update mmx code to latest snow changes
authorMichael Niedermayer <michaelni@gmx.at>
Sat, 25 Aug 2007 15:20:56 +0000 (15:20 +0000)
committerMichael Niedermayer <michaelni@gmx.at>
Sat, 25 Aug 2007 15:20:56 +0000 (15:20 +0000)
note, the code likely can overflow and thus needs some more changes
sse2 updated too but disabled as it is untested

Originally committed as revision 10223 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/i386/dsputil_mmx.c
libavcodec/i386/snowdsp_mmx.c

index 6540db2..3ebf148 100644 (file)
@@ -3621,8 +3621,7 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
 #endif
 
 #ifdef CONFIG_SNOW_DECODER
-#if 0
-        if(mm_flags & MM_SSE2){
+        if(mm_flags & MM_SSE2 & 0){
             c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2;
             c->vertical_compose97i = ff_snow_vertical_compose97i_sse2;
             c->inner_add_yblock = ff_snow_inner_add_yblock_sse2;
@@ -3632,7 +3631,6 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
             c->vertical_compose97i = ff_snow_vertical_compose97i_mmx;
             c->inner_add_yblock = ff_snow_inner_add_yblock_mmx;
         }
-#endif
 #endif
 
         if(mm_flags & MM_3DNOW){
index 03f622b..8a2262f 100644 (file)
 #include "snow.h"
 #include "x86_cpu.h"
 
-void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width){
+void ff_snow_horizontal_compose97i_sse2(IDWTELEM *b, int width){
     const int w2= (width+1)>>1;
     // SSE2 code runs faster with pointers aligned on a 32-byte boundary.
-    DWTELEM temp_buf[(width>>1) + 4];
-    DWTELEM * const temp = temp_buf + 4 - (((int)temp_buf & 0xF) >> 2);
+    IDWTELEM temp_buf[(width>>1) + 4];
+    IDWTELEM * const temp = temp_buf + 4 - (((int)temp_buf & 0xF) >> 2);
     const int w_l= (width>>1);
     const int w_r= w2 - 1;
     int i;
 
     { // Lift 0
-        DWTELEM * const ref = b + w2 - 1;
-        DWTELEM b_0 = b[0]; //By allowing the first entry in b[0] to be calculated twice
+        IDWTELEM * const ref = b + w2 - 1;
+        IDWTELEM b_0 = b[0]; //By allowing the first entry in b[0] to be calculated twice
         // (the first time erroneously), we allow the SSE2 code to run an extra pass.
         // The savings in code and time are well worth having to store this value and
         // calculate b[0] correctly afterwards.
@@ -42,31 +42,31 @@ void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width){
         i = 0;
         asm volatile(
             "pcmpeqd   %%xmm7, %%xmm7         \n\t"
-            "pslld        $31, %%xmm7         \n\t"
-            "psrld        $29, %%xmm7         \n\t"
+            "psllw        $15, %%xmm7         \n\t"
+            "psrlw        $13, %%xmm7         \n\t"
         ::);
-        for(; i<w_l-7; i+=8){
+        for(; i<w_l-15; i+=16){
             asm volatile(
                 "movdqu   (%1), %%xmm1        \n\t"
                 "movdqu 16(%1), %%xmm5        \n\t"
-                "movdqu  4(%1), %%xmm2        \n\t"
-                "movdqu 20(%1), %%xmm6        \n\t"
-                "paddd  %%xmm1, %%xmm2        \n\t"
-                "paddd  %%xmm5, %%xmm6        \n\t"
+                "movdqu  2(%1), %%xmm2        \n\t"
+                "movdqu 18(%1), %%xmm6        \n\t"
+                "paddw  %%xmm1, %%xmm2        \n\t"
+                "paddw  %%xmm5, %%xmm6        \n\t"
                 "movdqa %%xmm2, %%xmm0        \n\t"
                 "movdqa %%xmm6, %%xmm4        \n\t"
-                "paddd  %%xmm2, %%xmm2        \n\t"
-                "paddd  %%xmm6, %%xmm6        \n\t"
-                "paddd  %%xmm0, %%xmm2        \n\t"
-                "paddd  %%xmm4, %%xmm6        \n\t"
-                "paddd  %%xmm7, %%xmm2        \n\t"
-                "paddd  %%xmm7, %%xmm6        \n\t"
-                "psrad      $3, %%xmm2        \n\t"
-                "psrad      $3, %%xmm6        \n\t"
+                "paddw  %%xmm2, %%xmm2        \n\t"
+                "paddw  %%xmm6, %%xmm6        \n\t"
+                "paddw  %%xmm0, %%xmm2        \n\t"
+                "paddw  %%xmm4, %%xmm6        \n\t"
+                "paddw  %%xmm7, %%xmm2        \n\t"
+                "paddw  %%xmm7, %%xmm6        \n\t"
+                "psraw      $3, %%xmm2        \n\t"
+                "psraw      $3, %%xmm6        \n\t"
                 "movdqa   (%0), %%xmm0        \n\t"
                 "movdqa 16(%0), %%xmm4        \n\t"
-                "psubd  %%xmm2, %%xmm0        \n\t"
-                "psubd  %%xmm6, %%xmm4        \n\t"
+                "psubw  %%xmm2, %%xmm0        \n\t"
+                "psubw  %%xmm6, %%xmm4        \n\t"
                 "movdqa %%xmm0, (%0)          \n\t"
                 "movdqa %%xmm4, 16(%0)        \n\t"
                 :: "r"(&b[i]), "r"(&ref[i])
@@ -78,24 +78,24 @@ void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width){
     }
 
