brintos

brintos / llvm-project-archived public Read only

0
0
Text · 54.5 KiB · 7777bf4 Raw
706 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=AVX5; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -passes=slp-vectorizer -S | FileCheck %s --check-prefixes=AVX6; 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.usub.sat.i64(i64, i64)24declare i32 @llvm.usub.sat.i32(i32, i32)25declare i16 @llvm.usub.sat.i16(i16, i16)26declare i8  @llvm.usub.sat.i8 (i8 , i8 )27 28define void @sub_v8i64() {29; SSE2-LABEL: @sub_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.usub.sat.i64(i64 [[A0]], i64 [[B0]])47; SSE2-NEXT:    [[R1:%.*]] = call i64 @llvm.usub.sat.i64(i64 [[A1]], i64 [[B1]])48; SSE2-NEXT:    [[R2:%.*]] = call i64 @llvm.usub.sat.i64(i64 [[A2]], i64 [[B2]])49; SSE2-NEXT:    [[R3:%.*]] = call i64 @llvm.usub.sat.i64(i64 [[A3]], i64 [[B3]])50; SSE2-NEXT:    [[R4:%.*]] = call i64 @llvm.usub.sat.i64(i64 [[A4]], i64 [[B4]])51; SSE2-NEXT:    [[R5:%.*]] = call i64 @llvm.usub.sat.i64(i64 [[A5]], i64 [[B5]])52; SSE2-NEXT:    [[R6:%.*]] = call i64 @llvm.usub.sat.i64(i64 [[A6]], i64 [[B6]])53; SSE2-NEXT:    [[R7:%.*]] = call i64 @llvm.usub.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: @sub_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.usub.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.usub.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.usub.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.usub.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: @sub_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.usub.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.usub.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: @sub_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.usub.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  %a0 = load i64, ptr @a64, align 8102  %a1 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 1), align 8103  %a2 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 2), align 8104  %a3 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 3), align 8105  %a4 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 4), align 8106  %a5 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 5), align 8107  %a6 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 6), align 8108  %a7 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 7), align 8109  %b0 = load i64, ptr @b64, align 8110  %b1 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 1), align 8111  %b2 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 2), align 8112  %b3 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 3), align 8113  %b4 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 4), align 8114  %b5 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 5), align 8115  %b6 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 6), align 8116  %b7 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 7), align 8117  %r0 = call i64 @llvm.usub.sat.i64(i64 %a0, i64 %b0)118  %r1 = call i64 @llvm.usub.sat.i64(i64 %a1, i64 %b1)119  %r2 = call i64 @llvm.usub.sat.i64(i64 %a2, i64 %b2)120  %r3 = call i64 @llvm.usub.sat.i64(i64 %a3, i64 %b3)121  %r4 = call i64 @llvm.usub.sat.i64(i64 %a4, i64 %b4)122  %r5 = call i64 @llvm.usub.sat.i64(i64 %a5, i64 %b5)123  %r6 = call i64 @llvm.usub.sat.i64(i64 %a6, i64 %b6)124  %r7 = call i64 @llvm.usub.sat.i64(i64 %a7, i64 %b7)125  store i64 %r0, ptr @c64, align 8126  store i64 %r1, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 1), align 8127  store i64 %r2, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 2), align 8128  store i64 %r3, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 3), align 8129  store i64 %r4, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 4), align 8130  store i64 %r5, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 5), align 8131  store i64 %r6, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 6), align 8132  store i64 %r7, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 7), align 8133  ret void134}135 136define void @sub_v16i32() {137; SSE-LABEL: @sub_v16i32(138; SSE-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr @a32, align 4139; SSE-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr @b32, align 4140; SSE-NEXT:    [[TMP3:%.*]] = call <4 x i32> @llvm.usub.sat.v4i32(<4 x i32> [[TMP1]], <4 x i32> [[TMP2]])141; SSE-NEXT:    store <4 x i32> [[TMP3]], ptr @c32, align 4142; SSE-NEXT:    [[TMP4:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 4), align 4143; SSE-NEXT:    [[TMP5:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 4), align 4144; SSE-NEXT:    [[TMP6:%.*]] = call <4 x i32> @llvm.usub.sat.v4i32(<4 x i32> [[TMP4]], <4 x i32> [[TMP5]])145; SSE-NEXT:    store <4 x i32> [[TMP6]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 4), align 4146; SSE-NEXT:    [[TMP7:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 8), align 4147; SSE-NEXT:    [[TMP8:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 8), align 4148; SSE-NEXT:    [[TMP9:%.*]] = call <4 x i32> @llvm.usub.sat.v4i32(<4 x i32> [[TMP7]], <4 x i32> [[TMP8]])149; SSE-NEXT:    store <4 x i32> [[TMP9]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 8), align 4150; SSE-NEXT:    [[TMP10:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 12), align 4151; SSE-NEXT:    [[TMP11:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 12), align 4152; SSE-NEXT:    [[TMP12:%.*]] = call <4 x i32> @llvm.usub.sat.v4i32(<4 x i32> [[TMP10]], <4 x i32> [[TMP11]])153; SSE-NEXT:    store <4 x i32> [[TMP12]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 12), align 4154; SSE-NEXT:    ret void155;156; AVX-LABEL: @sub_v16i32(157; AVX-NEXT:    [[TMP1:%.