852 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-prefix=SSE3; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=slm -passes=slp-vectorizer -S | FileCheck %s --check-prefix=SLM4; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -mattr=+prefer-128-bit -passes=slp-vectorizer -S | FileCheck %s --check-prefix=AVX1285; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -mattr=-prefer-128-bit -passes=slp-vectorizer -S | FileCheck %s --check-prefix=AVX2566; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -mattr=+prefer-128-bit -passes=slp-vectorizer -S | FileCheck %s --check-prefix=AVX1287; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -mattr=-prefer-128-bit -passes=slp-vectorizer -S | FileCheck %s --check-prefix=AVX2568; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=knl -passes=slp-vectorizer -S | FileCheck %s --check-prefix=AVX5129; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skx -mattr=-prefer-256-bit -passes=slp-vectorizer -S | FileCheck %s --check-prefix=AVX51210; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skx -mattr=+prefer-256-bit -passes=slp-vectorizer -S | FileCheck %s --check-prefix=AVX25611 12@a64 = common global [8 x i64] zeroinitializer, align 6413@b64 = common global [8 x i64] zeroinitializer, align 6414@c64 = common global [8 x i64] zeroinitializer, align 6415@a32 = common global [16 x i32] zeroinitializer, align 6416@b32 = common global [16 x i32] zeroinitializer, align 6417@c32 = common global [16 x i32] zeroinitializer, align 6418@a16 = common global [32 x i16] zeroinitializer, align 6419@b16 = common global [32 x i16] zeroinitializer, align 6420@c16 = common global [32 x i16] zeroinitializer, align 6421@a8 = common global [64 x i8] zeroinitializer, align 6422@b8 = common global [64 x i8] zeroinitializer, align 6423@c8 = common global [64 x i8] zeroinitializer, align 6424 25define void @mul_v8i64() {26; SSE-LABEL: @mul_v8i64(27; SSE-NEXT: [[A0:%.*]] = load i64, ptr @a64, align 828; SSE-NEXT: [[A1:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 1), align 829; SSE-NEXT: [[A2:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 2), align 830; SSE-NEXT: [[A3:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 3), align 831; SSE-NEXT: [[A4:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 4), align 832; SSE-NEXT: [[A5:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 5), align 833; SSE-NEXT: [[A6:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 6), align 834; SSE-NEXT: [[A7:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 7), align 835; SSE-NEXT: [[B0:%.*]] = load i64, ptr @b64, align 836; SSE-NEXT: [[B1:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 1), align 837; SSE-NEXT: [[B2:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 2), align 838; SSE-NEXT: [[B3:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 3), align 839; SSE-NEXT: [[B4:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 4), align 840; SSE-NEXT: [[B5:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 5), align 841; SSE-NEXT: [[B6:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 6), align 842; SSE-NEXT: [[B7:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 7), align 843; SSE-NEXT: [[R0:%.*]] = mul i64 [[A0]], [[B0]]44; SSE-NEXT: [[R1:%.*]] = mul i64 [[A1]], [[B1]]45; SSE-NEXT: [[R2:%.*]] = mul i64 [[A2]], [[B2]]46; SSE-NEXT: [[R3:%.*]] = mul i64 [[A3]], [[B3]]47; SSE-NEXT: [[R4:%.*]] = mul i64 [[A4]], [[B4]]48; SSE-NEXT: [[R5:%.*]] = mul i64 [[A5]], [[B5]]49; SSE-NEXT: [[R6:%.*]] = mul i64 [[A6]], [[B6]]50; SSE-NEXT: [[R7:%.*]] = mul i64 [[A7]], [[B7]]51; SSE-NEXT: store i64 [[R0]], ptr @c64, align 852; SSE-NEXT: store i64 [[R1]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 1), align 853; SSE-NEXT: store i64 [[R2]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 2), align 854; SSE-NEXT: store i64 [[R3]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 3), align 855; SSE-NEXT: store i64 [[R4]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 4), align 856; SSE-NEXT: store i64 [[R5]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 5), align 857; SSE-NEXT: store i64 [[R6]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 6), align 858; SSE-NEXT: store i64 [[R7]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 7), align 859; SSE-NEXT: ret void60;61; SLM-LABEL: @mul_v8i64(62; SLM-NEXT: [[A0:%.*]] = load i64, ptr @a64, align 863; SLM-NEXT: [[A1:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 1), align 864; SLM-NEXT: [[A2:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 2), align 865; SLM-NEXT: [[A3:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 3), align 866; SLM-NEXT: [[A4:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 4), align 867; SLM-NEXT: [[A5:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 5), align 868; SLM-NEXT: [[A6:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 6), align 869; SLM-NEXT: [[A7:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 7), align 870; SLM-NEXT: [[B0:%.*]] = load i64, ptr @b64, align 871; SLM-NEXT: [[B1:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 1), align 872; SLM-NEXT: [[B2:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 2), align 873; SLM-NEXT: [[B3:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 3), align 874; SLM-NEXT: [[B4:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 4), align 875; SLM-NEXT: [[B5:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 5), align 876; SLM-NEXT: [[B6:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 6), align 877; SLM-NEXT: [[B7:%.*]] = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 7), align 878; SLM-NEXT: [[R0:%.*]] = mul i64 [[A0]], [[B0]]79; SLM-NEXT: [[R1:%.*]] = mul i64 [[A1]], [[B1]]80; SLM-NEXT: [[R2:%.*]] = mul i64 [[A2]], [[B2]]81; SLM-NEXT: [[R3:%.*]] = mul i64 [[A3]], [[B3]]82; SLM-NEXT: [[R4:%.*]] = mul i64 [[A4]], [[B4]]83; SLM-NEXT: [[R5:%.*]] = mul i64 [[A5]], [[B5]]84; SLM-NEXT: [[R6:%.*]] = mul i64 [[A6]], [[B6]]85; SLM-NEXT: [[R7:%.*]] = mul i64 [[A7]], [[B7]]86; SLM-NEXT: store i64 [[R0]], ptr @c64, align 887; SLM-NEXT: store i64 [[R1]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 1), align 888; SLM-NEXT: store i64 [[R2]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 2), align 889; SLM-NEXT: store i64 [[R3]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 3), align 890; SLM-NEXT: store i64 [[R4]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 4), align 891; SLM-NEXT: store i64 [[R5]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 5), align 892; SLM-NEXT: store i64 [[R6]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 6), align 893; SLM-NEXT: store i64 [[R7]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 7), align 894; SLM-NEXT: ret void95;96; AVX128-LABEL: @mul_v8i64(97; AVX128-NEXT: [[TMP1:%.*]] = load <2 x i64>, ptr @a64, align 898; AVX128-NEXT: [[TMP2:%.*]] = load <2 x i64>, ptr @b64, align 899; AVX128-NEXT: [[TMP3:%.*]] = mul <2 x i64> [[TMP1]], [[TMP2]]100; AVX128-NEXT: store <2 x i64> [[TMP3]], ptr @c64, align 8101; AVX128-NEXT: [[TMP4:%.