214 lines · c
1// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-cir %s -o %t.cir2// RUN: FileCheck --input-file=%t.cir %s --check-prefix=CIR3// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-llvm %s -o %t-cir.ll4// RUN: FileCheck --input-file=%t-cir.ll %s -check-prefix=LLVM5// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -emit-llvm %s -o %t.ll6// RUN: FileCheck --input-file=%t.ll %s -check-prefix=OGCG7 8typedef float __m128 __attribute__((__vector_size__(16), __aligned__(16)));9typedef double __m128d __attribute__((__vector_size__(16), __aligned__(16)));10 11__m128 test_cmpnleps(__m128 A, __m128 B) {12 // CIR-LABEL: cir.func dso_local @test_cmpnleps(13 // CIR: %[[ARG0:.*]]: !cir.vector<4 x !cir.float> {{.*}}, %[[ARG1:.*]]: !cir.vector<4 x !cir.float> {{.*}}) -> !cir.vector<4 x !cir.float> inline(never) {14 // CIR: %[[ALLOCA_0:.*]] = cir.alloca !cir.vector<4 x !cir.float>, !cir.ptr<!cir.vector<4 x !cir.float>>, ["A", init] {alignment = 16 : i64}15 // CIR: %[[ALLOCA_1:.*]] = cir.alloca !cir.vector<4 x !cir.float>, !cir.ptr<!cir.vector<4 x !cir.float>>, ["B", init] {alignment = 16 : i64}16 // CIR: %[[ALLOCA_2:.*]] = cir.alloca !cir.vector<4 x !cir.float>, !cir.ptr<!cir.vector<4 x !cir.float>>, ["__retval"] {alignment = 16 : i64}17 // CIR: cir.store %[[ARG0]], %[[ALLOCA_0]] : !cir.vector<4 x !cir.float>, !cir.ptr<!cir.vector<4 x !cir.float>>18 // CIR: cir.store %[[ARG1]], %[[ALLOCA_1]] : !cir.vector<4 x !cir.float>, !cir.ptr<!cir.vector<4 x !cir.float>>19 // CIR: %[[LOAD_0:.*]] = cir.load align(16) %[[ALLOCA_0]] : !cir.ptr<!cir.vector<4 x !cir.float>>, !cir.vector<4 x !cir.float>20 // CIR: %[[LOAD_1:.*]] = cir.load align(16) %[[ALLOCA_1]] : !cir.ptr<!cir.vector<4 x !cir.float>>, !cir.vector<4 x !cir.float>21 // CIR: %[[VEC_0:.*]] = cir.vec.cmp(le, %[[LOAD_0]], %[[LOAD_1]]) : !cir.vector<4 x !cir.float>, !cir.vector<4 x !s32i>22 // CIR: %[[UNARY_0:.*]] = cir.unary(not, %[[VEC_0]]) : !cir.vector<4 x !s32i>, !cir.vector<4 x !s32i>23 // CIR: %[[CAST_0:.*]] = cir.cast bitcast %[[UNARY_0]] : !cir.vector<4 x !s32i> -> !cir.vector<4 x !cir.float>24 // CIR: cir.store %[[CAST_0]], %[[ALLOCA_2]] : !cir.vector<4 x !cir.float>, !cir.ptr<!cir.vector<4 x !cir.float>>25 // CIR: %[[LOAD_2:.*]] = cir.load %[[ALLOCA_2]] : !cir.ptr<!cir.vector<4 x !cir.float>>, !cir.vector<4 x !cir.float>26 // CIR: cir.return %[[LOAD_2]] : !cir.vector<4 x !cir.float>27 // CIR: } 28 29 // LLVM-LABEL: define dso_local <4 x float> @test_cmpnleps(30 // LLVM-SAME: <4 x float> [[TMP0:%.*]], <4 x float> [[TMP1:%.*]]) #[[ATTR0:[0-9]+]] {31 // LLVM-NEXT: [[TMP3:%.*]] = alloca <4 x float>, i64 1, align 1632 // LLVM-NEXT: [[TMP4:%.*]] = alloca <4 x float>, i64 1, align 1633 // LLVM-NEXT: [[TMP5:%.