basic type mechanism working

1 files changed, 70 insertions, 216 deletions

diff --git a/ffmpeg/libavcodec/x86/hpeldsp_init.c b/ffmpeg/libavcodec/x86/hpeldsp_init.c
index 4b877b8..8ecf909 100644
--- a/ffmpeg/libavcodec/x86/hpeldsp_init.c
+++ b/ffmpeg/libavcodec/x86/hpeldsp_init.c
@@ -24,13 +24,10 @@
 
 #include "libavutil/cpu.h"
 #include "libavutil/x86/asm.h"
+#include "libavutil/x86/cpu.h"
 #include "libavcodec/hpeldsp.h"
-#include "dsputil_mmx.h"
+#include "dsputil_x86.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,
@@ -77,103 +74,45 @@ void ff_avg_pixels8_xy2_mmxext(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 */
 
+#define avg_pixels8_mmx         ff_avg_pixels8_mmx
+#define avg_pixels8_x2_mmx      ff_avg_pixels8_x2_mmx
+#define avg_pixels16_mmx        ff_avg_pixels16_mmx
+#define avg_pixels8_xy2_mmx     ff_avg_pixels8_xy2_mmx
+#define avg_pixels16_xy2_mmx    ff_avg_pixels16_xy2_mmx
+#define put_pixels8_mmx         ff_put_pixels8_mmx
+#define put_pixels16_mmx        ff_put_pixels16_mmx
+#define put_pixels8_xy2_mmx     ff_put_pixels8_xy2_mmx
+#define put_pixels16_xy2_mmx    ff_put_pixels16_xy2_mmx
+#define avg_no_rnd_pixels16_mmx ff_avg_pixels16_mmx
+#define put_no_rnd_pixels8_mmx  ff_put_pixels8_mmx
+#define put_no_rnd_pixels16_mmx ff_put_pixels16_mmx
 
 #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
 #define DEF(x, y) x ## _no_rnd_ ## y ## _mmx
 #define SET_RND  MOVQ_WONE
 #define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
 #define PAVGB(a, b, c, e)               PAVGB_MMX_NO_RND(a, b, c, e)
-#define OP_AVG(a, b, c, e)              PAVGB_MMX(a, b, c, e)
+#define STATIC static
 
+#include "rnd_template.c"
 #include "hpeldsp_rnd_template.c"
 
 #undef DEF
 #undef SET_RND
 #undef PAVGBP
 #undef PAVGB
-#undef NO_RND
+#undef STATIC
+
+PIXELS16(static, avg_no_rnd, , _y2, _mmx)
+PIXELS16(static, put_no_rnd, , _y2, _mmx)
+
+PIXELS16(static, avg_no_rnd, , _xy2, _mmx)
+PIXELS16(static, put_no_rnd, , _xy2, _mmx)
+
 /***********************************/
 /* MMX rounding */
 
@@ -188,112 +127,29 @@ void ff_avg_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels,
 #undef SET_RND
 #undef PAVGBP
 #undef PAVGB
-#undef OP_AVG
+
+PIXELS16(static, avg, , _y2, _mmx)
+PIXELS16(static, put, , _y2, _mmx)
 
 #endif /* HAVE_INLINE_ASM */
 
 
 #if HAVE_YASM
-#define ff_put_pixels8_mmx ff_put_pixels8_mmxext
-
-/***********************************/
-/* 3Dnow specific */
 
-#define DEF(x) x ## _3dnow
+#define HPELDSP_AVG_PIXELS16(CPUEXT)                \
+    PIXELS16(static, put_no_rnd, ff_,  _x2, CPUEXT) \
+    PIXELS16(static, put,        ff_,  _y2, CPUEXT) \
+    PIXELS16(static, put_no_rnd, ff_,  _y2, CPUEXT) \
+    PIXELS16(static, avg,        ff_,     , CPUEXT) \
+    PIXELS16(static, avg,        ff_,  _x2, CPUEXT) \
+    PIXELS16(static, avg,        ff_,  _y2, CPUEXT) \
+    PIXELS16(static, avg,        ff_, _xy2, CPUEXT)
 
-#include "hpeldsp_avg_template.c"
-
-#undef DEF
-
-/***********************************/
-/* MMXEXT specific */
-
-#define DEF(x) x ## _mmxext
-
-#include "hpeldsp_avg_template.c"
-
-#undef DEF
+HPELDSP_AVG_PIXELS16(_3dnow)
+HPELDSP_AVG_PIXELS16(_mmxext)
 
