Redo the forward 4x4 dct

The new fdct lowers the round trip sum squared error for a
4x4 block ~0.12. or ~0.008/pixel. For reference, the old
matrix multiply version has average round trip error 1.46
for a 4x4 block.

Thanks to "derf" for his suggestions and references.

Change-Id: I5559d1e81d333b319404ab16b336b739f87afc79

14 files changed, 118 insertions, 562 deletions

diff --git a/vp8/encoder/block.h b/vp8/encoder/block.h
index c1fcfe29a..b55bc51cb 100644
--- a/vp8/encoder/block.h
+++ b/vp8/encoder/block.h
@@ -100,14 +100,9 @@ typedef struct
 
     void (*vp8_short_fdct4x4)(short *input, short *output, int pitch);
     void (*vp8_short_fdct8x4)(short *input, short *output, int pitch);
-    void (*short_fdct4x4rd)(short *input, short *output, int pitch);
-    void (*short_fdct8x4rd)(short *input, short *output, int pitch);
     void (*short_walsh4x4)(short *input, short *output, int pitch);
-
     void (*quantize_b)(BLOCK *b, BLOCKD *d);
 
-
-
 } MACROBLOCK;
 
 
diff --git a/vp8/encoder/dct.c b/vp8/encoder/dct.c
index 3075e5853..58e36109c 100644
--- a/vp8/encoder/dct.c
+++ b/vp8/encoder/dct.c
@@ -11,163 +11,54 @@
 
 #include <math.h>
 
-
-static const short dct_matrix2[4][4] =
-{
-    { 23170,  30274,  23170, 12540 },
-    { 23170,  12540, -23170, -30274 },
-    { 23170, -12540, -23170, 30274 },
-    { 23170, -30274,  23170, -12540 }
-};
-
-static const short dct_matrix1[4][4] =
-{
-    { 23170,  23170,  23170,  23170 },
-    { 30274,  12540, -12540, -30274 },
-    { 23170, -23170, -23170,  23170 },
-    { 12540, -30274,  30274, -12540 }
-};
-
-
-#define _1STSTAGESHIFT           14
-#define _1STSTAGEROUNDING        (1<<( _1STSTAGESHIFT-1))
-#define _2NDSTAGESHIFT           16
-#define _2NDSTAGEROUNDING        (1<<( _2NDSTAGESHIFT-1))
-
-// using matrix multiply
 void vp8_short_fdct4x4_c(short *input, short *output, int pitch)
 {
-    int i, j, k;
-    short temp[4][4];
-    int sumtemp;
-    pitch >>= 1;
-
-    for (i = 0; i < 4; i++)
-    {
-        for (j = 0; j < 4; j++)
-        {
-            sumtemp = 0;
-
-            for (k = 0; k < 4; k++)
-            {
-                sumtemp += input[i*pitch+k] * dct_matrix2[k][j];
-
-            }
-
-            temp[i][j] = (short)((sumtemp + _1STSTAGEROUNDING) >> _1STSTAGESHIFT);
-        }
-    }
-
-
-    for (i = 0; i < 4; i++)
-    {
-        for (j = 0; j < 4; j++)
-        {
-            sumtemp = 0;
-
-            for (k = 0; k < 4; k++)
-            {
-                sumtemp += dct_matrix1[i][ k] * temp[k][ j];
-            }
-
-            output[i*4+j] = (short)((sumtemp + _2NDSTAGEROUNDING) >> _2NDSTAGESHIFT);
-        }
-    }
-
-}
-
-
-void vp8_short_fdct8x4_c(short *input, short *output, int pitch)
-{
-    vp8_short_fdct4x4_c(input,   output,    pitch);
-    vp8_short_fdct4x4_c(input + 4, output + 16, pitch);
-}
-
-
-static const signed short x_c1 = 60547;
-static const signed short x_c2 = 46341;
-static const signed short x_c3 = 25080;
-
-void vp8_fast_fdct4x4_c(short *input, short *output, int pitch)
-{
     int i;
     int a1, b1, c1, d1;
-    int a2, b2, c2, d2;
     short *ip = input;
-
     short *op = output;
-    int temp1, temp2;
 
     for (i = 0; i < 4; i++)
     {
-        a1 = (ip[0] + ip[3]) * 2;
-        b1 = (ip[1] + ip[2]) * 2;
-        c1 = (ip[1] - ip[2]) * 2;
-        d1 = (ip[0] - ip[3]) * 2;
-
-        temp1 = a1 + b1;
-        temp2 = a1 - b1;
-
-        op[0] = ((temp1 * x_c2) >> 16) + temp1;
-        op[2] = ((temp2 * x_c2) >> 16) + temp2;
-
-        temp1 = (c1 * x_c3) >> 16;
-        temp2 = ((d1 * x_c1) >> 16) + d1;
+        a1 = ((ip[0] + ip[3])<<3);
+        b1 = ((ip[1] + ip[2])<<3);
+        c1 = ((ip[1] - ip[2])<<3);
+        d1 = ((ip[0] - ip[3])<<3);
 
-        op[1] = temp1 + temp2;
-
-        temp1 = (d1 * x_c3) >> 16;
-        temp2 = ((c1 * x_c1) >> 16) + c1;
+        op[0] = a1 + b1;
+        op[2] = a1 - b1;
 
-        op[3] = temp1 - temp2;
+        op[1] = (c1 * 2217 + d1 * 5352 +  14500)>>12;
+        op[3] = (d1 * 2217 - c1 * 5352 +   7500)>>12;
 
         ip += pitch / 2;
         op += 4;
-    }
 
+    }
     ip = output;
     op = output;
-
     for (i = 0; i < 4; i++)
     {
-
         a1 = ip[0] + ip[12];
         b1 = ip[4] + ip[8];
         c1 = ip[4] - ip[8];
         d1 = ip[0] - ip[12];
 