*]] = load <8 x i32>, ptr @a32, align 4158; AVX-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr @b32, align 4159; AVX-NEXT:    [[TMP3:%.*]] = call <8 x i32> @llvm.usub.sat.v8i32(<8 x i32> [[TMP1]], <8 x i32> [[TMP2]])160; AVX-NEXT:    store <8 x i32> [[TMP3]], ptr @c32, align 4161; AVX-NEXT:    [[TMP4:%.*]] = load <8 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 8), align 4162; AVX-NEXT:    [[TMP5:%.*]] = load <8 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 8), align 4163; AVX-NEXT:    [[TMP6:%.*]] = call <8 x i32> @llvm.usub.sat.v8i32(<8 x i32> [[TMP4]], <8 x i32> [[TMP5]])164; AVX-NEXT:    store <8 x i32> [[TMP6]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 8), align 4165; AVX-NEXT:    ret void166;167; AVX512-LABEL: @sub_v16i32(168; AVX512-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @a32, align 4169; AVX512-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @b32, align 4170; AVX512-NEXT:    [[TMP3:%.*]] = call <16 x i32> @llvm.usub.sat.v16i32(<16 x i32> [[TMP1]], <16 x i32> [[TMP2]])171; AVX512-NEXT:    store <16 x i32> [[TMP3]], ptr @c32, align 4172; AVX512-NEXT:    ret void173;174  %a0  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 0 ), align 4175  %a1  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 1 ), align 4176  %a2  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 2 ), align 4177  %a3  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 3 ), align 4178  %a4  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 4 ), align 4179  %a5  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 5 ), align 4180  %a6  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 6 ), align 4181  %a7  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 7 ), align 4182  %a8  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 8 ), align 4183  %a9  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 9 ), align 4184  %a10 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 10), align 4185  %a11 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 11), align 4186  %a12 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 12), align 4187  %a13 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 13), align 4188  %a14 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 14), align 4189  %a15 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 15), align 4190  %b0  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 0 ), align 4191  %b1  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 1 ), align 4192  %b2  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 2 ), align 4193  %b3  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 3 ), align 4194  %b4  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 4 ), align 4195  %b5  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 5 ), align 4196  %b6  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 6 ), align 4197  %b7  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 7 ), align 4198  %b8  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 8 ), align 4199  %b9  = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 9 ), align 4200  %b10 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 10), align 4201  %b11 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 11), align 4202  %b12 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 12), align 4203  %b13 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 13), align 4204  %b14 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 14), align 4205  %b15 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 15), align 4206  %r0  = call i32 @llvm.usub.sat.i32(i32 %a0 , i32 %b0 )207  %r1  = call i32 @llvm.usub.sat.i32(i32 %a1 , i32 %b1 )208  %r2  = call i32 @llvm.usub.sat.i32(i32 %a2 , i32 %b2 )209  %r3  = call i32 @llvm.usub.sat.i32(i32 %a3 , i32 %b3 )210  %r4  = call i32 @llvm.usub.sat.i32(i32 %a4 , i32 %b4 )211  %r5  = call i32 @llvm.usub.sat.i32(i32 %a5 , i32 %b5 )212  %r6  = call i32 @llvm.usub.sat.i32(i32 %a6 , i32 %b6 )213  %r7  = call i32 @llvm.usub.sat.i32(i32 %a7 , i32 %b7 )214  %r8  = call i32 @llvm.usub.sat.i32(i32 %a8 , i32 %b8 )215  %r9  = call i32 @llvm.usub.sat.i32(i32 %a9 , i32 %b9 )216  %r10 = call i32 @llvm.usub.sat.i32(i32 %a10, i32 %b10)217  %r11 = call i32 @llvm.usub.sat.i32(i32 %a11, i32 %b11)218  %r12 = call i32 @llvm.usub.sat.i32(i32 %a12, i32 %b12)219  %r13 = call i32 @llvm.usub.sat.i32(i32 %a13, i32 %b13)220  %r14 = call i32 @llvm.usub.sat.i32(i32 %a14, i32 %b14)221  %r15 = call i32 @llvm.usub.sat.i32(i32 %a15, i32 %b15)222  store i32 %r0 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 0 ), align 4223  store i32 %r1 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 1 ), align 4224  store i32 %r2 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 2 ), align 4225  store i32 %r3 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 3 ), align 4226  store i32 %r4 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 4 ), align 4227  store i32 %r5 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 5 ), align 4228  store i32 %r6 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 6 ), align 4229  store i32 %r7 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 7 ), align 4230  store i32 %r8 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 8 ), align 4231  store i32 %r9 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 9 ), align 4232  store i32 %r10, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 10), align 4233  store i32 %r11, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 11), align 4234  store i32 %r12, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 12), align 4235  store i32 %r13, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 13), align 4236  store i32 %r14, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 14), align 4237  store i32 %r15, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 15), align 4238  ret void239}240 241define void @sub_v32i16() {242; SSE-LABEL: @sub_v32i16(243; SSE-NEXT:    [[TMP1:%.