 #endif /* HAVE_YASM */
 
-
-#if HAVE_INLINE_ASM
-#define put_no_rnd_pixels16_mmx put_pixels16_mmx
-#define put_no_rnd_pixels8_mmx put_pixels8_mmx
-#define put_pixels16_mmxext put_pixels16_mmx
-#define put_pixels8_mmxext put_pixels8_mmx
-#define put_pixels4_mmxext put_pixels4_mmx
-#define put_no_rnd_pixels16_mmxext put_no_rnd_pixels16_mmx
-#define put_no_rnd_pixels8_mmxext put_no_rnd_pixels8_mmx
-
-static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels,
-                            ptrdiff_t line_size, int h)
-{
-    __asm__ volatile (
-        "lea   (%3, %3), %%"REG_a"      \n\t"
-        ".p2align     3                 \n\t"
-        "1:                             \n\t"
-        "movq  (%1    ), %%mm0          \n\t"
-        "movq  (%1, %3), %%mm1          \n\t"
-        "movq     %%mm0, (%2)           \n\t"
-        "movq     %%mm1, (%2, %3)       \n\t"
-        "add  %%"REG_a", %1             \n\t"
-        "add  %%"REG_a", %2             \n\t"
-        "movq  (%1    ), %%mm0          \n\t"
-        "movq  (%1, %3), %%mm1          \n\t"
-        "movq     %%mm0, (%2)           \n\t"
-        "movq     %%mm1, (%2, %3)       \n\t"
-        "add  %%"REG_a", %1             \n\t"
-        "add  %%"REG_a", %2             \n\t"
-        "subl        $4, %0             \n\t"
-        "jnz         1b                 \n\t"
-        : "+g"(h), "+r"(pixels),  "+r"(block)
-        : "r"((x86_reg)line_size)
-        : "%"REG_a, "memory"
-        );
-}
-
-static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels,
-                             ptrdiff_t line_size, int h)
-{
-    __asm__ volatile (
-        "lea   (%3, %3), %%"REG_a"      \n\t"
-        ".p2align     3                 \n\t"
-        "1:                             \n\t"
-        "movq  (%1    ), %%mm0          \n\t"
-        "movq 8(%1    ), %%mm4          \n\t"
-        "movq  (%1, %3), %%mm1          \n\t"
-        "movq 8(%1, %3), %%mm5          \n\t"
-        "movq     %%mm0,  (%2)          \n\t"
-        "movq     %%mm4, 8(%2)          \n\t"
-        "movq     %%mm1,  (%2, %3)      \n\t"
-        "movq     %%mm5, 8(%2, %3)      \n\t"
-        "add  %%"REG_a", %1             \n\t"
-        "add  %%"REG_a", %2             \n\t"
-        "movq  (%1    ), %%mm0          \n\t"
-        "movq 8(%1    ), %%mm4          \n\t"
-        "movq  (%1, %3), %%mm1          \n\t"
-        "movq 8(%1, %3), %%mm5          \n\t"
-        "movq     %%mm0,  (%2)          \n\t"
-        "movq     %%mm4, 8(%2)          \n\t"
-        "movq     %%mm1,  (%2, %3)      \n\t"
-        "movq     %%mm5, 8(%2, %3)      \n\t"
-        "add  %%"REG_a", %1             \n\t"
-        "add  %%"REG_a", %2             \n\t"
-        "subl        $4, %0             \n\t"
-        "jnz         1b                 \n\t"
-        : "+g"(h), "+r"(pixels),  "+r"(block)
-        : "r"((x86_reg)line_size)
-        : "%"REG_a, "memory"
-        );
-}
-#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; \
@@ -302,9 +158,9 @@ void ff_avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
         c->PFX ## _pixels_tab IDX [3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU; \
     } while (0)
 
-static void hpeldsp_init_mmx(HpelDSPContext *c, int flags, int mm_flags)
+static void hpeldsp_init_mmx(HpelDSPContext *c, int flags, int cpu_flags)
 {
-#if HAVE_INLINE_ASM
+#if HAVE_MMX_INLINE
     SET_HPEL_FUNCS(put,        [0], 16, mmx);
     SET_HPEL_FUNCS(put_no_rnd, [0], 16, mmx);
     SET_HPEL_FUNCS(avg,        [0], 16, mmx);
@@ -312,18 +168,18 @@ static void hpeldsp_init_mmx(HpelDSPContext *c, int flags, int mm_flags)
     SET_HPEL_FUNCS(put,        [1],  8, mmx);
     SET_HPEL_FUNCS(put_no_rnd, [1],  8, mmx);
     SET_HPEL_FUNCS(avg,        [1],  8, mmx);
-#endif /* HAVE_INLINE_ASM */
+#endif /* HAVE_MMX_INLINE */
 }
 