+        op[0]  = ( a1 + b1 + 7)>>4;
+        op[8]  = ( a1 - b1 + 7)>>4;
 
-        temp1 = a1 + b1;
-        temp2 = a1 - b1;
-
-        a2 = ((temp1 * x_c2) >> 16) + temp1;
-        c2 = ((temp2 * x_c2) >> 16) + temp2;
-
-        temp1 = (c1 * x_c3) >> 16;
-        temp2 = ((d1 * x_c1) >> 16) + d1;
-
-        b2 = temp1 + temp2;
-
-        temp1 = (d1 * x_c3) >> 16;
-        temp2 = ((c1 * x_c1) >> 16) + c1;
-
-        d2 = temp1 - temp2;
-
-
-        op[0]   = (a2 + 1) >> 1;
-        op[4]   = (b2 + 1) >> 1;
-        op[8]   = (c2 + 1) >> 1;
-        op[12]  = (d2 + 1) >> 1;
+        op[4]  =((c1 * 2217 + d1 * 5352 +  12000)>>16) + (d1!=0);
+        op[12] = (d1 * 2217 - c1 * 5352 +  51000)>>16;
 
         ip++;
         op++;
     }
 }
 
-void vp8_fast_fdct8x4_c(short *input, short *output, int pitch)
+void vp8_short_fdct8x4_c(short *input, short *output, int pitch)
 {
-    vp8_fast_fdct4x4_c(input,   output,    pitch);
-    vp8_fast_fdct4x4_c(input + 4, output + 16, pitch);
+    vp8_short_fdct4x4_c(input,   output,    pitch);
+    vp8_short_fdct4x4_c(input + 4, output + 16, pitch);
 }
 
 void vp8_short_walsh4x4_c(short *input, short *output, int pitch)
diff --git a/vp8/encoder/dct.h b/vp8/encoder/dct.h
index f79dba4f2..0ab40b310 100644
--- a/vp8/encoder/dct.h
+++ b/vp8/encoder/dct.h
@@ -32,16 +32,6 @@ extern prototype_fdct(vp8_fdct_short4x4);
 #endif
 extern prototype_fdct(vp8_fdct_short8x4);
 
-#ifndef vp8_fdct_fast4x4
-#define vp8_fdct_fast4x4  vp8_fast_fdct4x4_c
-#endif
-extern prototype_fdct(vp8_fdct_fast4x4);
-
-#ifndef vp8_fdct_fast8x4
-#define vp8_fdct_fast8x4  vp8_fast_fdct8x4_c
-#endif
-extern prototype_fdct(vp8_fdct_fast8x4);
-
 #ifndef vp8_fdct_walsh_short4x4
 #define vp8_fdct_walsh_short4x4  vp8_short_walsh4x4_c
 #endif
diff --git a/vp8/encoder/encodeintra.c b/vp8/encoder/encodeintra.c
index 0e160930d..870cb5815 100644
--- a/vp8/encoder/encodeintra.c
+++ b/vp8/encoder/encodeintra.c
@@ -66,7 +66,7 @@ void vp8_encode_intra4x4block_rd(const VP8_ENCODER_RTCD *rtcd, MACROBLOCK *x, BL
 
     ENCODEMB_INVOKE(&rtcd->encodemb, subb)(be, b, 16);
 
-    x->short_fdct4x4rd(be->src_diff, be->coeff, 32);
+    x->vp8_short_fdct4x4(be->src_diff, be->coeff, 32);
 
     x->quantize_b(be, b);
 
diff --git a/vp8/encoder/encodemb.c b/vp8/encoder/encodemb.c
index 824850c41..8bc01df5b 100644
--- a/vp8/encoder/encodemb.c
+++ b/vp8/encoder/encodemb.c
@@ -130,7 +130,8 @@ void vp8_transform_mbuvrd(MACROBLOCK *x)
 
     for (i = 16; i < 24; i += 2)
     {
-        x->short_fdct8x4rd(&x->block[i].src_diff[0], &x->block[i].coeff[0], 16);
+        x->vp8_short_fdct8x4(&x->block[i].src_diff[0],
+            &x->block[i].coeff[0], 16);
     }
 }
 
@@ -140,14 +141,16 @@ void vp8_transform_intra_mby(MACROBLOCK *x)
 
     for (i = 0; i < 16; i += 2)
     {
-        x->vp8_short_fdct8x4(&x->block[i].src_diff[0], &x->block[i].coeff[0], 32);
+        x->vp8_short_fdct8x4(&x->block[i].src_diff[0],
+            &x->block[i].coeff[0], 32);
     }
 
     // build dc block from 16 y dc values
     vp8_build_dcblock(x);
 
     // do 2nd order transform on the dc block
-    x->short_walsh4x4(&x->block[24].src_diff[0], &x->block[24].coeff[0], 8);
+    x->short_walsh4x4(&x->block[24].src_diff[0],
+        &x->block[24].coeff[0], 8);
 
 }
 
@@ -157,14 +160,16 @@ void vp8_transform_intra_mbyrd(MACROBLOCK *x)
 
     for (i = 0; i < 16; i += 2)
     {
-        x->short_fdct8x4rd(&x->block[i].src_diff[0], &x->block[i].coeff[0], 32);
+        x->vp8_short_fdct8x4(&x->block[i].src_diff[0],
+            &x->block[i].coeff[0], 32);
     }
 
     // build dc block from 16 y dc values
     vp8_build_dcblock(x);
 
     // do 2nd order transform on the dc block
-    x->short_walsh4x4(&x->block[24].src_diff[0], &x->block[24].coeff[0], 8);
+    x->short_walsh4x4(&x->block[24].src_diff[0],
+        &x->block[24].coeff[0], 8);
 }
 
 void vp8_transform_mb(MACROBLOCK *x)
@@ -173,7 +178,8 @@ void vp8_transform_mb(MACROBLOCK *x)
 
     for (i = 0; i < 16; i += 2)
     {
-        x->vp8_short_fdct8x4(&x->block[i].src_diff[0], &x->block[i].coeff[0], 32);
+        x->vp8_short_fdct8x4(&x->block[i].src_diff[0],
+            &x->block[i].coeff[0], 32);
     }
 
