Merge remote-tracking branch 'qatar/master'

* qatar/master:
  dsputil: x86: Fix compile error
  dsputil: x86: Convert h263 loop filter to yasm

Merged-by: Michael Niedermayer <michaelni@gmx.at>

2 files changed, 171 insertions, 177 deletions

diff --git a/libavcodec/x86/dsputil.asm b/libavcodec/x86/dsputil.asm
index 587d5ee968..02bec2d516 100644
--- a/libavcodec/x86/dsputil.asm
+++ b/libavcodec/x86/dsputil.asm
@@ -22,6 +22,8 @@
 %include "libavutil/x86/x86util.asm"
 
 SECTION_RODATA
+cextern pb_FC
+cextern h263_loop_filter_strength
 pb_f: times 16 db 15
 pb_zzzzzzzz77777777: times 8 db -1
 pb_7: times 8 db 7
@@ -648,3 +650,164 @@ BSWAP32_BUF
 
 INIT_XMM ssse3
 BSWAP32_BUF
+
+
+%macro H263_LOOP_FILTER 5
+    pxor         m7, m7
+    mova         m0, [%1]
+    mova         m1, [%1]
+    mova         m2, [%4]
+    mova         m3, [%4]
+    punpcklbw    m0, m7
+    punpckhbw    m1, m7
+    punpcklbw    m2, m7
+    punpckhbw    m3, m7
+    psubw        m0, m2
+    psubw        m1, m3
+    mova         m2, [%2]
+    mova         m3, [%2]
+    mova         m4, [%3]
+    mova         m5, [%3]
+    punpcklbw    m2, m7
+    punpckhbw    m3, m7
+    punpcklbw    m4, m7
+    punpckhbw    m5, m7
+    psubw        m4, m2
+    psubw        m5, m3
+    psllw        m4, 2
+    psllw        m5, 2
+    paddw        m4, m0
+    paddw        m5, m1
+    pxor         m6, m6
+    pcmpgtw      m6, m4
+    pcmpgtw      m7, m5
+    pxor         m4, m6
+    pxor         m5, m7
+    psubw        m4, m6
+    psubw        m5, m7
+    psrlw        m4, 3
+    psrlw        m5, 3
+    packuswb     m4, m5
+    packsswb     m6, m7
+    pxor         m7, m7
+    movd         m2, %5
+    punpcklbw    m2, m2
+    punpcklbw    m2, m2
+    punpcklbw    m2, m2
+    psubusb      m2, m4
+    mova         m3, m2
+    psubusb      m3, m4
+    psubb        m2, m3
+    mova         m3, [%2]
+    mova         m4, [%3]
+    pxor         m3, m6
+    pxor         m4, m6
+    paddusb      m3, m2
+    psubusb      m4, m2
+    pxor         m3, m6
+    pxor         m4, m6
+    paddusb      m2, m2
+    packsswb     m0, m1
+    pcmpgtb      m7, m0
+    pxor         m0, m7
+    psubb        m0, m7
+    mova         m1, m0
+    psubusb      m0, m2
+    psubb        m1, m0
+    pand         m1, [pb_FC]
+    psrlw        m1, 2
+    pxor         m1, m7
+    psubb        m1, m7
+    mova         m5, [%1]
+    mova         m6, [%4]
+    psubb        m5, m1
