brintos

brintos / llvm-project-archived public Read only

0
0
Text · 16.8 KiB · 854cc09 Raw
356 lines · c
1// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512vbmi -emit-llvm -o - -Wall -Werror | FileCheck %s2// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=i386-apple-darwin -target-feature +avx512vbmi -emit-llvm -o - -Wall -Werror | FileCheck %s3// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512vbmi -emit-llvm -o - -Wall -Werror | FileCheck %s4// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=i386-apple-darwin -target-feature +avx512vbmi -emit-llvm -o - -Wall -Werror | FileCheck %s5 6// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512vbmi -emit-llvm -o - -Wall -Werror -fexperimental-new-constant-interpreter | FileCheck %s7// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=i386-apple-darwin -target-feature +avx512vbmi -emit-llvm -o - -Wall -Werror -fexperimental-new-constant-interpreter | FileCheck %s8// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512vbmi -emit-llvm -o - -Wall -Werror -fexperimental-new-constant-interpreter | FileCheck %s9// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=i386-apple-darwin -target-feature +avx512vbmi -emit-llvm -o - -Wall -Werror -fexperimental-new-constant-interpreter | FileCheck %s10 11 12#include <immintrin.h>13#include "builtin_test_helpers.h"14 15__m512i test_mm512_mask2_permutex2var_epi8(__m512i __A, __m512i __I, __mmask64 __U, __m512i __B) {16  // CHECK-LABEL: test_mm512_mask2_permutex2var_epi817  // CHECK: call <64 x i8> @llvm.x86.avx512.vpermi2var.qi.512(<64 x i8> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}})18  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}19  return _mm512_mask2_permutex2var_epi8(__A, __I, __U, __B); 20}21 22__m512i test_mm512_permutex2var_epi8(__m512i __A, __m512i __I, __m512i __B) {23  // CHECK-LABEL: test_mm512_permutex2var_epi824  // CHECK: call <64 x i8> @llvm.x86.avx512.vpermi2var.qi.512(<64 x i8> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}})25  return _mm512_permutex2var_epi8(__A, __I, __B); 26}27 28__m512i test_mm512_mask_permutex2var_epi8(__m512i __A, __mmask64 __U, __m512i __I, __m512i __B) {29  // CHECK-LABEL: test_mm512_mask_permutex2var_epi830  // CHECK: call <64 x i8> @llvm.x86.avx512.vpermi2var.qi.512(<64 x i8> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}})31  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}32  return _mm512_mask_permutex2var_epi8(__A, __U, __I, __B); 33}34 35__m512i test_mm512_maskz_permutex2var_epi8(__mmask64 __U, __m512i __A, __m512i __I, __m512i __B) {36  // CHECK-LABEL: test_mm512_maskz_permutex2var_epi837  // CHECK: call <64 x i8> @llvm.x86.avx512.vpermi2var.qi.512(<64 x i8> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}})38  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}39  return _mm512_maskz_permutex2var_epi8(__U, __A, __I, __B); 40}41 42TEST_CONSTEXPR(match_v64qu(43    _mm512_permutex2var_epi8((__m512i)(__v64qu){44            0, 1, 2, 3, 4, 5, 6, 7,45            8, 9, 10, 11, 12, 13, 14, 15,46            16, 17, 18, 19, 20, 21, 22, 23,47            24, 25, 26, 27, 28, 29, 30, 31,48            32, 33, 34, 35, 36, 37, 38, 39,49            40, 41, 42, 43, 44, 45, 46, 47,50            48, 49, 50, 51, 52, 53, 54, 55,51            56, 57, 58, 59, 60, 61, 62, 63},52        (__m512i)(__v64qu){53            0, 64, 1, 65, 2, 66, 3, 67,54            4, 68, 5, 69, 6, 70, 7, 71,55            8, 72, 9, 73, 10, 74, 11, 75,56            12, 76, 13, 77, 14, 78, 15, 79,57            16, 80, 17, 81, 18, 82, 19, 83,58            20, 84, 21, 