*]] = load <2 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 2), align 8102; AVX128-NEXT: [[TMP5:%.*]] = load <2 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 2), align 8103; AVX128-NEXT: [[TMP6:%.*]] = mul <2 x i64> [[TMP4]], [[TMP5]]104; AVX128-NEXT: store <2 x i64> [[TMP6]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 2), align 8105; AVX128-NEXT: [[TMP7:%.*]] = load <2 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 4), align 8106; AVX128-NEXT: [[TMP8:%.*]] = load <2 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 4), align 8107; AVX128-NEXT: [[TMP9:%.*]] = mul <2 x i64> [[TMP7]], [[TMP8]]108; AVX128-NEXT: store <2 x i64> [[TMP9]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 4), align 8109; AVX128-NEXT: [[TMP10:%.*]] = load <2 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 6), align 8110; AVX128-NEXT: [[TMP11:%.*]] = load <2 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 6), align 8111; AVX128-NEXT: [[TMP12:%.*]] = mul <2 x i64> [[TMP10]], [[TMP11]]112; AVX128-NEXT: store <2 x i64> [[TMP12]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 6), align 8113; AVX128-NEXT: ret void114;115; AVX256-LABEL: @mul_v8i64(116; AVX256-NEXT: [[TMP1:%.*]] = load <4 x i64>, ptr @a64, align 8117; AVX256-NEXT: [[TMP2:%.*]] = load <4 x i64>, ptr @b64, align 8118; AVX256-NEXT: [[TMP3:%.*]] = mul <4 x i64> [[TMP1]], [[TMP2]]119; AVX256-NEXT: store <4 x i64> [[TMP3]], ptr @c64, align 8120; AVX256-NEXT: [[TMP4:%.*]] = load <4 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 4), align 8121; AVX256-NEXT: [[TMP5:%.*]] = load <4 x i64>, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 4), align 8122; AVX256-NEXT: [[TMP6:%.*]] = mul <4 x i64> [[TMP4]], [[TMP5]]123; AVX256-NEXT: store <4 x i64> [[TMP6]], ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 4), align 8124; AVX256-NEXT: ret void125;126; AVX512-LABEL: @mul_v8i64(127; AVX512-NEXT: [[TMP1:%.*]] = load <8 x i64>, ptr @a64, align 8128; AVX512-NEXT: [[TMP2:%.*]] = load <8 x i64>, ptr @b64, align 8129; AVX512-NEXT: [[TMP3:%.*]] = mul <8 x i64> [[TMP1]], [[TMP2]]130; AVX512-NEXT: store <8 x i64> [[TMP3]], ptr @c64, align 8131; AVX512-NEXT: ret void132;133 %a0 = load i64, ptr @a64, align 8134 %a1 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 1), align 8135 %a2 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 2), align 8136 %a3 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 3), align 8137 %a4 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 4), align 8138 %a5 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 5), align 8139 %a6 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 6), align 8140 %a7 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @a64, i32 0, i64 7), align 8141 %b0 = load i64, ptr @b64, align 8142 %b1 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 1), align 8143 %b2 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 2), align 8144 %b3 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 3), align 8145 %b4 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 4), align 8146 %b5 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 5), align 8147 %b6 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 6), align 8148 %b7 = load i64, ptr getelementptr inbounds ([8 x i64], ptr @b64, i32 0, i64 7), align 8149 %r0 = mul i64 %a0, %b0150 %r1 = mul i64 %a1, %b1151 %r2 = mul i64 %a2, %b2152 %r3 = mul i64 %a3, %b3153 %r4 = mul i64 %a4, %b4154 %r5 = mul i64 %a5, %b5155 %r6 = mul i64 %a6, %b6156 %r7 = mul i64 %a7, %b7157 store i64 %r0, ptr @c64, align 8158 store i64 %r1, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 1), align 8159 store i64 %r2, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 2), align 8160 store i64 %r3, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 3), align 8161 store i64 %r4, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 4), align 8162 store i64 %r5, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 5), align 8163 store i64 %r6, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 6), align 8164 store i64 %r7, ptr getelementptr inbounds ([8 x i64], ptr @c64, i32 0, i64 7), align 8165 ret void166}167 168define void @mul_v16i32() {169; SSE-LABEL: @mul_v16i32(170; SSE-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @a32, align 4171; SSE-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr @b32, align 4172; SSE-NEXT: [[TMP3:%.*]] = mul <4 x i32> [[TMP1]], [[TMP2]]173; SSE-NEXT: store <4 x i32> [[TMP3]], ptr @c32, align 4174; SSE-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 4), align 4175; SSE-NEXT: [[TMP5:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 4), align 4176; SSE-NEXT: [[TMP6:%.*]] = mul <4 x i32> [[TMP4]], [[TMP5]]177; SSE-NEXT: store <4 x i32> [[TMP6]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 4), align 4178; SSE-NEXT: [[TMP7:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 8), align 4179; SSE-NEXT: [[TMP8:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 8), align 4180; SSE-NEXT: [[TMP9:%.*]] = mul <4 x i32> [[TMP7]], [[TMP8]]181; SSE-NEXT: store <4 x i32> [[TMP9]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 8), align 4182; SSE-NEXT: [[TMP10:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 12), align 4183; SSE-NEXT: [[TMP11:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 12), align 4184; SSE-NEXT: [[TMP12:%.*]] = mul <4 x i32> [[TMP10]], [[TMP11]]185; SSE-NEXT: store <4 x i32> [[TMP12]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 12), align 4186; SSE-NEXT: ret void187;188; SLM-LABEL: @mul_v16i32(189; SLM-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @a32, align 4190; SLM-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr @b32, align 4191; SLM-NEXT: [[TMP3:%.*]] = mul <4 x i32> [[TMP1]], [[TMP2]]192; SLM-NEXT: store <4 x i32> [[TMP3]], ptr @c32, align 4193; SLM-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 4), align 4194; SLM-NEXT: [[TMP5:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 4), align 4195; SLM-NEXT: [[TMP6:%.*]] = mul <4 x i32> [[TMP4]], [[TMP5]]196; SLM-NEXT: store <4 x i32> [[TMP6]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 4), align 4197; SLM-NEXT: [[TMP7:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 8), align 4198; SLM-NEXT: [[TMP8:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 8), align 4199; SLM-NEXT: [[TMP9:%.*]] = mul <4 x i32> [[TMP7]], [[TMP8]]200; SLM-NEXT: store <4 x i32> [[TMP9]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 8), align 4201; SLM-NEXT: [[TMP10:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 12), align 4202; SLM-NEXT: [[TMP11:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 12), align 4203; SLM-NEXT: [[TMP12:%.*]] = mul <4 x i32> [[TMP10]], [[TMP11]]204; SLM-NEXT: store <4 x i32> [[TMP12]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 12), align 4205; SLM-NEXT: ret void206;207; AVX128-LABEL: @mul_v16i32(208; AVX128-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @a32, align 4209; AVX128-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr @b32, align 4210; AVX128-NEXT: [[TMP3:%.