+    paddb        m6, m1
+%endmacro
+
+INIT_MMX mmx
+; void h263_v_loop_filter(uint8_t *src, int stride, int qscale)
+cglobal h263_v_loop_filter, 3,5
+    movsxdifnidn r1, r1d
+    movsxdifnidn r2, r2d
+
+    lea          r4, [h263_loop_filter_strength]
+    movzx       r3d, BYTE [r4+r2]
+    movsx        r2, r3b
+    shl          r2, 1
+
+    mov          r3, r0
+    sub          r3, r1
+    mov          r4, r3
+    sub          r4, r1
+    H263_LOOP_FILTER r4, r3, r0, r0+r1, r2d
+
+    mova       [r3], m3
+    mova       [r0], m4
+    mova       [r4], m5
+    mova    [r0+r1], m6
+    RET
+
+%macro TRANSPOSE4X4 2
+    movd      m0, [%1]
+    movd      m1, [%1+r1]
+    movd      m2, [%1+r1*2]
+    movd      m3, [%1+r3]
+    punpcklbw m0, m1
+    punpcklbw m2, m3
+    mova      m1, m0
+    punpcklwd m0, m2
+    punpckhwd m1, m2
+    movd [%2+ 0], m0
+    punpckhdq m0, m0
+    movd [%2+ 8], m0
+    movd [%2+16], m1
+    punpckhdq m1, m1
+    movd [%2+24], m1
+%endmacro
+
+
+; void h263_h_loop_filter(uint8_t *src, int stride, int qscale)
+INIT_MMX mmx
+cglobal h263_h_loop_filter, 3,5,0,32
+    movsxdifnidn r1, r1d
+    movsxdifnidn r2, r2d
+
+    lea          r4, [h263_loop_filter_strength]
+    movzx       r3d, BYTE [r4+r2]
+    movsx        r2, r3b
+    shl          r2, 1
+
+    sub          r0, 2
+    lea          r3, [r1*3]
+
+    TRANSPOSE4X4 r0, rsp
+    lea          r4, [r0+r1*4]
+    TRANSPOSE4X4 r4, rsp+4
+
+    H263_LOOP_FILTER rsp, rsp+8, rsp+16, rsp+24, r2d
+
+    mova         m1, m5
+    mova         m0, m4
+    punpcklbw    m5, m3
+    punpcklbw    m4, m6
+    punpckhbw    m1, m3
+    punpckhbw    m0, m6
+    mova         m3, m5
+    mova         m6, m1
+    punpcklwd    m5, m4
+    punpcklwd    m1, m0
+    punpckhwd    m3, m4
+    punpckhwd    m6, m0
+    movd       [r0], m5
+    punpckhdq    m5, m5
+    movd  [r0+r1*1], m5
+    movd  [r0+r1*2], m3
+    punpckhdq    m3, m3
+    movd    [r0+r3], m3
+    movd       [r4], m1
+    punpckhdq    m1, m1
+    movd  [r4+r1*1], m1
+    movd  [r4+r1*2], m6
+    punpckhdq    m6, m6
+    movd    [r4+r3], m6
+    RET
diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
index b528d4ba95..7155363fec 100644
--- a/libavcodec/x86/dsputil_mmx.c
+++ b/libavcodec/x86/dsputil_mmx.c
@@ -652,181 +652,12 @@ static void add_hfyu_median_prediction_cmov(uint8_t *dst, const uint8_t *top,
     *left_top = tl;
 }
 #endif
+#endif /* HAVE_INLINE_ASM */
 