     // build dc block from 16 y dc values
@@ -182,12 +188,14 @@ void vp8_transform_mb(MACROBLOCK *x)
 
     for (i = 16; i < 24; i += 2)
     {
-        x->vp8_short_fdct8x4(&x->block[i].src_diff[0], &x->block[i].coeff[0], 16);
+        x->vp8_short_fdct8x4(&x->block[i].src_diff[0],
+            &x->block[i].coeff[0], 16);
     }
 
     // do 2nd order transform on the dc block
     if (x->e_mbd.mbmi.mode != SPLITMV)
-        x->short_walsh4x4(&x->block[24].src_diff[0], &x->block[24].coeff[0], 8);
+        x->short_walsh4x4(&x->block[24].src_diff[0],
+        &x->block[24].coeff[0], 8);
 
 }
 
@@ -197,14 +205,16 @@ void vp8_transform_mby(MACROBLOCK *x)
 
     for (i = 0; i < 16; i += 2)
     {
-        x->vp8_short_fdct8x4(&x->block[i].src_diff[0], &x->block[i].coeff[0], 32);
+        x->vp8_short_fdct8x4(&x->block[i].src_diff[0],
+            &x->block[i].coeff[0], 32);
     }
 
     // build dc block from 16 y dc values
     if (x->e_mbd.mbmi.mode != SPLITMV)
     {
         vp8_build_dcblock(x);
-        x->short_walsh4x4(&x->block[24].src_diff[0], &x->block[24].coeff[0], 8);
+        x->short_walsh4x4(&x->block[24].src_diff[0],
+            &x->block[24].coeff[0], 8);
     }
 }
 
@@ -214,7 +224,8 @@ void vp8_transform_mbrd(MACROBLOCK *x)
 
     for (i = 0; i < 16; i += 2)
     {
-        x->short_fdct8x4rd(&x->block[i].src_diff[0], &x->block[i].coeff[0], 32);
+        x->vp8_short_fdct8x4(&x->block[i].src_diff[0],
+            &x->block[i].coeff[0], 32);
     }
 
     // build dc block from 16 y dc values
@@ -223,12 +234,14 @@ void vp8_transform_mbrd(MACROBLOCK *x)
 
     for (i = 16; i < 24; i += 2)
     {
-        x->short_fdct8x4rd(&x->block[i].src_diff[0], &x->block[i].coeff[0], 16);
+        x->vp8_short_fdct8x4(&x->block[i].src_diff[0],
+            &x->block[i].coeff[0], 16);
     }
 
     // do 2nd order transform on the dc block
     if (x->e_mbd.mbmi.mode != SPLITMV)
-        x->short_walsh4x4(&x->block[24].src_diff[0], &x->block[24].coeff[0], 8);
+        x->short_walsh4x4(&x->block[24].src_diff[0],
+            &x->block[24].coeff[0], 8);
 }
 
 void vp8_stuff_inter16x16(MACROBLOCK *x)
diff --git a/vp8/encoder/ethreading.c b/vp8/encoder/ethreading.c
index a205667dc..dd98a09d1 100644
--- a/vp8/encoder/ethreading.c
+++ b/vp8/encoder/ethreading.c
@@ -257,9 +257,6 @@ static void setup_mbby_copy(MACROBLOCK *mbdst, MACROBLOCK *mbsrc)
 
     z->vp8_short_fdct4x4     = x->vp8_short_fdct4x4;
     z->vp8_short_fdct8x4     = x->vp8_short_fdct8x4;
-    z->short_fdct4x4rd   = x->short_fdct4x4rd;
-    z->short_fdct8x4rd   = x->short_fdct8x4rd;
-    z->short_fdct8x4rd   = x->short_fdct8x4rd;
     z->short_walsh4x4    = x->short_walsh4x4;
     z->quantize_b        = x->quantize_b;
 
diff --git a/vp8/encoder/generic/csystemdependent.c b/vp8/encoder/generic/csystemdependent.c
index e68d65025..dd89f1a82 100644
--- a/vp8/encoder/generic/csystemdependent.c
+++ b/vp8/encoder/generic/csystemdependent.c
@@ -68,8 +68,8 @@ void vp8_cmachine_specific_config(VP8_COMP *cpi)
 
     cpi->rtcd.fdct.short4x4                  = vp8_short_fdct4x4_c;
     cpi->rtcd.fdct.short8x4                  = vp8_short_fdct8x4_c;
-    cpi->rtcd.fdct.fast4x4                   = vp8_fast_fdct4x4_c;
-    cpi->rtcd.fdct.fast8x4                   = vp8_fast_fdct8x4_c;
+    cpi->rtcd.fdct.fast4x4                   = vp8_short_fdct4x4_c;
+    cpi->rtcd.fdct.fast8x4                   = vp8_short_fdct8x4_c;
     cpi->rtcd.fdct.walsh_short4x4            = vp8_short_walsh4x4_c;
 
     cpi->rtcd.encodemb.berr                  = vp8_block_error_c;
diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c
index f3456a733..60d807c03 100644
--- a/vp8/encoder/onyx_if.c
+++ b/vp8/encoder/onyx_if.c
@@ -137,8 +137,6 @@ extern unsigned int inter_b_modes[15];
 
 extern void (*vp8_short_fdct4x4)(short *input, short *output, int pitch);
 extern void (*vp8_short_fdct8x4)(short *input, short *output, int pitch);
-extern void (*vp8_fast_fdct4x4)(short *input, short *output, int pitch);
-extern void (*vp8_fast_fdct8x4)(short *input, short *output, int pitch);
 
 extern const int vp8_bits_per_mb[2][QINDEX_RANGE];
 
