From 5e8b787afa46f0c0b7c865e8afca26652a4e06ed Mon Sep 17 00:00:00 2001
From: Loren Merritt <lorenm@u.washington.edu>
Date: Sat, 25 Mar 2006 08:41:14 +0000
Subject: [PATCH] simplified and slightly faster h264_chroma_mc8_mmx

Originally committed as revision 5214 to svn://svn.ffmpeg.org/ffmpeg/trunk
---
 libavcodec/i386/dsputil_h264_template_mmx.c | 204 +++++++-------------
 1 file changed, 72 insertions(+), 132 deletions(-)

diff --git a/libavcodec/i386/dsputil_h264_template_mmx.c b/libavcodec/i386/dsputil_h264_template_mmx.c
index d254337abe..b49c880a73 100644
--- a/libavcodec/i386/dsputil_h264_template_mmx.c
+++ b/libavcodec/i386/dsputil_h264_template_mmx.c
@@ -37,51 +37,54 @@ static void H264_CHROMA_MC8_TMPL(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*
 
     assert(x<8 && y<8 && x>=0 && y>=0);
 
-    if(y==0)
+    if(y==0 || x==0)
     {
-        /* horizontal filter only */
-        asm volatile("movd %0, %%mm5\n\t"
-                     "punpcklwd %%mm5, %%mm5\n\t"
-                     "punpckldq %%mm5, %%mm5\n\t" /* mm5 = B = x */
-                     "movq %1, %%mm4\n\t"
-                     "pxor %%mm7, %%mm7\n\t"
-                     "psubw %%mm5, %%mm4\n\t"     /* mm4 = A = 8-x */
-                     : : "rm" (x), "m" (ff_pw_8));
+        /* 1 dimensional filter only */
+        const int dxy = x ? 1 : stride;
+
+        asm volatile(
+            "movd %0, %%mm5\n\t"
+            "movq %1, %%mm4\n\t"
+            "punpcklwd %%mm5, %%mm5\n\t"
+            "punpckldq %%mm5, %%mm5\n\t" /* mm5 = B = x */
+            "movq %%mm4, %%mm6\n\t"
+            "pxor %%mm7, %%mm7\n\t"
+            "psubw %%mm5, %%mm4\n\t"     /* mm4 = A = 8-x */
+            "psrlw $1, %%mm6\n\t"        /* mm6 = 4 */
+            :: "rm"(x+y), "m"(ff_pw_8));
 
         for(i=0; i<h; i++) {
             asm volatile(
                 /* mm0 = src[0..7], mm1 = src[1..8] */
                 "movq %0, %%mm0\n\t"
-                "movq %1, %%mm1\n\t"
-                : : "m" (src[0]), "m" (src[1]));
+                "movq %1, %%mm2\n\t"
+                :: "m"(src[0]), "m"(src[dxy]));
 
             asm volatile(
-                /* [mm2,mm3] = A * src[0..7] */
-                "movq %%mm0, %%mm2\n\t"
-                "punpcklbw %%mm7, %%mm2\n\t"
-                "pmullw %%mm4, %%mm2\n\t"
-                "movq %%mm0, %%mm3\n\t"
-                "punpckhbw %%mm7, %%mm3\n\t"
-                "pmullw %%mm4, %%mm3\n\t"
-
-                /* [mm2,mm3] += B * src[1..8] */
-                "movq %%mm1, %%mm0\n\t"
+                /* [mm0,mm1] = A * src[0..7] */
+                /* [mm2,mm3] = B * src[1..8] */
+                "movq %%mm0, %%mm1\n\t"
+                "movq %%mm2, %%mm3\n\t"
                 "punpcklbw %%mm7, %%mm0\n\t"
-                "pmullw %%mm5, %%mm0\n\t"
                 "punpckhbw %%mm7, %%mm1\n\t"
-                "pmullw %%mm5, %%mm1\n\t"
-                "paddw %%mm0, %%mm2\n\t"
-                "paddw %%mm1, %%mm3\n\t"
+                "punpcklbw %%mm7, %%mm2\n\t"
+                "punpckhbw %%mm7, %%mm3\n\t"
+                "pmullw %%mm4, %%mm0\n\t"
+                "pmullw %%mm4, %%mm1\n\t"
+                "pmullw %%mm5, %%mm2\n\t"
+                "pmullw %%mm5, %%mm3\n\t"
 
