x86: dsputil: Coalesce all init files
[ffmpeg.git] / libavcodec / x86 / dsputil_init.c
index d867c7c..1303757 100644 (file)
@@ -28,7 +28,6 @@
 #include "libavutil/x86/cpu.h"
 #include "libavcodec/dsputil.h"
 #include "libavcodec/mpegvideo.h"
-#include "dsputil_x86.h"
 
 #if HAVE_INLINE_ASM
 
@@ -797,6 +796,419 @@ DCT_SAD_FUNC(ssse3)
 #undef HSUM
 #undef DCT_SAD
 
+
+DECLARE_ASM_CONST(8, uint64_t, round_tab)[3] = {
+    0x0000000000000000ULL,
+    0x0001000100010001ULL,
+    0x0002000200020002ULL,
+};
+
+DECLARE_ASM_CONST(8, uint64_t, bone) = 0x0101010101010101LL;
+
+static inline void sad8_1_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
+{
+    x86_reg len = -(stride * h);
+    __asm__ volatile (
+        ".p2align 4                     \n\t"
+        "1:                             \n\t"
+        "movq (%1, %%"REG_a"), %%mm0    \n\t"
+        "movq (%2, %%"REG_a"), %%mm2    \n\t"
+        "movq (%2, %%"REG_a"), %%mm4    \n\t"
+        "add %3, %%"REG_a"              \n\t"
+        "psubusb %%mm0, %%mm2           \n\t"
+        "psubusb %%mm4, %%mm0           \n\t"
+        "movq (%1, %%"REG_a"), %%mm1    \n\t"
+        "movq (%2, %%"REG_a"), %%mm3    \n\t"
+        "movq (%2, %%"REG_a"), %%mm5    \n\t"
+        "psubusb %%mm1, %%mm3           \n\t"
+        "psubusb %%mm5, %%mm1           \n\t"
+        "por %%mm2, %%mm0               \n\t"
+        "por %%mm1, %%mm3               \n\t"
+        "movq %%mm0, %%mm1              \n\t"
+        "movq %%mm3, %%mm2              \n\t"
+        "punpcklbw %%mm7, %%mm0         \n\t"
+        "punpckhbw %%mm7, %%mm1         \n\t"
+        "punpcklbw %%mm7, %%mm3         \n\t"
+        "punpckhbw %%mm7, %%mm2         \n\t"
+        "paddw %%mm1, %%mm0             \n\t"
+        "paddw %%mm3, %%mm2             \n\t"
+        "paddw %%mm2, %%mm0             \n\t"
+        "paddw %%mm0, %%mm6             \n\t"
+        "add %3, %%"REG_a"              \n\t"
+        " js 1b                         \n\t"
+        : "+a" (len)
+        : "r" (blk1 - len), "r" (blk2 - len), "r" ((x86_reg) stride));
+}
+
+static inline void sad8_1_mmxext(uint8_t *blk1, uint8_t *blk2,
+                                 int stride, int h)
+{
+    __asm__ volatile (
+        ".p2align 4                     \n\t"
+        "1:                             \n\t"
+        "movq (%1), %%mm0               \n\t"
+        "movq (%1, %3), %%mm1           \n\t"
+        "psadbw (%2), %%mm0             \n\t"
+        "psadbw (%2, %3), %%mm1         \n\t"
+        "paddw %%mm0, %%mm6             \n\t"
+        "paddw %%mm1, %%mm6             \n\t"
+        "lea (%1,%3,2), %1              \n\t"
+        "lea (%2,%3,2), %2              \n\t"
+        "sub $2, %0                     \n\t"
+        " jg 1b                         \n\t"
+        : "+r" (h), "+r" (blk1), "+r" (blk2)
+        : "r" ((x86_reg) stride));
+}
+
+static int sad16_sse2(MpegEncContext *v, uint8_t *blk2, uint8_t *blk1,
+                      int stride, int h)
+{
+    int ret;
+    __asm__ volatile (
+        "pxor %%xmm2, %%xmm2            \n\t"
+        ".p2align 4                     \n\t"
+        "1:                             \n\t"
+        "movdqu (%1), %%xmm0            \n\t"
+        "movdqu (%1, %4), %%xmm1        \n\t"
+        "psadbw (%2), %%xmm0            \n\t"
+        "psadbw (%2, %4), %%xmm1        \n\t"
+        "paddw %%xmm0, %%xmm2           \n\t"
+        "paddw %%xmm1, %%xmm2           \n\t"
+        "lea (%1,%4,2), %1              \n\t"
+        "lea (%2,%4,2), %2              \n\t"
+        "sub $2, %0                     \n\t"
+        " jg 1b                         \n\t"
+        "movhlps %%xmm2, %%xmm0         \n\t"
+        "paddw   %%xmm0, %%xmm2         \n\t"
+        "movd    %%xmm2, %3             \n\t"
+        : "+r" (h), "+r" (blk1), "+r" (blk2), "=r" (ret)
+        : "r" ((x86_reg) stride));