     { // Lift 1
-        DWTELEM * const dst = b+w2;
+        IDWTELEM * const dst = b+w2;
 
         i = 0;
-        for(; (((long)&dst[i]) & 0xF) && i<w_r; i++){
+        for(; (((long)&dst[i]) & 0x1F) && i<w_r; i++){
             dst[i] = dst[i] - (b[i] + b[i + 1]);
         }
-        for(; i<w_r-7; i+=8){
+        for(; i<w_r-15; i+=16){
             asm volatile(
                 "movdqu   (%1), %%xmm1        \n\t"
                 "movdqu 16(%1), %%xmm5        \n\t"
-                "movdqu  4(%1), %%xmm2        \n\t"
-                "movdqu 20(%1), %%xmm6        \n\t"
-                "paddd  %%xmm1, %%xmm2        \n\t"
-                "paddd  %%xmm5, %%xmm6        \n\t"
+                "movdqu  2(%1), %%xmm2        \n\t"
+                "movdqu 18(%1), %%xmm6        \n\t"
+                "paddw  %%xmm1, %%xmm2        \n\t"
+                "paddw  %%xmm5, %%xmm6        \n\t"
                 "movdqa   (%0), %%xmm0        \n\t"
                 "movdqa 16(%0), %%xmm4        \n\t"
-                "psubd  %%xmm2, %%xmm0        \n\t"
-                "psubd  %%xmm6, %%xmm4        \n\t"
+                "psubw  %%xmm2, %%xmm0        \n\t"
+                "psubw  %%xmm6, %%xmm4        \n\t"
                 "movdqa %%xmm0, (%0)          \n\t"
                 "movdqa %%xmm4, 16(%0)        \n\t"
                 :: "r"(&dst[i]), "r"(&b[i])
@@ -106,33 +106,33 @@ void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width){
     }
 
     { // Lift 2
-        DWTELEM * const ref = b+w2 - 1;
-        DWTELEM b_0 = b[0];
+        IDWTELEM * const ref = b+w2 - 1;
+        IDWTELEM b_0 = b[0];
 
         i = 0;
         asm volatile(
-            "pslld          $1, %%xmm7        \n\t"
+            "psllw          $1, %%xmm7        \n\t"
         ::);
-        for(; i<w_l-7; i+=8){
+        for(; i<w_l-15; i+=16){
             asm volatile(
                 "movdqu   (%1), %%xmm1        \n\t"
                 "movdqu 16(%1), %%xmm5        \n\t"
-                "movdqu  4(%1), %%xmm0        \n\t"
-                "movdqu 20(%1), %%xmm4        \n\t" //FIXME try aligned reads and shifts
-                "paddd  %%xmm1, %%xmm0        \n\t"
-                "paddd  %%xmm5, %%xmm4        \n\t"
-                "paddd  %%xmm7, %%xmm0        \n\t"
-                "paddd  %%xmm7, %%xmm4        \n\t"
+                "movdqu  2(%1), %%xmm0        \n\t"
+                "movdqu 18(%1), %%xmm4        \n\t" //FIXME try aligned reads and shifts
+                "paddw  %%xmm1, %%xmm0        \n\t"
+                "paddw  %%xmm5, %%xmm4        \n\t"
+                "paddw  %%xmm7, %%xmm0        \n\t"
+                "paddw  %%xmm7, %%xmm4        \n\t"
                 "movdqa   (%0), %%xmm1        \n\t"
                 "movdqa 16(%0), %%xmm5        \n\t"
-                "psrad      $2, %%xmm0        \n\t"
-                "psrad      $2, %%xmm4        \n\t"
-                "paddd  %%xmm1, %%xmm0        \n\t"
-                "paddd  %%xmm5, %%xmm4        \n\t"
-                "psrad      $2, %%xmm0        \n\t"
-                "psrad      $2, %%xmm4        \n\t"
-                "paddd  %%xmm1, %%xmm0        \n\t"
-                "paddd  %%xmm5, %%xmm4        \n\t"
+                "psraw      $2, %%xmm0        \n\t"
+                "psraw      $2, %%xmm4        \n\t"
+                "paddw  %%xmm1, %%xmm0        \n\t"
+                "paddw  %%xmm5, %%xmm4        \n\t"
+                "psraw      $2, %%xmm0        \n\t"
+                "psraw      $2, %%xmm4        \n\t"
+                "paddw  %%xmm1, %%xmm0        \n\t"
+                "paddw  %%xmm5, %%xmm4        \n\t"
                 "movdqa %%xmm0, (%0)          \n\t"
                 "movdqa %%xmm4, 16(%0)        \n\t"
                 :: "r"(&b[i]), "r"(&ref[i])
@@ -144,26 +144,26 @@ void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width){
     }
 
     { // Lift 3
-        DWTELEM * const src = b+w2;
+        IDWTELEM * const src = b+w2;
 
         i = 0;
-        for(; (((long)&temp[i]) & 0xF) && i<w_r; i++){
+        for(; (((long)&temp[i]) & 0x1F) && i<w_r; i++){
             temp[i] = src[i] - ((-W_AM*(b[i] + b[i+1]))>>W_AS);
         }
         for(; i<w_r-7; i+=8){
             asm volatile(
-                "movdqu  4(%1), %%xmm2        \n\t"
-                "movdqu 20(%1), %%xmm6        \n\t"
-                "paddd    (%1), %%xmm2        \n\t"
-                "paddd  16(%1), %%xmm6        \n\t"
+                "movdqu  2(%1), %%xmm2        \n\t"
+                "movdqu 18(%1), %%xmm6        \n\t"
+                "paddw    (%1), %%xmm2        \n\t"
+                "paddw  16(%1), %%xmm6        \n\t"
                 "movdqu   (%0), %%xmm0        \n\t"
                 "movdqu 16(%0), %%xmm4        \n\t"
-                "paddd  %%xmm2, %%xmm0        \n\t"
-                "paddd  %%xmm6, %%xmm4        \n\t"
-                "psrad      $1, %%xmm2        \n\t"
-                "psrad      $1, %%xmm6        \n\t"
-                "paddd  %%xmm0, %%xmm2        \n\t"
-                "paddd  %%xmm4, %%xmm6        \n\t"
+                "paddw  %%xmm2, %%xmm0        \n\t"
+                "paddw  %%xmm6, %%xmm4        \n\t"
+                "psraw      $1, %%xmm2        \n\t"
+                "psraw      $1, %%xmm6        \n\t"
+                "paddw  %%xmm0, %%xmm2        \n\t"
+                "paddw  %%xmm4, %%xmm6        \n\t"
                 "movdqa %%xmm2, (%2)          \n\t"
                 "movdqa %%xmm6, 16(%2)        \n\t"
                 :: "r"(&src[i]), "r"(&b[i]), "r"(&temp[i])
@@ -176,11 +176,11 @@ void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width){
     {
         snow_interleave_line_header(&i, width, b, temp);
 
