1 #include <assert.h> 2 #include <stdlib.h> 3 #include <string.h> 4 5 #define SAT(x) (((x) < -32768) ? -32768 : (((x) > 32767) ? 32767 : (x))) 6 7 static void test_pabsb_c(char *pDst, char *pSrc, int xmm) 8 { 9 int i; 10 11 for ( i = 0; i < (8 << xmm); i++ ) 12 pDst[ i ] = pSrc[ i ] > 0 ? pSrc[i ] : -pSrc[ i ]; 13 } 14 15 static void test_pabsw_c(short *pDst, short *pSrc, int xmm) 16 { 17 int i; 18 19 for ( i = 0; i < (4 << xmm); i++ ) 20 pDst[ i ] = pSrc[ i ] > 0 ? pSrc[i ] : -pSrc[ i ]; 21 } 22 23 static void test_pabsd_c(int *pDst, int *pSrc, int xmm) 24 { 25 int i; 26 27 for ( i = 0; i < (2 << xmm); i++ ) 28 pDst[ i ] = pSrc[ i ] > 0 ? pSrc[i ] : -pSrc[ i ]; 29 } 30 31 static void test_psignb_c(char *pDst, char *pSrc, int xmm) 32 { 33 int i; 34 35 for ( i = 0; i < (8 << xmm); i++ ) 36 pDst[ i ] = pSrc[i] ? ( pSrc[ i ] >= 0 ? pDst[i ] : -pDst[ i ] ) : 0; 37 } 38 39 static void test_psignw_c(short *pDst, short *pSrc, int xmm) 40 { 41 int i; 42 43 for ( i = 0; i < (4 << xmm); i++ ) 44 pDst[ i ] = pSrc[i] ? ( pSrc[ i ] >= 0 ? pDst[i ] : -pDst[ i ] ) : 0; 45 } 46 47 static void test_psignd_c(int *pDst, int *pSrc, int xmm) 48 { 49 int i; 50 51 for ( i = 0; i < (2 << xmm); i++ ) 52 pDst[ i ] = pSrc[i] ? ( pSrc[ i ] >= 0 ? pDst[i ] : -pDst[ i ] ) : 0; 53 } 54 55 static void test_phaddw_c(unsigned short *pDst,unsigned short *pSrc, int xmm) 56 { 57 int i; 58 59 for ( i = 0; i < (2 << xmm); i++ ) 60 pDst[ i ] = pDst[ i * 2 ] + pDst[ i * 2 + 1 ]; 61 62 for ( i = 0; i < (2 << xmm); i++ ) 63 pDst[ i + (2 << xmm) ] = pSrc[ i * 2 ] + pSrc[ i * 2 + 1 ]; 64 } 65 66 static void test_phaddsw_c(short *pDst, short *pSrc, int xmm) 67 { 68 int i; 69 70 for ( i = 0; i < (2 << xmm); i++ ) 71 pDst[ i ] = SAT( pDst[ i * 2 ] + pDst[ i * 2 + 1 ] ); 72 73 for ( i = 0; i < (2 << xmm); i++ ) 74 pDst[ i + (2 << xmm) ] = SAT( pSrc[ i * 2 ] + pSrc[ i * 2 + 1 ] ); 75 } 76 77 static void test_phaddd_c(unsigned int *pDst, unsigned int *pSrc, int xmm) 78 { 79 int i; 80 81 for ( i = 0; i < (1 << xmm); i++ ) 82 pDst[ i ] = pDst[ i * 2 ] + pDst[ i * 2 + 1 ]; 83 84 for ( i = 0; i < (1 << xmm); i++ ) 85 pDst[ i + (1 << xmm) ] = pSrc[ i * 2 ] + pSrc[ i * 2 + 1 ]; 86 } 87 88 static void test_phsubw_c(unsigned short *pDst,unsigned short *pSrc, int xmm) 89 { 90 int i; 91 92 for ( i = 0; i < (2 << xmm); i++ ) 93 pDst[ i ] = pDst[ i * 2 ] - pDst[ i * 2 + 1 ]; 94 95 for ( i = 0; i < (2 << xmm); i++ ) 96 pDst[ i + (2 << xmm) ] = pSrc[ i * 2 ] - pSrc[ i * 2 + 1 ]; 97 } 98 99 static void test_phsubsw_c(short *pDst, short *pSrc, int xmm) 100 { 101 int i; 102 103 for ( i = 0; i < (2 << xmm); i++ ) 104 pDst[ i ] = SAT( pDst[ i * 2 ] - pDst[ i * 2 + 1 ] ); 105 106 for ( i = 0; i < (2 << xmm); i++ ) 107 pDst[ i + (2 << xmm) ] = SAT( pSrc[ i * 2 ] - pSrc[ i * 2 + 1 ] ); 108 } 109 110 static void test_phsubd_c(unsigned int *pDst, unsigned int *pSrc, int xmm) 111 { 