55 unsigned long long uq;
78#if defined(ARCH_X86_X32) || defined(ARCH_X86_64)
81 register int rval = 0;
83 __asm__ __volatile__ (
89 "movl %%eax, %%ecx\n\t"
93 "xorl $0x200000, %%eax\n\t"
102 "xorl %%eax, %%ecx\n\t"
103 "testl $0x200000, %%ecx\n\t"
104 "jz NotSupported1\n\t"
113 "cmpl $0x756e6547, %%ebx\n\t"
115 "cmpl $0x49656e69, %%edx\n\t"
117 "cmpl $0x6c65746e, %%ecx\n"
123 "cmpl $0x68747541, %%ebx\n\t"
125 "cmpl $0x69746e65, %%edx\n\t"
127 "cmpl $0x444d4163, %%ecx\n"
133 "cmpl $0x69727943, %%ebx\n\t"
134 "jne NotSupported2\n\t"
135 "cmpl $0x736e4978, %%edx\n\t"
136 "jne NotSupported3\n\t"
137 "cmpl $0x64616574, %%ecx\n\t"
138 "jne NotSupported4\n\t"
151 "cmpl $0x2, %%eax\n\t"
156 "movl $0x80000001, %%eax\n\t"
158 "testl $0x00800000, %%eax\n\t"
159 "jz NotSupported5\n\t"
160 "testl $0x01000000, %%eax\n\t"
161 "jnz EMMXSupported\n\t"
173 "movl $0x80000000, %%eax\n\t"
175 "cmpl $0x80000000, %%eax\n\t"
179 "movl $0x80000001, %%eax\n\t"
181 "testl $0x00800000, %%edx\n\t"
182 "jz NotSupported6\n\t"
183 "testl $0x80000000, %%edx\n\t"
184 "jnz ThreeDNowSupported\n\t"
187 "ThreeDNowSupported:\n\t"
188 "testl $0x40000000, %%edx\n\t"
189 "jnz AMDXMMXSupported\n\t"
192 "AMDXMMXSupported:\n\t"
204 "testl $0x00800000, %%edx\n\t"
205 "jz NotSupported7\n\t"
210 "\nNotSupported1:\n\t"
211 "#movl $101, %0\n\n\t"
212 "\nNotSupported2:\n\t"
213 "#movl $102, %0\n\n\t"
214 "\nNotSupported3:\n\t"
215 "#movl $103, %0\n\n\t"
216 "\nNotSupported4:\n\t"
217 "#movl $104, %0\n\n\t"
218 "\nNotSupported5:\n\t"
219 "#movl $105, %0\n\n\t"
220 "\nNotSupported6:\n\t"
221 "#movl $106, %0\n\n\t"
222 "\nNotSupported7:\n\t"
223 "#movl $107, %0\n\n\t"
230 :
"eax",
"ecx",
"edx"
255 int *brutS,
int *brutD,
int buffratio,
int precalCoef[16][16]);
257 int *lbruS,
int *lbruD,
int buffratio,
int precalCoef[16][16]);
273#define mmx_i2r(op, imm, reg) \
276 mmx_trace.uq = (imm); \
277 printf(#op "_i2r(" #imm "=0x%08x%08x, ", \
278 mmx_trace.d[1], mmx_trace.d[0]); \
279 __asm__ __volatile__ ("movq %%" #reg ", %0" \
282 printf(#reg "=0x%08x%08x) => ", \
283 mmx_trace.d[1], mmx_trace.d[0]); \
284 __asm__ __volatile__ (#op " %0, %%" #reg \
287 __asm__ __volatile__ ("movq %%" #reg ", %0" \
290 printf(#reg "=0x%08x%08x\n", \
291 mmx_trace.d[1], mmx_trace.d[0]); \
294#define mmx_m2r(op, mem, reg) \
298 printf(#op "_m2r(" #mem "=0x%08x%08x, ", \
299 mmx_trace.d[1], mmx_trace.d[0]); \
300 __asm__ __volatile__ ("movq %%" #reg ", %0" \
303 printf(#reg "=0x%08x%08x) => ", \
304 mmx_trace.d[1], mmx_trace.d[0]); \
305 __asm__ __volatile__ (#op " %0, %%" #reg \
308 __asm__ __volatile__ ("movq %%" #reg ", %0" \
311 printf(#reg "=0x%08x%08x\n", \
312 mmx_trace.d[1], mmx_trace.d[0]); \
315#define mmx_r2m(op, reg, mem) \
318 __asm__ __volatile__ ("movq %%" #reg ", %0" \
321 printf(#op "_r2m(" #reg "=0x%08x%08x, ", \
322 mmx_trace.d[1], mmx_trace.d[0]); \
324 printf(#mem "=0x%08x%08x) => ", \