-        for (; (i & 0x1E) != 0x1E; i-=2){
+        for (; (i & 0x3E) != 0x3E; i-=2){
             b[i+1] = temp[i>>1];
             b[i] = b[i>>1];
         }
-        for (i-=30; i>=0; i-=32){
+        for (i-=62; i>=0; i-=64){
             asm volatile(
                 "movdqa      (%1), %%xmm0       \n\t"
                 "movdqa    16(%1), %%xmm2       \n\t"
@@ -190,18 +190,18 @@ void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width){
                 "movdqa    16(%1), %%xmm3       \n\t"
                 "movdqa    32(%1), %%xmm5       \n\t"
                 "movdqa    48(%1), %%xmm7       \n\t"
-                "punpckldq   (%2), %%xmm0       \n\t"
-                "punpckldq 16(%2), %%xmm2       \n\t"
-                "punpckldq 32(%2), %%xmm4       \n\t"
-                "punpckldq 48(%2), %%xmm6       \n\t"
+                "punpcklwd   (%2), %%xmm0       \n\t"
+                "punpcklwd 16(%2), %%xmm2       \n\t"
+                "punpcklwd 32(%2), %%xmm4       \n\t"
+                "punpcklwd 48(%2), %%xmm6       \n\t"
                 "movdqa    %%xmm0, (%0)         \n\t"
                 "movdqa    %%xmm2, 32(%0)       \n\t"
                 "movdqa    %%xmm4, 64(%0)       \n\t"
                 "movdqa    %%xmm6, 96(%0)       \n\t"
-                "punpckhdq   (%2), %%xmm1       \n\t"
-                "punpckhdq 16(%2), %%xmm3       \n\t"
-                "punpckhdq 32(%2), %%xmm5       \n\t"
-                "punpckhdq 48(%2), %%xmm7       \n\t"
+                "punpckhwd   (%2), %%xmm1       \n\t"
+                "punpckhwd 16(%2), %%xmm3       \n\t"
+                "punpckhwd 32(%2), %%xmm5       \n\t"
+                "punpckhwd 48(%2), %%xmm7       \n\t"
                 "movdqa    %%xmm1, 16(%0)       \n\t"
                 "movdqa    %%xmm3, 48(%0)       \n\t"
                 "movdqa    %%xmm5, 80(%0)       \n\t"
@@ -213,43 +213,43 @@ void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width){
     }
 }
 
-void ff_snow_horizontal_compose97i_mmx(DWTELEM *b, int width){
+void ff_snow_horizontal_compose97i_mmx(IDWTELEM *b, int width){
     const int w2= (width+1)>>1;
-    DWTELEM temp[width >> 1];
+    IDWTELEM temp[width >> 1];
     const int w_l= (width>>1);
     const int w_r= w2 - 1;
     int i;
 
     { // Lift 0
-        DWTELEM * const ref = b + w2 - 1;
+        IDWTELEM * const ref = b + w2 - 1;
 
         i = 1;
         b[0] = b[0] - ((W_DM * 2 * ref[1]+W_DO)>>W_DS);
         asm volatile(
-            "pcmpeqd    %%mm7, %%mm7         \n\t"
-            "pslld        $31, %%mm7         \n\t"
-            "psrld        $29, %%mm7         \n\t"
+            "pcmpeqw    %%mm7, %%mm7         \n\t"
+            "psllw        $15, %%mm7         \n\t"
+            "psrlw        $13, %%mm7         \n\t"
            ::);
-        for(; i<w_l-3; i+=4){
+        for(; i<w_l-7; i+=8){
             asm volatile(
                 "movq     (%1), %%mm2        \n\t"
                 "movq    8(%1), %%mm6        \n\t"
-                "paddd   4(%1), %%mm2        \n\t"
-                "paddd  12(%1), %%mm6        \n\t"
+                "paddw   2(%1), %%mm2        \n\t"
+                "paddw  10(%1), %%mm6        \n\t"
                 "movq    %%mm2, %%mm0        \n\t"
                 "movq    %%mm6, %%mm4        \n\t"
-                "paddd   %%mm2, %%mm2        \n\t"
-                "paddd   %%mm6, %%mm6        \n\t"
-                "paddd   %%mm0, %%mm2        \n\t"
-                "paddd   %%mm4, %%mm6        \n\t"
-                "paddd   %%mm7, %%mm2        \n\t"
-                "paddd   %%mm7, %%mm6        \n\t"
-                "psrad      $3, %%mm2        \n\t"
-                "psrad      $3, %%mm6        \n\t"
+                "paddw   %%mm2, %%mm2        \n\t"
+                "paddw   %%mm6, %%mm6        \n\t"
+                "paddw   %%mm0, %%mm2        \n\t"
+                "paddw   %%mm4, %%mm6        \n\t"
+                "paddw   %%mm7, %%mm2        \n\t"
+                "paddw   %%mm7, %%mm6        \n\t"
+                "psraw      $3, %%mm2        \n\t"
+                "psraw      $3, %%mm6        \n\t"
                 "movq     (%0), %%mm0        \n\t"
                 "movq    8(%0), %%mm4        \n\t"
-                "psubd   %%mm2, %%mm0        \n\t"
-                "psubd   %%mm6, %%mm4        \n\t"
+                "psubw   %%mm2, %%mm0        \n\t"
+                "psubw   %%mm6, %%mm4        \n\t"
                 "movq    %%mm0, (%0)         \n\t"
                 "movq    %%mm4, 8(%0)        \n\t"
                 :: "r"(&b[i]), "r"(&ref[i])
@@ -260,19 +260,19 @@ void ff_snow_horizontal_compose97i_mmx(DWTELEM *b, int width){
     }
 