@@ -1136,15 +1134,11 @@ void vp8_set_speed_features(VP8_COMP *cpi)
     {
         cpi->mb.vp8_short_fdct8x4 = FDCT_INVOKE(&cpi->rtcd.fdct, short8x4);
         cpi->mb.vp8_short_fdct4x4 = FDCT_INVOKE(&cpi->rtcd.fdct, short4x4);
-        cpi->mb.short_fdct8x4rd = FDCT_INVOKE(&cpi->rtcd.fdct, short8x4);
-        cpi->mb.short_fdct4x4rd = FDCT_INVOKE(&cpi->rtcd.fdct, short4x4);
     }
     else
     {
         cpi->mb.vp8_short_fdct8x4   = FDCT_INVOKE(&cpi->rtcd.fdct, fast8x4);
         cpi->mb.vp8_short_fdct4x4   = FDCT_INVOKE(&cpi->rtcd.fdct, fast4x4);
-        cpi->mb.short_fdct8x4rd = FDCT_INVOKE(&cpi->rtcd.fdct, fast8x4);
-        cpi->mb.short_fdct4x4rd = FDCT_INVOKE(&cpi->rtcd.fdct, fast4x4);
     }
 
     cpi->mb.short_walsh4x4 = FDCT_INVOKE(&cpi->rtcd.fdct, walsh_short4x4);
diff --git a/vp8/encoder/rdopt.c b/vp8/encoder/rdopt.c
index 2d6dee139..70cf122fa 100644
--- a/vp8/encoder/rdopt.c
+++ b/vp8/encoder/rdopt.c
@@ -1028,7 +1028,7 @@ static unsigned int vp8_encode_inter_mb_segment(MACROBLOCK *x, int const *labels
 
             vp8_build_inter_predictors_b(bd, 16, x->e_mbd.subpixel_predict);
             ENCODEMB_INVOKE(rtcd, subb)(be, bd, 16);
-            x->short_fdct4x4rd(be->src_diff, be->coeff, 32);
+            x->vp8_short_fdct4x4(be->src_diff, be->coeff, 32);
 
             // set to 0 no way to account for 2nd order DC so discount
             //be->coeff[0] = 0;
@@ -1056,7 +1056,7 @@ static void macro_block_yrd(MACROBLOCK *mb, int *Rate, int *Distortion, const vp
     // Fdct and building the 2nd order block
     for (beptr = mb->block; beptr < mb->block + 16; beptr += 2)
     {
-        mb->short_fdct8x4rd(beptr->src_diff, beptr->coeff, 32);
+        mb->vp8_short_fdct8x4(beptr->src_diff, beptr->coeff, 32);
         *Y2DCPtr++ = beptr->coeff[0];
         *Y2DCPtr++ = beptr->coeff[16];
     }
diff --git a/vp8/encoder/x86/csystemdependent.c b/vp8/encoder/x86/csystemdependent.c
index 6aeac508f..bf12fee54 100644
--- a/vp8/encoder/x86/csystemdependent.c
+++ b/vp8/encoder/x86/csystemdependent.c
@@ -181,10 +181,17 @@ void vp8_cmachine_specific_config(void)
         // Willamette instruction set available:
         vp8_mbuverror                = vp8_mbuverror_xmm;
         vp8_fast_quantize_b            = vp8_fast_quantize_b_sse;
+#if 0 //new fdct
         vp8_short_fdct4x4             = vp8_short_fdct4x4_mmx;
         vp8_short_fdct8x4             = vp8_short_fdct8x4_mmx;
-        vp8_fast_fdct4x4              = vp8_fast_fdct4x4_mmx;
-        vp8_fast_fdct8x4              = vp8_fast_fdct8x4_wmt;
+        vp8_fast_fdct4x4              = vp8_short_fdct4x4_mmx;
+        vp8_fast_fdct8x4              = vp8_short_fdct8x4_wmt;
+#else
+        vp8_short_fdct4x4             = vp8_short_fdct4x4_c;
+        vp8_short_fdct8x4             = vp8_short_fdct8x4_c;
+        vp8_fast_fdct4x4              = vp8_short_fdct4x4_c;
+        vp8_fast_fdct8x4              = vp8_fast_fdct8x4_c;
+#endif
         vp8_subtract_b                = vp8_subtract_b_mmx;
         vp8_subtract_mbuv             = vp8_subtract_mbuv_mmx;
         vp8_variance4x4              = vp8_variance4x4_mmx;
@@ -218,10 +225,17 @@ void vp8_cmachine_specific_config(void)
         // MMX instruction set available:
         vp8_mbuverror                = vp8_mbuverror_mmx;
         vp8_fast_quantize_b            = vp8_fast_quantize_b_mmx;
+#if 0 // new fdct
         vp8_short_fdct4x4             = vp8_short_fdct4x4_mmx;
         vp8_short_fdct8x4             = vp8_short_fdct8x4_mmx;
-        vp8_fast_fdct4x4              = vp8_fast_fdct4x4_mmx;
-        vp8_fast_fdct8x4              = vp8_fast_fdct8x4_mmx;
+        vp8_fast_fdct4x4              = vp8_short_fdct4x4_mmx;
+        vp8_fast_fdct8x4              = vp8_short_fdct8x4_mmx;
+#else
+        vp8_short_fdct4x4             = vp8_short_fdct4x4_c;
+        vp8_short_fdct8x4             = vp8_short_fdct8x4_c;
+        vp8_fast_fdct4x4              = vp8_short_fdct4x4_c;
+        vp8_fast_fdct8x4              = vp8_fast_fdct8x4_c;
+#endif
         vp8_subtract_b                = vp8_subtract_b_mmx;
         vp8_subtract_mbuv             = vp8_subtract_mbuv_mmx;
         vp8_variance4x4              = vp8_variance4x4_mmx;
@@ -254,10 +268,10 @@ void vp8_cmachine_specific_config(void)
     {
         // Pure C:
         vp8_mbuverror                = vp8_mbuverror_c;
-        vp8_fast_quantize_b            = vp8_fast_quantize_b_c;
+        vp8_fast_quantize_b          = vp8_fast_quantize_b_c;
         vp8_short_fdct4x4             = vp8_short_fdct4x4_c;
         vp8_short_fdct8x4             = vp8_short_fdct8x4_c;
-        vp8_fast_fdct4x4              = vp8_fast_fdct4x4_c;
+        vp8_fast_fdct4x4              = vp8_short_fdct4x4_c;
         vp8_fast_fdct8x4              = vp8_fast_fdct8x4_c;
         vp8_subtract_b                = vp8_subtract_b_c;
         vp8_subtract_mbuv             = vp8_subtract_mbuv_c;
diff --git a/vp8/encoder/x86/dct_mmx.asm b/vp8/encoder/x86/dct_mmx.asm
index 32d6610aa..ff96c49f3 100644
--- a/vp8/encoder/x86/dct_mmx.asm
+++ b/vp8/encoder/x86/dct_mmx.asm
@@ -13,8 +13,7 @@
 
