Differences From Artifact [a4f9a5f812]:
- File
src/OFMatrix4x4.m
— part of check-in
[5edf0d083d]
at
2023-11-06 20:17:44
on branch trunk
— OFMatrix4x4: Use an extra SSE register on AMD64
This gives another nice speed improvement. (user: js, size: 9943) [annotate] [blame] [check-ins using]
To Artifact [c5d68e8b03]:
- File src/OFMatrix4x4.m — part of check-in [1b22456db6] at 2023-11-09 21:09:29 on branch trunk — Use named operands for __asm__ (user: js, size: 10533) [annotate] [blame] [check-ins using]
︙ | ︙ | |||
37 38 39 40 41 42 43 | static void transformVectors_SSE(OFMatrix4x4 *self, SEL _cmd, OFVector4D *vectors, size_t count) { OF_ALIGN(16) float tmp[4]; __asm__ __volatile__ ( | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | > | | > | 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 | static void transformVectors_SSE(OFMatrix4x4 *self, SEL _cmd, OFVector4D *vectors, size_t count) { OF_ALIGN(16) float tmp[4]; __asm__ __volatile__ ( "test %[count], %[count]\n\t" "jz 0f\n" "\n\t" "movaps (%[matrix]), %%xmm0\n\t" "movaps 16(%[matrix]), %%xmm1\n\t" "movaps 32(%[matrix]), %%xmm2\n\t" # ifdef OF_AMD64 "movaps 48(%[matrix]), %%xmm8\n" # endif "\n\t" "0:\n\t" "movaps (%[vectors]), %%xmm3\n" "\n\t" "movaps %%xmm0, %%xmm4\n\t" "mulps %%xmm3, %%xmm4\n\t" "movaps %%xmm4, (%[tmp])\n\t" "addss 4(%[tmp]), %%xmm4\n\t" "addss 8(%[tmp]), %%xmm4\n\t" "addss 12(%[tmp]), %%xmm4\n" "\n\t" "movaps %%xmm1, %%xmm5\n\t" "mulps %%xmm3, %%xmm5\n\t" "movaps %%xmm5, (%[tmp])\n\t" "addss 4(%[tmp]), %%xmm5\n\t" "addss 8(%[tmp]), %%xmm5\n\t" "addss 12(%[tmp]), %%xmm5\n" "\n\t" "movaps %%xmm2, %%xmm6\n\t" "mulps %%xmm3, %%xmm6\n\t" "movaps %%xmm6, (%[tmp])\n\t" "addss 4(%[tmp]), %%xmm6\n\t" "addss 8(%[tmp]), %%xmm6\n\t" "addss 12(%[tmp]), %%xmm6\n" "\n\t" # ifdef OF_AMD64 "movaps %%xmm8, %%xmm7\n\t" # else "movaps 48(%[matrix]), %%xmm7\n\t" # endif "mulps %%xmm3, %%xmm7\n\t" "movaps %%xmm7, (%[tmp])\n\t" "addss 4(%[tmp]), %%xmm7\n\t" "addss 8(%[tmp]), %%xmm7\n\t" "addss 12(%[tmp]), %%xmm7\n" "\n\t" "movss %%xmm4, (%[vectors])\n\t" "movss %%xmm5, 4(%[vectors])\n\t" "movss %%xmm6, 8(%[vectors])\n\t" "movss %%xmm7, 12(%[vectors])\n" "\n\t" "add $16, %[vectors]\n\t" "dec %[count]\n\t" "jnz 0b\n" : [count] "+r" (count), [vectors] "+r" (vectors) : [matrix] "r" (self->_values), [tmp] "r" (&tmp) : "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7", # ifdef OF_AMD64 "xmm8", # endif "memory" ); } |
︙ | ︙ | |||
