[clang] [llvm] [HLSL] implement the rcp intrinsic (PR #83857)
Farzon Lotfi via cfe-commits
cfe-commits at lists.llvm.org
Mon Mar 4 07:48:39 PST 2024
https://github.com/farzonl created https://github.com/llvm/llvm-project/pull/83857
This PR implements the frontend for llvm#70100
This PR is part 1 of 2.
Part 2 depends on `DXILIntrinsicExpansion.cpp` in [dixl-lerp-intrinsic-lowering PR](https://github.com/llvm/llvm-project/compare/main...farzonl:llvm-project:dixl-lerp-intrinsic-lowering)
which will have an intrinsic to instruction expansion pass.
That pass is what we need to complete the DXIL lowering portion of this PR:
- `Builtins.td` - add an `rcp` builtin
- `CGBuiltin.cpp` - add the builtin to intrinsic lowering
- `hlsl_intrinsics.h` - add the `rcp` api
- `SemaChecking.cpp` - reuse frac's sema checks
- `IntrinsicsDirectX.td` - add the llvm intrinsic
>From 0060612b4fc1b7f5f5cc51a41a5fae7535ba2b02 Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzon at farzon.org>
Date: Sun, 3 Mar 2024 16:14:35 -0500
Subject: [PATCH] [HLSL] implement the rcp intrinsic This PR implements the
frontend for llvm#70100 This PR is part 1 of 2. Part 2 depends on
dixl-lerp-intrinsic-lowering PR which will have an intrinsic to instruction
expansion pass. THat pass is what we need to complete the DXIL lowering
portion of this PR:
llvm/llvm-project at main...farzonl:llvm-project:dixl-lerp-intrinsic-lowering
Builtins.td - add an rcp builtin
CGBuiltin.cpp add the builtin to intrinsic lowering
hlsl_intrinsics.h - add the rcp api
SemaChecking.cpp - reuse frac's semachecks
IntrinsicsDirectX.td add the llvm intrinsic
---
clang/include/clang/Basic/Builtins.td | 6 +++
clang/lib/CodeGen/CGBuiltin.cpp | 8 +++
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 41 +++++++++++++++
clang/lib/Sema/SemaChecking.cpp | 1 +
clang/test/CodeGenHLSL/builtins/rcp.hlsl | 53 ++++++++++++++++++++
clang/test/SemaHLSL/BuiltIns/rcp-errors.hlsl | 27 ++++++++++
llvm/include/llvm/IR/IntrinsicsDirectX.td | 2 +
7 files changed, 138 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/rcp.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/rcp-errors.hlsl
diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index 2c83dca248fb7d..a30953f4a94fa3 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4548,6 +4548,12 @@ def HLSLLerp : LangBuiltin<"HLSL_LANG"> {
let Prototype = "void(...)";
}
+def HLSLRcp : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_elementwise_rcp"];
+ let Attributes = [NoThrow, Const];
+ 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 e90014261217bc..aefc09767378a6 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18057,6 +18057,14 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
/*ReturnType*/ Op0->getType(), Intrinsic::dx_frac,
ArrayRef<Value *>{Op0}, nullptr, "dx.frac");
}
+ case Builtin::BI__builtin_hlsl_elementwise_rcp: {
+ Value *Op0 = EmitScalarExpr(E->getArg(0));
+ if (!E->getArg(0)->getType()->hasFloatingRepresentation())
+ llvm_unreachable("rcp operand must have a float representation");
+ return Builder.CreateIntrinsic(
+ /*ReturnType*/ Op0->getType(), Intrinsic::dx_rcp,
+ ArrayRef<Value *>{Op0}, nullptr, "dx.rcp");
+ }
}
return nullptr;
}
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 5180530363889f..6174461cfa8802 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -831,6 +831,47 @@ uint64_t3 reversebits(uint64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_bitreverse)
uint64_t4 reversebits(uint64_t4);
+//===----------------------------------------------------------------------===//
+// rcp builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn T rcp(T x)
+/// \brief Calculates a fast, approximate, per-component reciprocal ie 1 / \a x.
+/// \param x The specified input value.