-                /* dst[0..7] = pack(([mm2,mm3] + 32) >> 6) */
-                "paddw %1, %%mm2\n\t"
-                "paddw %1, %%mm3\n\t"
-                "psrlw $3, %%mm2\n\t"
-                "psrlw $3, %%mm3\n\t"
-                "packuswb %%mm3, %%mm2\n\t"
-                H264_CHROMA_OP(%0, %%mm2)
-                "movq %%mm2, %0\n\t"
-                : "=m" (dst[0]) : "m" (ff_pw_4));
+                /* dst[0..7] = (A * src[0..7] + B * src[1..8] + 4) >> 3 */
+                "paddw %%mm6, %%mm0\n\t"
+                "paddw %%mm6, %%mm1\n\t"
+                "paddw %%mm2, %%mm0\n\t"
+                "paddw %%mm3, %%mm1\n\t"
+                "psrlw $3, %%mm0\n\t"
+                "psrlw $3, %%mm1\n\t"
+                "packuswb %%mm1, %%mm0\n\t"
+                H264_CHROMA_OP(%0, %%mm0)
+                "movq %%mm0, %0\n\t"
+                : "=m" (dst[0]));
 
             src += stride;
             dst += stride;
@@ -89,65 +92,6 @@ static void H264_CHROMA_MC8_TMPL(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*
         return;
     }
 
-    if(x==0)
-    {
-        /* vertical filter only */
-        asm volatile("movd %0, %%mm6\n\t"
-                     "punpcklwd %%mm6, %%mm6\n\t"
-                     "punpckldq %%mm6, %%mm6\n\t" /* mm6 = C = y */
-                     "movq %1, %%mm4\n\t"
-                     "pxor %%mm7, %%mm7\n\t"
-                     "psubw %%mm6, %%mm4\n\t"     /* mm4 = A = 8-y */
-                     : : "rm" (y), "m" (ff_pw_8));
-
-        asm volatile(
-            /* mm0 = src[0..7] */
-            "movq %0, %%mm0\n\t"
-            : : "m" (src[0]));
-
-        for(i=0; i<h; i++) {
-            asm volatile(
-                /* [mm2,mm3] = A * src[0..7] */
-                "movq %mm0, %mm2\n\t"
-                "punpcklbw %mm7, %mm2\n\t"
-                "pmullw %mm4, %mm2\n\t"
-                "movq %mm0, %mm3\n\t"
-                "punpckhbw %mm7, %mm3\n\t"
-                "pmullw %mm4, %mm3\n\t");
-
-            src += stride;
-            asm volatile(
-                /* mm0 = src[0..7] */
-                "movq %0, %%mm0\n\t"
-                : : "m" (src[0]));
-
-            asm volatile(
-                /* [mm2,mm3] += C * src[0..7] */
-                "movq %mm0, %mm1\n\t"
-                "punpcklbw %mm7, %mm1\n\t"
-                "pmullw %mm6, %mm1\n\t"
-                "paddw %mm1, %mm2\n\t"
-                "movq %mm0, %mm5\n\t"
-                "punpckhbw %mm7, %mm5\n\t"
-                "pmullw %mm6, %mm5\n\t"
-                "paddw %mm5, %mm3\n\t");
-
-            asm volatile(
-                /* dst[0..7] = pack(([mm2,mm3] + 32) >> 6) */
-                "paddw %1, %%mm2\n\t"
-                "paddw %1, %%mm3\n\t"
-                "psrlw $3, %%mm2\n\t"
-                "psrlw $3, %%mm3\n\t"
-                "packuswb %%mm3, %%mm2\n\t"
-                H264_CHROMA_OP(%0, %%mm2)
-                "movq %%mm2, %0\n\t"
-                : "=m" (dst[0]) : "m" (ff_pw_4));
-
-            dst += stride;
-        }
-        return;
-    }
-
     /* general case, bilinear */
     asm volatile("movd %2, %%mm4\n\t"
                  "movd %3, %%mm6\n\t"
@@ -177,57 +121,53 @@ static void H264_CHROMA_MC8_TMPL(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*
         : : "m" (src[0]), "m" (src[1]));
 
     for(i=0; i<h; i++) {
-        asm volatile(
-            /* [mm2,mm3] = A * src[0..7] */
-            "movq %%mm0, %%mm2\n\t"
-            "punpcklbw %%mm7, %%mm2\n\t"
-            "pmullw %0, %%mm2\n\t"
-            "movq %%mm0, %%mm3\n\t"
-            "punpckhbw %%mm7, %%mm3\n\t"
-            "pmullw %0, %%mm3\n\t"
+        src += stride;
 