-static inline void transpose4x4(uint8_t *dst, uint8_t *src, x86_reg dst_stride, x86_reg src_stride){
-    __asm__ volatile( //FIXME could save 1 instruction if done as 8x4 ...
-        "movd  (%1), %%mm0              \n\t"
-        "add   %3, %1                   \n\t"
-        "movd  (%1), %%mm1              \n\t"
-        "movd  (%1,%3,1), %%mm2         \n\t"
-        "movd  (%1,%3,2), %%mm3         \n\t"
-        "punpcklbw %%mm1, %%mm0         \n\t"
-        "punpcklbw %%mm3, %%mm2         \n\t"
-        "movq %%mm0, %%mm1              \n\t"
-        "punpcklwd %%mm2, %%mm0         \n\t"
-        "punpckhwd %%mm2, %%mm1         \n\t"
-        "movd  %%mm0, (%0)              \n\t"
-        "add   %2, %0                   \n\t"
-        "punpckhdq %%mm0, %%mm0         \n\t"
-        "movd  %%mm0, (%0)              \n\t"
-        "movd  %%mm1, (%0,%2,1)         \n\t"
-        "punpckhdq %%mm1, %%mm1         \n\t"
-        "movd  %%mm1, (%0,%2,2)         \n\t"
-
-        :  "+&r" (dst),
-           "+&r" (src)
-        :  "r" (dst_stride),
-           "r" (src_stride)
-        :  "memory"
-    );
-}
-
-#define H263_LOOP_FILTER                        \
-    "pxor      %%mm7, %%mm7             \n\t"   \
-    "movq         %0, %%mm0             \n\t"   \
-    "movq         %0, %%mm1             \n\t"   \
-    "movq         %3, %%mm2             \n\t"   \
-    "movq         %3, %%mm3             \n\t"   \
-    "punpcklbw %%mm7, %%mm0             \n\t"   \
-    "punpckhbw %%mm7, %%mm1             \n\t"   \
-    "punpcklbw %%mm7, %%mm2             \n\t"   \
-    "punpckhbw %%mm7, %%mm3             \n\t"   \
-    "psubw     %%mm2, %%mm0             \n\t"   \
-    "psubw     %%mm3, %%mm1             \n\t"   \
-    "movq         %1, %%mm2             \n\t"   \
-    "movq         %1, %%mm3             \n\t"   \
-    "movq         %2, %%mm4             \n\t"   \
-    "movq         %2, %%mm5             \n\t"   \
-    "punpcklbw %%mm7, %%mm2             \n\t"   \
-    "punpckhbw %%mm7, %%mm3             \n\t"   \
-    "punpcklbw %%mm7, %%mm4             \n\t"   \
-    "punpckhbw %%mm7, %%mm5             \n\t"   \
-    "psubw     %%mm2, %%mm4             \n\t"   \
-    "psubw     %%mm3, %%mm5             \n\t"   \
-    "psllw        $2, %%mm4             \n\t"   \
-    "psllw        $2, %%mm5             \n\t"   \
-    "paddw     %%mm0, %%mm4             \n\t"   \
-    "paddw     %%mm1, %%mm5             \n\t"   \
-    "pxor      %%mm6, %%mm6             \n\t"   \
-    "pcmpgtw   %%mm4, %%mm6             \n\t"   \
-    "pcmpgtw   %%mm5, %%mm7             \n\t"   \
-    "pxor      %%mm6, %%mm4             \n\t"   \
-    "pxor      %%mm7, %%mm5             \n\t"   \
-    "psubw     %%mm6, %%mm4             \n\t"   \
-    "psubw     %%mm7, %%mm5             \n\t"   \
-    "psrlw        $3, %%mm4             \n\t"   \
-    "psrlw        $3, %%mm5             \n\t"   \
-    "packuswb  %%mm5, %%mm4             \n\t"   \
-    "packsswb  %%mm7, %%mm6             \n\t"   \
-    "pxor      %%mm7, %%mm7             \n\t"   \
-    "movd         %4, %%mm2             \n\t"   \
-    "punpcklbw %%mm2, %%mm2             \n\t"   \
-    "punpcklbw %%mm2, %%mm2             \n\t"   \
-    "punpcklbw %%mm2, %%mm2             \n\t"   \
-    "psubusb   %%mm4, %%mm2             \n\t"   \
-    "movq      %%mm2, %%mm3             \n\t"   \
-    "psubusb   %%mm4, %%mm3             \n\t"   \
-    "psubb     %%mm3, %%mm2             \n\t"   \
-    "movq         %1, %%mm3             \n\t"   \
-    "movq         %2, %%mm4             \n\t"   \
-    "pxor      %%mm6, %%mm3             \n\t"   \
-    "pxor      %%mm6, %%mm4             \n\t"   \
-    "paddusb   %%mm2, %%mm3             \n\t"   \
-    "psubusb   %%mm2, %%mm4             \n\t"   \
-    "pxor      %%mm6, %%mm3             \n\t"   \
-    "pxor      %%mm6, %%mm4             \n\t"   \
-    "paddusb   %%mm2, %%mm2             \n\t"   \
-    "packsswb  %%mm1, %%mm0             \n\t"   \
-    "pcmpgtb   %%mm0, %%mm7             \n\t"   \
-    "pxor      %%mm7, %%mm0             \n\t"   \
-    "psubb     %%mm7, %%mm0             \n\t"   \
-    "movq      %%mm0, %%mm1             \n\t"   \
-    "psubusb   %%mm2, %%mm0             \n\t"   \