*]] = load <8 x i16>, ptr @a16, align 2244; SSE-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr @b16, align 2245; SSE-NEXT:    [[TMP3:%.*]] = call <8 x i16> @llvm.usub.sat.v8i16(<8 x i16> [[TMP1]], <8 x i16> [[TMP2]])246; SSE-NEXT:    store <8 x i16> [[TMP3]], ptr @c16, align 2247; SSE-NEXT:    [[TMP4:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 8), align 2248; SSE-NEXT:    [[TMP5:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 8), align 2249; SSE-NEXT:    [[TMP6:%.*]] = call <8 x i16> @llvm.usub.sat.v8i16(<8 x i16> [[TMP4]], <8 x i16> [[TMP5]])250; SSE-NEXT:    store <8 x i16> [[TMP6]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 8), align 2251; SSE-NEXT:    [[TMP7:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 16), align 2252; SSE-NEXT:    [[TMP8:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 16), align 2253; SSE-NEXT:    [[TMP9:%.*]] = call <8 x i16> @llvm.usub.sat.v8i16(<8 x i16> [[TMP7]], <8 x i16> [[TMP8]])254; SSE-NEXT:    store <8 x i16> [[TMP9]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 16), align 2255; SSE-NEXT:    [[TMP10:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 24), align 2256; SSE-NEXT:    [[TMP11:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 24), align 2257; SSE-NEXT:    [[TMP12:%.*]] = call <8 x i16> @llvm.usub.sat.v8i16(<8 x i16> [[TMP10]], <8 x i16> [[TMP11]])258; SSE-NEXT:    store <8 x i16> [[TMP12]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 24), align 2259; SSE-NEXT:    ret void260;261; AVX-LABEL: @sub_v32i16(262; AVX-NEXT:    [[TMP1:%.*]] = load <16 x i16>, ptr @a16, align 2263; AVX-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr @b16, align 2264; AVX-NEXT:    [[TMP3:%.*]] = call <16 x i16> @llvm.usub.sat.v16i16(<16 x i16> [[TMP1]], <16 x i16> [[TMP2]])265; AVX-NEXT:    store <16 x i16> [[TMP3]], ptr @c16, align 2266; AVX-NEXT:    [[TMP4:%.*]] = load <16 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 16), align 2267; AVX-NEXT:    [[TMP5:%.*]] = load <16 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 16), align 2268; AVX-NEXT:    [[TMP6:%.*]] = call <16 x i16> @llvm.usub.sat.v16i16(<16 x i16> [[TMP4]], <16 x i16> [[TMP5]])269; AVX-NEXT:    store <16 x i16> [[TMP6]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 16), align 2270; AVX-NEXT:    ret void271;272; AVX512-LABEL: @sub_v32i16(273; AVX512-NEXT:    [[TMP1:%.*]] = load <32 x i16>, ptr @a16, align 2274; AVX512-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr @b16, align 2275; AVX512-NEXT:    [[TMP3:%.*]] = call <32 x i16> @llvm.usub.sat.v32i16(<32 x i16> [[TMP1]], <32 x i16> [[TMP2]])276; AVX512-NEXT:    store <32 x i16> [[TMP3]], ptr @c16, align 2277; AVX512-NEXT:    ret void278;279  %a0  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 0 ), align 2280  %a1  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 1 ), align 2281  %a2  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 2 ), align 2282  %a3  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 3 ), align 2283  %a4  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 4 ), align 2284  %a5  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 5 ), align 2285  %a6  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 6 ), align 2286  %a7  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 7 ), align 2287  %a8  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 8 ), align 2288  %a9  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 9 ), align 2289  %a10 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 10), align 2290  %a11 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 11), align 2291  %a12 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 12), align 2292  %a13 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 13), align 2293  %a14 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 14), align 2294  %a15 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 15), align 2295  %a16 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 16), align 2296  %a17 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 17), align 2297  %a18 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 18), align 2298  %a19 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 19), align 2299  %a20 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 20), align 2300  %a21 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 21), align 2301  %a22 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 22), align 2302  %a23 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 23), align 2303  %a24 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 24), align 2304  %a25 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 25), align 2305  %a26 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 26), align 2306  %a27 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 27), align 2307  %a28 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 28), align 2308  %a29 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 29), align 2309  %a30 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 30), align 2310  %a31 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 31), align 2311  %b0  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 0 ), align 2312  %b1  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 1 ), align 2313  %b2  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 