*]] = alloca <4 x float>, i64 1, align 1634 // LLVM-NEXT: store <4 x float> [[TMP0]], ptr [[TMP3]], align 1635 // LLVM-NEXT: store <4 x float> [[TMP1]], ptr [[TMP4]], align 1636 // LLVM-NEXT: [[TMP6:%.*]] = load <4 x float>, ptr [[TMP3]], align 1637 // LLVM-NEXT: [[TMP7:%.*]] = load <4 x float>, ptr [[TMP4]], align 1638 // LLVM-NEXT: [[TMP8:%.*]] = fcmp ole <4 x float> [[TMP6]], [[TMP7]]39 // LLVM-NEXT: [[TMP9:%.*]] = sext <4 x i1> [[TMP8]] to <4 x i32>40 // LLVM-NEXT: [[TMP10:%.*]] = xor <4 x i32> [[TMP9]], splat (i32 -1)41 // LLVM-NEXT: [[TMP11:%.*]] = bitcast <4 x i32> [[TMP10]] to <4 x float>42 // LLVM-NEXT: store <4 x float> [[TMP11]], ptr [[TMP5]], align 1643 // LLVM-NEXT: [[TMP12:%.*]] = load <4 x float>, ptr [[TMP5]], align 1644 // LLVM-NEXT: ret <4 x float> [[TMP12]]45 46 // OGCG-LABEL: define dso_local <4 x float> @test_cmpnleps(47 // OGCG-SAME: <4 x float> noundef [[A:%.*]], <4 x float> noundef [[B:%.*]]) #[[ATTR0:[0-9]+]] {48 // OGCG-NEXT: [[ENTRY:.*:]]49 // OGCG-NEXT: [[A_ADDR:%.*]] = alloca <4 x float>, align 1650 // OGCG-NEXT: [[B_ADDR:%.*]] = alloca <4 x float>, align 1651 // OGCG-NEXT: store <4 x float> [[A]], ptr [[A_ADDR]], align 1652 // OGCG-NEXT: store <4 x float> [[B]], ptr [[B_ADDR]], align 1653 // OGCG-NEXT: [[TMP0:%.*]] = load <4 x float>, ptr [[A_ADDR]], align 1654 // OGCG-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr [[B_ADDR]], align 1655 // OGCG-NEXT: [[TMP2:%.*]] = fcmp ugt <4 x float> [[TMP0]], [[TMP1]]56 // OGCG-NEXT: [[TMP3:%.*]] = sext <4 x i1> [[TMP2]] to <4 x i32>57 // OGCG-NEXT: [[TMP4:%.*]] = bitcast <4 x i32> [[TMP3]] to <4 x float>58 // OGCG-NEXT: ret <4 x float> [[TMP4]]59 return __builtin_ia32_cmpnleps(A, B);60}61 62__m128d test_cmpnlepd(__m128d A, __m128d B) {63 // CIR-LABEL: cir.func dso_local @test_cmpnlepd(64 // CIR： %[[ARG0:.*]]: !cir.vector<2 x !cir.double> {{.*}}, %[[ARG1:.*]]: !cir.vector<2 x !cir.double> {{.*}}) -> !cir.vector<2 x !cir.double> inline(never) {65 // CIR: %[[ALLOCA_0:.*]] = cir.alloca !cir.vector<2 x !cir.double>, !cir.ptr<!cir.vector<2 x !cir.double>>, ["A", init] {alignment = 16 : i64} 66 // CIR: %[[ALLOCA_1:.*]] = cir.alloca !cir.vector<2 x !cir.double>, !cir.ptr<!cir.vector<2 x !cir.double>>, ["B", init] {alignment = 16 : i64} 67 // CIR: %[[ALLOCA_2:.*]] = cir.alloca !cir.vector<2 x !cir.double>, !cir.ptr<!cir.vector<2 x !cir.double>>, ["__retval"] {alignment = 16 : i64} 68 // CIR: cir.store %[[ARG0]], %[[ALLOCA_0]] : !cir.vector<2 x !cir.double>, !cir.ptr<!cir.vector<2 x !cir.double>> 69 // CIR: cir.store %[[ARG1]], %[[ALLOCA_1]] : !cir.vector<2 x !cir.double>, !cir.ptr<!cir.vector<2 x !cir.double>> 70 // CIR: %[[LOAD_0:.