325 mmx_trace.d[1], mmx_trace.d[0]); \
326 __asm__ __volatile__ (#op " %%" #reg ", %0" \
330 printf(#mem "=0x%08x%08x\n", \
331 mmx_trace.d[1], mmx_trace.d[0]); \
334#define mmx_r2r(op, regs, regd) \
337 __asm__ __volatile__ ("movq %%" #regs ", %0" \
340 printf(#op "_r2r(" #regs "=0x%08x%08x, ", \
341 mmx_trace.d[1], mmx_trace.d[0]); \
342 __asm__ __volatile__ ("movq %%" #regd ", %0" \
345 printf(#regd "=0x%08x%08x) => ", \
346 mmx_trace.d[1], mmx_trace.d[0]); \
347 __asm__ __volatile__ (#op " %" #regs ", %" #regd); \
348 __asm__ __volatile__ ("movq %%" #regd ", %0" \
351 printf(#regd "=0x%08x%08x\n", \
352 mmx_trace.d[1], mmx_trace.d[0]); \
355#define mmx_m2m(op, mems, memd) \
358 mmx_trace = (mems); \
359 printf(#op "_m2m(" #mems "=0x%08x%08x, ", \
360 mmx_trace.d[1], mmx_trace.d[0]); \
361 mmx_trace = (memd); \
362 printf(#memd "=0x%08x%08x) => ", \
363 mmx_trace.d[1], mmx_trace.d[0]); \
364 __asm__ __volatile__ ("movq %0, %%mm0\n\t" \
365 #op " %1, %%mm0\n\t" \
369 mmx_trace = (memd); \
370 printf(#memd "=0x%08x%08x\n", \
371 mmx_trace.d[1], mmx_trace.d[0]); \
379#define mmx_i2r(op, imm, reg) \
380 __asm__ __volatile__ (#op " %0, %%" #reg \
384#define mmx_m2r(op, mem, reg) \
385 __asm__ __volatile__ (#op " %0, %%" #reg \
389#define mmx_r2m(op, reg, mem) \
390 __asm__ __volatile__ (#op " %%" #reg ", %0" \
394#define mmx_r2r(op, regs, regd) \
395 __asm__ __volatile__ (#op " %" #regs ", %" #regd)
397#define mmx_m2m(op, mems, memd) \
398 __asm__ __volatile__ ("movq %0, %%mm0\n\t" \
399 #op " %1, %%mm0\n\t" \
411#define movq_m2r(var, reg) mmx_m2r(movq, var, reg)
412#define movq_r2m(reg, var) mmx_r2m(movq, reg, var)
413#define movq_r2r(regs, regd) mmx_r2r(movq, regs, regd)
414#define movq(vars, vard) \
415 __asm__ __volatile__ ("movq %1, %%mm0\n\t" \
426#define movd_m2r(var, reg) mmx_m2r(movd, var, reg)
427#define movd_r2m(reg, var) mmx_r2m(movd, reg, var)
428#define movd_r2r(regs, regd) mmx_r2r(movd, regs, regd)
429#define movd(vars, vard) \
430 __asm__ __volatile__ ("movd %1, %%mm0\n\t" \
438#define paddd_m2r(var, reg) mmx_m2r(paddd, var, reg)
439#define paddd_r2r(regs, regd) mmx_r2r(paddd, regs, regd)
440#define paddd(vars, vard) mmx_m2m(paddd, vars, vard)
442#define paddw_m2r(var, reg) mmx_m2r(paddw, var, reg)
443#define paddw_r2r(regs, regd) mmx_r2r(paddw, regs, regd)
444#define paddw(vars, vard) mmx_m2m(paddw, vars, vard)
446#define paddb_m2r(var, reg) mmx_m2r(paddb, var, reg)
447#define paddb_r2r(regs, regd) mmx_r2r(paddb, regs, regd)
448#define paddb(vars, vard) mmx_m2m(paddb, vars, vard)
453#define paddsw_m2r(var, reg) mmx_m2r(paddsw, var, reg)
454#define paddsw_r2r(regs, regd) mmx_r2r(paddsw, regs, regd)
455#define paddsw(vars, vard) mmx_m2m(paddsw, vars, vard)
457#define paddsb_m2r(var, reg) mmx_m2r(paddsb, var, reg)
458#define paddsb_r2r(regs, regd) mmx_r2r(paddsb, regs, regd)
459#define paddsb(vars, vard) mmx_m2m(paddsb, vars, vard)
