719 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=SSE,SSE23; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=slm -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=SSE,SLM4; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=AVX,AVX15; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=AVX,AVX26; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=knl -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=AVX5127; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skx -mattr=-prefer-256-bit -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=AVX5128; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skx -mattr=+prefer-256-bit -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=AVX9 10@a64 = common global [8 x i64] zeroinitializer, align 6411@b64 = common global [8 x i64] zeroinitializer, align 6412@c64 = common global [8 x i64] zeroinitializer, align 6413@a32 = common global [16 x i32] zeroinitializer, align 6414@b32 = common global [16 x i32] zeroinitializer, align 6415@c32 = common global [16 x i32] zeroinitializer, align 6416@a16 = common global [32 x i16] zeroinitializer, align 6417@b16 = common global [32 x i16] zeroinitializer, align 6418@c16 = common global [32 x i16] zeroinitializer, align 6419@a8 = common global [64 x i8] zeroinitializer, align 6420@b8 = common global [64 x i8] zeroinitializer, align 6421@c8 = common global [64 x i8] zeroinitializer, align 6422 23declare i64 @llvm.sadd.sat.i64(i64, i64)24declare i32 @llvm.sadd.sat.i32(i32, i32)25declare i16 @llvm.sadd.sat.i16(i16, i16)26declare i8 @llvm.sadd.sat.i8 (i8 , i8 )27 28define void @add_v8i64() {29; SSE2-LABEL: @add_v8i64(30; SSE2-NEXT: [[A0:%.*]] = load i64, ptr @a64, align 831; SSE2-NEXT: [[A1:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 1), align 832; SSE2-NEXT: [[A2:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 2), align 833; SSE2-NEXT: [[A3:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 3), align 834; SSE2-NEXT: [[A4:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 4), align 835; SSE2-NEXT: [[A5:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 5), align 836; SSE2-NEXT: [[A6:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 6), align 837; SSE2-NEXT: [[A7:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 7), align 838; SSE2-NEXT: [[B0:%.*]] = load i64, ptr @b64, align 839; SSE2-NEXT: [[B1:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 1), align 840; SSE2-NEXT: [[B2:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 2), align 841; SSE2-NEXT: [[B3:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 3), align 842; SSE2-NEXT: [[B4:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 4), align 843; SSE2-NEXT: [[B5:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 5), align 844; SSE2-NEXT: [[B6:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 6), align 845; SSE2-NEXT: [[B7:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 7), align 846; SSE2-NEXT: [[R0:%.*]] = call i64 @llvm.sadd.sat.i64(i64 [[A0]], i64 [[B0]])47; SSE2-NEXT: [[R1:%.*]] = call i64 @llvm.sadd.sat.i64(i64 [[A1]], i64 [[B1]])48; SSE2-NEXT: [[R2:%.*]] = call i64 @llvm.sadd.sat.i64(i64 [[A2]], i64 [[B2]])49; SSE2-NEXT: [[R3:%.*]] = call i64 @llvm.sadd.sat.i64(i64 [[A3]], i64 [[B3]])50; SSE2-NEXT: [[R4:%.*]] = call i64 @llvm.sadd.sat.i64(i64 [[A4]], i64 [[B4]])51; SSE2-NEXT: [[R5:%.*]] = call i64 @llvm.sadd.sat.i64(i64 [[A5]], i64 [[B5]])52; SSE2-NEXT: [[R6:%.*]] = call i64 @llvm.sadd.sat.i64(i64 [[A6]], i64 [[B6]])53; SSE2-NEXT: [[R7:%.*]] = call i64 @llvm.sadd.sat.i64(i64 [[A7]], i64 [[B7]])54; SSE2-NEXT: store i64 [[R0]], ptr @c64, align 855; SSE2-NEXT: store i64 [[R1]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 1), align 856; SSE2-NEXT: store i64 [[R2]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 2), align 857; SSE2-NEXT: store i64 [[R3]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 3), align 858; SSE2-NEXT: store i64 [[R4]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 4), align 859; SSE2-NEXT: store i64 [[R5]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 5), align 860; SSE2-NEXT: store i64 [[R6]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 6), align 861; SSE2-NEXT: store i64 [[R7]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 7), align 862; SSE2-NEXT: ret void63;64; SLM-LABEL: @add_v8i64(65; SLM-NEXT: [[TMP1:%.*]] = load <2 x i64>, ptr @a64, align 866; SLM-NEXT: [[TMP2:%.*]] = load <2 x i64>, ptr @b64, align 867; SLM-NEXT: [[TMP3:%.*]] = call <2 x i64> @llvm.sadd.sat.v2i64(<2 x i64> [[TMP1]], <2 x i64> [[TMP2]])68; SLM-NEXT: store <2 x i64> [[TMP3]], ptr @c64, align 869; SLM-NEXT: [[TMP4:%.*]] = load <2 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 2), align 870; SLM-NEXT: [[TMP5:%.*]] = load <2 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 2), align 871; SLM-NEXT: [[TMP6:%.*]] = call <2 x i64> @llvm.sadd.sat.v2i64(<2 x i64> [[TMP4]], <2 x i64> [[TMP5]])72; SLM-NEXT: store <2 x i64> [[TMP6]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 2), align 873; SLM-NEXT: [[TMP7:%.*]] = load <2 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 4), align 874; SLM-NEXT: [[TMP8:%.*]] = load <2 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 4), align 875; SLM-NEXT: [[TMP9:%.*]] = call <2 x i64> @llvm.sadd.sat.v2i64(<2 x i64> [[TMP7]], <2 x i64> [[TMP8]])76; SLM-NEXT: store <2 x i64> [[TMP9]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 4), align 877; SLM-NEXT: [[TMP10:%.*]] = load <2 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 6), align 878; SLM-NEXT: [[TMP11:%.*]] = load <2 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 6), align 879; SLM-NEXT: [[TMP12:%.*]] = call <2 x i64> @llvm.sadd.sat.v2i64(<2 x i64> [[TMP10]], <2 x i64> [[TMP11]])80; SLM-NEXT: store <2 x i64> [[TMP12]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 6), align 881; SLM-NEXT: ret void82;83; AVX-LABEL: @add_v8i64(84; AVX-NEXT: [[TMP1:%.*]] = load <4 x i64>, ptr @a64, align 885; AVX-NEXT: [[TMP2:%.*]] = load <4 x i64>, ptr @b64, align 886; AVX-NEXT: [[TMP3:%.*]] = call <4 x i64> @llvm.sadd.sat.v4i64(<4 x i64> [[TMP1]], <4 x i64> [[TMP2]])87; AVX-NEXT: store <4 x i64> [[TMP3]], ptr @c64, align 888; AVX-NEXT: [[TMP4:%.*]] = load <4 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 4), align 889; AVX-NEXT: [[TMP5:%.