*]] = mul <4 x i32> [[TMP1]], [[TMP2]]211; AVX128-NEXT: store <4 x i32> [[TMP3]], ptr @c32, align 4212; AVX128-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 4), align 4213; AVX128-NEXT: [[TMP5:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 4), align 4214; AVX128-NEXT: [[TMP6:%.*]] = mul <4 x i32> [[TMP4]], [[TMP5]]215; AVX128-NEXT: store <4 x i32> [[TMP6]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 4), align 4216; AVX128-NEXT: [[TMP7:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 8), align 4217; AVX128-NEXT: [[TMP8:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 8), align 4218; AVX128-NEXT: [[TMP9:%.*]] = mul <4 x i32> [[TMP7]], [[TMP8]]219; AVX128-NEXT: store <4 x i32> [[TMP9]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 8), align 4220; AVX128-NEXT: [[TMP10:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 12), align 4221; AVX128-NEXT: [[TMP11:%.*]] = load <4 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 12), align 4222; AVX128-NEXT: [[TMP12:%.*]] = mul <4 x i32> [[TMP10]], [[TMP11]]223; AVX128-NEXT: store <4 x i32> [[TMP12]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 12), align 4224; AVX128-NEXT: ret void225;226; AVX256-LABEL: @mul_v16i32(227; AVX256-NEXT: [[TMP1:%.*]] = load <8 x i32>, ptr @a32, align 4228; AVX256-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr @b32, align 4229; AVX256-NEXT: [[TMP3:%.*]] = mul <8 x i32> [[TMP1]], [[TMP2]]230; AVX256-NEXT: store <8 x i32> [[TMP3]], ptr @c32, align 4231; AVX256-NEXT: [[TMP4:%.*]] = load <8 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 8), align 4232; AVX256-NEXT: [[TMP5:%.*]] = load <8 x i32>, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 8), align 4233; AVX256-NEXT: [[TMP6:%.*]] = mul <8 x i32> [[TMP4]], [[TMP5]]234; AVX256-NEXT: store <8 x i32> [[TMP6]], ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 8), align 4235; AVX256-NEXT: ret void236;237; AVX512-LABEL: @mul_v16i32(238; AVX512-NEXT: [[TMP1:%.*]] = load <16 x i32>, ptr @a32, align 4239; AVX512-NEXT: [[TMP2:%.*]] = load <16 x i32>, ptr @b32, align 4240; AVX512-NEXT: [[TMP3:%.*]] = mul <16 x i32> [[TMP1]], [[TMP2]]241; AVX512-NEXT: store <16 x i32> [[TMP3]], ptr @c32, align 4242; AVX512-NEXT: ret void243;244 %a0 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 0 ), align 4245 %a1 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 1 ), align 4246 %a2 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 2 ), align 4247 %a3 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 3 ), align 4248 %a4 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 4 ), align 4249 %a5 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 5 ), align 4250 %a6 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 6 ), align 4251 %a7 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 7 ), align 4252 %a8 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 8 ), align 4253 %a9 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 9 ), align 4254 %a10 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 10), align 4255 %a11 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 11), align 4256 %a12 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 12), align 4257 %a13 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 13), align 4258 %a14 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 14), align 4259 %a15 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @a32, i32 0, i64 15), align 4260 %b0 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 0 ), align 4261 %b1 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 1 ), align 4262 %b2 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 2 ), align 4263 %b3 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 3 ), align 4264 %b4 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 4 ), align 4265 %b5 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 5 ), align 4266 %b6 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 6 ), align 4267 %b7 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 7 ), align 4268 %b8 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 8 ), align 4269 %b9 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 9 ), align 4270 %b10 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 10), align 4271 %b11 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 11), align 4272 %b12 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 12), align 4273 %b13 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 13), align 4274 %b14 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 14), align 4275 %b15 = load i32, ptr getelementptr inbounds ([16 x i32], ptr @b32, i32 0, i64 15), align 4276 %r0 = mul i32 %a0 , %b0277 %r1 = mul i32 %a1 , %b1278 %r2 = mul i32 %a2 , %b2279 %r3 = mul i32 %a3 , %b3280 %r4 = mul i32 %a4 , %b4281 %r5 = mul i32 %a5 , %b5282 %r6 = mul i32 %a6 , %b6283 %r7 = mul i32 %a7 , %b7284 %r8 = mul i32 %a8 , %b8285 %r9 = mul i32 %a9 , %b9286 %r10 = mul i32 %a10, %b10287 %r11 = mul i32 %a11, %b11288 %r12 = mul i32 %a12, %b12289 %r13 = mul i32 %a13, %b13290 %r14 = mul i32 %a14, %b14291 %r15 = mul i32 %a15, %b15292 store i32 %r0 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 0 ), align 4293 store i32 %r1 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 1 ), align 4294 store i32 %r2 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 2 ), align 4295 store i32 %r3 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 3 ), align 4296 store i32 %r4 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 4 ), align 4297 store i32 %r5 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 5 ), align 4298 store i32 %r6 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 6 ), align 4299 store i32 %r7 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 7 ), align 4300 store i32 %r8 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 8 ), align 4301 store i32 %r9 , ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 9 ), align 4302 store i32 %r10, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 10), align 4303 store i32 %r11, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 11), align 4304 store i32 %r12, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 12), align 4305 store i32 %r13, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 13), align 4306 store i32 %r14, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 14), align 4307 store i32 %r15, ptr getelementptr inbounds ([16 x i32], ptr @c32, i32 0, i64 15), align 4308 ret void309}310 311define void @mul_v32i16() {312; SSE-LABEL: @mul_v32i16(313; SSE-NEXT: [[TMP1:%.*]] = load <8 x i16>, ptr @a16, align 2314; SSE-NEXT: [[TMP2:%.*]] = load <8 x i16>, ptr @b16, align 2315; SSE-NEXT: [[TMP3:%.*]] = mul <8 x i16> [[TMP1]], [[TMP2]]316; SSE-NEXT: store <8 x i16> [[TMP3]], ptr @c16, align 2317; SSE-NEXT: [[TMP4:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 8), align 2318; SSE-NEXT: [[TMP5:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 8), align 2319; SSE-NEXT: [[TMP6:%.