[clang] [llvm] [HLSL][DXIL] Implementation of round intrinsic (PR #83520)
Farzon Lotfi via cfe-commits
cfe-commits at lists.llvm.org
Thu Feb 29 19:30:46 PST 2024
https://github.com/farzonl created https://github.com/llvm/llvm-project/pull/83520
- Builtins.td - add the round builtin
- CGBuiltin.cpp - add the builtin to DirectX intrinsic mapping hlsl_intrinsics.h - add the round api
- SemaChecking.cpp - add type checks for builtin
- IntrinsicsDirectX.td - add the round intrinsic
- DXIL.td add the llvm intrinsic to DXIL lowering mapping
- This change implements: #70077
>From f9ce067ed26268cc2015109e1f9389d0321590e8 Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzon at farzon.org>
Date: Thu, 29 Feb 2024 21:48:47 -0500
Subject: [PATCH] [HLSL][DXIL] Implementation of round intrinsic Builtins.td -
add the round builtin CGBuiltin.cpp - add the builtin to DirectX intrinsic
mapping hlsl_intrinsics.h - add the round api SemaChecking.cpp - add type
checks for builtin IntrinsicsDirectX.td - add the round intrinsic DXIL.td add
the llvm intrinsic to DXIL lowering mapping This change implements: #70077
---
clang/include/clang/Basic/Builtins.td | 6 +++
clang/lib/CodeGen/CGBuiltin.cpp | 8 +++
clang/lib/Headers/hlsl/hlsl_intrinsics.h | 34 ++++++++++++
clang/lib/Sema/SemaChecking.cpp | 7 +++
clang/test/CodeGenHLSL/builtins/round.hlsl | 53 +++++++++++++++++++
.../test/SemaHLSL/BuiltIns/round-errors.hlsl | 27 ++++++++++
llvm/include/llvm/IR/IntrinsicsDirectX.td | 1 +
llvm/lib/Target/DirectX/DXIL.td | 3 ++
llvm/test/CodeGen/DirectX/round.ll | 43 +++++++++++++++
9 files changed, 182 insertions(+)
create mode 100644 clang/test/CodeGenHLSL/builtins/round.hlsl
create mode 100644 clang/test/SemaHLSL/BuiltIns/round-errors.hlsl
create mode 100644 llvm/test/CodeGen/DirectX/round.ll
diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index 2c83dca248fb7d..cb59a7b26f729f 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 HLSLRound : LangBuiltin<"HLSL_LANG"> {
+ let Spellings = ["__builtin_hlsl_elementwise_round"];
+ 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 98684448f4ff5c..a27390c4e85fb7 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18052,6 +18052,14 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
/*ReturnType*/ Op0->getType(), Intrinsic::dx_frac,
ArrayRef<Value *>{Op0}, nullptr, "dx.frac");
}
+ case Builtin::BI__builtin_hlsl_elementwise_round: {
+ Value *Op0 = EmitScalarExpr(E->getArg(0));
+ if (!E->getArg(0)->getType()->hasFloatingRepresentation())
+ llvm_unreachable("round operand must have a float representation");
+ return Builder.CreateIntrinsic(
+ /*ReturnType*/ Op0->getType(), Intrinsic::dx_round,
+ ArrayRef<Value *>{Op0}, nullptr, "dx.round");
+ }
}
return nullptr;
}
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 0aa8651ba80dc4..4f52b001440584 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -831,6 +831,40 @@ uint64_t3 reversebits(uint64_t3);
_HLSL_BUILTIN_ALIAS(__builtin_elementwise_bitreverse)
uint64_t4 reversebits(uint64_t4);
+//===----------------------------------------------------------------------===//
+// frac builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn T round(T x)
+/// \brief Rounds the specified value to the nearest integer. Halfway cases are
+/// rounded to the nearest even. \a x parameter. \param x The specified input
+/// value.
