swscale: refactor vertical scaler
authorPedro Arthur <bygrandao@gmail.com>
Tue, 18 Aug 2015 14:47:55 +0000 (11:47 -0300)
committerPedro Arthur <bygrandao@gmail.com>
Wed, 19 Aug 2015 13:43:52 +0000 (10:43 -0300)
libswscale/Makefile
libswscale/slice.c
libswscale/swscale.c
libswscale/swscale_internal.h
libswscale/vscale.c [new file with mode: 0644]
libswscale/x86/swscale.c

index b2b6381..e70e358 100644 (file)
@@ -17,6 +17,7 @@ OBJS = alphablend.o                                     \
        yuv2rgb.o                                        \
        slice.o                                          \
        hscale.o                                         \
+       vscale.o                                         \
 
 OBJS-$(CONFIG_SHARED)        += log2_tab.o
 
index 611e4e6..8fd16d3 100644 (file)
@@ -214,6 +214,7 @@ int ff_init_filters(SwsContext * c)
     int index;
     int num_ydesc;
     int num_cdesc;
+    int num_vdesc = isPlanarYUV(c->dstFormat) && !isGray(c->dstFormat) ? 2 : 1;
     int need_lum_conv = c->lumToYV12 || c->readLumPlanar || c->alpToYV12 || c->readAlpPlanar;
     int need_chr_conv = c->chrToYV12 || c->readChrPlanar;
     int srcIdx, dstIdx;
@@ -228,8 +229,8 @@ int ff_init_filters(SwsContext * c)
     num_ydesc = need_lum_conv ? 2 : 1;
     num_cdesc = need_chr_conv ? 2 : 1;
 
-    c->numSlice = FFMAX(num_ydesc, num_cdesc) + 1;
-    c->numDesc = num_ydesc + num_cdesc;
+    c->numSlice = FFMAX(num_ydesc, num_cdesc) + 2;
+    c->numDesc = num_ydesc + num_cdesc + num_vdesc;
     c->descIndex[0] = num_ydesc;
     c->descIndex[1] = num_ydesc + num_cdesc;
 
@@ -243,12 +244,13 @@ int ff_init_filters(SwsContext * c)
 
     res = alloc_slice(&c->slice[0], c->srcFormat, c->srcH, c->chrSrcH, c->chrSrcHSubSample, c->chrSrcVSubSample, 0);
     if (res < 0) goto cleanup;
-    for (i = 1; i < c->numSlice-1; ++i) {
+    for (i = 1; i < c->numSlice-2; ++i) {
         res = alloc_slice(&c->slice[i], c->srcFormat, c->vLumFilterSize + MAX_LINES_AHEAD, c->vChrFilterSize + MAX_LINES_AHEAD, c->chrSrcHSubSample, c->chrSrcVSubSample, 0);
         if (res < 0) goto cleanup;
         res = alloc_lines(&c->slice[i], FFALIGN(c->srcW*2+78, 16), c->srcW);
         if (res < 0) goto cleanup;
     }
+    // horizontal scaler output
     res = alloc_slice(&c->slice[i], c->srcFormat, c->vLumFilterSize + MAX_LINES_AHEAD, c->vChrFilterSize + MAX_LINES_AHEAD, c->chrDstHSubSample, c->chrDstVSubSample, 1);
     if (res < 0) goto cleanup;
     res = alloc_lines(&c->slice[i], dst_stride, c->dstW);
@@ -256,6 +258,11 @@ int ff_init_filters(SwsContext * c)
 
     fill_ones(&c->slice[i], dst_stride>>1, c->dstBpc == 16);
 
+    // vertical scaler output
+    ++i;
+    res = alloc_slice(&c->slice[i], c->dstFormat, c->dstH, c->chrDstH, c->chrDstHSubSample, c->chrDstVSubSample, 0);
+    if (res < 0) goto cleanup;
+
     index = 0;
     srcIdx = 0;
     dstIdx = 1;
@@ -290,6 +297,13 @@ int ff_init_filters(SwsContext * c)
             ff_init_desc_no_chr(&c->desc[index], &c->slice[srcIdx], &c->slice[dstIdx]);
     }
 
