x86: mpegvideo: more sensible names for optimization file and init function
authorDiego Biurrun <diego@biurrun.de>
Mon, 13 Aug 2012 11:13:57 +0000 (13:13 +0200)
committerDiego Biurrun <diego@biurrun.de>
Fri, 24 Aug 2012 00:23:16 +0000 (02:23 +0200)
libavcodec/mpegvideo.c
libavcodec/mpegvideo.h
libavcodec/x86/Makefile
libavcodec/x86/mpegvideo.c [new file with mode: 0644]
libavcodec/x86/mpegvideo_mmx.c [deleted file]

index 8f3544f..13e2acc 100644 (file)
@@ -185,8 +185,8 @@ av_cold int ff_dct_common_init(MpegEncContext *s)
         s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact;
     s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;
 
-#if HAVE_MMX
-    ff_MPV_common_init_mmx(s);
+#if ARCH_X86
+    ff_MPV_common_init_x86(s);
 #elif ARCH_ALPHA
     ff_MPV_common_init_axp(s);
 #elif HAVE_MMI
index 850681b..7ea7f82 100644 (file)
@@ -757,7 +757,7 @@ int ff_MPV_encode_end(AVCodecContext *avctx);
 int ff_MPV_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
                           const AVFrame *frame, int *got_packet);
 void ff_MPV_encode_init_x86(MpegEncContext *s);
-void ff_MPV_common_init_mmx(MpegEncContext *s);
+void ff_MPV_common_init_x86(MpegEncContext *s);
 void ff_MPV_common_init_axp(MpegEncContext *s);
 void ff_MPV_common_init_mmi(MpegEncContext *s);
 void ff_MPV_common_init_arm(MpegEncContext *s);
index d843bc9..4dffe6c 100644 (file)
@@ -22,7 +22,7 @@ MMX-OBJS-$(CONFIG_H264DSP)             += x86/h264dsp_init.o
 MMX-OBJS-$(CONFIG_H264PRED)            += x86/h264_intrapred_init.o
 MMX-OBJS-$(CONFIG_LPC)                 += x86/lpc_mmx.o
 MMX-OBJS-$(CONFIG_MPEGAUDIODSP)        += x86/mpegaudiodec_mmx.o
-MMX-OBJS-$(CONFIG_MPEGVIDEO)           += x86/mpegvideo_mmx.o
+MMX-OBJS-$(CONFIG_MPEGVIDEO)           += x86/mpegvideo.o
 MMX-OBJS-$(CONFIG_MPEGVIDEOENC)        += x86/mpegvideoenc.o
 MMX-OBJS-$(CONFIG_PNG_DECODER)         += x86/pngdsp_init.o
 MMX-OBJS-$(CONFIG_PRORES_DECODER)      += x86/proresdsp_init.o
diff --git a/libavcodec/x86/mpegvideo.c b/libavcodec/x86/mpegvideo.c
new file mode 100644 (file)
index 0000000..077031a
--- /dev/null
@@ -0,0 +1,608 @@
+/*
+ * Optimized for ia32 CPUs by Nick Kurshev <nickols_k@mail.ru>
+ * h263, mpeg1, mpeg2 dequantizer & draw_edges by Michael Niedermayer <michaelni@gmx.at>
+ *
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/cpu.h"
+#include "libavutil/x86/asm.h"
+#include "libavcodec/avcodec.h"
+#include "libavcodec/dsputil.h"
+#include "libavcodec/mpegvideo.h"
+#include "dsputil_mmx.h"
+
+#if HAVE_INLINE_ASM
+
+static void dct_unquantize_h263_intra_mmx(MpegEncContext *s,
+                                  DCTELEM *block, int n, int qscale)
+{
+    x86_reg level, qmul, qadd, nCoeffs;
+
+    qmul = qscale << 1;
+
+    assert(s->block_last_index[n]>=0 || s->h263_aic);
+
+    if (!s->h263_aic) {
+        if (n < 4)
+            level = block[0] * s->y_dc_scale;
+        else
+            level = block[0] * s->c_dc_scale;
+        qadd = (qscale - 1) | 1;
+    }else{
+        qadd = 0;
+        level= block[0];
+    }
+    if(s->ac_pred)
+        nCoeffs=63;
+    else
+        nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
+//printf("%d %d  ", qmul, qadd);
+__asm__ volatile(
+                "movd %1, %%mm6                 \n\t" //qmul
+                "packssdw %%mm6, %%mm6          \n\t"
+                "packssdw %%mm6, %%mm6          \n\t"
+                "movd %2, %%mm5                 \n\t" //qadd
+                "pxor %%mm7, %%mm7              \n\t"
+                "packssdw %%mm5, %%mm5          \n\t"
+                "packssdw %%mm5, %%mm5          \n\t"
+                "psubw %%mm5, %%mm7             \n\t"
+                "pxor %%mm4, %%mm4              \n\t"
+                ".p2align 4                     \n\t"
+                "1:                             \n\t"
+                "movq (%0, %3), %%mm0           \n\t"
+                "movq 8(%0, %3), %%mm1          \n\t"
+
+                "pmullw %%mm6, %%mm0            \n\t"
+                "pmullw %%mm6, %%mm1            \n\t"
+
+                "movq (%0, %3), %%mm2           \n\t"
+                "movq 8(%0, %3), %%mm3          \n\t"
+
+                "pcmpgtw %%mm4, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
+                "pcmpgtw %%mm4, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
+
+                "pxor %%mm2, %%mm0              \n\t"
+                "pxor %%mm3, %%mm1              \n\t"
+
+                "paddw %%mm7, %%mm0             \n\t"
+                "paddw %%mm7, %%mm1             \n\t"
+
+                "pxor %%mm0, %%mm2              \n\t"
+                "pxor %%mm1, %%mm3              \n\t"
+
+                "pcmpeqw %%mm7, %%mm0           \n\t" // block[i] == 0 ? -1 : 0
+                "pcmpeqw %%mm7, %%mm1           \n\t" // block[i] == 0 ? -1 : 0
+
+                "pandn %%mm2, %%mm0             \n\t"
+                "pandn %%mm3, %%mm1             \n\t"
+
+                "movq %%mm0, (%0, %3)           \n\t"
+                "movq %%mm1, 8(%0, %3)          \n\t"
+
+                "add $16, %3                    \n\t"
+                "jng 1b                         \n\t"
+                ::"r" (block+nCoeffs), "rm"(qmul), "rm" (qadd), "r" (2*(-nCoeffs))
+                : "memory"
+        );
+        block[0]= level;
+}
+
+
+static void dct_unquantize_h263_inter_mmx(MpegEncContext *s,
+                                  DCTELEM *block, int n, int qscale)
+{
+    x86_reg qmul, qadd, nCoeffs;
+
+    qmul = qscale << 1;
+    qadd = (qscale - 1) | 1;
+
+    assert(s->block_last_index[n]>=0 || s->h263_aic);
+
+    nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
+//printf("%d %d  ", qmul, qadd);
+__asm__ volatile(
+                "movd %1, %%mm6                 \n\t" //qmul
+                "packssdw %%mm6, %%mm6          \n\t"
+                "packssdw %%mm6, %%mm6          \n\t"
+                "movd %2, %%mm5                 \n\t" //qadd
+                "pxor %%mm7, %%mm7              \n\t"
+                "packssdw %%mm5, %%mm5          \n\t"
+                "packssdw %%mm5, %%mm5          \n\t"
+                "psubw %%mm5, %%mm7             \n\t"
+                "pxor %%mm4, %%mm4              \n\t"
+                ".p2align 4                     \n\t"
+                "1:                             \n\t"
+                "movq (%0, %3), %%mm0           \n\t"
+                "movq 8(%0, %3), %%mm1          \n\t"
+
+                "pmullw %%mm6, %%mm0            \n\t"
+                "pmullw %%mm6, %%mm1            \n\t"
+
+                "movq (%0, %3), %%mm2           \n\t"
+                "movq 8(%0, %3), %%mm3          \n\t"
+
+                "pcmpgtw %%mm4, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
+                "pcmpgtw %%mm4, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
+
+                "pxor %%mm2, %%mm0              \n\t"
+                "pxor %%mm3, %%mm1              \n\t"
+
+                "paddw %%mm7, %%mm0             \n\t"
+                "paddw %%mm7, %%mm1             \n\t"
+
+                "pxor %%mm0, %%mm2              \n\t"
+                "pxor %%mm1, %%mm3              \n\t"
+
+                "pcmpeqw %%mm7, %%mm0           \n\t" // block[i] == 0 ? -1 : 0
+                "pcmpeqw %%mm7, %%mm1           \n\t" // block[i] == 0 ? -1 : 0