     { // Lift 1
-        DWTELEM * const dst = b+w2;
+        IDWTELEM * const dst = b+w2;
 
         i = 0;
-        for(; i<w_r-3; i+=4){
+        for(; i<w_r-7; i+=8){
             asm volatile(
                 "movq     (%1), %%mm2        \n\t"
                 "movq    8(%1), %%mm6        \n\t"
-                "paddd   4(%1), %%mm2        \n\t"
-                "paddd  12(%1), %%mm6        \n\t"
+                "paddw   2(%1), %%mm2        \n\t"
+                "paddw  10(%1), %%mm6        \n\t"
                 "movq     (%0), %%mm0        \n\t"
                 "movq    8(%0), %%mm4        \n\t"
-                "psubd   %%mm2, %%mm0        \n\t"
-                "psubd   %%mm6, %%mm4        \n\t"
+                "psubw   %%mm2, %%mm0        \n\t"
+                "psubw   %%mm6, %%mm4        \n\t"
                 "movq    %%mm0, (%0)         \n\t"
                 "movq    %%mm4, 8(%0)        \n\t"
                 :: "r"(&dst[i]), "r"(&b[i])
@@ -283,31 +283,31 @@ void ff_snow_horizontal_compose97i_mmx(DWTELEM *b, int width){
     }
 
     { // Lift 2
-        DWTELEM * const ref = b+w2 - 1;
+        IDWTELEM * const ref = b+w2 - 1;
 
         i = 1;
         b[0] = b[0] + (((2 * ref[1] + W_BO) + 4 * b[0]) >> W_BS);
         asm volatile(
-            "pslld          $1, %%mm7        \n\t"
+            "psllw          $1, %%mm7        \n\t"
            ::);
-        for(; i<w_l-3; i+=4){
+        for(; i<w_l-7; i+=8){
             asm volatile(
                 "movq     (%1), %%mm0        \n\t"
                 "movq    8(%1), %%mm4        \n\t"
-                "paddd   4(%1), %%mm0        \n\t"
-                "paddd  12(%1), %%mm4        \n\t"
-                "paddd   %%mm7, %%mm0        \n\t"
-                "paddd   %%mm7, %%mm4        \n\t"
-                "psrad      $2, %%mm0        \n\t"
-                "psrad      $2, %%mm4        \n\t"
+                "paddw   2(%1), %%mm0        \n\t"
+                "paddw  10(%1), %%mm4        \n\t"
+                "paddw   %%mm7, %%mm0        \n\t"
+                "paddw   %%mm7, %%mm4        \n\t"
+                "psraw      $2, %%mm0        \n\t"
+                "psraw      $2, %%mm4        \n\t"
                 "movq     (%0), %%mm1        \n\t"
                 "movq    8(%0), %%mm5        \n\t"
-                "paddd   %%mm1, %%mm0        \n\t"
-                "paddd   %%mm5, %%mm4        \n\t"
-                "psrad      $2, %%mm0        \n\t"
-                "psrad      $2, %%mm4        \n\t"
-                "paddd   %%mm1, %%mm0        \n\t"
-                "paddd   %%mm5, %%mm4        \n\t"
+                "paddw   %%mm1, %%mm0        \n\t"
+                "paddw   %%mm5, %%mm4        \n\t"
+                "psraw      $2, %%mm0        \n\t"
+                "psraw      $2, %%mm4        \n\t"
+                "paddw   %%mm1, %%mm0        \n\t"
+                "paddw   %%mm5, %%mm4        \n\t"
                 "movq    %%mm0, (%0)         \n\t"
                 "movq    %%mm4, 8(%0)        \n\t"
                 :: "r"(&b[i]), "r"(&ref[i])
@@ -318,23 +318,23 @@ void ff_snow_horizontal_compose97i_mmx(DWTELEM *b, int width){
     }
 
     { // Lift 3
-        DWTELEM * const src = b+w2;
+        IDWTELEM * const src = b+w2;
         i = 0;
 
-        for(; i<w_r-3; i+=4){
+        for(; i<w_r-7; i+=8){
             asm volatile(
-                "movq    4(%1), %%mm2        \n\t"
-                "movq   12(%1), %%mm6        \n\t"
-                "paddd    (%1), %%mm2        \n\t"
-                "paddd   8(%1), %%mm6        \n\t"
+                "movq    2(%1), %%mm2        \n\t"
+                "movq   10(%1), %%mm6        \n\t"
+                "paddw    (%1), %%mm2        \n\t"
+                "paddw   8(%1), %%mm6        \n\t"
                 "movq     (%0), %%mm0        \n\t"
                 "movq    8(%0), %%mm4        \n\t"
-                "paddd   %%mm2, %%mm0        \n\t"
-                "paddd   %%mm6, %%mm4        \n\t"
-                "psrad      $1, %%mm2        \n\t"
-                "psrad      $1, %%mm6        \n\t"
-                "paddd   %%mm0, %%mm2        \n\t"
-                "paddd   %%mm4, %%mm6        \n\t"
+                "paddw   %%mm2, %%mm0        \n\t"
+                "paddw   %%mm6, %%mm4        \n\t"
+                "psraw      $1, %%mm2        \n\t"
+                "psraw      $1, %%mm6        \n\t"
+                "paddw   %%mm0, %%mm2        \n\t"
+                "paddw   %%mm4, %%mm6        \n\t"
                 "movq    %%mm2, (%2)         \n\t"
                 "movq    %%mm6, 8(%2)        \n\t"
                 :: "r"(&src[i]), "r"(&b[i]), "r"(&temp[i])
@@ -347,11 +347,11 @@ void ff_snow_horizontal_compose97i_mmx(DWTELEM *b, int width){
     {
         snow_interleave_line_header(&i, width, b, temp);
 