 section .text
     global sym(vp8_short_fdct4x4_mmx)
-    global sym(vp8_fast_fdct4x4_mmx)
-    global sym(vp8_fast_fdct8x4_wmt)
+    global sym(vp8_short_fdct8x4_wmt)
 
 
 %define         DCTCONSTANTSBITS         (16)
@@ -24,10 +23,6 @@ section .text
 %define         x_c3                      (25080)          ; cos(pi*3/8) * (1<<15)
 
 
-%define _1STSTAGESHIFT           14
-%define _2NDSTAGESHIFT           16
-
-; using matrix multiply with source and destbuffer has a pitch
 ;void vp8_short_fdct4x4_mmx(short *input, short *output, int pitch)
 sym(vp8_short_fdct4x4_mmx):
     push        rbp
@@ -37,333 +32,6 @@ sym(vp8_short_fdct4x4_mmx):
     push rsi
     push rdi
     ; end prolog
-
-        mov         rsi,    arg(0) ;input
-        mov         rdi,    arg(1) ;output
-
-        movsxd      rax,    dword ptr arg(2) ;pitch
-        lea         rdx,    [dct_matrix GLOBAL]
-
-        movq        mm0,    [rsi   ]
-        movq        mm1,    [rsi + rax]
-
-        movq        mm2,    [rsi + rax*2]
-        lea         rsi,    [rsi + rax*2]
-
-        movq        mm3,    [rsi + rax]
-
-        ; first column
-        movq        mm4,    mm0
-        movq        mm7,    [rdx]
-
-        pmaddwd     mm4,    mm7
-        movq        mm5,    mm1
-
-        pmaddwd     mm5,    mm7
-        movq        mm6,    mm4
-
-        punpckldq   mm4,    mm5
-        punpckhdq   mm6,    mm5
-
-        paddd       mm4,    mm6
-        movq        mm5,    mm2
-
-
-        pmaddwd     mm5,    mm7
-        movq        mm6,    mm3
-
-        pmaddwd     mm6,    mm7
-        movq        mm7,    mm5
-
-        punpckldq   mm5,    mm6
-        punpckhdq   mm7,    mm6
-
-        paddd       mm5,    mm7
-        movq        mm6,    [dct1st_stage_rounding_mmx GLOBAL]
-
-        paddd       mm4,    mm6
-        paddd       mm5,    mm6
-
-        psrad       mm4,    _1STSTAGESHIFT
-        psrad       mm5,    _1STSTAGESHIFT
-
-        packssdw    mm4,    mm5
-        movq        [rdi],  mm4
-
-        ;second column
-        movq        mm4,    mm0
-
-        pmaddwd     mm4,    [rdx+8]
-        movq        mm5,    mm1
-
-        pmaddwd     mm5,    [rdx+8]
-        movq        mm6,    mm4
-
-        punpckldq   mm4,    mm5
-        punpckhdq   mm6,    mm5
-
-        paddd       mm4,    mm6
-        movq        mm5,    mm2
-
-        pmaddwd     mm5,    [rdx+8]
-        movq        mm6,    mm3
-
-        pmaddwd     mm6,    [rdx+8]
-        movq        mm7,    mm5
-
-        punpckldq   mm5,    mm6
-        punpckhdq   mm7,    mm6
-
-        paddd       mm5,    mm7
-        movq        mm6,    [dct1st_stage_rounding_mmx GLOBAL]
-
-        paddd       mm4,    mm6
-        paddd       mm5,    mm6
-
-        psrad       mm4,    _1STSTAGESHIFT
-        psrad       mm5,    _1STSTAGESHIFT
-
-        packssdw    mm4,    mm5
-        movq        [rdi+8],  mm4
-
-
-        ;third column
-        movq        mm4,    mm0
-
-        pmaddwd     mm4,    [rdx+16]
-        movq        mm5,    mm1
-
-        pmaddwd     mm5,    [rdx+16]
-        movq        mm6,    mm4
-
-        punpckldq   mm4,    mm5
-        punpckhdq   mm6,    mm5
-
-        paddd       mm4,    mm6
-        movq        mm5,    mm2
-
-        pmaddwd     mm5,    [rdx+16]
-        movq        mm6,    mm3
-
-        pmaddwd     mm6,    [rdx+16]
-        movq        mm7,    mm5
-
-        punpckldq   mm5,    mm6
-        punpckhdq   mm7,    mm6
-
-        paddd       mm5,    mm7
-        movq        mm6,    [dct1st_stage_rounding_mmx GLOBAL]
-
-        paddd       mm4,    mm6
-        paddd       mm5,    mm6
-
-        psrad       mm4,    _1STSTAGESHIFT
-        psrad       mm5,    _1STSTAGESHIFT
-
-        packssdw    mm4,    mm5
-        movq        [rdi+16],  mm4
-
-        ;fourth column (this is the last column, so we do not have save the source any more)
-
-        pmaddwd     mm0,    [rdx+24]
-
-        pmaddwd     mm1,    [rdx+24]
-        movq        mm6,    mm0
-
-        punpckldq   mm0,    mm1
-        punpckhdq   mm6,    mm1
-
-        paddd       mm0,    mm6
-
-        pmaddwd     mm2,    [rdx+24]
-
-        pmaddwd     mm3,    [rdx+24]
-        movq        mm7,    mm2
-
-        punpckldq   mm2,    mm3
-        punpckhdq   mm7,    mm3
-
-        paddd       mm2,    mm7
-        movq        mm6,    [dct1st_stage_rounding_mmx GLOBAL]
