Differences From Artifact [6d7dc755b7]:
- File src/OFMatrix4x4.m — part of check-in [998478014d] at 2023-11-04 13:13:31 on branch trunk — Make GCC happy again (user: js, size: 11724) [annotate] [blame] [check-ins using]
To Artifact [66a4a58f6e]:
- File
src/OFMatrix4x4.m
— part of check-in
[34b48a4208]
at
2023-11-05 13:37:08
on branch trunk
— OFMatrix4x4: Improve 3DNow! implementation
The new 3DNow! implementation is now better than the Enhanced 3DNow!
implementation, hence this also deletes the Enhanced 3DNow!
implementation. (user: js, size: 8587) [annotate] [blame] [check-ins using]
| ︙ | ︙ | |||
70 71 72 73 74 75 76 | "memory" ); } # ifndef __clang__ # pragma GCC pop_options # endif | < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < | | < < | 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 |
"memory"
);
}
# ifndef __clang__
# pragma GCC pop_options
# endif
# ifndef __clang__
# pragma GCC push_options
# pragma GCC target("3dnow")
# endif
static void
multiplyWithMatrix_3DNow(OFMatrix4x4 *self, SEL _cmd, OFMatrix4x4 *matrix)
{
float (*left)[4] = matrix->_values, (*right)[4] = self->_values;
float result[4][4], (*resultPtr)[4] = result;
__asm__ __volatile__ (
"xorw %%cx, %%cx\n"
"\n\t"
"0:\n\t"
"movd (%2), %%mm0\n\t"
"punpckldq 16(%2), %%mm0\n\t"
"pfmul (%1), %%mm0\n\t"
"movd 32(%2), %%mm1\n\t"
"punpckldq 48(%2), %%mm1\n\t"
"pfmul 8(%1), %%mm1\n\t"
"pfacc %%mm1, %%mm0\n\t"
"pfacc %%mm0, %%mm0\n\t"
"movd %%mm0, (%0)\n"
"\n\t"
"add $4, %0\n\t"
"add $4, %2\n\t"
"incb %%cl\n\t"
"cmpb $4, %%cl\n\t"
"jb 0b\n"
|
| ︙ | ︙ | |||
243 244 245 246 247 248 249 | "movq (%1), %%mm0\n\t" "movq 8(%1), %%mm1\n" "\n\t" "movq %%mm0, %%mm2\n\t" "movq %%mm1, %%mm3\n\t" "pfmul (%2), %%mm2\n\t" "pfmul 8(%2), %%mm3\n\t" | | | < < | | < < | | < < | | < < | 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 | "movq (%1), %%mm0\n\t" "movq 8(%1), %%mm1\n" "\n\t" "movq %%mm0, %%mm2\n\t" "movq %%mm1, %%mm3\n\t" "pfmul (%2), %%mm2\n\t" "pfmul 8(%2), %%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(%2), %%mm3\n\t" "pfmul 24(%2), %%mm4\n\t" "pfacc %%mm4, %%mm3\n\t" "pfacc %%mm3, %%mm3\n\t" "\n\t" "punpckldq %%mm3, %%mm2\n\t" "movq %%mm2, (%1)\n" "\n\t" "movq %%mm0, %%mm2\n\t" "movq %%mm1, %%mm3\n\t" "pfmul 32(%2), %%mm2\n\t" "pfmul 40(%2), %%mm3\n\t" "pfacc %%mm3, %%mm2\n\t" "pfacc %%mm2, %%mm2\n\t" "\n\t" "pfmul 48(%2), %%mm0\n\t" "pfmul 56(%2), %%mm1\n\t" "pfacc %%mm1, %%mm0\n\t" "pfacc %%mm0, %%mm0\n\t" "\n\t" "punpckldq %%mm0, %%mm2\n\t" "movq %%mm2, 8(%1)\n" "\n\t" "add $16, %1\n\t" "dec %0\n\t" "jnz 0b\n" |
| ︙ | ︙ | |||
310 311 312 313 314 315 316 |
class_getInstanceMethod(self, selector)); \
class_replaceMethod(self, selector, (IMP)func, typeEncoding);
if ([OFSystemInfo supportsSSE41]) {
REPLACE(@selector(transformVectors:count:),
transformVectors_SSE41)
} else if ([OFSystemInfo supports3DNow]) {
| < | < < < < < | | | < | 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 |
class_getInstanceMethod(self, selector)); \
class_replaceMethod(self, selector, (IMP)func, typeEncoding);
if ([OFSystemInfo supportsSSE41]) {
REPLACE(@selector(transformVectors:count:),
transformVectors_SSE41)
} else if ([OFSystemInfo supports3DNow]) {
REPLACE(@selector(multiplyWithMatrix:),
multiplyWithMatrix_3DNow)
REPLACE(@selector(transformVectors:count:),
transformVectors_3DNow)
}
# undef REPLACE
}
#endif
+ (instancetype)alloc
|
| ︙ | ︙ |