[clang] [llvm] [HLSL][DXIL][SPIRV] Implementation of an abstraction for intrinsic selection of HLSL backends (PR #87171)
Farzon Lotfi via cfe-commits
cfe-commits at lists.llvm.org
Thu Apr 4 14:26:55 PDT 2024
https://github.com/farzonl updated https://github.com/llvm/llvm-project/pull/87171
>From 6e088142af4196f11bc006e0024d4ca0ac4e08f0 Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlotfi at microsoft.com>
Date: Thu, 28 Mar 2024 21:05:36 -0400
Subject: [PATCH 01/11] [HLSL][DXIL][SPIRV] Intrinsic unification PR
---
clang/include/clang/Basic/Builtins.td | 6 +
clang/lib/CodeGen/CGBuiltin.cpp | 19 ++
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 112 +++++++
clang/lib/Sema/SemaChecking.cpp | 1 +
clang/test/CodeGenHLSL/builtins/all.hlsl | 277 ++++++++++++++++++
llvm/include/llvm/IR/Intrinsics.td | 12 +
llvm/include/llvm/IR/IntrinsicsDirectX.td | 4 +-
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 3 +-
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 19 ++
.../test/CodeGen/SPIRV/hlsl-intrinsics/all.ll | 95 ++++++
10 files changed, 545 insertions(+), 3 deletions(-)
create mode 100644 clang/test/CodeGenHLSL/builtins/all.hlsl
create mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/all.ll
diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index f421223ff087de..d6ceb450bd106b 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4587,6 +4587,12 @@ def GetDeviceSideMangledName : LangBuiltin<"CUDA_LANG"> {
}
// HLSL
+def HLSLAll : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_elementwise_all"];
+ let Attributes = [NoThrow, Const];
+ let Prototype = "bool(...)";
+}
+
def HLSLAny : LangBuiltin<"HLSL_LANG"> {
let Spellings = ["__builtin_hlsl_elementwise_any"];
let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 2537e715b63ee4..2acf799f168785 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -51,6 +51,7 @@
#include "llvm/IR/IntrinsicsR600.h"
#include "llvm/IR/IntrinsicsRISCV.h"
#include "llvm/IR/IntrinsicsS390.h"
+#include "llvm/IR/IntrinsicsSPIRV.h"
#include "llvm/IR/IntrinsicsVE.h"
#include "llvm/IR/IntrinsicsWebAssembly.h"
#include "llvm/IR/IntrinsicsX86.h"
@@ -18166,12 +18167,30 @@ Intrinsic::ID getDotProductIntrinsic(QualType QT, int elementCount) {
return Intrinsic::dx_udot;
}
+Intrinsic::ID getAllIntrinsic(const llvm::Triple::ArchType Arch) {
+ switch (Arch) {
+ case llvm::Triple::dxil:
+ return Intrinsic::dx_all;
+ case llvm::Triple::spirv:
+ return Intrinsic::spv_all;
+ default:
+ llvm_unreachable("Input semantic not supported by target");
+ }
+}
+
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E) {
if (!getLangOpts().HLSL)
return nullptr;
switch (BuiltinID) {
+ case Builtin::BI__builtin_hlsl_elementwise_all: {
+ Value *Op0 = EmitScalarExpr(E->getArg(0));
+ return Builder.CreateIntrinsic(
+ /*ReturnType=*/llvm::Type::getInt1Ty(getLLVMContext()),
+ getAllIntrinsic(CGM.getTarget().getTriple().getArch()),
+ ArrayRef<Value *>{Op0}, nullptr, "hlsl.all");
+ }
case Builtin::BI__builtin_hlsl_elementwise_any: {
Value *Op0 = EmitScalarExpr(E->getArg(0));
return Builder.CreateIntrinsic(
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 9fb6204f90c9a8..06409c6fc77417 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -100,6 +100,118 @@ double3 abs(double3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_abs)
double4 abs(double4);
+//===----------------------------------------------------------------------===//
+// all builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn bool all(T x)
+/// \brief Returns True if all components of the \a x parameter are non-zero;
+/// otherwise, false. \param x The input value.
+
+#ifdef __HLSL_ENABLE_16_BIT
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(int16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(int16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(int16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(int16_t4);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(uint16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(uint16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(uint16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(uint16_t4);
+#endif
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(half4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(bool);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(bool2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(bool3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(bool4);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(int);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(int2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(int3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(int4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(uint);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(uint2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(uint3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(uint4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(float);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(float2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(float3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(float4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(int64_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(int64_t2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(int64_t3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(int64_t4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(uint64_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(uint64_t2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(uint64_t3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(uint64_t4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(double);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(double2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(double3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_all)
+bool all(double4);
+
//===----------------------------------------------------------------------===//
// any builtins
//===----------------------------------------------------------------------===//
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
index 3dcd18b3afc8b4..4c30cc3160ded5 100644
--- a/clang/lib/Sema/SemaChecking.cpp
+++ b/clang/lib/Sema/SemaChecking.cpp
@@ -5563,6 +5563,7 @@ void SetElementTypeAsReturnType(Sema *S, CallExpr *TheCall,
// returning an ExprError
bool Sema::CheckHLSLBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
switch (BuiltinID) {
+ case Builtin::BI__builtin_hlsl_elementwise_all:
case Builtin::BI__builtin_hlsl_elementwise_any: {
if (checkArgCount(*this, TheCall, 1))
return true;
diff --git a/clang/test/CodeGenHLSL/builtins/all.hlsl b/clang/test/CodeGenHLSL/builtins/all.hlsl
new file mode 100644
index 00000000000000..b48daa287480ff
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/all.hlsl
@@ -0,0 +1,277 @@
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
+// RUN: spirv-unknown-vulkan-compute %s -fnative-half-type \
+// RUN: -emit-llvm -disable-llvm-passes -o - | FileCheck %s \
+// RUN: --check-prefixes=CHECK,NATIVE_HALF,SPIR_NATIVE_HALF,SPIR_CHECK
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
+// RUN: spirv-unknown-vulkan-compute %s -emit-llvm -disable-llvm-passes \
+// RUN: -o - | FileCheck %s --check-prefixes=CHECK,SPIR_NO_HALF,SPIR_CHECK
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
+// RUN: dxil-pc-shadermodel6.3-library %s -fnative-half-type \
+// RUN: -emit-llvm -disable-llvm-passes -o - | FileCheck %s \
+// RUN: --check-prefixes=CHECK,NATIVE_HALF,DXIL_NATIVE_HALF,DXIL_CHECK
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple \
+// RUN: dxil-pc-shadermodel6.3-library %s -emit-llvm -disable-llvm-passes \
+// RUN: -o - | FileCheck %s --check-prefixes=CHECK,DXIL_NO_HALF,DXIL_CHECK
+
+#ifdef __HLSL_ENABLE_16_BIT
+// DXIL_NATIVE_HALF: define noundef i1 @
+// SPIR_NATIVE_HALF: define spir_func noundef i1 @
+// DXIL_NATIVE_HALF: %hlsl.all = call i1 @llvm.dx.all.i16
+// SPIR_NATIVE_HALF: %hlsl.all = call i1 @llvm.spv.all.i16
+// NATIVE_HALF: ret i1 %hlsl.all
+bool test_all_int16_t(int16_t p0) { return all(p0); }
+// DXIL_NATIVE_HALF: define noundef i1 @
+// SPIR_NATIVE_HALF: define spir_func noundef i1 @
+// DXIL_NATIVE_HALF: %hlsl.all = call i1 @llvm.dx.all.v2i16
+// SPIR_NATIVE_HALF: %hlsl.all = call i1 @llvm.spv.all.v2i16
+// NATIVE_HALF: ret i1 %hlsl.all
+bool test_all_int16_t2(int16_t2 p0) { return all(p0); }
+// DXIL_NATIVE_HALF: define noundef i1 @
+// SPIR_NATIVE_HALF: define spir_func noundef i1 @
+// DXIL_NATIVE_HALF: %hlsl.all = call i1 @llvm.dx.all.v3i16
+// SPIR_NATIVE_HALF: %hlsl.all = call i1 @llvm.spv.all.v3i16
+// NATIVE_HALF: ret i1 %hlsl.all
+bool test_all_int16_t3(int16_t3 p0) { return all(p0); }
+// DXIL_NATIVE_HALF: define noundef i1 @
+// SPIR_NATIVE_HALF: define spir_func noundef i1 @
+// DXIL_NATIVE_HALF: %hlsl.all = call i1 @llvm.dx.all.v4i16
