Overview
| Comment: | OFMatrix4x4: Restore SSE4.1 code
The Clang alignment bug has been worked around in the previous commit, |
|---|---|
| Downloads: | Tarball | ZIP archive | SQL archive |
| Timelines: | family | ancestors | descendants | both | trunk |
| Files: | files | file ages | folders |
| SHA3-256: |
0eb97e46123c1283123d0774270f36a9 |
| User & Date: | js on 2023-11-04 13:10:43 |
| Other Links: | manifest | tags |
Context
|
2023-11-04
| ||
| 13:13 | Make GCC happy again check-in: 998478014d user: js tags: trunk | |
| 13:10 | OFMatrix4x4: Restore SSE4.1 code check-in: 0eb97e4612 user: js tags: trunk | |
| 13:06 | Work around Clang not aligning ivars correctly check-in: 055e14fc75 user: js tags: trunk | |
|
2023-11-03
| ||
| 01:07 | OFMatrix4x4: Remove SSE4.1 due to Clang bugs check-in: 7e1dbda4b4 user: js tags: trunk | |
Changes
Modified src/OFMatrix4x4.m from [1cf8596c69] to [52a22d3f60].
| ︙ | ︙ | |||
26 27 28 29 30 31 32 33 34 35 36 37 38 39 |
{ 0, 1, 0, 0 },
{ 0, 0, 1, 0 },
{ 0, 0, 0, 1 }
};
@implementation OFMatrix4x4
#if (defined(OF_AMD64) || defined(OF_X86)) && defined(__GNUC__)
# ifndef __clang__
# pragma GCC push_options
# pragma GCC target("3dnow,3dnowa")
# endif
static void
multiplyWithMatrix_enhanced3DNow(OFMatrix4x4 *self, SEL _cmd,
OFMatrix4x4 *matrix)
| > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > | 26 27 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 |
{ 0, 1, 0, 0 },
{ 0, 0, 1, 0 },
{ 0, 0, 0, 1 }
};
@implementation OFMatrix4x4
#if (defined(OF_AMD64) || defined(OF_X86)) && defined(__GNUC__)
# ifndef __clang__
# pragma GCC push_options
# pragma GCC target("sse4.1")
# endif
static void
transformVectors_SSE41(OFMatrix4x4 *self, SEL _cmd, OFVector4D *vectors,
size_t count)
{
__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 48(%2), %%xmm3\n"
"\n\t"
"0:\n\t"
"movaps (%1), %%xmm4\n\t"
"movaps %%xmm4, %%xmm5\n\t"
"dpps $0xFF, %%xmm0, %%xmm4\n\t"
"movaps %%xmm5, %%xmm6\n\t"
"dpps $0xFF, %%xmm1, %%xmm5\n\t"
"movaps %%xmm6, %%xmm7\n\t"
"dpps $0xFF, %%xmm2, %%xmm6\n\t"
"dpps $0xFF, %%xmm3, %%xmm7\n\t"
"insertps $0x10, %%xmm5, %%xmm4\n\t"
"insertps $0x20, %%xmm6, %%xmm4\n\t"
"insertps $0x30, %%xmm7, %%xmm4\n\t"
"movaps %%xmm4, (%1)\n"
"\n\t"
"add $16, %1\n\t"
"dec %0\n\t"
"jnz 0b\n"
: "+r"(count), "+r"(vectors)
: "r"(self->_values)
: "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7",
"memory"
);
}
# ifndef __clang__
# pragma GCC pop_options
# endif
# ifndef __clang__
# pragma GCC push_options
# pragma GCC target("3dnow,3dnowa")
# endif
static void
multiplyWithMatrix_enhanced3DNow(OFMatrix4x4 *self, SEL _cmd,
OFMatrix4x4 *matrix)
|
| ︙ | ︙ | |||
262 263 264 265 266 267 268 | return; # define REPLACE(selector, func) \ typeEncoding = method_getTypeEncoding( \ class_getInstanceMethod(self, selector)); \ class_replaceMethod(self, selector, (IMP)func, typeEncoding); | | > > > | 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 |
return;
# define REPLACE(selector, func) \
typeEncoding = method_getTypeEncoding( \
class_getInstanceMethod(self, selector)); \
class_replaceMethod(self, selector, (IMP)func, typeEncoding);
if ([OFSystemInfo supportsSSE41]) {
REPLACE(@selector(transformVectors:count:),
transformVectors_SSE41)
} else if ([OFSystemInfo supports3DNow]) {
if ([OFSystemInfo supportsEnhanced3DNow]) {
REPLACE(@selector(multiplyWithMatrix:),
multiplyWithMatrix_enhanced3DNow)
REPLACE(@selector(transformVectors:count:),
transformVectors_enhanced3DNow)
} else {
REPLACE(@selector(multiplyWithMatrix:),
|
| ︙ | ︙ |