*]] = cir.load align(16) %[[ALLOCA_0]] : !cir.ptr<!cir.vector<2 x !cir.double>>, !cir.vector<2 x !cir.double> 71 // CIR: %[[LOAD_1:.*]] = cir.load align(16) %[[ALLOCA_1]] : !cir.ptr<!cir.vector<2 x !cir.double>>, !cir.vector<2 x !cir.double> 72 // CIR: %[[VEC_0:.*]] = cir.vec.cmp(le, %[[LOAD_0]], %[[LOAD_1]]) : !cir.vector<2 x !cir.double>, !cir.vector<2 x !s64i> 73 // CIR: %[[UNARY_0:.*]] = cir.unary(not, %[[VEC_0]]) : !cir.vector<2 x !s64i>, !cir.vector<2 x !s64i> 74 // CIR: %[[CAST_0:.*]] = cir.cast bitcast %[[UNARY_0]] : !cir.vector<2 x !s64i> -> !cir.vector<2 x !cir.double> 75 // CIR: cir.store %[[CAST_0]], %[[ALLOCA_2]] : !cir.vector<2 x !cir.double>, !cir.ptr<!cir.vector<2 x !cir.double>> 76 // CIR: %[[LOAD_2:.*]] = cir.load %[[ALLOCA_2]] : !cir.ptr<!cir.vector<2 x !cir.double>>, !cir.vector<2 x !cir.double> 77 // CIR: cir.return %[[LOAD_2]] : !cir.vector<2 x !cir.double> 78 // CIR: } 79 80 // LLVM-LABEL: define dso_local <2 x double> @test_cmpnlepd(81 // LLVM-SAME: <2 x double> [[TMP0:%.*]], <2 x double> [[TMP1:%.*]]) #[[ATTR0:[0-9]+]] {82 // LLVM-NEXT: [[TMP3:%.*]] = alloca <2 x double>, i64 1, align 1683 // LLVM-NEXT: [[TMP4:%.*]] = alloca <2 x double>, i64 1, align 1684 // LLVM-NEXT: [[TMP5:%.*]] = alloca <2 x double>, i64 1, align 1685 // LLVM-NEXT: store <2 x double> [[TMP0]], ptr [[TMP3]], align 1686 // LLVM-NEXT: store <2 x double> [[TMP1]], ptr [[TMP4]], align 1687 // LLVM-NEXT: [[TMP6:%.*]] = load <2 x double>, ptr [[TMP3]], align 1688 // LLVM-NEXT: [[TMP7:%.*]] = load <2 x double>, ptr [[TMP4]], align 1689 // LLVM-NEXT: [[TMP8:%.*]] = fcmp ole <2 x double> [[TMP6]], [[TMP7]]90 // LLVM-NEXT: [[TMP9:%.*]] = sext <2 x i1> [[TMP8]] to <2 x i64>91 // LLVM-NEXT: [[TMP10:%.*]] = xor <2 x i64> [[TMP9]], splat (i64 -1)92 // LLVM-NEXT: [[TMP11:%.*]] = bitcast <2 x i64> [[TMP10]] to <2 x double>93 // LLVM-NEXT: store <2 x double> [[TMP11]], ptr [[TMP5]], align 1694 // LLVM-NEXT: [[TMP12:%.*]] = load <2 x double>, ptr [[TMP5]], align 1695 // LLVM-NEXT: ret <2 x double> [[TMP12]]96 97 // OGCG-LABEL: define dso_local <2 x double> @test_cmpnlepd(98 // OGCG-SAME: <2 x double> noundef [[A:%.*]], <2 x double> noundef [[B:%.*]]) #[[ATTR0:[0-9]+]] {99 // OGCG-NEXT: [[ENTRY:.*:]]100 // OGCG-NEXT: [[A_ADDR:%.*]] = alloca <2 x double>, align 16101 // OGCG-NEXT: [[B_ADDR:%.*]] = alloca <2 x double>, align 16102 // OGCG-NEXT: store <2 x double> [[A]], ptr [[A_ADDR]], align 16103 // OGCG-NEXT: store <2 x double> [[B]], ptr [[B_ADDR]], align 16104 // OGCG-NEXT: [[TMP0:%.*]] = load <2 x double>, ptr [[A_ADDR]], align 16105 // OGCG-NEXT: [[TMP1:%.*]] = load <2 x double>, ptr [[B_ADDR]], align 16106 // OGCG-NEXT: [[TMP2:%.*]] = fcmp ugt <2 x double> [[TMP0]], [[TMP1]]107 // OGCG-NEXT: [[TMP3:%.