpp
+++ b/clang/lib/Sema/SemaDecl.cpp
@@ -30,6 +30,7 @@
#include "clang/AST/Type.h"
#include "clang/Basic/Builtins.h"
#include "clang/Basic/DiagnosticComment.h"
+#include "clang/Basic/HLSLRuntime.h"
#include "clang/Basic/PartialDiagnostic.h"
#include "clang/Basic/SourceManager.h"
#include "clang/Basic/TargetInfo.h"
@@ -14559,10 +14560,10 @@ void Sema::ActOnUninitializedDecl(Decl *RealDecl) {
if (getLangOpts().HLSL && HLSL().ActOnUninitializedVarDecl(Var))
return;
- // HLSL input variables are expected to be externally initialized, even
- // when marked `static`.
+ // HLSL input & push-constant variables are expected to be externally
+ // initialized, even when marked `static`.
if (getLangOpts().HLSL &&
- Var->getType().getAddressSpace() == LangAS::hlsl_input)
+ hlsl::isInitializedByPipeline(Var->getType().getAddressSpace()))
return;
// C++03 [dcl.init]p9:
diff --git a/clang/lib/Sema/SemaDeclAttr.cpp b/clang/lib/Sema/SemaDeclAttr.cpp
index a9e7b44ac9d73..0396155bd6a9d 100644
--- a/clang/lib/Sema/SemaDeclAttr.cpp
+++ b/clang/lib/Sema/SemaDeclAttr.cpp
@@ -7614,6 +7614,9 @@ ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D, const ParsedAttr &AL,
case ParsedAttr::AT_HLSLVkExtBuiltinInput:
S.HLSL().handleVkExtBuiltinInputAttr(D, AL);
break;
+ case ParsedAttr::AT_HLSLVkPushConstant:
+ S.HLSL().handleVkPushConstantAttr(D, AL);
+ break;
case ParsedAttr::AT_HLSLVkConstantId:
S.HLSL().handleVkConstantIdAttr(D, AL);
break;
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 2b9b3abbd5360..1831584c88697 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -1667,6 +1667,11 @@ void SemaHLSL::handleVkExtBuiltinInputAttr(Decl *D, const ParsedAttr &AL) {
HLSLVkExtBuiltinInputAttr(getASTContext(), AL, ID));
}
+void SemaHLSL::handleVkPushConstantAttr(Decl *D, const ParsedAttr &AL) {
+ D->addAttr(::new (getASTContext())
+ HLSLVkPushConstantAttr(getASTContext(), AL));
+}
+
void SemaHLSL::handleVkConstantIdAttr(Decl *D, const ParsedAttr &AL) {
uint32_t Id;
if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(0), Id))
@@ -3837,12 +3842,15 @@ QualType SemaHLSL::getInoutParameterType(QualType Ty) {
return Ty;
}
-static bool IsDefaultBufferConstantDecl(VarDecl *VD) {
+static bool IsDefaultBufferConstantDecl(const ASTContext &Ctx, VarDecl *VD) {
+ bool IsVulkan =
+ Ctx.getTargetInfo().getTriple().getOS() == llvm::Triple::Vulkan;
+ bool IsVKPushConstant = IsVulkan && VD->hasAttr<HLSLVkPushConstantAttr>();
QualType QT = VD->getType();
return VD->getDeclContext()->isTranslationUnit() &&
QT.getAddressSpace() == LangAS::Default &&
VD->getStorageClass() != SC_Static &&
- !VD->hasAttr<HLSLVkConstantIdAttr>() &&
+ !VD->hasAttr<HLSLVkConstantIdAttr>() && !IsVKPushConstant &&
!isInvalidConstantBufferLeafElementType(QT.getTypePtr());
}
@@ -3863,6 +3871,19 @@ void SemaHLSL::deduceAddressSpace(VarDecl *Decl) {
return;
}
+ bool IsVulkan = getASTContext().getTargetInfo().getTriple().getOS() ==
+ llvm::Triple::Vulkan;
+ if (IsVulkan && Decl->hasAttr<HLSLVkPushConstantAttr>()) {
+ if (HasDeclaredAPushConstant)
+ SemaRef.Diag(Decl->getLocation(), diag::err_hlsl_push_constant_unique);
+
+ LangAS ImplAS = LangAS::hlsl_push_constant;
+ Type = SemaRef.getASTContext().getAddrSpaceQualType(Type, ImplAS);
+ Decl->setType(Type);
+ HasDeclaredAPushConstant = true;
+ return;
+ }
+
if (Type->isSamplerT() || Type->isVoidType())
return;
@@ -3895,7 +3916,7 @@ void SemaHLSL::ActOnVariableDeclarator(VarDecl *VD) {
// Global variables outside a cbuffer block that are not a resource, static,
// groupshared, or an empty array or struct belong to the default constant
// buffer $Globals (to be created at the end of the translation unit).
- if (IsDefaultBufferConstantDecl(VD)) {
+ if (IsDefaultBufferConstantDecl(getASTContext(), VD)) {
// update address space to hlsl_constant
QualType NewTy = getASTContext().getAddrSpaceQualType(
VD->getType(), LangAS::hlsl_constant);
@@ -4196,8 +4217,11 @@ void SemaHLSL::processExplicitBindingsOnDecl(VarDecl *VD) {
bool HasBinding = false;
for (Attr *A : VD->attrs()) {
- if (isa<HLSLVkBindingAttr>(A))
+ if (isa<HLSLVkBindingAttr>(A)) {
HasBinding = true;
+ if (auto PA = VD->getAttr<HLSLVkPushConstantAttr>())
+ Diag(PA->getLoc(), diag::err_hlsl_attr_incompatible) << A << PA;
+ }
HLSLResourceBindingAttr *RBA = dyn_cast<HLSLResourceBindingAttr>(A);
if (!RBA || !RBA->hasRegisterSlot())
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl
new file mode 100644
index 0000000000000..412ec4dffc572
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl
@@ -0,0 +1,20 @@
+// RUN: %clang_cc1 -triple spirv-pc-vulkan-compute -x hlsl -emit-llvm -finclude-default-header -disable-llvm-passes -o - %s | FileCheck %s
+
+struct S {
+ uint32_t a : 1;
+ uint32_t b : 1;
+};
+// CHECK: %struct.S = type { i8 }
+
+[[vk::push_constant]] S buffer;
+// CHECK: @buffer = external hidden addrspace(13) externally_initialized global %struct.S, align 1
+
+[numthreads(1, 1, 1)]
+void main() {
+ uint32_t v = buffer.b;
+// CHECK: %bf.load = load i8, ptr addrspace(13) @buffer, align 1
+// CHECK: %bf.lshr = lshr i8 %bf.load, 1
+// CHECK: %bf.clear = and i8 %bf.lshr, 1
+// CHECK: %bf.cast = zext i8 %bf.clear to i32
+// CHECK: store i32 %bf.cast
+}
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl
new file mode 100644
index 0000000000000..2b2e9d09c7ab0
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl
@@ -0,0 +1,17 @@
+// RUN: %clang_cc1 -triple spirv-pc-vulkan-compute -x hlsl -emit-llvm -finclude-default-header -disable-llvm-passes -o - %s | FileCheck %s
+
+[[vk::push_constant]]
+struct {
+ int a;
+ float b;
+ float3 c;
+}
+PushConstants;
+
+// CHECK: %struct.anon = type <{ i32, float, <3 x float> }>
+// CHECK: @PushConstants = external hidden addrspace(13) externally_initialized global %struct.anon, align 1
+
+[numthreads(1, 1, 1)]
+void main() {
+ float tmp = PushConstants.b;
+}
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl
new file mode 100644
index 0000000000000..6b58decfa5188
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl
@@ -0,0 +1,13 @@
+// RUN: %clang_cc1 -triple spirv-unknown-vulkan-compute -x hlsl -emit-llvm -disable-llvm-passes -o - -hlsl-entry main %s -verify
+
+struct S {
+ float f;
+};
+
+// expected-error@+1 {{'vk::binding' attribute is not compatible with 'vk::push_constant' attribute}}
+[[vk::push_constant, vk::binding(5)]]
+S pcs;
+
+[numthreads(1, 1, 1)]
+void main() {
+}
diff...
[truncated]
|
Member
|
@llvm/pr-subscribers-backend-systemz Author: Nathan Gauër (Keenuts) ChangesImplements initial support for vk::push_constant. The old fix would be to use target specific types, but this is Patch is 27.93 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/166793.diff 33 Files Affected:
diff --git a/clang/include/clang/Basic/AddressSpaces.h b/clang/include/clang/Basic/AddressSpaces.h
index 48e4a1c61fe02..7280b8fc923d2 100644
--- a/clang/include/clang/Basic/AddressSpaces.h
+++ b/clang/include/clang/Basic/AddressSpaces.h
@@ -62,6 +62,7 @@ enum class LangAS : unsigned {
hlsl_private,
hlsl_device,
hlsl_input,
+ hlsl_push_constant,
// Wasm specific address spaces.