-            /* [mm2,mm3] += B * src[1..8] */
-            "movq %%mm1, %%mm0\n\t"
-            "punpcklbw %%mm7, %%mm0\n\t"
-            "pmullw %%mm5, %%mm0\n\t"
-            "punpckhbw %%mm7, %%mm1\n\t"
+        asm volatile(
+            /* mm2 = A * src[0..3] + B * src[1..4] */
+            /* mm3 = A * src[4..7] + B * src[5..8] */
+            "movq %%mm0, %%mm2\n\t"
+            "movq %%mm1, %%mm3\n\t"
+            "punpckhbw %%mm7, %%mm0\n\t"
+            "punpcklbw %%mm7, %%mm1\n\t"
+            "punpcklbw %%mm7, %%mm2\n\t"
+            "punpckhbw %%mm7, %%mm3\n\t"
+            "pmullw %0, %%mm0\n\t"
+            "pmullw %0, %%mm2\n\t"
             "pmullw %%mm5, %%mm1\n\t"
-            "paddw %%mm0, %%mm2\n\t"
-            "paddw %%mm1, %%mm3\n\t"
+            "pmullw %%mm5, %%mm3\n\t"
+            "paddw %%mm1, %%mm2\n\t"
+            "paddw %%mm0, %%mm3\n\t"
             : : "m" (AA));
 
-        src += stride;
         asm volatile(
-            /* mm0 = src[0..7], mm1 = src[1..8] */
+            /* [mm2,mm3] += C * src[0..7] */
             "movq %0, %%mm0\n\t"
+            "movq %%mm0, %%mm1\n\t"
+            "punpcklbw %%mm7, %%mm0\n\t"
+            "punpckhbw %%mm7, %%mm1\n\t"
+            "pmullw %%mm6, %%mm0\n\t"
+            "pmullw %%mm6, %%mm1\n\t"
+            "paddw %%mm0, %%mm2\n\t"
+            "paddw %%mm1, %%mm3\n\t"
+            : : "m" (src[0]));
+
+        asm volatile(
+            /* [mm2,mm3] += D * src[1..8] */
             "movq %1, %%mm1\n\t"
-            : : "m" (src[0]), "m" (src[1]));
-
-        asm volatile(
-            /* [mm2,mm3] += C *  src[0..7] */
-            "movq %mm0, %mm4\n\t"
-            "punpcklbw %mm7, %mm4\n\t"
-            "pmullw %mm6, %mm4\n\t"
-            "paddw %mm4, %mm2\n\t"
-            "movq %mm0, %mm4\n\t"
-            "punpckhbw %mm7, %mm4\n\t"
-            "pmullw %mm6, %mm4\n\t"
-            "paddw %mm4, %mm3\n\t");
-
-        asm volatile(
-            /* [mm2,mm3] += D *  src[1..8] */
-            "movq %%mm1, %%mm4\n\t"
-            "punpcklbw %%mm7, %%mm4\n\t"
-            "pmullw %0, %%mm4\n\t"
-            "paddw %%mm4, %%mm2\n\t"
+            "movq %%mm1, %%mm0\n\t"
             "movq %%mm1, %%mm4\n\t"
+            "punpcklbw %%mm7, %%mm0\n\t"
             "punpckhbw %%mm7, %%mm4\n\t"
-            "pmullw %0, %%mm4\n\t"
+            "pmullw %2, %%mm0\n\t"
+            "pmullw %2, %%mm4\n\t"
+            "paddw %%mm0, %%mm2\n\t"
             "paddw %%mm4, %%mm3\n\t"
-            : : "m" (DD));
+            "movq %0, %%mm0\n\t"
+            : : "m" (src[0]), "m" (src[1]), "m" (DD));
 
         asm volatile(
-            /* dst[0..7] = pack(([mm2,mm3] + 32) >> 6) */
+            /* dst[0..7] = ([mm2,mm3] + 32) >> 6 */
             "paddw %1, %%mm2\n\t"
             "paddw %1, %%mm3\n\t"
             "psrlw $6, %%mm2\n\t"