-    "psubb     %%mm0, %%mm1             \n\t"   \
-    "pand         %5, %%mm1             \n\t"   \
-    "psrlw        $2, %%mm1             \n\t"   \
-    "pxor      %%mm7, %%mm1             \n\t"   \
-    "psubb     %%mm7, %%mm1             \n\t"   \
-    "movq         %0, %%mm5             \n\t"   \
-    "movq         %3, %%mm6             \n\t"   \
-    "psubb     %%mm1, %%mm5             \n\t"   \
-    "paddb     %%mm1, %%mm6             \n\t"
-
-static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale)
-{
-    if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
-        const int strength = ff_h263_loop_filter_strength[qscale];
-
-        __asm__ volatile (
-            H263_LOOP_FILTER
-
-            "movq %%mm3, %1             \n\t"
-            "movq %%mm4, %2             \n\t"
-            "movq %%mm5, %0             \n\t"
-            "movq %%mm6, %3             \n\t"
-            : "+m"(*(uint64_t*)(src - 2 * stride)),
-              "+m"(*(uint64_t*)(src - 1 * stride)),
-              "+m"(*(uint64_t*)(src + 0 * stride)),
-              "+m"(*(uint64_t*)(src + 1 * stride))
-            : "g"(2 * strength), "m"(ff_pb_FC)
-            );
-    }
-}
-
-static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale)
-{
-    if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
-        const int strength = ff_h263_loop_filter_strength[qscale];
-        DECLARE_ALIGNED(8, uint64_t, temp)[4];
-        uint8_t *btemp = (uint8_t*)temp;
-
-        src -= 2;
-
-        transpose4x4(btemp,     src,              8, stride);
-        transpose4x4(btemp + 4, src + 4 * stride, 8, stride);
-        __asm__ volatile (
-            H263_LOOP_FILTER // 5 3 4 6
-
-            : "+m"(temp[0]),
-              "+m"(temp[1]),
-              "+m"(temp[2]),
-              "+m"(temp[3])
-            : "g"(2 * strength), "m"(ff_pb_FC)
-            );
-
-        __asm__ volatile (
-            "movq      %%mm5, %%mm1         \n\t"
-            "movq      %%mm4, %%mm0         \n\t"
-            "punpcklbw %%mm3, %%mm5         \n\t"
-            "punpcklbw %%mm6, %%mm4         \n\t"
-            "punpckhbw %%mm3, %%mm1         \n\t"
-            "punpckhbw %%mm6, %%mm0         \n\t"
-            "movq      %%mm5, %%mm3         \n\t"
-            "movq      %%mm1, %%mm6         \n\t"
-            "punpcklwd %%mm4, %%mm5         \n\t"
-            "punpcklwd %%mm0, %%mm1         \n\t"
-            "punpckhwd %%mm4, %%mm3         \n\t"
-            "punpckhwd %%mm0, %%mm6         \n\t"
-            "movd      %%mm5, (%0)          \n\t"
-            "punpckhdq %%mm5, %%mm5         \n\t"
-            "movd      %%mm5, (%0, %2)      \n\t"
-            "movd      %%mm3, (%0, %2, 2)   \n\t"
-            "punpckhdq %%mm3, %%mm3         \n\t"
-            "movd      %%mm3, (%0, %3)      \n\t"
-            "movd      %%mm1, (%1)          \n\t"
-            "punpckhdq %%mm1, %%mm1         \n\t"
-            "movd      %%mm1, (%1, %2)      \n\t"
-            "movd      %%mm6, (%1, %2, 2)   \n\t"
-            "punpckhdq %%mm6, %%mm6         \n\t"
-            "movd      %%mm6, (%1, %3)      \n\t"
-            :: "r"(src),
-               "r"(src + 4 * stride),
-               "r"((x86_reg)stride),
-               "r"((x86_reg)(3 * stride))
-            );
-    }
-}
+void ff_h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale);
+void ff_h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale);
 
+#if HAVE_INLINE_ASM
 /* Draw the edges of width 'w' of an image of size width, height
  * this MMX version can only handle w == 8 || w == 16. */
 static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height,
@@ -1808,14 +1639,14 @@ static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
 #endif
 
     c->add_bytes = add_bytes_mmx;
+#endif /* HAVE_INLINE_ASM */
 
+#if HAVE_YASM
     if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
-        c->h263_v_loop_filter = h263_v_loop_filter_mmx;
-        c->h263_h_loop_filter = h263_h_loop_filter_mmx;
+        c->h263_v_loop_filter = ff_h263_v_loop_filter_mmx;
+        c->h263_h_loop_filter = ff_h263_h_loop_filter_mmx;
     }
-#endif /* HAVE_INLINE_ASM */
 
-#if HAVE_YASM
     c->vector_clip_int32 = ff_vector_clip_int32_mmx;
 #endif