vp3/x86: use full transpose for all IDCTs.

This way, the special IDCT permutations are no longer needed. Bfin code
is disabled until someone updates it. This is similar to how H264 does
it, and removes the dsputil dependency imposed by the scantable code.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

13 files changed, 141 insertions, 122 deletions

diff --git a/libavcodec/arm/vp3dsp_init_arm.c b/libavcodec/arm/vp3dsp_init_arm.c
index 5ffc8fd440..5af795bc29 100644
--- a/libavcodec/arm/vp3dsp_init_arm.c
+++ b/libavcodec/arm/vp3dsp_init_arm.c
@@ -41,6 +41,5 @@ av_cold void ff_vp3dsp_init_arm(VP3DSPContext *c, int flags)
         c->idct_dc_add   = ff_vp3_idct_dc_add_neon;
         c->v_loop_filter = ff_vp3_v_loop_filter_neon;
         c->h_loop_filter = ff_vp3_h_loop_filter_neon;
-        c->idct_perm     = FF_TRANSPOSE_IDCT_PERM;
     }
 }
diff --git a/libavcodec/bfin/vp3_bfin.c b/libavcodec/bfin/vp3_bfin.c
index e823a045b1..366955b58c 100644
--- a/libavcodec/bfin/vp3_bfin.c
+++ b/libavcodec/bfin/vp3_bfin.c
@@ -28,7 +28,7 @@
 #include "vp3_bfin.h"
 
 /* Intra iDCT offset 128 */
-void ff_bfin_vp3_idct_put (uint8_t *dest, int line_size, int16_t *block)
+static void ff_bfin_vp3_idct_put (uint8_t *dest, int line_size, int16_t *block)
 {
     uint8_t *cm = ff_cropTbl + MAX_NEG_CROP + 128;
     int i,j;
@@ -43,7 +43,7 @@ void ff_bfin_vp3_idct_put (uint8_t *dest, int line_size, int16_t *block)
 }
 
 /* Inter iDCT */
-void ff_bfin_vp3_idct_add (uint8_t *dest, int line_size, int16_t *block)
+static void ff_bfin_vp3_idct_add (uint8_t *dest, int line_size, int16_t *block)
 {
     ff_bfin_vp3_idct (block);
     ff_bfin_add_pixels_clamped (block, dest, line_size);
@@ -53,6 +53,8 @@ void ff_bfin_vp3_idct_add (uint8_t *dest, int line_size, int16_t *block)
 
 av_cold void ff_vp3dsp_init_bfin(VP3DSPContext *c, int flags)
 {
-    c->idct_add = ff_bfin_vp3_idct_add;
-    c->idct_put = ff_bfin_vp3_idct_put;
+    // FIXME: these functions are disabled because they expect unpermutated
+    // IDCT coefficients as input, but the coefficients are transposed
+    //c->idct_add = ff_bfin_vp3_idct_add;
+    //c->idct_put = ff_bfin_vp3_idct_put;
 }
diff --git a/libavcodec/bfin/vp3_bfin.h b/libavcodec/bfin/vp3_bfin.h
index f0bf824951..e832de749a 100644
--- a/libavcodec/bfin/vp3_bfin.h
+++ b/libavcodec/bfin/vp3_bfin.h
@@ -23,7 +23,5 @@
 #include <stdint.h>
 
 void ff_bfin_vp3_idct(int16_t *block);
-void ff_bfin_vp3_idct_put(uint8_t *dest, int line_size, int16_t *block);
-void ff_bfin_vp3_idct_add(uint8_t *dest, int line_size, int16_t *block);
 