*]] = mul <8 x i16> [[TMP4]], [[TMP5]]320; SSE-NEXT: store <8 x i16> [[TMP6]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 8), align 2321; SSE-NEXT: [[TMP7:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 16), align 2322; SSE-NEXT: [[TMP8:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 16), align 2323; SSE-NEXT: [[TMP9:%.*]] = mul <8 x i16> [[TMP7]], [[TMP8]]324; SSE-NEXT: store <8 x i16> [[TMP9]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 16), align 2325; SSE-NEXT: [[TMP10:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 24), align 2326; SSE-NEXT: [[TMP11:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 24), align 2327; SSE-NEXT: [[TMP12:%.*]] = mul <8 x i16> [[TMP10]], [[TMP11]]328; SSE-NEXT: store <8 x i16> [[TMP12]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 24), align 2329; SSE-NEXT: ret void330;331; SLM-LABEL: @mul_v32i16(332; SLM-NEXT: [[TMP1:%.*]] = load <8 x i16>, ptr @a16, align 2333; SLM-NEXT: [[TMP2:%.*]] = load <8 x i16>, ptr @b16, align 2334; SLM-NEXT: [[TMP3:%.*]] = mul <8 x i16> [[TMP1]], [[TMP2]]335; SLM-NEXT: store <8 x i16> [[TMP3]], ptr @c16, align 2336; SLM-NEXT: [[TMP4:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 8), align 2337; SLM-NEXT: [[TMP5:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 8), align 2338; SLM-NEXT: [[TMP6:%.*]] = mul <8 x i16> [[TMP4]], [[TMP5]]339; SLM-NEXT: store <8 x i16> [[TMP6]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 8), align 2340; SLM-NEXT: [[TMP7:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 16), align 2341; SLM-NEXT: [[TMP8:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 16), align 2342; SLM-NEXT: [[TMP9:%.*]] = mul <8 x i16> [[TMP7]], [[TMP8]]343; SLM-NEXT: store <8 x i16> [[TMP9]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 16), align 2344; SLM-NEXT: [[TMP10:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 24), align 2345; SLM-NEXT: [[TMP11:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 24), align 2346; SLM-NEXT: [[TMP12:%.*]] = mul <8 x i16> [[TMP10]], [[TMP11]]347; SLM-NEXT: store <8 x i16> [[TMP12]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 24), align 2348; SLM-NEXT: ret void349;350; AVX128-LABEL: @mul_v32i16(351; AVX128-NEXT: [[TMP1:%.*]] = load <8 x i16>, ptr @a16, align 2352; AVX128-NEXT: [[TMP2:%.*]] = load <8 x i16>, ptr @b16, align 2353; AVX128-NEXT: [[TMP3:%.*]] = mul <8 x i16> [[TMP1]], [[TMP2]]354; AVX128-NEXT: store <8 x i16> [[TMP3]], ptr @c16, align 2355; AVX128-NEXT: [[TMP4:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 8), align 2356; AVX128-NEXT: [[TMP5:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 8), align 2357; AVX128-NEXT: [[TMP6:%.*]] = mul <8 x i16> [[TMP4]], [[TMP5]]358; AVX128-NEXT: store <8 x i16> [[TMP6]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 8), align 2359; AVX128-NEXT: [[TMP7:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 16), align 2360; AVX128-NEXT: [[TMP8:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 16), align 2361; AVX128-NEXT: [[TMP9:%.*]] = mul <8 x i16> [[TMP7]], [[TMP8]]362; AVX128-NEXT: store <8 x i16> [[TMP9]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 16), align 2363; AVX128-NEXT: [[TMP10:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 24), align 2364; AVX128-NEXT: [[TMP11:%.*]] = load <8 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 24), align 2365; AVX128-NEXT: [[TMP12:%.*]] = mul <8 x i16> [[TMP10]], [[TMP11]]366; AVX128-NEXT: store <8 x i16> [[TMP12]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 24), align 2367; AVX128-NEXT: ret void368;369; AVX256-LABEL: @mul_v32i16(370; AVX256-NEXT: [[TMP1:%.*]] = load <16 x i16>, ptr @a16, align 2371; AVX256-NEXT: [[TMP2:%.*]] = load <16 x i16>, ptr @b16, align 2372; AVX256-NEXT: [[TMP3:%.*]] = mul <16 x i16> [[TMP1]], [[TMP2]]373; AVX256-NEXT: store <16 x i16> [[TMP3]], ptr @c16, align 2374; AVX256-NEXT: [[TMP4:%.*]] = load <16 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 16), align 2375; AVX256-NEXT: [[TMP5:%.*]] = load <16 x i16>, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 16), align 2376; AVX256-NEXT: [[TMP6:%.*]] = mul <16 x i16> [[TMP4]], [[TMP5]]377; AVX256-NEXT: store <16 x i16> [[TMP6]], ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 16), align 2378; AVX256-NEXT: ret void379;380; AVX512-LABEL: @mul_v32i16(381; AVX512-NEXT: [[TMP1:%.*]] = load <32 x i16>, ptr @a16, align 2382; AVX512-NEXT: [[TMP2:%.*]] = load <32 x i16>, ptr @b16, align 2383; AVX512-NEXT: [[TMP3:%.*]] = mul <32 x i16> [[TMP1]], [[TMP2]]384; AVX512-NEXT: store <32 x i16> [[TMP3]], ptr @c16, align 2385; AVX512-NEXT: ret void386;387 %a0 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 0 ), align 2388 %a1 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 1 ), align 2389 %a2 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 2 ), align 2390 %a3 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 3 ), align 2391 %a4 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 4 ), align 2392 %a5 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 5 ), align 2393 %a6 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 6 ), align 2394 %a7 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 7 ), align 2395 %a8 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 8 ), align 2396 %a9 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 9 ), align 2397 %a10 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 10), align 2398 %a11 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 11), align 2399 %a12 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 12), align 2400 %a13 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 13), align 2401 %a14 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 14), align 2402 %a15 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 15), align 2403 %a16 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 16), align 2404 %a17 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 17), align 2405 %a18 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 18), align 2406 %a19 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 19), align 2407 %a20 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 20), align 2408 %a21 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 21), align 2409 %a22 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 22), align 2410 %a23 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 23), align 2411 %a24 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 24), align 2412 %a25 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 25), align 2413 %a26 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 26), align 2414 %a27 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 27), align 2415 %a28 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 28), align 2416 %a29 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 29), align 2417 %a30 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 