# HG changeset patch
# User darkshikari
# Date 1277782464 0
# Node ID ec7be1d7d5b4a90a2cff26329a6233aac2dbe600
# Parent  d584c7373a648436c2949e0a7e271304c4ded67d
Use crop table in C implementations of VP8 DSP functions.
Much faster VP8 C DSP functions; ~5-10% faster overall with asm off.

diff -r d584c7373a64 -r ec7be1d7d5b4 vp8dsp.c
--- a/vp8dsp.c	Tue Jun 29 01:41:59 2010 +0000
+++ b/vp8dsp.c	Tue Jun 29 03:34:24 2010 +0000
@@ -61,6 +61,7 @@
 static void vp8_idct_add_c(uint8_t *dst, DCTELEM block[16], int stride)
 {
     int i, t0, t1, t2, t3;
+    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
     DCTELEM tmp[16];
 
     for (i = 0; i < 4; i++) {
@@ -81,10 +82,10 @@
         t2 = MUL_35468(tmp[1*4+i]) - MUL_20091(tmp[3*4+i]);
         t3 = MUL_20091(tmp[1*4+i]) + MUL_35468(tmp[3*4+i]);
 
-        dst[0] = av_clip_uint8(dst[0] + ((t0 + t3 + 4) >> 3));
-        dst[1] = av_clip_uint8(dst[1] + ((t1 + t2 + 4) >> 3));
-        dst[2] = av_clip_uint8(dst[2] + ((t1 - t2 + 4) >> 3));
-        dst[3] = av_clip_uint8(dst[3] + ((t0 - t3 + 4) >> 3));
+        dst[0] = cm[dst[0] + ((t0 + t3 + 4) >> 3)];
+        dst[1] = cm[dst[1] + ((t1 + t2 + 4) >> 3)];
+        dst[2] = cm[dst[2] + ((t1 - t2 + 4) >> 3)];
+        dst[3] = cm[dst[3] + ((t0 - t3 + 4) >> 3)];
         dst += stride;
     }
 }
@@ -92,12 +93,13 @@
 static void vp8_idct_dc_add_c(uint8_t *dst, DCTELEM block[16], int stride)
 {
     int i, dc = (block[0] + 4) >> 3;
+    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP + dc;
 
     for (i = 0; i < 4; i++) {
-        dst[0] = av_clip_uint8(dst[0] + dc);
-        dst[1] = av_clip_uint8(dst[1] + dc);
-        dst[2] = av_clip_uint8(dst[2] + dc);
-        dst[3] = av_clip_uint8(dst[3] + dc);
+        dst[0] = cm[dst[0]];
+        dst[1] = cm[dst[1]];
+        dst[2] = cm[dst[2]];
+        dst[3] = cm[dst[3]];
         dst += stride;
     }
 }
@@ -114,17 +116,20 @@
     int av_unused q2 = p[ 2*stride];\
     int av_unused q3 = p[ 3*stride];
 
+#define clip_int8(n) (cm[n+0x80]-0x80)
+
 static av_always_inline void filter_common(uint8_t *p, int stride, int is4tap)
 {
     LOAD_PIXELS
     int a, f1, f2;
+    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
 
     a = 3*(q0 - p0);
 
     if (is4tap)
-        a += av_clip_int8(p1 - q1);
+        a += clip_int8(p1 - q1);
 
-    a = av_clip_int8(a);
+    a = clip_int8(a);
 
     // We deviate from the spec here with c(a+3) >> 3
     // since that's what libvpx does.
@@ -133,14 +138,14 @@
 
     // Despite what the spec says, we do need to clamp here to
     // be bitexact with libvpx.
-    p[-1*stride] = av_clip_uint8(p0 + f2);
-    p[ 0*stride] = av_clip_uint8(q0 - f1);
+    p[-1*stride] = cm[p0 + f2];
+    p[ 0*stride] = cm[q0 - f1];
 
     // only used for _inner on blocks without high edge variance
     if (!is4tap) {
         a = (f1+1)>>1;
-        p[-2*stride] = av_clip_uint8(p1 + a);
-        p[ 1*stride] = av_clip_uint8(q1 - a);
+        p[-2*stride] = cm[p1 + a];
+        p[ 1*stride] = cm[q1 - a];
     }
 }
 
@@ -172,22 +177,23 @@
 static av_always_inline void filter_mbedge(uint8_t *p, int stride)
 {
     int a0, a1, a2, w;
+    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
 
     LOAD_PIXELS
 
-    w = av_clip_int8(p1-q1);
-    w = av_clip_int8(w + 3*(q0-p0));
+    w = clip_int8(p1-q1);
+    w = clip_int8(w + 3*(q0-p0));
 
     a0 = (27*w + 63) >> 7;
     a1 = (18*w + 63) >> 7;
     a2 = ( 9*w + 63) >> 7;
 
-    p[-3*stride] = av_clip_uint8(p2 + a2);
-    p[-2*stride] = av_clip_uint8(p1 + a1);
-    p[-1*stride] = av_clip_uint8(p0 + a0);
-    p[ 0*stride] = av_clip_uint8(q0 - a0);
-    p[ 1*stride] = av_clip_uint8(q1 - a1);
-    p[ 2*stride] = av_clip_uint8(q2 - a2);
+    p[-3*stride] = cm[p2 + a2];
+    p[-2*stride] = cm[p1 + a1];
+    p[-1*stride] = cm[p0 + a0];
+    p[ 0*stride] = cm[q0 - a0];
+    p[ 1*stride] = cm[q1 - a1];
+    p[ 2*stride] = cm[q2 - a2];
 }
 
 #define LOOP_FILTER(dir, size, stridea, strideb) \
@@ -263,17 +269,18 @@
 PUT_PIXELS(4)
 
 #define FILTER_6TAP(src, F, stride) \
-    av_clip_uint8((F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + F[0]*src[x-2*stride] + \
-                   F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + F[5]*src[x+3*stride] + 64) >> 7)
+    cm[(F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + F[0]*src[x-2*stride] + \
+        F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + F[5]*src[x+3*stride] + 64) >> 7]
 
 #define FILTER_4TAP(src, F, stride) \
-    av_clip_uint8((F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + \
-                   F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + 64) >> 7)
+    cm[(F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + \
+        F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + 64) >> 7]
 
 #define VP8_EPEL_H(SIZE, FILTER, FILTERNAME) \
 static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
 { \
     const uint8_t *filter = subpel_filters[mx-1]; \
+    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
     int x, y; \
 \
     for (y = 0; y < h; y++) { \
@@ -287,6 +294,7 @@
 static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
 { \
     const uint8_t *filter = subpel_filters[my-1]; \
+    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
     int x, y; \
 \
     for (y = 0; y < h; y++) { \
@@ -300,6 +308,7 @@
 static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
 { \
     const uint8_t *filter = subpel_filters[mx-1]; \
+    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
     int x, y; \
     uint8_t tmp_array[(2*SIZE+5)*SIZE]; \
     uint8_t *tmp = tmp_array; \