+// SPIR_NATIVE_HALF: %hlsl.all = call i1 @llvm.spv.all.v4i16
+// NATIVE_HALF: ret i1 %hlsl.all
+bool test_all_int16_t4(int16_t4 p0) { return all(p0); }
+
+// DXIL_NATIVE_HALF: define noundef i1 @
+// SPIR_NATIVE_HALF: define spir_func noundef i1 @
+// DXIL_NATIVE_HALF: %hlsl.all = call i1 @llvm.dx.all.i16
+// SPIR_NATIVE_HALF: %hlsl.all = call i1 @llvm.spv.all.i16
+// NATIVE_HALF: ret i1 %hlsl.all
+bool test_all_uint16_t(uint16_t p0) { return all(p0); }
+// DXIL_NATIVE_HALF: define noundef i1 @
+// SPIR_NATIVE_HALF: define spir_func noundef i1 @
+// DXIL_NATIVE_HALF: %hlsl.all = call i1 @llvm.dx.all.v2i16
+// SPIR_NATIVE_HALF: %hlsl.all = call i1 @llvm.spv.all.v2i16
+// NATIVE_HALF: ret i1 %hlsl.all
+bool test_all_uint16_t2(uint16_t2 p0) { return all(p0); }
+// DXIL_NATIVE_HALF: define noundef i1 @
+// SPIR_NATIVE_HALF: define spir_func noundef i1 @
+// DXIL_NATIVE_HALF: %hlsl.all = call i1 @llvm.dx.all.v3i16
+// SPIR_NATIVE_HALF: %hlsl.all = call i1 @llvm.spv.all.v3i16
+// NATIVE_HALF: ret i1 %hlsl.all
+bool test_all_uint16_t3(uint16_t3 p0) { return all(p0); }
+// DXIL_NATIVE_HALF: define noundef i1 @
+// SPIR_NATIVE_HALF: define spir_func noundef i1 @
+// DXIL_NATIVE_HALF: %hlsl.all = call i1 @llvm.dx.all.v4i16
+// SPIR_NATIVE_HALF: %hlsl.all = call i1 @llvm.spv.all.v4i16
+// NATIVE_HALF: ret i1 %hlsl.all
+bool test_all_uint16_t4(uint16_t4 p0) { return all(p0); }
+#endif // __HLSL_ENABLE_16_BIT
+
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_NATIVE_HALF: %hlsl.all = call i1 @llvm.dx.all.f16
+// SPIR_NATIVE_HALF: %hlsl.all = call i1 @llvm.spv.all.f16
+// DXIL_NO_HALF: %hlsl.all = call i1 @llvm.dx.all.f32
+// SPIR_NO_HALF: %hlsl.all = call i1 @llvm.spv.all.f32
+// CHECK: ret i1 %hlsl.all
+bool test_all_half(half p0) { return all(p0); }
+
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_NATIVE_HALF: %hlsl.all = call i1 @llvm.dx.all.v2f16
+// SPIR_NATIVE_HALF: %hlsl.all = call i1 @llvm.spv.all.v2f16
+// DXIL_NO_HALF: %hlsl.all = call i1 @llvm.dx.all.v2f32
+// SPIR_NO_HALF: %hlsl.all = call i1 @llvm.spv.all.v2f32
+// CHECK: ret i1 %hlsl.all
+bool test_all_half2(half2 p0) { return all(p0); }
+
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_NATIVE_HALF: %hlsl.all = call i1 @llvm.dx.all.v3f16
+// SPIR_NATIVE_HALF: %hlsl.all = call i1 @llvm.spv.all.v3f16
+// DXIL_NO_HALF: %hlsl.all = call i1 @llvm.dx.all.v3f32
+// SPIR_NO_HALF: %hlsl.all = call i1 @llvm.spv.all.v3f32
+// CHECK: ret i1 %hlsl.all
+bool test_all_half3(half3 p0) { return all(p0); }
+
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_NATIVE_HALF: %hlsl.all = call i1 @llvm.dx.all.v4f16
+// SPIR_NATIVE_HALF: %hlsl.all = call i1 @llvm.spv.all.v4f16
+// DXIL_NO_HALF: %hlsl.all = call i1 @llvm.dx.all.v4f32
+// SPIR_NO_HALF: %hlsl.all = call i1 @llvm.spv.all.v4f32
+// CHECK: ret i1 %hlsl.all
+bool test_all_half4(half4 p0) { return all(p0); }
+
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.f32
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.f32
+// CHECK: ret i1 %hlsl.all
+bool test_all_float(float p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v2f32
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v2f32
+// CHECK: ret i1 %hlsl.all
+bool test_all_float2(float2 p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v3f32
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v3f32
+// CHECK: ret i1 %hlsl.all
+bool test_all_float3(float3 p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v4f32
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v4f32
+// CHECK: ret i1 %hlsl.all
+bool test_all_float4(float4 p0) { return all(p0); }
+
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.f64
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.f64
+// CHECK: ret i1 %hlsl.all
+bool test_all_double(double p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v2f64
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v2f64
+// CHECK: ret i1 %hlsl.all
+bool test_all_double2(double2 p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v3f64
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v3f64
+// CHECK: ret i1 %hlsl.all
+bool test_all_double3(double3 p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v4f64
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v4f64
+// CHECK: ret i1 %hlsl.all
+bool test_all_double4(double4 p0) { return all(p0); }
+
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.i32
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.i32
+// CHECK: ret i1 %hlsl.all
+bool test_all_int(int p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v2i32
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v2i32
+// CHECK: ret i1 %hlsl.all
+bool test_all_int2(int2 p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v3i32
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v3i32
+// CHECK: ret i1 %hlsl.all
+bool test_all_int3(int3 p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v4i32
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v4i32
+// CHECK: ret i1 %hlsl.all
+bool test_all_int4(int4 p0) { return all(p0); }
+
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.i32
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.i32
+// CHECK: ret i1 %hlsl.all
+bool test_all_uint(uint p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v2i32
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v2i32
+// CHECK: ret i1 %hlsl.all
+bool test_all_uint2(uint2 p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v3i32
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v3i32
+// CHECK: ret i1 %hlsl.all
+bool test_all_uint3(uint3 p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v4i32
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v4i32
+// CHECK: ret i1 %hlsl.all
+bool test_all_uint4(uint4 p0) { return all(p0); }
+
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.i64
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.i64
+// CHECK: ret i1 %hlsl.all
+bool test_all_int64_t(int64_t p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v2i64
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v2i64
+// CHECK: ret i1 %hlsl.all
+bool test_all_int64_t2(int64_t2 p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v3i64
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v3i64
+// CHECK: ret i1 %hlsl.all
+bool test_all_int64_t3(int64_t3 p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v4i64
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v4i64
+// CHECK: ret i1 %hlsl.all
+bool test_all_int64_t4(int64_t4 p0) { return all(p0); }
+
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.i64
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.i64
+// CHECK: ret i1 %hlsl.all
+bool test_all_uint64_t(uint64_t p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v2i64
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v2i64
+// CHECK: ret i1 %hlsl.all
+bool test_all_uint64_t2(uint64_t2 p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v3i64
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v3i64
+// CHECK: ret i1 %hlsl.all
+bool test_all_uint64_t3(uint64_t3 p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v4i64
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v4i64
+// CHECK: ret i1 %hlsl.all
+bool test_all_uint64_t4(uint64_t4 p0) { return all(p0); }
+
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.i1
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.i1
+// CHECK: ret i1 %hlsl.all
+bool test_all_bool(bool p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v2i1
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v2i1
+// CHECK: ret i1 %hlsl.all
+bool test_all_bool2(bool2 p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v3i1
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v3i1
+// CHECK: ret i1 %hlsl.all
+bool test_all_bool3(bool3 p0) { return all(p0); }
+// DXIL_CHECK: define noundef i1 @
+// SPIR_CHECK: define spir_func noundef i1 @
+// DXIL_CHECK: %hlsl.all = call i1 @llvm.dx.all.v4i1
+// SPIR_CHECK: %hlsl.all = call i1 @llvm.spv.all.v4i1
+// CHECK: ret i1 %hlsl.all
+bool test_all_bool4(bool4 p0) { return all(p0); }
diff --git a/llvm/include/llvm/IR/Intrinsics.td b/llvm/include/llvm/IR/Intrinsics.td
index c04f4c52692158..1163bd6c8fbaba 100644
--- a/llvm/include/llvm/IR/Intrinsics.td
+++ b/llvm/include/llvm/IR/Intrinsics.td
@@ -684,6 +684,18 @@ class DefaultAttrsIntrinsic<list<LLVMType> ret_types,
intr_properties, name,
sd_properties, /*disable_default_attributes*/ 0> {}
+// HLSL Intrinsic Classes
+class HLSLThreadId
+ : Intrinsic</*ret_types*/ [llvm_i32_ty],
+ /*param_types*/[llvm_i32_ty],
+ /*intr_properties*/ [IntrNoMem, IntrWillReturn],
+ /*name*/ ""> {}
+class HLSLAll : Intrinsic</*ret_types*/ [llvm_i1_ty],
+ /*param_types*/[llvm_any_ty],
+ /*intr_properties*/ [IntrNoMem, IntrWillReturn],
+ /*name*/ ""> {}
+
+
/// ClangBuiltin - If this intrinsic exactly corresponds to a Clang builtin, this
/// specifies the name of the builtin. This provides automatic CBE and CFE
/// support.
diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td
index a871fac46b9fd5..c67102a14387f4 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -11,8 +11,7 @@
//===----------------------------------------------------------------------===//
let TargetPrefix = "dx" in {
-
-def int_dx_thread_id : Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem, IntrWillReturn]>;
+def int_dx_thread_id : HLSLThreadId;
def int_dx_group_id : Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem, IntrWillReturn]>;
def int_dx_thread_id_in_group : Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem, IntrWillReturn]>;
def int_dx_flattened_thread_id_in_group : Intrinsic<[llvm_i32_ty], [], [IntrNoMem, IntrWillReturn]>;
@@ -20,6 +19,7 @@ def int_dx_flattened_thread_id_in_group : Intrinsic<[llvm_i32_ty], [], [IntrNoMe
def int_dx_create_handle : ClangBuiltin<"__builtin_hlsl_create_handle">,
Intrinsic<[ llvm_ptr_ty ], [llvm_i8_ty], [IntrWillReturn]>;
+def int_dx_all : HLSLAll;
def int_dx_any : DefaultAttrsIntrinsic<[llvm_i1_ty], [llvm_any_ty]>;
def int_dx_clamp : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>]>;
def int_dx_uclamp : DefaultAttrsIntrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>]>;
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 0eb09b1699aff4..f987d7928ab23f 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -53,7 +53,8 @@ let TargetPrefix = "spv" in {
ImmArg<ArgIndex<0>>]>;
// The following intrinsic(s) are mirrored from IntrinsicsDirectX.td for HLSL support.
- def int_spv_thread_id : Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem, IntrWillReturn]>;
+ def int_spv_thread_id : HLSLThreadId;
+ def int_spv_all : HLSLAll;
def int_spv_create_handle : ClangBuiltin<"__builtin_hlsl_create_handle">,
Intrinsic<[ llvm_ptr_ty ], [llvm_i8_ty], [IntrWillReturn]>;
}
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index 49749b56345306..0a8383c3df4c76 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -144,6 +144,9 @@ class SPIRVInstructionSelector : public InstructionSelector {
bool selectAddrSpaceCast(Register ResVReg, const SPIRVType *ResType,
MachineInstr &I) const;
+ bool selectAll(Register ResVReg, const SPIRVType *ResType,
+ MachineInstr &I) const;
+
bool selectBitreverse(Register ResVReg, const SPIRVType *ResType,
MachineInstr &I) const;
@@ -1155,6 +1158,20 @@ static unsigned getBoolCmpOpcode(unsigned PredNum) {
}
}
+bool SPIRVInstructionSelector::selectAll(Register ResVReg,
+ const SPIRVType *ResType,
+ MachineInstr &I) const {
+ assert(I.getNumOperands() == 3);
+ assert(I.getOperand(2).isReg());
+
+ MachineBasicBlock &BB = *I.getParent();
+ return BuildMI(BB, I, I.getDebugLoc(), TII.get(SPIRV::OpAll))
+ .addDef(ResVReg)
+ .addUse(GR.getSPIRVTypeID(ResType))
+ .addUse(I.getOperand(2).getReg())
+ .constrainAllUses(TII, TRI, RBI);
+}
+
bool SPIRVInstructionSelector::selectBitreverse(Register ResVReg,
const SPIRVType *ResType,
MachineInstr &I) const {
@@ -1785,6 +1802,8 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg,
break;
case Intrinsic::spv_thread_id:
return selectSpvThreadId(ResVReg, ResType, I);
+ case Intrinsic::spv_all:
+ return selectAll(ResVReg, ResType, I);
case Intrinsic::spv_lifetime_start:
case Intrinsic::spv_lifetime_end: {
unsigned Op = IID == Intrinsic::spv_lifetime_start ? SPIRV::OpLifetimeStart
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/all.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/all.ll
new file mode 100644
index 00000000000000..c904ac08699355
--- /dev/null
+++ b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/all.ll
@@ -0,0 +1,95 @@
+; RUN: llc -O0 -mtriple=spirv-unknown-unknown %s -o - | FileCheck %s
+; Note: The validator is wrong it wants the return to be a bool vector when it is bool scalar return
+; TODO: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o - -filetype=obj | spirv-val %}
+
+; Make sure spirv operation function calls for all are generated.
+
+; CHECK: OpMemoryModel Logical GLSL450
+
+define noundef i1 @all_bool(i1 noundef %a) {
+entry:
+ ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
+ %hlsl.all = call i1 @llvm.spv.all.i1(i1 %a)
+ ret i1 %hlsl.all
+}
+
+define noundef i1 @all_int64_t(i64 noundef %p0) {
+entry:
+ %p0.addr = alloca i64, align 8
+ store i64 %p0, ptr %p0.addr, align 8
+ %0 = load i64, ptr %p0.addr, align 8
+ ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
+ %hlsl.all = call i1 @llvm.spv.all.i64(i64 %0)
+ ret i1 %hlsl.all
+}
+
+
+define noundef i1 @all_int(i32 noundef %p0) {
+entry:
+ %p0.addr = alloca i32, align 4
+ store i32 %p0, ptr %p0.addr, align 4
+ %0 = load i32, ptr %p0.addr, align 4
+ ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
+ %hlsl.all = call i1 @llvm.spv.all.i32(i32 %0)
+ ret i1 %hlsl.all
+}
+
+
+define noundef i1 @all_int16_t(i16 noundef %p0) {
+entry:
+ %p0.addr = alloca i16, align 2
+ store i16 %p0, ptr %p0.addr, align 2
+ %0 = load i16, ptr %p0.addr, align 2
+ ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
+ %hlsl.all = call i1 @llvm.spv.all.i16(i16 %0)
+ ret i1 %hlsl.all
+}
+
+define noundef i1 @all_double(double noundef %p0) {
+entry:
+ %p0.addr = alloca double, align 8
+ store double %p0, ptr %p0.addr, align 8
+ %0 = load double, ptr %p0.addr, align 8
+ ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
+ %hlsl.all = call i1 @llvm.spv.all.f64(double %0)
+ ret i1 %hlsl.all
+}
+
+
+define noundef i1 @all_float(float noundef %p0) {
+entry:
+ %p0.addr = alloca float, align 4
+ store float %p0, ptr %p0.addr, align 4
+ %0 = load float, ptr %p0.addr, align 4
+ ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
+ %hlsl.all = call i1 @llvm.spv.all.f32(float %0)
+ ret i1 %hlsl.all
+}
+
+
+define noundef i1 @all_half(half noundef %p0) {
+entry:
+ %p0.addr = alloca half, align 2
+ store half %p0, ptr %p0.addr, align 2
+ %0 = load half, ptr %p0.addr, align 2
+ ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
+ %hlsl.all = call i1 @llvm.spv.all.f16(half %0)
+ ret i1 %hlsl.all
+}
+
+
+define noundef i1 @all_bool4(<4 x i1> noundef %p0) {
+entry:
+ ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
+ %hlsl.all = call i1 @llvm.spv.all.v4i1(<4 x i1> %p0)
+ ret i1 %hlsl.all
+}
+
+declare i1 @llvm.spv.all.v4i1(<4 x i1>)
+declare i1 @llvm.spv.all.i1(i1)
+declare i1 @llvm.spv.all.i16(i16)
+declare i1 @llvm.spv.all.i32(i32)
+declare i1 @llvm.spv.all.i64(i64)
+declare i1 @llvm.spv.all.f16(half)
+declare i1 @llvm.spv.all.f32(float)
+declare i1 @llvm.spv.all.f64(double)
>From c9f74a036cabb369603aa7c652c0a54fd17df6ce Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlotfi at microsoft.com>
Date: Mon, 1 Apr 2024 19:57:34 -0400
Subject: [PATCH 02/11] add a macro for switch statement, remove type classes
---
clang/lib/CodeGen/CGBuiltin.cpp | 16 ++-------
clang/lib/CodeGen/CGHLSLRuntime.cpp | 18 +++-------
clang/lib/CodeGen/CGHLSLUtils.h | 44 +++++++++++++++++++++++
llvm/include/llvm/IR/Intrinsics.td | 12 -------
llvm/include/llvm/IR/IntrinsicsDirectX.td | 7 ++--