-
-        paddd       mm0,    mm6
-        paddd       mm2,    mm6
-
-        psrad       mm0,    _1STSTAGESHIFT
-        psrad       mm2,    _1STSTAGESHIFT
-
-        packssdw    mm0,    mm2
-
-        movq        mm3,    mm0
-
-        ; done with one pass
-        ; now start second pass
-        movq        mm0,    [rdi   ]
-        movq        mm1,    [rdi+ 8]
-        movq        mm2,    [rdi+ 16]
-
-        movq        mm4,    mm0
-
-        pmaddwd     mm4,    [rdx]
-        movq        mm5,    mm1
-
-        pmaddwd     mm5,    [rdx]
-        movq        mm6,    mm4
-
-        punpckldq   mm4,    mm5
-        punpckhdq   mm6,    mm5
-
-        paddd       mm4,    mm6
-        movq        mm5,    mm2
-
-        pmaddwd     mm5,    [rdx]
-        movq        mm6,    mm3
-
-        pmaddwd     mm6,    [rdx]
-        movq        mm7,    mm5
-
-        punpckldq   mm5,    mm6
-        punpckhdq   mm7,    mm6
-
-        paddd       mm5,    mm7
-        movq        mm6,    [dct2nd_stage_rounding_mmx GLOBAL]
-
-        paddd       mm4,    mm6
-        paddd       mm5,    mm6
-
-        psrad       mm4,    _2NDSTAGESHIFT
-        psrad       mm5,    _2NDSTAGESHIFT
-
-        packssdw    mm4,    mm5
-        movq        [rdi],  mm4
-
-        ;second column
-        movq        mm4,    mm0
-
-        pmaddwd     mm4,    [rdx+8]
-        movq        mm5,    mm1
-
-        pmaddwd     mm5,    [rdx+8]
-        movq        mm6,    mm4
-
-        punpckldq   mm4,    mm5
-        punpckhdq   mm6,    mm5
-
-        paddd       mm4,    mm6
-        movq        mm5,    mm2
-
-        pmaddwd     mm5,    [rdx+8]
-        movq        mm6,    mm3
-
-        pmaddwd     mm6,    [rdx+8]
-        movq        mm7,    mm5
-
-        punpckldq   mm5,    mm6
-        punpckhdq   mm7,    mm6
-
-        paddd       mm5,    mm7
-        movq        mm6,    [dct2nd_stage_rounding_mmx GLOBAL]
-
-        paddd       mm4,    mm6
-        paddd       mm5,    mm6
-
-        psrad       mm4,    _2NDSTAGESHIFT
-        psrad       mm5,    _2NDSTAGESHIFT
-
-        packssdw    mm4,    mm5
-        movq        [rdi+8],  mm4
-
-
-        ;third column
-        movq        mm4,    mm0
-
-        pmaddwd     mm4,    [rdx+16]
-        movq        mm5,    mm1
-
-        pmaddwd     mm5,    [rdx+16]
-        movq        mm6,    mm4
-
-        punpckldq   mm4,    mm5
-        punpckhdq   mm6,    mm5
-
-        paddd       mm4,    mm6
-        movq        mm5,    mm2
-
-        pmaddwd     mm5,    [rdx+16]
-        movq        mm6,    mm3
-
-        pmaddwd     mm6,    [rdx+16]
-        movq        mm7,    mm5
-
-        punpckldq   mm5,    mm6
-        punpckhdq   mm7,    mm6
-
-        paddd       mm5,    mm7
-        movq        mm6,    [dct2nd_stage_rounding_mmx GLOBAL]
-
-        paddd       mm4,    mm6
-        paddd       mm5,    mm6
-
-        psrad       mm4,    _2NDSTAGESHIFT
-        psrad       mm5,    _2NDSTAGESHIFT
-
-        packssdw    mm4,    mm5
-        movq        [rdi+16],  mm4
-
-        ;fourth column
-        movq        mm4,    mm0
-
-        pmaddwd     mm4,    [rdx+24]
-        movq        mm5,    mm1
-
-        pmaddwd     mm5,    [rdx+24]
-        movq        mm6,    mm4
-
-        punpckldq   mm4,    mm5
-        punpckhdq   mm6,    mm5
-
-        paddd       mm4,    mm6
-        movq        mm5,    mm2
-
-        pmaddwd     mm5,    [rdx+24]
-        movq        mm6,    mm3
-
-        pmaddwd     mm6,    [rdx+24]
-        movq        mm7,    mm5
-
-        punpckldq   mm5,    mm6
-        punpckhdq   mm7,    mm6
-
-        paddd       mm5,    mm7
-        movq        mm6,    [dct2nd_stage_rounding_mmx GLOBAL]
-
-        paddd       mm4,    mm6
-        paddd       mm5,    mm6
-
-        psrad       mm4,    _2NDSTAGESHIFT
-        psrad       mm5,    _2NDSTAGESHIFT
-
-        packssdw    mm4,    mm5
-        movq        [rdi+24],  mm4
-
-    ; begin epilog
-    pop rdi
-    pop rsi
-    RESTORE_GOT
-    UNSHADOW_ARGS
-    pop         rbp
-    ret
-
-
-;void vp8_fast_fdct4x4_mmx(short *input, short *output, int pitch)
-sym(vp8_fast_fdct4x4_mmx):
-    push        rbp
-    mov         rbp, rsp
-    SHADOW_ARGS_TO_STACK 3
-    GET_GOT     rbx
-    push rsi
-    push rdi
-    ; end prolog
         mov     rsi,    arg(0) ;input
         mov     rdi,    arg(1) ;output
 