+    ++index;
+    {
+        srcIdx = c->numSlice - 2;
+        dstIdx = c->numSlice - 1;
+        ff_init_vscale(c, c->desc + index, c->slice + srcIdx, c->slice + dstIdx);
+    }
+
     return 0;
 
 cleanup:
index 03019d4..d87efda 100644 (file)
@@ -326,8 +326,8 @@ static int swscale(SwsContext *c, const uint8_t *src[],
 #endif
     const int dstW                   = c->dstW;
     const int dstH                   = c->dstH;
-    const int chrDstW                = c->chrDstW;
 #ifndef NEW_FILTER
+    const int chrDstW                = c->chrDstW;
     const int chrSrcW                = c->chrSrcW;
     const int lumXInc                = c->lumXInc;
     const int chrXInc                = c->chrXInc;
@@ -341,9 +341,9 @@ static int swscale(SwsContext *c, const uint8_t *src[],
     int32_t *hChrFilterPos           = c->hChrFilterPos;
     int16_t *hLumFilter              = c->hLumFilter;
     int16_t *hChrFilter              = c->hChrFilter;
-#endif
     int32_t *lumMmxFilter            = c->lumMmxFilter;
     int32_t *chrMmxFilter            = c->chrMmxFilter;
+#endif
     const int vLumFilterSize         = c->vLumFilterSize;
     const int vChrFilterSize         = c->vChrFilterSize;
 #ifndef NEW_FILTER
@@ -381,13 +381,18 @@ static int swscale(SwsContext *c, const uint8_t *src[],
     int lastInChrBuf = c->lastInChrBuf;
 //    int perform_gamma = c->is_internal_gamma;
 
+#ifdef NEW_FILTER
     int lumStart = 0;
     int lumEnd = c->descIndex[0];
     int chrStart = lumEnd;
     int chrEnd = c->descIndex[1];
+    int vStart = chrEnd;
+    int vEnd = c->numDesc;
     SwsSlice *src_slice = &c->slice[lumStart];
-    SwsSlice *dst_slice = &c->slice[c->numSlice-1];
+    SwsSlice *hout_slice = &c->slice[c->numSlice-2];
+    SwsSlice *vout_slice = &c->slice[c->numSlice-1];
     SwsFilterDescriptor *desc = c->desc;
+#endif
     int hasLumHoles = 1;
     int hasChrHoles = 1;
 
@@ -460,20 +465,28 @@ static int swscale(SwsContext *c, const uint8_t *src[],
     }
     lastDstY = dstY;
 
-    ff_init_slice_from_src(src_slice, (uint8_t**)src, srcStride, c->srcW,
-            srcSliceY, srcSliceH,
-            chrSrcSliceY, chrSrcSliceH);
-
-    dst_slice->plane[0].sliceY = lastInLumBuf + 1;
-    dst_slice->plane[1].sliceY = lastInChrBuf + 1;
-    dst_slice->plane[2].sliceY = lastInChrBuf + 1;
-    dst_slice->plane[3].sliceY = lastInLumBuf + 1;
+#ifdef NEW_FILTER
+    ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
+                   yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, c->use_mmx_vfilter);
 
-    dst_slice->plane[0].sliceH =
-    dst_slice->plane[1].sliceH =
-    dst_slice->plane[2].sliceH =
-    dst_slice->plane[3].sliceH = 0;
-    dst_slice->width = dstW;
+    ff_init_slice_from_src(src_slice, (uint8_t**)src, srcStride, c->srcW,
+            srcSliceY, srcSliceH, chrSrcSliceY, chrSrcSliceH);
+
+    ff_init_slice_from_src(vout_slice, (uint8_t**)dst, dstStride, c->dstW,
+            dstY, dstH, dstY >> c->chrDstVSubSample,
+            FF_CEIL_RSHIFT(dstH, c->chrDstVSubSample));
+
+    hout_slice->plane[0].sliceY = lastInLumBuf + 1;
+    hout_slice->plane[1].sliceY = lastInChrBuf + 1;
+    hout_slice->plane[2].sliceY = lastInChrBuf + 1;
+    hout_slice->plane[3].sliceY = lastInLumBuf + 1;
+
+    hout_slice->plane[0].sliceH =
+    hout_slice->plane[1].sliceH =
+    hout_slice->plane[2].sliceH =
+    hout_slice->plane[3].sliceH = 0;
+    hout_slice->width = dstW;
+#endif
 
     for (; dstY < dstH; dstY++) {
         const int chrDstY = dstY >> c->chrDstVSubSample;
@@ -496,29 +509,35 @@ static int swscale(SwsContext *c, const uint8_t *src[],
         int lastLumSrcY2 = FFMIN(c->srcH,    firstLumSrcY2 + vLumFilterSize) - 1;
         int lastChrSrcY  = FFMIN(c->chrSrcH, firstChrSrcY  + vChrFilterSize) - 1;
         int enough_lines;
+#ifdef NEW_FILTER
         int i;
         int posY, cPosY, firstPosY, lastPosY, firstCPosY, lastCPosY;
+#endif
 