2 ), align 2314  %b3  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 3 ), align 2315  %b4  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 4 ), align 2316  %b5  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 5 ), align 2317  %b6  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 6 ), align 2318  %b7  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 7 ), align 2319  %b8  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 8 ), align 2320  %b9  = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 9 ), align 2321  %b10 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 10), align 2322  %b11 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 11), align 2323  %b12 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 12), align 2324  %b13 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 13), align 2325  %b14 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 14), align 2326  %b15 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 15), align 2327  %b16 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 16), align 2328  %b17 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 17), align 2329  %b18 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 18), align 2330  %b19 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 19), align 2331  %b20 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 20), align 2332  %b21 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 21), align 2333  %b22 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 22), align 2334  %b23 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 23), align 2335  %b24 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 24), align 2336  %b25 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 25), align 2337  %b26 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 26), align 2338  %b27 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 27), align 2339  %b28 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 28), align 2340  %b29 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 29), align 2341  %b30 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 30), align 2342  %b31 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 31), align 2343  %r0  = call i16 @llvm.usub.sat.i16(i16 %a0 , i16 %b0 )344  %r1  = call i16 @llvm.usub.sat.i16(i16 %a1 , i16 %b1 )345  %r2  = call i16 @llvm.usub.sat.i16(i16 %a2 , i16 %b2 )346  %r3  = call i16 @llvm.usub.sat.i16(i16 %a3 , i16 %b3 )347  %r4  = call i16 @llvm.usub.sat.i16(i16 %a4 , i16 %b4 )348  %r5  = call i16 @llvm.usub.sat.i16(i16 %a5 , i16 %b5 )349  %r6  = call i16 @llvm.usub.sat.i16(i16 %a6 , i16 %b6 )350  %r7  = call i16 @llvm.usub.sat.i16(i16 %a7 , i16 %b7 )351  %r8  = call i16 @llvm.usub.sat.i16(i16 %a8 , i16 %b8 )352  %r9  = call i16 @llvm.usub.sat.i16(i16 %a9 , i16 %b9 )353  %r10 = call i16 @llvm.usub.sat.i16(i16 %a10, i16 %b10)354  %r11 = call i16 @llvm.usub.sat.i16(i16 %a11, i16 %b11)355  %r12 = call i16 @llvm.usub.sat.i16(i16 %a12, i16 %b12)356  %r13 = call i16 @llvm.usub.sat.i16(i16 %a13, i16 %b13)357  %r14 = call i16 @llvm.usub.sat.i16(i16 %a14, i16 %b14)358  %r15 = call i16 @llvm.usub.sat.i16(i16 %a15, i16 %b15)359  %r16 = call i16 @llvm.usub.sat.i16(i16 %a16, i16 %b16)360  %r17 = call i16 @llvm.usub.sat.i16(i16 %a17, i16 %b17)361  %r18 = call i16 @llvm.usub.sat.i16(i16 %a18, i16 %b18)362  %r19 = call i16 @llvm.usub.sat.i16(i16 %a19, i16 %b19)363  %r20 = call i16 @llvm.usub.sat.i16(i16 %a20, i16 %b20)364  %r21 = call i16 @llvm.usub.sat.i16(i16 %a21, i16 %b21)365  %r22 = call i16 @llvm.usub.sat.i16(i16 %a22, i16 %b22)366  %r23 = call i16 @llvm.usub.sat.i16(i16 %a23, i16 %b23)367  %r24 = call i16 @llvm.usub.sat.i16(i16 %a24, i16 %b24)368  %r25 = call i16 @llvm.usub.sat.i16(i16 %a25, i16 %b25)369  %r26 = call i16 @llvm.usub.sat.i16(i16 %a26, i16 %b26)370  %r27 = call i16 @llvm.usub.sat.i16(i16 %a27, i16 %b27)371  %r28 = call i16 @llvm.usub.sat.i16(i16 %a28, i16 %b28)372  %r29 = call i16 @llvm.usub.sat.i16(i16 %a29, i16 %b29)373  %r30 = call i16 @llvm.usub.sat.i16(i16 %a30, i16 %b30)374  %r31 = call i16 @llvm.usub.sat.i16(i16 %a31, i16 %b31)375  store i16 %r0 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 0 ), align 2376  store i16 %r1 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 1 ), align 2377  store i16 %r2 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 2 ), align 2378  store i16 %r3 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 3 ), align 2379  store i16 %r4 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 4 ), align 2380  store i16 %r5 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 5 ), align 2381  store i16 %r6 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 6 ), align 2382  store i16 %r7 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 7 ), align 2383  store i16 %r8 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 8 ), align 2384  store i16 %r9 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 9 ), align 2385  store i16 %r10, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 10), align 2386  store i16 %r11, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 11), align 2387  store i16 %r12, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 12), align 2388  store i16 %r13, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 13), align 2389  store i16 %r14, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 14), align 2390  store i16 %r15, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 15), align 2391  store i16 %r16, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 16), align 2392  store i16 %r17, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 17), align 