+
+                "pandn %%mm2, %%mm0             \n\t"
+                "pandn %%mm3, %%mm1             \n\t"
+
+                "movq %%mm0, (%0, %3)           \n\t"
+                "movq %%mm1, 8(%0, %3)          \n\t"
+
+                "add $16, %3                    \n\t"
+                "jng 1b                         \n\t"
+                ::"r" (block+nCoeffs), "rm"(qmul), "rm" (qadd), "r" (2*(-nCoeffs))
+                : "memory"
+        );
+}
+
+
+/*
+  NK:
+  Note: looking at PARANOID:
+  "enable all paranoid tests for rounding, overflows, etc..."
+
+#ifdef PARANOID
+                if (level < -2048 || level > 2047)
+                    fprintf(stderr, "unquant error %d %d\n", i, level);
+#endif
+  We can suppose that result of two multiplications can't be greater than 0xFFFF
+  i.e. is 16-bit, so we use here only PMULLW instruction and can avoid
+  a complex multiplication.
+=====================================================
+ Full formula for multiplication of 2 integer numbers
+ which are represent as high:low words:
+ input: value1 = high1:low1
+        value2 = high2:low2
+ output: value3 = value1*value2
+ value3=high3:low3 (on overflow: modulus 2^32 wrap-around)
+ this mean that for 0x123456 * 0x123456 correct result is 0x766cb0ce4
+ but this algorithm will compute only 0x66cb0ce4
+ this limited by 16-bit size of operands
+ ---------------------------------
+ tlow1 = high1*low2
+ tlow2 = high2*low1
+ tlow1 = tlow1 + tlow2
+ high3:low3 = low1*low2
+ high3 += tlow1
+*/
+static void dct_unquantize_mpeg1_intra_mmx(MpegEncContext *s,
+                                     DCTELEM *block, int n, int qscale)
+{
+    x86_reg nCoeffs;
+    const uint16_t *quant_matrix;
+    int block0;
+
+    assert(s->block_last_index[n]>=0);
+
+    nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]+1;
+
+    if (n < 4)
+        block0 = block[0] * s->y_dc_scale;
+    else
+        block0 = block[0] * s->c_dc_scale;
+    /* XXX: only mpeg1 */
+    quant_matrix = s->intra_matrix;
+__asm__ volatile(
+                "pcmpeqw %%mm7, %%mm7           \n\t"
+                "psrlw $15, %%mm7               \n\t"
+                "movd %2, %%mm6                 \n\t"
+                "packssdw %%mm6, %%mm6          \n\t"
+                "packssdw %%mm6, %%mm6          \n\t"
+                "mov %3, %%"REG_a"              \n\t"
+                ".p2align 4                     \n\t"
+                "1:                             \n\t"
+                "movq (%0, %%"REG_a"), %%mm0    \n\t"
+                "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
+                "movq (%1, %%"REG_a"), %%mm4    \n\t"
+                "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
+                "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
+                "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
+                "pxor %%mm2, %%mm2              \n\t"
+                "pxor %%mm3, %%mm3              \n\t"
+                "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
+                "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
+                "pxor %%mm2, %%mm0              \n\t"
+                "pxor %%mm3, %%mm1              \n\t"
+                "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
+                "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
+                "pmullw %%mm4, %%mm0            \n\t" // abs(block[i])*q
+                "pmullw %%mm5, %%mm1            \n\t" // abs(block[i])*q
+                "pxor %%mm4, %%mm4              \n\t"
+                "pxor %%mm5, %%mm5              \n\t" // FIXME slow
+                "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
+                "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
+                "psraw $3, %%mm0                \n\t"
+                "psraw $3, %%mm1                \n\t"
+                "psubw %%mm7, %%mm0             \n\t"
+                "psubw %%mm7, %%mm1             \n\t"
+                "por %%mm7, %%mm0               \n\t"
+                "por %%mm7, %%mm1               \n\t"
+                "pxor %%mm2, %%mm0              \n\t"
+                "pxor %%mm3, %%mm1              \n\t"
+                "psubw %%mm2, %%mm0             \n\t"
+                "psubw %%mm3, %%mm1             \n\t"
+                "pandn %%mm0, %%mm4             \n\t"
+                "pandn %%mm1, %%mm5             \n\t"
+                "movq %%mm4, (%0, %%"REG_a")    \n\t"
+                "movq %%mm5, 8(%0, %%"REG_a")   \n\t"
+
+                "add $16, %%"REG_a"             \n\t"
+                "js 1b                          \n\t"
+                ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
+                : "%"REG_a, "memory"
+        );
+    block[0]= block0;
+}
+
+static void dct_unquantize_mpeg1_inter_mmx(MpegEncContext *s,
+                                     DCTELEM *block, int n, int qscale)
+{
+    x86_reg nCoeffs;
+    const uint16_t *quant_matrix;
+
+    assert(s->block_last_index[n]>=0);
+
+    nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]+1;
+
+        quant_matrix = s->inter_matrix;
+__asm__ volatile(
+                "pcmpeqw %%mm7, %%mm7           \n\t"
+                "psrlw $15, %%mm7               \n\t"
+                "movd %2, %%mm6                 \n\t"
+                "packssdw %%mm6, %%mm6          \n\t"
+                "packssdw %%mm6, %%mm6          \n\t"
+                "mov %3, %%"REG_a"              \n\t"
+                ".p2align 4                     \n\t"
+                "1:                             \n\t"
+                "movq (%0, %%"REG_a"), %%mm0    \n\t"
+                "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
+                "movq (%1, %%"REG_a"), %%mm4    \n\t"
+                "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
+                "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
+                "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
+                "pxor %%mm2, %%mm2              \n\t"
+                "pxor %%mm3, %%mm3              \n\t"
+                "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
+                "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
+                "pxor %%mm2, %%mm0              \n\t"
+                "pxor %%mm3, %%mm1              \n\t"
+                "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
+                "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
+                "paddw %%mm0, %%mm0             \n\t" // abs(block[i])*2
+                "paddw %%mm1, %%mm1             \n\t" // abs(block[i])*2
+                "paddw %%mm7, %%mm0             \n\t" // abs(block[i])*2 + 1
+                "paddw %%mm7, %%mm1             \n\t" // abs(block[i])*2 + 1
+                "pmullw %%mm4, %%mm0            \n\t" // (abs(block[i])*2 + 1)*q
+                "pmullw %%mm5, %%mm1            \n\t" // (abs(block[i])*2 + 1)*q
+                "pxor %%mm4, %%mm4              \n\t"
+                "pxor %%mm5, %%mm5              \n\t" // FIXME slow
+                "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
+                "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
+                "psraw $4, %%mm0                \n\t"