         // handle holes (FAST_BILINEAR & weird filters)
         if (firstLumSrcY > lastInLumBuf) {
+#ifdef NEW_FILTER
             hasLumHoles = lastInLumBuf != firstLumSrcY - 1;
-            lastInLumBuf = firstLumSrcY - 1;
             if (hasLumHoles) {
-                dst_slice->plane[0].sliceY = lastInLumBuf + 1;
-                dst_slice->plane[3].sliceY = lastInLumBuf + 1;
-                dst_slice->plane[0].sliceH =
-                dst_slice->plane[3].sliceH = 0;
+                hout_slice->plane[0].sliceY = lastInLumBuf + 1;
+                hout_slice->plane[3].sliceY = lastInLumBuf + 1;
+                hout_slice->plane[0].sliceH =
+                hout_slice->plane[3].sliceH = 0;
             }
+#endif
+            lastInLumBuf = firstLumSrcY - 1;
         }
         if (firstChrSrcY > lastInChrBuf) {
+#ifdef NEW_FILTER
             hasChrHoles = lastInChrBuf != firstChrSrcY - 1;
-            lastInChrBuf = firstChrSrcY - 1;
             if (hasChrHoles) {
-                dst_slice->plane[1].sliceY = lastInChrBuf + 1;
-                dst_slice->plane[2].sliceY = lastInChrBuf + 1;
-                dst_slice->plane[1].sliceH =
-                dst_slice->plane[2].sliceH = 0;
+                hout_slice->plane[1].sliceY = lastInChrBuf + 1;
+                hout_slice->plane[2].sliceY = lastInChrBuf + 1;
+                hout_slice->plane[1].sliceH =
+                hout_slice->plane[2].sliceH = 0;
             }
+#endif
+            lastInChrBuf = firstChrSrcY - 1;
         }
         av_assert0(firstLumSrcY >= lastInLumBuf - vLumBufSize + 1);
         av_assert0(firstChrSrcY >= lastInChrBuf - vChrBufSize + 1);
@@ -541,7 +560,7 @@ static int swscale(SwsContext *c, const uint8_t *src[],
         }
 
 #ifdef NEW_FILTER
-        posY = dst_slice->plane[0].sliceY + dst_slice->plane[0].sliceH;
+        posY = hout_slice->plane[0].sliceY + hout_slice->plane[0].sliceH;
         if (posY <= lastLumSrcY && !hasLumHoles) {
             firstPosY = FFMAX(firstLumSrcY, posY);
             lastPosY = FFMIN(lastLumSrcY + MAX_LINES_AHEAD, srcSliceY + srcSliceH - 1);
@@ -550,7 +569,7 @@ static int swscale(SwsContext *c, const uint8_t *src[],
             lastPosY = lastLumSrcY;
         }
 
-        cPosY = dst_slice->plane[1].sliceY + dst_slice->plane[1].sliceH;
+        cPosY = hout_slice->plane[1].sliceY + hout_slice->plane[1].sliceH;
         if (cPosY <= lastChrSrcY && !hasChrHoles) {
             firstCPosY = FFMAX(firstChrSrcY, cPosY);
             lastCPosY = FFMIN(lastChrSrcY + MAX_LINES_AHEAD, FF_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample) - 1);
@@ -559,7 +578,7 @@ static int swscale(SwsContext *c, const uint8_t *src[],
             lastCPosY = lastChrSrcY;
         }
 
-        ff_rotate_slice(dst_slice, lastPosY, lastCPosY);
+        ff_rotate_slice(hout_slice, lastPosY, lastCPosY);
 
         if (posY < lastLumSrcY + 1) {
             for (i = lumStart; i < lumEnd; ++i)
@@ -650,22 +669,20 @@ static int swscale(SwsContext *c, const uint8_t *src[],
             ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX,
                                      &yuv2packed1, &yuv2packed2, &yuv2packedX, &yuv2anyX);
             use_mmx_vfilter= 0;
+            ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
+                           yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, use_mmx_vfilter);
         }
 
