129 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -mtriple=x86_64-unknown -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=SSE3; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=slm -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=SLM4; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=AVX5; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=AVX26; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=knl -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=AVX27; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skx -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=AVX28 9define <8 x float> @ceil_floor(<8 x float> %a) {10; SSE-LABEL: @ceil_floor(11; SSE-NEXT: [[A0:%.*]] = extractelement <8 x float> [[A:%.*]], i32 012; SSE-NEXT: [[A1:%.*]] = extractelement <8 x float> [[A]], i32 113; SSE-NEXT: [[A2:%.*]] = extractelement <8 x float> [[A]], i32 214; SSE-NEXT: [[A3:%.*]] = extractelement <8 x float> [[A]], i32 315; SSE-NEXT: [[A4:%.*]] = extractelement <8 x float> [[A]], i32 416; SSE-NEXT: [[A5:%.*]] = extractelement <8 x float> [[A]], i32 517; SSE-NEXT: [[A6:%.*]] = extractelement <8 x float> [[A]], i32 618; SSE-NEXT: [[A7:%.*]] = extractelement <8 x float> [[A]], i32 719; SSE-NEXT: [[AB0:%.*]] = call float @llvm.ceil.f32(float [[A0]])20; SSE-NEXT: [[AB1:%.*]] = call float @llvm.floor.f32(float [[A1]])21; SSE-NEXT: [[AB2:%.*]] = call float @llvm.floor.f32(float [[A2]])22; SSE-NEXT: [[AB3:%.*]] = call float @llvm.ceil.f32(float [[A3]])23; SSE-NEXT: [[AB4:%.*]] = call float @llvm.ceil.f32(float [[A4]])24; SSE-NEXT: [[AB5:%.*]] = call float @llvm.ceil.f32(float [[A5]])25; SSE-NEXT: [[AB6:%.*]] = call float @llvm.floor.f32(float [[A6]])26; SSE-NEXT: [[AB7:%.*]] = call float @llvm.floor.f32(float [[A7]])27; SSE-NEXT: [[R0:%.*]] = insertelement <8 x float> zeroinitializer, float [[AB0]], i32 028; SSE-NEXT: [[R1:%.*]] = insertelement <8 x float> [[R0]], float [[AB1]], i32 129; SSE-NEXT: [[R2:%.*]] = insertelement <8 x float> [[R1]], float [[AB2]], i32 230; SSE-NEXT: [[R3:%.*]] = insertelement <8 x float> [[R2]], float [[AB3]], i32 331; SSE-NEXT: [[R4:%.*]] = insertelement <8 x float> [[R3]], float [[AB4]], i32 432; SSE-NEXT: [[R5:%.*]] = insertelement <8 x float> [[R4]], float [[AB5]], i32 533; SSE-NEXT: [[R6:%.*]] = insertelement <8 x float> [[R5]], float [[AB6]], i32 634; SSE-NEXT: [[R71:%.*]] = insertelement <8 x float> [[R6]], float [[AB7]], i32 735; SSE-NEXT: ret <8 x float> [[R71]]36;37; SLM-LABEL: @ceil_floor(38; SLM-NEXT: [[A0:%.*]] = extractelement <8 x float> [[TMP1:%.*]], i32 039; SLM-NEXT: [[A3:%.*]] = extractelement <8 x float> [[TMP1]], i32 340; SLM-NEXT: [[AB0:%.*]] = call float @llvm.ceil.f32(float [[A0]])41; SLM-NEXT: [[TMP2:%.*]] = shufflevector <8 x float> [[TMP1]], <8 x float> poison, <2 x i32> <i32 1, i32 2>42; SLM-NEXT: [[TMP10:%.*]] = call <2 x float> @llvm.floor.v2f32(<2 x float> [[TMP2]])43; SLM-NEXT: [[AB3:%.*]] = call float @llvm.ceil.f32(float [[A3]])44; SLM-NEXT: [[TMP3:%.*]] = shufflevector <8 x float> [[TMP1]], <8 x float> poison, <2 x i32> <i32 4, i32 5>45; SLM-NEXT: [[TMP4:%.