wasm_funcref,
diff --git a/clang/include/clang/Basic/Attr.td b/clang/include/clang/Basic/Attr.td
index 8dfe4bc08c48e..e00765a57cb23 100644
--- a/clang/include/clang/Basic/Attr.td
+++ b/clang/include/clang/Basic/Attr.td
@@ -5146,6 +5146,14 @@ def HLSLVkExtBuiltinInput : InheritableAttr {
let Documentation = [HLSLVkExtBuiltinInputDocs];
}
+def HLSLVkPushConstant : InheritableAttr {
+ let Spellings = [CXX11<"vk", "push_constant">];
+ let Args = [];
+ let Subjects = SubjectList<[GlobalVar], ErrorDiag>;
+ let LangOpts = [HLSL];
+ let Documentation = [HLSLVkPushConstantDocs];
+}
+
def HLSLVkConstantId : InheritableAttr {
let Spellings = [CXX11<"vk", "constant_id">];
let Args = [IntArgument<"Id">];
diff --git a/clang/include/clang/Basic/AttrDocs.td b/clang/include/clang/Basic/AttrDocs.td
index 4813191d2d602..3938c624c9d0c 100644
--- a/clang/include/clang/Basic/AttrDocs.td
+++ b/clang/include/clang/Basic/AttrDocs.td
@@ -8777,6 +8777,11 @@ https://github.com/microsoft/hlsl-specs/blob/main/proposals/0011-inline-spirv.md
}];
}
+def HLSLVkPushConstantDocs : Documentation {
+ let Category = DocCatVariable;
+ let Content = [{ FIXME }];
+}
+
def AnnotateTypeDocs : Documentation {
let Category = DocCatType;
let Heading = "annotate_type";
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index a6e60fe4692ee..c9d128d77c90f 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -13183,6 +13183,9 @@ def err_hlsl_attr_invalid_type : Error<
"attribute %0 only applies to a field or parameter of type '%1'">;
def err_hlsl_attr_invalid_ast_node : Error<
"attribute %0 only applies to %1">;
+def err_hlsl_attr_incompatible
+ : Error<"%0 attribute is not compatible with %1 attribute">;
+
def err_hlsl_entry_shader_attr_mismatch : Error<
"%0 attribute on entry function does not match the target profile">;
def err_hlsl_numthreads_argument_oor : Error<"argument '%select{X|Y|Z}0' to numthreads attribute cannot exceed %1">;
@@ -13294,6 +13297,9 @@ def err_hlsl_incomplete_resource_array_in_function_param: Error<
def err_hlsl_assign_to_global_resource: Error<
"assignment to global resource variable %0 is not allowed">;
+def err_hlsl_push_constant_unique
+ : Error<"cannot have more than one push constant block">;
+
// Layout randomization diagnostics.
def err_non_designated_init_used : Error<
"a randomized struct can only be initialized with a designated initializer">;
diff --git a/clang/include/clang/Basic/HLSLRuntime.h b/clang/include/clang/Basic/HLSLRuntime.h
index 03166805daa6a..f6a1cf9636467 100644
--- a/clang/include/clang/Basic/HLSLRuntime.h
+++ b/clang/include/clang/Basic/HLSLRuntime.h
@@ -14,6 +14,7 @@
#ifndef CLANG_BASIC_HLSLRUNTIME_H
#define CLANG_BASIC_HLSLRUNTIME_H
+#include "clang/Basic/AddressSpaces.h"
#include "clang/Basic/LangOptions.h"
#include <cstdint>
@@ -30,6 +31,10 @@ getStageFromEnvironment(const llvm::Triple::EnvironmentType &E) {
return static_cast<ShaderStage>(Pipeline);
}
+constexpr bool isInitializedByPipeline(LangAS AS) {
+ return AS == LangAS::hlsl_input || AS == LangAS::hlsl_push_constant;
+}
+
#define ENUM_COMPARE_ASSERT(Value) \
static_assert( \
getStageFromEnvironment(llvm::Triple::Value) == ShaderStage::Value, \
diff --git a/clang/include/clang/Sema/SemaHLSL.h b/clang/include/clang/Sema/SemaHLSL.h
index 86da323892f98..2fcac237eba1c 100644
--- a/clang/include/clang/Sema/SemaHLSL.h
+++ b/clang/include/clang/Sema/SemaHLSL.h
@@ -190,6 +190,7 @@ class SemaHLSL : public SemaBase {
void handleSemanticAttr(Decl *D, const ParsedAttr &AL);
void handleVkExtBuiltinInputAttr(Decl *D, const ParsedAttr &AL);
+ void handleVkPushConstantAttr(Decl *D, const ParsedAttr &AL);
bool CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall);
QualType ProcessResourceTypeAttributes(QualType Wrapped);
@@ -239,6 +240,8 @@ class SemaHLSL : public SemaBase {
IdentifierInfo *RootSigOverrideIdent = nullptr;
+ bool HasDeclaredAPushConstant = false;
+
struct SemanticInfo {
HLSLParsedSemanticAttr *Semantic;
std::optional<uint32_t> Index;
diff --git a/clang/lib/AST/Type.cpp b/clang/lib/AST/Type.cpp
index 4548af17e37f2..53082bcf78f6a 100644
--- a/clang/lib/AST/Type.cpp
+++ b/clang/lib/AST/Type.cpp
@@ -101,6 +101,7 @@ bool Qualifiers::isTargetAddressSpaceSupersetOf(LangAS A, LangAS B,
(A == LangAS::Default && B == LangAS::hlsl_private) ||
(A == LangAS::Default && B == LangAS::hlsl_device) ||
(A == LangAS::Default && B == LangAS::hlsl_input) ||
+ (A == LangAS::Default && B == LangAS::hlsl_push_constant) ||
// Conversions from target specific address spaces may be legal
// depending on the target information.
Ctx.getTargetInfo().isAddressSpaceSupersetOf(A, B);
diff --git a/clang/lib/AST/TypePrinter.cpp b/clang/lib/AST/TypePrinter.cpp
index c18b2eafc722c..8448dd3748e28 100644
--- a/clang/lib/AST/TypePrinter.cpp
+++ b/clang/lib/AST/TypePrinter.cpp
@@ -2749,6 +2749,8 @@ std::string Qualifiers::getAddrSpaceAsString(LangAS AS) {
return "hlsl_device";
case LangAS::hlsl_input:
return "hlsl_input";
+ case LangAS::hlsl_push_constant:
+ return "hlsl_push_constant";
case LangAS::wasm_funcref:
return "__funcref";
default:
diff --git a/clang/lib/Basic/TargetInfo.cpp b/clang/lib/Basic/TargetInfo.cpp
index ffaf98bf9c366..92ca7a66a9593 100644
--- a/clang/lib/Basic/TargetInfo.cpp
+++ b/clang/lib/Basic/TargetInfo.cpp
@@ -52,6 +52,7 @@ static const LangASMap FakeAddrSpaceMap = {
15, // hlsl_private
16, // hlsl_device
17, // hlsl_input
+ 18, // hlsl_push_constant
20, // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/AArch64.h b/clang/lib/Basic/Targets/AArch64.h
index 1a7aa658e9d87..8e8d8e6ae86b5 100644
--- a/clang/lib/Basic/Targets/AArch64.h
+++ b/clang/lib/Basic/Targets/AArch64.h
@@ -48,6 +48,7 @@ static const unsigned ARM64AddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/AMDGPU.cpp b/clang/lib/Basic/Targets/AMDGPU.cpp
index d4d696b8456b6..993a73a89c9e9 100644
--- a/clang/lib/Basic/Targets/AMDGPU.cpp
+++ b/clang/lib/Basic/Targets/AMDGPU.cpp
@@ -63,6 +63,7 @@ const LangASMap AMDGPUTargetInfo::AMDGPUDefIsGenMap = {
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_private
llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_device
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_input
+ llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_push_constant
};
const LangASMap AMDGPUTargetInfo::AMDGPUDefIsPrivMap = {
@@ -91,6 +92,7 @@ const LangASMap AMDGPUTargetInfo::AMDGPUDefIsPrivMap = {
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_private
llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_device
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_input
+ llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_push_constant
};
} // namespace targets
} // namespace clang
diff --git a/clang/lib/Basic/Targets/DirectX.h b/clang/lib/Basic/Targets/DirectX.h
index a21a593365773..c0799a6f7610f 100644
--- a/clang/lib/Basic/Targets/DirectX.h
+++ b/clang/lib/Basic/Targets/DirectX.h
@@ -46,6 +46,7 @@ static const unsigned DirectXAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/NVPTX.h b/clang/lib/Basic/Targets/NVPTX.h
index f5c8396f398aa..6338a4f2f9036 100644
--- a/clang/lib/Basic/Targets/NVPTX.h
+++ b/clang/lib/Basic/Targets/NVPTX.h
@@ -50,6 +50,7 @@ static const unsigned NVPTXAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/SPIR.h b/clang/lib/Basic/Targets/SPIR.h
index 22b2799518dd0..94449231efb94 100644
--- a/clang/lib/Basic/Targets/SPIR.h
+++ b/clang/lib/Basic/Targets/SPIR.h
@@ -51,6 +51,7 @@ static const unsigned SPIRDefIsPrivMap[] = {
10, // hlsl_private
11, // hlsl_device
7, // hlsl_input
+ 13, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
@@ -87,6 +88,7 @@ static const unsigned SPIRDefIsGenMap[] = {
10, // hlsl_private
11, // hlsl_device
7, // hlsl_input
+ 13, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/SystemZ.h b/clang/lib/Basic/Targets/SystemZ.h
index 4e15d5af1cde6..4ce515b31a001 100644
--- a/clang/lib/Basic/Targets/SystemZ.h
+++ b/clang/lib/Basic/Targets/SystemZ.h
@@ -46,6 +46,7 @@ static const unsigned ZOSAddressMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
0 // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/TCE.h b/clang/lib/Basic/Targets/TCE.h
index 005cab9819472..161025378c471 100644
--- a/clang/lib/Basic/Targets/TCE.h
+++ b/clang/lib/Basic/Targets/TCE.h
@@ -55,6 +55,7 @@ static const unsigned TCEOpenCLAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/WebAssembly.h b/clang/lib/Basic/Targets/WebAssembly.h
index 4de6ce6bb5a21..c8065843aeb42 100644
--- a/clang/lib/Basic/Targets/WebAssembly.h
+++ b/clang/lib/Basic/Targets/WebAssembly.h
@@ -46,6 +46,7 @@ static const unsigned WebAssemblyAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
20, // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/X86.h b/clang/lib/Basic/Targets/X86.h
index e7da2622e78b5..7b88ac70e234f 100644
--- a/clang/lib/Basic/Targets/X86.h
+++ b/clang/lib/Basic/Targets/X86.h
@@ -50,6 +50,7 @@ static const unsigned X86AddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 3eeb1718e455a..c3b536c7a267f 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -6049,9 +6049,11 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D,
getCUDARuntime().handleVarRegistration(D, *GV);
}
- if (LangOpts.HLSL && GetGlobalVarAddressSpace(D) == LangAS::hlsl_input) {
+ if (LangOpts.HLSL &&
+ hlsl::isInitializedByPipeline(GetGlobalVarAddressSpace(D))) {
// HLSL Input variables are considered to be set by the driver/pipeline, but
- // only visible to a single thread/wave.