464#define paddusw_m2r(var, reg) mmx_m2r(paddusw, var, reg)
465#define paddusw_r2r(regs, regd) mmx_r2r(paddusw, regs, regd)
466#define paddusw(vars, vard) mmx_m2m(paddusw, vars, vard)
468#define paddusb_m2r(var, reg) mmx_m2r(paddusb, var, reg)
469#define paddusb_r2r(regs, regd) mmx_r2r(paddusb, regs, regd)
470#define paddusb(vars, vard) mmx_m2m(paddusb, vars, vard)
475#define psubd_m2r(var, reg) mmx_m2r(psubd, var, reg)
476#define psubd_r2r(regs, regd) mmx_r2r(psubd, regs, regd)
477#define psubd(vars, vard) mmx_m2m(psubd, vars, vard)
479#define psubw_m2r(var, reg) mmx_m2r(psubw, var, reg)
480#define psubw_r2r(regs, regd) mmx_r2r(psubw, regs, regd)
481#define psubw(vars, vard) mmx_m2m(psubw, vars, vard)
483#define psubb_m2r(var, reg) mmx_m2r(psubb, var, reg)
484#define psubb_r2r(regs, regd) mmx_r2r(psubb, regs, regd)
485#define psubb(vars, vard) mmx_m2m(psubb, vars, vard)
490#define psubsw_m2r(var, reg) mmx_m2r(psubsw, var, reg)
491#define psubsw_r2r(regs, regd) mmx_r2r(psubsw, regs, regd)
492#define psubsw(vars, vard) mmx_m2m(psubsw, vars, vard)
494#define psubsb_m2r(var, reg) mmx_m2r(psubsb, var, reg)
495#define psubsb_r2r(regs, regd) mmx_r2r(psubsb, regs, regd)
496#define psubsb(vars, vard) mmx_m2m(psubsb, vars, vard)
501#define psubusw_m2r(var, reg) mmx_m2r(psubusw, var, reg)
502#define psubusw_r2r(regs, regd) mmx_r2r(psubusw, regs, regd)
503#define psubusw(vars, vard) mmx_m2m(psubusw, vars, vard)
505#define psubusb_m2r(var, reg) mmx_m2r(psubusb, var, reg)
506#define psubusb_r2r(regs, regd) mmx_r2r(psubusb, regs, regd)
507#define psubusb(vars, vard) mmx_m2m(psubusb, vars, vard)
512#define pmullw_m2r(var, reg) mmx_m2r(pmullw, var, reg)
513#define pmullw_r2r(regs, regd) mmx_r2r(pmullw, regs, regd)
514#define pmullw(vars, vard) mmx_m2m(pmullw, vars, vard)
519#define pmulhw_m2r(var, reg) mmx_m2r(pmulhw, var, reg)
520#define pmulhw_r2r(regs, regd) mmx_r2r(pmulhw, regs, regd)
521#define pmulhw(vars, vard) mmx_m2m(pmulhw, vars, vard)
528#define pmaddwd_m2r(var, reg) mmx_m2r(pmaddwd, var, reg)
529#define pmaddwd_r2r(regs, regd) mmx_r2r(pmaddwd, regs, regd)
530#define pmaddwd(vars, vard) mmx_m2m(pmaddwd, vars, vard)
536#define pand_m2r(var, reg) \
538 mmx_m2r(pandn, (mmx_t) -1LL, reg); \
539 mmx_m2r(pandn, var, reg); \
541#define pand_r2r(regs, regd) \
543 mmx_m2r(pandn, (mmx_t) -1LL, regd); \
544 mmx_r2r(pandn, regs, regd) \
546#define pand(vars, vard) \
548 movq_m2r(vard, mm0); \
549 mmx_m2r(pandn, (mmx_t) -1LL, mm0); \
550 mmx_m2r(pandn, vars, mm0); \
551 movq_r2m(mm0, vard); \
554#define pand_m2r(var, reg) mmx_m2r(pand, var, reg)
555#define pand_r2r(regs, regd) mmx_r2r(pand, regs, regd)
556#define pand(vars, vard) mmx_m2m(pand, vars, vard)
562#define pandn_m2r(var, reg) mmx_m2r(pandn, var, reg)
563#define pandn_r2r(regs, regd) mmx_r2r(pandn, regs, regd)
564#define pandn(vars, vard) mmx_m2m(pandn, vars, vard)
569#define por_m2r(var, reg) mmx_m2r(por, var, reg)