 #endif /* AVCODEC_BFIN_VP3_BFIN_H */
diff --git a/libavcodec/ppc/vp3dsp_altivec.c b/libavcodec/ppc/vp3dsp_altivec.c
index 42ae00b8ba..cc587b0755 100644
--- a/libavcodec/ppc/vp3dsp_altivec.c
+++ b/libavcodec/ppc/vp3dsp_altivec.c
@@ -184,7 +184,6 @@ av_cold void ff_vp3dsp_init_ppc(VP3DSPContext *c, int flags)
     if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) {
         c->idct_put  = vp3_idct_put_altivec;
         c->idct_add  = vp3_idct_add_altivec;
-        c->idct_perm = FF_TRANSPOSE_IDCT_PERM;
     }
 #endif
 }
diff --git a/libavcodec/vp3.c b/libavcodec/vp3.c
index a77bf8fe59..5f5a3e3b22 100644
--- a/libavcodec/vp3.c
+++ b/libavcodec/vp3.c
@@ -140,6 +140,7 @@ typedef struct Vp3DecodeContext {
     ThreadFrame current_frame;
     int keyframe;
     uint8_t idct_permutation[64];
+    uint8_t idct_scantable[64];
     DSPContext dsp;
     VideoDSPContext vdsp;
     VP3DSPContext vp3dsp;
@@ -177,8 +178,6 @@ typedef struct Vp3DecodeContext {
 
     int8_t (*motion_val[2])[2];
 
-    ScanTable scantable;
-
     /* tables */
     uint16_t coded_dc_scale_factor[64];
     uint32_t coded_ac_scale_factor[64];
@@ -1356,7 +1355,7 @@ static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
                               int plane, int inter, int16_t block[64])
 {
     int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
-    uint8_t *perm = s->scantable.permutated;
+    uint8_t *perm = s->idct_scantable;
     int i = 0;
 
     do {
@@ -1699,8 +1698,12 @@ static av_cold int vp3_decode_init(AVCodecContext *avctx)
     ff_videodsp_init(&s->vdsp, 8);
     ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
 
-    ff_init_scantable_permutation(s->idct_permutation, s->vp3dsp.idct_perm);
-    ff_init_scantable(s->idct_permutation, &s->scantable, ff_zigzag_direct);
+    for (i = 0; i < 64; i++) {
+#define T(x) (x >> 3) | ((x & 7) << 3)
+        s->idct_permutation[i] = T(i);
+        s->idct_scantable[i] = T(ff_zigzag_direct[i]);
+#undef T
+    }
 
     /* initialize to an impossible value which will force a recalculation
      * in the first frame decode */
diff --git a/libavcodec/vp3dsp.c b/libavcodec/vp3dsp.c
index 051812e72d..93489637ba 100644
--- a/libavcodec/vp3dsp.c
+++ b/libavcodec/vp3dsp.c
@@ -54,11 +54,12 @@ static av_always_inline void idct(uint8_t *dst, int stride, int16_t *input, int
     /* Inverse DCT on the rows now */
     for (i = 0; i < 8; i++) {
         /* Check for non-zero values */
-        if ( ip[0] | ip[1] | ip[2] | ip[3] | ip[4] | ip[5] | ip[6] | ip[7] ) {
-            A = M(xC1S7, ip[1]) + M(xC7S1, ip[7]);
-            B = M(xC7S1, ip[1]) - M(xC1S7, ip[7]);
-            C = M(xC3S5, ip[3]) + M(xC5S3, ip[5]);
-            D = M(xC3S5, ip[5]) - M(xC5S3, ip[3]);
+        if ( ip[0 * 8] | ip[1 * 8] | ip[2 * 8] | ip[3 * 8] |
+             ip[4 * 8] | ip[5 * 8] | ip[6 * 8] | ip[7 * 8] ) {
+            A = M(xC1S7, ip[1 * 8]) + M(xC7S1, ip[7 * 8]);
+            B = M(xC7S1, ip[1 * 8]) - M(xC1S7, ip[7 * 8]);
+            C = M(xC3S5, ip[3 * 8]) + M(xC5S3, ip[5 * 8]);
+            D = M(xC3S5, ip[5 * 8]) - M(xC5S3, ip[3 * 8]);
 