-        for (; (i & 0xE) != 0xE; i-=2){
+        for (; (i & 0x1E) != 0x1E; i-=2){
             b[i+1] = temp[i>>1];
             b[i] = b[i>>1];
         }
-        for (i-=14; i>=0; i-=16){
+        for (i-=30; i>=0; i-=32){
             asm volatile(
                 "movq        (%1), %%mm0       \n\t"
                 "movq       8(%1), %%mm2       \n\t"
@@ -361,18 +361,18 @@ void ff_snow_horizontal_compose97i_mmx(DWTELEM *b, int width){
                 "movq       8(%1), %%mm3       \n\t"
                 "movq      16(%1), %%mm5       \n\t"
                 "movq      24(%1), %%mm7       \n\t"
-                "punpckldq   (%2), %%mm0       \n\t"
-                "punpckldq  8(%2), %%mm2       \n\t"
-                "punpckldq 16(%2), %%mm4       \n\t"
-                "punpckldq 24(%2), %%mm6       \n\t"
+                "punpcklwd   (%2), %%mm0       \n\t"
+                "punpcklwd  8(%2), %%mm2       \n\t"
+                "punpcklwd 16(%2), %%mm4       \n\t"
+                "punpcklwd 24(%2), %%mm6       \n\t"
                 "movq       %%mm0, (%0)        \n\t"
                 "movq       %%mm2, 16(%0)      \n\t"
                 "movq       %%mm4, 32(%0)      \n\t"
                 "movq       %%mm6, 48(%0)      \n\t"
-                "punpckhdq   (%2), %%mm1       \n\t"
-                "punpckhdq  8(%2), %%mm3       \n\t"
-                "punpckhdq 16(%2), %%mm5       \n\t"
-                "punpckhdq 24(%2), %%mm7       \n\t"
+                "punpckhwd   (%2), %%mm1       \n\t"
+                "punpckhwd  8(%2), %%mm3       \n\t"
+                "punpckhwd 16(%2), %%mm5       \n\t"
+                "punpckhwd 24(%2), %%mm7       \n\t"
                 "movq       %%mm1, 8(%0)       \n\t"
                 "movq       %%mm3, 24(%0)      \n\t"
                 "movq       %%mm5, 40(%0)      \n\t"
@@ -385,40 +385,40 @@ void ff_snow_horizontal_compose97i_mmx(DWTELEM *b, int width){
 }
 
 #define snow_vertical_compose_sse2_load_add(op,r,t0,t1,t2,t3)\
-        ""op" (%%"r",%%"REG_d",4), %%"t0"      \n\t"\
-        ""op" 16(%%"r",%%"REG_d",4), %%"t1"    \n\t"\
-        ""op" 32(%%"r",%%"REG_d",4), %%"t2"    \n\t"\
-        ""op" 48(%%"r",%%"REG_d",4), %%"t3"    \n\t"
+        ""op" (%%"r",%%"REG_d",2), %%"t0"      \n\t"\
+        ""op" 16(%%"r",%%"REG_d",2), %%"t1"    \n\t"\
+        ""op" 32(%%"r",%%"REG_d",2), %%"t2"    \n\t"\
+        ""op" 48(%%"r",%%"REG_d",2), %%"t3"    \n\t"
 
 #define snow_vertical_compose_sse2_load(r,t0,t1,t2,t3)\
         snow_vertical_compose_sse2_load_add("movdqa",r,t0,t1,t2,t3)
 
 #define snow_vertical_compose_sse2_add(r,t0,t1,t2,t3)\
-        snow_vertical_compose_sse2_load_add("paddd",r,t0,t1,t2,t3)
+        snow_vertical_compose_sse2_load_add("paddw",r,t0,t1,t2,t3)
 
 #define snow_vertical_compose_sse2_sub(s0,s1,s2,s3,t0,t1,t2,t3)\
-        "psubd %%"s0", %%"t0" \n\t"\
-        "psubd %%"s1", %%"t1" \n\t"\
-        "psubd %%"s2", %%"t2" \n\t"\
-        "psubd %%"s3", %%"t3" \n\t"
+        "psubw %%"s0", %%"t0" \n\t"\
+        "psubw %%"s1", %%"t1" \n\t"\
+        "psubw %%"s2", %%"t2" \n\t"\
+        "psubw %%"s3", %%"t3" \n\t"
 
 #define snow_vertical_compose_sse2_store(w,s0,s1,s2,s3)\
-        "movdqa %%"s0", (%%"w",%%"REG_d",4)      \n\t"\
-        "movdqa %%"s1", 16(%%"w",%%"REG_d",4)    \n\t"\
-        "movdqa %%"s2", 32(%%"w",%%"REG_d",4)    \n\t"\
-        "movdqa %%"s3", 48(%%"w",%%"REG_d",4)    \n\t"
+        "movdqa %%"s0", (%%"w",%%"REG_d",2)      \n\t"\
+        "movdqa %%"s1", 16(%%"w",%%"REG_d",2)    \n\t"\
+        "movdqa %%"s2", 32(%%"w",%%"REG_d",2)    \n\t"\
+        "movdqa %%"s3", 48(%%"w",%%"REG_d",2)    \n\t"
 
 #define snow_vertical_compose_sse2_sra(n,t0,t1,t2,t3)\
-        "psrad $"n", %%"t0" \n\t"\
-        "psrad $"n", %%"t1" \n\t"\
-        "psrad $"n", %%"t2" \n\t"\
-        "psrad $"n", %%"t3" \n\t"
+        "psraw $"n", %%"t0" \n\t"\
+        "psraw $"n", %%"t1" \n\t"\
+        "psraw $"n", %%"t2" \n\t"\
+        "psraw $"n", %%"t3" \n\t"
 