112 int i; 113 114 for ( i = 0; i < (1 << xmm); i++ ) 115 pDst[ i ] = pDst[ i * 2 ] - pDst[ i * 2 + 1 ]; 116 117 for ( i = 0; i < (1 << xmm); i++ ) 118 pDst[ i + (1 << xmm) ] = pSrc[ i * 2 ] - pSrc[ i * 2 + 1 ]; 119 } 120 121 static void test_pmulhrsw_c(short *pDst, short *pSrc, int xmm) 122 { 123 int i; 124 125 for ( i = 0; i < (4 << xmm); i++ ) 126 { 127 int a = pSrc[ i ] * pDst[ i ]; 128 pDst[i] = (short)(((a >> 14) + 1) >> 1); 129 } 130 } 131 132 static void test_pmaddubsw_c(unsigned char *pDst, signed char *pSrc, int xmm) 133 { 134 int i; 135 136 for ( i = 0; i < (4 << xmm); i++ ) 137 { 138 int a = pSrc[ 2 * i ] * pDst[ 2 * i ] + pSrc[ 2 * i + 1 ] * pDst[ 2 * i + 1]; 139 ((signed short *)pDst)[i] = SAT(a); 140 } 141 } 142 143 static void test_pshufb_c(unsigned char *pDst, unsigned char *pSrc, int xmm) 144 { 145 unsigned char bla[16]; 146 int i; 147 148 memcpy( bla, pDst, ( 8 << xmm ) ); 149 150 for ( i = 0; i < (8 << xmm); i++ ) 151 pDst[ i ] = (pSrc[ i ] >= 0x80) ? 0 : bla[ pSrc[ i ] & ((1 << (xmm + 3)) - 1) ]; 152 } 153 154 static void test_palignr_c(unsigned char *pDst, unsigned char *pSrc, int xmm) 155 { 156 int i; 157 158 for ( i = 0; i < 3; i++ ) 159 pDst[ i + (8 << xmm) - 3 ] = pDst[ i ]; 160 161 for ( i = 3; i < (8 << xmm); i++ ) 162 pDst[ i - 3 ] = pSrc[ i ]; 163 } 164 165 static void randomize_args(unsigned char *pDst, unsigned char *pSrc) 166 { 167 int j; 168 for ( j = 0; j < 16; j++ ) 169 { 170 pDst[ j ] = rand() % 256; 171 pSrc[ j ] = rand() % 256; 172 } 173 } 174 175 #define CHECK_FUNCTION(instruction, extension, additionnal, pDst, pSrc) \ 176 do { \ 177 unsigned char temp_dst[16]; \ 178 unsigned char temp_src[16]; \ 179 randomize_args( pDst, pSrc ); \ 180 memcpy( temp_dst, pDst, 16 ); \ 181 memcpy( temp_src, pSrc, 16 ); \ 182 test_##instruction##_c( pDst, pSrc, additionnal ); \ 183 test_##instruction##_##extension( temp_dst, temp_src ); \ 184 assert( !memcmp( pDst, temp_dst, (8 << additionnal) ) ); \ 185 } while( 0 ) 186 187 #define CHECK_FUNCTIONS(instruction) \ 188 CHECK_FUNCTION(instruction, mmx, 0, pDst, pSrc); \ 189 CHECK_FUNCTION(instruction, xmm, 1, pDst, pSrc) 190 191 192 void main(int nArgC, char *pArgv[]) 193 { 194 void *pSrc = malloc(16); 195 void *pDst = malloc(16); 196 int nIter = atoi( pArgv[ 1 ] ); 197 int i; 198 199 for ( i = 0; i < nIter; i++ ) 200 { 201 CHECK_FUNCTIONS( psignb ); 202 CHECK_FUNCTIONS( psignw ); 203 CHECK_FUNCTIONS( psignd ); 204 205 CHECK_FUNCTIONS( pabsb ); 206 CHECK_FUNCTIONS( pabsw ); 207 CHECK_FUNCTIONS( pabsd ); 208 209 CHECK_FUNCTIONS( phaddw ); 210 CHECK_FUNCTIONS( phaddsw ); 211 CHECK_FUNCTIONS( phaddd ); 212 213 CHECK_FUNCTIONS( phsubw ); 214 CHECK_FUNCTIONS( phsubsw ); 215 CHECK_FUNCTIONS( phsubd ); 216 217 CHECK_FUNCTIONS( pmulhrsw ); 218 CHECK_FUNCTIONS( pmaddubsw ); 219 220 CHECK_FUNCTIONS( pshufb ); 221 CHECK_FUNCTIONS( palignr ); 222 } 223 }