+///
+/// The return value is the \a x parameter, rounded to the nearest integer
+/// within a floating-point type.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_round)
+half round(half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_round)
+half2 round(half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_round)
+half3 round(half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_round)
+half4 round(half4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_round)
+float round(float);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_round)
+float2 round(float2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_round)
+float3 round(float3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_round)
+float4 round(float4);
+
//===----------------------------------------------------------------------===//
// sin builtins
//===----------------------------------------------------------------------===//
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
index 690bdaa63d058b..56e4264f85ccdd 100644
--- a/clang/lib/Sema/SemaChecking.cpp
+++ b/clang/lib/Sema/SemaChecking.cpp
@@ -5294,6 +5294,13 @@ bool Sema::CheckHLSLBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
return true;
break;
}
+ case Builtin::BI__builtin_hlsl_elementwise_round: {
+ if (PrepareBuiltinElementwiseMathOneArgCall(TheCall))
+ return true;
+ if (CheckAllArgsHaveFloatRepresentation(this, TheCall))
+ return true;
+ break;
+ }
}
return false;
}
diff --git a/clang/test/CodeGenHLSL/builtins/round.hlsl b/clang/test/CodeGenHLSL/builtins/round.hlsl
new file mode 100644
index 00000000000000..6d582791447af2
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/round.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.round = call half @llvm.dx.round.f16(
+// NATIVE_HALF: ret half %dx.round
+// NO_HALF: define noundef float @"?test_round_half@@YA$halff@$halff@@Z"(
+// NO_HALF: %dx.round = call float @llvm.dx.round.f32(
+// NO_HALF: ret float %dx.round
+half test_round_half(half p0) { return round(p0); }
+// NATIVE_HALF: define noundef <2 x half> @
+// NATIVE_HALF: %dx.round = call <2 x half> @llvm.dx.round.v2f16
+// NATIVE_HALF: ret <2 x half> %dx.round
+// NO_HALF: define noundef <2 x float> @
+// NO_HALF: %dx.round = call <2 x float> @llvm.dx.round.v2f32(
+// NO_HALF: ret <2 x float> %dx.round
+half2 test_round_half2(half2 p0) { return round(p0); }
+// NATIVE_HALF: define noundef <3 x half> @
+// NATIVE_HALF: %dx.round = call <3 x half> @llvm.dx.round.v3f16
+// NATIVE_HALF: ret <3 x half> %dx.round
+// NO_HALF: define noundef <3 x float> @
+// NO_HALF: %dx.round = call <3 x float> @llvm.dx.round.v3f32(
+// NO_HALF: ret <3 x float> %dx.round
+half3 test_round_half3(half3 p0) { return round(p0); }
+// NATIVE_HALF: define noundef <4 x half> @
+// NATIVE_HALF: %dx.round = call <4 x half> @llvm.dx.round.v4f16
+// NATIVE_HALF: ret <4 x half> %dx.round
+// NO_HALF: define noundef <4 x float> @
+// NO_HALF: %dx.round = call <4 x float> @llvm.dx.round.v4f32(
+// NO_HALF: ret <4 x float> %dx.round
+half4 test_round_half4(half4 p0) { return round(p0); }
+
+// CHECK: define noundef float @
+// CHECK: %dx.round = call float @llvm.dx.round.f32(
+// CHECK: ret float %dx.round
+float test_round_float(float p0) { return round(p0); }
+// CHECK: define noundef <2 x float> @
+// CHECK: %dx.round = call <2 x float> @llvm.dx.round.v2f32
+// CHECK: ret <2 x float> %dx.round
+float2 test_round_float2(float2 p0) { return round(p0); }
+// CHECK: define noundef <3 x float> @
+// CHECK: %dx.round = call <3 x float> @llvm.dx.round.v3f32
+// CHECK: ret <3 x float> %dx.round
+float3 test_round_float3(float3 p0) { return round(p0); }
+// CHECK: define noundef <4 x float> @