+                "psraw $4, %%mm1                \n\t"
+                "psubw %%mm7, %%mm0             \n\t"
+                "psubw %%mm7, %%mm1             \n\t"
+                "por %%mm7, %%mm0               \n\t"
+                "por %%mm7, %%mm1               \n\t"
+                "pxor %%mm2, %%mm0              \n\t"
+                "pxor %%mm3, %%mm1              \n\t"
+                "psubw %%mm2, %%mm0             \n\t"
+                "psubw %%mm3, %%mm1             \n\t"
+                "pandn %%mm0, %%mm4             \n\t"
+                "pandn %%mm1, %%mm5             \n\t"
+                "movq %%mm4, (%0, %%"REG_a")    \n\t"
+                "movq %%mm5, 8(%0, %%"REG_a")   \n\t"
+
+                "add $16, %%"REG_a"             \n\t"
+                "js 1b                          \n\t"
+                ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
+                : "%"REG_a, "memory"
+        );
+}
+
+static void dct_unquantize_mpeg2_intra_mmx(MpegEncContext *s,
+                                     DCTELEM *block, int n, int qscale)
+{
+    x86_reg nCoeffs;
+    const uint16_t *quant_matrix;
+    int block0;
+
+    assert(s->block_last_index[n]>=0);
+
+    if(s->alternate_scan) nCoeffs= 63; //FIXME
+    else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
+
+    if (n < 4)
+        block0 = block[0] * s->y_dc_scale;
+    else
+        block0 = block[0] * s->c_dc_scale;
+    quant_matrix = s->intra_matrix;
+__asm__ volatile(
+                "pcmpeqw %%mm7, %%mm7           \n\t"
+                "psrlw $15, %%mm7               \n\t"
+                "movd %2, %%mm6                 \n\t"
+                "packssdw %%mm6, %%mm6          \n\t"
+                "packssdw %%mm6, %%mm6          \n\t"
+                "mov %3, %%"REG_a"              \n\t"
+                ".p2align 4                     \n\t"
+                "1:                             \n\t"
+                "movq (%0, %%"REG_a"), %%mm0    \n\t"
+                "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
+                "movq (%1, %%"REG_a"), %%mm4    \n\t"
+                "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
+                "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
+                "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
+                "pxor %%mm2, %%mm2              \n\t"
+                "pxor %%mm3, %%mm3              \n\t"
+                "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
+                "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
+                "pxor %%mm2, %%mm0              \n\t"
+                "pxor %%mm3, %%mm1              \n\t"
+                "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
+                "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
+                "pmullw %%mm4, %%mm0            \n\t" // abs(block[i])*q
+                "pmullw %%mm5, %%mm1            \n\t" // abs(block[i])*q
+                "pxor %%mm4, %%mm4              \n\t"
+                "pxor %%mm5, %%mm5              \n\t" // FIXME slow
+                "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
+                "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
+                "psraw $3, %%mm0                \n\t"
+                "psraw $3, %%mm1                \n\t"
+                "pxor %%mm2, %%mm0              \n\t"
+                "pxor %%mm3, %%mm1              \n\t"
+                "psubw %%mm2, %%mm0             \n\t"
+                "psubw %%mm3, %%mm1             \n\t"
+                "pandn %%mm0, %%mm4             \n\t"
+                "pandn %%mm1, %%mm5             \n\t"
+                "movq %%mm4, (%0, %%"REG_a")    \n\t"
+                "movq %%mm5, 8(%0, %%"REG_a")   \n\t"
+
+                "add $16, %%"REG_a"             \n\t"
+                "jng 1b                         \n\t"
+                ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
+                : "%"REG_a, "memory"
+        );
+    block[0]= block0;
+        //Note, we do not do mismatch control for intra as errors cannot accumulate
+}
+
+static void dct_unquantize_mpeg2_inter_mmx(MpegEncContext *s,
+                                     DCTELEM *block, int n, int qscale)
+{
+    x86_reg nCoeffs;
+    const uint16_t *quant_matrix;
+
+    assert(s->block_last_index[n]>=0);
+
+    if(s->alternate_scan) nCoeffs= 63; //FIXME
+    else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
+
+        quant_matrix = s->inter_matrix;
+__asm__ volatile(
+                "pcmpeqw %%mm7, %%mm7           \n\t"
+                "psrlq $48, %%mm7               \n\t"
+                "movd %2, %%mm6                 \n\t"
+                "packssdw %%mm6, %%mm6          \n\t"
+                "packssdw %%mm6, %%mm6          \n\t"
+                "mov %3, %%"REG_a"              \n\t"
+                ".p2align 4                     \n\t"
+                "1:                             \n\t"
+                "movq (%0, %%"REG_a"), %%mm0    \n\t"
+                "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
+                "movq (%1, %%"REG_a"), %%mm4    \n\t"
+                "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
+                "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
+                "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
+                "pxor %%mm2, %%mm2              \n\t"
+                "pxor %%mm3, %%mm3              \n\t"
+                "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
+                "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
+                "pxor %%mm2, %%mm0              \n\t"
+                "pxor %%mm3, %%mm1              \n\t"
+                "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
+                "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
+                "paddw %%mm0, %%mm0             \n\t" // abs(block[i])*2
+                "paddw %%mm1, %%mm1             \n\t" // abs(block[i])*2
+                "pmullw %%mm4, %%mm0            \n\t" // abs(block[i])*2*q
+                "pmullw %%mm5, %%mm1            \n\t" // abs(block[i])*2*q
+                "paddw %%mm4, %%mm0             \n\t" // (abs(block[i])*2 + 1)*q
+                "paddw %%mm5, %%mm1             \n\t" // (abs(block[i])*2 + 1)*q
+                "pxor %%mm4, %%mm4              \n\t"
+                "pxor %%mm5, %%mm5              \n\t" // FIXME slow
+                "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
+                "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
+                "psrlw $4, %%mm0                \n\t"
+                "psrlw $4, %%mm1                \n\t"
+                "pxor %%mm2, %%mm0              \n\t"
+                "pxor %%mm3, %%mm1              \n\t"
+                "psubw %%mm2, %%mm0             \n\t"
+                "psubw %%mm3, %%mm1             \n\t"
+                "pandn %%mm0, %%mm4             \n\t"
+                "pandn %%mm1, %%mm5             \n\t"
+                "pxor %%mm4, %%mm7              \n\t"
+                "pxor %%mm5, %%mm7              \n\t"
+                "movq %%mm4, (%0, %%"REG_a")    \n\t"
+                "movq %%mm5, 8(%0, %%"REG_a")   \n\t"
+
+                "add $16, %%"REG_a"             \n\t"