*]] = sext <2 x i1> [[TMP2]] to <2 x i64>108 // OGCG-NEXT: [[TMP4:%.*]] = bitcast <2 x i64> [[TMP3]] to <2 x double>109 // OGCG-NEXT: ret <2 x double> [[TMP4]]110 return __builtin_ia32_cmpnlepd(A, B);111}112 113__m128 test_cmpnltps(__m128 A, __m128 B) {114 // CIR-LABEL: cir.func dso_local @test_cmpnltps(115 // CIR-SAME: %[[ARG0:.*]]: !cir.vector<4 x !cir.float> {{.*}}, %[[ARG1:.*]]: !cir.vector<4 x !cir.float> {{.*}}) -> !cir.vector<4 x !cir.float> inline(never) {116 // CIR: %[[ALLOCA_0:.*]] = cir.alloca !cir.vector<4 x !cir.float>, !cir.ptr<!cir.vector<4 x !cir.float>>, ["A", init] {alignment = 16 : i64} 117 // CIR: %[[ALLOCA_1:.*]] = cir.alloca !cir.vector<4 x !cir.float>, !cir.ptr<!cir.vector<4 x !cir.float>>, ["B", init] {alignment = 16 : i64} 118 // CIR: %[[ALLOCA_2:.*]] = cir.alloca !cir.vector<4 x !cir.float>, !cir.ptr<!cir.vector<4 x !cir.float>>, ["__retval"] {alignment = 16 : i64} 119 // CIR: cir.store %[[ARG0]], %[[ALLOCA_0]] : !cir.vector<4 x !cir.float>, !cir.ptr<!cir.vector<4 x !cir.float>> 120 // CIR: cir.store %[[ARG1]], %[[ALLOCA_1]] : !cir.vector<4 x !cir.float>, !cir.ptr<!cir.vector<4 x !cir.float>> 121 // CIR: %[[LOAD_0:.*]] = cir.load align(16) %[[ALLOCA_0]] : !cir.ptr<!cir.vector<4 x !cir.float>>, !cir.vector<4 x !cir.float> 122 // CIR: %[[LOAD_1:.*]] = cir.load align(16) %[[ALLOCA_1]] : !cir.ptr<!cir.vector<4 x !cir.float>>, !cir.vector<4 x !cir.float> 123 // CIR: %[[VEC_0:.*]] = cir.vec.cmp(lt, %[[LOAD_0]], %[[LOAD_1]]) : !cir.vector<4 x !cir.float>, !cir.vector<4 x !s32i> 124 // CIR: %[[UNARY_0:.*]] = cir.unary(not, %[[VEC_0]]) : !cir.vector<4 x !s32i>, !cir.vector<4 x !s32i> 125 // CIR: %[[CAST_0:.*]] = cir.cast bitcast %[[UNARY_0]] : !cir.vector<4 x !s32i> -> !cir.vector<4 x !cir.float> 126 // CIR: cir.store %[[CAST_0]], %[[ALLOCA_2]] : !cir.vector<4 x !cir.float>, !cir.ptr<!cir.vector<4 x !cir.float>> 127 // CIR: %[[LOAD_2:.*]] = cir.load %[[ALLOCA_2]] : !cir.ptr<!cir.vector<4 x !cir.float>>, !cir.vector<4 x !cir.float> 128 // CIR: cir.return %[[LOAD_2]] : !cir.vector<4 x !cir.float> 129 // CIR: } 130 131 // LLVM-LABEL: define dso_local <4 x float> @test_cmpnltps(132 // LLVM-SAME: <4 x float> [[TMP0:%.*]], <4 x float> [[TMP1:%.*]]) #[[ATTR0:[0-9]+]] {133 // LLVM-NEXT: [[TMP3:%.*]] = alloca <4 x float>, i64 1, align 16134 // LLVM-NEXT: [[TMP4:%.*]] = alloca <4 x float>, i64 1, align 16135 // LLVM-NEXT: [[TMP5:%.*]] = alloca <4 x float>, i64 1, align 16136 // LLVM-NEXT: store <4 x float> [[TMP0]], ptr [[TMP3]], align 16137 // LLVM-NEXT: store <4 x float> [[TMP1]], ptr [[TMP4]], align 16138 // LLVM-NEXT: [[TMP6:%.*]] = load <4 x float>, ptr [[TMP3]], align 16139 // LLVM-NEXT: [[TMP7:%.