 #define snow_vertical_compose_sse2_r2r_add(s0,s1,s2,s3,t0,t1,t2,t3)\
-        "paddd %%"s0", %%"t0" \n\t"\
-        "paddd %%"s1", %%"t1" \n\t"\
-        "paddd %%"s2", %%"t2" \n\t"\
-        "paddd %%"s3", %%"t3" \n\t"
+        "paddw %%"s0", %%"t0" \n\t"\
+        "paddw %%"s1", %%"t1" \n\t"\
+        "paddw %%"s2", %%"t2" \n\t"\
+        "paddw %%"s3", %%"t3" \n\t"
 
 #define snow_vertical_compose_sse2_move(s0,s1,s2,s3,t0,t1,t2,t3)\
         "movdqa %%"s0", %%"t0" \n\t"\
@@ -426,10 +426,10 @@ void ff_snow_horizontal_compose97i_mmx(DWTELEM *b, int width){
         "movdqa %%"s2", %%"t2" \n\t"\
         "movdqa %%"s3", %%"t3" \n\t"
 
-void ff_snow_vertical_compose97i_sse2(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width){
+void ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
     long i = width;
 
-    while(i & 0xF)
+    while(i & 0x1F)
     {
         i--;
         b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
@@ -452,8 +452,8 @@ void ff_snow_vertical_compose97i_sse2(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWT
         snow_vertical_compose_sse2_r2r_add("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")
 
         "pcmpeqd %%xmm1, %%xmm1                      \n\t"
-        "pslld $31, %%xmm1                           \n\t"
-        "psrld $29, %%xmm1                           \n\t"
+        "psllw $15, %%xmm1                           \n\t"
+        "psrlw $13, %%xmm1                           \n\t"
         "mov %5, %%"REG_a"                           \n\t"
 
         snow_vertical_compose_sse2_r2r_add("xmm1","xmm1","xmm1","xmm1","xmm0","xmm2","xmm4","xmm6")
@@ -472,8 +472,8 @@ void ff_snow_vertical_compose97i_sse2(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWT
         snow_vertical_compose_sse2_add(REG_c,"xmm0","xmm2","xmm4","xmm6")
 
         "pcmpeqd %%xmm1, %%xmm1                      \n\t"
-        "pslld $31, %%xmm1                           \n\t"
-        "psrld $30, %%xmm1                           \n\t"
+        "psllw $15, %%xmm1                           \n\t"
+        "psrlw $14, %%xmm1                           \n\t"
         "mov %1, %%"REG_S"                           \n\t"
 
         snow_vertical_compose_sse2_r2r_add("xmm1","xmm1","xmm1","xmm1","xmm0","xmm2","xmm4","xmm6")
@@ -488,7 +488,7 @@ void ff_snow_vertical_compose97i_sse2(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWT
         snow_vertical_compose_sse2_store(REG_a,"xmm0","xmm2","xmm4","xmm6")
 
         "2:                                          \n\t"
-        "sub $16, %%"REG_d"                          \n\t"
+        "sub $32, %%"REG_d"                          \n\t"
         "jge 1b                                      \n\t"
         :"+d"(i)
         :
@@ -497,25 +497,25 @@ void ff_snow_vertical_compose97i_sse2(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWT
 }
 
 #define snow_vertical_compose_mmx_load_add(op,r,t0,t1,t2,t3)\
-        ""op" (%%"r",%%"REG_d",4), %%"t0"   \n\t"\
-        ""op" 8(%%"r",%%"REG_d",4), %%"t1"  \n\t"\
-        ""op" 16(%%"r",%%"REG_d",4), %%"t2" \n\t"\
-        ""op" 24(%%"r",%%"REG_d",4), %%"t3" \n\t"
+        ""op" (%%"r",%%"REG_d",2), %%"t0"   \n\t"\
+        ""op" 8(%%"r",%%"REG_d",2), %%"t1"  \n\t"\
+        ""op" 16(%%"r",%%"REG_d",2), %%"t2" \n\t"\
+        ""op" 24(%%"r",%%"REG_d",2), %%"t3" \n\t"
 
 #define snow_vertical_compose_mmx_load(r,t0,t1,t2,t3)\
         snow_vertical_compose_mmx_load_add("movq",r,t0,t1,t2,t3)
 
 #define snow_vertical_compose_mmx_add(r,t0,t1,t2,t3)\
-        snow_vertical_compose_mmx_load_add("paddd",r,t0,t1,t2,t3)
+        snow_vertical_compose_mmx_load_add("paddw",r,t0,t1,t2,t3)
 
 #define snow_vertical_compose_mmx_sub(s0,s1,s2,s3,t0,t1,t2,t3)\
         snow_vertical_compose_sse2_sub(s0,s1,s2,s3,t0,t1,t2,t3)
 
 #define snow_vertical_compose_mmx_store(w,s0,s1,s2,s3)\
-        "movq %%"s0", (%%"w",%%"REG_d",4)   \n\t"\
-        "movq %%"s1", 8(%%"w",%%"REG_d",4)  \n\t"\
-        "movq %%"s2", 16(%%"w",%%"REG_d",4) \n\t"\
-        "movq %%"s3", 24(%%"w",%%"REG_d",4) \n\t"
+        "movq %%"s0", (%%"w",%%"REG_d",2)   \n\t"\
+        "movq %%"s1", 8(%%"w",%%"REG_d",2)  \n\t"\
+        "movq %%"s2", 16(%%"w",%%"REG_d",2) \n\t"\
+        "movq %%"s3", 24(%%"w",%%"REG_d",2) \n\t"
 
 #define snow_vertical_compose_mmx_sra(n,t0,t1,t2,t3)\
         snow_vertical_compose_sse2_sra(n,t0,t1,t2,t3)
@@ -529,9 +529,9 @@ void ff_snow_vertical_compose97i_sse2(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWT
         "movq %%"s2", %%"t2" \n\t"\
         "movq %%"s3", %%"t3" \n\t"
 