             Ad = M(xC4S4, (A - C));
             Bd = M(xC4S4, (B - D));
@@ -66,11 +67,11 @@ static av_always_inline void idct(uint8_t *dst, int stride, int16_t *input, int
             Cd = A + C;
             Dd = B + D;
 
-            E = M(xC4S4, (ip[0] + ip[4]));
-            F = M(xC4S4, (ip[0] - ip[4]));
+            E = M(xC4S4, (ip[0 * 8] + ip[4 * 8]));
+            F = M(xC4S4, (ip[0 * 8] - ip[4 * 8]));
 
-            G = M(xC2S6, ip[2]) + M(xC6S2, ip[6]);
-            H = M(xC6S2, ip[2]) - M(xC2S6, ip[6]);
+            G = M(xC2S6, ip[2 * 8]) + M(xC6S2, ip[6 * 8]);
+            H = M(xC6S2, ip[2 * 8]) - M(xC2S6, ip[6 * 8]);
 
             Ed = E - G;
             Gd = E + G;
@@ -82,33 +83,33 @@ static av_always_inline void idct(uint8_t *dst, int stride, int16_t *input, int
             Hd = Bd + H;
 
             /*  Final sequence of operations over-write original inputs. */
-            ip[0] = Gd + Cd ;
-            ip[7] = Gd - Cd ;
+            ip[0 * 8] = Gd + Cd ;
+            ip[7 * 8] = Gd - Cd ;
 
-            ip[1] = Add + Hd;
-            ip[2] = Add - Hd;
+            ip[1 * 8] = Add + Hd;
+            ip[2 * 8] = Add - Hd;
 
-            ip[3] = Ed + Dd ;
-            ip[4] = Ed - Dd ;
+            ip[3 * 8] = Ed + Dd ;
+            ip[4 * 8] = Ed - Dd ;
 
-            ip[5] = Fd + Bdd;
-            ip[6] = Fd - Bdd;
+            ip[5 * 8] = Fd + Bdd;
+            ip[6 * 8] = Fd - Bdd;
         }
 
-        ip += 8;            /* next row */
+        ip += 1;            /* next row */
     }
 
     ip = input;
 
     for ( i = 0; i < 8; i++) {
         /* Check for non-zero values (bitwise or faster than ||) */
-        if ( ip[1 * 8] | ip[2 * 8] | ip[3 * 8] |
-             ip[4 * 8] | ip[5 * 8] | ip[6 * 8] | ip[7 * 8] ) {
+        if ( ip[1] | ip[2] | ip[3] |
+             ip[4] | ip[5] | ip[6] | ip[7] ) {
 
-            A = M(xC1S7, ip[1*8]) + M(xC7S1, ip[7*8]);
-            B = M(xC7S1, ip[1*8]) - M(xC1S7, ip[7*8]);
-            C = M(xC3S5, ip[3*8]) + M(xC5S3, ip[5*8]);
-            D = M(xC3S5, ip[5*8]) - M(xC5S3, ip[3*8]);
+            A = M(xC1S7, ip[1]) + M(xC7S1, ip[7]);
+            B = M(xC7S1, ip[1]) - M(xC1S7, ip[7]);
+            C = M(xC3S5, ip[3]) + M(xC5S3, ip[5]);
+            D = M(xC3S5, ip[5]) - M(xC5S3, ip[3]);
 
             Ad = M(xC4S4, (A - C));
             Bd = M(xC4S4, (B - D));
@@ -116,16 +117,16 @@ static av_always_inline void idct(uint8_t *dst, int stride, int16_t *input, int
             Cd = A + C;
             Dd = B + D;
 
-            E = M(xC4S4, (ip[0*8] + ip[4*8])) + 8;
-            F = M(xC4S4, (ip[0*8] - ip[4*8])) + 8;
+            E = M(xC4S4, (ip[0] + ip[4])) + 8;
+            F = M(xC4S4, (ip[0] - ip[4])) + 8;
 
             if(type==1){  //HACK
                 E += 16*128;
                 F += 16*128;
             }
 