llvm/include/llvm/IR/IntrinsicsSPIRV.td | 4 +--
6 files changed, 57 insertions(+), 44 deletions(-)
create mode 100644 clang/lib/CodeGen/CGHLSLUtils.h
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 2acf799f168785..92e9089cbe7bb7 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -13,6 +13,7 @@
#include "ABIInfo.h"
#include "CGCUDARuntime.h"
#include "CGCXXABI.h"
+#include "CGHLSLUtils.h"
#include "CGObjCRuntime.h"
#include "CGOpenCLRuntime.h"
#include "CGRecordLayout.h"
@@ -51,7 +52,6 @@
#include "llvm/IR/IntrinsicsR600.h"
#include "llvm/IR/IntrinsicsRISCV.h"
#include "llvm/IR/IntrinsicsS390.h"
-#include "llvm/IR/IntrinsicsSPIRV.h"
#include "llvm/IR/IntrinsicsVE.h"
#include "llvm/IR/IntrinsicsWebAssembly.h"
#include "llvm/IR/IntrinsicsX86.h"
@@ -18167,17 +18167,6 @@ Intrinsic::ID getDotProductIntrinsic(QualType QT, int elementCount) {
return Intrinsic::dx_udot;
}
-Intrinsic::ID getAllIntrinsic(const llvm::Triple::ArchType Arch) {
- switch (Arch) {
- case llvm::Triple::dxil:
- return Intrinsic::dx_all;
- case llvm::Triple::spirv:
- return Intrinsic::spv_all;
- default:
- llvm_unreachable("Input semantic not supported by target");
- }
-}
-
Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
const CallExpr *E) {
if (!getLangOpts().HLSL)
@@ -18188,7 +18177,8 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
Value *Op0 = EmitScalarExpr(E->getArg(0));
return Builder.CreateIntrinsic(
/*ReturnType=*/llvm::Type::getInt1Ty(getLLVMContext()),
- getAllIntrinsic(CGM.getTarget().getTriple().getArch()),
+ CGHLSLUtils::get_hlsl_all_intrinsic(
+ CGM.getTarget().getTriple().getArch()),
ArrayRef<Value *>{Op0}, nullptr, "hlsl.all");
}
case Builtin::BI__builtin_hlsl_elementwise_any: {
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.cpp b/clang/lib/CodeGen/CGHLSLRuntime.cpp
index 794d93358b0a4c..c772049dbe3acc 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.cpp
+++ b/clang/lib/CodeGen/CGHLSLRuntime.cpp
@@ -14,11 +14,10 @@
#include "CGHLSLRuntime.h"
#include "CGDebugInfo.h"
+#include "CGHLSLUtils.h"
#include "CodeGenModule.h"
#include "clang/AST/Decl.h"
#include "clang/Basic/TargetOptions.h"
-#include "llvm/IR/IntrinsicsDirectX.h"
-#include "llvm/IR/IntrinsicsSPIRV.h"
#include "llvm/IR/Metadata.h"
#include "llvm/IR/Module.h"
#include "llvm/Support/FormatVariadic.h"
@@ -343,18 +342,9 @@ llvm::Value *CGHLSLRuntime::emitInputSemantic(IRBuilder<> &B,
return B.CreateCall(FunctionCallee(DxGroupIndex));
}
if (D.hasAttr<HLSLSV_DispatchThreadIDAttr>()) {
- llvm::Function *ThreadIDIntrinsic;
- switch (CGM.getTarget().getTriple().getArch()) {
- case llvm::Triple::dxil:
- ThreadIDIntrinsic = CGM.getIntrinsic(Intrinsic::dx_thread_id);
- break;
- case llvm::Triple::spirv:
- ThreadIDIntrinsic = CGM.getIntrinsic(Intrinsic::spv_thread_id);
- break;
- default:
- llvm_unreachable("Input semantic not supported by target");
- break;
- }
+ llvm::Function *ThreadIDIntrinsic =
+ CGM.getIntrinsic(CGHLSLUtils::get_hlsl_thread_id_intrinsic(
+ CGM.getTarget().getTriple().getArch()));
return buildVectorInput(B, ThreadIDIntrinsic, Ty);
}
assert(false && "Unhandled parameter attribute");
diff --git a/clang/lib/CodeGen/CGHLSLUtils.h b/clang/lib/CodeGen/CGHLSLUtils.h
new file mode 100644
index 00000000000000..ca29835105aa87
--- /dev/null
+++ b/clang/lib/CodeGen/CGHLSLUtils.h
@@ -0,0 +1,44 @@
+
+//===----- CGHLSLUtils.h - Utility functions for HLSL CodeGen ---*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This File Provides utility function for HLSL code generation.
+// It is used to abstract away implementation details of backends.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_CLANG_LIB_CODEGEN_CGHLSLUTILS_H
+#define LLVM_CLANG_LIB_CODEGEN_CGHLSLUTILS_H
+
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicsDirectX.h"
+#include "llvm/IR/IntrinsicsSPIRV.h"
+
+// Define the function generator macro
+#define GENERATE_HLSL_INTRINSIC_FUNCTION(name) \
+ static llvm::Intrinsic::ID get_hlsl_##name##_intrinsic( \
+ const llvm::Triple::ArchType Arch) { \
+ switch (Arch) { \
+ case llvm::Triple::dxil: \
+ return llvm::Intrinsic::dx_##name; \
+ case llvm::Triple::spirv: \
+ return llvm::Intrinsic::spv_##name; \
+ default: \
+ llvm_unreachable("Input semantic not supported by target"); \
+ } \
+ }
+
+class CGHLSLUtils {
+public:
+ GENERATE_HLSL_INTRINSIC_FUNCTION(all)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(thread_id)
+private:
+ CGHLSLUtils() = delete;
+};
+
+#endif
diff --git a/llvm/include/llvm/IR/Intrinsics.td b/llvm/include/llvm/IR/Intrinsics.td
index 1163bd6c8fbaba..c04f4c52692158 100644
--- a/llvm/include/llvm/IR/Intrinsics.td
+++ b/llvm/include/llvm/IR/Intrinsics.td
@@ -684,18 +684,6 @@ class DefaultAttrsIntrinsic<list<LLVMType> ret_types,
intr_properties, name,
sd_properties, /*disable_default_attributes*/ 0> {}
-// HLSL Intrinsic Classes
-class HLSLThreadId
- : Intrinsic</*ret_types*/ [llvm_i32_ty],
- /*param_types*/[llvm_i32_ty],
- /*intr_properties*/ [IntrNoMem, IntrWillReturn],
- /*name*/ ""> {}
-class HLSLAll : Intrinsic</*ret_types*/ [llvm_i1_ty],
- /*param_types*/[llvm_any_ty],
- /*intr_properties*/ [IntrNoMem, IntrWillReturn],
- /*name*/ ""> {}
-
-
/// ClangBuiltin - If this intrinsic exactly corresponds to a Clang builtin, this
/// specifies the name of the builtin. This provides automatic CBE and CFE
/// support.
diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td
index c67102a14387f4..a7f212da2f5b63 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -11,7 +11,8 @@
//===----------------------------------------------------------------------===//
let TargetPrefix = "dx" in {
-def int_dx_thread_id : HLSLThreadId;
+
+def int_dx_thread_id : Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem, IntrWillReturn]>;
def int_dx_group_id : Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem, IntrWillReturn]>;
def int_dx_thread_id_in_group : Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem, IntrWillReturn]>;
def int_dx_flattened_thread_id_in_group : Intrinsic<[llvm_i32_ty], [], [IntrNoMem, IntrWillReturn]>;
@@ -19,8 +20,8 @@ def int_dx_flattened_thread_id_in_group : Intrinsic<[llvm_i32_ty], [], [IntrNoMe
def int_dx_create_handle : ClangBuiltin<"__builtin_hlsl_create_handle">,
Intrinsic<[ llvm_ptr_ty ], [llvm_i8_ty], [IntrWillReturn]>;
-def int_dx_all : HLSLAll;
-def int_dx_any : DefaultAttrsIntrinsic<[llvm_i1_ty], [llvm_any_ty]>;
+def int_dx_all : DefaultAttrsIntrinsic<[llvm_i1_ty], [llvm_any_ty]>;
+def int_dx_any : DefaultAttrsIntrinsic<[llvm_i1_ty], [llvm_any_ty]>;
def int_dx_clamp : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>]>;
def int_dx_uclamp : DefaultAttrsIntrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>]>;
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index f987d7928ab23f..f843383f0b00ed 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -53,8 +53,8 @@ let TargetPrefix = "spv" in {
ImmArg<ArgIndex<0>>]>;
// The following intrinsic(s) are mirrored from IntrinsicsDirectX.td for HLSL support.