+    return ret;
+}
+
+static inline void sad8_x2a_mmxext(uint8_t *blk1, uint8_t *blk2,
+                                   int stride, int h)
+{
+    __asm__ volatile (
+        ".p2align 4                     \n\t"
+        "1:                             \n\t"
+        "movq (%1), %%mm0               \n\t"
+        "movq (%1, %3), %%mm1           \n\t"
+        "pavgb 1(%1), %%mm0             \n\t"
+        "pavgb 1(%1, %3), %%mm1         \n\t"
+        "psadbw (%2), %%mm0             \n\t"
+        "psadbw (%2, %3), %%mm1         \n\t"
+        "paddw %%mm0, %%mm6             \n\t"
+        "paddw %%mm1, %%mm6             \n\t"
+        "lea (%1,%3,2), %1              \n\t"
+        "lea (%2,%3,2), %2              \n\t"
+        "sub $2, %0                     \n\t"
+        " jg 1b                         \n\t"
+        : "+r" (h), "+r" (blk1), "+r" (blk2)
+        : "r" ((x86_reg) stride));
+}
+
+static inline void sad8_y2a_mmxext(uint8_t *blk1, uint8_t *blk2,
+                                   int stride, int h)
+{
+    __asm__ volatile (
+        "movq (%1), %%mm0               \n\t"
+        "add %3, %1                     \n\t"
+        ".p2align 4                     \n\t"
+        "1:                             \n\t"
+        "movq (%1), %%mm1               \n\t"
+        "movq (%1, %3), %%mm2           \n\t"
+        "pavgb %%mm1, %%mm0             \n\t"
+        "pavgb %%mm2, %%mm1             \n\t"
+        "psadbw (%2), %%mm0             \n\t"
+        "psadbw (%2, %3), %%mm1         \n\t"
+        "paddw %%mm0, %%mm6             \n\t"
+        "paddw %%mm1, %%mm6             \n\t"
+        "movq %%mm2, %%mm0              \n\t"
+        "lea (%1,%3,2), %1              \n\t"
+        "lea (%2,%3,2), %2              \n\t"
+        "sub $2, %0                     \n\t"
+        " jg 1b                         \n\t"
+        : "+r" (h), "+r" (blk1), "+r" (blk2)
+        : "r" ((x86_reg) stride));
+}
+
+static inline void sad8_4_mmxext(uint8_t *blk1, uint8_t *blk2,
+                                 int stride, int h)
+{
+    __asm__ volatile (
+        "movq "MANGLE(bone)", %%mm5     \n\t"
+        "movq (%1), %%mm0               \n\t"
+        "pavgb 1(%1), %%mm0             \n\t"
+        "add %3, %1                     \n\t"
+        ".p2align 4                     \n\t"
+        "1:                             \n\t"
+        "movq (%1), %%mm1               \n\t"
+        "movq (%1,%3), %%mm2            \n\t"
+        "pavgb 1(%1), %%mm1             \n\t"
+        "pavgb 1(%1,%3), %%mm2          \n\t"
+        "psubusb %%mm5, %%mm1           \n\t"
+        "pavgb %%mm1, %%mm0             \n\t"
+        "pavgb %%mm2, %%mm1             \n\t"
+        "psadbw (%2), %%mm0             \n\t"
+        "psadbw (%2,%3), %%mm1          \n\t"
+        "paddw %%mm0, %%mm6             \n\t"
+        "paddw %%mm1, %%mm6             \n\t"
+        "movq %%mm2, %%mm0              \n\t"
+        "lea (%1,%3,2), %1              \n\t"
+        "lea (%2,%3,2), %2              \n\t"
+        "sub $2, %0                     \n\t"
+        " jg 1b                         \n\t"
+        : "+r" (h), "+r" (blk1), "+r" (blk2)
+        : "r" ((x86_reg) stride));
+}
+
+static inline void sad8_2_mmx(uint8_t *blk1a, uint8_t *blk1b, uint8_t *blk2,
+                              int stride, int h)
+{
+    x86_reg len = -(stride * h);
+    __asm__ volatile (
+        ".p2align 4                     \n\t"
+        "1:                             \n\t"
+        "movq (%1, %%"REG_a"), %%mm0    \n\t"
+        "movq (%2, %%"REG_a"), %%mm1    \n\t"
+        "movq (%1, %%"REG_a"), %%mm2    \n\t"
+        "movq (%2, %%"REG_a"), %%mm3    \n\t"
+        "punpcklbw %%mm7, %%mm0         \n\t"
+        "punpcklbw %%mm7, %%mm1         \n\t"
+        "punpckhbw %%mm7, %%mm2         \n\t"
+        "punpckhbw %%mm7, %%mm3         \n\t"