85, 22, 86, 23, 87,59            24, 88, 25, 89, 26, 90, 27, 91,60            28, 92, 29, 93, 30, 94, 31, 95},61        (__m512i)(__v64qu){62            200, 201, 202, 203, 204, 205, 206, 207,63            208, 209, 210, 211, 212, 213, 214, 215,64            216, 217, 218, 219, 220, 221, 222, 223,65            224, 225, 226, 227, 228, 229, 230, 231,66            232, 233, 234, 235, 236, 237, 238, 239,67            240, 241, 242, 243, 244, 245, 246, 247,68            248, 249, 250, 251, 252, 253, 254, 255,69            0, 1, 2, 3, 4, 5, 6, 7}),70    0, 200, 1, 201, 2, 202, 3, 203,71    4, 204, 5, 205, 6, 206, 7, 207,72    8, 208, 9, 209, 10, 210, 11, 211,73    12, 212, 13, 213, 14, 214, 15, 215,74    16, 216, 17, 217, 18, 218, 19, 219,75    20, 220, 21, 221, 22, 222, 23, 223,76    24, 224, 25, 225, 26, 226, 27, 227,77    28, 228, 29, 229, 30, 230, 31, 231));78TEST_CONSTEXPR(match_v64qu(79    _mm512_mask_permutex2var_epi8((__m512i)(__v64qu){80            10, 11, 12, 13, 14, 15, 16, 17,81            18, 19, 20, 21, 22, 23, 24, 25,82            26, 27, 28, 29, 30, 31, 32, 33,83            34, 35, 36, 37, 38, 39, 40, 41,84            42, 43, 44, 45, 46, 47, 48, 49,85            50, 51, 52, 53, 54, 55, 56, 57,86            58, 59, 60, 61, 62, 63, 64, 65,87            66, 67, 68, 69, 70, 71, 72, 73},88        0xAAAAAAAAAAAAAAAAULL,89        (__m512i)(__v64qu){90            0, 64, 1, 65, 2, 66, 3, 67,91            4, 68, 5, 69, 6, 70, 7, 71,92            8, 72, 9, 73, 10, 74, 11, 75,93            12, 76, 13, 77, 14, 78, 15, 79,94            16, 80, 17, 81, 18, 82, 19, 83,95            20, 84, 21, 85, 22, 86, 23, 87,96            24, 88, 25, 89, 26, 90, 27, 91,97            28, 92, 29, 93, 30, 94, 31, 95},98        (__m512i)(__v64qu){99            200, 201, 202, 203, 204, 205, 206, 207,100            208, 209, 210, 211, 212, 213, 214, 215,101            216, 217, 218, 219, 220, 221, 222, 223,102            224, 225, 226, 227, 228, 229, 230, 231,103            232, 233, 234, 235, 236, 237, 238, 239,104            240, 241, 242, 243, 244, 245, 246, 247,105            248, 249, 250, 251, 252, 253, 254, 255,106            0, 1, 2, 3, 4, 5, 6, 7}),107    10, 200, 12, 201, 14, 202, 16, 203,108    18, 204, 20, 205, 22, 206, 24, 207,109    26, 208, 28, 209, 30, 210, 32, 211,110    34, 212, 36, 213, 38, 214, 40, 215,111    42, 216, 44, 217, 46, 218, 48, 219,112    50, 220, 52, 221, 54, 222, 56, 223,113    58, 224, 60, 225, 62, 226, 64, 227,114    66, 228, 68, 229, 70, 230, 72, 231));115TEST_CONSTEXPR(match_v64qu(116    _mm512_maskz_permutex2var_epi8(0x5555555555555555ULL,117        (__m512i)(__v64qu){118            0, 1, 2, 3, 4, 5, 6, 7,119            8, 9, 10, 11, 12, 13, 14, 15,120            16, 17, 18, 19, 20, 21, 22, 23,121            24, 25, 26, 27, 28, 29, 30, 31,122            32, 33, 34, 35, 36, 37, 38, 39,123            40, 41, 42, 43, 44, 45, 46, 47,124            48, 49, 50, 51, 52, 53, 54, 55,125            56, 57, 58, 59, 60, 61, 62, 63},126        (__m512i)(__v64qu){127            0, 64, 1, 65, 2, 66, 3, 67,128            4, 68, 5, 69, 6, 70, 7, 71,129            8, 72, 9, 73, 10, 74, 11, 75,130            12, 76, 13, 77, 14, 78, 15, 79,131            16, 80, 17, 81, 18, 82, 19, 83,132            20, 84, 21, 85, 22, 86, 23, 87,133            24, 88, 25, 89, 26, 90, 27, 91,134            28, 92, 29, 93, 30, 94, 31, 95},135        (__m512i)(__v64qu){136            200, 201, 202, 203, 204, 205, 206, 207,137            208, 209, 210, 211, 212, 213, 214, 215,138            216, 217, 218, 219, 220, 221, 222, 223,139            224, 225, 226, 