*]] = load <4 x float>, ptr [[TMP4]], align 16140 // LLVM-NEXT: [[TMP8:%.*]] = fcmp olt <4 x float> [[TMP6]], [[TMP7]]141 // LLVM-NEXT: [[TMP9:%.*]] = sext <4 x i1> [[TMP8]] to <4 x i32>142 // LLVM-NEXT: [[TMP10:%.*]] = xor <4 x i32> [[TMP9]], splat (i32 -1)143 // LLVM-NEXT: [[TMP11:%.*]] = bitcast <4 x i32> [[TMP10]] to <4 x float>144 // LLVM-NEXT: store <4 x float> [[TMP11]], ptr [[TMP5]], align 16145 // LLVM-NEXT: [[TMP12:%.*]] = load <4 x float>, ptr [[TMP5]], align 16146 // LLVM-NEXT: ret <4 x float> [[TMP12]]147 148 // OGCG-LABEL: define dso_local <4 x float> @test_cmpnltps(149 // OGCG-SAME: <4 x float> noundef [[A:%.*]], <4 x float> noundef [[B:%.*]]) #[[ATTR0:[0-9]+]] {150 // OGCG-NEXT: [[ENTRY:.*:]]151 // OGCG-NEXT: [[A_ADDR:%.*]] = alloca <4 x float>, align 16152 // OGCG-NEXT: [[B_ADDR:%.*]] = alloca <4 x float>, align 16153 // OGCG-NEXT: store <4 x float> [[A]], ptr [[A_ADDR]], align 16154 // OGCG-NEXT: store <4 x float> [[B]], ptr [[B_ADDR]], align 16155 // OGCG-NEXT: [[TMP0:%.*]] = load <4 x float>, ptr [[A_ADDR]], align 16156 // OGCG-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr [[B_ADDR]], align 16157 // OGCG-NEXT: [[TMP2:%.*]] = fcmp uge <4 x float> [[TMP0]], [[TMP1]]158 // OGCG-NEXT: [[TMP3:%.*]] = sext <4 x i1> [[TMP2]] to <4 x i32>159 // OGCG-NEXT: [[TMP4:%.*]] = bitcast <4 x i32> [[TMP3]] to <4 x float>160 // OGCG-NEXT: ret <4 x float> [[TMP4]]161 return __builtin_ia32_cmpnltps(A, B);162}163 164__m128d test_cmpnltpd(__m128d A, __m128d B) {165 // CIR-LABEL: cir.func dso_local @test_cmpnltpd(166 // CIR: %[[ARG0:.*]]: !cir.vector<2 x !cir.double> {{.*}}, %[[ARG1:.*]]: !cir.vector<2 x !cir.double> {{.*}}) -> !cir.vector<2 x !cir.double> inline(never) {167 // CIR: %[[ALLOCA_0:.*]] = cir.alloca !cir.vector<2 x !cir.double>, !cir.ptr<!cir.vector<2 x !cir.double>>, ["A", init] {alignment = 16 : i64} 168 // CIR: %[[ALLOCA_1:.*]] = cir.alloca !cir.vector<2 x !cir.double>, !cir.ptr<!cir.vector<2 x !cir.double>>, ["B", init] {alignment = 16 : i64} 169 // CIR: %[[ALLOCA_2:.*]] = cir.alloca !cir.vector<2 x !cir.double>, !cir.ptr<!cir.vector<2 x !cir.double>>, ["__retval"] {alignment = 16 : i64} 170 // CIR: cir.store %[[ARG0]], %[[ALLOCA_0]] : !cir.vector<2 x !cir.double>, !cir.ptr<!cir.vector<2 x !cir.double>> 171 // CIR: cir.store %[[ARG1]], %[[ALLOCA_1]] : !cir.vector<2 x !cir.double>, !cir.ptr<!cir.vector<2 x !cir.double>> 172 // CIR: %[[LOAD_0:.*]] = cir.load align(16) %[[ALLOCA_0]] : !cir.ptr<!cir.vector<2 x !cir.double>>, !cir.vector<2 x !cir.double> 173 // CIR: %[[LOAD_1:.*]] = cir.load align(16) %[[ALLOCA_1]] : !cir.ptr<!cir.vector<2 x !cir.double>>, !cir.vector<2 x !cir.double> 174 // CIR: %[[VEC_0:.