-            G = M(xC2S6, ip[2*8]) + M(xC6S2, ip[6*8]);
-            H = M(xC6S2, ip[2*8]) - M(xC2S6, ip[6*8]);
+            G = M(xC2S6, ip[2]) + M(xC6S2, ip[6]);
+            H = M(xC6S2, ip[2]) - M(xC2S6, ip[6]);
 
             Ed = E - G;
             Gd = E + G;
@@ -137,19 +138,7 @@ static av_always_inline void idct(uint8_t *dst, int stride, int16_t *input, int
             Hd = Bd + H;
 
             /* Final sequence of operations over-write original inputs. */
-            if(type==0){
-                ip[0*8] = (Gd + Cd )  >> 4;
-                ip[7*8] = (Gd - Cd )  >> 4;
-
-                ip[1*8] = (Add + Hd ) >> 4;
-                ip[2*8] = (Add - Hd ) >> 4;
-
-                ip[3*8] = (Ed + Dd )  >> 4;
-                ip[4*8] = (Ed - Dd )  >> 4;
-
-                ip[5*8] = (Fd + Bdd ) >> 4;
-                ip[6*8] = (Fd - Bdd ) >> 4;
-            }else if(type==1){
+            if(type==1){
                 dst[0*stride] = av_clip_uint8((Gd + Cd )  >> 4);
                 dst[7*stride] = av_clip_uint8((Gd - Cd )  >> 4);
 
@@ -176,16 +165,7 @@ static av_always_inline void idct(uint8_t *dst, int stride, int16_t *input, int
             }
 
         } else {
-            if(type==0){
-                ip[0*8] =
-                ip[1*8] =
-                ip[2*8] =
-                ip[3*8] =
-                ip[4*8] =
-                ip[5*8] =
-                ip[6*8] =
-                ip[7*8] = ((xC4S4 * ip[0*8] + (IdctAdjustBeforeShift<<16))>>20);
-            }else if(type==1){
+            if(type==1){
                 dst[0*stride]=
                 dst[1*stride]=
                 dst[2*stride]=
@@ -193,10 +173,10 @@ static av_always_inline void idct(uint8_t *dst, int stride, int16_t *input, int
                 dst[4*stride]=
                 dst[5*stride]=
                 dst[6*stride]=
-                dst[7*stride]= av_clip_uint8(128 + ((xC4S4 * ip[0*8] + (IdctAdjustBeforeShift<<16))>>20));
+                dst[7*stride]= av_clip_uint8(128 + ((xC4S4 * ip[0] + (IdctAdjustBeforeShift<<16))>>20));
             }else{
-                if(ip[0*8]){
-                    int v= ((xC4S4 * ip[0*8] + (IdctAdjustBeforeShift<<16))>>20);
+                if(ip[0]){
+                    int v= ((xC4S4 * ip[0] + (IdctAdjustBeforeShift<<16))>>20);
                     dst[0*stride] = av_clip_uint8(dst[0*stride] + v);
                     dst[1*stride] = av_clip_uint8(dst[1*stride] + v);
                     dst[2*stride] = av_clip_uint8(dst[2*stride] + v);
@@ -209,7 +189,7 @@ static av_always_inline void idct(uint8_t *dst, int stride, int16_t *input, int
             }
         }
 
-        ip++;            /* next column */
+        ip += 8;            /* next column */
         dst++;
     }
 }
@@ -307,8 +287,6 @@ av_cold void ff_vp3dsp_init(VP3DSPContext *c, int flags)
     c->v_loop_filter = vp3_v_loop_filter_c;
     c->h_loop_filter = vp3_h_loop_filter_c;
 
