[clang] [llvm] [HLSL] Implement `rsqrt` intrinsic (PR #84820)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Mar 11 13:05:02 PDT 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-clang-codegen
Author: Farzon Lotfi (farzonl)
<details>
<summary>Changes</summary>
This change implements #<!-- -->70074
- `hlsl_intrinsics.h` - add the `rsqrt` api
- `DXIL.td` add the llvm intrinsic to DXIL op lowering map.
- `Builtins.td` - add an hlsl builtin for rsqrt.
- `CGBuiltin.cpp` add the ir generation for the rsqrt intrinsic.
- `SemaChecking.cpp` - reuse the one arg float only checks.
- `IntrinsicsDirectX.td` -add an `rsqrt` intrinsic.
---
Full diff: https://github.com/llvm/llvm-project/pull/84820.diff
10 Files Affected:
- (modified) clang/include/clang/Basic/Builtins.td (+6)
- (modified) clang/lib/CodeGen/CGBuiltin.cpp (+8)
- (modified) clang/lib/Headers/hlsl/hlsl_intrinsics.h (+32)
- (modified) clang/lib/Sema/SemaChecking.cpp (+1)
- (added) clang/test/CodeGenHLSL/builtins/rsqrt.hlsl (+53)
- (added) clang/test/SemaHLSL/BuiltIns/dot-warning.ll (+49)
- (added) clang/test/SemaHLSL/BuiltIns/rsqrt-errors.hlsl (+27)
- (modified) llvm/include/llvm/IR/IntrinsicsDirectX.td (+1)
- (modified) llvm/lib/Target/DirectX/DXIL.td (+3)
- (added) llvm/test/CodeGen/DirectX/rsqrt.ll (+31)
``````````diff
diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index 9c703377ca8d3e..de0cfb4e46b8bd 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4590,6 +4590,12 @@ def HLSLRcp : LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
+def HLSLRSqrt : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_elementwise_rsqrt"];
+ let Attributes = [NoThrow, Const, CustomTypeChecking];
+ let Prototype = "void(...)";
+}
+
// Builtins for XRay.
def XRayCustomEvent : Builtin {
let Spellings = ["__xray_customevent"];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 20c35757939152..d2c83a5e405f42 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18077,6 +18077,14 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
/*ReturnType=*/Op0->getType(), Intrinsic::dx_rcp,
ArrayRef<Value *>{Op0}, nullptr, "dx.rcp");
}
+ case Builtin::BI__builtin_hlsl_elementwise_rsqrt: {
+ Value *Op0 = EmitScalarExpr(E->getArg(0));
+ if (!E->getArg(0)->getType()->hasFloatingRepresentation())
+ llvm_unreachable("rsqrt operand must have a float representation");
+ return Builder.CreateIntrinsic(
+ /*ReturnType=*/Op0->getType(), Intrinsic::dx_rsqrt,
+ ArrayRef<Value *>{Op0}, nullptr, "dx.rsqrt");
+ }
}
return nullptr;
}
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 45f8544392584e..f88aa4a5d5c644 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -1153,6 +1153,38 @@ double3 rcp(double3);
_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rcp)
double4 rcp(double4);
+//===----------------------------------------------------------------------===//
+// rsqrt builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn T rsqrt(T x)
+/// \brief RReturns the reciprocal of the square root of the specified value \a x.
+/// \param x The specified input value.
+///
+/// This function uses the following formula: 1 / sqrt(x).
