1; RUN: opt -S -dxil-intrinsic-expansion < %s | FileCheck %s --check-prefixes=CHECK,EXPCHECK 2; RUN: opt -S -dxil-intrinsic-expansion -dxil-op-lower -mtriple=dxil-pc-shadermodel6.3-library < %s | FileCheck %s --check-prefixes=CHECK,DOPCHECK 3 4; Make sure dxil operation function calls for normalize are generated for half/float. 5 6declare half @llvm.dx.normalize.f16(half) 7declare <2 x half> @llvm.dx.normalize.v2f16(<2 x half>) 8declare <3 x half> @llvm.dx.normalize.v3f16(<3 x half>) 9declare <4 x half> @llvm.dx.normalize.v4f16(<4 x half>) 10 11declare float @llvm.dx.normalize.f32(float) 12declare <2 x float> @llvm.dx.normalize.v2f32(<2 x float>) 13declare <3 x float> @llvm.dx.normalize.v3f32(<3 x float>) 14declare <4 x float> @llvm.dx.normalize.v4f32(<4 x float>) 15 16define noundef half @test_normalize_half(half noundef %p0) { 17entry: 18 ; CHECK: fdiv half %p0, %p0 19 %hlsl.normalize = call half @llvm.dx.normalize.f16(half %p0) 20 ret half %hlsl.normalize 21} 22 23define noundef <2 x half> @test_normalize_half2(<2 x half> noundef %p0) { 24entry: 25 ; EXPCHECK: [[doth2:%.*]] = call half @llvm.dx.dot2.v2f16(<2 x half> %{{.*}}, <2 x half> %{{.*}}) 26 ; DOPCHECK: [[doth2:%.*]] = call half @dx.op.dot2.f16(i32 54, half %{{.*}}, half %{{.*}}, half %{{.*}}, half %{{.*}}) 27 ; EXPCHECK: [[rsqrt:%.*]] = call half @llvm.dx.rsqrt.f16(half [[doth2]]) 28 ; DOPCHECK: [[rsqrt:%.*]] = call half @dx.op.unary.f16(i32 25, half [[doth2]]) 29 ; CHECK: [[splatinserth2:%.*]] = insertelement <2 x half> poison, half [[rsqrt]], i64 0 30 ; CHECK: [[splat:%.*]] = shufflevector <2 x half> [[splatinserth2]], <2 x half> poison, <2 x i32> zeroinitializer 31 ; CHECK: fmul <2 x half> %p0, [[splat]] 32 33 %hlsl.normalize = call <2 x half> @llvm.dx.normalize.v2f16(<2 x half> %p0) 34 ret <2 x half> %hlsl.normalize 35} 36 37define noundef <3 x half> @test_normalize_half3(<3 x half> noundef %p0) { 38entry: 39 ; EXPCHECK: [[doth3:%.*]] = call half @llvm.dx.dot3.v3f16(<3 x half> %{{.*}}, <3 x half> %{{.*}}) 40 ; DOPCHECK: [[doth3:%.*]] = call half @dx.op.dot3.f16(i32 55, half %{{.*}}, half %{{.*}}, half %{{.*}}, half %{{.*}}) 41 ; EXPCHECK: [[rsqrt:%.*]] = call half @llvm.dx.rsqrt.f16(half [[doth3]]) 42 ; DOPCHECK: [[rsqrt:%.*]] = call half @dx.op.unary.f16(i32 25, half [[doth3]]) 43 ; CHECK: [[splatinserth3:%.*]] = insertelement <3 x half> poison, half [[rsqrt]], i64 0 44 ; CHECK: [[splat:%.*]] shufflevector <3 x half> [[splatinserth3]], <3 x half> poison, <3 x i32> zeroinitializer 45 ; CHECK: fmul <3 x half> %p0, %.splat 46 47 %hlsl.normalize = call <3 x half> @llvm.dx.normalize.v3f16(<3 x half> %p0) 48 ret <3 x half> %hlsl.normalize 49} 50 51define noundef <4 x half> @test_normalize_half4(<4 x half> noundef %p0) { 52entry: 53 ; EXPCHECK: [[doth4:%.*]] = call half @llvm.dx.dot4.v4f16(<4 x half> %{{.*}}, <4 x half> %{{.*}}) 54 ; DOPCHECK: [[doth4:%.*]] = call half @dx.op.dot4.f16(i32 56, half %{{.*}}, half %{{.*}}, half %{{.*}}, half %{{.*}}) 55 ; EXPCHECK: [[rsqrt:%.*]] = call half @llvm.dx.rsqrt.f16(half [[doth4]]) 56 ; DOPCHECK: [[rsqrt:%.*]] = call half @dx.op.unary.f16(i32 25, half [[doth4]]) 57 ; CHECK: [[splatinserth4:%.