227, 228, 229, 230, 231,140            232, 233, 234, 235, 236, 237, 238, 239,141            240, 241, 242, 243, 244, 245, 246, 247,142            248, 249, 250, 251, 252, 253, 254, 255,143            0, 1, 2, 3, 4, 5, 6, 7}),144    0, 0, 1, 0, 2, 0, 3, 0,145    4, 0, 5, 0, 6, 0, 7, 0,146    8, 0, 9, 0, 10, 0, 11, 0,147    12, 0, 13, 0, 14, 0, 15, 0,148    16, 0, 17, 0, 18, 0, 19, 0,149    20, 0, 21, 0, 22, 0, 23, 0,150    24, 0, 25, 0, 26, 0, 27, 0,151    28, 0, 29, 0, 30, 0, 31, 0));152TEST_CONSTEXPR(match_v64qu(153    _mm512_mask2_permutex2var_epi8((__m512i)(__v64qu){154            0, 1, 2, 3, 4, 5, 6, 7,155            8, 9, 10, 11, 12, 13, 14, 15,156            16, 17, 18, 19, 20, 21, 22, 23,157            24, 25, 26, 27, 28, 29, 30, 31,158            32, 33, 34, 35, 36, 37, 38, 39,159            40, 41, 42, 43, 44, 45, 46, 47,160            48, 49, 50, 51, 52, 53, 54, 55,161            56, 57, 58, 59, 60, 61, 62, 63},162        (__m512i)(__v64qu){163            0, 64, 1, 65, 2, 66, 3, 67,164            4, 68, 5, 69, 6, 70, 7, 71,165            8, 72, 9, 73, 10, 74, 11, 75,166            12, 76, 13, 77, 14, 78, 15, 79,167            16, 80, 17, 81, 18, 82, 19, 83,168            20, 84, 21, 85, 22, 86, 23, 87,169            24, 88, 25, 89, 26, 90, 27, 91,170            28, 92, 29, 93, 30, 94, 31, 95},171        0x5555555555555555ULL,172        (__m512i)(__v64qu){173            200, 201, 202, 203, 204, 205, 206, 207,174            208, 209, 210, 211, 212, 213, 214, 215,175            216, 217, 218, 219, 220, 221, 222, 223,176            224, 225, 226, 227, 228, 229, 230, 231,177            232, 233, 234, 235, 236, 237, 238, 239,178            240, 241, 242, 243, 244, 245, 246, 247,179            248, 249, 250, 251, 252, 253, 254, 255,180            0, 1, 2, 3, 4, 5, 6, 7}),181    0, 64, 1, 65, 2, 66, 3, 67,182    4, 68, 5, 69, 6, 70, 7, 71,183    8, 72, 9, 73, 10, 74, 11, 75,184    12, 76, 13, 77, 14, 78, 15, 79,185    16, 80, 17, 81, 18, 82, 19, 83,186    20, 84, 21, 85, 22, 86, 23, 87,187    24, 88, 25, 89, 26, 90, 27, 91,188    28, 92, 29, 93, 30, 94, 31, 95));189 190__m512i test_mm512_permutexvar_epi8(__m512i __A, __m512i __B) {191  // CHECK-LABEL: test_mm512_permutexvar_epi8192  // CHECK: call <64 x i8> @llvm.x86.avx512.permvar.qi.512(<64 x i8> %{{.*}}, <64 x i8> %{{.*}})193  return _mm512_permutexvar_epi8(__A, __B); 194}195 196__m512i test_mm512_maskz_permutexvar_epi8(__mmask64 __M, __m512i __A, __m512i __B) {197  // CHECK-LABEL: test_mm512_maskz_permutexvar_epi8198  // CHECK: call <64 x i8> @llvm.x86.avx512.permvar.qi.512(<64 x i8> %{{.*}}, <64 x i8> %{{.*}})199  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}200  return _mm512_maskz_permutexvar_epi8(__M, __A, __B); 201}202 203__m512i test_mm512_mask_permutexvar_epi8(__m512i __W, __mmask64 __M, __m512i __A, __m512i __B) {204  // CHECK-LABEL: test_mm512_mask_permutexvar_epi8205  // CHECK: call <64 x i8> @llvm.x86.avx512.permvar.qi.512(<64 x i8> %{{.*}}, <64 x i8> %{{.*}})206  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}207  return _mm512_mask_permutexvar_epi8(__W, __M, __A, __B); 208}209 210__m512i test_mm512_multishift_epi64_epi8(__m512i __X, __m512i __Y) {211  // CHECK-LABEL: test_mm512_multishift_epi64_epi8212  // CHECK: call <64 x i8> @llvm.x86.avx512.pmultishift.qb.512(<64 x i8> %{{.*}}, <64 x i8> %{{.