570#define por_r2r(regs, regd) mmx_r2r(por, regs, regd)
571#define por(vars, vard) mmx_m2m(por, vars, vard)
576#define pxor_m2r(var, reg) mmx_m2r(pxor, var, reg)
577#define pxor_r2r(regs, regd) mmx_r2r(pxor, regs, regd)
578#define pxor(vars, vard) mmx_m2m(pxor, vars, vard)
584#define pcmpeqd_m2r(var, reg) mmx_m2r(pcmpeqd, var, reg)
585#define pcmpeqd_r2r(regs, regd) mmx_r2r(pcmpeqd, regs, regd)
586#define pcmpeqd(vars, vard) mmx_m2m(pcmpeqd, vars, vard)
588#define pcmpeqw_m2r(var, reg) mmx_m2r(pcmpeqw, var, reg)
589#define pcmpeqw_r2r(regs, regd) mmx_r2r(pcmpeqw, regs, regd)
590#define pcmpeqw(vars, vard) mmx_m2m(pcmpeqw, vars, vard)
592#define pcmpeqb_m2r(var, reg) mmx_m2r(pcmpeqb, var, reg)
593#define pcmpeqb_r2r(regs, regd) mmx_r2r(pcmpeqb, regs, regd)
594#define pcmpeqb(vars, vard) mmx_m2m(pcmpeqb, vars, vard)
600#define pcmpgtd_m2r(var, reg) mmx_m2r(pcmpgtd, var, reg)
601#define pcmpgtd_r2r(regs, regd) mmx_r2r(pcmpgtd, regs, regd)
602#define pcmpgtd(vars, vard) mmx_m2m(pcmpgtd, vars, vard)
604#define pcmpgtw_m2r(var, reg) mmx_m2r(pcmpgtw, var, reg)
605#define pcmpgtw_r2r(regs, regd) mmx_r2r(pcmpgtw, regs, regd)
606#define pcmpgtw(vars, vard) mmx_m2m(pcmpgtw, vars, vard)
608#define pcmpgtb_m2r(var, reg) mmx_m2r(pcmpgtb, var, reg)
609#define pcmpgtb_r2r(regs, regd) mmx_r2r(pcmpgtb, regs, regd)
610#define pcmpgtb(vars, vard) mmx_m2m(pcmpgtb, vars, vard)
615#define psllq_i2r(imm, reg) mmx_i2r(psllq, imm, reg)
616#define psllq_m2r(var, reg) mmx_m2r(psllq, var, reg)
617#define psllq_r2r(regs, regd) mmx_r2r(psllq, regs, regd)
618#define psllq(vars, vard) mmx_m2m(psllq, vars, vard)
620#define pslld_i2r(imm, reg) mmx_i2r(pslld, imm, reg)
621#define pslld_m2r(var, reg) mmx_m2r(pslld, var, reg)
622#define pslld_r2r(regs, regd) mmx_r2r(pslld, regs, regd)
623#define pslld(vars, vard) mmx_m2m(pslld, vars, vard)
625#define psllw_i2r(imm, reg) mmx_i2r(psllw, imm, reg)
626#define psllw_m2r(var, reg) mmx_m2r(psllw, var, reg)
627#define psllw_r2r(regs, regd) mmx_r2r(psllw, regs, regd)
628#define psllw(vars, vard) mmx_m2m(psllw, vars, vard)
633#define psrlq_i2r(imm, reg) mmx_i2r(psrlq, imm, reg)
634#define psrlq_m2r(var, reg) mmx_m2r(psrlq, var, reg)
635#define psrlq_r2r(regs, regd) mmx_r2r(psrlq, regs, regd)
636#define psrlq(vars, vard) mmx_m2m(psrlq, vars, vard)
638#define psrld_i2r(imm, reg) mmx_i2r(psrld, imm, reg)
639#define psrld_m2r(var, reg) mmx_m2r(psrld, var, reg)
640#define psrld_r2r(regs, regd) mmx_r2r(psrld, regs, regd)
641#define psrld(vars, vard) mmx_m2m(psrld, vars, vard)
643#define psrlw_i2r(imm, reg) mmx_i2r(psrlw, imm, reg)
644#define psrlw_m2r(var, reg) mmx_m2r(psrlw, var, reg)
645#define psrlw_r2r(regs, regd) mmx_r2r(psrlw, regs, regd)
646#define psrlw(vars, vard) mmx_m2m(psrlw, vars, vard)
651#define psrad_i2r(imm, reg) mmx_i2r(psrad, imm, reg)
652#define psrad_m2r(var, reg) mmx_m2r(psrad, var, reg)
653#define psrad_r2r(regs, regd) mmx_r2r(psrad, regs, regd)