*]] = cir.vec.cmp(lt, %[[LOAD_0]], %[[LOAD_1]]) : !cir.vector<2 x !cir.double>, !cir.vector<2 x !s64i> 175 // CIR: %[[UNARY_0:.*]] = cir.unary(not, %[[VEC_0]]) : !cir.vector<2 x !s64i>, !cir.vector<2 x !s64i> 176 // CIR: %[[CAST_0:.*]] = cir.cast bitcast %[[UNARY_0]] : !cir.vector<2 x !s64i> -> !cir.vector<2 x !cir.double> 177 // CIR: cir.store %[[CAST_0]], %[[ALLOCA_2]] : !cir.vector<2 x !cir.double>, !cir.ptr<!cir.vector<2 x !cir.double>> 178 // CIR: %[[LOAD_2:.*]] = cir.load %[[ALLOCA_2]] : !cir.ptr<!cir.vector<2 x !cir.double>>, !cir.vector<2 x !cir.double> 179 // CIR: cir.return %[[LOAD_2]] : !cir.vector<2 x !cir.double> 180 // CIR: } 181 182 // LLVM-LABEL: define dso_local <2 x double> @test_cmpnltpd(183 // LLVM-SAME: <2 x double> [[TMP0:%.*]], <2 x double> [[TMP1:%.*]]) #[[ATTR0:[0-9]+]] {184 // LLVM-NEXT: [[TMP3:%.*]] = alloca <2 x double>, i64 1, align 16185 // LLVM-NEXT: [[TMP4:%.*]] = alloca <2 x double>, i64 1, align 16186 // LLVM-NEXT: [[TMP5:%.*]] = alloca <2 x double>, i64 1, align 16187 // LLVM-NEXT: store <2 x double> [[TMP0]], ptr [[TMP3]], align 16188 // LLVM-NEXT: store <2 x double> [[TMP1]], ptr [[TMP4]], align 16189 // LLVM-NEXT: [[TMP6:%.*]] = load <2 x double>, ptr [[TMP3]], align 16190 // LLVM-NEXT: [[TMP7:%.*]] = load <2 x double>, ptr [[TMP4]], align 16191 // LLVM-NEXT: [[TMP8:%.*]] = fcmp olt <2 x double> [[TMP6]], [[TMP7]]192 // LLVM-NEXT: [[TMP9:%.*]] = sext <2 x i1> [[TMP8]] to <2 x i64>193 // LLVM-NEXT: [[TMP10:%.*]] = xor <2 x i64> [[TMP9]], splat (i64 -1)194 // LLVM-NEXT: [[TMP11:%.*]] = bitcast <2 x i64> [[TMP10]] to <2 x double>195 // LLVM-NEXT: store <2 x double> [[TMP11]], ptr [[TMP5]], align 16196 // LLVM-NEXT: [[TMP12:%.*]] = load <2 x double>, ptr [[TMP5]], align 16197 // LLVM-NEXT: ret <2 x double> [[TMP12]]198 199 // OGCG-LABEL: define dso_local <2 x double> @test_cmpnltpd(200 // OGCG-SAME: <2 x double> noundef [[A:%.*]], <2 x double> noundef [[B:%.*]]) #[[ATTR0:[0-9]+]] {201 // OGCG-NEXT: [[ENTRY:.*:]]202 // OGCG-NEXT: [[A_ADDR:%.*]] = alloca <2 x double>, align 16203 // OGCG-NEXT: [[B_ADDR:%.*]] = alloca <2 x double>, align 16204 // OGCG-NEXT: store <2 x double> [[A]], ptr [[A_ADDR]], align 16205 // OGCG-NEXT: store <2 x double> [[B]], ptr [[B_ADDR]], align 16206 // OGCG-NEXT: [[TMP0:%.*]] = load <2 x double>, ptr [[A_ADDR]], align 16207 // OGCG-NEXT: [[TMP1:%.*]] = load <2 x double>, ptr [[B_ADDR]], align 16208 // OGCG-NEXT: [[TMP2:%.*]] = fcmp uge <2 x double> [[TMP0]], [[TMP1]]209 // OGCG-NEXT: [[TMP3:%.*]] = sext <2 x i1> [[TMP2]] to <2 x i64>210 // OGCG-NEXT: [[TMP4:%.*]] = bitcast <2 x i64> [[TMP3]] to <2 x double>211 // OGCG-NEXT: ret <2 x double> [[TMP4]]212 return __builtin_ia32_cmpnltpd(A, B);213}214