*]] = call <2 x float> @llvm.ceil.v2f32(<2 x float> [[TMP3]])46; SLM-NEXT: [[TMP5:%.*]] = shufflevector <8 x float> [[TMP1]], <8 x float> poison, <2 x i32> <i32 6, i32 7>47; SLM-NEXT: [[TMP6:%.*]] = call <2 x float> @llvm.floor.v2f32(<2 x float> [[TMP5]])48; SLM-NEXT: [[R0:%.*]] = insertelement <8 x float> zeroinitializer, float [[AB0]], i32 049; SLM-NEXT: [[TMP7:%.*]] = shufflevector <2 x float> [[TMP10]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>50; SLM-NEXT: [[R23:%.*]] = shufflevector <8 x float> [[R0]], <8 x float> [[TMP7]], <8 x i32> <i32 0, i32 8, i32 9, i32 3, i32 4, i32 5, i32 6, i32 7>51; SLM-NEXT: [[R3:%.*]] = insertelement <8 x float> [[R23]], float [[AB3]], i32 352; SLM-NEXT: [[TMP8:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>53; SLM-NEXT: [[R52:%.*]] = shufflevector <8 x float> [[R3]], <8 x float> [[TMP8]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 6, i32 7>54; SLM-NEXT: [[TMP9:%.*]] = shufflevector <2 x float> [[TMP6]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>55; SLM-NEXT: [[R71:%.*]] = shufflevector <8 x float> [[R52]], <8 x float> [[TMP9]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 8, i32 9>56; SLM-NEXT: ret <8 x float> [[R71]]57;58; AVX-LABEL: @ceil_floor(59; AVX-NEXT: [[A0:%.*]] = extractelement <8 x float> [[TMP1:%.*]], i32 060; AVX-NEXT: [[A3:%.*]] = extractelement <8 x float> [[TMP1]], i32 361; AVX-NEXT: [[AB0:%.*]] = call float @llvm.ceil.f32(float [[A0]])62; AVX-NEXT: [[TMP9:%.*]] = shufflevector <8 x float> [[TMP1]], <8 x float> poison, <2 x i32> <i32 1, i32 2>63; AVX-NEXT: [[TMP8:%.*]] = call <2 x float> @llvm.floor.v2f32(<2 x float> [[TMP9]])64; AVX-NEXT: [[AB3:%.*]] = call float @llvm.ceil.f32(float [[A3]])65; AVX-NEXT: [[TMP3:%.*]] = shufflevector <8 x float> [[TMP1]], <8 x float> poison, <2 x i32> <i32 4, i32 5>66; AVX-NEXT: [[TMP2:%.*]] = call <2 x float> @llvm.ceil.v2f32(<2 x float> [[TMP3]])67; AVX-NEXT: [[TMP10:%.*]] = shufflevector <8 x float> [[TMP1]], <8 x float> poison, <2 x i32> <i32 6, i32 7>68; AVX-NEXT: [[TMP4:%.*]] = call <2 x float> @llvm.floor.v2f32(<2 x float> [[TMP10]])69; AVX-NEXT: [[R0:%.*]] = insertelement <8 x float> zeroinitializer, float [[AB0]], i32 070; AVX-NEXT: [[TMP7:%.*]] = shufflevector <2 x float> [[TMP8]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>71; AVX-NEXT: [[R23:%.*]] = shufflevector <8 x float> [[R0]], <8 x float> [[TMP7]], <8 x i32> <i32 0, i32 8, i32 9, i32 3, i32 4, i32 5, i32 6, i32 7>72; AVX-NEXT: [[R3:%.*]] = insertelement <8 x float> [[R23]], float [[AB3]], i32 373; AVX-NEXT: [[TMP5:%.*]] = shufflevector <2 x float> [[TMP2]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>74; AVX-NEXT: [[R52:%.*]] = shufflevector <8 x float> [[R3]], <8 x float> [[TMP5]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 6, i32 7>75; AVX-NEXT: [[TMP6:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>76; AVX-NEXT: [[R71:%.*]] = shufflevector <8 x float> [[R52]], <8 x float> [[TMP6]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 8, i32 9>77; AVX-NEXT: ret <8 x float> [[R71]]78;79; AVX2-LABEL: @ceil_floor(80; AVX2-NEXT: [[A0:%.*]] = extractelement <8 x float> [[TMP1:%.