+ // only visible to a single thread/wave. Push constants are also externally
+ // initialized, but constant, hence cross-wave visibility is not relevant.
GV->setExternallyInitialized(true);
} else {
GV->setInitializer(Init);
@@ -6102,10 +6104,11 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D,
!D->hasAttr<ConstInitAttr>())
Linkage = llvm::GlobalValue::InternalLinkage;
- // HLSL variables in the input address space maps like memory-mapped
- // variables. Even if they are 'static', they are externally initialized and
- // read/write by the hardware/driver/pipeline.
- if (LangOpts.HLSL && GetGlobalVarAddressSpace(D) == LangAS::hlsl_input)
+ // HLSL variables in the input or push-constant address space maps are like
+ // memory-mapped variables. Even if they are 'static', they are externally
+ // initialized and read/write by the hardware/driver/pipeline.
+ if (LangOpts.HLSL &&
+ hlsl::isInitializedByPipeline(GetGlobalVarAddressSpace(D)))
Linkage = llvm::GlobalValue::ExternalLinkage;
GV->setLinkage(Linkage);
diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
index 25b89d65847ad..5c29bf5e77414 100644
--- a/clang/lib/Sema/SemaDecl.cpp
+++ b/clang/lib/Sema/SemaDecl.cpp
@@ -30,6 +30,7 @@
#include "clang/AST/Type.h"
#include "clang/Basic/Builtins.h"
#include "clang/Basic/DiagnosticComment.h"
+#include "clang/Basic/HLSLRuntime.h"
#include "clang/Basic/PartialDiagnostic.h"
#include "clang/Basic/SourceManager.h"
#include "clang/Basic/TargetInfo.h"
@@ -14559,10 +14560,10 @@ void Sema::ActOnUninitializedDecl(Decl *RealDecl) {
if (getLangOpts().HLSL && HLSL().ActOnUninitializedVarDecl(Var))
return;
- // HLSL input variables are expected to be externally initialized, even
- // when marked `static`.
+ // HLSL input & push-constant variables are expected to be externally
+ // initialized, even when marked `static`.
if (getLangOpts().HLSL &&
- Var->getType().getAddressSpace() == LangAS::hlsl_input)
+ hlsl::isInitializedByPipeline(Var->getType().getAddressSpace()))
return;
// C++03 [dcl.init]p9:
diff --git a/clang/lib/Sema/SemaDeclAttr.cpp b/clang/lib/Sema/SemaDeclAttr.cpp
index a9e7b44ac9d73..0396155bd6a9d 100644
--- a/clang/lib/Sema/SemaDeclAttr.cpp
+++ b/clang/lib/Sema/SemaDeclAttr.cpp
@@ -7614,6 +7614,9 @@ ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D, const ParsedAttr &AL,
case ParsedAttr::AT_HLSLVkExtBuiltinInput:
S.HLSL().handleVkExtBuiltinInputAttr(D, AL);
break;
+ case ParsedAttr::AT_HLSLVkPushConstant:
+ S.HLSL().handleVkPushConstantAttr(D, AL);
+ break;
case ParsedAttr::AT_HLSLVkConstantId:
S.HLSL().handleVkConstantIdAttr(D, AL);
break;
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 2b9b3abbd5360..1831584c88697 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -1667,6 +1667,11 @@ void SemaHLSL::handleVkExtBuiltinInputAttr(Decl *D, const ParsedAttr &AL) {
HLSLVkExtBuiltinInputAttr(getASTContext(), AL, ID));
}
+void SemaHLSL::handleVkPushConstantAttr(Decl *D, const ParsedAttr &AL) {
+ D->addAttr(::new (getASTContext())
+ HLSLVkPushConstantAttr(getASTContext(), AL));
+}
+
void SemaHLSL::handleVkConstantIdAttr(Decl *D, const ParsedAttr &AL) {
uint32_t Id;
if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(0), Id))
@@ -3837,12 +3842,15 @@ QualType SemaHLSL::getInoutParameterType(QualType Ty) {
return Ty;
}
-static bool IsDefaultBufferConstantDecl(VarDecl *VD) {
+static bool IsDefaultBufferConstantDecl(const ASTContext &Ctx, VarDecl *VD) {
+ bool IsVulkan =
+ Ctx.getTargetInfo().getTriple().getOS() == llvm::Triple::Vulkan;
+ bool IsVKPushConstant = IsVulkan && VD->hasAttr<HLSLVkPushConstantAttr>();
QualType QT = VD->getType();
return VD->getDeclContext()->isTranslationUnit() &&
QT.getAddressSpace() == LangAS::Default &&
VD->getStorageClass() != SC_Static &&
- !VD->hasAttr<HLSLVkConstantIdAttr>() &&
+ !VD->hasAttr<HLSLVkConstantIdAttr>() && !IsVKPushConstant &&
!isInvalidConstantBufferLeafElementType(QT.getTypePtr());
}
@@ -3863,6 +3871,19 @@ void SemaHLSL::deduceAddressSpace(VarDecl *Decl) {
return;
}
+ bool IsVulkan = getASTContext().getTargetInfo().getTriple().getOS() ==
+ llvm::Triple::Vulkan;
+ if (IsVulkan && Decl->hasAttr<HLSLVkPushConstantAttr>()) {
+ if (HasDeclaredAPushConstant)
+ SemaRef.Diag(Decl->getLocation(), diag::err_hlsl_push_constant_unique);
+
+ LangAS ImplAS = LangAS::hlsl_push_constant;
+ Type = SemaRef.getASTContext().getAddrSpaceQualType(Type, ImplAS);
+ Decl->setType(Type);
+ HasDeclaredAPushConstant = true;
+ return;
+ }
+
if (Type->isSamplerT() || Type->isVoidType())
return;
@@ -3895,7 +3916,7 @@ void SemaHLSL::ActOnVariableDeclarator(VarDecl *VD) {
// Global variables outside a cbuffer block that are not a resource, static,
// groupshared, or an empty array or struct belong to the default constant
// buffer $Globals (to be created at the end of the translation unit).
- if (IsDefaultBufferConstantDecl(VD)) {
+ if (IsDefaultBufferConstantDecl(getASTContext(), VD)) {
// update address space to hlsl_constant
QualType NewTy = getASTContext().getAddrSpaceQualType(
VD->getType(), LangAS::hlsl_constant);
@@ -4196,8 +4217,11 @@ void SemaHLSL::processExplicitBindingsOnDecl(VarDecl *VD) {
bool HasBinding = false;
for (Attr *A : VD->attrs()) {
- if (isa<HLSLVkBindingAttr>(A))
+ if (isa<HLSLVkBindingAttr>(A)) {
HasBinding = true;
+ if (auto PA = VD->getAttr<HLSLVkPushConstantAttr>())
+ Diag(PA->getLoc(), diag::err_hlsl_attr_incompatible) << A << PA;
+ }
HLSLResourceBindingAttr *RBA = dyn_cast<HLSLResourceBindingAttr>(A);
if (!RBA || !RBA->hasRegisterSlot())
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl
new file mode 100644
index 0000000000000..412ec4dffc572
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl
@@ -0,0 +1,20 @@
+// RUN: %clang_cc1 -triple spirv-pc-vulkan-compute -x hlsl -emit-llvm -finclude-default-header -disable-llvm-passes -o - %s | FileCheck %s
+
+struct S {
+ uint32_t a : 1;
+ uint32_t b : 1;
+};
+// CHECK: %struct.S = type { i8 }
+
+[[vk::push_constant]] S buffer;
+// CHECK: @buffer = external hidden addrspace(13) externally_initialized global %struct.S, align 1
+
+[numthreads(1, 1, 1)]
+void main() {
+ uint32_t v = buffer.b;
+// CHECK: %bf.load = load i8, ptr addrspace(13) @buffer, align 1
+// CHECK: %bf.lshr = lshr i8 %bf.load, 1
+// CHECK: %bf.clear = and i8 %bf.lshr, 1
+// CHECK: %bf.cast = zext i8 %bf.clear to i32
+// CHECK: store i32 %bf.cast
+}
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl
new file mode 100644
index 0000000000000..2b2e9d09c7ab0
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl
@@ -0,0 +1,17 @@
+// RUN: %clang_cc1 -triple spirv-pc-vulkan-compute -x hlsl -emit-llvm -finclude-default-header -disable-llvm-passes -o - %s | FileCheck %s
+
+[[vk::push_constant]]
+struct {
+ int a;
+ float b;
+ float3 c;
+}
+PushConstants;
+
+// CHECK: %struct.anon = type <{ i32, float, <3 x float> }>
+// CHECK: @PushConstants = external hidden addrspace(13) externally_initialized global %struct.anon, align 1
+
+[numthreads(1, 1, 1)]
+void main() {
+ float tmp = PushConstants.b;
+}
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl
new file mode 100644
index 0000000000000..6b58decfa5188
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl
@@ -0,0 +1,13 @@
+// RUN: %clang_cc1 -triple spirv-unknown-vulkan-compute -x hlsl -emit-llvm -disable-llvm-passes -o - -hlsl-entry main %s -verify
+
+struct S {
+ float f;
+};
+
+// expected-error@+1 {{'vk::binding' attribute is not compatible with 'vk::push_constant' attribute}}
+[[vk::push_constant, vk::binding(5)]]
+S pcs;
+
+[numthreads(1, 1, 1)]
+void main() {
+}
diff...