+        "paddw %%mm0, %%mm1             \n\t"
+        "paddw %%mm2, %%mm3             \n\t"
+        "movq (%3, %%"REG_a"), %%mm4    \n\t"
+        "movq (%3, %%"REG_a"), %%mm2    \n\t"
+        "paddw %%mm5, %%mm1             \n\t"
+        "paddw %%mm5, %%mm3             \n\t"
+        "psrlw $1, %%mm1                \n\t"
+        "psrlw $1, %%mm3                \n\t"
+        "packuswb %%mm3, %%mm1          \n\t"
+        "psubusb %%mm1, %%mm4           \n\t"
+        "psubusb %%mm2, %%mm1           \n\t"
+        "por %%mm4, %%mm1               \n\t"
+        "movq %%mm1, %%mm0              \n\t"
+        "punpcklbw %%mm7, %%mm0         \n\t"
+        "punpckhbw %%mm7, %%mm1         \n\t"
+        "paddw %%mm1, %%mm0             \n\t"
+        "paddw %%mm0, %%mm6             \n\t"
+        "add %4, %%"REG_a"              \n\t"
+        " js 1b                         \n\t"
+        : "+a" (len)
+        : "r" (blk1a - len), "r" (blk1b - len), "r" (blk2 - len),
+          "r" ((x86_reg) stride));
+}
+
+static inline void sad8_4_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
+{
+    x86_reg len = -(stride * h);
+    __asm__ volatile (
+        "movq  (%1, %%"REG_a"), %%mm0   \n\t"
+        "movq 1(%1, %%"REG_a"), %%mm2   \n\t"
+        "movq %%mm0, %%mm1              \n\t"
+        "movq %%mm2, %%mm3              \n\t"
+        "punpcklbw %%mm7, %%mm0         \n\t"
+        "punpckhbw %%mm7, %%mm1         \n\t"
+        "punpcklbw %%mm7, %%mm2         \n\t"
+        "punpckhbw %%mm7, %%mm3         \n\t"
+        "paddw %%mm2, %%mm0             \n\t"
+        "paddw %%mm3, %%mm1             \n\t"
+        ".p2align 4                     \n\t"
+        "1:                             \n\t"
+        "movq  (%2, %%"REG_a"), %%mm2   \n\t"
+        "movq 1(%2, %%"REG_a"), %%mm4   \n\t"
+        "movq %%mm2, %%mm3              \n\t"
+        "movq %%mm4, %%mm5              \n\t"
+        "punpcklbw %%mm7, %%mm2         \n\t"
+        "punpckhbw %%mm7, %%mm3         \n\t"
+        "punpcklbw %%mm7, %%mm4         \n\t"
+        "punpckhbw %%mm7, %%mm5         \n\t"
+        "paddw %%mm4, %%mm2             \n\t"
+        "paddw %%mm5, %%mm3             \n\t"
+        "movq 16+"MANGLE(round_tab)", %%mm5 \n\t"
+        "paddw %%mm2, %%mm0             \n\t"
+        "paddw %%mm3, %%mm1             \n\t"
+        "paddw %%mm5, %%mm0             \n\t"
+        "paddw %%mm5, %%mm1             \n\t"
+        "movq (%3, %%"REG_a"), %%mm4    \n\t"
+        "movq (%3, %%"REG_a"), %%mm5    \n\t"
+        "psrlw $2, %%mm0                \n\t"
+        "psrlw $2, %%mm1                \n\t"
+        "packuswb %%mm1, %%mm0          \n\t"
+        "psubusb %%mm0, %%mm4           \n\t"
+        "psubusb %%mm5, %%mm0           \n\t"
+        "por %%mm4, %%mm0               \n\t"
+        "movq %%mm0, %%mm4              \n\t"
+        "punpcklbw %%mm7, %%mm0         \n\t"
+        "punpckhbw %%mm7, %%mm4         \n\t"
+        "paddw %%mm0, %%mm6             \n\t"
+        "paddw %%mm4, %%mm6             \n\t"
+        "movq  %%mm2, %%mm0             \n\t"
+        "movq  %%mm3, %%mm1             \n\t"
+        "add %4, %%"REG_a"              \n\t"
+        " js 1b                         \n\t"
+        : "+a" (len)
+        : "r" (blk1 - len), "r" (blk1 - len + stride), "r" (blk2 - len),
+          "r" ((x86_reg) stride));
+}
+
+static inline int sum_mmx(void)
+{
+    int ret;
+    __asm__ volatile (
+        "movq %%mm6, %%mm0              \n\t"
+        "psrlq $32, %%mm6               \n\t"
+        "paddw %%mm0, %%mm6             \n\t"
+        "movq %%mm6, %%mm0              \n\t"
+        "psrlq $16, %%mm6               \n\t"
+        "paddw %%mm0, %%mm6             \n\t"
+        "movd %%mm6, %0                 \n\t"
+        : "=r" (ret));
+    return ret & 0xFFFF;
+}
+