-void ff_snow_vertical_compose97i_mmx(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width){
+void ff_snow_vertical_compose97i_mmx(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
     long i = width;
-    while(i & 0x7)
+    while(i & 15)
     {
         i--;
         b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
@@ -553,9 +553,9 @@ void ff_snow_vertical_compose97i_mmx(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTE
         snow_vertical_compose_mmx_r2r_add("mm0","mm2","mm4","mm6","mm0","mm2","mm4","mm6")
         snow_vertical_compose_mmx_r2r_add("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")
 
-        "pcmpeqd %%mm1, %%mm1                        \n\t"
-        "pslld $31, %%mm1                            \n\t"
-        "psrld $29, %%mm1                            \n\t"
+        "pcmpeqw %%mm1, %%mm1                        \n\t"
+        "psllw $15, %%mm1                            \n\t"
+        "psrlw $13, %%mm1                            \n\t"
         "mov %5, %%"REG_a"                           \n\t"
 
         snow_vertical_compose_mmx_r2r_add("mm1","mm1","mm1","mm1","mm0","mm2","mm4","mm6")
@@ -573,9 +573,9 @@ void ff_snow_vertical_compose97i_mmx(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTE
         snow_vertical_compose_mmx_sra("2","mm0","mm2","mm4","mm6")
         snow_vertical_compose_mmx_add(REG_c,"mm0","mm2","mm4","mm6")
 
-        "pcmpeqd %%mm1, %%mm1                        \n\t"
-        "pslld $31, %%mm1                            \n\t"
-        "psrld $30, %%mm1                            \n\t"
+        "pcmpeqw %%mm1, %%mm1                        \n\t"
+        "psllw $15, %%mm1                            \n\t"
+        "psrlw $14, %%mm1                            \n\t"
         "mov %1, %%"REG_S"                           \n\t"
 
         snow_vertical_compose_mmx_r2r_add("mm1","mm1","mm1","mm1","mm0","mm2","mm4","mm6")
@@ -590,7 +590,7 @@ void ff_snow_vertical_compose97i_mmx(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTE
         snow_vertical_compose_mmx_store(REG_a,"mm0","mm2","mm4","mm6")
 
         "2:                                          \n\t"
-        "sub $8, %%"REG_d"                           \n\t"
+        "sub $16, %%"REG_d"                          \n\t"
         "jge 1b                                      \n\t"
         :"+d"(i)
         :
@@ -599,7 +599,7 @@ void ff_snow_vertical_compose97i_mmx(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTE
 }
 
 #define snow_inner_add_yblock_sse2_header \
-    DWTELEM * * dst_array = sb->line + src_y;\
+    IDWTELEM * * dst_array = sb->line + src_y;\
     long tmp;\
     asm volatile(\
              "mov  %7, %%"REG_c"             \n\t"\
@@ -607,8 +607,8 @@ void ff_snow_vertical_compose97i_mmx(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTE
              "mov  %4, %%"REG_S"             \n\t"\
              "pxor %%xmm7, %%xmm7            \n\t" /* 0 */\
              "pcmpeqd %%xmm3, %%xmm3         \n\t"\
-             "pslld $31, %%xmm3              \n\t"\
-             "psrld $24, %%xmm3              \n\t" /* FRAC_BITS >> 1 */\
+             "psllw $15, %%xmm3              \n\t"\
+             "psrlw $12, %%xmm3              \n\t" /* FRAC_BITS >> 1 */\
              "1:                             \n\t"\
              "mov %1, %%"REG_D"              \n\t"\
              "mov (%%"REG_D"), %%"REG_D"     \n\t"\
@@ -662,7 +662,7 @@ void ff_snow_vertical_compose97i_mmx(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTE
              "jnz 1b                         \n\t"\
              :"+m"(dst8),"+m"(dst_array),"=&r"(tmp)\
              :\
-             "rm"((long)(src_x<<2)),"m"(obmc),"a"(block),"m"((long)b_h),"m"((long)src_stride):\
+             "rm"((long)(src_x<<1)),"m"(obmc),"a"(block),"m"((long)b_h),"m"((long)src_stride):\
              "%"REG_c"","%"REG_S"","%"REG_D"","%"REG_d"");
 
 #define snow_inner_add_yblock_sse2_end_8\
@@ -736,36 +736,23 @@ snow_inner_add_yblock_sse2_accum_16("1", "512")
 snow_inner_add_yblock_sse2_accum_16("0", "528")
 
              "mov %0, %%"REG_d"              \n\t"
-             "movdqa %%xmm1, %%xmm0          \n\t"
-             "movdqa %%xmm5, %%xmm4          \n\t"
-             "punpcklwd %%xmm7, %%xmm0       \n\t"
-             "paddd (%%"REG_D"), %%xmm0      \n\t"
-             "punpckhwd %%xmm7, %%xmm1       \n\t"
-             "paddd 16(%%"REG_D"), %%xmm1    \n\t"
-             "punpcklwd %%xmm7, %%xmm4       \n\t"
-             "paddd 32(%%"REG_D"), %%xmm4    \n\t"
-             "punpckhwd %%xmm7, %%xmm5       \n\t"
-             "paddd 48(%%"REG_D"), %%xmm5    \n\t"
-             "paddd %%xmm3, %%xmm0           \n\t"
-             "paddd %%xmm3, %%xmm1           \n\t"
-             "paddd %%xmm3, %%xmm4           \n\t"
-             "paddd %%xmm3, %%xmm5           \n\t"
-             "psrad $8, %%xmm0               \n\t" /* FRAC_BITS. */
-             "psrad $8, %%xmm1               \n\t" /* FRAC_BITS. */
-             "psrad $8, %%xmm4               \n\t" /* FRAC_BITS. */
-             "psrad $8, %%xmm5               \n\t" /* FRAC_BITS. */
-
-             "packssdw %%xmm1, %%xmm0        \n\t"
-             "packssdw %%xmm5, %%xmm4        \n\t"
-             "packuswb %%xmm4, %%xmm0        \n\t"
-
-             "movdqu %%xmm0, (%%"REG_d")       \n\t"
+             "psrlw $4, %%xmm1               \n\t"
+             "psrlw $4, %%xmm5               \n\t"
+             "paddw   (%%"REG_D"), %%xmm1    \n\t"
+             "paddw 16(%%"REG_D"), %%xmm5    \n\t"
+             "paddw %%xmm3, %%xmm1           \n\t"
+             "paddw %%xmm3, %%xmm5           \n\t"
+             "psraw $4, %%xmm1               \n\t" /* FRAC_BITS. */
+             "psraw $4, %%xmm5               \n\t" /* FRAC_BITS. */
+             "packuswb %%xmm5, %%xmm1        \n\t"
+
+             "movdqu %%xmm1, (%%"REG_d")       \n\t"
 