*]] = load <4 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 4), align 890; AVX-NEXT: [[TMP6:%.*]] = call <4 x i64> @llvm.sadd.sat.v4i64(<4 x i64> [[TMP4]], <4 x i64> [[TMP5]])91; AVX-NEXT: store <4 x i64> [[TMP6]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 4), align 892; AVX-NEXT: ret void93;94; AVX512-LABEL: @add_v8i64(95; AVX512-NEXT: [[TMP1:%.*]] = load <8 x i64>, ptr @a64, align 896; AVX512-NEXT: [[TMP2:%.*]] = load <8 x i64>, ptr @b64, align 897; AVX512-NEXT: [[TMP3:%.*]] = call <8 x i64> @llvm.sadd.sat.v8i64(<8 x i64> [[TMP1]], <8 x i64> [[TMP2]])98; AVX512-NEXT: store <8 x i64> [[TMP3]], ptr @c64, align 899; AVX512-NEXT: ret void100;101; AVX256BW-LABEL: @add_v8i64(102; AVX256BW-NEXT: [[TMP1:%.*]] = load <4 x i64>, ptr @a64, align 8103; AVX256BW-NEXT: [[TMP2:%.*]] = load <4 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 4), align 8104; AVX256BW-NEXT: [[TMP3:%.*]] = load <4 x i64>, ptr @b64, align 8105; AVX256BW-NEXT: [[TMP4:%.*]] = load <4 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 4), align 8106; AVX256BW-NEXT: [[TMP5:%.*]] = call <4 x i64> @llvm.sadd.sat.v4i64(<4 x i64> [[TMP1]], <4 x i64> [[TMP3]])107; AVX256BW-NEXT: [[TMP6:%.*]] = call <4 x i64> @llvm.sadd.sat.v4i64(<4 x i64> [[TMP2]], <4 x i64> [[TMP4]])108; AVX256BW-NEXT: store <4 x i64> [[TMP5]], ptr @c64, align 8109; AVX256BW-NEXT: store <4 x i64> [[TMP6]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 4), align 8110; AVX256BW-NEXT: ret void111 %a0 = load i64, ptr @a64, align 8112 %a1 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 1), align 8113 %a2 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 2), align 8114 %a3 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 3), align 8115 %a4 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 4), align 8116 %a5 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 5), align 8117 %a6 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 6), align 8118 %a7 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 7), align 8119 %b0 = load i64, ptr @b64, align 8120 %b1 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 1), align 8121 %b2 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 2), align 8122 %b3 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 3), align 8123 %b4 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 4), align 8124 %b5 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 5), align 8125 %b6 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 6), align 8126 %b7 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 7), align 8127 %r0 = call i64 @llvm.sadd.sat.i64(i64 %a0, i64 %b0)128 %r1 = call i64 @llvm.sadd.sat.i64(i64 %a1, i64 %b1)129 %r2 = call i64 @llvm.sadd.sat.i64(i64 %a2, i64 %b2)130 %r3 = call i64 @llvm.sadd.sat.i64(i64 %a3, i64 %b3)131 %r4 = call i64 @llvm.sadd.sat.i64(i64 %a4, i64 %b4)132 %r5 = call i64 @llvm.sadd.sat.i64(i64 %a5, i64 %b5)133 %r6 = call i64 @llvm.sadd.sat.i64(i64 %a6, i64 %b6)134 %r7 = call i64 @llvm.sadd.sat.i64(i64 %a7, i64 %b7)135 store i64 %r0, ptr @c64, align 8136 store i64 %r1, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 1), align 8137 store i64 %r2, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 2), align 8138 store i64 %r3, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 3), align 8139 store i64 %r4, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 4), align 8140 store i64 %r5, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 5), align 8141 store i64 %r6, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 6), align 8142 store i64 %r7, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 7), align 8143 ret void144}145 146define void @add_v16i32() {147; SSE-LABEL: @add_v16i32(148; SSE-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @a32, align 4149; SSE-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr @b32, align 4150; SSE-NEXT: [[TMP3:%.*]] = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> [[TMP1]], <4 x i32> [[TMP2]])151; SSE-NEXT: store <4 x i32> [[TMP3]], ptr @c32, align 4152; SSE-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 4), align 4153; SSE-NEXT: [[TMP5:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 4), align 4154; SSE-NEXT: [[TMP6:%.*]] = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> [[TMP4]], <4 x i32> [[TMP5]])155; SSE-NEXT: store <4 x i32> [[TMP6]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 4), align 4156; SSE-NEXT: [[TMP7:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 8), align 4157; SSE-NEXT: [[TMP8:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 8), align 4158; SSE-NEXT: [[TMP9:%.*]] = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> [[TMP7]], <4 x i32> [[TMP8]])159; SSE-NEXT: store <4 x i32> [[TMP9]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 8), align 4160; SSE-NEXT: [[TMP10:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 12), align 4161; SSE-NEXT: [[TMP11:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 12), align 4162; SSE-NEXT: [[TMP12:%.*]] = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> [[TMP10]], <4 x i32> [[TMP11]])163; SSE-NEXT: store <4 x i32> [[TMP12]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 12), align 4164; SSE-NEXT: ret void165;166; AVX-LABEL: @add_v16i32(167; AVX-NEXT: [[TMP1:%.*]] = load <8 x i32>, ptr @a32, align 4168; AVX-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr @b32, align 4169; AVX-NEXT: [[TMP3:%.*]] = call <8 x i32> @llvm.sadd.sat.v8i32(<8 x i32> [[TMP1]], <8 x i32> [[TMP2]])170; AVX-NEXT: store <8 x i32> [[TMP3]], ptr @c32, align 4171; AVX-NEXT: [[TMP4:%.*]] = load <8 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 8), align 4172; AVX-NEXT: [[TMP5:%.*]] = load <8 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 8), align 4173; AVX-NEXT: [[TMP6:%.*]] = call <8 x i32> @llvm.sadd.sat.v8i32(<8 x i32> [[TMP4]], <8 x i32> [[TMP5]])174; AVX-NEXT: store <8 x i32> [[TMP6]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 8), align 4175; AVX-NEXT: ret void176;177; AVX512-LABEL: @add_v16i32(178; AVX512-NEXT: [[TMP1:%.*]] = load <16 x i32>, ptr @a32, align 4179; AVX512-NEXT: [[TMP2:%.*]] = load <16 x i32>, ptr @b32, align 4180; AVX512-NEXT: [[TMP3:%.