30), align 2418 %a31 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @a16, i32 0, i64 31), align 2419 %b0 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 0 ), align 2420 %b1 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 1 ), align 2421 %b2 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 2 ), align 2422 %b3 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 3 ), align 2423 %b4 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 4 ), align 2424 %b5 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 5 ), align 2425 %b6 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 6 ), align 2426 %b7 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 7 ), align 2427 %b8 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 8 ), align 2428 %b9 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 9 ), align 2429 %b10 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 10), align 2430 %b11 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 11), align 2431 %b12 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 12), align 2432 %b13 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 13), align 2433 %b14 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 14), align 2434 %b15 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 15), align 2435 %b16 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 16), align 2436 %b17 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 17), align 2437 %b18 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 18), align 2438 %b19 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 19), align 2439 %b20 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 20), align 2440 %b21 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 21), align 2441 %b22 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 22), align 2442 %b23 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 23), align 2443 %b24 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 24), align 2444 %b25 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 25), align 2445 %b26 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 26), align 2446 %b27 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 27), align 2447 %b28 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 28), align 2448 %b29 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 29), align 2449 %b30 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 30), align 2450 %b31 = load i16, ptr getelementptr inbounds ([32 x i16], ptr @b16, i32 0, i64 31), align 2451 %r0 = mul i16 %a0 , %b0452 %r1 = mul i16 %a1 , %b1453 %r2 = mul i16 %a2 , %b2454 %r3 = mul i16 %a3 , %b3455 %r4 = mul i16 %a4 , %b4456 %r5 = mul i16 %a5 , %b5457 %r6 = mul i16 %a6 , %b6458 %r7 = mul i16 %a7 , %b7459 %r8 = mul i16 %a8 , %b8460 %r9 = mul i16 %a9 , %b9461 %r10 = mul i16 %a10, %b10462 %r11 = mul i16 %a11, %b11463 %r12 = mul i16 %a12, %b12464 %r13 = mul i16 %a13, %b13465 %r14 = mul i16 %a14, %b14466 %r15 = mul i16 %a15, %b15467 %r16 = mul i16 %a16, %b16468 %r17 = mul i16 %a17, %b17469 %r18 = mul i16 %a18, %b18470 %r19 = mul i16 %a19, %b19471 %r20 = mul i16 %a20, %b20472 %r21 = mul i16 %a21, %b21473 %r22 = mul i16 %a22, %b22474 %r23 = mul i16 %a23, %b23475 %r24 = mul i16 %a24, %b24476 %r25 = mul i16 %a25, %b25477 %r26 = mul i16 %a26, %b26478 %r27 = mul i16 %a27, %b27479 %r28 = mul i16 %a28, %b28480 %r29 = mul i16 %a29, %b29481 %r30 = mul i16 %a30, %b30482 %r31 = mul i16 %a31, %b31483 store i16 %r0 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 0 ), align 2484 store i16 %r1 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 1 ), align 2485 store i16 %r2 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 2 ), align 2486 store i16 %r3 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 3 ), align 2487 store i16 %r4 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 4 ), align 2488 store i16 %r5 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 5 ), align 2489 store i16 %r6 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 6 ), align 2490 store i16 %r7 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 7 ), align 2491 store i16 %r8 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 8 ), align 2492 store i16 %r9 , ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 9 ), align 2493 store i16 %r10, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 10), align 2494 store i16 %r11, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 11), align 2495 store i16 %r12, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 12), align 2496 store i16 %r13, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 13), align 2497 store i16 %r14, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 14), align 2498 store i16 %r15, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 15), align 2499 store i16 %r16, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 16), align 2500 store i16 %r17, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 17), align 2501 store i16 %r18, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 18), align 2502 store i16 %r19, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 19), align 2503 store i16 %r20, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 20), align 2504 store i16 %r21, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 21), align 2505 store i16 %r22, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 22), align 2506 store i16 %r23, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 23), align 2507 store i16 %r24, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 24), align 2508 store i16 %r25, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 25), align 2509 store i16 %r26, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 26), align 2510 store i16 %r27, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 27), align 2511 store i16 %r28, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 28), align 2512 store i16 %r29, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 29), align 2513 store i16 %r30, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 30), align 2514 store i16 %r31, ptr getelementptr inbounds ([32 x i16], ptr @c16, i32 0, i64 31), align 2515 ret void516}517 518define void @mul_v64i8() {519; SSE-LABEL: @mul_v64i8(520; SSE-NEXT: [[TMP1:%.*]] = load <16 x i8>, ptr @a8, align 1521; SSE-NEXT: [[TMP2:%.*]] = load <16 x i8>, ptr @b8, align 1522; SSE-NEXT: [[TMP3:%.*]] = mul <16 x i8> [[TMP1]], [[TMP2]]523; SSE-NEXT: store <16 x i8> [[TMP3]], ptr @c8, align 1524; SSE-NEXT: [[TMP4:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 16), align 1525; SSE-NEXT: [[TMP5:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 16), align 1526; SSE-NEXT: [[TMP6:%.*]] = mul <16 x i8> [[TMP4]], [[TMP5]]527; SSE-NEXT: store <16 x i8> [[TMP6]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 16), align 1528; SSE-NEXT: [[TMP7:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 32), align 1529; SSE-NEXT: [[TMP8:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 32), align 1530; SSE-NEXT: [[TMP9:%.