-    c->idct_perm = FF_NO_IDCT_PERM;
-
     if (ARCH_ARM)
         ff_vp3dsp_init_arm(c, flags);
     if (ARCH_BFIN)
diff --git a/libavcodec/vp3dsp.h b/libavcodec/vp3dsp.h
index 558077f908..3aded9927c 100644
--- a/libavcodec/vp3dsp.h
+++ b/libavcodec/vp3dsp.h
@@ -43,8 +43,6 @@ typedef struct VP3DSPContext {
     void (*idct_dc_add)(uint8_t *dest, int line_size, int16_t *block);
     void (*v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
     void (*h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
-
-    int idct_perm;
 } VP3DSPContext;
 
 void ff_vp3dsp_init(VP3DSPContext *c, int flags);
diff --git a/libavcodec/vp5.c b/libavcodec/vp5.c
index b2d7de94f1..4eecbe3a87 100644
--- a/libavcodec/vp5.c
+++ b/libavcodec/vp5.c
@@ -172,7 +172,7 @@ static void vp5_parse_coeff(VP56Context *s)
 {
     VP56RangeCoder *c = &s->c;
     VP56Model *model = s->modelp;
-    uint8_t *permute = s->scantable.permutated;
+    uint8_t *permute = s->idct_scantable;
     uint8_t *model1, *model2;
     int coeff, sign, coeff_idx;
     int b, i, cg, idx, ctx, ctx_last;
diff --git a/libavcodec/vp56.c b/libavcodec/vp56.c
index 00334d1282..15799b7ba2 100644
--- a/libavcodec/vp56.c
+++ b/libavcodec/vp56.c
@@ -263,7 +263,7 @@ static VP56mb vp56_decode_mv(VP56Context *s, int row, int col)
 
 static void vp56_add_predictors_dc(VP56Context *s, VP56Frame ref_frame)
 {
-    int idx = s->scantable.permutated[0];
+    int idx = s->idct_scantable[0];
     int b;
 
     for (b=0; b<6; b++) {
@@ -692,8 +692,11 @@ av_cold int ff_vp56_init_context(AVCodecContext *avctx, VP56Context *s,
     ff_videodsp_init(&s->vdsp, 8);
     ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
     ff_vp56dsp_init(&s->vp56dsp, avctx->codec->id);
-    ff_init_scantable_permutation(s->dsp.idct_permutation, s->vp3dsp.idct_perm);
-    ff_init_scantable(s->dsp.idct_permutation, &s->scantable,ff_zigzag_direct);
+    for (i = 0; i < 64; i++) {
+#define T(x) (x >> 3) | ((x & 7) << 3)
+        s->idct_scantable[i] = T(ff_zigzag_direct[i]);
+#undef T
+    }
 
     for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++) {
         s->frames[i] = av_frame_alloc();
diff --git a/libavcodec/vp56.h b/libavcodec/vp56.h
index e9a53a1667..926d28b5ae 100644
--- a/libavcodec/vp56.h
+++ b/libavcodec/vp56.h
@@ -100,7 +100,7 @@ struct vp56_context {
     VideoDSPContext vdsp;
     VP3DSPContext vp3dsp;
     VP56DSPContext vp56dsp;
-    ScanTable scantable;
+    uint8_t idct_scantable[64];
     AVFrame *frames[4];
     uint8_t *edge_emu_buffer_alloc;
     uint8_t *edge_emu_buffer;
diff --git a/libavcodec/vp6.c b/libavcodec/vp6.c
index 4d27f7023d..ea4841f2f8 100644
--- a/libavcodec/vp6.c
+++ b/libavcodec/vp6.c
@@ -368,7 +368,7 @@ static unsigned vp6_get_nb_null(VP56Context *s)
 static void vp6_parse_coeff_huffman(VP56Context *s)
 {
     VP56Model *model = s->modelp;
-    uint8_t *permute = s->scantable.permutated;
+    uint8_t *permute = s->idct_scantable;
     VLC *vlc_coeff;
     int coeff, sign, coeff_idx;
     int b, cg, idx;
@@ -428,7 +428,7 @@ static void vp6_parse_coeff(VP56Context *s)
 {
     VP56RangeCoder *c = s->ccp;
     VP56Model *model = s->modelp;
-    uint8_t *permute = s->scantable.permutated;
+    uint8_t *permute = s->idct_scantable;
     uint8_t *model1, *model2, *model3;
     int coeff, sign, coeff_idx;
     int b, i, cg, idx, ctx;
diff --git a/libavcodec/x86/vp3dsp.asm b/libavcodec/x86/vp3dsp.asm
index be5f21b2d8..423866cd2f 100644
--- a/libavcodec/x86/vp3dsp.asm
+++ b/libavcodec/x86/vp3dsp.asm
@@ -500,22 +500,22 @@ cglobal vp3_h_loop_filter, 3, 4
 