+                "jng 1b                         \n\t"
+                "movd 124(%0, %3), %%mm0        \n\t"
+                "movq %%mm7, %%mm6              \n\t"
+                "psrlq $32, %%mm7               \n\t"
+                "pxor %%mm6, %%mm7              \n\t"
+                "movq %%mm7, %%mm6              \n\t"
+                "psrlq $16, %%mm7               \n\t"
+                "pxor %%mm6, %%mm7              \n\t"
+                "pslld $31, %%mm7               \n\t"
+                "psrlq $15, %%mm7               \n\t"
+                "pxor %%mm7, %%mm0              \n\t"
+                "movd %%mm0, 124(%0, %3)        \n\t"
+
+                ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "r" (-2*nCoeffs)
+                : "%"REG_a, "memory"
+        );
+}
+
+static void  denoise_dct_mmx(MpegEncContext *s, DCTELEM *block){
+    const int intra= s->mb_intra;
+    int *sum= s->dct_error_sum[intra];
+    uint16_t *offset= s->dct_offset[intra];
+
+    s->dct_count[intra]++;
+
+    __asm__ volatile(
+        "pxor %%mm7, %%mm7                      \n\t"
+        "1:                                     \n\t"
+        "pxor %%mm0, %%mm0                      \n\t"
+        "pxor %%mm1, %%mm1                      \n\t"
+        "movq (%0), %%mm2                       \n\t"
+        "movq 8(%0), %%mm3                      \n\t"
+        "pcmpgtw %%mm2, %%mm0                   \n\t"
+        "pcmpgtw %%mm3, %%mm1                   \n\t"
+        "pxor %%mm0, %%mm2                      \n\t"
+        "pxor %%mm1, %%mm3                      \n\t"
+        "psubw %%mm0, %%mm2                     \n\t"
+        "psubw %%mm1, %%mm3                     \n\t"
+        "movq %%mm2, %%mm4                      \n\t"
+        "movq %%mm3, %%mm5                      \n\t"
+        "psubusw (%2), %%mm2                    \n\t"
+        "psubusw 8(%2), %%mm3                   \n\t"
+        "pxor %%mm0, %%mm2                      \n\t"
+        "pxor %%mm1, %%mm3                      \n\t"
+        "psubw %%mm0, %%mm2                     \n\t"
+        "psubw %%mm1, %%mm3                     \n\t"
+        "movq %%mm2, (%0)                       \n\t"
+        "movq %%mm3, 8(%0)                      \n\t"
+        "movq %%mm4, %%mm2                      \n\t"
+        "movq %%mm5, %%mm3                      \n\t"
+        "punpcklwd %%mm7, %%mm4                 \n\t"
+        "punpckhwd %%mm7, %%mm2                 \n\t"
+        "punpcklwd %%mm7, %%mm5                 \n\t"
+        "punpckhwd %%mm7, %%mm3                 \n\t"
+        "paddd (%1), %%mm4                      \n\t"
+        "paddd 8(%1), %%mm2                     \n\t"
+        "paddd 16(%1), %%mm5                    \n\t"
+        "paddd 24(%1), %%mm3                    \n\t"
+        "movq %%mm4, (%1)                       \n\t"
+        "movq %%mm2, 8(%1)                      \n\t"
+        "movq %%mm5, 16(%1)                     \n\t"
+        "movq %%mm3, 24(%1)                     \n\t"
+        "add $16, %0                            \n\t"
+        "add $32, %1                            \n\t"
+        "add $16, %2                            \n\t"
+        "cmp %3, %0                             \n\t"
+            " jb 1b                             \n\t"
+        : "+r" (block), "+r" (sum), "+r" (offset)
+        : "r"(block+64)
+    );
+}
+
+static void  denoise_dct_sse2(MpegEncContext *s, DCTELEM *block){
+    const int intra= s->mb_intra;
+    int *sum= s->dct_error_sum[intra];
+    uint16_t *offset= s->dct_offset[intra];
+
+    s->dct_count[intra]++;
+
+    __asm__ volatile(
+        "pxor %%xmm7, %%xmm7                    \n\t"
+        "1:                                     \n\t"
+        "pxor %%xmm0, %%xmm0                    \n\t"
+        "pxor %%xmm1, %%xmm1                    \n\t"
+        "movdqa (%0), %%xmm2                    \n\t"
+        "movdqa 16(%0), %%xmm3                  \n\t"
+        "pcmpgtw %%xmm2, %%xmm0                 \n\t"
+        "pcmpgtw %%xmm3, %%xmm1                 \n\t"
+        "pxor %%xmm0, %%xmm2                    \n\t"
+        "pxor %%xmm1, %%xmm3                    \n\t"
+        "psubw %%xmm0, %%xmm2                   \n\t"
+        "psubw %%xmm1, %%xmm3                   \n\t"
+        "movdqa %%xmm2, %%xmm4                  \n\t"
+        "movdqa %%xmm3, %%xmm5                  \n\t"
+        "psubusw (%2), %%xmm2                   \n\t"
+        "psubusw 16(%2), %%xmm3                 \n\t"
+        "pxor %%xmm0, %%xmm2                    \n\t"
+        "pxor %%xmm1, %%xmm3                    \n\t"
+        "psubw %%xmm0, %%xmm2                   \n\t"
+        "psubw %%xmm1, %%xmm3                   \n\t"
+        "movdqa %%xmm2, (%0)                    \n\t"
+        "movdqa %%xmm3, 16(%0)                  \n\t"
+        "movdqa %%xmm4, %%xmm6                  \n\t"
+        "movdqa %%xmm5, %%xmm0                  \n\t"
+        "punpcklwd %%xmm7, %%xmm4               \n\t"
+        "punpckhwd %%xmm7, %%xmm6               \n\t"
+        "punpcklwd %%xmm7, %%xmm5               \n\t"
+        "punpckhwd %%xmm7, %%xmm0               \n\t"
+        "paddd (%1), %%xmm4                     \n\t"
+        "paddd 16(%1), %%xmm6                   \n\t"
+        "paddd 32(%1), %%xmm5                   \n\t"
+        "paddd 48(%1), %%xmm0                   \n\t"
+        "movdqa %%xmm4, (%1)                    \n\t"
+        "movdqa %%xmm6, 16(%1)                  \n\t"
+        "movdqa %%xmm5, 32(%1)                  \n\t"
+        "movdqa %%xmm0, 48(%1)                  \n\t"
+        "add $32, %0                            \n\t"
+        "add $64, %1                            \n\t"
+        "add $32, %2                            \n\t"
+        "cmp %3, %0                             \n\t"
+            " jb 1b                             \n\t"
+        : "+r" (block), "+r" (sum), "+r" (offset)
+        : "r"(block+64)
+          XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3",
+                            "%xmm4", "%xmm5", "%xmm6", "%xmm7")
+    );
+}
+
+#endif /* HAVE_INLINE_ASM */
+
+void ff_MPV_common_init_x86(MpegEncContext *s)
+{
+#if HAVE_INLINE_ASM
+    int mm_flags = av_get_cpu_flags();
+
+    if (mm_flags & AV_CPU_FLAG_MMX) {
+        s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_mmx;
+        s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_mmx;
+        s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_mmx;
+        s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_mmx;
+        if(!(s->flags & CODEC_FLAG_BITEXACT))
+            s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_mmx;
+        s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_mmx;
+
+        if (mm_flags & AV_CPU_FLAG_SSE2) {
+            s->denoise_dct= denoise_dct_sse2;
+        } else {
+                s->denoise_dct= denoise_dct_mmx;
+        }
+    }
+#endif /* HAVE_INLINE_ASM */
+}
diff --git a/libavcodec/x86/mpegvideo_mmx.c b/libavcodec/x86/mpegvideo_mmx.c
deleted file mode 100644 (file)
index 720deb3..0000000
+++ /dev/null
@@ -1,608 +0,0 @@
-/*
- * Optimized for ia32 CPUs by Nick Kurshev <nickols_k@mail.ru>
- * h263, mpeg1, mpeg2 dequantizer & draw_edges by Michael Niedermayer <michaelni@gmx.at>
- *
- * This file is part of Libav.