+// CHECK: %dx.round = call <4 x float> @llvm.dx.round.v4f32
+// CHECK: ret <4 x float> %dx.round
+float4 test_round_float4(float4 p0) { return round(p0); }
diff --git a/clang/test/SemaHLSL/BuiltIns/round-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/round-errors.hlsl
new file mode 100644
index 00000000000000..9af07df5e40499
--- /dev/null
+++ b/clang/test/SemaHLSL/BuiltIns/round-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_round();
+ // 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_round(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_round(p1);
+ // expected-error at -1 {{1st argument must be a vector, integer or floating point type (was 'bool')}}
+}
+
+float builtin_round_int_to_float_promotion(int p1) {
+ return __builtin_hlsl_elementwise_round(p1);
+ // expected-error at -1 {{passing 'int' to parameter of incompatible type 'float'}}
+}
+
+float2 builtin_round_int2_to_float2_promotion(int2 p1) {
+ return __builtin_hlsl_elementwise_round(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..d389193fe97e37 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -26,6 +26,7 @@ def int_dx_dot :
[IntrNoMem, IntrWillReturn, Commutative] >;
def int_dx_frac : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>]>;
+def int_dx_round : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>]>;
def int_dx_lerp :
Intrinsic<[LLVMScalarOrSameVectorWidth<0, LLVMVectorElementType<0>>],
diff --git a/llvm/lib/Target/DirectX/DXIL.td b/llvm/lib/Target/DirectX/DXIL.td
index 67ef7986622092..bab2c252c99d68 100644
--- a/llvm/lib/Target/DirectX/DXIL.td
+++ b/llvm/lib/Target/DirectX/DXIL.td
@@ -220,6 +220,9 @@ def Sin : DXILOpMapping<13, unary, int_sin,
"Returns sine(theta) for theta in radians.">;
def UMax : DXILOpMapping<39, binary, int_umax,
"Unsigned integer maximum. UMax(a,b) = a > b ? a : b">;
+def Round : DXILOpMapping<26, unary, int_dx_round,
+ "Returns the input rounded to the nearest integer"
+ "within a floating-point type.">;
def ThreadId : DXILOpMapping<93, threadId, int_dx_thread_id,
"Reads the thread ID">;
def GroupId : DXILOpMapping<94, groupId, int_dx_group_id,
diff --git a/llvm/test/CodeGen/DirectX/round.ll b/llvm/test/CodeGen/DirectX/round.ll
new file mode 100644
index 00000000000000..6b4a74f2513589
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/round.ll
@@ -0,0 +1,43 @@
+; RUN: opt -S -dxil-op-lower < %s | FileCheck %s
+
+; Make sure dxil operation function calls for round are generated for float and half.
+; CHECK:call float @dx.op.unary.f32(i32 26, float %{{.*}})
+; CHECK:call half @dx.op.unary.f16(i32 26, 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 @_Z3foof(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.round = call float @llvm.dx.round.f32(float %0)
+ ret float %dx.round
+}
+
+; Function Attrs: nocallback nofree nosync nounwind readnone speculatable willreturn
+declare float @llvm.dx.round.f32(float) #1
+
+; Function Attrs: noinline nounwind optnone
+define noundef half @_Z3barDh(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.round = call half @llvm.dx.round.f16(half %0)
+ ret half %dx.round
+}
+
+; Function Attrs: nocallback nofree nosync nounwind readnone speculatable willreturn
+declare half @llvm.dx.round.f16(half) #1
+
+attributes #0 = { noinline nounwind optnone "frame-pointer"="none" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #1 = { nocallback nofree nosync nounwind readnone speculatable willreturn }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 15.0.0 (https://github.com/llvm/llvm-project.git 73417c517644db5c419c85c0b3cb6750172fcab5)"}
\ No newline at end of file
More information about the cfe-commits
mailing list