-static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags, int mm_flags)
+static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags, int cpu_flags)
 {
-#if HAVE_YASM
+#if HAVE_MMXEXT_EXTERNAL
     c->put_pixels_tab[0][1] = ff_put_pixels16_x2_mmxext;
-    c->put_pixels_tab[0][2] = ff_put_pixels16_y2_mmxext;
+    c->put_pixels_tab[0][2] = put_pixels16_y2_mmxext;
 
-    c->avg_pixels_tab[0][0] = ff_avg_pixels16_mmxext;
-    c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_mmxext;
-    c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_mmxext;
+    c->avg_pixels_tab[0][0] = avg_pixels16_mmxext;
+    c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmxext;
+    c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmxext;
 
     c->put_pixels_tab[1][1] = ff_put_pixels8_x2_mmxext;
     c->put_pixels_tab[1][2] = ff_put_pixels8_y2_mmxext;
@@ -333,17 +189,15 @@ static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags, int mm_flags)
     c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_mmxext;
 
     if (!(flags & CODEC_FLAG_BITEXACT)) {
-        c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_mmxext;
-        c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_mmxext;
+        c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmxext;
+        c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmxext;
         c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_mmxext;
         c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_mmxext;
 
-        c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_mmxext;
+        c->avg_pixels_tab[0][3] = 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;
@@ -351,15 +205,15 @@ static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags, int mm_flags)
 #endif /* HAVE_MMXEXT_EXTERNAL */
 }
 
-static void hpeldsp_init_3dnow(HpelDSPContext *c, int flags, int mm_flags)
+static void hpeldsp_init_3dnow(HpelDSPContext *c, int flags, int cpu_flags)
 {
-#if HAVE_YASM
+#if HAVE_AMD3DNOW_EXTERNAL
     c->put_pixels_tab[0][1] = ff_put_pixels16_x2_3dnow;
-    c->put_pixels_tab[0][2] = ff_put_pixels16_y2_3dnow;
+    c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
 
-    c->avg_pixels_tab[0][0] = ff_avg_pixels16_3dnow;
-    c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_3dnow;
-    c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_3dnow;
+    c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
+    c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
+    c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
 
     c->put_pixels_tab[1][1] = ff_put_pixels8_x2_3dnow;
     c->put_pixels_tab[1][2] = ff_put_pixels8_y2_3dnow;
@@ -369,12 +223,12 @@ static void hpeldsp_init_3dnow(HpelDSPContext *c, int flags, int mm_flags)
     c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_3dnow;
 
     if (!(flags & CODEC_FLAG_BITEXACT)){
-        c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_3dnow;
-        c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_3dnow;
+        c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
+        c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
         c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_3dnow;
         c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_3dnow;
 
-        c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_3dnow;
+        c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
         c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_3dnow;
     }
 
@@ -382,13 +236,13 @@ static void hpeldsp_init_3dnow(HpelDSPContext *c, int flags, int mm_flags)
         c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_exact_3dnow;
         c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_exact_3dnow;
     }
-#endif /* HAVE_YASM */
+#endif /* HAVE_AMD3DNOW_EXTERNAL */
 }
 
-static void hpeldsp_init_sse2(HpelDSPContext *c, int flags, int mm_flags)
+static void hpeldsp_init_sse2(HpelDSPContext *c, int flags, int cpu_flags)
 {
 #if HAVE_SSE2_EXTERNAL
-    if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {
+    if (!(cpu_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;
@@ -399,17 +253,17 @@ static void hpeldsp_init_sse2(HpelDSPContext *c, int flags, int mm_flags)
 
 void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)
 {
-    int mm_flags = av_get_cpu_flags();
+    int cpu_flags = av_get_cpu_flags();
 
-    if (mm_flags & AV_CPU_FLAG_MMX)
-        hpeldsp_init_mmx(c, flags, mm_flags);
+    if (INLINE_MMX(cpu_flags))
+        hpeldsp_init_mmx(c, flags, cpu_flags);
 
-    if (mm_flags & AV_CPU_FLAG_MMXEXT)
-        hpeldsp_init_mmxext(c, flags, mm_flags);
+    if (EXTERNAL_AMD3DNOW(cpu_flags))
+        hpeldsp_init_3dnow(c, flags, cpu_flags);
 
-    if (mm_flags & AV_CPU_FLAG_3DNOW)
-        hpeldsp_init_3dnow(c, flags, mm_flags);
+    if (EXTERNAL_MMXEXT(cpu_flags))
+        hpeldsp_init_mmxext(c, flags, cpu_flags);
 
-    if (mm_flags & AV_CPU_FLAG_SSE2)
-        hpeldsp_init_sse2(c, flags, mm_flags);
+    if (EXTERNAL_SSE2(cpu_flags))
+        hpeldsp_init_sse2(c, flags, cpu_flags);
 }