+static inline int sum_mmxext(void)
+{
+    int ret;
+    __asm__ volatile (
+        "movd %%mm6, %0                 \n\t"
+        : "=r" (ret));
+    return ret;
+}
+
+static inline void sad8_x2a_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
+{
+    sad8_2_mmx(blk1, blk1 + 1, blk2, stride, h);
+}
+
+static inline void sad8_y2a_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
+{
+    sad8_2_mmx(blk1, blk1 + stride, blk2, stride, h);
+}
+
+#define PIX_SAD(suf)                                                    \
+static int sad8_ ## suf(MpegEncContext *v, uint8_t *blk2,               \
+                        uint8_t *blk1, int stride, int h)               \
+{                                                                       \
+    assert(h == 8);                                                     \
+    __asm__ volatile (                                                  \
+        "pxor %%mm7, %%mm7     \n\t"                                    \
+        "pxor %%mm6, %%mm6     \n\t"                                    \
+        :);                                                             \
+                                                                        \
+    sad8_1_ ## suf(blk1, blk2, stride, 8);                              \
+                                                                        \
+    return sum_ ## suf();                                               \
+}                                                                       \
+                                                                        \
+static int sad8_x2_ ## suf(MpegEncContext *v, uint8_t *blk2,            \
+                           uint8_t *blk1, int stride, int h)            \
+{                                                                       \
+    assert(h == 8);                                                     \
+    __asm__ volatile (                                                  \
+        "pxor %%mm7, %%mm7     \n\t"                                    \
+        "pxor %%mm6, %%mm6     \n\t"                                    \
+        "movq %0, %%mm5        \n\t"                                    \
+        :: "m" (round_tab[1]));                                         \
+                                                                        \
+    sad8_x2a_ ## suf(blk1, blk2, stride, 8);                            \
+                                                                        \
+    return sum_ ## suf();                                               \
+}                                                                       \
+                                                                        \
+static int sad8_y2_ ## suf(MpegEncContext *v, uint8_t *blk2,            \
+                           uint8_t *blk1, int stride, int h)            \
+{                                                                       \
+    assert(h == 8);                                                     \
+    __asm__ volatile (                                                  \
+        "pxor %%mm7, %%mm7     \n\t"                                    \
+        "pxor %%mm6, %%mm6     \n\t"                                    \
+        "movq %0, %%mm5        \n\t"                                    \
+        :: "m" (round_tab[1]));                                         \
+                                                                        \
+    sad8_y2a_ ## suf(blk1, blk2, stride, 8);                            \
+                                                                        \
+    return sum_ ## suf();                                               \
+}                                                                       \
+                                                                        \
+static int sad8_xy2_ ## suf(MpegEncContext *v, uint8_t *blk2,           \
