add SSE2 version of vp6_filter_diag
authorZuxy Meng <zuxy.meng@gmail.com>
Fri, 13 Feb 2009 00:02:33 +0000 (00:02 +0000)
committerAurelien Jacobs <aurel@gnuage.org>
Fri, 13 Feb 2009 00:02:33 +0000 (00:02 +0000)
original patch by Zuxy Meng  zuxy.meng _at_ gmail _dot_ com

Originally committed as revision 17195 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/Makefile
libavcodec/x86/dsputil_mmx.c
libavcodec/x86/vp6dsp_sse2.c [new file with mode: 0644]
libavcodec/x86/vp6dsp_sse2.h [new file with mode: 0644]

index 8a45f65..5df16d0 100644 (file)
@@ -415,11 +415,11 @@ MMX-OBJS-$(CONFIG_VC1_DECODER)         += x86/vc1dsp_mmx.o
 MMX-OBJS-$(CONFIG_VP3_DECODER)         += x86/vp3dsp_mmx.o x86/vp3dsp_sse2.o
 MMX-OBJS-$(CONFIG_VP5_DECODER)         += x86/vp3dsp_mmx.o x86/vp3dsp_sse2.o
 MMX-OBJS-$(CONFIG_VP6_DECODER)         += x86/vp3dsp_mmx.o x86/vp3dsp_sse2.o \
-                                          x86/vp6dsp_mmx.o
+                                          x86/vp6dsp_mmx.o x86/vp6dsp_sse2.o
 MMX-OBJS-$(CONFIG_VP6A_DECODER)        += x86/vp3dsp_mmx.o x86/vp3dsp_sse2.o \
-                                          x86/vp6dsp_mmx.o
+                                          x86/vp6dsp_mmx.o x86/vp6dsp_sse2.o
 MMX-OBJS-$(CONFIG_VP6F_DECODER)        += x86/vp3dsp_mmx.o x86/vp3dsp_sse2.o \
-                                          x86/vp6dsp_mmx.o
+                                          x86/vp6dsp_mmx.o x86/vp6dsp_sse2.o
 MMX-OBJS-$(CONFIG_WMV3_DECODER)        += x86/vc1dsp_mmx.o
 MMX-OBJS-$(HAVE_YASM)                  += x86/dsputil_yasm.o            \
                                           $(YASM-OBJS-yes)
index 4d0503d..2f47f5f 100644 (file)
@@ -32,6 +32,7 @@
 #include "vp3dsp_mmx.h"
 #include "vp3dsp_sse2.h"
 #include "vp6dsp_mmx.h"
+#include "vp6dsp_sse2.h"
 #include "idct_xvid.h"
 