2393  store i16 %r18, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 18), align 2394  store i16 %r19, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 19), align 2395  store i16 %r20, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 20), align 2396  store i16 %r21, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 21), align 2397  store i16 %r22, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 22), align 2398  store i16 %r23, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 23), align 2399  store i16 %r24, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 24), align 2400  store i16 %r25, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 25), align 2401  store i16 %r26, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 26), align 2402  store i16 %r27, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 27), align 2403  store i16 %r28, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 28), align 2404  store i16 %r29, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 29), align 2405  store i16 %r30, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 30), align 2406  store i16 %r31, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 31), align 2407  ret void408}409 410define void @sub_v64i8() {411; SSE-LABEL: @sub_v64i8(412; SSE-NEXT:    [[TMP1:%.*]] = load <16 x i8>, ptr @a8, align 1413; SSE-NEXT:    [[TMP2:%.*]] = load <16 x i8>, ptr @b8, align 1414; SSE-NEXT:    [[TMP3:%.*]] = call <16 x i8> @llvm.usub.sat.v16i8(<16 x i8> [[TMP1]], <16 x i8> [[TMP2]])415; SSE-NEXT:    store <16 x i8> [[TMP3]], ptr @c8, align 1416; SSE-NEXT:    [[TMP4:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 16), align 1417; SSE-NEXT:    [[TMP5:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 16), align 1418; SSE-NEXT:    [[TMP6:%.*]] = call <16 x i8> @llvm.usub.sat.v16i8(<16 x i8> [[TMP4]], <16 x i8> [[TMP5]])419; SSE-NEXT:    store <16 x i8> [[TMP6]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 16), align 1420; SSE-NEXT:    [[TMP7:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 32), align 1421; SSE-NEXT:    [[TMP8:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 32), align 1422; SSE-NEXT:    [[TMP9:%.*]] = call <16 x i8> @llvm.usub.sat.v16i8(<16 x i8> [[TMP7]], <16 x i8> [[TMP8]])423; SSE-NEXT:    store <16 x i8> [[TMP9]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 32), align 1424; SSE-NEXT:    [[TMP10:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 48), align 1425; SSE-NEXT:    [[TMP11:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 48), align 1426; SSE-NEXT:    [[TMP12:%.*]] = call <16 x i8> @llvm.usub.sat.v16i8(<16 x i8> [[TMP10]], <16 x i8> [[TMP11]])427; SSE-NEXT:    store <16 x i8> [[TMP12]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 48), align 1428; SSE-NEXT:    ret void429;430; AVX-LABEL: @sub_v64i8(431; AVX-NEXT:    [[TMP1:%.*]] = load <32 x i8>, ptr @a8, align 1432; AVX-NEXT:    [[TMP2:%.*]] = load <32 x i8>, ptr @b8, align 1433; AVX-NEXT:    [[TMP3:%.*]] = call <32 x i8> @llvm.usub.sat.v32i8(<32 x i8> [[TMP1]], <32 x i8> [[TMP2]])434; AVX-NEXT:    store <32 x i8> [[TMP3]], ptr @c8, align 1435; AVX-NEXT:    [[TMP4:%.*]] = load <32 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 32), align 1436; AVX-NEXT:    [[TMP5:%.*]] = load <32 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 32), align 1437; AVX-NEXT:    [[TMP6:%.*]] = call <32 x i8> @llvm.usub.sat.v32i8(<32 x i8> [[TMP4]], <32 x i8> [[TMP5]])438; AVX-NEXT:    store <32 x i8> [[TMP6]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 32), align 1439; AVX-NEXT:    ret void440;441; AVX512-LABEL: @sub_v64i8(442; AVX512-NEXT:    [[TMP1:%.*]] = load <64 x i8>, ptr @a8, align 1443; AVX512-NEXT:    [[TMP2:%.*]] = load <64 x i8>, ptr @b8, align 1444; AVX512-NEXT:    [[TMP3:%.*]] = call <64 x i8> @llvm.usub.sat.v64i8(<64 x i8> [[TMP1]], <64 x i8> [[TMP2]])445; AVX512-NEXT:    store <64 x i8> [[TMP3]], ptr @c8, align 1446; AVX512-NEXT:    ret void447;448  %a0  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 0 ), align 1449  %a1  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 1 ), align 1450  %a2  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 2 ), align 1451  %a3  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 3 ), align 1452  %a4  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 4 ), align 1453  %a5  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 5 ), align 1454  %a6  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 6 ), align 1455  %a7  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 7 ), align 1456  %a8  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 8 ), align 1457  %a9  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 9 ), align 1458  %a10 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 10), align 1459  %a11 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 11), align 1460  %a12 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 12), align 1461  %a13 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 13), align 1462  %a14 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 14), align 1463  %a15 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 15), align 1464  %a16 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 16), align 1465  %a17 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 17), align 1466  %a18 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 18), align 1467  %a19 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 19), align 1468  %a20 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 20), align 1469  %a21 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 21), align 1470  %a22 