+                            uint8_t *blk1, int stride, int h)           \
+{                                                                       \
+    assert(h == 8);                                                     \
+    __asm__ volatile (                                                  \
+        "pxor %%mm7, %%mm7     \n\t"                                    \
+        "pxor %%mm6, %%mm6     \n\t"                                    \
+        ::);                                                            \
+                                                                        \
+    sad8_4_ ## suf(blk1, blk2, stride, 8);                              \
+                                                                        \
+    return sum_ ## suf();                                               \
+}                                                                       \
+                                                                        \
+static int sad16_ ## suf(MpegEncContext *v, uint8_t *blk2,              \
+                         uint8_t *blk1, int stride, int h)              \
+{                                                                       \
+    __asm__ volatile (                                                  \
+        "pxor %%mm7, %%mm7     \n\t"                                    \
+        "pxor %%mm6, %%mm6     \n\t"                                    \
+        :);                                                             \
+                                                                        \
+    sad8_1_ ## suf(blk1,     blk2,     stride, h);                      \
+    sad8_1_ ## suf(blk1 + 8, blk2 + 8, stride, h);                      \
+                                                                        \
+    return sum_ ## suf();                                               \
+}                                                                       \
+                                                                        \
+static int sad16_x2_ ## suf(MpegEncContext *v, uint8_t *blk2,           \
+                            uint8_t *blk1, int stride, int h)           \
+{                                                                       \
+    __asm__ volatile (                                                  \
+        "pxor %%mm7, %%mm7     \n\t"                                    \
+        "pxor %%mm6, %%mm6     \n\t"                                    \
+        "movq %0, %%mm5        \n\t"                                    \
+        :: "m" (round_tab[1]));                                         \
+                                                                        \
+    sad8_x2a_ ## suf(blk1,     blk2,     stride, h);                    \
+    sad8_x2a_ ## suf(blk1 + 8, blk2 + 8, stride, h);                    \
+                                                                        \
+    return sum_ ## suf();                                               \
+}                                                                       \
+                                                                        \
+static int sad16_y2_ ## suf(MpegEncContext *v, uint8_t *blk2,           \
+                            uint8_t *blk1, int stride, int h)           \
+{                                                                       \
+    __asm__ volatile (                                                  \
+        "pxor %%mm7, %%mm7     \n\t"                                    \
+        "pxor %%mm6, %%mm6     \n\t"                                    \
+        "movq %0, %%mm5        \n\t"                                    \
+        :: "m" (round_tab[1]));                                         \
+                                                                        \
