Differences From Artifact [b31d5e9464]:
- File
src/OFMatrix4x4.m
— part of check-in
[3280466d35]
at
2023-11-05 18:18:05
on branch trunk
— OFMatrix4x4: Unroll inner loop in 3DNow! version
This results in a ~ 16% performance improvement on a Duron 750. (user: js, size: 9256) [annotate] [blame] [check-ins using]
To Artifact [ac56c8b767]:
- File
src/OFMatrix4x4.m
— part of check-in
[cf955413ab]
at
2023-11-06 00:59:37
on branch trunk
— OFMatrix4x4: SSE1 for -[transformVectors:count:]
This new SSE1 implementation is better than the SSE4.1 implementation,
hence this also deletes the SSE4.1 implementation. (user: js, size: 9770) [annotate] [blame] [check-ins using]
| ︙ | ︙ | |||
28 29 30 31 32 33 34 |
{ 0, 0, 0, 1 }
};
@implementation OFMatrix4x4
#if (defined(OF_AMD64) || defined(OF_X86)) && defined(__GNUC__)
# ifndef __clang__
# pragma GCC push_options
| | | > > | > | | | > > > > > | | > > > > > | | > > > > | > | | | > | > > > | | 28 29 30 31 32 33 34 35 36 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 |
{ 0, 0, 0, 1 }
};
@implementation OFMatrix4x4
#if (defined(OF_AMD64) || defined(OF_X86)) && defined(__GNUC__)
# ifndef __clang__
# pragma GCC push_options
# pragma GCC target("sse")
# endif
static void
transformVectors_SSE(OFMatrix4x4 *self, SEL _cmd, OFVector4D *vectors,
size_t count)
{
OF_ALIGN(16) float tmp[4];
__asm__ __volatile__ (
"test %0, %0\n\t"
"jz 0f\n"
"\n\t"
"movaps (%2), %%xmm0\n\t"
"movaps 16(%2), %%xmm1\n\t"
"movaps 32(%2), %%xmm2\n\t"
"movaps (%1), %%xmm3\n"
"\n\t"
"0:\n\t"
"movaps %%xmm0, %%xmm4\n\t"
"mulps %%xmm3, %%xmm4\n\t"
"movaps %%xmm4, (%3)\n\t"
"addss 4(%3), %%xmm4\n\t"
"addss 8(%3), %%xmm4\n\t"
"addss 12(%3), %%xmm4\n"
"\n\t"
"movaps %%xmm1, %%xmm5\n\t"
"mulps %%xmm3, %%xmm5\n\t"
"movaps %%xmm5, (%3)\n\t"
"addss 4(%3), %%xmm5\n\t"
"addss 8(%3), %%xmm5\n\t"
"addss 12(%3), %%xmm5\n"
"\n\t"
"movaps %%xmm2, %%xmm6\n\t"
"mulps %%xmm3, %%xmm6\n\t"
"movaps %%xmm6, (%3)\n\t"
"addss 4(%3), %%xmm6\n\t"
"addss 8(%3), %%xmm6\n\t"
"addss 12(%3), %%xmm6\n"
"\n\t"
"movaps 48(%2), %%xmm7\n\t"
"mulps %%xmm3, %%xmm7\n\t"
"movaps %%xmm7, (%3)\n\t"
"addss 4(%3), %%xmm7\n\t"
"addss 8(%3), %%xmm7\n\t"
"addss 12(%3), %%xmm7\n"
"\n\t"
"movss %%xmm4, (%1)\n\t"
"movss %%xmm5, 4(%1)\n\t"
"movss %%xmm6, 8(%1)\n\t"
"movss %%xmm7, 12(%1)\n"
"\n\t"
"add $16, %1\n\t"
"dec %0\n\t"
"jnz 0b\n"
: "+r"(count), "+r"(vectors)
: "r"(self->_values), "r"(&tmp)
: "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7",
"memory"
);
}
# ifndef __clang__
# pragma GCC pop_options
# endif
|
| ︙ | ︙ | |||
205 206 207 208 209 210 211 | return; # define REPLACE(selector, func) \ typeEncoding = method_getTypeEncoding( \ class_getInstanceMethod(self, selector)); \ class_replaceMethod(self, selector, (IMP)func, typeEncoding); | | | | 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 |
return;
# define REPLACE(selector, func) \
typeEncoding = method_getTypeEncoding( \
class_getInstanceMethod(self, selector)); \
class_replaceMethod(self, selector, (IMP)func, typeEncoding);
if ([OFSystemInfo supportsSSE]) {
REPLACE(@selector(transformVectors:count:),
transformVectors_SSE)
} else if ([OFSystemInfo supports3DNow]) {
REPLACE(@selector(multiplyWithMatrix:),
multiplyWithMatrix_3DNow)
REPLACE(@selector(transformVectors:count:),
transformVectors_3DNow)
}
|
| ︙ | ︙ |