[HLSL][DXIL] HLSL's `round` should follow `roundeven` behavior #87078

farzonl · 2024-03-29T15:24:34Z

fixes #86999

fixes llvm#86999

llvmbot · 2024-03-29T15:25:03Z

@llvm/pr-subscribers-hlsl
@llvm/pr-subscribers-backend-directx
@llvm/pr-subscribers-backend-x86

@llvm/pr-subscribers-clang

Author: Farzon Lotfi (farzonl)

Changes

fixes #86999

Full diff: https://github.com/llvm/llvm-project/pull/87078.diff

7 Files Affected:

(modified) clang/lib/Headers/hlsl/hlsl_intrinsics.h (+8-8)
(modified) clang/lib/Sema/SemaChecking.cpp (+1)
(modified) clang/test/CodeGenHLSL/builtins/round.hlsl (+24-24)
(modified) clang/test/SemaHLSL/BuiltIns/half-float-only-errors.hlsl (+1)
(modified) llvm/lib/Target/DirectX/DXIL.td (+1-1)
(modified) llvm/test/CodeGen/DirectX/round.ll (+13-22)
(modified) llvm/test/CodeGen/DirectX/round_error.ll (+2-2)

diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index a34e72402c0e64..9fb6204f90c9a8 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -1248,25 +1248,25 @@ float4 rsqrt(float4);
 /// rounded to the nearest even value.
 
 _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_round)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_roundeven)
 half round(half);
 _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_round)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_roundeven)
 half2 round(half2);
 _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_round)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_roundeven)
 half3 round(half3);
 _HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_round)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_roundeven)
 half4 round(half4);
 
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_round)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_roundeven)
 float round(float);
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_round)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_roundeven)
 float2 round(float2);
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_round)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_roundeven)
 float3 round(float3);
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_round)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_roundeven)
 float4 round(float4);
 
 //===----------------------------------------------------------------------===//
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
index 2e4e18a3ebf759..2684535d8e53d1 100644
--- a/clang/lib/Sema/SemaChecking.cpp
+++ b/clang/lib/Sema/SemaChecking.cpp
@@ -5650,6 +5650,7 @@ bool Sema::CheckHLSLBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   case Builtin::BI__builtin_elementwise_log2:
   case Builtin::BI__builtin_elementwise_log10:
   case Builtin::BI__builtin_elementwise_pow:
+  case Builtin::BI__builtin_elementwise_roundeven:
   case Builtin::BI__builtin_elementwise_sin:
   case Builtin::BI__builtin_elementwise_sqrt:
   case Builtin::BI__builtin_elementwise_trunc: {
diff --git a/clang/test/CodeGenHLSL/builtins/round.hlsl b/clang/test/CodeGenHLSL/builtins/round.hlsl
index b9f35bd3712d18..33d761dbdfbeae 100644
--- a/clang/test/CodeGenHLSL/builtins/round.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/round.hlsl
@@ -7,47 +7,47 @@
 // RUN:   -o - | FileCheck %s --check-prefixes=CHECK,NO_HALF
 
 // NATIVE_HALF: define noundef half @
-// NATIVE_HALF: %elt.round = call half @llvm.round.f16(
-// NATIVE_HALF: ret half %elt.round
+// NATIVE_HALF: %elt.roundeven = call half @llvm.roundeven.f16(
+// NATIVE_HALF: ret half %elt.roundeven
 // NO_HALF: define noundef float @"?test_round_half@@YA$halff@$halff@@Z"(
-// NO_HALF: %elt.round = call float @llvm.round.f32(
-// NO_HALF: ret float %elt.round
+// NO_HALF: %elt.roundeven = call float @llvm.roundeven.f32(
+// NO_HALF: ret float %elt.roundeven
 half test_round_half(half p0) { return round(p0); }
 // NATIVE_HALF: define noundef <2 x half> @
-// NATIVE_HALF: %elt.round = call <2 x half> @llvm.round.v2f16
-// NATIVE_HALF: ret <2 x half> %elt.round
+// NATIVE_HALF: %elt.roundeven = call <2 x half> @llvm.roundeven.v2f16
+// NATIVE_HALF: ret <2 x half> %elt.roundeven
 // NO_HALF: define noundef <2 x float> @
-// NO_HALF: %elt.round = call <2 x float> @llvm.round.v2f32(
-// NO_HALF: ret <2 x float> %elt.round
+// NO_HALF: %elt.roundeven = call <2 x float> @llvm.roundeven.v2f32(
+// NO_HALF: ret <2 x float> %elt.roundeven
 half2 test_round_half2(half2 p0) { return round(p0); }
 // NATIVE_HALF: define noundef <3 x half> @
-// NATIVE_HALF: %elt.round = call <3 x half> @llvm.round.v3f16
-// NATIVE_HALF: ret <3 x half> %elt.round
+// NATIVE_HALF: %elt.roundeven = call <3 x half> @llvm.roundeven.v3f16
+// NATIVE_HALF: ret <3 x half> %elt.roundeven
 // NO_HALF: define noundef <3 x float> @
-// NO_HALF: %elt.round = call <3 x float> @llvm.round.v3f32(
-// NO_HALF: ret <3 x float> %elt.round
+// NO_HALF: %elt.roundeven = call <3 x float> @llvm.roundeven.v3f32(
+// NO_HALF: ret <3 x float> %elt.roundeven
 half3 test_round_half3(half3 p0) { return round(p0); }
 // NATIVE_HALF: define noundef <4 x half> @
-// NATIVE_HALF: %elt.round = call <4 x half> @llvm.round.v4f16
-// NATIVE_HALF: ret <4 x half> %elt.round
+// NATIVE_HALF: %elt.roundeven = call <4 x half> @llvm.roundeven.v4f16
+// NATIVE_HALF: ret <4 x half> %elt.roundeven
 // NO_HALF: define noundef <4 x float> @
-// NO_HALF: %elt.round = call <4 x float> @llvm.round.v4f32(
-// NO_HALF: ret <4 x float> %elt.round
+// NO_HALF: %elt.roundeven = call <4 x float> @llvm.roundeven.v4f32(
+// NO_HALF: ret <4 x float> %elt.roundeven
 half4 test_round_half4(half4 p0) { return round(p0); }
 