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rsqrt)
+half rsqrt(half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rsqrt)
+half2 rsqrt(half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rsqrt)
+half3 rsqrt(half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rsqrt)
+half4 rsqrt(half4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rsqrt)
+float rsqrt(float);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rsqrt)
+float2 rsqrt(float2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rsqrt)
+float3 rsqrt(float3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rsqrt)
+float4 rsqrt(float4);
+
//===----------------------------------------------------------------------===//
// round builtins
//===----------------------------------------------------------------------===//
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
index a5f42b630c3fa2..0dafff47ab4040 100644
--- a/clang/lib/Sema/SemaChecking.cpp
+++ b/clang/lib/Sema/SemaChecking.cpp
@@ -5285,6 +5285,7 @@ bool Sema::CheckHLSLBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
return true;
break;
}
+ case Builtin::BI__builtin_hlsl_elementwise_rsqrt:
case Builtin::BI__builtin_hlsl_elementwise_rcp:
case Builtin::BI__builtin_hlsl_elementwise_frac: {
if (PrepareBuiltinElementwiseMathOneArgCall(TheCall))
diff --git a/clang/test/CodeGenHLSL/builtins/rsqrt.hlsl b/clang/test/CodeGenHLSL/builtins/rsqrt.hlsl
new file mode 100644
index 00000000000000..c87a8c404b08e1
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/rsqrt.hlsl
@@ -0,0 +1,53 @@
+// 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
+// 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,NO_HALF
+
+// NATIVE_HALF: define noundef half @
+// NATIVE_HALF: %dx.rsqrt = call half @llvm.dx.rsqrt.f16(
+// NATIVE_HALF: ret half %dx.rsqrt
+// NO_HALF: define noundef float @"?test_rsqrt_half@@YA$halff@$halff@@Z"(
+// NO_HALF: %dx.rsqrt = call float @llvm.dx.rsqrt.f32(
+// NO_HALF: ret float %dx.rsqrt
+half test_rsqrt_half(half p0) { return rsqrt(p0); }
+// NATIVE_HALF: define noundef <2 x half> @
+// NATIVE_HALF: %dx.rsqrt = call <2 x half> @llvm.dx.rsqrt.v2f16
+// NATIVE_HALF: ret <2 x half> %dx.rsqrt
+// NO_HALF: define noundef <2 x float> @
+// NO_HALF: %dx.rsqrt = call <2 x float> @llvm.dx.rsqrt.v2f32(
+// NO_HALF: ret <2 x float> %dx.rsqrt
+half2 test_rsqrt_half2(half2 p0) { return rsqrt(p0); }
+// NATIVE_HALF: define noundef <3 x half> @
+// NATIVE_HALF: %dx.rsqrt = call <3 x half> @llvm.dx.rsqrt.v3f16
+// NATIVE_HALF: ret <3 x half> %dx.rsqrt
+// NO_HALF: define noundef <3 x float> @
+// NO_HALF: %dx.rsqrt = call <3 x float> @llvm.dx.rsqrt.v3f32(
+// NO_HALF: ret <3 x float> %dx.rsqrt
+half3 test_rsqrt_half3(half3 p0) { return rsqrt(p0); }
+// NATIVE_HALF: define noundef <4 x half> @
+// NATIVE_HALF: %dx.rsqrt = call <4 x half> @llvm.dx.rsqrt.v4f16
+// NATIVE_HALF: ret <4 x half> %dx.rsqrt
+// NO_HALF: define noundef <4 x float> @
+// NO_HALF: %dx.rsqrt = call <4 x float> @llvm.dx.rsqrt.v4f32(
+// NO_HALF: ret <4 x float> %dx.rsqrt
+half4 test_rsqrt_half4(half4 p0) { return rsqrt(p0); }
+
+// CHECK: define noundef float @
+// CHECK: %dx.rsqrt = call float @llvm.dx.rsqrt.f32(
+// CHECK: ret float %dx.rsqrt
+float test_rsqrt_float(float p0) { return rsqrt(p0); }
+// CHECK: define noundef <2 x float> @
+// CHECK: %dx.rsqrt = call <2 x float> @llvm.dx.rsqrt.v2f32
+// CHECK: ret <2 x float> %dx.rsqrt
+float2 test_rsqrt_float2(float2 p0) { return rsqrt(p0); }
+// CHECK: define noundef <3 x float> @
+// CHECK: %dx.rsqrt = call <3 x float> @llvm.dx.rsqrt.v3f32