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 22), align 1471  %a23 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 23), align 1472  %a24 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 24), align 1473  %a25 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 25), align 1474  %a26 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 26), align 1475  %a27 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 27), align 1476  %a28 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 28), align 1477  %a29 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 29), align 1478  %a30 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 30), align 1479  %a31 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 31), align 1480  %a32 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 32), align 1481  %a33 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 33), align 1482  %a34 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 34), align 1483  %a35 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 35), align 1484  %a36 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 36), align 1485  %a37 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 37), align 1486  %a38 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 38), align 1487  %a39 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 39), align 1488  %a40 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 40), align 1489  %a41 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 41), align 1490  %a42 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 42), align 1491  %a43 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 43), align 1492  %a44 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 44), align 1493  %a45 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 45), align 1494  %a46 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 46), align 1495  %a47 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 47), align 1496  %a48 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 48), align 1497  %a49 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 49), align 1498  %a50 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 50), align 1499  %a51 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 51), align 1500  %a52 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 52), align 1501  %a53 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 53), align 1502  %a54 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 54), align 1503  %a55 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 55), align 1504  %a56 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 56), align 1505  %a57 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 57), align 1506  %a58 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 58), align 1507  %a59 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 59), align 1508  %a60 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 60), align 1509  %a61 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 61), align 1510  %a62 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 62), align 1511  %a63 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 63), align 1512  %b0  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 0 ), align 1513  %b1  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 1 ), align 1514  %b2  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 2 ), align 1515  %b3  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 3 ), align 1516  %b4  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 4 ), align 1517  %b5  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 5 ), align 1518  %b6  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 6 ), align 1519  %b7  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 7 ), align 1520  %b8  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 8 ), align 1521  %b9  = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 9 ), align 1522  %b10 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 10), align 1523  %b11 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 11), align 1524  %b12 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 12), align 1525  %b13 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 13), align 1526  %b14 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 14), align 1527  %b15 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 15), align 1528  %b16 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 16), align 1529  %b17 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 17), align 1530  %b18 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 18), align 1531  %b19 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 19), align 1532  %b20 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 20), align 1533  %b21 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 21), align 1534  %b22 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 22), align 1535  %b23 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 23), align 1536  %b24 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 24), align 1537  %b25 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 25), align 1538  %b26 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 26), align 1539  %b27 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 27), align 1540  %b28 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 28), align 1541  %b29 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 29), align 1542  %b30 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 30), align 1543  %b31 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 31), align 1544  %b32 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 32), align 