[FFmpeg-devel] [PATCH 6/7] Merged first and second inner loop

[Tucker DiNapoli]

Now instead of 3 loops of 4 blocks there's only one.
Also removed some variables that became unused because of this
---
 libpostproc/postprocess_template.c | 29 +++++++++--------------------
 1 file changed, 9 insertions(+), 20 deletions(-)

diff --git a/libpostproc/postprocess_template.c b/libpostproc/postprocess_template.c
index 9096586..794ea17 100644
--- a/libpostproc/postprocess_template.c
+++ b/libpostproc/postprocess_template.c
@@ -3445,10 +3445,7 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[
         // finish 1 block before the next otherwise we might have a problem
         // with the L1 Cache of the P4 ... or only a few blocks at a time or something
         for(x=0; x<width; ){
-            int startx = x;
             int endx = FFMIN(width, x+32);
-            uint8_t *dstBlockStart = dstBlock;
-            const uint8_t *srcBlockStart = srcBlock;
             int qp_index = 0;
             for(qp_index=0; qp_index < 4; qp_index+=1){
                 QP = QPptr[(x+qp_index*8)>>qpHShift];
@@ -3472,7 +3469,16 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[
                 );
 #endif
             }
+            qp_index = 0;
             for(; x < endx; x+=BLOCK_SIZE){
+                const int stride= dstStride;
+                av_unused uint8_t *tmpXchg;
+                //temporary while changing QP stuff to make things continue to work
+                c.QP = c.QP_block[qp_index];
+                c.nonBQP = c.nonBQP_block[qp_index];
+                c.pQPb = c.pQPb_block[qp_index];
+                c.pQPb2 = c.pQPb2_block[qp_index++];
+
                 prefetchnta(srcBlock + (((x>>2)&6) + copyAhead)*srcStride + 32);
                 prefetchnta(srcBlock + (((x>>2)&6) + copyAhead+1)*srcStride + 32);
                 prefetcht0(dstBlock + (((x>>2)&6) + copyAhead)*dstStride + 32);
@@ -3496,23 +3502,6 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[
                 /*          else if(mode & CUBIC_BLEND_DEINT_FILTER)
                             RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
                 */
-                dstBlock+=8;
-                srcBlock+=8;
-            }
-
-            qp_index = 0;
-            dstBlock = dstBlockStart;
-            srcBlock = srcBlockStart;
-
-            for(x = startx; x < endx; x+=BLOCK_SIZE){
-                const int stride= dstStride;
-                av_unused uint8_t *tmpXchg;
-                //temporary while changing QP stuff to make things continue to work
-                c.QP = c.QP_block[qp_index];
-                c.nonBQP = c.nonBQP_block[qp_index];
-                c.pQPb = c.pQPb_block[qp_index];
-                c.pQPb2 = c.pQPb2_block[qp_index++];
-
                 /* only deblock if we have 2 blocks */
                 if(y + 8 < height){
                     if(mode & V_X1_FILTER)
-- 
2.3.3