         {
 #ifdef NEW_FILTER
-            const int16_t **lumSrcPtr  = (const int16_t **)(void*) dst_slice->plane[0].line + firstLumSrcY - dst_slice->plane[0].sliceY;
-            const int16_t **chrUSrcPtr = (const int16_t **)(void*) dst_slice->plane[1].line + firstChrSrcY - dst_slice->plane[1].sliceY;
-            const int16_t **chrVSrcPtr = (const int16_t **)(void*) dst_slice->plane[2].line + firstChrSrcY - dst_slice->plane[2].sliceY;
-            const int16_t **alpSrcPtr  = (CONFIG_SWSCALE_ALPHA && alpPixBuf) ?
-                                         (const int16_t **)(void*) dst_slice->plane[3].line + firstLumSrcY - dst_slice->plane[3].sliceY : NULL;
+            for (i = vStart; i < vEnd; ++i)
+                desc[i].process(c, &desc[i], dstY, 1);
 #else
             const int16_t **lumSrcPtr  = (const int16_t **)(void*) lumPixBuf  + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;
             const int16_t **chrUSrcPtr = (const int16_t **)(void*) chrUPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
             const int16_t **chrVSrcPtr = (const int16_t **)(void*) chrVPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
             const int16_t **alpSrcPtr  = (CONFIG_SWSCALE_ALPHA && alpPixBuf) ?
                                          (const int16_t **)(void*) alpPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize : NULL;
-#endif
             int16_t *vLumFilter = c->vLumFilter;
             int16_t *vChrFilter = c->vChrFilter;
 
@@ -766,6 +783,7 @@ static int swscale(SwsContext *c, const uint8_t *src[],
                          chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
                          alpSrcPtr, dest, dstW, dstY);
             }
+#endif
             //if (perform_gamma)
             //    gamma_convert(dest, dstW, c->gamma);
         }
index 4e1cfc1..0d4ce33 100644 (file)
@@ -1002,6 +1002,15 @@ typedef struct FilterContext
     int xInc;
 } FilterContext;
 
+typedef struct VScalerContext
+{
+    uint16_t *filter[2];
+    int32_t  *filter_pos;
+    int filter_size;
+    int isMMX;
+    void *pfn;
+} VScalerContext;
+
 // warp input lines in the form (src + width*i + j) to slice format (line[i][j])
 int ff_init_slice_from_src(SwsSlice * s, uint8_t *src[4], int stride[4], int srcW, int lumY, int lumH, int chrY, int chrH);
 
@@ -1033,9 +1042,18 @@ int ff_init_desc_chscale(SwsFilterDescriptor *desc, SwsSlice *src, SwsSlice *dst
 
 int ff_init_desc_no_chr(SwsFilterDescriptor *desc, SwsSlice * src, SwsSlice *dst);
 
+/// initializes vertical scaling descriptors
+int ff_init_vscale(SwsContext *c, SwsFilterDescriptor *desc, SwsSlice *src, SwsSlice *dst);
+
+/// setup vertical scaler functions
+void ff_init_vscale_pfn(SwsContext *c, yuv2planar1_fn yuv2plane1, yuv2planarX_fn yuv2planeX,
+    yuv2interleavedX_fn yuv2nv12cX, yuv2packed1_fn yuv2packed1, yuv2packed2_fn yuv2packed2,
+    yuv2packedX_fn yuv2packedX, yuv2anyX_fn yuv2anyX, int use_mmx);
+
 //number of extra lines to process
 #define MAX_LINES_AHEAD 4
 