*]] = call <16 x i32> @llvm.sadd.sat.v16i32(<16 x i32> [[TMP1]], <16 x i32> [[TMP2]])181; AVX512-NEXT: store <16 x i32> [[TMP3]], ptr @c32, align 4182; AVX512-NEXT: ret void183;184 %a0 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 0 ), align 4185 %a1 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 1 ), align 4186 %a2 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 2 ), align 4187 %a3 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 3 ), align 4188 %a4 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 4 ), align 4189 %a5 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 5 ), align 4190 %a6 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 6 ), align 4191 %a7 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 7 ), align 4192 %a8 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 8 ), align 4193 %a9 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 9 ), align 4194 %a10 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 10), align 4195 %a11 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 11), align 4196 %a12 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 12), align 4197 %a13 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 13), align 4198 %a14 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 14), align 4199 %a15 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 15), align 4200 %b0 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 0 ), align 4201 %b1 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 1 ), align 4202 %b2 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 2 ), align 4203 %b3 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 3 ), align 4204 %b4 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 4 ), align 4205 %b5 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 5 ), align 4206 %b6 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 6 ), align 4207 %b7 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 7 ), align 4208 %b8 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 8 ), align 4209 %b9 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 9 ), align 4210 %b10 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 10), align 4211 %b11 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 11), align 4212 %b12 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 12), align 4213 %b13 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 13), align 4214 %b14 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 14), align 4215 %b15 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 15), align 4216 %r0 = call i32 @llvm.sadd.sat.i32(i32 %a0 , i32 %b0 )217 %r1 = call i32 @llvm.sadd.sat.i32(i32 %a1 , i32 %b1 )218 %r2 = call i32 @llvm.sadd.sat.i32(i32 %a2 , i32 %b2 )219 %r3 = call i32 @llvm.sadd.sat.i32(i32 %a3 , i32 %b3 )220 %r4 = call i32 @llvm.sadd.sat.i32(i32 %a4 , i32 %b4 )221 %r5 = call i32 @llvm.sadd.sat.i32(i32 %a5 , i32 %b5 )222 %r6 = call i32 @llvm.sadd.sat.i32(i32 %a6 , i32 %b6 )223 %r7 = call i32 @llvm.sadd.sat.i32(i32 %a7 , i32 %b7 )224 %r8 = call i32 @llvm.sadd.sat.i32(i32 %a8 , i32 %b8 )225 %r9 = call i32 @llvm.sadd.sat.i32(i32 %a9 , i32 %b9 )226 %r10 = call i32 @llvm.sadd.sat.i32(i32 %a10, i32 %b10)227 %r11 = call i32 @llvm.sadd.sat.i32(i32 %a11, i32 %b11)228 %r12 = call i32 @llvm.sadd.sat.i32(i32 %a12, i32 %b12)229 %r13 = call i32 @llvm.sadd.sat.i32(i32 %a13, i32 %b13)230 %r14 = call i32 @llvm.sadd.sat.i32(i32 %a14, i32 %b14)231 %r15 = call i32 @llvm.sadd.sat.i32(i32 %a15, i32 %b15)232 store i32 %r0 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 0 ), align 4233 store i32 %r1 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 1 ), align 4234 store i32 %r2 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 2 ), align 4235 store i32 %r3 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 3 ), align 4236 store i32 %r4 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 4 ), align 4237 store i32 %r5 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 5 ), align 4238 store i32 %r6 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 6 ), align 4239 store i32 %r7 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 7 ), align 4240 store i32 %r8 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 8 ), align 4241 store i32 %r9 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 9 ), align 4242 store i32 %r10, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 10), align 4243 store i32 %r11, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 11), align 4244 store i32 %r12, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 12), align 4245 store i32 %r13, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 13), align 4246 store i32 %r14, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 14), align 4247 store i32 %r15, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 15), align 4248 ret void249}250 251define void @add_v32i16() {252; SSE-LABEL: @add_v32i16(253; SSE-NEXT: [[TMP1:%.*]] = load <8 x i16>, ptr @a16, align 2254; SSE-NEXT: [[TMP2:%.*]] = load <8 x i16>, ptr @b16, align 2255; SSE-NEXT: [[TMP3:%.*]] = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> [[TMP1]], <8 x i16> [[TMP2]])256; SSE-NEXT: store <8 x i16> [[TMP3]], ptr @c16, align 2257; SSE-NEXT: [[TMP4:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 8), align 2258; SSE-NEXT: [[TMP5:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 8), align 2259; SSE-NEXT: [[TMP6:%.*]] = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> [[TMP4]], <8 x i16> [[TMP5]])260; SSE-NEXT: store <8 x i16> [[TMP6]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 8), align 2261; SSE-NEXT: [[TMP7:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 16), align 2262; SSE-NEXT: [[TMP8:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 16), align 2263; SSE-NEXT: [[TMP9:%.*]] = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> [[TMP7]], <8 x i16> [[TMP8]])264; SSE-NEXT: store <8 x i16> [[TMP9]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 16), align 2265; SSE-NEXT: [[TMP10:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 24), align 2266; SSE-NEXT: [[TMP11:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 24), align 2267; SSE-NEXT: [[TMP12:%.*]] = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> [[TMP10]], <8 x i16> [[TMP11]])268; SSE-NEXT: store <8 x i16> [[TMP12]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 24), align 2269; SSE-NEXT: ret void270;271; AVX-LABEL: @add_v32i16(272; AVX-NEXT: [[TMP1:%.