1545  %b33 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 33), align 1546  %b34 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 34), align 1547  %b35 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 35), align 1548  %b36 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 36), align 1549  %b37 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 37), align 1550  %b38 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 38), align 1551  %b39 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 39), align 1552  %b40 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 40), align 1553  %b41 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 41), align 1554  %b42 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 42), align 1555  %b43 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 43), align 1556  %b44 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 44), align 1557  %b45 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 45), align 1558  %b46 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 46), align 1559  %b47 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 47), align 1560  %b48 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 48), align 1561  %b49 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 49), align 1562  %b50 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 50), align 1563  %b51 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 51), align 1564  %b52 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 52), align 1565  %b53 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 53), align 1566  %b54 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 54), align 1567  %b55 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 55), align 1568  %b56 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 56), align 1569  %b57 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 57), align 1570  %b58 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 58), align 1571  %b59 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 59), align 1572  %b60 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 60), align 1573  %b61 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 61), align 1574  %b62 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 62), align 1575  %b63 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 63), align 1576  %r0  = call i8 @llvm.usub.sat.i8(i8 %a0 , i8 %b0 )577  %r1  = call i8 @llvm.usub.sat.i8(i8 %a1 , i8 %b1 )578  %r2  = call i8 @llvm.usub.sat.i8(i8 %a2 , i8 %b2 )579  %r3  = call i8 @llvm.usub.sat.i8(i8 %a3 , i8 %b3 )580  %r4  = call i8 @llvm.usub.sat.i8(i8 %a4 , i8 %b4 )581  %r5  = call i8 @llvm.usub.sat.i8(i8 %a5 , i8 %b5 )582  %r6  = call i8 @llvm.usub.sat.i8(i8 %a6 , i8 %b6 )583  %r7  = call i8 @llvm.usub.sat.i8(i8 %a7 , i8 %b7 )584  %r8  = call i8 @llvm.usub.sat.i8(i8 %a8 , i8 %b8 )585  %r9  = call i8 @llvm.usub.sat.i8(i8 %a9 , i8 %b9 )586  %r10 = call i8 @llvm.usub.sat.i8(i8 %a10, i8 %b10)587  %r11 = call i8 @llvm.usub.sat.i8(i8 %a11, i8 %b11)588  %r12 = call i8 @llvm.usub.sat.i8(i8 %a12, i8 %b12)589  %r13 = call i8 @llvm.usub.sat.i8(i8 %a13, i8 %b13)590  %r14 = call i8 @llvm.usub.sat.i8(i8 %a14, i8 %b14)591  %r15 = call i8 @llvm.usub.sat.i8(i8 %a15, i8 %b15)592  %r16 = call i8 @llvm.usub.sat.i8(i8 %a16, i8 %b16)593  %r17 = call i8 @llvm.usub.sat.i8(i8 %a17, i8 %b17)594  %r18 = call i8 @llvm.usub.sat.i8(i8 %a18, i8 %b18)595  %r19 = call i8 @llvm.usub.sat.i8(i8 %a19, i8 %b19)596  %r20 = call i8 @llvm.usub.sat.i8(i8 %a20, i8 %b20)597  %r21 = call i8 @llvm.usub.sat.i8(i8 %a21, i8 %b21)598  %r22 = call i8 @llvm.usub.sat.i8(i8 %a22, i8 %b22)599  %r23 = call i8 @llvm.usub.sat.i8(i8 %a23, i8 %b23)600  %r24 = call i8 @llvm.usub.sat.i8(i8 %a24, i8 %b24)601  %r25 = call i8 @llvm.usub.sat.i8(i8 %a25, i8 %b25)602  %r26 = call i8 @llvm.usub.sat.i8(i8 %a26, i8 %b26)603  %r27 = call i8 @llvm.usub.sat.i8(i8 %a27, i8 %b27)604  %r28 = call i8 @llvm.usub.sat.i8(i8 %a28, i8 %b28)605  %r29 = call i8 @llvm.usub.sat.i8(i8 %a29, i8 %b29)606  %r30 = call i8 @llvm.usub.sat.i8(i8 %a30, i8 %b30)607  %r31 = call i8 @llvm.usub.sat.i8(i8 %a31, i8 %b31)608  %r32 = call i8 @llvm.usub.sat.i8(i8 %a32, i8 %b32)609  %r33 = call i8 @llvm.usub.sat.i8(i8 %a33, i8 %b33)610  %r34 = call i8 @llvm.usub.sat.i8(i8 %a34, i8 %b34)611  %r35 = call i8 @llvm.usub.sat.i8(i8 %a35, i8 %b35)612  %r36 = call i8 @llvm.usub.sat.i8(i8 %a36, i8 %b36)613  %r37 = call i8 @llvm.usub.sat.i8(i8 %a37, i8 %b37)614  %r38 = call i8 @llvm.usub.sat.i8(i8 %a38, i8 %b38)615  %r39 = call i8 @llvm.usub.sat.i8(i8 %a39, i8 %b39)616  %r40 = call i8 @llvm.usub.sat.i8(i8 %a40, i8 %b40)617  %r41 = call i8 @llvm.usub.sat.i8(i8 %a41, i8 %b41)618  %r42 = call i8 @llvm.usub.sat.i8(i8 %a42, i8 %b42)619  %r43 = call i8 @llvm.usub.sat.i8(i8 %a43, i8 %b43)620  %r44 = call i8 @llvm.usub.sat.i8(i8 %a44, i8 %b44)621  %r45 = call i8 @llvm.usub.sat.i8(i8 %a45, i8 %b45)622  %r46 = call i8 @llvm.usub.sat.i8(i8 %a46, i8 %b46)623  %r47 = call i8 @llvm.usub.sat.i8(i8 %a47, i8 %b47)624  %r48 = call i8 @llvm.usub.sat.i8(i8 %a48, i8 %b48)625  %r49 = call i8 @llvm.usub.sat.i8(i8 %a49, i8 %b49)626  %r50 = call i8 @llvm.usub.sat.i8(i8 %a50, i8 %b50)627  %r51 = call i8 @llvm.usub.sat.i8(i8 %a51, i8 %b51)628  %r52 = call i8 @llvm.usub.sat.i8(i8 %a52, i8 %b52)629  %r53 = call i8 @llvm.usub.sat.i8(i8 %a53, i8 %b53)630  %r54 = call i8 @llvm.usub.sat.i8(i8 %a54, i8 %b54)631  %r55 = call i8 @llvm.usub.sat.i8(i8 %a55, i8 %b55)632  %r56 = call i8 @llvm.usub.sat.i8(i8 %a56, i8 %b56)633  %r57 = call i8 @llvm.usub.sat.i8(i8 %a57, i8 %b57)634  %r58 = call i8 @llvm.usub.sat.i8(i8 %a58, i8 %b58)635  %r59 = call i8 @llvm.usub.sat.i8(i8 %a59, i8 %b59)636  %r60 = call i8 @llvm.usub.sat.i8(i8 %a60, i8 %b60)637  %r61 = call i8 @llvm.usub.sat.i8(i8 %a61, i8 %b61)638  %r62 = call i8 @llvm.usub.sat.i8(i8 %a62, i8 %b62)639  %r63 = call i8 @llvm.usub.sat.i8(i8 %a63, i8 %b63)640  store i8 %r0 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 0 ), align 1641  store i8 %r1 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 1 ), align 1642  store i8 %r2 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 2 ), align 1643  store i8 %r3 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 3 ), align 1644  store i8 %r4 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 4 ), align 1645  store i8 %r5 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 5 ), align 1646  store i8 %r6 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 6 ), align 1647  store i8 %r7 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 7 ), align 1648  store i8 %r8 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 8 ), align 1649  store i8 %r9 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 9 ), align 1650  store i8 %r10, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 10), align 1651  store i8 %r11, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 11), align 1652  store i8 %r12, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 12), align 1653  store i8 %r13, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 13), align 1654  store i8 %r14, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 14), align 1655  store i8 %r15, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 15), align 1656  store i8 %r16, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 16), align 1657  store i8 %r17, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 17), align 1658  store i8 %r18, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 18), align 1659  store i8 %r19, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 19), align 1660  store i8 %r20, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 20), align 1661  store i8 %r21, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 21), align 1662  store i8 %r22, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 22), align 1663  store i8 %r23, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 23), align 1664  store i8 %r24, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 24), align 1665  store i8 %r25, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 25), align 1666  store i8 %r26, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 26), align 1667  store i8 %r27, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 27), align 1668  store i8 %r28, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 28), align 1669  store i8 %r29, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 29), align 1670  store i8 %r30, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 30), align 1671  store i8 %r31, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 31), align 1672  store i8 %r32, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 32), align 1673  store i8 %r33, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 33), align 1674  store i8 %r34, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 34), align 1675  store i8 %r35, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 35), align 1676  store i8 %r36, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 36), align 1677  store i8 %r37, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 37), align 1678  store i8 %r38, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 38), align 1679  store i8 %r39, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 39), align 1680  store i8 %r40, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 40), align 1681  store i8 %r41, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 41), align 1682  store i8 %r42, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 42), align 1683  store i8 %r43, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 43), align 1684  store i8 %r44, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 44), align 1685  store i8 %r45, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 45), align 1686  store i8 %r46, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 46), align 1687  store i8 %r47, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 47), align 1688  store i8 %r48, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 48), align 1689  store i8 %r49, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 49), align 1690  store i8 %r50, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 50), align 1691  store i8 %r51, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 51), align 1692  store i8 %r52, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 52), align 1693  store i8 %r53, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 53), align 1694  store i8 %r54, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 54), align 1695  store i8 %r55, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 55), align 1696  store i8 %r56, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 56), align 1697  store i8 %r57, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 57), align 1698  store i8 %r58, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 58), align 1699  store i8 %r59, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 59), align 1700  store i8 %r60, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 60), align 1701  store i8 %r61, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 61), align 1702  store i8 %r62, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 62), align 1703  store i8 %r63, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 63), align 1704  ret void705}706