-
+// enable use of refactored scaler code
 #define NEW_FILTER
+
 #endif /* SWSCALE_SWSCALE_INTERNAL_H */
diff --git a/libswscale/vscale.c b/libswscale/vscale.c
new file mode 100644 (file)
index 0000000..18d5ad1
--- /dev/null
@@ -0,0 +1,287 @@
+/*
+ * Copyright (C) 2015 Pedro Arthur <bygrandao@gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+#include "swscale_internal.h"
+
+static int lum_planar_vscale(SwsContext *c, SwsFilterDescriptor *desc, int sliceY, int sliceH)
+{
+    VScalerContext *inst = desc->instance;
+    int dstW = desc->dst->width;
+
+    int first = FFMAX(1-inst->filter_size, inst->filter_pos[sliceY]);
+    int sp = first - desc->src->plane[0].sliceY;
+    int dp = sliceY - desc->dst->plane[0].sliceY;
+    uint8_t **src = desc->src->plane[0].line + sp;
+    uint8_t **dst = desc->dst->plane[0].line + dp;
+    uint16_t *filter = inst->filter[0] + (inst->isMMX ? 0 : sliceY * inst->filter_size);
+
+    if (inst->filter_size == 1)
+        ((yuv2planar1_fn)inst->pfn)((const int16_t*)src[0], dst[0], dstW, c->lumDither8, 0);
+    else
+        ((yuv2planarX_fn)inst->pfn)(filter, inst->filter_size, (const int16_t**)src, dst[0], dstW, c->lumDither8, 0);
+
+    if (desc->alpha) {
+        int sp = first - desc->src->plane[3].sliceY;
+        int dp = sliceY - desc->dst->plane[3].sliceY;
+        uint8_t **src = desc->src->plane[3].line + sp;
+        uint8_t **dst = desc->dst->plane[3].line + dp;
+        uint16_t *filter = inst->filter[1] + (inst->isMMX ? 0 : sliceY * inst->filter_size);
+
+        if (inst->filter_size == 1)
+            ((yuv2planar1_fn)inst->pfn)((const int16_t*)src[0], dst[0], dstW, c->lumDither8, 0);
+        else
+            ((yuv2planarX_fn)inst->pfn)(filter, inst->filter_size, (const int16_t**)src, dst[0], dstW, c->lumDither8, 0);
+    }
+
+    return 1;
+}
+
+static int chr_planar_vscale(SwsContext *c, SwsFilterDescriptor *desc, int sliceY, int sliceH)
+{
+    const int chrSkipMask = (1 << desc->dst->v_chr_sub_sample) - 1;
+    if (sliceY & chrSkipMask)
+        return 0;
+    else {
+        VScalerContext *inst = desc->instance;
+        int dstW = FF_CEIL_RSHIFT(desc->dst->width, desc->dst->h_chr_sub_sample);
+        int chrSliceY = sliceY >> desc->dst->v_chr_sub_sample;
+
+        int first = FFMAX(1-inst->filter_size, inst->filter_pos[chrSliceY]);
+        int sp1 = first - desc->src->plane[1].sliceY;
+        int sp2 = first - desc->src->plane[2].sliceY;
+        int dp1 = chrSliceY - desc->dst->plane[1].sliceY;
+        int dp2 = chrSliceY - desc->dst->plane[2].sliceY;
+        uint8_t **src1 = desc->src->plane[1].line + sp1;
+        uint8_t **src2 = desc->src->plane[2].line + sp2;
+        uint8_t **dst1 = desc->dst->plane[1].line + dp1;
+        uint8_t **dst2 = desc->dst->plane[2].line + dp2;
+        uint16_t *filter = inst->filter[0] + (inst->isMMX ? 0 : chrSliceY * inst->filter_size);
+
+        if (c->yuv2nv12cX) {
+            ((yuv2interleavedX_fn)inst->pfn)(c, filter, inst->filter_size, (const int16_t**)src1, (const int16_t**)src2, dst1[0], dstW);
+        } else if (inst->filter_size == 1) {
+            ((yuv2planar1_fn)inst->pfn)((const int16_t*)src1[0], dst1[0], dstW, c->chrDither8, 0);
+            ((yuv2planar1_fn)inst->pfn)((const int16_t*)src2[0], dst2[0], dstW, c->chrDither8, 3);
+        } else {
+            ((yuv2planarX_fn)inst->pfn)(filter, inst->filter_size, (const int16_t**)src1, dst1[0], dstW, c->chrDither8, 0);
+            ((yuv2planarX_fn)inst->pfn)(filter, inst->filter_size, (const int16_t**)src2, dst2[0], dstW, c->chrDither8, inst->isMMX ? (c->uv_offx2 >> 1) : 3);
+        }
+    }
+
+    return 1;
+}
+
+static int packed_vscale(SwsContext *c, SwsFilterDescriptor *desc, int sliceY, int sliceH)
+{