*}})213  return _mm512_multishift_epi64_epi8(__X, __Y);214}215 216TEST_CONSTEXPR(match_v64qu(217    _mm512_multishift_epi64_epi8(218        (__m512i)(__v64qu){219            0, 8, 16, 24, 32, 40, 48, 56,220            0, 8, 16, 24, 32, 40, 48, 56,221            0, 8, 16, 24, 32, 40, 48, 56,222            0, 8, 16, 24, 32, 40, 48, 56,223            0, 8, 16, 24, 32, 40, 48, 56,224            0, 8, 16, 24, 32, 40, 48, 56,225            0, 8, 16, 24, 32, 40, 48, 56,226            0, 8, 16, 24, 32, 40, 48, 56},227        (__m512i)(__v64qu){228            0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,229            0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18,230            0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28,231            0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38,232            0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48,233            0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58,234            0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68,235            0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78}),236    0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,237    0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18,238    0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28,239    0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38,240    0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48,241    0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58,242    0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68,243    0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78));244 245TEST_CONSTEXPR(match_v64qu(246    _mm512_multishift_epi64_epi8(247        (__m512i)(__v64qu){248            4, 4, 4, 4, 4, 4, 4, 4,249            4, 4, 4, 4, 4, 4, 4, 4,250            4, 4, 4, 4, 4, 4, 4, 4,251            4, 4, 4, 4, 4, 4, 4, 4,252            4, 4, 4, 4, 4, 4, 4, 4,253            4, 4, 4, 4, 4, 4, 4, 4,254            4, 4, 4, 4, 4, 4, 4, 4,255            4, 4, 4, 4, 4, 4, 4, 4},256        (__m512i)(__v64qu){257            0x10, 0x32, 0x54, 0x76, 0x98, 0xBA, 0xDC, 0xFE,258            0x10, 0x32, 0x54, 0x76, 0x98, 0xBA, 0xDC, 0xFE,259            0x10, 0x32, 0x54, 0x76, 0x98, 0xBA, 0xDC, 0xFE,260            0x10, 0x32, 0x54, 0x76, 0x98, 0xBA, 0xDC, 0xFE,261            0x10, 0x32, 0x54, 0x76, 0x98, 0xBA, 0xDC, 0xFE,262            0x10, 0x32, 0x54, 0x76, 0x98, 0xBA, 0xDC, 0xFE,263            0x10, 0x32, 0x54, 0x76, 0x98, 0xBA, 0xDC, 0xFE,264            0x10, 0x32, 0x54, 0x76, 0x98, 0xBA, 0xDC, 0xFE}),265    0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21,266    0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21,267    0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21,268    0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21,269    0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21,270    0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21,271    0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21,272    0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21, 0x21));273 274__m512i test_mm512_mask_multishift_epi64_epi8(__m512i __W, __mmask64 __M, __m512i __X, __m512i __Y) {275  // CHECK-LABEL: test_mm512_mask_multishift_epi64_epi8276  // CHECK: call <64 x i8> @llvm.x86.avx512.pmultishift.qb.512(<64 x i8> %{{.*}}, <64 x i8> %{{.