- def int_spv_thread_id : HLSLThreadId;
- def int_spv_all : HLSLAll;
+ def int_spv_thread_id : Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem, IntrWillReturn]>;
def int_spv_create_handle : ClangBuiltin<"__builtin_hlsl_create_handle">,
Intrinsic<[ llvm_ptr_ty ], [llvm_i8_ty], [IntrWillReturn]>;
+ def int_spv_all : DefaultAttrsIntrinsic<[llvm_i1_ty], [llvm_any_ty]>;
}
>From 4fb0e4dab3bf79c6cbd210b1217cfb2eef6da83a Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlotfi at microsoft.com>
Date: Tue, 2 Apr 2024 20:38:15 -0400
Subject: [PATCH 03/11] address pr comments
---
clang/lib/CodeGen/CGBuiltin.cpp | 2 +-
clang/lib/CodeGen/CGHLSLRuntime.cpp | 2 +-
clang/lib/CodeGen/CGHLSLUtils.h | 20 +++++++++++---------
3 files changed, 13 insertions(+), 11 deletions(-)
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 92e9089cbe7bb7..c31576afcfc28b 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18177,7 +18177,7 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
Value *Op0 = EmitScalarExpr(E->getArg(0));
return Builder.CreateIntrinsic(
/*ReturnType=*/llvm::Type::getInt1Ty(getLLVMContext()),
- CGHLSLUtils::get_hlsl_all_intrinsic(
+ HLSLUtils::get_hlsl_all_intrinsic(
CGM.getTarget().getTriple().getArch()),
ArrayRef<Value *>{Op0}, nullptr, "hlsl.all");
}
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.cpp b/clang/lib/CodeGen/CGHLSLRuntime.cpp
index c772049dbe3acc..78c98ed4e13788 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.cpp
+++ b/clang/lib/CodeGen/CGHLSLRuntime.cpp
@@ -343,7 +343,7 @@ llvm::Value *CGHLSLRuntime::emitInputSemantic(IRBuilder<> &B,
}
if (D.hasAttr<HLSLSV_DispatchThreadIDAttr>()) {
llvm::Function *ThreadIDIntrinsic =
- CGM.getIntrinsic(CGHLSLUtils::get_hlsl_thread_id_intrinsic(
+ CGM.getIntrinsic(HLSLUtils::get_hlsl_thread_id_intrinsic(
CGM.getTarget().getTriple().getArch()));
return buildVectorInput(B, ThreadIDIntrinsic, Ty);
}
diff --git a/clang/lib/CodeGen/CGHLSLUtils.h b/clang/lib/CodeGen/CGHLSLUtils.h
index ca29835105aa87..deb2ca4f2d0ed8 100644
--- a/clang/lib/CodeGen/CGHLSLUtils.h
+++ b/clang/lib/CodeGen/CGHLSLUtils.h
@@ -7,7 +7,7 @@
//
//===----------------------------------------------------------------------===//
//
-// This File Provides utility function for HLSL code generation.
+// This file provides utility functions for HLSL code generation.
// It is used to abstract away implementation details of backends.
//
//===----------------------------------------------------------------------===//
@@ -33,12 +33,14 @@
} \
}
-class CGHLSLUtils {
-public:
- GENERATE_HLSL_INTRINSIC_FUNCTION(all)
- GENERATE_HLSL_INTRINSIC_FUNCTION(thread_id)
-private:
- CGHLSLUtils() = delete;
-};
+namespace clang::CodeGen {
+ class HLSLUtils {
+ public:
+ GENERATE_HLSL_INTRINSIC_FUNCTION(all)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(thread_id)
+ private:
+ HLSLUtils() = delete;
+ };
+} // namespace clang::CodeGen
-#endif
+#endif // LLVM_CLANG_LIB_CODEGEN_CGHLSLUTILS_H
>From 4236523ae52b0e8d44c2dd31a687bc64cda5e237 Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlotfi at microsoft.com>
Date: Tue, 2 Apr 2024 20:44:48 -0400
Subject: [PATCH 04/11] remove spirv lowering for all intrinsic
---
.../Target/SPIRV/SPIRVInstructionSelector.cpp | 19 ----
.../test/CodeGen/SPIRV/hlsl-intrinsics/all.ll | 95 -------------------
2 files changed, 114 deletions(-)
delete mode 100644 llvm/test/CodeGen/SPIRV/hlsl-intrinsics/all.ll
diff --git a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
index 0a8383c3df4c76..49749b56345306 100644
--- a/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVInstructionSelector.cpp
@@ -144,9 +144,6 @@ class SPIRVInstructionSelector : public InstructionSelector {
bool selectAddrSpaceCast(Register ResVReg, const SPIRVType *ResType,
MachineInstr &I) const;
- bool selectAll(Register ResVReg, const SPIRVType *ResType,
- MachineInstr &I) const;
-
bool selectBitreverse(Register ResVReg, const SPIRVType *ResType,
MachineInstr &I) const;
@@ -1158,20 +1155,6 @@ static unsigned getBoolCmpOpcode(unsigned PredNum) {
}
}
-bool SPIRVInstructionSelector::selectAll(Register ResVReg,
- const SPIRVType *ResType,
- MachineInstr &I) const {
- assert(I.getNumOperands() == 3);
- assert(I.getOperand(2).isReg());
-
- MachineBasicBlock &BB = *I.getParent();
- return BuildMI(BB, I, I.getDebugLoc(), TII.get(SPIRV::OpAll))
- .addDef(ResVReg)
- .addUse(GR.getSPIRVTypeID(ResType))
- .addUse(I.getOperand(2).getReg())
- .constrainAllUses(TII, TRI, RBI);
-}
-
bool SPIRVInstructionSelector::selectBitreverse(Register ResVReg,
const SPIRVType *ResType,
MachineInstr &I) const {
@@ -1802,8 +1785,6 @@ bool SPIRVInstructionSelector::selectIntrinsic(Register ResVReg,
break;
case Intrinsic::spv_thread_id:
return selectSpvThreadId(ResVReg, ResType, I);
- case Intrinsic::spv_all:
- return selectAll(ResVReg, ResType, I);
case Intrinsic::spv_lifetime_start:
case Intrinsic::spv_lifetime_end: {
unsigned Op = IID == Intrinsic::spv_lifetime_start ? SPIRV::OpLifetimeStart
diff --git a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/all.ll b/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/all.ll
deleted file mode 100644
index c904ac08699355..00000000000000
--- a/llvm/test/CodeGen/SPIRV/hlsl-intrinsics/all.ll
+++ /dev/null
@@ -1,95 +0,0 @@
-; RUN: llc -O0 -mtriple=spirv-unknown-unknown %s -o - | FileCheck %s
-; Note: The validator is wrong it wants the return to be a bool vector when it is bool scalar return
-; TODO: %if spirv-tools %{ llc -O0 -mtriple=spirv-unknown-unknown %s -o - -filetype=obj | spirv-val %}
-
-; Make sure spirv operation function calls for all are generated.