+    sad8_y2a_ ## suf(blk1,     blk2,     stride, h);                    \
+    sad8_y2a_ ## suf(blk1 + 8, blk2 + 8, stride, h);                    \
+                                                                        \
+    return sum_ ## suf();                                               \
+}                                                                       \
+                                                                        \
+static int sad16_xy2_ ## suf(MpegEncContext *v, uint8_t *blk2,          \
+                             uint8_t *blk1, int stride, int h)          \
+{                                                                       \
+    __asm__ volatile (                                                  \
+        "pxor %%mm7, %%mm7     \n\t"                                    \
+        "pxor %%mm6, %%mm6     \n\t"                                    \
+        ::);                                                            \
+                                                                        \
+    sad8_4_ ## suf(blk1,     blk2,     stride, h);                      \
+    sad8_4_ ## suf(blk1 + 8, blk2 + 8, stride, h);                      \
+                                                                        \
+    return sum_ ## suf();                                               \
+}                                                                       \
+
+PIX_SAD(mmx)
+PIX_SAD(mmxext)
+
 #endif /* HAVE_INLINE_ASM */
 
 int ff_sse16_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
@@ -821,22 +1233,49 @@ av_cold void ff_dsputil_init_x86(DSPContext *c, AVCodecContext *avctx)
     if (INLINE_MMX(cpu_flags)) {
         c->sum_abs_dctelem = sum_abs_dctelem_mmx;
 
+        c->pix_abs[0][0] = sad16_mmx;
+        c->pix_abs[0][1] = sad16_x2_mmx;
+        c->pix_abs[0][2] = sad16_y2_mmx;
+        c->pix_abs[0][3] = sad16_xy2_mmx;
+        c->pix_abs[1][0] = sad8_mmx;
+        c->pix_abs[1][1] = sad8_x2_mmx;
+        c->pix_abs[1][2] = sad8_y2_mmx;
+        c->pix_abs[1][3] = sad8_xy2_mmx;
+
+        c->sad[0] = sad16_mmx;
+        c->sad[1] = sad8_mmx;
+
         c->sse[0]  = sse16_mmx;
         c->sse[1]  = sse8_mmx;
         c->vsad[4] = vsad_intra16_mmx;
 
         c->nsse[0] = nsse16_mmx;
         c->nsse[1] = nsse8_mmx;
+
         if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
-            c->vsad[0]      = vsad16_mmx;
+            c->vsad[0] = vsad16_mmx;
         }
     }
 
     if (INLINE_MMXEXT(cpu_flags)) {
         c->sum_abs_dctelem = sum_abs_dctelem_mmxext;
-        c->vsad[4]         = vsad_intra16_mmxext;
+
+        c->vsad[4] = vsad_intra16_mmxext;
+
+        c->pix_abs[0][0] = sad16_mmxext;
+        c->pix_abs[1][0] = sad8_mmxext;
+
+        c->sad[0] = sad16_mmxext;
+        c->sad[1] = sad8_mmxext;
 
         if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
+            c->pix_abs[0][1] = sad16_x2_mmxext;
+            c->pix_abs[0][2] = sad16_y2_mmxext;
+            c->pix_abs[0][3] = sad16_xy2_mmxext;
+            c->pix_abs[1][1] = sad8_x2_mmxext;
+            c->pix_abs[1][2] = sad8_y2_mmxext;
+            c->pix_abs[1][3] = sad8_xy2_mmxext;
+
             c->vsad[0] = vsad16_mmxext;
         }
     }
@@ -845,6 +1284,10 @@ av_cold void ff_dsputil_init_x86(DSPContext *c, AVCodecContext *avctx)
         c->sum_abs_dctelem = sum_abs_dctelem_sse2;
     }
 
+    if (INLINE_SSE2(cpu_flags) && !(cpu_flags & AV_CPU_FLAG_3DNOW)) {
+        c->sad[0] = sad16_sse2;
+    }
+
 #if HAVE_SSSE3_INLINE
     if (INLINE_SSSE3(cpu_flags)) {
         c->sum_abs_dctelem = sum_abs_dctelem_ssse3;
@@ -875,6 +1318,4 @@ av_cold void ff_dsputil_init_x86(DSPContext *c, AVCodecContext *avctx)
         c->hadamard8_diff[0] = ff_hadamard8_diff16_ssse3;
         c->hadamard8_diff[1] = ff_hadamard8_diff_ssse3;
     }
-
-    ff_dsputil_init_pix_mmx(c, avctx);
 }