*}})277  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}278  return _mm512_mask_multishift_epi64_epi8(__W, __M, __X, __Y);279}280 281TEST_CONSTEXPR(match_v64qu(282    _mm512_mask_multishift_epi64_epi8(283        (__m512i)(__v64qu){284            0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,285            0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,286            0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,287            0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,288            0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,289            0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,290            0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,291            0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF},292        0xAAAAAAAAAAAAAAAAULL,293        (__m512i)(__v64qu){294            0, 8, 16, 24, 32, 40, 48, 56,295            0, 8, 16, 24, 32, 40, 48, 56,296            0, 8, 16, 24, 32, 40, 48, 56,297            0, 8, 16, 24, 32, 40, 48, 56,298            0, 8, 16, 24, 32, 40, 48, 56,299            0, 8, 16, 24, 32, 40, 48, 56,300            0, 8, 16, 24, 32, 40, 48, 56,301            0, 8, 16, 24, 32, 40, 48, 56},302        (__m512i)(__v64qu){303            0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,304            0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18,305            0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28,306            0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38,307            0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48,308            0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58,309            0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68,310            0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78}),311    0xFF, 0x02, 0xFF, 0x04, 0xFF, 0x06, 0xFF, 0x08,312    0xFF, 0x12, 0xFF, 0x14, 0xFF, 0x16, 0xFF, 0x18,313    0xFF, 0x22, 0xFF, 0x24, 0xFF, 0x26, 0xFF, 0x28,314    0xFF, 0x32, 0xFF, 0x34, 0xFF, 0x36, 0xFF, 0x38,315    0xFF, 0x42, 0xFF, 0x44, 0xFF, 0x46, 0xFF, 0x48,316    0xFF, 0x52, 0xFF, 0x54, 0xFF, 0x56, 0xFF, 0x58,317    0xFF, 0x62, 0xFF, 0x64, 0xFF, 0x66, 0xFF, 0x68,318    0xFF, 0x72, 0xFF, 0x74, 0xFF, 0x76, 0xFF, 0x78));319 320__m512i test_mm512_maskz_multishift_epi64_epi8(__mmask64 __M, __m512i __X, __m512i __Y) {321  // CHECK-LABEL: test_mm512_maskz_multishift_epi64_epi8322  // CHECK: call <64 x i8> @llvm.x86.avx512.pmultishift.qb.512(<64 x i8> %{{.*}}, <64 x i8> %{{.*}})323  // CHECK: select <64 x i1> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}}324  return _mm512_maskz_multishift_epi64_epi8(__M, __X, __Y);325}326 327TEST_CONSTEXPR(match_v64qu(328    _mm512_maskz_multishift_epi64_epi8(329        0x5555555555555555ULL,330        (__m512i)(__v64qu){331            0, 8, 16, 24, 32, 40, 48, 56,332            0, 8, 16, 24, 32, 40, 48, 56,333            0, 8, 16, 24, 32, 40, 48, 56,334            0, 8, 16, 24, 32, 40, 48, 56,335            0, 8, 16, 24, 32, 40, 48, 56,336            0, 8, 16, 24, 32, 40, 48, 56,337            0, 8, 16, 24, 32, 40, 48, 56,338            0, 8, 16, 24, 32, 40, 48, 56},339        (__m512i)(__v64qu){340            0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,341            0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18,342            0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28,343            0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38,344            0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48,345            0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58,346            0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68,347            0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78}),348    0x01, 0, 0x03, 0, 0x05, 0, 0x07, 0,349    0x11, 0, 0x13, 0, 0x15, 0, 0x17, 0,350    0x21, 0, 0x23, 0, 0x25, 0, 0x27, 0,351    0x31, 0, 0x33, 0, 0x35, 0, 0x37, 0,352    0x41, 0, 0x43, 0, 0x45, 0, 0x47, 0,353    0x51, 0, 0x53, 0, 0x55, 0, 0x57, 0,354    0x61, 0, 0x63, 0, 0x65, 0, 0x67, 0,355    0x71, 0, 0x73, 0, 0x75, 0, 0x77, 0));356