 // CHECK: define noundef float @
-// CHECK: %elt.round = call float @llvm.round.f32(
-// CHECK: ret float %elt.round
+// CHECK: %elt.roundeven = call float @llvm.roundeven.f32(
+// CHECK: ret float %elt.roundeven
 float test_round_float(float p0) { return round(p0); }
 // CHECK: define noundef <2 x float> @
-// CHECK: %elt.round = call <2 x float> @llvm.round.v2f32
-// CHECK: ret <2 x float> %elt.round
+// CHECK: %elt.roundeven = call <2 x float> @llvm.roundeven.v2f32
+// CHECK: ret <2 x float> %elt.roundeven
 float2 test_round_float2(float2 p0) { return round(p0); }
 // CHECK: define noundef <3 x float> @
-// CHECK: %elt.round = call <3 x float> @llvm.round.v3f32
-// CHECK: ret <3 x float> %elt.round
+// CHECK: %elt.roundeven = call <3 x float> @llvm.roundeven.v3f32
+// CHECK: ret <3 x float> %elt.roundeven
 float3 test_round_float3(float3 p0) { return round(p0); }
 // CHECK: define noundef <4 x float> @
-// CHECK: %elt.round = call <4 x float> @llvm.round.v4f32
-// CHECK: ret <4 x float> %elt.round
+// CHECK: %elt.roundeven = call <4 x float> @llvm.roundeven.v4f32
+// CHECK: ret <4 x float> %elt.roundeven
 float4 test_round_float4(float4 p0) { return round(p0); }
diff --git a/clang/test/SemaHLSL/BuiltIns/half-float-only-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/half-float-only-errors.hlsl
index c56986b7f86225..98c02c38675f4e 100644
--- a/clang/test/SemaHLSL/BuiltIns/half-float-only-errors.hlsl
+++ b/clang/test/SemaHLSL/BuiltIns/half-float-only-errors.hlsl
@@ -8,6 +8,7 @@
 // RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm -disable-llvm-passes -verify -DTEST_FUNC=__builtin_elementwise_log10
 // RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm -disable-llvm-passes -verify -DTEST_FUNC=__builtin_elementwise_sin
 // RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm -disable-llvm-passes -verify -DTEST_FUNC=__builtin_elementwise_sqrt
+// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm -disable-llvm-passes -verify -DTEST_FUNC=__builtin_elementwise_roundeven
 // RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm -disable-llvm-passes -verify -DTEST_FUNC=__builtin_elementwise_trunc
 