+// CHECK: ret <3 x float> %dx.rsqrt
+float3 test_rsqrt_float3(float3 p0) { return rsqrt(p0); }
+// CHECK: define noundef <4 x float> @
+// CHECK: %dx.rsqrt = call <4 x float> @llvm.dx.rsqrt.v4f32
+// CHECK: ret <4 x float> %dx.rsqrt
+float4 test_rsqrt_float4(float4 p0) { return rsqrt(p0); }
diff --git a/clang/test/SemaHLSL/BuiltIns/dot-warning.ll b/clang/test/SemaHLSL/BuiltIns/dot-warning.ll
new file mode 100644
index 00000000000000..5ecdde4d70e51f
--- /dev/null
+++ b/clang/test/SemaHLSL/BuiltIns/dot-warning.ll
@@ -0,0 +1,49 @@
+; ModuleID = 'D:\projects\llvm-project\clang\test\SemaHLSL\BuiltIns\dot-warning.hlsl'
+source_filename = "D:\\projects\\llvm-project\\clang\\test\\SemaHLSL\\BuiltIns\\dot-warning.hlsl"
+target datalayout = "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:32-f64:64-n8:16:32:64"
+target triple = "dxil-pc-shadermodel6.3-library"
+
+; Function Attrs: noinline nounwind optnone
+define noundef float @"?test_dot_builtin_vector_elem_size_reduction@@YAMT?$__vector at J$01 at __clang@@M at Z"(<2 x i64> noundef %p0, float noundef %p1) #0 {
+entry:
+ %p1.addr = alloca float, align 4
+ %p0.addr = alloca <2 x i64>, align 16
+ store float %p1, ptr %p1.addr, align 4
+ store <2 x i64> %p0, ptr %p0.addr, align 16
+ %0 = load <2 x i64>, ptr %p0.addr, align 16
+ %conv = sitofp <2 x i64> %0 to <2 x float>
+ %1 = load float, ptr %p1.addr, align 4
+ %splat.splatinsert = insertelement <2 x float> poison, float %1, i64 0
+ %splat.splat = shufflevector <2 x float> %splat.splatinsert, <2 x float> poison, <2 x i32> zeroinitializer
+ %dx.dot = call float @llvm.dx.dot.v2f32(<2 x float> %conv, <2 x float> %splat.splat)
+ ret float %dx.dot
+}
+
+; Function Attrs: nounwind willreturn memory(none)
+declare float @llvm.dx.dot.v2f32(<2 x float>, <2 x float>) #1
+
+; Function Attrs: noinline nounwind optnone
+define noundef float @"?test_dot_builtin_int_vector_elem_size_reduction@@YAMT?$__vector at H$01 at __clang@@M at Z"(<2 x i32> noundef %p0, float noundef %p1) #0 {
+entry:
+ %p1.addr = alloca float, align 4
+ %p0.addr = alloca <2 x i32>, align 8
+ store float %p1, ptr %p1.addr, align 4
+ store <2 x i32> %p0, ptr %p0.addr, align 8
+ %0 = load <2 x i32>, ptr %p0.addr, align 8
+ %conv = sitofp <2 x i32> %0 to <2 x float>
+ %1 = load float, ptr %p1.addr, align 4
+ %splat.splatinsert = insertelement <2 x float> poison, float %1, i64 0
+ %splat.splat = shufflevector <2 x float> %splat.splatinsert, <2 x float> poison, <2 x i32> zeroinitializer
+ %dx.dot = call float @llvm.dx.dot.v2f32(<2 x float> %conv, <2 x float> %splat.splat)
+ ret float %dx.dot
+}
+
+attributes #0 = { noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #1 = { nounwind willreturn memory(none) }
+
+!llvm.module.flags = !{!0, !1}
+!llvm.ident = !{!2}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 4, !"dx.disable_optimizations", i32 1}
+!2 = !{!"clang version 19.0.0git (https://github.com/farzonl/llvm-project.git f40562c7b4224e00da2ff2e13d175abfaac68532)"}
diff --git a/clang/test/SemaHLSL/BuiltIns/rsqrt-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/rsqrt-errors.hlsl
new file mode 100644
index 00000000000000..c74c502bd7a26f
--- /dev/null
+++ b/clang/test/SemaHLSL/BuiltIns/rsqrt-errors.hlsl
@@ -0,0 +1,27 @@
+
+// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm -disable-llvm-passes -verify -verify-ignore-unexpected
+
+float test_too_few_arg() {
+ return __builtin_hlsl_elementwise_rsqrt();
+ // expected-error at -1 {{too few arguments to function call, expected 1, have 0}}
+}
+