[truncated]
|
Member
|
@llvm/pr-subscribers-backend-amdgpu Author: Nathan Gauër (Keenuts) ChangesImplements initial support for vk::push_constant. The old fix would be to use target specific types, but this is Patch is 27.93 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/166793.diff 33 Files Affected:
diff --git a/clang/include/clang/Basic/AddressSpaces.h b/clang/include/clang/Basic/AddressSpaces.h
index 48e4a1c61fe02..7280b8fc923d2 100644
--- a/clang/include/clang/Basic/AddressSpaces.h
+++ b/clang/include/clang/Basic/AddressSpaces.h
@@ -62,6 +62,7 @@ enum class LangAS : unsigned {
hlsl_private,
hlsl_device,
hlsl_input,
+ hlsl_push_constant,
// Wasm specific address spaces.
wasm_funcref,
diff --git a/clang/include/clang/Basic/Attr.td b/clang/include/clang/Basic/Attr.td
index 8dfe4bc08c48e..e00765a57cb23 100644
--- a/clang/include/clang/Basic/Attr.td
+++ b/clang/include/clang/Basic/Attr.td
@@ -5146,6 +5146,14 @@ def HLSLVkExtBuiltinInput : InheritableAttr {
let Documentation = [HLSLVkExtBuiltinInputDocs];
}
+def HLSLVkPushConstant : InheritableAttr {
+ let Spellings = [CXX11<"vk", "push_constant">];
+ let Args = [];
+ let Subjects = SubjectList<[GlobalVar], ErrorDiag>;
+ let LangOpts = [HLSL];
+ let Documentation = [HLSLVkPushConstantDocs];
+}
+
def HLSLVkConstantId : InheritableAttr {
let Spellings = [CXX11<"vk", "constant_id">];
let Args = [IntArgument<"Id">];
diff --git a/clang/include/clang/Basic/AttrDocs.td b/clang/include/clang/Basic/AttrDocs.td
index 4813191d2d602..3938c624c9d0c 100644
--- a/clang/include/clang/Basic/AttrDocs.td
+++ b/clang/include/clang/Basic/AttrDocs.td
@@ -8777,6 +8777,11 @@ https://github.com/microsoft/hlsl-specs/blob/main/proposals/0011-inline-spirv.md
}];
}
+def HLSLVkPushConstantDocs : Documentation {
+ let Category = DocCatVariable;
+ let Content = [{ FIXME }];
+}
+
def AnnotateTypeDocs : Documentation {
let Category = DocCatType;
let Heading = "annotate_type";
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index a6e60fe4692ee..c9d128d77c90f 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -13183,6 +13183,9 @@ def err_hlsl_attr_invalid_type : Error<
"attribute %0 only applies to a field or parameter of type '%1'">;
def err_hlsl_attr_invalid_ast_node : Error<
"attribute %0 only applies to %1">;
+def err_hlsl_attr_incompatible
+ : Error<"%0 attribute is not compatible with %1 attribute">;
+
def err_hlsl_entry_shader_attr_mismatch : Error<
"%0 attribute on entry function does not match the target profile">;
def err_hlsl_numthreads_argument_oor : Error<"argument '%select{X|Y|Z}0' to numthreads attribute cannot exceed %1">;
@@ -13294,6 +13297,9 @@ def err_hlsl_incomplete_resource_array_in_function_param: Error<
def err_hlsl_assign_to_global_resource: Error<
"assignment to global resource variable %0 is not allowed">;
+def err_hlsl_push_constant_unique
+ : Error<"cannot have more than one push constant block">;
+
// Layout randomization diagnostics.
def err_non_designated_init_used : Error<
"a randomized struct can only be initialized with a designated initializer">;
diff --git a/clang/include/clang/Basic/HLSLRuntime.h b/clang/include/clang/Basic/HLSLRuntime.h
index 03166805daa6a..f6a1cf9636467 100644
--- a/clang/include/clang/Basic/HLSLRuntime.h
+++ b/clang/include/clang/Basic/HLSLRuntime.h
@@ -14,6 +14,7 @@
#ifndef CLANG_BASIC_HLSLRUNTIME_H
#define CLANG_BASIC_HLSLRUNTIME_H
+#include "clang/Basic/AddressSpaces.h"
#include "clang/Basic/LangOptions.h"
#include <cstdint>
@@ -30,6 +31,10 @@ getStageFromEnvironment(const llvm::Triple::EnvironmentType &E) {
return static_cast<ShaderStage>(Pipeline);
}
+constexpr bool isInitializedByPipeline(LangAS AS) {
+ return AS == LangAS::hlsl_input || AS == LangAS::hlsl_push_constant;
+}
+
#define ENUM_COMPARE_ASSERT(Value) \
static_assert( \
getStageFromEnvironment(llvm::Triple::Value) == ShaderStage::Value, \
diff --git a/clang/include/clang/Sema/SemaHLSL.h b/clang/include/clang/Sema/SemaHLSL.h
index 86da323892f98..2fcac237eba1c 100644
--- a/clang/include/clang/Sema/SemaHLSL.h
+++ b/clang/include/clang/Sema/SemaHLSL.h
@@ -190,6 +190,7 @@ class SemaHLSL : public SemaBase {
void handleSemanticAttr(Decl *D, const ParsedAttr &AL);
void handleVkExtBuiltinInputAttr(Decl *D, const ParsedAttr &AL);
+ void handleVkPushConstantAttr(Decl *D, const ParsedAttr &AL);
bool CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall);
QualType ProcessResourceTypeAttributes(QualType Wrapped);
@@ -239,6 +240,8 @@ class SemaHLSL : public SemaBase {
IdentifierInfo *RootSigOverrideIdent = nullptr;
+ bool HasDeclaredAPushConstant = false;
+
struct SemanticInfo {
HLSLParsedSemanticAttr *Semantic;
std::optional<uint32_t> Index;
diff --git a/clang/lib/AST/Type.cpp b/clang/lib/AST/Type.cpp
index 4548af17e37f2..53082bcf78f6a 100644
--- a/clang/lib/AST/Type.cpp
+++ b/clang/lib/AST/Type.cpp
@@ -101,6 +101,7 @@ bool Qualifiers::isTargetAddressSpaceSupersetOf(LangAS A, LangAS B,
(A == LangAS::Default && B == LangAS::hlsl_private) ||
(A == LangAS::Default && B == LangAS::hlsl_device) ||
(A == LangAS::Default && B == LangAS::hlsl_input) ||
+ (A == LangAS::Default && B == LangAS::hlsl_push_constant) ||
// Conversions from target specific address spaces may be legal
// depending on the target information.
Ctx.getTargetInfo().isAddressSpaceSupersetOf(A, B);
diff --git a/clang/lib/AST/TypePrinter.cpp b/clang/lib/AST/TypePrinter.cpp
index c18b2eafc722c..8448dd3748e28 100644
--- a/clang/lib/AST/TypePrinter.cpp
+++ b/clang/lib/AST/TypePrinter.cpp
@@ -2749,6 +2749,8 @@ std::string Qualifiers::getAddrSpaceAsString(LangAS AS) {
return "hlsl_device";
case LangAS::hlsl_input:
return "hlsl_input";
+ case LangAS::hlsl_push_constant:
+ return "hlsl_push_constant";
case LangAS::wasm_funcref:
return "__funcref";
default:
diff --git a/clang/lib/Basic/TargetInfo.cpp b/clang/lib/Basic/TargetInfo.cpp
index ffaf98bf9c366..92ca7a66a9593 100644
--- a/clang/lib/Basic/TargetInfo.cpp
+++ b/clang/lib/Basic/TargetInfo.cpp
@@ -52,6 +52,7 @@ static const LangASMap FakeAddrSpaceMap = {
15, // hlsl_private
16, // hlsl_device
17, // hlsl_input
+ 18, // hlsl_push_constant
20, // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/AArch64.h b/clang/lib/Basic/Targets/AArch64.h
index 1a7aa658e9d87..8e8d8e6ae86b5 100644
--- a/clang/lib/Basic/Targets/AArch64.h
+++ b/clang/lib/Basic/Targets/AArch64.h
@@ -48,6 +48,7 @@ static const unsigned ARM64AddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/AMDGPU.cpp b/clang/lib/Basic/Targets/AMDGPU.cpp
index d4d696b8456b6..993a73a89c9e9 100644
--- a/clang/lib/Basic/Targets/AMDGPU.cpp
+++ b/clang/lib/Basic/Targets/AMDGPU.cpp
@@ -63,6 +63,7 @@ const LangASMap AMDGPUTargetInfo::AMDGPUDefIsGenMap = {
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_private
llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_device
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_input
+ llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_push_constant
};
const LangASMap AMDGPUTargetInfo::AMDGPUDefIsPrivMap = {
@@ -91,6 +92,7 @@ const LangASMap AMDGPUTargetInfo::AMDGPUDefIsPrivMap = {
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_private
llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_device
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_input
+ llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_push_constant
};
} // namespace targets
} // namespace clang
diff --git a/clang/lib/Basic/Targets/DirectX.h b/clang/lib/Basic/Targets/DirectX.h
index a21a593365773..c0799a6f7610f 100644
--- a/clang/lib/Basic/Targets/DirectX.h
+++ b/clang/lib/Basic/Targets/DirectX.h
@@ -46,6 +46,7 @@ static const unsigned DirectXAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/NVPTX.h b/clang/lib/Basic/Targets/NVPTX.h
index f5c8396f398aa..6338a4f2f9036 100644
--- a/clang/lib/Basic/Targets/NVPTX.h
+++ b/clang/lib/Basic/Targets/NVPTX.h
@@ -50,6 +50,7 @@ static const unsigned NVPTXAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/SPIR.h b/clang/lib/Basic/Targets/SPIR.h
index 22b2799518dd0..94449231efb94 100644
--- a/clang/lib/Basic/Targets/SPIR.h
+++ b/clang/lib/Basic/Targets/SPIR.h
@@ -51,6 +51,7 @@ static const unsigned SPIRDefIsPrivMap[] = {
10, // hlsl_private
11, // hlsl_device
7, // hlsl_input
+ 13, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
@@ -87,6 +88,7 @@ static const unsigned SPIRDefIsGenMap[] = {
10, // hlsl_private
11, // hlsl_device
7, // hlsl_input
+ 13, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/SystemZ.h b/clang/lib/Basic/Targets/SystemZ.h
index 4e15d5af1cde6..4ce515b31a001 100644
--- a/clang/lib/Basic/Targets/SystemZ.h
+++ b/clang/lib/Basic/Targets/SystemZ.h
@@ -46,6 +46,7 @@ static const unsigned ZOSAddressMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
0 // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/TCE.h b/clang/lib/Basic/Targets/TCE.h
index 005cab9819472..161025378c471 100644
--- a/clang/lib/Basic/Targets/TCE.h
+++ b/clang/lib/Basic/Targets/TCE.h
@@ -55,6 +55,7 @@ static const unsigned TCEOpenCLAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/WebAssembly.h b/clang/lib/Basic/Targets/WebAssembly.h
index 4de6ce6bb5a21..c8065843aeb42 100644
--- a/clang/lib/Basic/Targets/WebAssembly.h
+++ b/clang/lib/Basic/Targets/WebAssembly.h
@@ -46,6 +46,7 @@ static const unsigned WebAssemblyAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
20, // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/X86.h b/clang/lib/Basic/Targets/X86.h
index e7da2622e78b5..7b88ac70e234f 100644
--- a/clang/lib/Basic/Targets/X86.h
+++ b/clang/lib/Basic/Targets/X86.h
@@ -50,6 +50,7 @@ static const unsigned X86AddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 3eeb1718e455a..c3b536c7a267f 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -6049,9 +6049,11 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D,
getCUDARuntime().handleVarRegistration(D, *GV);
}
- if (LangOpts.HLSL && GetGlobalVarAddressSpace(D) == LangAS::hlsl_input) {
+ if (LangOpts.HLSL &&
+ hlsl::isInitializedByPipeline(GetGlobalVarAddressSpace(D))) {
// HLSL Input variables are considered to be set by the driver/pipeline, but
- // only visible to a single thread/wave.