-
-; CHECK: OpMemoryModel Logical GLSL450
-
-define noundef i1 @all_bool(i1 noundef %a) {
-entry:
- ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
- %hlsl.all = call i1 @llvm.spv.all.i1(i1 %a)
- ret i1 %hlsl.all
-}
-
-define noundef i1 @all_int64_t(i64 noundef %p0) {
-entry:
- %p0.addr = alloca i64, align 8
- store i64 %p0, ptr %p0.addr, align 8
- %0 = load i64, ptr %p0.addr, align 8
- ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
- %hlsl.all = call i1 @llvm.spv.all.i64(i64 %0)
- ret i1 %hlsl.all
-}
-
-
-define noundef i1 @all_int(i32 noundef %p0) {
-entry:
- %p0.addr = alloca i32, align 4
- store i32 %p0, ptr %p0.addr, align 4
- %0 = load i32, ptr %p0.addr, align 4
- ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
- %hlsl.all = call i1 @llvm.spv.all.i32(i32 %0)
- ret i1 %hlsl.all
-}
-
-
-define noundef i1 @all_int16_t(i16 noundef %p0) {
-entry:
- %p0.addr = alloca i16, align 2
- store i16 %p0, ptr %p0.addr, align 2
- %0 = load i16, ptr %p0.addr, align 2
- ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
- %hlsl.all = call i1 @llvm.spv.all.i16(i16 %0)
- ret i1 %hlsl.all
-}
-
-define noundef i1 @all_double(double noundef %p0) {
-entry:
- %p0.addr = alloca double, align 8
- store double %p0, ptr %p0.addr, align 8
- %0 = load double, ptr %p0.addr, align 8
- ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
- %hlsl.all = call i1 @llvm.spv.all.f64(double %0)
- ret i1 %hlsl.all
-}
-
-
-define noundef i1 @all_float(float noundef %p0) {
-entry:
- %p0.addr = alloca float, align 4
- store float %p0, ptr %p0.addr, align 4
- %0 = load float, ptr %p0.addr, align 4
- ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
- %hlsl.all = call i1 @llvm.spv.all.f32(float %0)
- ret i1 %hlsl.all
-}
-
-
-define noundef i1 @all_half(half noundef %p0) {
-entry:
- %p0.addr = alloca half, align 2
- store half %p0, ptr %p0.addr, align 2
- %0 = load half, ptr %p0.addr, align 2
- ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
- %hlsl.all = call i1 @llvm.spv.all.f16(half %0)
- ret i1 %hlsl.all
-}
-
-
-define noundef i1 @all_bool4(<4 x i1> noundef %p0) {
-entry:
- ; CHECK: %[[#]] = OpAll %[[#]] %[[#]]
- %hlsl.all = call i1 @llvm.spv.all.v4i1(<4 x i1> %p0)
- ret i1 %hlsl.all
-}
-
-declare i1 @llvm.spv.all.v4i1(<4 x i1>)
-declare i1 @llvm.spv.all.i1(i1)
-declare i1 @llvm.spv.all.i16(i16)
-declare i1 @llvm.spv.all.i32(i32)
-declare i1 @llvm.spv.all.i64(i64)
-declare i1 @llvm.spv.all.f16(half)
-declare i1 @llvm.spv.all.f32(float)
-declare i1 @llvm.spv.all.f64(double)
>From 176b025b2f57f5c34497c8ae8ae0450a1c70ac36 Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlotfi at microsoft.com>
Date: Tue, 2 Apr 2024 20:55:44 -0400
Subject: [PATCH 05/11] fix format
---
clang/lib/CodeGen/CGHLSLUtils.h | 17 +++++++++--------
1 file changed, 9 insertions(+), 8 deletions(-)
diff --git a/clang/lib/CodeGen/CGHLSLUtils.h b/clang/lib/CodeGen/CGHLSLUtils.h
index deb2ca4f2d0ed8..189a37cc6d98b4 100644
--- a/clang/lib/CodeGen/CGHLSLUtils.h
+++ b/clang/lib/CodeGen/CGHLSLUtils.h
@@ -29,18 +29,19 @@
case llvm::Triple::spirv: \
return llvm::Intrinsic::spv_##name; \
default: \
- llvm_unreachable("Input semantic not supported by target"); \
+ llvm_unreachable("Intrinsic " #name \
+ " not supported by target architecture"); \
} \
}
namespace clang::CodeGen {
- class HLSLUtils {
- public:
- GENERATE_HLSL_INTRINSIC_FUNCTION(all)
- GENERATE_HLSL_INTRINSIC_FUNCTION(thread_id)
- private:
- HLSLUtils() = delete;
- };
+class HLSLUtils {
+public:
+ GENERATE_HLSL_INTRINSIC_FUNCTION(all)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(thread_id)
+private:
+ HLSLUtils() = delete;
+};
} // namespace clang::CodeGen
#endif // LLVM_CLANG_LIB_CODEGEN_CGHLSLUTILS_H
>From 23e15ced6bd47996056a17a7fa33a1b22726bed0 Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlotfi at microsoft.com>
Date: Wed, 3 Apr 2024 13:07:12 -0400
Subject: [PATCH 06/11] move HLSLUtils functions to CGHLSLRuntime
---
clang/lib/CodeGen/CGBuiltin.cpp | 4 +--
clang/lib/CodeGen/CGHLSLRuntime.cpp | 3 +-
clang/lib/CodeGen/CGHLSLRuntime.h | 29 ++++++++++++++++++
clang/lib/CodeGen/CGHLSLUtils.h | 47 -----------------------------
4 files changed, 32 insertions(+), 51 deletions(-)
delete mode 100644 clang/lib/CodeGen/CGHLSLUtils.h
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c31576afcfc28b..11748444f12743 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -13,7 +13,7 @@
#include "ABIInfo.h"
#include "CGCUDARuntime.h"
#include "CGCXXABI.h"
-#include "CGHLSLUtils.h"
+#include "CGHLSLRuntime.h"
#include "CGObjCRuntime.h"
#include "CGOpenCLRuntime.h"
#include "CGRecordLayout.h"
@@ -18177,7 +18177,7 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
Value *Op0 = EmitScalarExpr(E->getArg(0));
return Builder.CreateIntrinsic(
/*ReturnType=*/llvm::Type::getInt1Ty(getLLVMContext()),
- HLSLUtils::get_hlsl_all_intrinsic(
+ CGHLSLRuntime::get_hlsl_all_intrinsic(
CGM.getTarget().getTriple().getArch()),
ArrayRef<Value *>{Op0}, nullptr, "hlsl.all");
}
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.cpp b/clang/lib/CodeGen/CGHLSLRuntime.cpp
index 78c98ed4e13788..13143998e9dd0a 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.cpp
+++ b/clang/lib/CodeGen/CGHLSLRuntime.cpp
@@ -14,7 +14,6 @@
#include "CGHLSLRuntime.h"
#include "CGDebugInfo.h"
-#include "CGHLSLUtils.h"
#include "CodeGenModule.h"
#include "clang/AST/Decl.h"
#include "clang/Basic/TargetOptions.h"
@@ -343,7 +342,7 @@ llvm::Value *CGHLSLRuntime::emitInputSemantic(IRBuilder<> &B,
}
if (D.hasAttr<HLSLSV_DispatchThreadIDAttr>()) {
llvm::Function *ThreadIDIntrinsic =
- CGM.getIntrinsic(HLSLUtils::get_hlsl_thread_id_intrinsic(
+ CGM.getIntrinsic(CGHLSLRuntime::get_hlsl_thread_id_intrinsic(
CGM.getTarget().getTriple().getArch()));
return buildVectorInput(B, ThreadIDIntrinsic, Ty);
}
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index bffefb66740a00..26bdb6ab38f0ba 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -16,6 +16,9 @@
#define LLVM_CLANG_LIB_CODEGEN_CGHLSLRUNTIME_H
#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicsDirectX.h"
+#include "llvm/IR/IntrinsicsSPIRV.h"
#include "clang/Basic/HLSLRuntime.h"
@@ -26,6 +29,21 @@
#include <optional>
#include <vector>
+// Define the function generator macro
+#define GENERATE_HLSL_INTRINSIC_FUNCTION(name) \
+ static llvm::Intrinsic::ID get_hlsl_##name##_intrinsic( \
+ const llvm::Triple::ArchType Arch) { \
+ switch (Arch) { \
+ case llvm::Triple::dxil: \
+ return llvm::Intrinsic::dx_##name; \
+ case llvm::Triple::spirv: \
+ return llvm::Intrinsic::spv_##name; \
+ default: \
+ llvm_unreachable("Intrinsic " #name \
+ " not supported by target architecture"); \
+ } \
+ }
+
namespace llvm {
class GlobalVariable;
class Function;
@@ -48,6 +66,17 @@ class CodeGenModule;
class CGHLSLRuntime {
public:
+ //===----------------------------------------------------------------------===//
+ // Start of reserved area for HLSL intrinsic getters.
+ //===----------------------------------------------------------------------===//
+
+ GENERATE_HLSL_INTRINSIC_FUNCTION(all)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(thread_id)
+
+ //===----------------------------------------------------------------------===//
+ // End of reserved area for HLSL intrinsic getters.