@@ -379,11 +47,11 @@ sym(vp8_fast_fdct4x4_mmx):
         movq    mm3,    [rcx + rax]
         ; get the constants
         ;shift to left by 1 for prescision
-        paddw   mm0,    mm0
-        paddw   mm1,    mm1
+        psllw   mm0,    3
+        psllw   mm1,    3
 
-        psllw   mm2,    1
-        psllw   mm3,    1
+        psllw   mm2,    3
+        psllw   mm3,    3
 
         ; transpose for the second stage
         movq    mm4,    mm0         ; 00 01 02 03
@@ -531,20 +199,23 @@ sym(vp8_fast_fdct4x4_mmx):
         movq    mm3,    mm5
         ; done with vertical
 
-		pcmpeqw	mm4,	mm4
-		pcmpeqw	mm5,	mm5
-		psrlw	mm4,	15
-		psrlw	mm5,	15
+        pcmpeqw mm4,    mm4
+        pcmpeqw mm5,    mm5
+        psrlw   mm4,    15
+        psrlw   mm5,    15
+
+        psllw   mm4,    2
+        psllw   mm5,    2
 
         paddw   mm0,    mm4
         paddw   mm1,    mm5
         paddw   mm2,    mm4
         paddw   mm3,    mm5
 
-        psraw   mm0, 1
-        psraw   mm1, 1
-        psraw   mm2, 1
-        psraw   mm3, 1
+        psraw   mm0, 3
+        psraw   mm1, 3
+        psraw   mm2, 3
+        psraw   mm3, 3
 
         movq        [rdi   ],   mm0
         movq        [rdi+ 8],   mm1
@@ -560,8 +231,8 @@ sym(vp8_fast_fdct4x4_mmx):
     ret
 
 
-;void vp8_fast_fdct8x4_wmt(short *input, short *output, int pitch)
-sym(vp8_fast_fdct8x4_wmt):
+;void vp8_short_fdct8x4_wmt(short *input, short *output, int pitch)
+sym(vp8_short_fdct8x4_wmt):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 3
@@ -584,11 +255,11 @@ sym(vp8_fast_fdct8x4_wmt):
         movdqa      xmm3,       [rcx + rax]
         ; get the constants
         ;shift to left by 1 for prescision
-        psllw       xmm0,        1
-        psllw       xmm2,        1
+        psllw       xmm0,        3
+        psllw       xmm2,        3
 
-        psllw       xmm4,        1
-        psllw       xmm3,        1
+        psllw       xmm4,        3
+        psllw       xmm3,        3
 
         ; transpose for the second stage
         movdqa      xmm1,       xmm0         ; 00 01 02 03 04 05 06 07
@@ -758,20 +429,23 @@ sym(vp8_fast_fdct8x4_wmt):
         ; done with vertical
 
 
-        pcmpeqw		xmm4,		xmm4
-        pcmpeqw		xmm5,		xmm5;
-        psrlw		xmm4,		15
-        psrlw		xmm5,		15
+        pcmpeqw     xmm4,       xmm4
+        pcmpeqw     xmm5,       xmm5;
+        psrlw       xmm4,       15
+        psrlw       xmm5,       15
+
+        psllw       xmm4,       2
+        psllw       xmm5,       2
 
         paddw       xmm0,       xmm4
         paddw       xmm1,       xmm5
         paddw       xmm2,       xmm4
         paddw       xmm3,       xmm5
 
-        psraw       xmm0,       1
-        psraw       xmm1,       1
-        psraw       xmm2,       1
-        psraw       xmm3,       1
+        psraw       xmm0,       3
+        psraw       xmm1,       3
+        psraw       xmm2,       3
+        psraw       xmm3,       3
 
         movq        QWORD PTR[rdi   ],   xmm0
         movq        QWORD PTR[rdi+ 8],   xmm1
diff --git a/vp8/encoder/x86/dct_x86.h b/vp8/encoder/x86/dct_x86.h
index 05d018043..ada16d34f 100644
--- a/vp8/encoder/x86/dct_x86.h
+++ b/vp8/encoder/x86/dct_x86.h
@@ -22,31 +22,22 @@
 #if HAVE_MMX
 extern prototype_fdct(vp8_short_fdct4x4_mmx);
 extern prototype_fdct(vp8_short_fdct8x4_mmx);
-extern prototype_fdct(vp8_fast_fdct4x4_mmx);
-extern prototype_fdct(vp8_fast_fdct8x4_mmx);
 
 #if !CONFIG_RUNTIME_CPU_DETECT
+#if 0 new c version,
 #undef  vp8_fdct_short4x4
 #define vp8_fdct_short4x4 vp8_short_fdct4x4_mmx
 
 #undef  vp8_fdct_short8x4
 #define vp8_fdct_short8x4 vp8_short_fdct8x4_mmx
-
-#undef  vp8_fdct_fast4x4
-#define vp8_fdct_fast4x4 vp8_fast_fdct4x4_mmx
-
-#undef  vp8_fdct_fast8x4
-#define vp8_fdct_fast8x4 vp8_fast_fdct8x4_mmx
+#endif
 