*]] = mul <16 x i8> [[TMP7]], [[TMP8]]531; SSE-NEXT: store <16 x i8> [[TMP9]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 32), align 1532; SSE-NEXT: [[TMP10:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 48), align 1533; SSE-NEXT: [[TMP11:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 48), align 1534; SSE-NEXT: [[TMP12:%.*]] = mul <16 x i8> [[TMP10]], [[TMP11]]535; SSE-NEXT: store <16 x i8> [[TMP12]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 48), align 1536; SSE-NEXT: ret void537;538; SLM-LABEL: @mul_v64i8(539; SLM-NEXT: [[TMP1:%.*]] = load <16 x i8>, ptr @a8, align 1540; SLM-NEXT: [[TMP2:%.*]] = load <16 x i8>, ptr @b8, align 1541; SLM-NEXT: [[TMP3:%.*]] = mul <16 x i8> [[TMP1]], [[TMP2]]542; SLM-NEXT: store <16 x i8> [[TMP3]], ptr @c8, align 1543; SLM-NEXT: [[TMP4:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 16), align 1544; SLM-NEXT: [[TMP5:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 16), align 1545; SLM-NEXT: [[TMP6:%.*]] = mul <16 x i8> [[TMP4]], [[TMP5]]546; SLM-NEXT: store <16 x i8> [[TMP6]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 16), align 1547; SLM-NEXT: [[TMP7:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 32), align 1548; SLM-NEXT: [[TMP8:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 32), align 1549; SLM-NEXT: [[TMP9:%.*]] = mul <16 x i8> [[TMP7]], [[TMP8]]550; SLM-NEXT: store <16 x i8> [[TMP9]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 32), align 1551; SLM-NEXT: [[TMP10:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 48), align 1552; SLM-NEXT: [[TMP11:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 48), align 1553; SLM-NEXT: [[TMP12:%.*]] = mul <16 x i8> [[TMP10]], [[TMP11]]554; SLM-NEXT: store <16 x i8> [[TMP12]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 48), align 1555; SLM-NEXT: ret void556;557; AVX128-LABEL: @mul_v64i8(558; AVX128-NEXT: [[TMP1:%.*]] = load <16 x i8>, ptr @a8, align 1559; AVX128-NEXT: [[TMP2:%.*]] = load <16 x i8>, ptr @b8, align 1560; AVX128-NEXT: [[TMP3:%.*]] = mul <16 x i8> [[TMP1]], [[TMP2]]561; AVX128-NEXT: store <16 x i8> [[TMP3]], ptr @c8, align 1562; AVX128-NEXT: [[TMP4:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 16), align 1563; AVX128-NEXT: [[TMP5:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 16), align 1564; AVX128-NEXT: [[TMP6:%.*]] = mul <16 x i8> [[TMP4]], [[TMP5]]565; AVX128-NEXT: store <16 x i8> [[TMP6]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 16), align 1566; AVX128-NEXT: [[TMP7:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 32), align 1567; AVX128-NEXT: [[TMP8:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 32), align 1568; AVX128-NEXT: [[TMP9:%.*]] = mul <16 x i8> [[TMP7]], [[TMP8]]569; AVX128-NEXT: store <16 x i8> [[TMP9]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 32), align 1570; AVX128-NEXT: [[TMP10:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 48), align 1571; AVX128-NEXT: [[TMP11:%.*]] = load <16 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 48), align 1572; AVX128-NEXT: [[TMP12:%.*]] = mul <16 x i8> [[TMP10]], [[TMP11]]573; AVX128-NEXT: store <16 x i8> [[TMP12]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 48), align 1574; AVX128-NEXT: ret void575;576; AVX256-LABEL: @mul_v64i8(577; AVX256-NEXT: [[TMP1:%.*]] = load <32 x i8>, ptr @a8, align 1578; AVX256-NEXT: [[TMP2:%.*]] = load <32 x i8>, ptr @b8, align 1579; AVX256-NEXT: [[TMP3:%.*]] = mul <32 x i8> [[TMP1]], [[TMP2]]580; AVX256-NEXT: store <32 x i8> [[TMP3]], ptr @c8, align 1581; AVX256-NEXT: [[TMP4:%.*]] = load <32 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 32), align 1582; AVX256-NEXT: [[TMP5:%.*]] = load <32 x i8>, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 32), align 1583; AVX256-NEXT: [[TMP6:%.*]] = mul <32 x i8> [[TMP4]], [[TMP5]]584; AVX256-NEXT: store <32 x i8> [[TMP6]], ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 32), align 1585; AVX256-NEXT: ret void586;587; AVX512-LABEL: @mul_v64i8(588; AVX512-NEXT: [[TMP1:%.*]] = load <64 x i8>, ptr @a8, align 1589; AVX512-NEXT: [[TMP2:%.*]] = load <64 x i8>, ptr @b8, align 1590; AVX512-NEXT: [[TMP3:%.*]] = mul <64 x i8> [[TMP1]], [[TMP2]]591; AVX512-NEXT: store <64 x i8> [[TMP3]], ptr @c8, align 1592; AVX512-NEXT: ret void593;594 %a0 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 0 ), align 1595 %a1 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 1 ), align 1596 %a2 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 2 ), align 1597 %a3 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 3 ), align 1598 %a4 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 4 ), align 1599 %a5 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 5 ), align 1600 %a6 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 6 ), align 1601 %a7 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 7 ), align 1602 %a8 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 8 ), align 1603 %a9 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 9 ), align 1604 %a10 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 10), align 1605 %a11 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 11), align 1606 %a12 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 12), align 1607 %a13 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 13), align 1608 %a14 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 14), align 1609 %a15 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 15), align 1610 %a16 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 16), align 1611 %a17 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 17), align 1612 %a18 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 18), align 1613 %a19 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 19), align 1614 %a20 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 20), align 1615 %a21 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 21), align 1616 %a22 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 22), align 1617 %a23 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 23), align 1618 %a24 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 24), align 1619 %a25 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 25), align 1620 %a26 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 26), align 1621 %a27 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 27), align 1622 %a28 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 28), align 1623 %a29 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 29), align 1624 %a30 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 30), align 1625 %a31 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 31), align 1626 %a32 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 32), align 1627 %a33 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 33), align 1628 %a34 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 34), align 1629 %a35 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 35), align 1630 %a36 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 