 double2 test_double_builtin(double2 p0) {
diff --git a/llvm/lib/Target/DirectX/DXIL.td b/llvm/lib/Target/DirectX/DXIL.td
index c5d7ee76275f86..a131bcc7dafe9a 100644
--- a/llvm/lib/Target/DirectX/DXIL.td
+++ b/llvm/lib/Target/DirectX/DXIL.td
@@ -285,7 +285,7 @@ def RSqrt : DXILOpMapping<25, unary, int_dx_rsqrt,
                          "Returns the reciprocal of the square root of the specified value."
                          "rsqrt(x) = 1 / sqrt(x).",
                          [llvm_halforfloat_ty, LLVMMatchType<0>]>;
-def Round : DXILOpMapping<26, unary, int_round,
+def Round : DXILOpMapping<26, unary, int_roundeven,
                          "Returns the input rounded to the nearest integer"
                          "within a floating-point type.",
                          [llvm_halforfloat_ty, LLVMMatchType<0>]>;
diff --git a/llvm/test/CodeGen/DirectX/round.ll b/llvm/test/CodeGen/DirectX/round.ll
index 5d53a794b763a6..e0a3772ebca8fa 100644
--- a/llvm/test/CodeGen/DirectX/round.ll
+++ b/llvm/test/CodeGen/DirectX/round.ll
@@ -1,31 +1,22 @@
 ; 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 @round_float(float noundef %a) #0 {
+; CHECK-LABEL: round_half
+define noundef half @round_half(half noundef %a) {
 entry:
-  %a.addr = alloca float, align 4
-  store float %a, ptr %a.addr, align 4
-  %0 = load float, ptr %a.addr, align 4
-  %elt.round = call float @llvm.round.f32(float %0)
-  ret float %elt.round
+; CHECK: call half @dx.op.unary.f16(i32 26, half %{{.*}})
+  %elt.roundeven = call half @llvm.roundeven.f16(half %a)
+  ret half %elt.roundeven
 }
 
-; Function Attrs: nocallback nofree nosync nounwind readnone speculatable willreturn
-declare float @llvm.round.f32(float) #1
-
-; Function Attrs: noinline nounwind optnone
-define noundef half @round_half(half noundef %a) #0 {
+; CHECK-LABEL: round_float
+define noundef float @round_float(float noundef %a) {
 entry:
-  %a.addr = alloca half, align 2
-  store half %a, ptr %a.addr, align 2
-  %0 = load half, ptr %a.addr, align 2
-  %elt.round = call half @llvm.round.f16(half %0)
-  ret half %elt.round
+; CHECK: call float @dx.op.unary.f32(i32 26, float %{{.*}})
+  %elt.roundeven = call float @llvm.roundeven.f32(float %a)
+  ret float %elt.roundeven
 }
+
+declare half @llvm.roundeven.f16(half)
+declare float @llvm.roundeven.f32(float)
diff --git a/llvm/test/CodeGen/DirectX/round_error.ll b/llvm/test/CodeGen/DirectX/round_error.ll
index 3bd87b2bbf0200..2d27fbb5ee20de 100644
--- a/llvm/test/CodeGen/DirectX/round_error.ll
+++ b/llvm/test/CodeGen/DirectX/round_error.ll
@@ -8,6 +8,6 @@ entry:
   %a.addr = alloca double, align 8
   store double %a, ptr %a.addr, align 8
   %0 = load double, ptr %a.addr, align 8
-  %elt.round = call double @llvm.round.f64(double %0)
-  ret double %elt.round
+  %elt.roundeven = call double @llvm.roundeven.f64(double %0)
+  ret double %elt.roundeven
 }

[HLSL][DXIL] HLSL's round should follow roundeven behavior

bbb30f7

fixes llvm#86999

farzonl requested review from bogner, llvm-beanz, bharadwajy and python3kgae March 29, 2024 15:24

farzonl self-assigned this Mar 29, 2024

llvmbot added clang Clang issues not falling into any other category backend:X86 clang:frontend Language frontend issues, e.g. anything involving "Sema" clang:headers Headers provided by Clang, e.g. for intrinsics backend:DirectX HLSL HLSL Language Support labels Mar 29, 2024

farzonl mentioned this pull request Mar 29, 2024

[SPIRV][HLSL] Add HLSL intrinsic tests #86844

Merged

python3kgae approved these changes Mar 29, 2024

View reviewed changes

bogner approved these changes Mar 29, 2024

View reviewed changes

bharadwajy approved these changes Mar 29, 2024

View reviewed changes

farzonl merged commit e74332a into llvm:main Mar 29, 2024
11 checks passed

farzonl deleted the bugfix-hlsl-round-intrinsic-mapping branch March 29, 2024 20:32

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[HLSL][DXIL] HLSL's `round` should follow `roundeven` behavior #87078

[HLSL][DXIL] HLSL's `round` should follow `roundeven` behavior #87078

farzonl commented Mar 29, 2024

llvmbot commented Mar 29, 2024 •

edited

Loading

[HLSL][DXIL] HLSL's round should follow roundeven behavior #87078

[HLSL][DXIL] HLSL's round should follow roundeven behavior #87078

Conversation

farzonl commented Mar 29, 2024

llvmbot commented Mar 29, 2024 • edited Loading

[HLSL][DXIL] HLSL's `round` should follow `roundeven` behavior #87078

[HLSL][DXIL] HLSL's `round` should follow `roundeven` behavior #87078

llvmbot commented Mar 29, 2024 •

edited

Loading