|
| 1 | +// We need optimization to convert target-specific masked loads/stores to llvm |
| 2 | +// generic loads/stores |
| 3 | +// RUN: %clangxx_asan -o %t %s -mavx -O1 |
| 4 | +// RUN: not %run %t l1 2>&1 | FileCheck -check-prefix=CHECK-L1 %s |
| 5 | +// RUN: %run %t l6 2>&1 | FileCheck -check-prefix=CHECK-L6 %s |
| 6 | +// RUN: %run %t la 2>&1 | FileCheck -check-prefix=CHECK-LA %s |
| 7 | +// RUN: not %run %t s1 2>&1 | FileCheck -check-prefix=CHECK-S1 %s |
| 8 | +// RUN: %run %t s6 2>&1 | FileCheck -check-prefix=CHECK-S6 %s |
| 9 | +// RUN: %run %t sa 2>&1 | FileCheck -check-prefix=CHECK-SA %s |
| 10 | +// REQUIRES: x86-target-arch |
| 11 | +#include <assert.h> |
| 12 | +#include <stdio.h> |
| 13 | +#include <x86intrin.h> |
| 14 | + |
| 15 | +float g_vec3[3] = {1802398064.0, 1881171305.0, 25961.0}; |
| 16 | + |
| 17 | +void maskedstore_ps_0110(__m128 a) { |
| 18 | + _mm_maskstore_ps(g_vec3, (__v4si){0, -1, -1, 0}, a); |
| 19 | +} |
| 20 | + |
| 21 | +void maskedstore_ps_0001(__m128 a) { |
| 22 | + _mm_maskstore_ps(g_vec3, (__v4si){0, 0, 0, -1}, a); |
| 23 | +} |
| 24 | + |
| 25 | +void maskedstore_ps_1010(__m128 a) { |
| 26 | + _mm_maskstore_ps(g_vec3, (__v4si){-1, 0, -1, 0}, a); |
| 27 | +} |
| 28 | + |
| 29 | +__m128i maskedload_ps_0110() { |
| 30 | + return _mm_maskload_ps(g_vec3, (__v4si){0, -1, -1, 0}); |
| 31 | +} |
| 32 | + |
| 33 | +__m128i maskedload_ps_0001() { |
| 34 | + return _mm_maskload_ps(g_vec3, (__v4si){0, 0, 0, -1}); |
| 35 | +} |
| 36 | + |
| 37 | +__m128i maskedload_ps_1010() { |
| 38 | + return _mm_maskload_ps(g_vec3, (__v4si){-1, 0, -1, 0}); |
| 39 | +} |
| 40 | + |
| 41 | +__m128 a = (__v4sf){1.0, 2.0, 3.0, 4.0}; |
| 42 | + |
| 43 | +void print_vector(__v4sf v) { |
| 44 | + printf("%d,%d,%d,%d\n", (int)v[0], (int)v[1], (int)v[2], (int)v[3]); |
| 45 | +} |
| 46 | +void print_vector3(float * v) { |
| 47 | + printf("%d,%d,%d\n", (int)v[0], (int)v[1], (int)v[2]); |
| 48 | +} |
| 49 | + |
| 50 | +int main(int argc, char **argv) { |
| 51 | + assert(argc > 1); |
| 52 | + bool isLoad = argv[1][0] == 'l'; |
| 53 | + assert(isLoad || argv[1][0] == 's'); |
| 54 | + if (isLoad) { |
| 55 | + switch (argv[1][1]) { |
| 56 | + case '1': { |
| 57 | + // CHECK-L1: ERROR: AddressSanitizer |
| 58 | + __v4sf v = maskedload_ps_0001(); |
| 59 | + print_vector(v); |
| 60 | + // Should have blown up |
| 61 | + break; |
| 62 | + } |
| 63 | + case '6': { |
| 64 | + // Safe |
| 65 | + __v4sf v = maskedload_ps_0110(); |
| 66 | + // CHECK-L6: 0,1881171328,25961,0 |
| 67 | + print_vector(v); |
| 68 | + return 0; |
| 69 | + } |
| 70 | + case 'a': { |
| 71 | + // TODO: Poison middle element |
| 72 | + // Safe |
| 73 | + __v4sf v = maskedload_ps_1010(); |
| 74 | + // CHECK-LA: 1802398080,0,25961,0 |
| 75 | + print_vector(v); |
| 76 | + return 0; |
| 77 | + } |
| 78 | + } |
| 79 | + } else { |
| 80 | + switch (argv[1][1]) { |
| 81 | + case '1': |
| 82 | + // CHECK-S1: ERROR: AddressSanitizer |
| 83 | + maskedstore_ps_0001(a); |
| 84 | + // Should have blown up |
| 85 | + break; |
| 86 | + case '6': |
| 87 | + // Safe |
| 88 | + maskedstore_ps_0110(a); |
| 89 | + // CHECK-S6: 1802398080,2,3 |
| 90 | + print_vector3(g_vec3); |
| 91 | + return 0; |
| 92 | + case 'a': |
| 93 | + // TODO: Poison middle element |
| 94 | + // Safe |
| 95 | + maskedstore_ps_1010(a); |
| 96 | + // CHECK-SA: 1,1881171328,3 |
| 97 | + print_vector3(g_vec3); |
| 98 | + return 0; |
| 99 | + } |
| 100 | + } |
| 101 | + assert(0); |
| 102 | +} |
0 commit comments