     ; at this point, function has completed dequantization + dezigzag +
     ; partial transposition; now do the idct itself
-%define I(x) [%1+16* x     ]
-%define J(x) [%1+16*(x-4)+8]
+%define I(x) [%1+16*x]
+%define J(x) [%1+16*x]
     RowIDCT
     Transpose
 
-%define I(x) [%1+16* x   +64]
-%define J(x) [%1+16*(x-4)+72]
+%define I(x) [%1+16*x+8]
+%define J(x) [%1+16*x+8]
     RowIDCT
     Transpose
 
-%define I(x) [%1+16*x]
-%define J(x) [%1+16*x]
+%define I(x) [%1+16* x]
+%define J(x) [%1+16*(x-4)+8]
     ColumnIDCT
 
-%define I(x) [%1+16*x+8]
-%define J(x) [%1+16*x+8]
+%define I(x) [%1+16* x   +64]
+%define J(x) [%1+16*(x-4)+72]
     ColumnIDCT
 %endif ; mmsize == 16/8
 %endmacro
@@ -533,10 +533,17 @@ cglobal vp3_idct_put, 3, 4, 9
     mova          m1, [r2+mmsize*2+%%i]
     mova          m2, [r2+mmsize*4+%%i]
     mova          m3, [r2+mmsize*6+%%i]
+%if mmsize == 8
+    packsswb      m0, [r2+mmsize*8+%%i]
+    packsswb      m1, [r2+mmsize*10+%%i]
+    packsswb      m2, [r2+mmsize*12+%%i]
+    packsswb      m3, [r2+mmsize*14+%%i]
+%else
     packsswb      m0, [r2+mmsize*1+%%i]
     packsswb      m1, [r2+mmsize*3+%%i]
     packsswb      m2, [r2+mmsize*5+%%i]
     packsswb      m3, [r2+mmsize*7+%%i]
+%endif
     paddb         m0, m4
     paddb         m1, m4
     paddb         m2, m4
@@ -560,7 +567,7 @@ cglobal vp3_idct_put, 3, 4, 9
     movq   [r0+r1*2], m3
     movhps [r0+r3  ], m3
 %endif
-%assign %%i %%i+64
+%assign %%i %%i+8
 %endrep
 
     pxor          m0, m0
@@ -574,47 +581,81 @@ cglobal vp3_idct_put, 3, 4, 9
 cglobal vp3_idct_add, 3, 4, 9
     VP3_IDCT      r2
 