*]] = load <16 x i16>, ptr @a16, align 2273; AVX-NEXT: [[TMP2:%.*]] = load <16 x i16>, ptr @b16, align 2274; AVX-NEXT: [[TMP3:%.*]] = call <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16> [[TMP1]], <16 x i16> [[TMP2]])275; AVX-NEXT: store <16 x i16> [[TMP3]], ptr @c16, align 2276; AVX-NEXT: [[TMP4:%.*]] = load <16 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 16), align 2277; AVX-NEXT: [[TMP5:%.*]] = load <16 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 16), align 2278; AVX-NEXT: [[TMP6:%.*]] = call <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16> [[TMP4]], <16 x i16> [[TMP5]])279; AVX-NEXT: store <16 x i16> [[TMP6]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 16), align 2280; AVX-NEXT: ret void281;282; AVX512-LABEL: @add_v32i16(283; AVX512-NEXT: [[TMP1:%.*]] = load <32 x i16>, ptr @a16, align 2284; AVX512-NEXT: [[TMP2:%.*]] = load <32 x i16>, ptr @b16, align 2285; AVX512-NEXT: [[TMP3:%.*]] = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> [[TMP1]], <32 x i16> [[TMP2]])286; AVX512-NEXT: store <32 x i16> [[TMP3]], ptr @c16, align 2287; AVX512-NEXT: ret void288;289 %a0 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 0 ), align 2290 %a1 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 1 ), align 2291 %a2 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 2 ), align 2292 %a3 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 3 ), align 2293 %a4 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 4 ), align 2294 %a5 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 5 ), align 2295 %a6 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 6 ), align 2296 %a7 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 7 ), align 2297 %a8 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 8 ), align 2298 %a9 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 9 ), align 2299 %a10 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 10), align 2300 %a11 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 11), align 2301 %a12 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 12), align 2302 %a13 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 13), align 2303 %a14 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 14), align 2304 %a15 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 15), align 2305 %a16 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 16), align 2306 %a17 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 17), align 2307 %a18 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 18), align 2308 %a19 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 19), align 2309 %a20 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 20), align 2310 %a21 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 21), align 2311 %a22 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 22), align 2312 %a23 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 23), align 2313 %a24 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 24), align 2314 %a25 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 25), align 2315 %a26 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 26), align 2316 %a27 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 27), align 2317 %a28 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 28), align 2318 %a29 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 29), align 2319 %a30 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 30), align 2320 %a31 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 31), align 2321 %b0 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 0 ), align 2322 %b1 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 1 ), align 2323 %b2 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 2 ), align 2324 %b3 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 3 ), align 2325 %b4 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 4 ), align 2326 %b5 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 5 ), align 2327 %b6 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 6 ), align 2328 %b7 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 7 ), align 2329 %b8 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 8 ), align 2330 %b9 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 9 ), align 2331 %b10 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 10), align 2332 %b11 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 11), align 2333 %b12 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 12), align 2334 %b13 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 13), align 2335 %b14 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 14), align 2336 %b15 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 15), align 2337 %b16 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 16), align 2338 %b17 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 17), align 2339 %b18 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 18), align 2340 %b19 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 19), align 2341 %b20 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 20), align 2342 %b21 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 21), align 2343 %b22 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 22), align 2344 %b23 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 23), align 2345 %b24 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 24), align 2346 %b25 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 25), align 2347 %b26 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 26), align 2348 %b27 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 27), align 2349 %b28 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 28), align 2350 %b29 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 29), align 2351 %b30 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 30), align 2352 %b31 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 31), align 2353 %r0 = call i16 @llvm.sadd.sat.i16(i16 %a0 , i16 %b0 )354 %r1 = call i16 @llvm.sadd.sat.i16(i16 %a1 , i16 %b1 )355 %r2 = call i16 @llvm.sadd.sat.i16(i16 %a2 , i16 %b2 )356 %r3 = call i16 @llvm.sadd.sat.i16(i16 %a3 , i16 %b3 )357 %r4 = call i16 @llvm.sadd.sat.i16(i16 %a4 , i16 %b4 )358 %r5 = call i16 @llvm.sadd.sat.i16(i16 %a5 , i16 %b5 )359 %r6 = call i16 @llvm.sadd.sat.i16(i16 %a6 , i16 %b6 )360 %r7 = call i16 @llvm.sadd.sat.i16(i16 %a7 , i16 %b7 )361 %r8 = call i16 @llvm.sadd.sat.i16(i16 %a8 , i16 %b8 )362 %r9 = call i16 @llvm.sadd.sat.i16(i16 %a9 , i16 %b9 )363 %r10 = call i16 @llvm.sadd.sat.i16(i16 %a10, i16 %b10)364 %r11 = call i16 @llvm.sadd.sat.i16(i16 %a11, i16 %b11)365 %r12 = call i16 @llvm.sadd.sat.i16(i16 %a12, i16 %b12)366 %r13 = call i16 @llvm.sadd.sat.i16(i16 %a13, i16 %b13)367 %r14 = call i16 @llvm.sadd.sat.i16(i16 %a14, i16 %b14)368 %r15 = call i16 @llvm.sadd.sat.i16(i16 %a15, i16 %b15)369 %r16 = call i16 @llvm.sadd.sat.i16(i16 %a16, i16 %b16)370 %r17 = call i16 @llvm.sadd.sat.i16(i16 %a17, i16 %b17)371 %r18 = call i16 @llvm.sadd.sat.i16(i16 %a18, i16 %b18)372 %r19 = call i16 @llvm.sadd.sat.i16(i16 %a19, i16 %b19)373 %r20 = call i16 @llvm.sadd.sat.i16(i16 %a20, i16 %b20)374 %r21 = call i16 @llvm.sadd.sat.i16(i16 %a21, i16 %b21)375 %r22 = call i16 @llvm.sadd.sat.i16(i16 %a22, i16 %b22)376 %r23 = call i16 @llvm.sadd.sat.i16(i16 %a23, i16 %b23)377 %r24 = call i16 @llvm.sadd.sat.i16(i16 %a24, i16 %b24)378 %r25 = call i16 @llvm.sadd.sat.i16(i16 %a25, i16 %b25)379 %r26 = call i16 @llvm.sadd.sat.i16(i16 %a26, i16 %b26)380 %r27 = call i16 @llvm.sadd.sat.i16(i16 %a27, i16 %b27)381 %r28 = call i16 @llvm.sadd.sat.i16(i16 %a28, i16 %b28)382 %r29 = call i16 @llvm.sadd.sat.i16(i16 %a29, i16 %b29)383 %r30 = call i16 @llvm.sadd.sat.i16(i16 %a30, i16 %b30)384 %r31 = call i16 @llvm.sadd.sat.i16(i16 %a31, i16 %b31)385 store i16 %r0 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 0 ), align 2386 store i16 %r1 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 1 ), align 2387 store i16 %r2 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 2 ), align 2388 store i16 %r3 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 3 ), align 2389 store i16 %r4 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 4 ), align 2390 store i16 %r5 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 5 ), align 2391 store i16 %r6 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 6 ), align 2392 store i16 %r7 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 7 ), align 2393 store i16 %r8 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 8 ), align 2394 store i16 %r9 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 9 ), align 2395 store i16 %r10, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 10), align 2396 store i16 %r11, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 11), align 2397 store i16 %r12, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 12), align 2398 store i16 %r13, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 13), align 2399 store i16 %r14, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 14), align 2400 store i16 %r15, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 15), align 2401 store i16 %r16, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 16), align 2402 store i16 %r17, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 17), align 2403 store i16 %r18, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 18), align 2404 store i16 %r19, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 19), align 2405 store i16 %r20, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 20), align 2406 store i16 %r21, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 21), align 2407 store i16 %r22, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 22), align 2408 store i16 %r23, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 23), align 2409 store i16 %r24, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 24), align 2410 store i16 %r25, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 25), align 2411 store i16 %r26, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 26), align 2412 store i16 %r27, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 27), align 2413 store i16 %r28, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 28), align 2414 store i16 %r29, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 29), align 2415 store i16 %r30, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 30), align 2416 store i16 %r31, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 31), align 2417 ret void418}419 420define void @add_v64i8() {421; SSE-LABEL: @add_v64i8(422; SSE-NEXT: [[TMP1:%.*]] = load <16 x i8>, ptr @a8, align 1423; SSE-NEXT: [[TMP2:%.*]] = load <16 x i8>, ptr @b8, align 1424; SSE-NEXT: [[TMP3:%.*]] = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> [[TMP1]], <16 x i8> [[TMP2]])425; SSE-NEXT: store <16 x i8> [[TMP3]], ptr @c8, align 1426; SSE-NEXT: [[TMP4:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 16), align 1427; SSE-NEXT: [[TMP5:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 16), align 1428; SSE-NEXT: [[TMP6:%.*]] = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> [[TMP4]], <16 x i8> [[TMP5]])429; SSE-NEXT: store <16 x i8> [[TMP6]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 16), align 1430; SSE-NEXT: [[TMP7:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 32), align 1431; SSE-NEXT: [[TMP8:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 32), align 1432; SSE-NEXT: [[TMP9:%.*]] = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> [[TMP7]], <16 x i8> [[TMP8]])433; SSE-NEXT: store <16 x i8> [[TMP9]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 32), align 1434; SSE-NEXT: [[TMP10:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 48), align 1435; SSE-NEXT: [[TMP11:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 48), align 1436; SSE-NEXT: [[TMP12:%.*]] = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> [[TMP10]], <16 x i8> [[TMP11]])437; SSE-NEXT: store <16 x i8> [[TMP12]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 48), align 1438; SSE-NEXT: ret void439;440; AVX-LABEL: @add_v64i8(441; AVX-NEXT: [[TMP1:%.*]] = load <32 x i8>, ptr @a8, align 1442; AVX-NEXT: [[TMP2:%.*]] = load <32 x i8>, ptr @b8, align 1443; AVX-NEXT: [[TMP3:%.*]] = call <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8> [[TMP1]], <32 x i8> [[TMP2]])444; AVX-NEXT: store <32 x i8> [[TMP3]], ptr @c8, align 1445; AVX-NEXT: [[TMP4:%.*]] = load <32 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 32), align 1446; AVX-NEXT: [[TMP5:%.*]] = load <32 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 32), align 1447; AVX-NEXT: [[TMP6:%.