117 118 119 120 121 122 123 | float (*left)[4] = matrix->_values, (*right)[4] = self->_values; float result[4][4], (*resultPtr)[4] = result; __asm__ __volatile__ ( "movl $4, %%ecx\n\t" "\n\t" "0:\n\t" | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | > > > | | | | | | | | | | | | | | | | > | | | 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 | float (*left)[4] = matrix->_values, (*right)[4] = self->_values; float result[4][4], (*resultPtr)[4] = result; __asm__ __volatile__ ( "movl $4, %%ecx\n\t" "\n\t" "0:\n\t" "movd (%[right]), %%mm0\n\t" "punpckldq 16(%[right]), %%mm0\n\t" "pfmul (%[left]), %%mm0\n\t" "movd 32(%[right]), %%mm1\n\t" "punpckldq 48(%[right]), %%mm1\n\t" "pfmul 8(%[left]), %%mm1\n\t" "pfacc %%mm1, %%mm0\n\t" "pfacc %%mm0, %%mm0\n\t" "movd %%mm0, (%[result])\n\t" "movd 4(%[right]), %%mm0\n\t" "punpckldq 20(%[right]), %%mm0\n\t" "pfmul (%[left]), %%mm0\n\t" "movd 36(%[right]), %%mm1\n\t" "punpckldq 52(%[right]), %%mm1\n\t" "pfmul 8(%[left]), %%mm1\n\t" "pfacc %%mm1, %%mm0\n\t" "pfacc %%mm0, %%mm0\n\t" "movd %%mm0, 4(%[result])\n\t" "movd 8(%[right]), %%mm0\n\t" "punpckldq 24(%[right]), %%mm0\n\t" "pfmul (%[left]), %%mm0\n\t" "movd 40(%[right]), %%mm1\n\t" "punpckldq 56(%[right]), %%mm1\n\t" "pfmul 8(%[left]), %%mm1\n\t" "pfacc %%mm1, %%mm0\n\t" "pfacc %%mm0, %%mm0\n\t" "movd %%mm0, 8(%[result])\n\t" "movd 12(%[right]), %%mm0\n\t" "punpckldq 28(%[right]), %%mm0\n\t" "pfmul (%[left]), %%mm0\n\t" "movd 44(%[right]), %%mm1\n\t" "punpckldq 60(%[right]), %%mm1\n\t" "pfmul 8(%[left]), %%mm1\n\t" "pfacc %%mm1, %%mm0\n\t" "pfacc %%mm0, %%mm0\n\t" "movd %%mm0, 12(%[result])\n" "\n\t" "add $16, %[result]\n\t" "add $16, %[left]\n\t" "decl %%ecx\n\t" "jnz 0b\n" "\n\t" "femms" : [result] "+r" (resultPtr), [left] "+r" (left), [right] "+r" (right) : : "ecx", "mm0", "mm1", "memory" ); memcpy(self->_values, result, 16 * sizeof(float)); } static void transformVectors_3DNow(OFMatrix4x4 *self, SEL _cmd, OFVector4D *vectors, size_t count) { __asm__ __volatile__ ( "test %[count], %[count]\n\t" "jz 0f\n" "\n\t" "0:\n\t" "movq (%[vectors]), %%mm0\n\t" "movq 8(%[vectors]), %%mm1\n" "\n\t" "movq %%mm0, %%mm2\n\t" "movq %%mm1, %%mm3\n\t" "pfmul (%[matrix]), %%mm2\n\t" "pfmul 8(%[matrix]), %%mm3\n\t" "pfacc %%mm3, %%mm2\n\t" "pfacc %%mm2, %%mm2\n\t" "\n\t" "movq %%mm0, %%mm3\n\t" "movq %%mm1, %%mm4\n\t" "pfmul 16(%[matrix]), %%mm3\n\t" "pfmul 24(%[matrix]), %%mm4\n\t" "pfacc %%mm4, %%mm3\n\t" "pfacc %%mm3, %%mm3\n\t" "\n\t" "punpckldq %%mm3, %%mm2\n\t" "movq %%mm2, (%[vectors])\n" "\n\t" "movq %%mm0, %%mm2\n\t" "movq %%mm1, %%mm3\n\t" "pfmul 32(%[matrix]), %%mm2\n\t" "pfmul 40(%[matrix]), %%mm3\n\t" "pfacc %%mm3, %%mm2\n\t" "pfacc %%mm2, %%mm2\n\t" "\n\t" "pfmul 48(%[matrix]), %%mm0\n\t" "pfmul 56(%[matrix]), %%mm1\n\t" "pfacc %%mm1, %%mm0\n\t" "pfacc %%mm0, %%mm0\n\t" "\n\t" "punpckldq %%mm0, %%mm2\n\t" "movq %%mm2, 8(%[vectors])\n" "\n\t" "add $16, %[vectors]\n\t" "dec %[count]\n\t" "jnz 0b\n" "\n\t" "0:\n\t" "femms" : [count] "+r" (count), [vectors] "+r" (vectors) : [matrix] "r" (self->_values) : "mm0", "mm1", "mm2", "mm3", "mm4", "memory" ); } # ifndef __clang__ # pragma GCC pop_options # endif |
︙ | ︙ |