+///
+/// The return value is the reciprocal of the \a x parameter.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rcp)
+half rcp(half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rcp)
+half2 rcp(half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rcp)
+half3 rcp(half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rcp)
+half4 rcp(half4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rcp)
+float rcp(float);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rcp)
+float2 rcp(float2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rcp)
+float3 rcp(float3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rcp)
+float4 rcp(float4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rcp)
+double rcp(double);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rcp)
+double2 rcp(double2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rcp)
+double3 rcp(double3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_rcp)
+double4 rcp(double4);
+
//===----------------------------------------------------------------------===//
// round builtins
//===----------------------------------------------------------------------===//
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
index 0d4d57db01c93a..237c838cb8c255 100644
--- a/clang/lib/Sema/SemaChecking.cpp
+++ b/clang/lib/Sema/SemaChecking.cpp
@@ -5280,6 +5280,7 @@ bool Sema::CheckHLSLBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
return true;
break;
}
+ case Builtin::BI__builtin_hlsl_elementwise_rcp:
case Builtin::BI__builtin_hlsl_elementwise_frac: {
if (PrepareBuiltinElementwiseMathOneArgCall(TheCall))
return true;
diff --git a/clang/test/CodeGenHLSL/builtins/rcp.hlsl b/clang/test/CodeGenHLSL/builtins/rcp.hlsl
new file mode 100644
index 00000000000000..9b8406e1f0b605
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/rcp.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.rcp = call half @llvm.dx.rcp.f16(
+// NATIVE_HALF: ret half %dx.rcp
+// NO_HALF: define noundef float @"?test_rcp_half@@YA$halff@$halff@@Z"(
+// NO_HALF: %dx.rcp = call float @llvm.dx.rcp.f32(
+// NO_HALF: ret float %dx.rcp
+half test_rcp_half(half p0) { return rcp(p0); }
+// NATIVE_HALF: define noundef <2 x half> @
+// NATIVE_HALF: %dx.rcp = call <2 x half> @llvm.dx.rcp.v2f16
+// NATIVE_HALF: ret <2 x half> %dx.rcp
+// NO_HALF: define noundef <2 x float> @
+// NO_HALF: %dx.rcp = call <2 x float> @llvm.dx.rcp.v2f32(
+// NO_HALF: ret <2 x float> %dx.rcp
+half2 test_rcp_half2(half2 p0) { return rcp(p0); }
+// NATIVE_HALF: define noundef <3 x half> @
+// NATIVE_HALF: %dx.rcp = call <3 x half> @llvm.dx.rcp.v3f16
+// NATIVE_HALF: ret <3 x half> %dx.rcp
+// NO_HALF: define noundef <3 x float> @
+// NO_HALF: %dx.rcp = call <3 x float> @llvm.dx.rcp.v3f32(
+// NO_HALF: ret <3 x float> %dx.rcp
+half3 test_rcp_half3(half3 p0) { return rcp(p0); }
+// NATIVE_HALF: define noundef <4 x half> @
+// NATIVE_HALF: %dx.rcp = call <4 x half> @llvm.dx.rcp.v4f16
+// NATIVE_HALF: ret <4 x half> %dx.rcp
+// NO_HALF: define noundef <4 x float> @
+// NO_HALF: %dx.rcp = call <4 x float> @llvm.dx.rcp.v4f32(
+// NO_HALF: ret <4 x float> %dx.rcp
+half4 test_rcp_half4(half4 p0) { return rcp(p0); }
+
+// CHECK: define noundef float @
+// CHECK: %dx.rcp = call float @llvm.dx.rcp.f32(
+// CHECK: ret float %dx.rcp
+float test_rcp_float(float p0) { return rcp(p0); }
+// CHECK: define noundef <2 x float> @
+// CHECK: %dx.rcp = call <2 x float> @llvm.dx.rcp.v2f32
+// CHECK: ret <2 x float> %dx.rcp
+float2 test_rcp_float2(float2 p0) { return rcp(p0); }
+// CHECK: define noundef <3 x float> @
+// CHECK: %dx.rcp = call <3 x float> @llvm.dx.rcp.v3f32
+// CHECK: ret <3 x float> %dx.rcp
+float3 test_rcp_float3(float3 p0) { return rcp(p0); }
+// CHECK: define noundef <4 x float> @
+// CHECK: %dx.rcp = call <4 x float> @llvm.dx.rcp.v4f32
+// CHECK: ret <4 x float> %dx.rcp
+float4 test_rcp_float4(float4 p0) { return rcp(p0); }
diff --git a/clang/test/SemaHLSL/BuiltIns/rcp-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/rcp-errors.hlsl
new file mode 100644
index 00000000000000..dc4501dbd6d15a
--- /dev/null
+++ b/clang/test/SemaHLSL/BuiltIns/rcp-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_rcp();
+ // 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_rcp(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_rcp(p1);
+ // expected-error at -1 {{1st argument must be a vector, integer or floating point type (was 'bool')}}
+}
+
+float builtin_rcp_int_to_float_promotion(int p1) {
+ return __builtin_hlsl_elementwise_rcp(p1);
+ // expected-error at -1 {{passing 'int' to parameter of incompatible type 'float'}}
+}
+
+float2 builtin_rcp_int2_to_float2_promotion(int2 p1) {
+ return __builtin_hlsl_elementwise_rcp(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 b44d1c6d3d2f06..86defee4a563a7 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -31,4 +31,6 @@ def int_dx_lerp :
Intrinsic<[LLVMScalarOrSameVectorWidth<0, LLVMVectorElementType<0>>],
[llvm_anyvector_ty, LLVMScalarOrSameVectorWidth<0, LLVMVectorElementType<0>>,LLVMScalarOrSameVectorWidth<0, LLVMVectorElementType<0>>],
[IntrNoMem, IntrWillReturn] >;
+
+def int_dx_rcp : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>]>;
}
More information about the cfe-commits
mailing list