*]] = call <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8> [[TMP4]], <32 x i8> [[TMP5]])448; AVX-NEXT: store <32 x i8> [[TMP6]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 32), align 1449; AVX-NEXT: ret void450;451; AVX512-LABEL: @add_v64i8(452; AVX512-NEXT: [[TMP1:%.*]] = load <64 x i8>, ptr @a8, align 1453; AVX512-NEXT: [[TMP2:%.*]] = load <64 x i8>, ptr @b8, align 1454; AVX512-NEXT: [[TMP3:%.*]] = call <64 x i8> @llvm.sadd.sat.v64i8(<64 x i8> [[TMP1]], <64 x i8> [[TMP2]])455; AVX512-NEXT: store <64 x i8> [[TMP3]], ptr @c8, align 1456; AVX512-NEXT: ret void457;458 %a0 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 0 ), align 1459 %a1 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 1 ), align 1460 %a2 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 2 ), align 1461 %a3 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 3 ), align 1462 %a4 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 4 ), align 1463 %a5 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 5 ), align 1464 %a6 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 6 ), align 1465 %a7 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 7 ), align 1466 %a8 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 8 ), align 1467 %a9 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 9 ), align 1468 %a10 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 10), align 1469 %a11 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 11), align 1470 %a12 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 12), align 1471 %a13 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 13), align 1472 %a14 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 14), align 1473 %a15 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 15), align 1474 %a16 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 16), align 1475 %a17 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 17), align 1476 %a18 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 18), align 1477 %a19 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 19), align 1478 %a20 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 20), align 1479 %a21 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 21), align 1480 %a22 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 22), align 1481 %a23 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 23), align 1482 %a24 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 24), align 1483 %a25 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 25), align 1484 %a26 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 26), align 1485 %a27 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 27), align 1486 %a28 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 28), align 1487 %a29 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 29), align 1488 %a30 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 30), align 1489 %a31 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 31), align 1490 %a32 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 32), align 1491 %a33 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 33), align 1492 %a34 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 34), align 1493 %a35 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 35), align 1494 %a36 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 36), align 1495 %a37 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 37), align 1496 %a38 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 38), align 1497 %a39 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 39), align 1498 %a40 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 40), align 1499 %a41 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 41), align 1500 %a42 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 42), align 1501 %a43 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 43), align 1502 %a44 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 44), align 1503 %a45 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 45), align 1504 %a46 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 46), align 1505 %a47 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 47), align 1506 %a48 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 48), align 1507 %a49 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 49), align 1508 %a50 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 50), align 1509 %a51 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 51), align 1510 %a52 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 52), align 1511 %a53 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 53), align 1512 %a54 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 54), align 1513 %a55 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 55), align 1514 %a56 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 56), align 1515 %a57 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 57), align 1516 %a58 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 58), align 1517 %a59 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 59), align 1518 %a60 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 60), align 1519 %a61 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 61), align 1520 %a62 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 62), align 1521 %a63 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 63), align 1522 %b0 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 0 ), align 1523 %b1 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 1 ), align 1524 %b2 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 2 ), align 1525 %b3 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 3 ), align 1526 %b4 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 4 ), align 1527 %b5 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 5 ), align 1528 %b6 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 6 ), align 1529 %b7 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 7 ), align 1530 %b8 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 8 ), align 1531 %b9 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 9 ), align 1532 %b10 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 10), align 1533 %b11 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 11), align 1534 %b12 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 12), align 1535 %b13 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 13), align 1536 %b14 