-    mov           r3, 4
-    pxor          m4, m4
     movsxdifnidn  r1, r1d
-.loop:
+    lea           r3, [r1*3]
+    pxor          m4, m4
+%if mmsize == 16
+%assign %%i 0
+%rep 2
     movq          m0, [r0]
     movq          m1, [r0+r1]
-%if mmsize == 8
-    mova          m2, m0
-    mova          m3, m1
-%endif
+    movq          m2, [r0+r1*2]
+    movq          m3, [r0+r3]
     punpcklbw     m0, m4
     punpcklbw     m1, m4
-%if mmsize == 8
-    punpckhbw     m2, m4
-    punpckhbw     m3, m4
-%endif
-    paddsw        m0, [r2+ 0]
-    paddsw        m1, [r2+16]
-%if mmsize == 8
-    paddsw        m2, [r2+ 8]
-    paddsw        m3, [r2+24]
-    packuswb      m0, m2
-    packuswb      m1, m3
-%else ; mmsize == 16
+    punpcklbw     m2, m4
+    punpcklbw     m3, m4
+    paddsw        m0, [r2+ 0+%%i]
+    paddsw        m1, [r2+16+%%i]
+    paddsw        m2, [r2+32+%%i]
+    paddsw        m3, [r2+48+%%i]
     packuswb      m0, m1
+    packuswb      m2, m3
+    movq   [r0     ], m0
+    movhps [r0+r1  ], m0
+    movq   [r0+r1*2], m2
+    movhps [r0+r3  ], m2
+%if %%i == 0
+    lea           r0, [r0+r1*4]
 %endif
-    movq     [r0   ], m0
-%if mmsize == 8
-    movq     [r0+r1], m1
-%else ; mmsize == 16
-    movhps   [r0+r1], m0
+%assign %%i %%i+64
+%endrep
+%else
+%assign %%i 0
+%rep 2
+    movq          m0, [r0]
+    movq          m1, [r0+r1]
+    movq          m2, [r0+r1*2]
+    movq          m3, [r0+r3]
+    movq          m5, m0
+    movq          m6, m1
+    movq          m7, m2
+    punpcklbw     m0, m4
+    punpcklbw     m1, m4
+    punpcklbw     m2, m4
+    punpckhbw     m5, m4
+    punpckhbw     m6, m4
+    punpckhbw     m7, m4
+    paddsw        m0, [r2+ 0+%%i]
+    paddsw        m1, [r2+16+%%i]
+    paddsw        m2, [r2+32+%%i]
+    paddsw        m5, [r2+64+%%i]
+    paddsw        m6, [r2+80+%%i]
+    paddsw        m7, [r2+96+%%i]
+    packuswb      m0, m5
+    movq          m5, m3
+    punpcklbw     m3, m4
+    punpckhbw     m5, m4
+    packuswb      m1, m6
+    paddsw        m3, [r2+48+%%i]
+    paddsw        m5, [r2+112+%%i]
+    packuswb      m2, m7
+    packuswb      m3, m5
+    movq   [r0     ], m0
+    movq   [r0+r1  ], m1
+    movq   [r0+r1*2], m2
+    movq   [r0+r3  ], m3
+%if %%i == 0
+    lea           r0, [r0+r1*4]
 %endif
-    lea           r0, [r0+r1*2]
-%assign %%offset 0
-%rep 32/mmsize
-    mova [r2+%%offset], m4
-%assign %%offset %%offset+mmsize
+%assign %%i %%i+8
+%endrep
+%endif
+%assign %%i 0
+%rep 128/mmsize
+    mova    [r2+%%i], m4
+%assign %%i %%i+mmsize
 %endrep
-    add           r2, 32
-    dec           r3
-    jg .loop
     RET
 %endmacro
 
diff --git a/libavcodec/x86/vp3dsp_init.c b/libavcodec/x86/vp3dsp_init.c
index 288dbcc6f3..252b40a432 100644
--- a/libavcodec/x86/vp3dsp_init.c
+++ b/libavcodec/x86/vp3dsp_init.c
@@ -109,7 +109,6 @@ av_cold void ff_vp3dsp_init_x86(VP3DSPContext *c, int flags)
     if (EXTERNAL_MMX(cpuflags)) {
         c->idct_put  = ff_vp3_idct_put_mmx;
         c->idct_add  = ff_vp3_idct_add_mmx;
-        c->idct_perm = FF_PARTTRANS_IDCT_PERM;
     }
 #endif
 
@@ -125,6 +124,5 @@ av_cold void ff_vp3dsp_init_x86(VP3DSPContext *c, int flags)
     if (EXTERNAL_SSE2(cpuflags)) {
         c->idct_put  = ff_vp3_idct_put_sse2;
         c->idct_add  = ff_vp3_idct_add_sse2;
-        c->idct_perm = FF_TRANSPOSE_IDCT_PERM;
     }
 }