+ // only visible to a single thread/wave. Push constants are also externally
+ // initialized, but constant, hence cross-wave visibility is not relevant.
GV->setExternallyInitialized(true);
} else {
GV->setInitializer(Init);
@@ -6102,10 +6104,11 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D,
!D->hasAttr<ConstInitAttr>())
Linkage = llvm::GlobalValue::InternalLinkage;
- // HLSL variables in the input address space maps like memory-mapped
- // variables. Even if they are 'static', they are externally initialized and
- // read/write by the hardware/driver/pipeline.
- if (LangOpts.HLSL && GetGlobalVarAddressSpace(D) == LangAS::hlsl_input)
+ // HLSL variables in the input or push-constant address space maps are like
+ // memory-mapped variables. Even if they are 'static', they are externally
+ // initialized and read/write by the hardware/driver/pipeline.
+ if (LangOpts.HLSL &&
+ hlsl::isInitializedByPipeline(GetGlobalVarAddressSpace(D)))
Linkage = llvm::GlobalValue::ExternalLinkage;
GV->setLinkage(Linkage);
diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
index 25b89d65847ad..5c29bf5e77414 100644
--- a/clang/lib/Sema/SemaDecl.cpp
+++ b/clang/lib/Sema/SemaDecl.cpp
@@ -30,6 +30,7 @@
#include "clang/AST/Type.h"
#include "clang/Basic/Builtins.h"
#include "clang/Basic/DiagnosticComment.h"
+#include "clang/Basic/HLSLRuntime.h"
#include "clang/Basic/PartialDiagnostic.h"
#include "clang/Basic/SourceManager.h"
#include "clang/Basic/TargetInfo.h"
@@ -14559,10 +14560,10 @@ void Sema::ActOnUninitializedDecl(Decl *RealDecl) {
if (getLangOpts().HLSL && HLSL().ActOnUninitializedVarDecl(Var))
return;
- // HLSL input variables are expected to be externally initialized, even
- // when marked `static`.
+ // HLSL input & push-constant variables are expected to be externally
+ // initialized, even when marked `static`.
if (getLangOpts().HLSL &&
- Var->getType().getAddressSpace() == LangAS::hlsl_input)
+ hlsl::isInitializedByPipeline(Var->getType().getAddressSpace()))
return;
// C++03 [dcl.init]p9:
diff --git a/clang/lib/Sema/SemaDeclAttr.cpp b/clang/lib/Sema/SemaDeclAttr.cpp
index a9e7b44ac9d73..0396155bd6a9d 100644
--- a/clang/lib/Sema/SemaDeclAttr.cpp
+++ b/clang/lib/Sema/SemaDeclAttr.cpp
@@ -7614,6 +7614,9 @@ ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D, const ParsedAttr &AL,
case ParsedAttr::AT_HLSLVkExtBuiltinInput:
S.HLSL().handleVkExtBuiltinInputAttr(D, AL);
break;
+ case ParsedAttr::AT_HLSLVkPushConstant:
+ S.HLSL().handleVkPushConstantAttr(D, AL);
+ break;
case ParsedAttr::AT_HLSLVkConstantId:
S.HLSL().handleVkConstantIdAttr(D, AL);
break;
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 2b9b3abbd5360..1831584c88697 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -1667,6 +1667,11 @@ void SemaHLSL::handleVkExtBuiltinInputAttr(Decl *D, const ParsedAttr &AL) {
HLSLVkExtBuiltinInputAttr(getASTContext(), AL, ID));
}
+void SemaHLSL::handleVkPushConstantAttr(Decl *D, const ParsedAttr &AL) {
+ D->addAttr(::new (getASTContext())
+ HLSLVkPushConstantAttr(getASTContext(), AL));
+}
+
void SemaHLSL::handleVkConstantIdAttr(Decl *D, const ParsedAttr &AL) {
uint32_t Id;
if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(0), Id))
@@ -3837,12 +3842,15 @@ QualType SemaHLSL::getInoutParameterType(QualType Ty) {
return Ty;
}
-static bool IsDefaultBufferConstantDecl(VarDecl *VD) {
+static bool IsDefaultBufferConstantDecl(const ASTContext &Ctx, VarDecl *VD) {
+ bool IsVulkan =
+ Ctx.getTargetInfo().getTriple().getOS() == llvm::Triple::Vulkan;
+ bool IsVKPushConstant = IsVulkan && VD->hasAttr<HLSLVkPushConstantAttr>();
QualType QT = VD->getType();
return VD->getDeclContext()->isTranslationUnit() &&
QT.getAddressSpace() == LangAS::Default &&
VD->getStorageClass() != SC_Static &&
- !VD->hasAttr<HLSLVkConstantIdAttr>() &&
+ !VD->hasAttr<HLSLVkConstantIdAttr>() && !IsVKPushConstant &&
!isInvalidConstantBufferLeafElementType(QT.getTypePtr());
}
@@ -3863,6 +3871,19 @@ void SemaHLSL::deduceAddressSpace(VarDecl *Decl) {
return;
}
+ bool IsVulkan = getASTContext().getTargetInfo().getTriple().getOS() ==
+ llvm::Triple::Vulkan;
+ if (IsVulkan && Decl->hasAttr<HLSLVkPushConstantAttr>()) {
+ if (HasDeclaredAPushConstant)
+ SemaRef.Diag(Decl->getLocation(), diag::err_hlsl_push_constant_unique);
+
+ LangAS ImplAS = LangAS::hlsl_push_constant;
+ Type = SemaRef.getASTContext().getAddrSpaceQualType(Type, ImplAS);
+ Decl->setType(Type);
+ HasDeclaredAPushConstant = true;
+ return;
+ }
+
if (Type->isSamplerT() || Type->isVoidType())
return;
@@ -3895,7 +3916,7 @@ void SemaHLSL::ActOnVariableDeclarator(VarDecl *VD) {
// Global variables outside a cbuffer block that are not a resource, static,
// groupshared, or an empty array or struct belong to the default constant
// buffer $Globals (to be created at the end of the translation unit).
- if (IsDefaultBufferConstantDecl(VD)) {
+ if (IsDefaultBufferConstantDecl(getASTContext(), VD)) {
// update address space to hlsl_constant
QualType NewTy = getASTContext().getAddrSpaceQualType(
VD->getType(), LangAS::hlsl_constant);
@@ -4196,8 +4217,11 @@ void SemaHLSL::processExplicitBindingsOnDecl(VarDecl *VD) {
bool HasBinding = false;
for (Attr *A : VD->attrs()) {
- if (isa<HLSLVkBindingAttr>(A))
+ if (isa<HLSLVkBindingAttr>(A)) {
HasBinding = true;
+ if (auto PA = VD->getAttr<HLSLVkPushConstantAttr>())
+ Diag(PA->getLoc(), diag::err_hlsl_attr_incompatible) << A << PA;
+ }
HLSLResourceBindingAttr *RBA = dyn_cast<HLSLResourceBindingAttr>(A);
if (!RBA || !RBA->hasRegisterSlot())
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl
new file mode 100644
index 0000000000000..412ec4dffc572
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl
@@ -0,0 +1,20 @@
+// RUN: %clang_cc1 -triple spirv-pc-vulkan-compute -x hlsl -emit-llvm -finclude-default-header -disable-llvm-passes -o - %s | FileCheck %s
+
+struct S {
+ uint32_t a : 1;
+ uint32_t b : 1;
+};
+// CHECK: %struct.S = type { i8 }
+
+[[vk::push_constant]] S buffer;
+// CHECK: @buffer = external hidden addrspace(13) externally_initialized global %struct.S, align 1
+
+[numthreads(1, 1, 1)]
+void main() {
+ uint32_t v = buffer.b;
+// CHECK: %bf.load = load i8, ptr addrspace(13) @buffer, align 1
+// CHECK: %bf.lshr = lshr i8 %bf.load, 1
+// CHECK: %bf.clear = and i8 %bf.lshr, 1
+// CHECK: %bf.cast = zext i8 %bf.clear to i32
+// CHECK: store i32 %bf.cast
+}
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl
new file mode 100644
index 0000000000000..2b2e9d09c7ab0
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl
@@ -0,0 +1,17 @@
+// RUN: %clang_cc1 -triple spirv-pc-vulkan-compute -x hlsl -emit-llvm -finclude-default-header -disable-llvm-passes -o - %s | FileCheck %s
+
+[[vk::push_constant]]
+struct {
+ int a;
+ float b;
+ float3 c;
+}
+PushConstants;
+
+// CHECK: %struct.anon = type <{ i32, float, <3 x float> }>
+// CHECK: @PushConstants = external hidden addrspace(13) externally_initialized global %struct.anon, align 1
+
+[numthreads(1, 1, 1)]
+void main() {
+ float tmp = PushConstants.b;
+}
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl
new file mode 100644
index 0000000000000..6b58decfa5188
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl
@@ -0,0 +1,13 @@
+// RUN: %clang_cc1 -triple spirv-unknown-vulkan-compute -x hlsl -emit-llvm -disable-llvm-passes -o - -hlsl-entry main %s -verify
+
+struct S {
+ float f;
+};
+
+// expected-error@+1 {{'vk::binding' attribute is not compatible with 'vk::push_constant' attribute}}
+[[vk::push_constant, vk::binding(5)]]
+S pcs;
+
+[numthreads(1, 1, 1)]
+void main() {
+}
diff...