 #endif
 #endif
 
 
 #if HAVE_SSE2
-extern prototype_fdct(vp8_short_fdct4x4_wmt);
 extern prototype_fdct(vp8_short_fdct8x4_wmt);
-extern prototype_fdct(vp8_fast_fdct8x4_wmt);
-
 extern prototype_fdct(vp8_short_walsh4x4_sse2);
 
 #if !CONFIG_RUNTIME_CPU_DETECT
diff --git a/vp8/encoder/x86/x86_csystemdependent.c b/vp8/encoder/x86/x86_csystemdependent.c
index f3750455b..0fb82e60e 100644
--- a/vp8/encoder/x86/x86_csystemdependent.c
+++ b/vp8/encoder/x86/x86_csystemdependent.c
@@ -18,15 +18,10 @@
 #if HAVE_MMX
 void vp8_short_fdct8x4_mmx(short *input, short *output, int pitch)
 {
-    vp8_short_fdct4x4_mmx(input,   output,    pitch);
-    vp8_short_fdct4x4_mmx(input + 4, output + 16, pitch);
+    vp8_short_fdct4x4_c(input,   output,    pitch);
+    vp8_short_fdct4x4_c(input + 4, output + 16, pitch);
 }
 
-void vp8_fast_fdct8x4_mmx(short *input, short *output, int pitch)
-{
-    vp8_fast_fdct4x4_mmx(input,   output   , pitch);
-    vp8_fast_fdct4x4_mmx(input + 4, output + 16, pitch);
-}
 
 int vp8_fast_quantize_b_impl_mmx(short *coeff_ptr, short *zbin_ptr,
                                  short *qcoeff_ptr, short *dequant_ptr,
@@ -87,11 +82,6 @@ void vp8_subtract_b_mmx(BLOCK *be, BLOCKD *bd, int pitch)
 #endif
 
 #if HAVE_SSE2
-void vp8_short_fdct8x4_wmt(short *input, short *output, int pitch)
-{
-    vp8_short_fdct4x4_wmt(input,   output,    pitch);
-    vp8_short_fdct4x4_wmt(input + 4, output + 16, pitch);
-}
 
 int vp8_fast_quantize_b_impl_sse(short *coeff_ptr, short *zbin_ptr,
                                  short *qcoeff_ptr, short *dequant_ptr,
@@ -221,11 +211,19 @@ void vp8_arch_x86_encoder_init(VP8_COMP *cpi)
         cpi->rtcd.variance.get8x8var             = vp8_get8x8var_mmx;
         cpi->rtcd.variance.get16x16var           = vp8_get16x16var_mmx;
         cpi->rtcd.variance.get4x4sse_cs          = vp8_get4x4sse_cs_mmx;
-
+#if 0 // new fdct
         cpi->rtcd.fdct.short4x4                  = vp8_short_fdct4x4_mmx;
         cpi->rtcd.fdct.short8x4                  = vp8_short_fdct8x4_mmx;
-        cpi->rtcd.fdct.fast4x4                   = vp8_fast_fdct4x4_mmx;
-        cpi->rtcd.fdct.fast8x4                   = vp8_fast_fdct8x4_mmx;
+        cpi->rtcd.fdct.fast4x4                   = vp8_short_fdct4x4_mmx;
+        cpi->rtcd.fdct.fast8x4                   = vp8_short_fdct8x4_mmx;
+#else
+        cpi->rtcd.fdct.short4x4                  = vp8_short_fdct4x4_c;
+        cpi->rtcd.fdct.short8x4                  = vp8_short_fdct8x4_c;
+        cpi->rtcd.fdct.fast4x4                   = vp8_short_fdct4x4_c;
+        cpi->rtcd.fdct.fast8x4                   = vp8_short_fdct8x4_c;
+
+#endif
+
         cpi->rtcd.fdct.walsh_short4x4            = vp8_short_walsh4x4_c;
 
         cpi->rtcd.encodemb.berr                  = vp8_block_error_mmx;
@@ -270,13 +268,13 @@ void vp8_arch_x86_encoder_init(VP8_COMP *cpi)
         cpi->rtcd.variance.get16x16var           = vp8_get16x16var_sse2;
         /* cpi->rtcd.variance.get4x4sse_cs  not implemented for wmt */;
 
-#if 0
+#if 0 //new fdct
         /* short SSE2 DCT currently disabled, does not match the MMX version */
         cpi->rtcd.fdct.short4x4                  = vp8_short_fdct4x4_wmt;
         cpi->rtcd.fdct.short8x4                  = vp8_short_fdct8x4_wmt;
-#endif
         /* cpi->rtcd.fdct.fast4x4  not implemented for wmt */;
-        cpi->rtcd.fdct.fast8x4                   = vp8_fast_fdct8x4_wmt;
+        cpi->rtcd.fdct.fast8x4                   = vp8_short_fdct8x4_wmt;
+#endif
         cpi->rtcd.fdct.walsh_short4x4            = vp8_short_walsh4x4_sse2;
 
         cpi->rtcd.encodemb.berr                  = vp8_block_error_xmm;
diff --git a/vp8/vp8cx.mk b/vp8/vp8cx.mk
index f09f25852..f86a0b2aa 100644
--- a/vp8/vp8cx.mk
+++ b/vp8/vp8cx.mk
@@ -96,7 +96,6 @@ VP8_CX_SRCS-$(HAVE_MMX) += encoder/x86/subtract_mmx.asm
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/variance_sse2.c
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/variance_impl_sse2.asm
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/sad_sse2.asm
-VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/dct_sse2.asm
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/fwalsh_sse2.asm
 VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/quantize_sse2.asm
 VP8_CX_SRCS-$(HAVE_SSE3) += encoder/x86/sad_sse3.asm