- *
- * Libav 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.
- *
- * Libav 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 Libav; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- */
-
-#include "libavutil/cpu.h"
-#include "libavutil/x86/asm.h"
-#include "libavcodec/avcodec.h"
-#include "libavcodec/dsputil.h"
-#include "libavcodec/mpegvideo.h"
-#include "dsputil_mmx.h"
-
-#if HAVE_INLINE_ASM
-
-static void dct_unquantize_h263_intra_mmx(MpegEncContext *s,
-                                  DCTELEM *block, int n, int qscale)
-{
-    x86_reg level, qmul, qadd, nCoeffs;
-
-    qmul = qscale << 1;
-
-    assert(s->block_last_index[n]>=0 || s->h263_aic);
-
-    if (!s->h263_aic) {
-        if (n < 4)
-            level = block[0] * s->y_dc_scale;
-        else
-            level = block[0] * s->c_dc_scale;
-        qadd = (qscale - 1) | 1;
-    }else{
-        qadd = 0;
-        level= block[0];
-    }
-    if(s->ac_pred)
-        nCoeffs=63;
-    else
-        nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
-//printf("%d %d  ", qmul, qadd);
-__asm__ volatile(
-                "movd %1, %%mm6                 \n\t" //qmul
-                "packssdw %%mm6, %%mm6          \n\t"
-                "packssdw %%mm6, %%mm6          \n\t"
-                "movd %2, %%mm5                 \n\t" //qadd
-                "pxor %%mm7, %%mm7              \n\t"
-                "packssdw %%mm5, %%mm5          \n\t"
-                "packssdw %%mm5, %%mm5          \n\t"
-                "psubw %%mm5, %%mm7             \n\t"
-                "pxor %%mm4, %%mm4              \n\t"
-                ".p2align 4                     \n\t"
-                "1:                             \n\t"
-                "movq (%0, %3), %%mm0           \n\t"
-                "movq 8(%0, %3), %%mm1          \n\t"
-
-                "pmullw %%mm6, %%mm0            \n\t"
-                "pmullw %%mm6, %%mm1            \n\t"
-
-                "movq (%0, %3), %%mm2           \n\t"
-                "movq 8(%0, %3), %%mm3          \n\t"
-
-                "pcmpgtw %%mm4, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
-                "pcmpgtw %%mm4, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
-
-                "pxor %%mm2, %%mm0              \n\t"
-                "pxor %%mm3, %%mm1              \n\t"
-
-                "paddw %%mm7, %%mm0             \n\t"
-                "paddw %%mm7, %%mm1             \n\t"
-
-                "pxor %%mm0, %%mm2              \n\t"
-                "pxor %%mm1, %%mm3              \n\t"
-
-                "pcmpeqw %%mm7, %%mm0           \n\t" // block[i] == 0 ? -1 : 0
-                "pcmpeqw %%mm7, %%mm1           \n\t" // block[i] == 0 ? -1 : 0
-
-                "pandn %%mm2, %%mm0             \n\t"
-                "pandn %%mm3, %%mm1             \n\t"
-
-                "movq %%mm0, (%0, %3)           \n\t"
-                "movq %%mm1, 8(%0, %3)          \n\t"
-
-                "add $16, %3                    \n\t"
-                "jng 1b                         \n\t"
-                ::"r" (block+nCoeffs), "rm"(qmul), "rm" (qadd), "r" (2*(-nCoeffs))
-                : "memory"
-        );
-        block[0]= level;
-}
-
-
-static void dct_unquantize_h263_inter_mmx(MpegEncContext *s,
-                                  DCTELEM *block, int n, int qscale)
-{
-    x86_reg qmul, qadd, nCoeffs;
-
-    qmul = qscale << 1;
-    qadd = (qscale - 1) | 1;
-
-    assert(s->block_last_index[n]>=0 || s->h263_aic);
-
-    nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
-//printf("%d %d  ", qmul, qadd);
-__asm__ volatile(
-                "movd %1, %%mm6                 \n\t" //qmul
-                "packssdw %%mm6, %%mm6          \n\t"
-                "packssdw %%mm6, %%mm6          \n\t"
-                "movd %2, %%mm5                 \n\t" //qadd
-                "pxor %%mm7, %%mm7              \n\t"
-                "packssdw %%mm5, %%mm5          \n\t"
-                "packssdw %%mm5, %%mm5          \n\t"
-                "psubw %%mm5, %%mm7             \n\t"
-                "pxor %%mm4, %%mm4              \n\t"
-                ".p2align 4                     \n\t"
-                "1:                             \n\t"
-                "movq (%0, %3), %%mm0           \n\t"
-                "movq 8(%0, %3), %%mm1          \n\t"
-
-                "pmullw %%mm6, %%mm0            \n\t"
-                "pmullw %%mm6, %%mm1            \n\t"
-
-                "movq (%0, %3), %%mm2           \n\t"
-                "movq 8(%0, %3), %%mm3          \n\t"
-
-                "pcmpgtw %%mm4, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
-                "pcmpgtw %%mm4, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
-
-                "pxor %%mm2, %%mm0              \n\t"
-                "pxor %%mm3, %%mm1              \n\t"
-
-                "paddw %%mm7, %%mm0             \n\t"
-                "paddw %%mm7, %%mm1             \n\t"
-
-                "pxor %%mm0, %%mm2              \n\t"
-                "pxor %%mm1, %%mm3              \n\t"
-
-                "pcmpeqw %%mm7, %%mm0           \n\t" // block[i] == 0 ? -1 : 0
-                "pcmpeqw %%mm7, %%mm1           \n\t" // block[i] == 0 ? -1 : 0
-
-                "pandn %%mm2, %%mm0             \n\t"
-                "pandn %%mm3, %%mm1             \n\t"
-
-                "movq %%mm0, (%0, %3)           \n\t"
-                "movq %%mm1, 8(%0, %3)          \n\t"
-
-                "add $16, %3                    \n\t"
-                "jng 1b                         \n\t"
-                ::"r" (block+nCoeffs), "rm"(qmul), "rm" (qadd), "r" (2*(-nCoeffs))
-                : "memory"
-        );
-}
-
-
-/*
-  NK:
-  Note: looking at PARANOID:
-  "enable all paranoid tests for rounding, overflows, etc..."