+float2 test_too_many_arg(float2 p0) {
+ return __builtin_hlsl_elementwise_rsqrt(p0, p0);
+ // expected-error at -1 {{too many arguments to function call, expected 1, have 2}}
+}
+
+float builtin_bool_to_float_type_promotion(bool p1) {
+ return __builtin_hlsl_elementwise_rsqrt(p1);
+ // expected-error at -1 {{1st argument must be a vector, integer or floating point type (was 'bool')}}
+}
+
+float builtin_rsqrt_int_to_float_promotion(int p1) {
+ return __builtin_hlsl_elementwise_rsqrt(p1);
+ // expected-error at -1 {{passing 'int' to parameter of incompatible type 'float'}}
+}
+
+float2 builtin_rsqrt_int2_to_float2_promotion(int2 p1) {
+ return __builtin_hlsl_elementwise_rsqrt(p1);
+ // expected-error at -1 {{passing 'int2' (aka 'vector<int, 2>') to parameter of incompatible type '__attribute__((__vector_size__(2 * sizeof(float)))) float' (vector of 2 'float' values)}}
+}
diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td
index 7229292e377a83..366dedda2b3f73 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -37,4 +37,5 @@ def int_dx_lerp :
def int_dx_imad : DefaultAttrsIntrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>]>;
def int_dx_umad : DefaultAttrsIntrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>]>;
def int_dx_rcp : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>]>;
+def int_dx_rsqrt : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>]>;
}
diff --git a/llvm/lib/Target/DirectX/DXIL.td b/llvm/lib/Target/DirectX/DXIL.td
index 9536a01e125bb3..942715f6ad80d3 100644
--- a/llvm/lib/Target/DirectX/DXIL.td
+++ b/llvm/lib/Target/DirectX/DXIL.td
@@ -224,6 +224,9 @@ def Exp2 : DXILOpMapping<21, unary, int_exp2,
def Frac : DXILOpMapping<22, unary, int_dx_frac,
"Returns a fraction from 0 to 1 that represents the "
"decimal part of the input.">;
+def RSqrt : DXILOpMapping<25, unary, int_dx_rsqrt,
+ "Returns the reciprocal of the square root of the specified value."
+ "rsqrt(x) = 1 / sqrt(x).">;
def Round : DXILOpMapping<26, unary, int_round,
"Returns the input rounded to the nearest integer"
"within a floating-point type.">;
diff --git a/llvm/test/CodeGen/DirectX/rsqrt.ll b/llvm/test/CodeGen/DirectX/rsqrt.ll
new file mode 100644
index 00000000000000..818b5985422173
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/rsqrt.ll
@@ -0,0 +1,31 @@
+; RUN: opt -S -dxil-op-lower < %s | FileCheck %s
+
+; Make sure dxil operation function calls for rsqrt are generated for float and half.
+; CHECK:call float @dx.op.unary.f32(i32 25, float %{{.*}})
+; CHECK:call half @dx.op.unary.f16(i32 25, half %{{.*}})
+
+target datalayout = "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:32-f64:64-n8:16:32:64"
+target triple = "dxil-pc-shadermodel6.7-library"
+
+; Function Attrs: noinline nounwind optnone
+define noundef float @rsqrt_float(float noundef %a) #0 {
+entry:
+ %a.addr = alloca float, align 4
+ store float %a, ptr %a.addr, align 4
+ %0 = load float, ptr %a.addr, align 4
+ %dx.rsqrt = call float @llvm.dx.rsqrt.f32(float %0)
+ ret float %dx.rsqrt
+}
+
+; Function Attrs: nocallback nofree nosync nounwind readnone speculatable willreturn
+declare float @llvm.dx.rsqrt.f32(float) #1
+
+; Function Attrs: noinline nounwind optnone
+define noundef half @rsqrt_half(half noundef %a) #0 {
+entry:
+ %a.addr = alloca half, align 2
+ store half %a, ptr %a.addr, align 2
+ %0 = load half, ptr %a.addr, align 2
+ %dx.rsqrt = call half @llvm.dx.rsqrt.f16(half %0)
+ ret half %dx.rsqrt
+}
``````````
</details>
https://github.com/llvm/llvm-project/pull/84820
More information about the llvm-commits
mailing list