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 14), align 1537 %b15 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 15), align 1538 %b16 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 16), align 1539 %b17 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 17), align 1540 %b18 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 18), align 1541 %b19 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 19), align 1542 %b20 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 20), align 1543 %b21 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 21), align 1544 %b22 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 22), align 1545 %b23 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 23), align 1546 %b24 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 24), align 1547 %b25 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 25), align 1548 %b26 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 26), align 1549 %b27 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 27), align 1550 %b28 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 28), align 1551 %b29 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 29), align 1552 %b30 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 30), align 1553 %b31 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 31), align 1554 %b32 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 32), align 1555 %b33 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 33), align 1556 %b34 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 34), align 1557 %b35 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 35), align 1558 %b36 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 36), align 1559 %b37 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 37), align 1560 %b38 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 38), align 1561 %b39 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 39), align 1562 %b40 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 40), align 1563 %b41 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 41), align 1564 %b42 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 42), align 1565 %b43 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 43), align 1566 %b44 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 44), align 1567 %b45 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 45), align 1568 %b46 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 46), align 1569 %b47 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 47), align 1570 %b48 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 48), align 1571 %b49 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 49), align 1572 %b50 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 50), align 1573 %b51 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 51), align 1574 %b52 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 52), align 1575 %b53 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 53), align 1576 %b54 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 54), align 1577 %b55 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 55), align 1578 %b56 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 56), align 1579 %b57 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 57), align 1580 %b58 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 58), align 1581 %b59 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 59), align 1582 %b60 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 60), align 1583 %b61 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 61), align 1584 %b62 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 62), align 1585 %b63 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 63), align 1586 %r0 = call i8 @llvm.sadd.sat.i8(i8 %a0 , i8 %b0 )587 %r1 = call i8 @llvm.sadd.sat.i8(i8 %a1 , i8 %b1 )588 %r2 = call i8 @llvm.sadd.sat.i8(i8 %a2 , i8 %b2 )589 %r3 = call i8 @llvm.sadd.sat.i8(i8 %a3 , i8 %b3 )590 %r4 = call i8 @llvm.sadd.sat.i8(i8 %a4 , i8 %b4 )591 %r5 = call i8 @llvm.sadd.sat.i8(i8 %a5 , i8 %b5 )592 %r6 = call i8 @llvm.sadd.sat.i8(i8 %a6 , i8 %b6 )593 %r7 = call i8 @llvm.sadd.sat.i8(i8 %a7 , i8 %b7 )594 %r8 = call i8 @llvm.sadd.sat.i8(i8 %a8 , i8 %b8 )595 %r9 = call i8 @llvm.sadd.sat.i8(i8 %a9 , i8 %b9 )596 %r10 = call i8 @llvm.sadd.sat.i8(i8 %a10, i8 %b10)597 %r11 = call i8 @llvm.sadd.sat.i8(i8 %a11, i8 %b11)598 %r12 = call i8 @llvm.sadd.sat.i8(i8 %a12, i8 %b12)599 %r13 = call i8 @llvm.sadd.sat.i8(i8 %a13, i8 %b13)600 %r14 = call i8 @llvm.sadd.sat.i8(i8 %a14, i8 %b14)601 %r15 = call i8 @llvm.sadd.sat.i8(i8 %a15, i8 %b15)602 %r16 = call i8 @llvm.sadd.sat.i8(i8 %a16, i8 %b16)603 %r17 = call i8 @llvm.sadd.sat.i8(i8 %a17, i8 %b17)604 %r18 = call i8 @llvm.sadd.sat.i8(i8 %a18, i8 %b18)605 %r19 = call i8 @llvm.sadd.sat.i8(i8 %a19, i8 %b19)606 %r20 = call i8 @llvm.sadd.sat.i8(i8 %a20, i8 %b20)607 %r21 = call i8 @llvm.sadd.sat.i8(i8 %a21, i8 %b21)608 %r22 = call i8 @llvm.sadd.sat.i8(i8 %a22, i8 %b22)609 %r23 = call i8 @llvm.sadd.sat.i8(i8 %a23, i8 %b23)610 %r24 = call i8 @llvm.sadd.sat.i8(i8 %a24, i8 %b24)611 %r25 = call i8 @llvm.sadd.sat.i8(i8 %a25, i8 %b25)612 %r26 = call i8 @llvm.sadd.sat.i8(i8 %a26, i8 %b26)613 %r27 = call i8 @llvm.sadd.sat.i8(i8 %a27, i8 %b27)614 %r28 = call i8 @llvm.sadd.sat.i8(i8 %a28, i8 %b28)615 %r29 = call i8 @llvm.sadd.sat.i8(i8 %a29, i8 %b29)616 %r30 = call i8 @llvm.sadd.sat.i8(i8 %a30, i8 %b30)617 %r31 = call i8 @llvm.sadd.sat.i8(i8 %a31, i8 %b31)618 %r32 = call i8 @llvm.sadd.sat.i8(i8 %a32, i8 %b32)619 %r33 = call i8 @llvm.sadd.sat.i8(i8 %a33, i8 %b33)620 %r34 = call i8 @llvm.sadd.sat.i8(i8 %a34, i8 %b34)621 %r35 = call i8 @llvm.sadd.sat.i8(i8 %a35, i8 %b35)622 %r36 = call i8 @llvm.sadd.sat.i8(i8 %a36, i8 %b36)623 %r37 = call i8 @llvm.sadd.sat.i8(i8 %a37, i8 %b37)624 %r38 = call i8 @llvm.sadd.sat.i8(i8 %a38, i8 %b38)625 %r39 = call i8 @llvm.sadd.sat.i8(i8 %a39, i8 %b39)626 %r40 = call i8 @llvm.sadd.sat.i8(i8 %a40, i8 %b40)627 %r41 = call i8 @llvm.sadd.sat.i8(i8 %a41, i8 %b41)628 %r42 = call i8 @llvm.sadd.sat.i8(i8 %a42, i8 %b42)629 %r43 = call i8 @llvm.sadd.sat.i8(i8 %a43, i8 %b43)630 %r44 = call i8 @llvm.sadd.sat.i8(i8 %a44, i8 %b44)631 %r45 = call i8 @llvm.sadd.sat.i8(i8 %a45, i8 %b45)632 %r46 = call i8 @llvm.sadd.sat.i8(i8 %a46, i8 %b46)633 %r47 = call i8 @llvm.sadd.sat.i8(i8 %a47, i8 %b47)634 %r48 = call i8 @llvm.sadd.sat.i8(i8 %a48, i8 %b48)635 %r49 = call i8 @llvm.sadd.sat.i8(i8 %a49, i8 %b49)636 %r50 = call i8 @llvm.sadd.sat.i8(i8 %a50, i8 %b50)637 %r51 = call i8 @llvm.sadd.sat.i8(i8 %a51, i8 %b51)638 %r52 = call i8 @llvm.sadd.sat.i8(i8 %a52, i8 %b52)639 %r53 = call i8 @llvm.sadd.sat.i8(i8 %a53, i8 %b53)640 %r54 = call i8 @llvm.sadd.sat.i8(i8 %a54, i8 %b54)641 %r55 = call i8 @llvm.sadd.sat.i8(i8 %a55, i8 %b55)642 %r56 = call i8 @llvm.sadd.sat.i8(i8 %a56, i8 %b56)643 %r57 = call i8 @llvm.sadd.sat.i8(i8 %a57, i8 %b57)644 %r58 = call i8 @llvm.sadd.sat.i8(i8 %a58, i8 %b58)645 %r59 = call i8 @llvm.sadd.sat.i8(i8 %a59, i8 %b59)646 %r60 = call i8 @llvm.sadd.sat.i8(i8 %a60, i8 %b60)647 %r61 = call i8 @llvm.sadd.sat.i8(i8 %a61, i8 %b61)648 %r62 = call i8 @llvm.sadd.sat.i8(i8 %a62, i8 %b62)649 %r63 = call i8 @llvm.sadd.sat.i8(i8 %a63, i8 %b63)650 store i8 %r0 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 0 ), align 1651 store i8 %r1 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 1 ), align 1652 store i8 %r2 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 2 ), align 1653 store i8 %r3 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 3 ), align 1654 store i8 %r4 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 4 ), align 1655 store i8 %r5 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 5 ), align 1656 store i8 %r6 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 6 ), align 1657 store i8 %r7 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 7 ), align 1658 store i8 %r8 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 8 ), align 1659 store i8 %r9 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 9 ), align 1660 store i8 %r10, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 10), align 1661 store i8 %r11, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 11), align 1662 store i8 %r12, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 12), align 1663 store i8 %r13, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 13), align 1664 store i8 %r14, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 14), align 1665 store i8 %r15, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 15), align 1666 store i8 %r16, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 16), align 1667 store i8 %r17, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 17), align 1668 store i8 %r18, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 18), align 1669 store i8 %r19, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 19), align 1670 store i8 %r20, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 20), align 1671 store i8 %r21, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 21), align 1672 store i8 %r22, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 22), align 1673 store i8 %r23, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 23), align 1674 store i8 %r24, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 24), align 1675 store i8 %r25, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 25), align 1676 store i8 %r26, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 26), align 1677 store i8 %r27, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 27), align 1678 store i8 %r28, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 28), align 1679 store i8 %r29, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 29), align 1680 store i8 %r30, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 30), align 1681 store i8 %r31, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 31), align 1682 store i8 %r32, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 32), align 1683 store i8 %r33, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 33), align 1684 store i8 %r34, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 34), align 1685 store i8 %r35, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 35), align 1686 store i8 %r36, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 36), align 1687 store i8 %r37, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 37), align 1688 store i8 %r38, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 38), align 1689 store i8 %r39, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 39), align 1690 store i8 %r40, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 40), align 1691 store i8 %r41, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 41), align 1692 store i8 %r42, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 42), align 1693 store i8 %r43, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 43), align 1694 store i8 %r44, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 44), align 1695 store i8 %r45, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 45), align 1696 store i8 %r46, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 46), align 1697 store i8 %r47, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 47), align 1698 store i8 %r48, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 48), align 1699 store i8 %r49, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 49), align 1700 store i8 %r50, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 50), align 1701 store i8 %r51, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 51), align 1702 store i8 %r52, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 52), align 1703 store i8 %r53, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 53), align 1704 store i8 %r54, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 54), align 1705 store i8 %r55, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 55), align 1706 store i8 %r56, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 56), align 1707 store i8 %r57, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 57), align 1708 store i8 %r58, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 58), align 1709 store i8 %r59, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 59), align 1710 store i8 %r60, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 60), align 1711 store i8 %r61, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 61), align 1712 store i8 %r62, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 62), align 1713 store i8 %r63, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 63), align 1714 ret void715}716;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:717; AVX1: {{.*}}718; AVX2: {{.*}}719