[truncated]
|
Member
|
@llvm/pr-subscribers-backend-directx Author: Nathan Gauër (Keenuts) ChangesImplements initial support for vk::push_constant. The old fix would be to use target specific types, but this is Patch is 27.93 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/166793.diff 33 Files Affected:
diff --git a/clang/include/clang/Basic/AddressSpaces.h b/clang/include/clang/Basic/AddressSpaces.h
index 48e4a1c61fe02..7280b8fc923d2 100644
--- a/clang/include/clang/Basic/AddressSpaces.h
+++ b/clang/include/clang/Basic/AddressSpaces.h
@@ -62,6 +62,7 @@ enum class LangAS : unsigned {
hlsl_private,
hlsl_device,
hlsl_input,
+ hlsl_push_constant,
// Wasm specific address spaces.
wasm_funcref,
diff --git a/clang/include/clang/Basic/Attr.td b/clang/include/clang/Basic/Attr.td
index 8dfe4bc08c48e..e00765a57cb23 100644
--- a/clang/include/clang/Basic/Attr.td
+++ b/clang/include/clang/Basic/Attr.td
@@ -5146,6 +5146,14 @@ def HLSLVkExtBuiltinInput : InheritableAttr {
let Documentation = [HLSLVkExtBuiltinInputDocs];
}
+def HLSLVkPushConstant : InheritableAttr {
+ let Spellings = [CXX11<"vk", "push_constant">];
+ let Args = [];
+ let Subjects = SubjectList<[GlobalVar], ErrorDiag>;
+ let LangOpts = [HLSL];
+ let Documentation = [HLSLVkPushConstantDocs];
+}
+
def HLSLVkConstantId : InheritableAttr {
let Spellings = [CXX11<"vk", "constant_id">];
let Args = [IntArgument<"Id">];
diff --git a/clang/include/clang/Basic/AttrDocs.td b/clang/include/clang/Basic/AttrDocs.td
index 4813191d2d602..3938c624c9d0c 100644
--- a/clang/include/clang/Basic/AttrDocs.td
+++ b/clang/include/clang/Basic/AttrDocs.td
@@ -8777,6 +8777,11 @@ https://github.com/microsoft/hlsl-specs/blob/main/proposals/0011-inline-spirv.md
}];
}
+def HLSLVkPushConstantDocs : Documentation {
+ let Category = DocCatVariable;
+ let Content = [{ FIXME }];
+}
+
def AnnotateTypeDocs : Documentation {
let Category = DocCatType;
let Heading = "annotate_type";
diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index a6e60fe4692ee..c9d128d77c90f 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -13183,6 +13183,9 @@ def err_hlsl_attr_invalid_type : Error<
"attribute %0 only applies to a field or parameter of type '%1'">;
def err_hlsl_attr_invalid_ast_node : Error<
"attribute %0 only applies to %1">;
+def err_hlsl_attr_incompatible
+ : Error<"%0 attribute is not compatible with %1 attribute">;
+
def err_hlsl_entry_shader_attr_mismatch : Error<
"%0 attribute on entry function does not match the target profile">;
def err_hlsl_numthreads_argument_oor : Error<"argument '%select{X|Y|Z}0' to numthreads attribute cannot exceed %1">;
@@ -13294,6 +13297,9 @@ def err_hlsl_incomplete_resource_array_in_function_param: Error<
def err_hlsl_assign_to_global_resource: Error<
"assignment to global resource variable %0 is not allowed">;
+def err_hlsl_push_constant_unique
+ : Error<"cannot have more than one push constant block">;
+
// Layout randomization diagnostics.
def err_non_designated_init_used : Error<
"a randomized struct can only be initialized with a designated initializer">;
diff --git a/clang/include/clang/Basic/HLSLRuntime.h b/clang/include/clang/Basic/HLSLRuntime.h
index 03166805daa6a..f6a1cf9636467 100644
--- a/clang/include/clang/Basic/HLSLRuntime.h
+++ b/clang/include/clang/Basic/HLSLRuntime.h
@@ -14,6 +14,7 @@
#ifndef CLANG_BASIC_HLSLRUNTIME_H
#define CLANG_BASIC_HLSLRUNTIME_H
+#include "clang/Basic/AddressSpaces.h"
#include "clang/Basic/LangOptions.h"
#include <cstdint>
@@ -30,6 +31,10 @@ getStageFromEnvironment(const llvm::Triple::EnvironmentType &E) {
return static_cast<ShaderStage>(Pipeline);
}
+constexpr bool isInitializedByPipeline(LangAS AS) {
+ return AS == LangAS::hlsl_input || AS == LangAS::hlsl_push_constant;
+}
+
#define ENUM_COMPARE_ASSERT(Value) \
static_assert( \
getStageFromEnvironment(llvm::Triple::Value) == ShaderStage::Value, \
diff --git a/clang/include/clang/Sema/SemaHLSL.h b/clang/include/clang/Sema/SemaHLSL.h
index 86da323892f98..2fcac237eba1c 100644
--- a/clang/include/clang/Sema/SemaHLSL.h
+++ b/clang/include/clang/Sema/SemaHLSL.h
@@ -190,6 +190,7 @@ class SemaHLSL : public SemaBase {
void handleSemanticAttr(Decl *D, const ParsedAttr &AL);
void handleVkExtBuiltinInputAttr(Decl *D, const ParsedAttr &AL);
+ void handleVkPushConstantAttr(Decl *D, const ParsedAttr &AL);
bool CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall);
QualType ProcessResourceTypeAttributes(QualType Wrapped);
@@ -239,6 +240,8 @@ class SemaHLSL : public SemaBase {
IdentifierInfo *RootSigOverrideIdent = nullptr;
+ bool HasDeclaredAPushConstant = false;
+
struct SemanticInfo {
HLSLParsedSemanticAttr *Semantic;
std::optional<uint32_t> Index;
diff --git a/clang/lib/AST/Type.cpp b/clang/lib/AST/Type.cpp
index 4548af17e37f2..53082bcf78f6a 100644
--- a/clang/lib/AST/Type.cpp
+++ b/clang/lib/AST/Type.cpp
@@ -101,6 +101,7 @@ bool Qualifiers::isTargetAddressSpaceSupersetOf(LangAS A, LangAS B,
(A == LangAS::Default && B == LangAS::hlsl_private) ||
(A == LangAS::Default && B == LangAS::hlsl_device) ||
(A == LangAS::Default && B == LangAS::hlsl_input) ||
+ (A == LangAS::Default && B == LangAS::hlsl_push_constant) ||
// Conversions from target specific address spaces may be legal
// depending on the target information.
