#include "libavcodec/hpeldsp.h"
#include "dsputil_mmx.h"
-//#undef NDEBUG
-//#include <assert.h>
-
-#if HAVE_YASM
void ff_put_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
ptrdiff_t line_size, int h);
void ff_put_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,
ptrdiff_t line_size, int h);
void ff_avg_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels,
ptrdiff_t line_size, int h);
-#endif /* HAVE_YASM */
#if HAVE_INLINE_ASM
-#define JUMPALIGN() __asm__ volatile (".p2align 3"::)
-#define MOVQ_ZERO(regd) __asm__ volatile ("pxor %%"#regd", %%"#regd ::)
-
-#define MOVQ_BFE(regd) \
- __asm__ volatile ( \
- "pcmpeqd %%"#regd", %%"#regd" \n\t" \
- "paddb %%"#regd", %%"#regd" \n\t" ::)
-
-#ifndef PIC
-#define MOVQ_BONE(regd) __asm__ volatile ("movq %0, %%"#regd" \n\t" :: "m"(ff_bone))
-#define MOVQ_WTWO(regd) __asm__ volatile ("movq %0, %%"#regd" \n\t" :: "m"(ff_wtwo))
-#else
-// for shared library it's better to use this way for accessing constants
-// pcmpeqd -> -1
-#define MOVQ_BONE(regd) \
- __asm__ volatile ( \
- "pcmpeqd %%"#regd", %%"#regd" \n\t" \
- "psrlw $15, %%"#regd" \n\t" \
- "packuswb %%"#regd", %%"#regd" \n\t" ::)
-
-#define MOVQ_WTWO(regd) \
- __asm__ volatile ( \
- "pcmpeqd %%"#regd", %%"#regd" \n\t" \
- "psrlw $15, %%"#regd" \n\t" \
- "psllw $1, %%"#regd" \n\t"::)
-
-#endif
-
-// using regr as temporary and for the output result
-// first argument is unmodifed and second is trashed
-// regfe is supposed to contain 0xfefefefefefefefe
-#define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \
- "movq "#rega", "#regr" \n\t" \
- "pand "#regb", "#regr" \n\t" \
- "pxor "#rega", "#regb" \n\t" \
- "pand "#regfe", "#regb" \n\t" \
- "psrlq $1, "#regb" \n\t" \
- "paddb "#regb", "#regr" \n\t"
-
-#define PAVGB_MMX(rega, regb, regr, regfe) \
- "movq "#rega", "#regr" \n\t" \
- "por "#regb", "#regr" \n\t" \
- "pxor "#rega", "#regb" \n\t" \
- "pand "#regfe", "#regb" \n\t" \
- "psrlq $1, "#regb" \n\t" \
- "psubb "#regb", "#regr" \n\t"
-
-// mm6 is supposed to contain 0xfefefefefefefefe
-#define PAVGBP_MMX_NO_RND(rega, regb, regr, regc, regd, regp) \
- "movq "#rega", "#regr" \n\t" \
- "movq "#regc", "#regp" \n\t" \
- "pand "#regb", "#regr" \n\t" \
- "pand "#regd", "#regp" \n\t" \
- "pxor "#rega", "#regb" \n\t" \
- "pxor "#regc", "#regd" \n\t" \
- "pand %%mm6, "#regb" \n\t" \
- "pand %%mm6, "#regd" \n\t" \
- "psrlq $1, "#regb" \n\t" \
- "psrlq $1, "#regd" \n\t" \
- "paddb "#regb", "#regr" \n\t" \
- "paddb "#regd", "#regp" \n\t"
-
-#define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \
- "movq "#rega", "#regr" \n\t" \
- "movq "#regc", "#regp" \n\t" \
- "por "#regb", "#regr" \n\t" \
- "por "#regd", "#regp" \n\t" \
- "pxor "#rega", "#regb" \n\t" \
- "pxor "#regc", "#regd" \n\t" \
- "pand %%mm6, "#regb" \n\t" \
- "pand %%mm6, "#regd" \n\t" \
- "psrlq $1, "#regd" \n\t" \
- "psrlq $1, "#regb" \n\t" \
- "psubb "#regb", "#regr" \n\t" \
- "psubb "#regd", "#regp" \n\t"
-
/***********************************/
/* MMX no rounding */
#define NO_RND 1
}
#endif /* HAVE_INLINE_ASM */
-void ff_put_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
- ptrdiff_t line_size, int h);
-void ff_avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
- ptrdiff_t line_size, int h);
-
#define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \
do { \
c->PFX ## _pixels_tab IDX [0] = PFX ## _pixels ## SIZE ## _ ## CPU; \
c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_mmxext;
c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_mmxext;
}
-#endif /* HAVE_YASM */
-#if HAVE_MMXEXT_EXTERNAL
if (flags & CODEC_FLAG_BITEXACT && CONFIG_VP3_DECODER) {
c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_exact_mmxext;
c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_exact_mmxext;
}
-#endif /* HAVE_MMXEXT_EXTERNAL */
+#endif /* HAVE_YASM */
}
static void hpeldsp_init_3dnow(HpelDSPContext *c, int flags, int mm_flags)
static void hpeldsp_init_sse2(HpelDSPContext *c, int flags, int mm_flags)
{
-#if HAVE_SSE2_EXTERNAL
+#if HAVE_YASM
if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {
// these functions are slower than mmx on AMD, but faster on Intel
c->put_pixels_tab[0][0] = ff_put_pixels16_sse2;
c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2;
c->avg_pixels_tab[0][0] = ff_avg_pixels16_sse2;
}
-#endif /* HAVE_SSE2_EXTERNAL */
+#endif /* HAVE_YASM */
}
void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)