-
-#ifdef PARANOID
-                if (level < -2048 || level > 2047)
-                    fprintf(stderr, "unquant error %d %d\n", i, level);
-#endif
-  We can suppose that result of two multiplications can't be greater than 0xFFFF
-  i.e. is 16-bit, so we use here only PMULLW instruction and can avoid
-  a complex multiplication.
-=====================================================
- Full formula for multiplication of 2 integer numbers
- which are represent as high:low words:
- input: value1 = high1:low1
-        value2 = high2:low2
- output: value3 = value1*value2
- value3=high3:low3 (on overflow: modulus 2^32 wrap-around)
- this mean that for 0x123456 * 0x123456 correct result is 0x766cb0ce4
- but this algorithm will compute only 0x66cb0ce4
- this limited by 16-bit size of operands
- ---------------------------------
- tlow1 = high1*low2
- tlow2 = high2*low1
- tlow1 = tlow1 + tlow2
- high3:low3 = low1*low2
- high3 += tlow1
-*/
-static void dct_unquantize_mpeg1_intra_mmx(MpegEncContext *s,
-                                     DCTELEM *block, int n, int qscale)
-{
-    x86_reg nCoeffs;
-    const uint16_t *quant_matrix;
-    int block0;
-
-    assert(s->block_last_index[n]>=0);
-
-    nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]+1;
-
-    if (n < 4)
-        block0 = block[0] * s->y_dc_scale;
-    else
-        block0 = block[0] * s->c_dc_scale;
-    /* XXX: only mpeg1 */
-    quant_matrix = s->intra_matrix;
-__asm__ volatile(
-                "pcmpeqw %%mm7, %%mm7           \n\t"
-                "psrlw $15, %%mm7               \n\t"
-                "movd %2, %%mm6                 \n\t"
-                "packssdw %%mm6, %%mm6          \n\t"
-                "packssdw %%mm6, %%mm6          \n\t"
-                "mov %3, %%"REG_a"              \n\t"
-                ".p2align 4                     \n\t"
-                "1:                             \n\t"
-                "movq (%0, %%"REG_a"), %%mm0    \n\t"
-                "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
-                "movq (%1, %%"REG_a"), %%mm4    \n\t"
-                "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
-                "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
-                "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
-                "pxor %%mm2, %%mm2              \n\t"
-                "pxor %%mm3, %%mm3              \n\t"
-                "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
-                "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
-                "pxor %%mm2, %%mm0              \n\t"
-                "pxor %%mm3, %%mm1              \n\t"
-                "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
-                "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
-                "pmullw %%mm4, %%mm0            \n\t" // abs(block[i])*q
-                "pmullw %%mm5, %%mm1            \n\t" // abs(block[i])*q
-                "pxor %%mm4, %%mm4              \n\t"
-                "pxor %%mm5, %%mm5              \n\t" // FIXME slow
-                "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
-                "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
-                "psraw $3, %%mm0                \n\t"
-                "psraw $3, %%mm1                \n\t"
-                "psubw %%mm7, %%mm0             \n\t"
-                "psubw %%mm7, %%mm1             \n\t"
-                "por %%mm7, %%mm0               \n\t"
-                "por %%mm7, %%mm1               \n\t"
-                "pxor %%mm2, %%mm0              \n\t"
-                "pxor %%mm3, %%mm1              \n\t"
-                "psubw %%mm2, %%mm0             \n\t"
-                "psubw %%mm3, %%mm1             \n\t"
-                "pandn %%mm0, %%mm4             \n\t"
-                "pandn %%mm1, %%mm5             \n\t"
-                "movq %%mm4, (%0, %%"REG_a")    \n\t"
-                "movq %%mm5, 8(%0, %%"REG_a")   \n\t"
-
-                "add $16, %%"REG_a"             \n\t"
-                "js 1b                          \n\t"
-                ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
-                : "%"REG_a, "memory"
-        );
-    block[0]= block0;
-}
-
-static void dct_unquantize_mpeg1_inter_mmx(MpegEncContext *s,
-                                     DCTELEM *block, int n, int qscale)
-{
-    x86_reg nCoeffs;
-    const uint16_t *quant_matrix;
-
-    assert(s->block_last_index[n]>=0);
-
-    nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]+1;
-
-        quant_matrix = s->inter_matrix;
-__asm__ volatile(
-                "pcmpeqw %%mm7, %%mm7           \n\t"
-                "psrlw $15, %%mm7               \n\t"
-                "movd %2, %%mm6                 \n\t"
-                "packssdw %%mm6, %%mm6          \n\t"
-                "packssdw %%mm6, %%mm6          \n\t"
-                "mov %3, %%"REG_a"              \n\t"
-                ".p2align 4                     \n\t"
-                "1:                             \n\t"
-                "movq (%0, %%"REG_a"), %%mm0    \n\t"
-                "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
-                "movq (%1, %%"REG_a"), %%mm4    \n\t"
-                "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
-                "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
-                "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
-                "pxor %%mm2, %%mm2              \n\t"
-                "pxor %%mm3, %%mm3              \n\t"
-                "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
-                "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
-                "pxor %%mm2, %%mm0              \n\t"
-                "pxor %%mm3, %%mm1              \n\t"
-                "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
-                "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
-                "paddw %%mm0, %%mm0             \n\t" // abs(block[i])*2
-                "paddw %%mm1, %%mm1             \n\t" // abs(block[i])*2
-                "paddw %%mm7, %%mm0             \n\t" // abs(block[i])*2 + 1
-                "paddw %%mm7, %%mm1             \n\t" // abs(block[i])*2 + 1
-                "pmullw %%mm4, %%mm0            \n\t" // (abs(block[i])*2 + 1)*q
-                "pmullw %%mm5, %%mm1            \n\t" // (abs(block[i])*2 + 1)*q
-                "pxor %%mm4, %%mm4              \n\t"
-                "pxor %%mm5, %%mm5              \n\t" // FIXME slow
-                "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
-                "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
-                "psraw $4, %%mm0                \n\t"
-                "psraw $4, %%mm1                \n\t"
-                "psubw %%mm7, %%mm0             \n\t"
-                "psubw %%mm7, %%mm1             \n\t"
-                "por %%mm7, %%mm0               \n\t"
-                "por %%mm7, %%mm1               \n\t"