Ctx.getTargetInfo().isAddressSpaceSupersetOf(A, B);
diff --git a/clang/lib/AST/TypePrinter.cpp b/clang/lib/AST/TypePrinter.cpp
index c18b2eafc722c..8448dd3748e28 100644
--- a/clang/lib/AST/TypePrinter.cpp
+++ b/clang/lib/AST/TypePrinter.cpp
@@ -2749,6 +2749,8 @@ std::string Qualifiers::getAddrSpaceAsString(LangAS AS) {
return "hlsl_device";
case LangAS::hlsl_input:
return "hlsl_input";
+ case LangAS::hlsl_push_constant:
+ return "hlsl_push_constant";
case LangAS::wasm_funcref:
return "__funcref";
default:
diff --git a/clang/lib/Basic/TargetInfo.cpp b/clang/lib/Basic/TargetInfo.cpp
index ffaf98bf9c366..92ca7a66a9593 100644
--- a/clang/lib/Basic/TargetInfo.cpp
+++ b/clang/lib/Basic/TargetInfo.cpp
@@ -52,6 +52,7 @@ static const LangASMap FakeAddrSpaceMap = {
15, // hlsl_private
16, // hlsl_device
17, // hlsl_input
+ 18, // hlsl_push_constant
20, // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/AArch64.h b/clang/lib/Basic/Targets/AArch64.h
index 1a7aa658e9d87..8e8d8e6ae86b5 100644
--- a/clang/lib/Basic/Targets/AArch64.h
+++ b/clang/lib/Basic/Targets/AArch64.h
@@ -48,6 +48,7 @@ static const unsigned ARM64AddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/AMDGPU.cpp b/clang/lib/Basic/Targets/AMDGPU.cpp
index d4d696b8456b6..993a73a89c9e9 100644
--- a/clang/lib/Basic/Targets/AMDGPU.cpp
+++ b/clang/lib/Basic/Targets/AMDGPU.cpp
@@ -63,6 +63,7 @@ const LangASMap AMDGPUTargetInfo::AMDGPUDefIsGenMap = {
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_private
llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_device
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_input
+ llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_push_constant
};
const LangASMap AMDGPUTargetInfo::AMDGPUDefIsPrivMap = {
@@ -91,6 +92,7 @@ const LangASMap AMDGPUTargetInfo::AMDGPUDefIsPrivMap = {
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_private
llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_device
llvm::AMDGPUAS::PRIVATE_ADDRESS, // hlsl_input
+ llvm::AMDGPUAS::GLOBAL_ADDRESS, // hlsl_push_constant
};
} // namespace targets
} // namespace clang
diff --git a/clang/lib/Basic/Targets/DirectX.h b/clang/lib/Basic/Targets/DirectX.h
index a21a593365773..c0799a6f7610f 100644
--- a/clang/lib/Basic/Targets/DirectX.h
+++ b/clang/lib/Basic/Targets/DirectX.h
@@ -46,6 +46,7 @@ static const unsigned DirectXAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/NVPTX.h b/clang/lib/Basic/Targets/NVPTX.h
index f5c8396f398aa..6338a4f2f9036 100644
--- a/clang/lib/Basic/Targets/NVPTX.h
+++ b/clang/lib/Basic/Targets/NVPTX.h
@@ -50,6 +50,7 @@ static const unsigned NVPTXAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/SPIR.h b/clang/lib/Basic/Targets/SPIR.h
index 22b2799518dd0..94449231efb94 100644
--- a/clang/lib/Basic/Targets/SPIR.h
+++ b/clang/lib/Basic/Targets/SPIR.h
@@ -51,6 +51,7 @@ static const unsigned SPIRDefIsPrivMap[] = {
10, // hlsl_private
11, // hlsl_device
7, // hlsl_input
+ 13, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
@@ -87,6 +88,7 @@ static const unsigned SPIRDefIsGenMap[] = {
10, // hlsl_private
11, // hlsl_device
7, // hlsl_input
+ 13, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/SystemZ.h b/clang/lib/Basic/Targets/SystemZ.h
index 4e15d5af1cde6..4ce515b31a001 100644
--- a/clang/lib/Basic/Targets/SystemZ.h
+++ b/clang/lib/Basic/Targets/SystemZ.h
@@ -46,6 +46,7 @@ static const unsigned ZOSAddressMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
0 // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/TCE.h b/clang/lib/Basic/Targets/TCE.h
index 005cab9819472..161025378c471 100644
--- a/clang/lib/Basic/Targets/TCE.h
+++ b/clang/lib/Basic/Targets/TCE.h
@@ -55,6 +55,7 @@ static const unsigned TCEOpenCLAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/Basic/Targets/WebAssembly.h b/clang/lib/Basic/Targets/WebAssembly.h
index 4de6ce6bb5a21..c8065843aeb42 100644
--- a/clang/lib/Basic/Targets/WebAssembly.h
+++ b/clang/lib/Basic/Targets/WebAssembly.h
@@ -46,6 +46,7 @@ static const unsigned WebAssemblyAddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
20, // wasm_funcref
};
diff --git a/clang/lib/Basic/Targets/X86.h b/clang/lib/Basic/Targets/X86.h
index e7da2622e78b5..7b88ac70e234f 100644
--- a/clang/lib/Basic/Targets/X86.h
+++ b/clang/lib/Basic/Targets/X86.h
@@ -50,6 +50,7 @@ static const unsigned X86AddrSpaceMap[] = {
0, // hlsl_private
0, // hlsl_device
0, // hlsl_input
+ 0, // hlsl_push_constant
// Wasm address space values for this target are dummy values,
// as it is only enabled for Wasm targets.
20, // wasm_funcref
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 3eeb1718e455a..c3b536c7a267f 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -6049,9 +6049,11 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D,
getCUDARuntime().handleVarRegistration(D, *GV);
}
- if (LangOpts.HLSL && GetGlobalVarAddressSpace(D) == LangAS::hlsl_input) {
+ if (LangOpts.HLSL &&
+ hlsl::isInitializedByPipeline(GetGlobalVarAddressSpace(D))) {
// HLSL Input variables are considered to be set by the driver/pipeline, but
- // only visible to a single thread/wave.
+ // only visible to a single thread/wave. Push constants are also externally
+ // initialized, but constant, hence cross-wave visibility is not relevant.
GV->setExternallyInitialized(true);
} else {
GV->setInitializer(Init);
@@ -6102,10 +6104,11 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D,
!D->hasAttr<ConstInitAttr>())
Linkage = llvm::GlobalValue::InternalLinkage;
- // HLSL variables in the input address space maps like memory-mapped
- // variables. Even if they are 'static', they are externally initialized and
- // read/write by the hardware/driver/pipeline.
- if (LangOpts.HLSL && GetGlobalVarAddressSpace(D) == LangAS::hlsl_input)
+ // HLSL variables in the input or push-constant address space maps are like
+ // memory-mapped variables. Even if they are 'static', they are externally
+ // initialized and read/write by the hardware/driver/pipeline.
+ if (LangOpts.HLSL &&
+ hlsl::isInitializedByPipeline(GetGlobalVarAddressSpace(D)))
Linkage = llvm::GlobalValue::ExternalLinkage;
GV->setLinkage(Linkage);
diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
index 25b89d65847ad..5c29bf5e77414 100644
--- a/clang/lib/Sema/SemaDecl.cpp
+++ b/clang/lib/Sema/SemaDecl.cpp
@@ -30,6 +30,7 @@
#include "clang/AST/Type.h"
#include "clang/Basic/Builtins.h"
#include "clang/Basic/DiagnosticComment.h"
+#include "clang/Basic/HLSLRuntime.h"
#include "clang/Basic/PartialDiagnostic.h"
#include "clang/Basic/SourceManager.h"
#include "clang/Basic/TargetInfo.h"
@@ -14559,10 +14560,10 @@ void Sema::ActOnUninitializedDecl(Decl *RealDecl) {
if (getLangOpts().HLSL && HLSL().ActOnUninitializedVarDecl(Var))
return;
- // HLSL input variables are expected to be externally initialized, even
- // when marked `static`.
+ // HLSL input & push-constant variables are expected to be externally
+ // initialized, even when marked `static`.
if (getLangOpts().HLSL &&
- Var->getType().getAddressSpace() == LangAS::hlsl_input)
+ hlsl::isInitializedByPipeline(Var->getType().getAddressSpace()))
return;
// C++03 [dcl.init]p9:
diff --git a/clang/lib/Sema/SemaDeclAttr.cpp b/clang/lib/Sema/SemaDeclAttr.cpp
index a9e7b44ac9d73..0396155bd6a9d 100644
--- a/clang/lib/Sema/SemaDeclAttr.cpp
+++ b/clang/lib/Sema/SemaDeclAttr.cpp
@@ -7614,6 +7614,9 @@ ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D, const ParsedAttr &AL,
case ParsedAttr::AT_HLSLVkExtBuiltinInput:
S.HLSL().handleVkExtBuiltinInputAttr(D, AL);
break;
+ case ParsedAttr::AT_HLSLVkPushConstant:
+ S.HLSL().handleVkPushConstantAttr(D, AL);
+ break;
case ParsedAttr::AT_HLSLVkConstantId:
S.HLSL().handleVkConstantIdAttr(D, AL);
break;
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 2b9b3abbd5360..1831584c88697 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -1667,6 +1667,11 @@ void SemaHLSL::handleVkExtBuiltinInputAttr(Decl *D, const ParsedAttr &AL) {
HLSLVkExtBuiltinInputAttr(getASTContext(), AL, ID));
}
+void SemaHLSL::handleVkPushConstantAttr(Decl *D, const ParsedAttr &AL) {
+ D->addAttr(::new (getASTContext())
+ HLSLVkPushConstantAttr(getASTContext(), AL));
+}
+
void SemaHLSL::handleVkConstantIdAttr(Decl *D, const ParsedAttr &AL) {
uint32_t Id;
if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(0), Id))
@@ -3837,12 +3842,15 @@ QualType SemaHLSL::getInoutParameterType(QualType Ty) {
return Ty;
}
-static bool IsDefaultBufferConstantDecl(VarDecl *VD) {
+static bool IsDefaultBufferConstantDecl(const ASTContext &Ctx, VarDecl *VD) {
+ bool IsVulkan =
+ Ctx.getTargetInfo().getTriple().getOS() == llvm::Triple::Vulkan;
+ bool IsVKPushConstant = IsVulkan && VD->hasAttr<HLSLVkPushConstantAttr>();
QualType QT = VD->getType();
return VD->getDeclContext()->isTranslationUnit() &&
QT.getAddressSpace() == LangAS::Default &&
VD->getStorageClass() != SC_Static &&
- !VD->hasAttr<HLSLVkConstantIdAttr>() &&
+ !VD->hasAttr<HLSLVkConstantIdAttr>() && !IsVKPushConstant &&
!isInvalidConstantBufferLeafElementType(QT.getTypePtr());
}
@@ -3863,6 +3871,19 @@ void SemaHLSL::deduceAddressSpace(VarDecl *Decl) {
return;
}
+ bool IsVulkan = getASTContext().getTargetInfo().getTriple().getOS() ==
+ llvm::Triple::Vulkan;
+ if (IsVulkan && Decl->hasAttr<HLSLVkPushConstantAttr>()) {
+ if (HasDeclaredAPushConstant)
+ SemaRef.Diag(Decl->getLocation(), diag::err_hlsl_push_constant_unique);
+
+ LangAS ImplAS = LangAS::hlsl_push_constant;
+ Type = SemaRef.getASTContext().getAddrSpaceQualType(Type, ImplAS);
+ Decl->setType(Type);
+ HasDeclaredAPushConstant = true;
+ return;
+ }
+
if (Type->isSamplerT() || Type->isVoidType())
return;
@@ -3895,7 +3916,7 @@ void SemaHLSL::ActOnVariableDeclarator(VarDecl *VD) {
// Global variables outside a cbuffer block that are not a resource, static,
// groupshared, or an empty array or struct belong to the default constant
// buffer $Globals (to be created at the end of the translation unit).