 snow_inner_add_yblock_sse2_end_16
 }
 
 #define snow_inner_add_yblock_mmx_header \
-    DWTELEM * * dst_array = sb->line + src_y;\
+    IDWTELEM * * dst_array = sb->line + src_y;\
     long tmp;\
     asm volatile(\
              "mov  %7, %%"REG_c"             \n\t"\
@@ -773,8 +760,8 @@ snow_inner_add_yblock_sse2_end_16
              "mov  %4, %%"REG_S"             \n\t"\
              "pxor %%mm7, %%mm7              \n\t" /* 0 */\
              "pcmpeqd %%mm3, %%mm3           \n\t"\
-             "pslld $31, %%mm3               \n\t"\
-             "psrld $24, %%mm3               \n\t" /* FRAC_BITS >> 1 */\
+             "psllw $15, %%mm3               \n\t"\
+             "psrlw $12, %%mm3               \n\t" /* FRAC_BITS >> 1 */\
              "1:                             \n\t"\
              "mov %1, %%"REG_D"              \n\t"\
              "mov (%%"REG_D"), %%"REG_D"     \n\t"\
@@ -800,29 +787,16 @@ snow_inner_add_yblock_sse2_end_16
 
 #define snow_inner_add_yblock_mmx_mix(read_offset, write_offset)\
              "mov %0, %%"REG_d"              \n\t"\
-             "movq %%mm1, %%mm0              \n\t"\
-             "movq %%mm5, %%mm4              \n\t"\
-             "punpcklwd %%mm7, %%mm0         \n\t"\
-             "paddd "read_offset"(%%"REG_D"), %%mm0 \n\t"\
-             "punpckhwd %%mm7, %%mm1         \n\t"\
-             "paddd "read_offset"+8(%%"REG_D"), %%mm1 \n\t"\
-             "punpcklwd %%mm7, %%mm4         \n\t"\
-             "paddd "read_offset"+16(%%"REG_D"), %%mm4 \n\t"\
-             "punpckhwd %%mm7, %%mm5         \n\t"\
-             "paddd "read_offset"+24(%%"REG_D"), %%mm5 \n\t"\
-             "paddd %%mm3, %%mm0             \n\t"\
-             "paddd %%mm3, %%mm1             \n\t"\
-             "paddd %%mm3, %%mm4             \n\t"\
-             "paddd %%mm3, %%mm5             \n\t"\
-             "psrad $8, %%mm0                \n\t"\
-             "psrad $8, %%mm1                \n\t"\
-             "psrad $8, %%mm4                \n\t"\
-             "psrad $8, %%mm5                \n\t"\
-\
-             "packssdw %%mm1, %%mm0          \n\t"\
-             "packssdw %%mm5, %%mm4          \n\t"\
-             "packuswb %%mm4, %%mm0          \n\t"\
-             "movq %%mm0, "write_offset"(%%"REG_d") \n\t"
+             "psrlw $4, %%mm1                \n\t"\
+             "psrlw $4, %%mm5                \n\t"\
+             "paddw "read_offset"(%%"REG_D"), %%mm1 \n\t"\
+             "paddw "read_offset"+8(%%"REG_D"), %%mm5 \n\t"\
+             "paddw %%mm3, %%mm1             \n\t"\
+             "paddw %%mm3, %%mm5             \n\t"\
+             "psraw $4, %%mm1                \n\t"\
+             "psraw $4, %%mm5                \n\t"\
+             "packuswb %%mm5, %%mm1          \n\t"\
+             "movq %%mm1, "write_offset"(%%"REG_d") \n\t"
 
 #define snow_inner_add_yblock_mmx_end(s_step)\
              "add $"s_step", %%"REG_S"             \n\t"\
@@ -836,7 +810,7 @@ snow_inner_add_yblock_sse2_end_16
              "jnz 1b                         \n\t"\
              :"+m"(dst8),"+m"(dst_array),"=&r"(tmp)\
              :\
-             "rm"((long)(src_x<<2)),"m"(obmc),"a"(block),"m"((long)b_h),"m"((long)src_stride):\
+             "rm"((long)(src_x<<1)),"m"(obmc),"a"(block),"m"((long)b_h),"m"((long)src_stride):\
              "%"REG_c"","%"REG_S"","%"REG_D"","%"REG_d"");
 
 static void inner_add_yblock_bw_8_obmc_16_mmx(const uint8_t *obmc, const long obmc_stride, uint8_t * * block, int b_w, long b_h,
@@ -863,7 +837,7 @@ snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "8", "8")
 snow_inner_add_yblock_mmx_accum("2", "24", "8")
 snow_inner_add_yblock_mmx_accum("1", "520", "8")
 snow_inner_add_yblock_mmx_accum("0", "536", "8")
-snow_inner_add_yblock_mmx_mix("32", "8")
+snow_inner_add_yblock_mmx_mix("16", "8")
 snow_inner_add_yblock_mmx_end("32")
 }