36), align 1631 %a37 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 37), align 1632 %a38 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 38), align 1633 %a39 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 39), align 1634 %a40 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 40), align 1635 %a41 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 41), align 1636 %a42 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 42), align 1637 %a43 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 43), align 1638 %a44 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 44), align 1639 %a45 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 45), align 1640 %a46 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 46), align 1641 %a47 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 47), align 1642 %a48 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 48), align 1643 %a49 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 49), align 1644 %a50 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 50), align 1645 %a51 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 51), align 1646 %a52 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 52), align 1647 %a53 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 53), align 1648 %a54 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 54), align 1649 %a55 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 55), align 1650 %a56 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 56), align 1651 %a57 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 57), align 1652 %a58 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 58), align 1653 %a59 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 59), align 1654 %a60 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 60), align 1655 %a61 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 61), align 1656 %a62 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 62), align 1657 %a63 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @a8, i32 0, i64 63), align 1658 %b0 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 0 ), align 1659 %b1 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 1 ), align 1660 %b2 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 2 ), align 1661 %b3 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 3 ), align 1662 %b4 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 4 ), align 1663 %b5 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 5 ), align 1664 %b6 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 6 ), align 1665 %b7 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 7 ), align 1666 %b8 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 8 ), align 1667 %b9 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 9 ), align 1668 %b10 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 10), align 1669 %b11 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 11), align 1670 %b12 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 12), align 1671 %b13 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 13), align 1672 %b14 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 14), align 1673 %b15 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 15), align 1674 %b16 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 16), align 1675 %b17 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 17), align 1676 %b18 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 18), align 1677 %b19 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 19), align 1678 %b20 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 20), align 1679 %b21 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 21), align 1680 %b22 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 22), align 1681 %b23 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 23), align 1682 %b24 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 24), align 1683 %b25 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 25), align 1684 %b26 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 26), align 1685 %b27 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 27), align 1686 %b28 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 28), align 1687 %b29 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 29), align 1688 %b30 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 30), align 1689 %b31 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 31), align 1690 %b32 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 32), align 1691 %b33 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 33), align 1692 %b34 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 34), align 1693 %b35 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 35), align 1694 %b36 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 36), align 1695 %b37 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 37), align 1696 %b38 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 38), align 1697 %b39 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 39), align 1698 %b40 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 40), align 1699 %b41 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 41), align 1700 %b42 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 42), align 1701 %b43 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 43), align 1702 %b44 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 44), align 1703 %b45 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 45), align 1704 %b46 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 46), align 1705 %b47 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 47), align 1706 %b48 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 48), align 1707 %b49 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 49), align 1708 %b50 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 50), align 1709 %b51 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 51), align 1710 %b52 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 52), align 1711 %b53 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 53), align 1712 %b54 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 54), align 1713 %b55 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 55), align 1714 %b56 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 56), align 1715 %b57 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 57), align 1716 %b58 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 58), align 1717 %b59 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 59), align 1718 %b60 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 60), align 1719 %b61 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 61), align 1720 %b62 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 62), align 1721 %b63 = load i8, ptr getelementptr inbounds ([64 x i8], ptr @b8, i32 0, i64 63), align 1722 %r0 = mul i8 %a0 , %b0723 %r1 = mul i8 %a1 , %b1724 %r2 = mul i8 %a2 , %b2725 %r3 = mul i8 %a3 , %b3726 %r4 = mul i8 %a4 , %b4727 %r5 = mul i8 %a5 , %b5728 %r6 = mul i8 %a6 , %b6729 %r7 = mul i8 %a7 , %b7730 %r8 = mul i8 %a8 , %b8731 %r9 = mul i8 %a9 , %b9732 %r10 = mul i8 %a10, %b10733 %r11 = mul i8 %a11, %b11734 %r12 = mul i8 %a12, %b12735 %r13 = mul i8 %a13, %b13736 %r14 = mul i8 %a14, %b14737 %r15 = mul i8 %a15, %b15738 %r16 = mul i8 %a16, %b16739 %r17 = mul i8 %a17, %b17740 %r18 = mul i8 %a18, %b18741 %r19 = mul i8 %a19, %b19742 %r20 = mul i8 %a20, %b20743 %r21 = mul i8 %a21, %b21744 %r22 = mul i8 %a22, %b22745 %r23 = mul i8 %a23, %b23746 %r24 = mul i8 %a24, %b24747 %r25 = mul i8 %a25, %b25748 %r26 = mul i8 %a26, %b26749 %r27 = mul i8 %a27, %b27750 %r28 = mul i8 %a28, %b28751 %r29 = mul i8 %a29, %b29752 %r30 = mul i8 %a30, %b30753 %r31 = mul i8 %a31, %b31754 %r32 = mul i8 %a32, %b32755 %r33 = mul i8 %a33, %b33756 %r34 = mul i8 %a34, %b34757 %r35 = mul i8 %a35, %b35758 %r36 = mul i8 %a36, %b36759 %r37 = mul i8 %a37, %b37760 %r38 = mul i8 %a38, %b38761 %r39 = mul i8 %a39, %b39762 %r40 = mul i8 %a40, %b40763 %r41 = mul i8 %a41, %b41764 %r42 = mul i8 %a42, %b42765 %r43 = mul i8 %a43, %b43766 %r44 = mul i8 %a44, %b44767 %r45 = mul i8 %a45, %b45768 %r46 = mul i8 %a46, %b46769 %r47 = mul i8 %a47, %b47770 %r48 = mul i8 %a48, %b48771 %r49 = mul i8 %a49, %b49772 %r50 = mul i8 %a50, %b50773 %r51 = mul i8 %a51, %b51774 %r52 = mul i8 %a52, %b52775 %r53 = mul i8 %a53, %b53776 %r54 = mul i8 %a54, %b54777 %r55 = mul i8 %a55, %b55778 %r56 = mul i8 %a56, %b56779 %r57 = mul i8 %a57, %b57780 %r58 = mul i8 %a58, %b58781 %r59 = mul i8 %a59, %b59782 %r60 = mul i8 %a60, %b60783 %r61 = mul i8 %a61, %b61784 %r62 = mul i8 %a62, %b62785 %r63 = mul i8 %a63, %b63786 store i8 %r0 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 0 ), align 1787 store i8 %r1 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 1 ), align 1788 store i8 %r2 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 2 ), align 1789 store i8 %r3 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 3 ), align 1790 store i8 %r4 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 4 ), align 1791 store i8 %r5 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 5 ), align 1792 store i8 %r6 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 6 ), align 1793 store i8 %r7 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 7 ), align 1794 store i8 %r8 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 8 ), align 1795 store i8 %r9 , ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 9 ), align 1796 store i8 %r10, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 10), align 1797 store i8 %r11, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 11), align 1798 store i8 %r12, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 12), align 1799 store i8 %r13, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 13), align 1800 store i8 %r14, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 14), align 1801 store i8 %r15, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 15), align 1802 store i8 %r16, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 16), align 1803 store i8 %r17, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 17), align 1804 store i8 %r18, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 18), align 1805 store i8 %r19, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 19), align 1806 store i8 %r20, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 20), align 1807 store i8 %r21, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 21), align 1808 store i8 %r22, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 22), align 1809 store i8 %r23, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 23), align 1810 store i8 %r24, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 24), align 1811 store i8 %r25, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 25), align 1812 store i8 %r26, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 26), align 1813 store i8 %r27, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 27), align 1814 store i8 %r28, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 28), align 1815 store i8 %r29, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 29), align 1816 store i8 %r30, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 30), align 1817 store i8 %r31, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 31), align 1818 store i8 %r32, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 32), align 1819 store i8 %r33, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 33), align 1820 store i8 %r34, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 34), align 1821 store i8 %r35, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 35), align 1822 store i8 %r36, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 36), align 1823 store i8 %r37, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 37), align 1824 store i8 %r38, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 38), align 1825 store i8 %r39, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 39), align 1826 store i8 %r40, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 40), align 1827 store i8 %r41, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 41), align 1828 store i8 %r42, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 42), align 1829 store i8 %r43, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 43), align 1830 store i8 %r44, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 44), align 1831 store i8 %r45, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 45), align 1832 store i8 %r46, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 46), align 1833 store i8 %r47, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 47), align 1834 store i8 %r48, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 48), align 1835 store i8 %r49, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 49), align 1836 store i8 %r50, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 50), align 1837 store i8 %r51, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 51), align 1838 store i8 %r52, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 52), align 1839 store i8 %r53, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 53), align 1840 store i8 %r54, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 54), align 1841 store i8 %r55, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 55), align 1842 store i8 %r56, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 56), align 1843 store i8 %r57, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 57), align 1844 store i8 %r58, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 58), align 1845 store i8 %r59, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 59), align 1846 store i8 %r60, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 60), align 1847 store i8 %r61, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 61), align 1848 store i8 %r62, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 62), align 1849 store i8 %r63, ptr getelementptr inbounds ([64 x i8], ptr @c8, i32 0, i64 63), align 1850 ret void851}852