+ //===----------------------------------------------------------------------===//
+
struct BufferResBinding {
// The ID like 2 in register(b2, space1).
std::optional<unsigned> Reg;
diff --git a/clang/lib/CodeGen/CGHLSLUtils.h b/clang/lib/CodeGen/CGHLSLUtils.h
deleted file mode 100644
index 189a37cc6d98b4..00000000000000
--- a/clang/lib/CodeGen/CGHLSLUtils.h
+++ /dev/null
@@ -1,47 +0,0 @@
-
-//===----- CGHLSLUtils.h - Utility functions for HLSL CodeGen ---*- C++ -*-===//
-//
-// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-// See https://llvm.org/LICENSE.txt for license information.
-// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-//
-//===----------------------------------------------------------------------===//
-//
-// This file provides utility functions for HLSL code generation.
-// It is used to abstract away implementation details of backends.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef LLVM_CLANG_LIB_CODEGEN_CGHLSLUTILS_H
-#define LLVM_CLANG_LIB_CODEGEN_CGHLSLUTILS_H
-
-#include "llvm/IR/Intrinsics.h"
-#include "llvm/IR/IntrinsicsDirectX.h"
-#include "llvm/IR/IntrinsicsSPIRV.h"
-
-// Define the function generator macro
-#define GENERATE_HLSL_INTRINSIC_FUNCTION(name) \
- static llvm::Intrinsic::ID get_hlsl_##name##_intrinsic( \
- const llvm::Triple::ArchType Arch) { \
- switch (Arch) { \
- case llvm::Triple::dxil: \
- return llvm::Intrinsic::dx_##name; \
- case llvm::Triple::spirv: \
- return llvm::Intrinsic::spv_##name; \
- default: \
- llvm_unreachable("Intrinsic " #name \
- " not supported by target architecture"); \
- } \
- }
-
-namespace clang::CodeGen {
-class HLSLUtils {
-public:
- GENERATE_HLSL_INTRINSIC_FUNCTION(all)
- GENERATE_HLSL_INTRINSIC_FUNCTION(thread_id)
-private:
- HLSLUtils() = delete;
-};
-} // namespace clang::CodeGen
-
-#endif // LLVM_CLANG_LIB_CODEGEN_CGHLSLUTILS_H
>From 2d839242e3da501ed80440c3497dc747ebb71299 Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlotfi at microsoft.com>
Date: Wed, 3 Apr 2024 14:03:55 -0400
Subject: [PATCH 07/11] make intrinsic getters member functions. Add a getArch
abstraction b\c CodeGenModule is incomplete in the header\CGHLSLRuntime.h
---
clang/lib/CodeGen/CGBuiltin.cpp | 5 ++---
clang/lib/CodeGen/CGHLSLRuntime.cpp | 7 +++++--
clang/lib/CodeGen/CGHLSLRuntime.h | 5 +++--
3 files changed, 10 insertions(+), 7 deletions(-)
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 11748444f12743..effa76429decef 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18177,9 +18177,8 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
Value *Op0 = EmitScalarExpr(E->getArg(0));
return Builder.CreateIntrinsic(
/*ReturnType=*/llvm::Type::getInt1Ty(getLLVMContext()),
- CGHLSLRuntime::get_hlsl_all_intrinsic(
- CGM.getTarget().getTriple().getArch()),
- ArrayRef<Value *>{Op0}, nullptr, "hlsl.all");
+ CGM.getHLSLRuntime().get_hlsl_all_intrinsic(), ArrayRef<Value *>{Op0},
+ nullptr, "hlsl.all");
}
case Builtin::BI__builtin_hlsl_elementwise_any: {
Value *Op0 = EmitScalarExpr(E->getArg(0));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.cpp b/clang/lib/CodeGen/CGHLSLRuntime.cpp
index 13143998e9dd0a..c42fb4e140d56b 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.cpp
+++ b/clang/lib/CodeGen/CGHLSLRuntime.cpp
@@ -115,6 +115,10 @@ GlobalVariable *replaceBuffer(CGHLSLRuntime::Buffer &Buf) {
} // namespace
+llvm::Triple::ArchType CGHLSLRuntime::getArch() {
+ return CGM.getTarget().getTriple().getArch();
+}
+
void CGHLSLRuntime::addConstant(VarDecl *D, Buffer &CB) {
if (D->getStorageClass() == SC_Static) {
// For static inside cbuffer, take as global static.
@@ -342,8 +346,7 @@ llvm::Value *CGHLSLRuntime::emitInputSemantic(IRBuilder<> &B,
}
if (D.hasAttr<HLSLSV_DispatchThreadIDAttr>()) {
llvm::Function *ThreadIDIntrinsic =
- CGM.getIntrinsic(CGHLSLRuntime::get_hlsl_thread_id_intrinsic(
- CGM.getTarget().getTriple().getArch()));
+ CGM.getIntrinsic(get_hlsl_thread_id_intrinsic());
return buildVectorInput(B, ThreadIDIntrinsic, Ty);
}
assert(false && "Unhandled parameter attribute");
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index 26bdb6ab38f0ba..ffc0de0450452b 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -31,8 +31,8 @@
// Define the function generator macro
#define GENERATE_HLSL_INTRINSIC_FUNCTION(name) \
- static llvm::Intrinsic::ID get_hlsl_##name##_intrinsic( \
- const llvm::Triple::ArchType Arch) { \
+ llvm::Intrinsic::ID get_hlsl_##name##_intrinsic() { \
+ llvm::Triple::ArchType Arch = getArch(); \
switch (Arch) { \
case llvm::Triple::dxil: \
return llvm::Intrinsic::dx_##name; \
@@ -125,6 +125,7 @@ class CGHLSLRuntime {
BufferResBinding &Binding);
void addConstant(VarDecl *D, Buffer &CB);
void addBufferDecls(const DeclContext *DC, Buffer &CB);
+ llvm::Triple::ArchType getArch();
llvm::SmallVector<Buffer> Buffers;
};
>From a950dc6b3592a08ccac18da88626a987394fd252 Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlotfi at microsoft.com>
Date: Wed, 3 Apr 2024 18:55:15 -0400
Subject: [PATCH 08/11] add a templated macro for when we have builtins and
make GENERATE_HLSL_INTRINSIC_FUNCTION take a function name
---
clang/lib/CodeGen/CGBuiltin.cpp | 5 ++--
clang/lib/CodeGen/CGHLSLRuntime.cpp | 2 +-
clang/lib/CodeGen/CGHLSLRuntime.h | 43 ++++++++++++++++++-----------
3 files changed, 31 insertions(+), 19 deletions(-)
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index effa76429decef..f12312be83422f 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18177,8 +18177,9 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
Value *Op0 = EmitScalarExpr(E->getArg(0));
return Builder.CreateIntrinsic(
/*ReturnType=*/llvm::Type::getInt1Ty(getLLVMContext()),
- CGM.getHLSLRuntime().get_hlsl_all_intrinsic(), ArrayRef<Value *>{Op0},
- nullptr, "hlsl.all");
+ CGM.getHLSLRuntime()
+ .getIntrinsic<Builtin::BI__builtin_hlsl_elementwise_all>(),
+ ArrayRef<Value *>{Op0}, nullptr, "hlsl.all");
}
case Builtin::BI__builtin_hlsl_elementwise_any: {
Value *Op0 = EmitScalarExpr(E->getArg(0));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.cpp b/clang/lib/CodeGen/CGHLSLRuntime.cpp
index c42fb4e140d56b..5e6a3dd4878f46 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.cpp
+++ b/clang/lib/CodeGen/CGHLSLRuntime.cpp
@@ -346,7 +346,7 @@ llvm::Value *CGHLSLRuntime::emitInputSemantic(IRBuilder<> &B,
}
if (D.hasAttr<HLSLSV_DispatchThreadIDAttr>()) {
llvm::Function *ThreadIDIntrinsic =
- CGM.getIntrinsic(get_hlsl_thread_id_intrinsic());
+ CGM.getIntrinsic(getThreadIdIntrinsic());
return buildVectorInput(B, ThreadIDIntrinsic, Ty);
}
assert(false && "Unhandled parameter attribute");
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index ffc0de0450452b..17f6516fb276dd 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -20,6 +20,7 @@
#include "llvm/IR/IntrinsicsDirectX.h"
#include "llvm/IR/IntrinsicsSPIRV.h"
+#include "clang/Basic/Builtins.h"
#include "clang/Basic/HLSLRuntime.h"
#include "llvm/ADT/SmallVector.h"
@@ -29,19 +30,29 @@
#include <optional>
#include <vector>
-// Define the function generator macro
-#define GENERATE_HLSL_INTRINSIC_FUNCTION(name) \
- llvm::Intrinsic::ID get_hlsl_##name##_intrinsic() { \
- llvm::Triple::ArchType Arch = getArch(); \
- switch (Arch) { \
- case llvm::Triple::dxil: \
- return llvm::Intrinsic::dx_##name; \
- case llvm::Triple::spirv: \
- return llvm::Intrinsic::spv_##name; \
- default: \
- llvm_unreachable("Intrinsic " #name \
- " not supported by target architecture"); \
- } \
+#define GENERATE_HLSL_INTRINSIC_BASE(IntrinsicPostfix) \
+ llvm::Triple::ArchType Arch = getArch(); \
+ switch (Arch) { \
+ case llvm::Triple::dxil: \
+ return llvm::Intrinsic::dx_##IntrinsicPostfix; \
+ case llvm::Triple::spirv: \
+ return llvm::Intrinsic::spv_##IntrinsicPostfix; \
+ default: \
+ llvm_unreachable("Intrinsic " #IntrinsicPostfix \
+ " not supported by target architecture"); \
+ }
+
+// A function generator macro for when there is no builtins or
+// when builtins are mapped to a set of intrinsics for different types.
