Mercurial > libavcodec.hg

changeset 2732:473ee06ec3a1 libavcodec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
MMX code for (put|avg)_h264_chroma_mc8
author hzoli
date Wed, 25 May 2005 08:05:41 +0000
parents 0b4b57c4a8f5
children 32336384162e
files i386/dsputil_h264_template_mmx.c i386/dsputil_mmx.c
diffstat 2 files changed, 168 insertions(+), 0 deletions(-) [+]
line wrap: on
line diff
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/i386/dsputil_h264_template_mmx.c	Wed May 25 08:05:41 2005 +0000
@@ -0,0 +1,140 @@
+/*
+ * Copyright (c) 2005 Zoltan Hidvegi <hzoli -a- hzoli -d- com>
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+/**
+ * MMX optimized version of (put|avg)_h264_chroma_mc8.
+ * H264_CHROMA_MC8_TMPL must be defined to the desired function name and
+ * H264_CHROMA_OP must be defined to empty for put and pavgb/pavgusb for avg.
+ */
+static void H264_CHROMA_MC8_TMPL(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
+{
+    uint64_t AA __align8;
+    uint64_t DD __align8;
+    unsigned long srcos = (long)src & 7;
+    uint64_t sh1 __align8 = srcos * 8;
+    uint64_t sh2 __align8 = 56 - sh1;
+    int i;
+
+    assert(x<8 && y<8 && x>=0 && y>=0);
+
+    asm volatile("movd %1, %%mm4\n\t"
+                 "movd %2, %%mm6\n\t"
+                 "punpcklwd %%mm4, %%mm4\n\t"
+                 "punpcklwd %%mm6, %%mm6\n\t"
+                 "punpckldq %%mm4, %%mm4\n\t" /* mm4 = x words */
+                 "punpckldq %%mm6, %%mm6\n\t" /* mm6 = y words */
+                 "movq %%mm4, %%mm5\n\t"
+                 "pmullw %%mm6, %%mm4\n\t"    /* mm4 = x * y */
+                 "psllw $3, %%mm5\n\t"
+                 "psllw $3, %%mm6\n\t"
+                 "movq %%mm5, %%mm7\n\t"
+                 "paddw %%mm6, %%mm7\n\t"
+                 "movq %%mm4, %0\n\t"         /* DD = x * y */
+                 "psubw %%mm4, %%mm5\n\t"     /* mm5 = B = 8x - xy */
+                 "psubw %%mm4, %%mm6\n\t"     /* mm6 = C = 8y - xy */
+                 "paddw %3, %%mm4\n\t"
+                 "psubw %%mm7, %%mm4\n\t"     /* mm4 = A = xy - (8x+8y) + 64 */
+                 "pxor %%mm7, %%mm7\n\t"
+                 : "=m" (DD) : "rm" (x), "rm" (y), "m" (sixtyfour));
+
+    asm volatile("movq %%mm4, %0" : "=m" (AA));
+
+    src -= srcos;
+    asm volatile(
+        /* mm0 = src[0..7], mm1 = src[1..8] */
+        "movq %0, %%mm1\n\t"
+        "movq %1, %%mm0\n\t"
+        "psrlq %2, %%mm1\n\t"
+        "psllq %3, %%mm0\n\t"
+        "movq %%mm0, %%mm4\n\t"
+        "psllq $8, %%mm0\n\t"
+        "por %%mm1, %%mm0\n\t"
+        "psrlq $8, %%mm1\n\t"
+        "por %%mm4, %%mm1\n\t"
+        : : "m" (src[0]), "m" (src[8]), "m" (sh1), "m" (sh2));
+
+    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"
+
+            /* [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"
+            "pmullw %%mm5, %%mm1\n\t"
+            "paddw %%mm0, %%mm2\n\t"
+            "paddw %%mm1, %%mm3\n\t"
+            : : "m" (AA));
+
+        src += stride;
+        asm volatile(
+            /* mm0 = src[0..7], mm1 = src[1..8] */
+            "movq %0, %%mm1\n\t"
+            "movq %1, %%mm0\n\t"
+            "psrlq %2, %%mm1\n\t"
+            "psllq %3, %%mm0\n\t"
+            "movq %%mm0, %%mm4\n\t"
+            "psllq $8, %%mm0\n\t"
+            "por %%mm1, %%mm0\n\t"
+            "psrlq $8, %%mm1\n\t"
+            "por %%mm4, %%mm1\n\t"
+            : : "m" (src[0]), "m" (src[8]), "m" (sh1), "m" (sh2));
+
+        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, %%mm4\n\t"
+            "punpckhbw %%mm7, %%mm4\n\t"
+            "pmullw %0, %%mm4\n\t"
+            "paddw %%mm4, %%mm3\n\t"
+            : : "m" (DD));
+
+        asm volatile(
+            /* dst[0..7] = pack(([mm2,mm3] + 32) >> 6) */
+            "paddw %1, %%mm2\n\t"
+            "paddw %1, %%mm3\n\t"
+            "psrlw $6, %%mm2\n\t"
+            "psrlw $6, %%mm3\n\t"
+            "packuswb %%mm3, %%mm2\n\t"
+            H264_CHROMA_OP(%0, %%mm2)
+            "movq %%mm2, %0\n\t"
+            : "=m" (dst[0]) : "m" (thirtytwo));
+        dst+= stride;
+    }
+}
--- a/i386/dsputil_mmx.c	Tue May 24 14:21:42 2005 +0000
+++ b/i386/dsputil_mmx.c	Wed May 25 08:05:41 2005 +0000
@@ -3078,6 +3078,30 @@
 H264_MC(avg_, 8, mmx2)
 H264_MC(avg_, 16,mmx2)
 
+
+/** These are used by *_h264_chroma_mc8_* */
+static const uint64_t thirtytwo __align8 = 0x0020002000200020ULL;
+static const uint64_t sixtyfour __align8 = 0x0040004000400040ULL;
+
+#define H264_CHROMA_OP(S,D)
+#define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx
+#include "dsputil_h264_template_mmx.c"
+#undef H264_CHROMA_OP
+#undef H264_CHROMA_MC8_TMPL
+
+#define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"
+#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2
+#include "dsputil_h264_template_mmx.c"
+#undef H264_CHROMA_OP
+#undef H264_CHROMA_MC8_TMPL
+
+#define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"
+#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow
+#include "dsputil_h264_template_mmx.c"
+#undef H264_CHROMA_OP
+#undef H264_CHROMA_MC8_TMPL
+
+
 #if 0
 static void just_return() { return; }
 #endif
@@ -3377,6 +3401,7 @@
 
         c->h263_v_loop_filter= h263_v_loop_filter_mmx;
         c->h263_h_loop_filter= h263_h_loop_filter_mmx;        
+	c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx;
         
         if (mm_flags & MM_MMXEXT) {
             c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
@@ -3473,6 +3498,7 @@
             dspfunc(avg_h264_qpel, 2, 4);
 #undef dspfunc
 
+	    c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_mmx2;
             c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_mmx2;
             c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_mmx2;
             c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_mmx2;
@@ -3564,6 +3590,8 @@
             dspfunc(avg_h264_qpel, 0, 16);
             dspfunc(avg_h264_qpel, 1, 8);
             dspfunc(avg_h264_qpel, 2, 4);
+
+	    c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_3dnow;
         }
     }