-                "pxor %%mm2, %%mm0              \n\t"
-                "pxor %%mm3, %%mm1              \n\t"
-                "psubw %%mm2, %%mm0             \n\t"
-                "psubw %%mm3, %%mm1             \n\t"
-                "pandn %%mm0, %%mm4             \n\t"
-                "pandn %%mm1, %%mm5             \n\t"
-                "movq %%mm4, (%0, %%"REG_a")    \n\t"
-                "movq %%mm5, 8(%0, %%"REG_a")   \n\t"
-
-                "add $16, %%"REG_a"             \n\t"
-                "js 1b                          \n\t"
-                ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
-                : "%"REG_a, "memory"
-        );
-}
-
-static void dct_unquantize_mpeg2_intra_mmx(MpegEncContext *s,
-                                     DCTELEM *block, int n, int qscale)
-{
-    x86_reg nCoeffs;
-    const uint16_t *quant_matrix;
-    int block0;
-
-    assert(s->block_last_index[n]>=0);
-
-    if(s->alternate_scan) nCoeffs= 63; //FIXME
-    else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
-
-    if (n < 4)
-        block0 = block[0] * s->y_dc_scale;
-    else
-        block0 = block[0] * s->c_dc_scale;
-    quant_matrix = s->intra_matrix;
-__asm__ volatile(
-                "pcmpeqw %%mm7, %%mm7           \n\t"
-                "psrlw $15, %%mm7               \n\t"
-                "movd %2, %%mm6                 \n\t"
-                "packssdw %%mm6, %%mm6          \n\t"
-                "packssdw %%mm6, %%mm6          \n\t"
-                "mov %3, %%"REG_a"              \n\t"
-                ".p2align 4                     \n\t"
-                "1:                             \n\t"
-                "movq (%0, %%"REG_a"), %%mm0    \n\t"
-                "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
-                "movq (%1, %%"REG_a"), %%mm4    \n\t"
-                "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
-                "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
-                "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
-                "pxor %%mm2, %%mm2              \n\t"
-                "pxor %%mm3, %%mm3              \n\t"
-                "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
-                "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
-                "pxor %%mm2, %%mm0              \n\t"
-                "pxor %%mm3, %%mm1              \n\t"
-                "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
-                "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
-                "pmullw %%mm4, %%mm0            \n\t" // abs(block[i])*q
-                "pmullw %%mm5, %%mm1            \n\t" // abs(block[i])*q
-                "pxor %%mm4, %%mm4              \n\t"
-                "pxor %%mm5, %%mm5              \n\t" // FIXME slow
-                "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
-                "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
-                "psraw $3, %%mm0                \n\t"
-                "psraw $3, %%mm1                \n\t"
-                "pxor %%mm2, %%mm0              \n\t"
-                "pxor %%mm3, %%mm1              \n\t"
-                "psubw %%mm2, %%mm0             \n\t"
-                "psubw %%mm3, %%mm1             \n\t"
-                "pandn %%mm0, %%mm4             \n\t"
-                "pandn %%mm1, %%mm5             \n\t"
-                "movq %%mm4, (%0, %%"REG_a")    \n\t"
-                "movq %%mm5, 8(%0, %%"REG_a")   \n\t"
-
-                "add $16, %%"REG_a"             \n\t"
-                "jng 1b                         \n\t"
-                ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
-                : "%"REG_a, "memory"
-        );
-    block[0]= block0;
-        //Note, we do not do mismatch control for intra as errors cannot accumulate
-}
-
-static void dct_unquantize_mpeg2_inter_mmx(MpegEncContext *s,
-                                     DCTELEM *block, int n, int qscale)
-{
-    x86_reg nCoeffs;
-    const uint16_t *quant_matrix;
-
-    assert(s->block_last_index[n]>=0);
-
-    if(s->alternate_scan) nCoeffs= 63; //FIXME
-    else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
-
-        quant_matrix = s->inter_matrix;
-__asm__ volatile(
-                "pcmpeqw %%mm7, %%mm7           \n\t"
-                "psrlq $48, %%mm7               \n\t"
-                "movd %2, %%mm6                 \n\t"
-                "packssdw %%mm6, %%mm6          \n\t"
-                "packssdw %%mm6, %%mm6          \n\t"
-                "mov %3, %%"REG_a"              \n\t"
-                ".p2align 4                     \n\t"
-                "1:                             \n\t"
-                "movq (%0, %%"REG_a"), %%mm0    \n\t"
-                "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
-                "movq (%1, %%"REG_a"), %%mm4    \n\t"
-                "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
-                "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
-                "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
-                "pxor %%mm2, %%mm2              \n\t"
-                "pxor %%mm3, %%mm3              \n\t"
-                "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
-                "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
-                "pxor %%mm2, %%mm0              \n\t"
-                "pxor %%mm3, %%mm1              \n\t"
-                "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
-                "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
-                "paddw %%mm0, %%mm0             \n\t" // abs(block[i])*2
-                "paddw %%mm1, %%mm1             \n\t" // abs(block[i])*2
-                "pmullw %%mm4, %%mm0            \n\t" // abs(block[i])*2*q
-                "pmullw %%mm5, %%mm1            \n\t" // abs(block[i])*2*q
-                "paddw %%mm4, %%mm0             \n\t" // (abs(block[i])*2 + 1)*q
-                "paddw %%mm5, %%mm1             \n\t" // (abs(block[i])*2 + 1)*q
-                "pxor %%mm4, %%mm4              \n\t"
-                "pxor %%mm5, %%mm5              \n\t" // FIXME slow
-                "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
-                "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
-                "psrlw $4, %%mm0                \n\t"
-                "psrlw $4, %%mm1                \n\t"
-                "pxor %%mm2, %%mm0              \n\t"
-                "pxor %%mm3, %%mm1              \n\t"
-                "psubw %%mm2, %%mm0             \n\t"
-                "psubw %%mm3, %%mm1             \n\t"
-                "pandn %%mm0, %%mm4             \n\t"
-                "pandn %%mm1, %%mm5             \n\t"
-                "pxor %%mm4, %%mm7              \n\t"
-                "pxor %%mm5, %%mm7              \n\t"
-                "movq %%mm4, (%0, %%"REG_a")    \n\t"
-                "movq %%mm5, 8(%0, %%"REG_a")   \n\t"