+#define GENERATE_HLSL_INTRINSIC_FUNCTION(FunctionName, IntrinsicPostfix) \
+ llvm::Intrinsic::ID get##FunctionName##Intrinsic() { \
+ GENERATE_HLSL_INTRINSIC_BASE(IntrinsicPostfix) \
+ }
+
+// A template function generator macro for when we have builtins.
+#define GENERATE_HLSL_INTRINSIC_TEMPLATE(BuiltinName, IntrinsicPostfix) \
+ template <> llvm::Intrinsic::ID getIntrinsic<Builtin::BI__##BuiltinName>() { \
+ GENERATE_HLSL_INTRINSIC_BASE(IntrinsicPostfix) \
}
namespace llvm {
@@ -69,9 +80,9 @@ class CGHLSLRuntime {
//===----------------------------------------------------------------------===//
// Start of reserved area for HLSL intrinsic getters.
//===----------------------------------------------------------------------===//
-
- GENERATE_HLSL_INTRINSIC_FUNCTION(all)
- GENERATE_HLSL_INTRINSIC_FUNCTION(thread_id)
+ template <unsigned BI> llvm::Intrinsic::ID getIntrinsic();
+ GENERATE_HLSL_INTRINSIC_TEMPLATE(builtin_hlsl_elementwise_all, all)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(ThreadId, thread_id)
//===----------------------------------------------------------------------===//
// End of reserved area for HLSL intrinsic getters.
>From 998f8cada38e7708ad0454bb6a685e5801e5b676 Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlotfi at microsoft.com>
Date: Wed, 3 Apr 2024 19:37:58 -0400
Subject: [PATCH 09/11] prevent linker error gore when getIntrinsic is used but
specialized type isn't defined yet.
---
clang/lib/CodeGen/CGHLSLRuntime.h | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index 17f6516fb276dd..10fa5f860df50f 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -80,6 +80,7 @@ class CGHLSLRuntime {
//===----------------------------------------------------------------------===//
// Start of reserved area for HLSL intrinsic getters.
//===----------------------------------------------------------------------===//
+
template <unsigned BI> llvm::Intrinsic::ID getIntrinsic();
GENERATE_HLSL_INTRINSIC_TEMPLATE(builtin_hlsl_elementwise_all, all)
GENERATE_HLSL_INTRINSIC_FUNCTION(ThreadId, thread_id)
@@ -140,6 +141,11 @@ class CGHLSLRuntime {
llvm::SmallVector<Buffer> Buffers;
};
+// Default implementation
+template <unsigned BI> llvm::Intrinsic::ID CGHLSLRuntime::getIntrinsic() {
+ static_assert(false, "getIntrinsic is only allowed on specialized templates");
+}
+
} // namespace CodeGen
} // namespace clang
>From 3978c8c0d329db5e0dfb20013025fe148b74346c Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlotfi at microsoft.com>
Date: Thu, 4 Apr 2024 16:51:02 -0400
Subject: [PATCH 10/11] remove default template
---
clang/lib/CodeGen/CGHLSLRuntime.h | 5 -----
1 file changed, 5 deletions(-)
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index 10fa5f860df50f..54310b68aad8f6 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -141,11 +141,6 @@ class CGHLSLRuntime {
llvm::SmallVector<Buffer> Buffers;
};
-// Default implementation
-template <unsigned BI> llvm::Intrinsic::ID CGHLSLRuntime::getIntrinsic() {
- static_assert(false, "getIntrinsic is only allowed on specialized templates");
-}
-
} // namespace CodeGen
} // namespace clang
>From 4d0ebe8dc88ba8e6ee1bb9290bc407f22af9ff70 Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlotfi at microsoft.com>
Date: Thu, 4 Apr 2024 16:54:27 -0400
Subject: [PATCH 11/11] remove template
---
clang/lib/CodeGen/CGBuiltin.cpp | 5 ++---
clang/lib/CodeGen/CGHLSLRuntime.h | 36 +++++++++++--------------------
2 files changed, 15 insertions(+), 26 deletions(-)
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index f12312be83422f..df7502b8def531 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18177,9 +18177,8 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
Value *Op0 = EmitScalarExpr(E->getArg(0));
return Builder.CreateIntrinsic(
/*ReturnType=*/llvm::Type::getInt1Ty(getLLVMContext()),
- CGM.getHLSLRuntime()
- .getIntrinsic<Builtin::BI__builtin_hlsl_elementwise_all>(),
- ArrayRef<Value *>{Op0}, nullptr, "hlsl.all");
+ CGM.getHLSLRuntime().getAllIntrinsic(), ArrayRef<Value *>{Op0}, nullptr,
+ "hlsl.all");
}
case Builtin::BI__builtin_hlsl_elementwise_any: {
Value *Op0 = EmitScalarExpr(E->getArg(0));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index 54310b68aad8f6..2b8073aef973f8 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -30,29 +30,20 @@
#include <optional>
#include <vector>
-#define GENERATE_HLSL_INTRINSIC_BASE(IntrinsicPostfix) \
- llvm::Triple::ArchType Arch = getArch(); \
- switch (Arch) { \
- case llvm::Triple::dxil: \
- return llvm::Intrinsic::dx_##IntrinsicPostfix; \
- case llvm::Triple::spirv: \
- return llvm::Intrinsic::spv_##IntrinsicPostfix; \
- default: \
- llvm_unreachable("Intrinsic " #IntrinsicPostfix \
- " not supported by target architecture"); \
- }
-
-// A function generator macro for when there is no builtins or
-// when builtins are mapped to a set of intrinsics for different types.
+// A function generator macro for picking the right intrinsic
+// for the target backend
#define GENERATE_HLSL_INTRINSIC_FUNCTION(FunctionName, IntrinsicPostfix) \
llvm::Intrinsic::ID get##FunctionName##Intrinsic() { \
- GENERATE_HLSL_INTRINSIC_BASE(IntrinsicPostfix) \
- }
-
-// A template function generator macro for when we have builtins.
-#define GENERATE_HLSL_INTRINSIC_TEMPLATE(BuiltinName, IntrinsicPostfix) \
- template <> llvm::Intrinsic::ID getIntrinsic<Builtin::BI__##BuiltinName>() { \
- GENERATE_HLSL_INTRINSIC_BASE(IntrinsicPostfix) \
+ llvm::Triple::ArchType Arch = getArch(); \
+ switch (Arch) { \
+ case llvm::Triple::dxil: \
+ return llvm::Intrinsic::dx_##IntrinsicPostfix; \
+ case llvm::Triple::spirv: \
+ return llvm::Intrinsic::spv_##IntrinsicPostfix; \
+ default: \
+ llvm_unreachable("Intrinsic " #IntrinsicPostfix \
+ " not supported by target architecture"); \
+ } \
}
namespace llvm {
@@ -81,8 +72,7 @@ class CGHLSLRuntime {
// Start of reserved area for HLSL intrinsic getters.
//===----------------------------------------------------------------------===//
- template <unsigned BI> llvm::Intrinsic::ID getIntrinsic();
- GENERATE_HLSL_INTRINSIC_TEMPLATE(builtin_hlsl_elementwise_all, all)
+ GENERATE_HLSL_INTRINSIC_FUNCTION(All, all)
GENERATE_HLSL_INTRINSIC_FUNCTION(ThreadId, thread_id)
//===----------------------------------------------------------------------===//
More information about the cfe-commits
mailing list