*]] = insertelement <4 x half> poison, half [[rsqrt]], i64 0 58 ; CHECK: [[splat:%.*]] shufflevector <4 x half> [[splatinserth4]], <4 x half> poison, <4 x i32> zeroinitializer 59 ; CHECK: fmul <4 x half> %p0, %.splat 60 61 %hlsl.normalize = call <4 x half> @llvm.dx.normalize.v4f16(<4 x half> %p0) 62 ret <4 x half> %hlsl.normalize 63} 64 65define noundef float @test_normalize_float(float noundef %p0) { 66entry: 67 ; CHECK: fdiv float %p0, %p0 68 %hlsl.normalize = call float @llvm.dx.normalize.f32(float %p0) 69 ret float %hlsl.normalize 70} 71 72define noundef <2 x float> @test_normalize_float2(<2 x float> noundef %p0) { 73entry: 74 ; EXPCHECK: [[dotf2:%.*]] = call float @llvm.dx.dot2.v2f32(<2 x float> %{{.*}}, <2 x float> %{{.*}}) 75 ; DOPCHECK: [[dotf2:%.*]] = call float @dx.op.dot2.f32(i32 54, float %{{.*}}, float %{{.*}}, float %{{.*}}, float %{{.*}}) 76 ; EXPCHECK: [[rsqrt:%.*]] = call float @llvm.dx.rsqrt.f32(float [[dotf2]]) 77 ; DOPCHECK: [[rsqrt:%.*]] = call float @dx.op.unary.f32(i32 25, float [[dotf2]]) 78 ; CHECK: [[splatinsertf2:%.*]] = insertelement <2 x float> poison, float [[rsqrt]], i64 0 79 ; CHECK: [[splat:%.*]] shufflevector <2 x float> [[splatinsertf2]], <2 x float> poison, <2 x i32> zeroinitializer 80 ; CHECK: fmul <2 x float> %p0, %.splat 81 82 %hlsl.normalize = call <2 x float> @llvm.dx.normalize.v2f32(<2 x float> %p0) 83 ret <2 x float> %hlsl.normalize 84} 85 86define noundef <3 x float> @test_normalize_float3(<3 x float> noundef %p0) { 87entry: 88 ; EXPCHECK: [[dotf3:%.*]] = call float @llvm.dx.dot3.v3f32(<3 x float> %{{.*}}, <3 x float> %{{.*}}) 89 ; DOPCHECK: [[dotf3:%.*]] = call float @dx.op.dot3.f32(i32 55, float %{{.*}}, float %{{.*}}, float %{{.*}}, float %{{.*}}) 90 ; EXPCHECK: [[rsqrt:%.*]] = call float @llvm.dx.rsqrt.f32(float [[dotf3]]) 91 ; DOPCHECK: [[rsqrt:%.*]] = call float @dx.op.unary.f32(i32 25, float [[dotf3]]) 92 ; CHECK: [[splatinsertf3:%.*]] = insertelement <3 x float> poison, float [[rsqrt]], i64 0 93 ; CHECK: [[splat:%.*]] shufflevector <3 x float> [[splatinsertf3]], <3 x float> poison, <3 x i32> zeroinitializer 94 ; CHECK: fmul <3 x float> %p0, %.splat 95 96 %hlsl.normalize = call <3 x float> @llvm.dx.normalize.v3f32(<3 x float> %p0) 97 ret <3 x float> %hlsl.normalize 98} 99 100define noundef <4 x float> @test_normalize_float4(<4 x float> noundef %p0) { 101entry: 102 ; EXPCHECK: [[dotf4:%.*]] = call float @llvm.dx.dot4.v4f32(<4 x float> %{{.*}}, <4 x float> %{{.*}}) 103 ; DOPCHECK: [[dotf4:%.*]] = call float @dx.op.dot4.f32(i32 56, float %{{.*}}, float %{{.*}}, float %{{.*}}, float %{{.*}}) 104 ; EXPCHECK: [[rsqrt:%.*]] = call float @llvm.dx.rsqrt.f32(float [[dotf4]]) 105 ; DOPCHECK: [[rsqrt:%.*]] = call float @dx.op.unary.f32(i32 25, float [[dotf4]]) 106 ; CHECK: [[splatinsertf4:%.*]] = insertelement <4 x float> poison, float [[rsqrt]], i64 0 107 ; CHECK: [[splat:%.*]] shufflevector <4 x float> [[splatinsertf4]], <4 x float> poison, <4 x i32> zeroinitializer 108 ; CHECK: fmul <4 x float> %p0, %.splat 109 110 %hlsl.normalize = call <4 x float> @llvm.dx.normalize.v4f32(<4 x float> %p0) 111 ret <4 x float> %hlsl.normalize 112} 113