*]], i32 081; AVX2-NEXT: [[A3:%.*]] = extractelement <8 x float> [[TMP1]], i32 382; AVX2-NEXT: [[AB0:%.*]] = call float @llvm.ceil.f32(float [[A0]])83; AVX2-NEXT: [[TMP2:%.*]] = shufflevector <8 x float> [[TMP1]], <8 x float> poison, <2 x i32> <i32 1, i32 2>84; AVX2-NEXT: [[TMP10:%.*]] = call <2 x float> @llvm.floor.v2f32(<2 x float> [[TMP2]])85; AVX2-NEXT: [[AB3:%.*]] = call float @llvm.ceil.f32(float [[A3]])86; AVX2-NEXT: [[TMP3:%.*]] = shufflevector <8 x float> [[TMP1]], <8 x float> poison, <2 x i32> <i32 4, i32 5>87; AVX2-NEXT: [[TMP4:%.*]] = call <2 x float> @llvm.ceil.v2f32(<2 x float> [[TMP3]])88; AVX2-NEXT: [[TMP5:%.*]] = shufflevector <8 x float> [[TMP1]], <8 x float> poison, <2 x i32> <i32 6, i32 7>89; AVX2-NEXT: [[TMP6:%.*]] = call <2 x float> @llvm.floor.v2f32(<2 x float> [[TMP5]])90; AVX2-NEXT: [[R0:%.*]] = insertelement <8 x float> zeroinitializer, float [[AB0]], i32 091; AVX2-NEXT: [[TMP7:%.*]] = shufflevector <2 x float> [[TMP10]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>92; AVX2-NEXT: [[R23:%.*]] = shufflevector <8 x float> [[R0]], <8 x float> [[TMP7]], <8 x i32> <i32 0, i32 8, i32 9, i32 3, i32 4, i32 5, i32 6, i32 7>93; AVX2-NEXT: [[R3:%.*]] = insertelement <8 x float> [[R23]], float [[AB3]], i32 394; AVX2-NEXT: [[TMP8:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>95; AVX2-NEXT: [[R52:%.*]] = shufflevector <8 x float> [[R3]], <8 x float> [[TMP8]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 6, i32 7>96; AVX2-NEXT: [[TMP9:%.*]] = shufflevector <2 x float> [[TMP6]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>97; AVX2-NEXT: [[R71:%.*]] = shufflevector <8 x float> [[R52]], <8 x float> [[TMP9]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 8, i32 9>98; AVX2-NEXT: ret <8 x float> [[R71]]99;100 %a0 = extractelement <8 x float> %a, i32 0101 %a1 = extractelement <8 x float> %a, i32 1102 %a2 = extractelement <8 x float> %a, i32 2103 %a3 = extractelement <8 x float> %a, i32 3104 %a4 = extractelement <8 x float> %a, i32 4105 %a5 = extractelement <8 x float> %a, i32 5106 %a6 = extractelement <8 x float> %a, i32 6107 %a7 = extractelement <8 x float> %a, i32 7108 %ab0 = call float @llvm.ceil.f32(float %a0)109 %ab1 = call float @llvm.floor.f32(float %a1)110 %ab2 = call float @llvm.floor.f32(float %a2)111 %ab3 = call float @llvm.ceil.f32(float %a3)112 %ab4 = call float @llvm.ceil.f32(float %a4)113 %ab5 = call float @llvm.ceil.f32(float %a5)114 %ab6 = call float @llvm.floor.f32(float %a6)115 %ab7 = call float @llvm.floor.f32(float %a7)116 %r0 = insertelement <8 x float> zeroinitializer, float %ab0, i32 0117 %r1 = insertelement <8 x float> %r0, float %ab1, i32 1118 %r2 = insertelement <8 x float> %r1, float %ab2, i32 2119 %r3 = insertelement <8 x float> %r2, float %ab3, i32 3120 %r4 = insertelement <8 x float> %r3, float %ab4, i32 4121 %r5 = insertelement <8 x float> %r4, float %ab5, i32 5122 %r6 = insertelement <8 x float> %r5, float %ab6, i32 6123 %r7 = insertelement <8 x float> %r6, float %ab7, i32 7124 ret <8 x float> %r7125}126 127declare float @llvm.ceil.f32(float)128declare float @llvm.floor.f32(float)129