-
-                "add $16, %%"REG_a"             \n\t"
-                "jng 1b                         \n\t"
-                "movd 124(%0, %3), %%mm0        \n\t"
-                "movq %%mm7, %%mm6              \n\t"
-                "psrlq $32, %%mm7               \n\t"
-                "pxor %%mm6, %%mm7              \n\t"
-                "movq %%mm7, %%mm6              \n\t"
-                "psrlq $16, %%mm7               \n\t"
-                "pxor %%mm6, %%mm7              \n\t"
-                "pslld $31, %%mm7               \n\t"
-                "psrlq $15, %%mm7               \n\t"
-                "pxor %%mm7, %%mm0              \n\t"
-                "movd %%mm0, 124(%0, %3)        \n\t"
-
-                ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "r" (-2*nCoeffs)
-                : "%"REG_a, "memory"
-        );
-}
-
-static void  denoise_dct_mmx(MpegEncContext *s, DCTELEM *block){
-    const int intra= s->mb_intra;
-    int *sum= s->dct_error_sum[intra];
-    uint16_t *offset= s->dct_offset[intra];
-
-    s->dct_count[intra]++;
-
-    __asm__ volatile(
-        "pxor %%mm7, %%mm7                      \n\t"
-        "1:                                     \n\t"
-        "pxor %%mm0, %%mm0                      \n\t"
-        "pxor %%mm1, %%mm1                      \n\t"
-        "movq (%0), %%mm2                       \n\t"
-        "movq 8(%0), %%mm3                      \n\t"
-        "pcmpgtw %%mm2, %%mm0                   \n\t"
-        "pcmpgtw %%mm3, %%mm1                   \n\t"
-        "pxor %%mm0, %%mm2                      \n\t"
-        "pxor %%mm1, %%mm3                      \n\t"
-        "psubw %%mm0, %%mm2                     \n\t"
-        "psubw %%mm1, %%mm3                     \n\t"
-        "movq %%mm2, %%mm4                      \n\t"
-        "movq %%mm3, %%mm5                      \n\t"
-        "psubusw (%2), %%mm2                    \n\t"
-        "psubusw 8(%2), %%mm3                   \n\t"
-        "pxor %%mm0, %%mm2                      \n\t"
-        "pxor %%mm1, %%mm3                      \n\t"
-        "psubw %%mm0, %%mm2                     \n\t"
-        "psubw %%mm1, %%mm3                     \n\t"
-        "movq %%mm2, (%0)                       \n\t"
-        "movq %%mm3, 8(%0)                      \n\t"
-        "movq %%mm4, %%mm2                      \n\t"
-        "movq %%mm5, %%mm3                      \n\t"
-        "punpcklwd %%mm7, %%mm4                 \n\t"
-        "punpckhwd %%mm7, %%mm2                 \n\t"
-        "punpcklwd %%mm7, %%mm5                 \n\t"
-        "punpckhwd %%mm7, %%mm3                 \n\t"
-        "paddd (%1), %%mm4                      \n\t"
-        "paddd 8(%1), %%mm2                     \n\t"
-        "paddd 16(%1), %%mm5                    \n\t"
-        "paddd 24(%1), %%mm3                    \n\t"
-        "movq %%mm4, (%1)                       \n\t"
-        "movq %%mm2, 8(%1)                      \n\t"
-        "movq %%mm5, 16(%1)                     \n\t"
-        "movq %%mm3, 24(%1)                     \n\t"
-        "add $16, %0                            \n\t"
-        "add $32, %1                            \n\t"
-        "add $16, %2                            \n\t"
-        "cmp %3, %0                             \n\t"
-            " jb 1b                             \n\t"
-        : "+r" (block), "+r" (sum), "+r" (offset)
-        : "r"(block+64)
-    );
-}
-
-static void  denoise_dct_sse2(MpegEncContext *s, DCTELEM *block){
-    const int intra= s->mb_intra;
-    int *sum= s->dct_error_sum[intra];
-    uint16_t *offset= s->dct_offset[intra];
-
-    s->dct_count[intra]++;
-
-    __asm__ volatile(
-        "pxor %%xmm7, %%xmm7                    \n\t"
-        "1:                                     \n\t"
-        "pxor %%xmm0, %%xmm0                    \n\t"
-        "pxor %%xmm1, %%xmm1                    \n\t"
-        "movdqa (%0), %%xmm2                    \n\t"
-        "movdqa 16(%0), %%xmm3                  \n\t"
-        "pcmpgtw %%xmm2, %%xmm0                 \n\t"
-        "pcmpgtw %%xmm3, %%xmm1                 \n\t"
-        "pxor %%xmm0, %%xmm2                    \n\t"
-        "pxor %%xmm1, %%xmm3                    \n\t"
-        "psubw %%xmm0, %%xmm2                   \n\t"
-        "psubw %%xmm1, %%xmm3                   \n\t"
-        "movdqa %%xmm2, %%xmm4                  \n\t"
-        "movdqa %%xmm3, %%xmm5                  \n\t"
-        "psubusw (%2), %%xmm2                   \n\t"
-        "psubusw 16(%2), %%xmm3                 \n\t"
-        "pxor %%xmm0, %%xmm2                    \n\t"
-        "pxor %%xmm1, %%xmm3                    \n\t"
-        "psubw %%xmm0, %%xmm2                   \n\t"
-        "psubw %%xmm1, %%xmm3                   \n\t"
-        "movdqa %%xmm2, (%0)                    \n\t"
-        "movdqa %%xmm3, 16(%0)                  \n\t"
-        "movdqa %%xmm4, %%xmm6                  \n\t"
-        "movdqa %%xmm5, %%xmm0                  \n\t"
-        "punpcklwd %%xmm7, %%xmm4               \n\t"
-        "punpckhwd %%xmm7, %%xmm6               \n\t"
-        "punpcklwd %%xmm7, %%xmm5               \n\t"
-        "punpckhwd %%xmm7, %%xmm0               \n\t"
-        "paddd (%1), %%xmm4                     \n\t"
-        "paddd 16(%1), %%xmm6                   \n\t"
-        "paddd 32(%1), %%xmm5                   \n\t"
-        "paddd 48(%1), %%xmm0                   \n\t"
-        "movdqa %%xmm4, (%1)                    \n\t"
-        "movdqa %%xmm6, 16(%1)                  \n\t"
-        "movdqa %%xmm5, 32(%1)                  \n\t"
-        "movdqa %%xmm0, 48(%1)                  \n\t"
-        "add $32, %0                            \n\t"
-        "add $64, %1                            \n\t"
-        "add $32, %2                            \n\t"
-        "cmp %3, %0                             \n\t"
-            " jb 1b                             \n\t"
-        : "+r" (block), "+r" (sum), "+r" (offset)
-        : "r"(block+64)
-          XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3",
-                            "%xmm4", "%xmm5", "%xmm6", "%xmm7")
-    );
-}
-
-#endif /* HAVE_INLINE_ASM */
-
-void ff_MPV_common_init_mmx(MpegEncContext *s)
-{
-#if HAVE_INLINE_ASM
-    int mm_flags = av_get_cpu_flags();
-
-    if (mm_flags & AV_CPU_FLAG_MMX) {
-        s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_mmx;
-        s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_mmx;
-        s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_mmx;
-        s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_mmx;
-        if(!(s->flags & CODEC_FLAG_BITEXACT))
-            s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_mmx;
-        s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_mmx;
-
-        if (mm_flags & AV_CPU_FLAG_SSE2) {
-            s->denoise_dct= denoise_dct_sse2;
-        } else {
-                s->denoise_dct= denoise_dct_mmx;
-        }
-    }
-#endif /* HAVE_INLINE_ASM */
-}