 //#undef NDEBUG
@@ -2901,6 +2902,10 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
             H264_QPEL_FUNCS(3, 1, sse2);
             H264_QPEL_FUNCS(3, 2, sse2);
             H264_QPEL_FUNCS(3, 3, sse2);
+
+            if (CONFIG_VP6_DECODER) {
+                c->vp6_filter_diag4 = ff_vp6_filter_diag4_sse2;
+            }
         }
 #if HAVE_SSSE3
         if(mm_flags & FF_MM_SSSE3){
diff --git a/libavcodec/x86/vp6dsp_sse2.c b/libavcodec/x86/vp6dsp_sse2.c
new file mode 100644 (file)
index 0000000..c72a0ca
--- /dev/null
@@ -0,0 +1,98 @@
+/**
+ * @file libavcodec/x86/vp6dsp_mmx.c
+ * SSE2-optimized functions for the VP6 decoder
+ *
+ * Copyright (C) 2009  Zuxy Meng <zuxy.meng@gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg 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 FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/x86_cpu.h"
+#include "libavcodec/dsputil.h"
+#include "dsputil_mmx.h"
+#include "vp6dsp_sse2.h"
+
+#define DIAG4_SSE2(in1,in2,in3,in4)                                           \
+    "movq  "#in1"(%0), %%xmm0            \n\t"                                \
+    "movq  "#in2"(%0), %%xmm1            \n\t"                                \
+    "punpcklbw %%xmm7, %%xmm0            \n\t"                                \
+    "punpcklbw %%xmm7, %%xmm1            \n\t"                                \
+    "pmullw    %%xmm4, %%xmm0            \n\t" /* src[x-8 ] * biweight [0] */ \
+    "pmullw    %%xmm5, %%xmm1            \n\t" /* src[x   ] * biweight [1] */ \
+    "paddw     %%xmm1, %%xmm0            \n\t"                                \
+    "movq  "#in3"(%0), %%xmm1            \n\t"                                \
+    "movq  "#in4"(%0), %%xmm2            \n\t"                                \
+    "punpcklbw %%xmm7, %%xmm1            \n\t"                                \
+    "punpcklbw %%xmm7, %%xmm2            \n\t"                                \
+    "pmullw    %%xmm6, %%xmm1            \n\t" /* src[x+8 ] * biweight [2] */ \
+    "pmullw    %%xmm3, %%xmm2            \n\t" /* src[x+16] * biweight [3] */ \
+    "paddw     %%xmm2, %%xmm1            \n\t"                                \
+    "paddsw     %%xmm1, %%xmm0           \n\t"                                \
+    "paddsw "MANGLE(ff_pw_64)", %%xmm0   \n\t" /* Add 64 */                   \
+    "psraw         $7, %%xmm0            \n\t"                                \
+    "packuswb  %%xmm0, %%xmm0            \n\t"                                \
+    "movq      %%xmm0,   (%1)            \n\t"                                \
+
+void ff_vp6_filter_diag4_sse2(uint8_t *dst, uint8_t *src, int stride,
+                              const int16_t *h_weights,const int16_t *v_weights)
+{
+    uint8_t tmp[8*11], *t = tmp;
+    src -= stride;
+
+    __asm__ volatile(
+    "pxor           %%xmm7, %%xmm7       \n\t"
+    "movq               %4, %%xmm3       \n\t"
+    "pshuflw    $0, %%xmm3, %%xmm4       \n\t"
+    "punpcklqdq     %%xmm4, %%xmm4       \n\t"
+    "pshuflw   $85, %%xmm3, %%xmm5       \n\t"
+    "punpcklqdq     %%xmm5, %%xmm5       \n\t"
+    "pshuflw  $170, %%xmm3, %%xmm6       \n\t"
+    "punpcklqdq     %%xmm6, %%xmm6       \n\t"
+    "pshuflw  $255, %%xmm3, %%xmm3       \n\t"
+    "punpcklqdq     %%xmm3, %%xmm3       \n\t"
+    "1:                                  \n\t"
+    DIAG4_SSE2(-1,0,1,2)
+    "add  $8, %1                         \n\t"
+    "add  %2, %0                         \n\t"
+    "decl %3                             \n\t"
+    "jnz 1b                              \n\t"
+    : "+r"(src), "+r"(t)
+    : "g"((x86_reg)stride), "r"(11), "m"(*(const int64_t*)h_weights)
+    : "memory");
+
+    t = tmp + 8;
+
+    __asm__ volatile(
+    "movq               %4, %%xmm3       \n\t"
+    "pshuflw    $0, %%xmm3, %%xmm4       \n\t"
+    "punpcklqdq     %%xmm4, %%xmm4       \n\t"
+    "pshuflw   $85, %%xmm3, %%xmm5       \n\t"
+    "punpcklqdq     %%xmm5, %%xmm5       \n\t"
+    "pshuflw  $170, %%xmm3, %%xmm6       \n\t"
+    "punpcklqdq     %%xmm6, %%xmm6       \n\t"
+    "pshuflw  $255, %%xmm3, %%xmm3       \n\t"
+    "punpcklqdq     %%xmm3, %%xmm3       \n\t"
+    "1:                                  \n\t"
+    DIAG4_SSE2(-8,0,8,16)
+    "add  $8, %0                         \n\t"
+    "add  %2, %1                         \n\t"
+    "decl %3                             \n\t"
+    "jnz 1b                              \n\t"
+    : "+r"(t), "+r"(dst)
+    : "g"((x86_reg)stride), "r"(8), "m"(*(const int64_t*)v_weights)
+    : "memory");
+}
diff --git a/libavcodec/x86/vp6dsp_sse2.h b/libavcodec/x86/vp6dsp_sse2.h
new file mode 100644 (file)
index 0000000..a30089a
--- /dev/null
@@ -0,0 +1,30 @@
+/*
+ * vp6dsp SSE2 function declarations
+ * Copyright (c) 2009  Zuxy Meng <zuxy.meng@gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg 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 FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_X86_VP6DSP_SSE2_H
+#define AVCODEC_X86_VP6DSP_SSE2_H
+
+#include <stdint.h>
+
+void ff_vp6_filter_diag4_sse2(uint8_t *dst, uint8_t *src, int stride,
+                             const int16_t *h_weights,const int16_t *v_weights);
+
+#endif /* AVCODEC_X86_VP6DSP_SSE2_H */