- if (IsDefaultBufferConstantDecl(VD)) {
+ if (IsDefaultBufferConstantDecl(getASTContext(), VD)) {
// update address space to hlsl_constant
QualType NewTy = getASTContext().getAddrSpaceQualType(
VD->getType(), LangAS::hlsl_constant);
@@ -4196,8 +4217,11 @@ void SemaHLSL::processExplicitBindingsOnDecl(VarDecl *VD) {
bool HasBinding = false;
for (Attr *A : VD->attrs()) {
- if (isa<HLSLVkBindingAttr>(A))
+ if (isa<HLSLVkBindingAttr>(A)) {
HasBinding = true;
+ if (auto PA = VD->getAttr<HLSLVkPushConstantAttr>())
+ Diag(PA->getLoc(), diag::err_hlsl_attr_incompatible) << A << PA;
+ }
HLSLResourceBindingAttr *RBA = dyn_cast<HLSLResourceBindingAttr>(A);
if (!RBA || !RBA->hasRegisterSlot())
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl
new file mode 100644
index 0000000000000..412ec4dffc572
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.access.bitfield.hlsl
@@ -0,0 +1,20 @@
+// RUN: %clang_cc1 -triple spirv-pc-vulkan-compute -x hlsl -emit-llvm -finclude-default-header -disable-llvm-passes -o - %s | FileCheck %s
+
+struct S {
+ uint32_t a : 1;
+ uint32_t b : 1;
+};
+// CHECK: %struct.S = type { i8 }
+
+[[vk::push_constant]] S buffer;
+// CHECK: @buffer = external hidden addrspace(13) externally_initialized global %struct.S, align 1
+
+[numthreads(1, 1, 1)]
+void main() {
+ uint32_t v = buffer.b;
+// CHECK: %bf.load = load i8, ptr addrspace(13) @buffer, align 1
+// CHECK: %bf.lshr = lshr i8 %bf.load, 1
+// CHECK: %bf.clear = and i8 %bf.lshr, 1
+// CHECK: %bf.cast = zext i8 %bf.clear to i32
+// CHECK: store i32 %bf.cast
+}
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl
new file mode 100644
index 0000000000000..2b2e9d09c7ab0
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.anon-struct.hlsl
@@ -0,0 +1,17 @@
+// RUN: %clang_cc1 -triple spirv-pc-vulkan-compute -x hlsl -emit-llvm -finclude-default-header -disable-llvm-passes -o - %s | FileCheck %s
+
+[[vk::push_constant]]
+struct {
+ int a;
+ float b;
+ float3 c;
+}
+PushConstants;
+
+// CHECK: %struct.anon = type <{ i32, float, <3 x float> }>
+// CHECK: @PushConstants = external hidden addrspace(13) externally_initialized global %struct.anon, align 1
+
+[numthreads(1, 1, 1)]
+void main() {
+ float tmp = PushConstants.b;
+}
diff --git a/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl
new file mode 100644
index 0000000000000..6b58decfa5188
--- /dev/null
+++ b/clang/test/CodeGenHLSL/vk-features/vk.pushconstant.invalid.hlsl
@@ -0,0 +1,13 @@
+// RUN: %clang_cc1 -triple spirv-unknown-vulkan-compute -x hlsl -emit-llvm -disable-llvm-passes -o - -hlsl-entry main %s -verify
+
+struct S {
+ float f;
+};
+
+// expected-error@+1 {{'vk::binding' attribute is not compatible with 'vk::push_constant' attribute}}
+[[vk::push_constant, vk::binding(5)]]
+S pcs;
+
+[numthreads(1, 1, 1)]
+void main() {
+}
diff...
[truncated]
|
s-perron
reviewed
Nov 17, 2025
Contributor
s-perron
left a comment
s-perron
left a comment
There was a problem hiding this comment.
LGTM except I would like to see a test that checks that the AST is adding the attribute correctly.
Also, add a test for DXIL. Make sure it properly ignores the attribute.
Keenuts
added a commit
to llvm/wg-hlsl
that referenced
this pull request
Nov 18, 2025
Adds proposal to implement VK push constant support in Clang. E2E implementation draft: llvm/llvm-project#166793
🐧 Linux x64 Test Results
✅ The build succeeded and all tests passed. |
s-perron
approved these changes
Nov 18, 2025
farzonl
reviewed
Nov 18, 2025
farzonl
approved these changes
Nov 18, 2025
Member
|
LGTM once you fix the ci test failures and update documentation in tablegen. |
Keenuts
force-pushed
the
push-constants
branch
from
8d82a7f to
3488da4
Compare
Keenuts
force-pushed
the
push-constants
branch
2 times, most recently
from
fe95d22 to
c196755
Compare
Contributor
Author
|
Thanks for the reviews! Turns out the missing decoration was a bigger change than anticipated. If you prefer I could split this into 2 PRs: FE & BE. |
Implements initial support for vk::push_constant. As is, this allows handling simple push constants, but has one main issue: layout can be incorrect (See llvm#168401). The layout issue being not only push-constant related, it's ignored for this PR. The frontend part of the implementation is straightforward: - adding a new attribute - when targeting vulkan/spirv, we process it - global variables with this attribute gets a new AS: hlsl_push_constant The IR has nothing specific, only some RO globals in this new AS. On the SPIR-V side, we not convert this AS into a PushConstant storage class. But this creates some issues: the variables in this storage class must have a specific set of decoration to define their layout. Current infra to create the SPIR-V types lacks the context required to make this decision: no indication on the AS or context around the type being created. Refactoring this would be a heavy task as it would require getting this information in every place using the GR for type creation. Instead, we do something similar to CBuffers: - find all globals with this address space, and change their type to a target-specific type. - insert a new intrinsic in place of every reference to this global variable. This allow the backend to handle both layout variables loads and type lowering independently. Type lowering has nothing specific: when we encounter a target extension type with spirv.PushConstant, we lower this to the correct SPIR-V type with the proper offset & block decorations. As for the intrinsic, it's mostly a no-op, but required since we have this target-specific type. Note: this implementation prevents the static declaration of multiple push constants in a single shader module. The actual specification is more relaxed: there can be only one **used** push constant block per entrypoint. To correctly implement this, we'd require to keep some additional state to determine the list of statically used resources per entrypoint. This shall be addressed as a follow-up (see llvm#170310)
Keenuts
force-pushed
the
push-constants
branch
from
c196755 to
3fc3e29
Compare
Contributor
Author
|
@s-perron is this OK for you? |
s-perron
reviewed
Dec 10, 2025
| @@ -0,0 +1,12 @@ | |||
| // RUN: %clang_cc1 -triple spirv-unknown-vulkan1.3-compute -x hlsl -ast-dump -o - %s | FileCheck %s | |||
Contributor
There was a problem hiding this comment.
Could you add an extra run command to make sure the attribute is added when targeting DXIL too?
| II->getIntrinsicID() == Intrinsic::spv_pushconstant_getpointer) { | ||
| auto *GV = cast<GlobalVariable>(II->getOperand(0)); | ||
| auto *HandleType = cast<TargetExtType>(GV->getValueType()); | ||
| if (HandleType->getTargetExtName() == "spirv.PushConstant") { |
Contributor
There was a problem hiding this comment.
should this be an assert or error of some type? What else could the handle type be?
Comment on lines
+854
to
+858
| for (User *U : II->users()) { | ||
| Ty = cast<Instruction>(U)->getAccessType(); | ||
| if (Ty) | ||
| break; | ||
| } |
Contributor
There was a problem hiding this comment.
Why do you look at the users? Can't you get the type from target type as we do for a vulkan buffer?
My concern is that the first uses will be a load of an integer at the start of the puch constant block. LLVM-opt will remove the GEP because it is not needed in llvm-ir. The second use will be a gep to access a float in the second position. Then that will generate an access chain on an int.
Comment on lines
+51
to
+64
| for (llvm::User *U : Users) { | ||
| Instruction *I = dyn_cast<Instruction>(U); | ||
| if (!I) | ||
| continue; | ||
|
|
||
| IRBuilder<> Builder(I); | ||
| Value *GetPointerCall = Builder.CreateIntrinsic( | ||
| NewGV->getType(), Intrinsic::spv_pushconstant_getpointer, {NewGV}); | ||
| GR->buildAssignPtr(Builder, GV->getValueType(), GetPointerCall); | ||
|
|
||
| for (unsigned N = 0; N < I->getNumOperands(); ++N) { | ||
| if (I->getOperand(N) == GV) | ||
| I->setOperand(N, GetPointerCall); | ||
| } |
Contributor
There was a problem hiding this comment.
Why can you use GV->replacesAllUsesWith(NewGV)?
Comment on lines
+52
to
+54
| Instruction *I = dyn_cast<Instruction>(U); | ||
| if (!I) | ||
| continue; |
Contributor
There was a problem hiding this comment.
What are the non instruction uses, and why is it okay to leave them as is? Looking at the tests, I don't see any case where this happens.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Implements initial support for vk::push_constant.
As is, this allows handling simple push constants, but has one
main issue: layout can be incorrect (See #168401). The layout
issue being not only push-constant related, it's ignored for this PR.
The frontend part of the implementation is straightforward:
hlsl_push_constant
The IR has nothing specific, only some RO globals in this new AS.
On the SPIR-V side, we not convert this AS into a PushConstant storage
class. But this creates some issues: the variables in this storage class
must have a specific set of decoration to define their layout.
Current infra to create the SPIR-V types lacks the context required to
make this decision: no indication on the AS or context around the type
being created. Refactoring this would be a heavy task as it would
require getting this information in every place using the GR for type
creation.
Instead, we do something similar to CBuffers:
a target-specific type.
variable.
This allow the backend to handle both layout variables loads and type
lowering independently.
Type lowering has nothing specific: when we encounter a target extension
type with spirv.PushConstant, we lower this to the correct SPIR-V type
with the proper offset & block decorations.
As for the intrinsic, it's mostly a no-op, but required since we have
this target-specific type.
Note: this implementation prevents the static declaration of multiple
push constants in a single shader module. The actual specification is
more relaxed: there can be only one used push constant block per
entrypoint. To correctly implement this, we'd require to keep some
additional state to determine the list of statically used resources per
entrypoint. This shall be addressed as a follow-up (see #170310)