654#define psrad(vars, vard) mmx_m2m(psrad, vars, vard)
656#define psraw_i2r(imm, reg) mmx_i2r(psraw, imm, reg)
657#define psraw_m2r(var, reg) mmx_m2r(psraw, var, reg)
658#define psraw_r2r(regs, regd) mmx_r2r(psraw, regs, regd)
659#define psraw(vars, vard) mmx_m2m(psraw, vars, vard)
665#define packssdw_m2r(var, reg) mmx_m2r(packssdw, var, reg)
666#define packssdw_r2r(regs, regd) mmx_r2r(packssdw, regs, regd)
667#define packssdw(vars, vard) mmx_m2m(packssdw, vars, vard)
669#define packsswb_m2r(var, reg) mmx_m2r(packsswb, var, reg)
670#define packsswb_r2r(regs, regd) mmx_r2r(packsswb, regs, regd)
671#define packsswb(vars, vard) mmx_m2m(packsswb, vars, vard)
677#define packuswb_m2r(var, reg) mmx_m2r(packuswb, var, reg)
678#define packuswb_r2r(regs, regd) mmx_r2r(packuswb, regs, regd)
679#define packuswb(vars, vard) mmx_m2m(packuswb, vars, vard)
686#define punpckldq_m2r(var, reg) mmx_m2r(punpckldq, var, reg)
687#define punpckldq_r2r(regs, regd) mmx_r2r(punpckldq, regs, regd)
688#define punpckldq(vars, vard) mmx_m2m(punpckldq, vars, vard)
690#define punpcklwd_m2r(var, reg) mmx_m2r(punpcklwd, var, reg)
691#define punpcklwd_r2r(regs, regd) mmx_r2r(punpcklwd, regs, regd)
692#define punpcklwd(vars, vard) mmx_m2m(punpcklwd, vars, vard)
694#define punpcklbw_m2r(var, reg) mmx_m2r(punpcklbw, var, reg)
695#define punpcklbw_r2r(regs, regd) mmx_r2r(punpcklbw, regs, regd)
696#define punpcklbw(vars, vard) mmx_m2m(punpcklbw, vars, vard)
703#define punpckhdq_m2r(var, reg) mmx_m2r(punpckhdq, var, reg)
704#define punpckhdq_r2r(regs, regd) mmx_r2r(punpckhdq, regs, regd)
705#define punpckhdq(vars, vard) mmx_m2m(punpckhdq, vars, vard)
707#define punpckhwd_m2r(var, reg) mmx_m2r(punpckhwd, var, reg)
708#define punpckhwd_r2r(regs, regd) mmx_r2r(punpckhwd, regs, regd)
709#define punpckhwd(vars, vard) mmx_m2m(punpckhwd, vars, vard)
711#define punpckhbw_m2r(var, reg) mmx_m2r(punpckhbw, var, reg)
712#define punpckhbw_r2r(regs, regd) mmx_r2r(punpckhbw, regs, regd)
713#define punpckhbw(vars, vard) mmx_m2m(punpckhbw, vars, vard)
726 printf("emms()\n"); \
727 __asm__ __volatile__ ("emms");
732#define emms() __asm__ __volatile__ ("emms")
#define ATTR_ALIGN(align)
Definition: attributes.h:66
void zoom_filter_mmx(int prevX, int prevY, Pixel *expix1, Pixel *expix2, int *brutS, int *brutD, int buffratio, int precalCoef[16][16])
static int mmx_ok(void)
Definition: mmx.h:241
void zoom_filter_xmmx(int prevX, int prevY, Pixel *expix1, Pixel *expix2, int *lbruS, int *lbruD, int buffratio, int precalCoef[16][16])
void draw_line_xmmx(Pixel *data, int x1, int y1, int x2, int y2, int col, int screenx, int screeny)
static int mm_support(void)
Definition: mmx.h:70
void draw_line_mmx(Pixel *data, int x1, int y1, int x2, int y2, int col, int screenx, int screeny)
Definition: goom_graphic.h:55
long long q
Definition: mmx.h:54
unsigned long long uq
Definition: mmx.h:55