+    VScalerContext *inst = desc->instance;
+    int dstW = desc->dst->width;
+    int chrSliceY = sliceY >> desc->dst->v_chr_sub_sample;
+
+    int lum_fsize = inst[0].filter_size;
+    int chr_fsize = inst[1].filter_size;
+    uint16_t *lum_filter = inst[0].filter[0];
+    uint16_t *chr_filter = inst[1].filter[0];
+
+    int firstLum = FFMAX(1-lum_fsize, inst[0].filter_pos[chrSliceY]);
+    int firstChr = FFMAX(1-chr_fsize, inst[1].filter_pos[chrSliceY]);
+
+    int sp0 = firstLum - desc->src->plane[0].sliceY;
+    int sp1 = firstChr - desc->src->plane[1].sliceY;
+    int sp2 = firstChr - desc->src->plane[2].sliceY;
+    int sp3 = firstLum - desc->src->plane[3].sliceY;
+    int dp = sliceY - desc->dst->plane[0].sliceY;
+    uint8_t **src0 = desc->src->plane[0].line + sp0;
+    uint8_t **src1 = desc->src->plane[1].line + sp1;
+    uint8_t **src2 = desc->src->plane[2].line + sp2;
+    uint8_t **src3 = desc->alpha ? desc->src->plane[3].line + sp3 : NULL;
+    uint8_t **dst = desc->dst->plane[0].line + dp;
+
+
+    if (c->yuv2packed1 && lum_fsize == 1 && chr_fsize <= 2) { // unscaled RGB
+        int chrAlpha = chr_fsize == 1 ? 0 : chr_filter[2 * sliceY + 1];
+        ((yuv2packed1_fn)inst->pfn)(c, (const int16_t*)*src0, (const int16_t**)src1, (const int16_t**)src2, (const int16_t*)(desc->alpha ? *src3 : NULL),  *dst, dstW, chrAlpha, sliceY);
+    } else if (c->yuv2packed2 && lum_fsize == 2 && chr_fsize == 2) { // bilinear upscale RGB
+        int lumAlpha = lum_filter[2 * sliceY + 1];
+        int chrAlpha = chr_filter[2 * sliceY + 1];
+        c->lumMmxFilter[2] =
+        c->lumMmxFilter[3] = lum_filter[2 * sliceY]    * 0x10001;
+        c->chrMmxFilter[2] =
+        c->chrMmxFilter[3] = chr_filter[2 * chrSliceY] * 0x10001;
+        ((yuv2packed2_fn)inst->pfn)(c, (const int16_t**)src0, (const int16_t**)src1, (const int16_t**)src2, (const int16_t**)src3,
+                    *dst, dstW, lumAlpha, chrAlpha, sliceY);
+    } else { // general RGB
+        ((yuv2packedX_fn)inst->pfn)(c, lum_filter + sliceY * lum_fsize,
+                    (const int16_t**)src0, lum_fsize, chr_filter + sliceY * chr_fsize,
+                    (const int16_t**)src1, (const int16_t**)src2, chr_fsize, (const int16_t**)src3, *dst, dstW, sliceY);
+    }
+    return 1;
+}
+
+static int any_vscale(SwsContext *c, SwsFilterDescriptor *desc, int sliceY, int sliceH)
+{
+    VScalerContext *inst = desc->instance;
+    int dstW = desc->dst->width;
+    int chrSliceY = sliceY >> desc->dst->v_chr_sub_sample;
+
+    int lum_fsize = inst[0].filter_size;
+    int chr_fsize = inst[1].filter_size;
+    uint16_t *lum_filter = inst[0].filter[0];
+    uint16_t *chr_filter = inst[1].filter[0];
+
+    int firstLum = FFMAX(1-lum_fsize, inst[0].filter_pos[chrSliceY]);
+    int firstChr = FFMAX(1-chr_fsize, inst[1].filter_pos[chrSliceY]);
+
+    int sp0 = firstLum - desc->src->plane[0].sliceY;
+    int sp1 = firstChr - desc->src->plane[1].sliceY;
+    int sp2 = firstChr - desc->src->plane[2].sliceY;
+    int sp3 = firstLum - desc->src->plane[3].sliceY;
+    int dp0 = sliceY - desc->dst->plane[0].sliceY;
+    int dp1 = chrSliceY - desc->dst->plane[1].sliceY;
+    int dp2 = chrSliceY - desc->dst->plane[2].sliceY;
+    int dp3 = sliceY - desc->dst->plane[3].sliceY;
+
+    uint8_t **src0 = desc->src->plane[0].line + sp0;
+    uint8_t **src1 = desc->src->plane[1].line + sp1;
+    uint8_t **src2 = desc->src->plane[2].line + sp2;
+    uint8_t **src3 = desc->alpha ? desc->src->plane[3].line + sp3 : NULL;
+    uint8_t *dst[4] = { desc->dst->plane[0].line[dp0],
+                        desc->dst->plane[1].line[dp1],
+                        desc->dst->plane[2].line[dp2],
+                        desc->alpha ? desc->dst->plane[3].line[dp3] : NULL };
+
+    av_assert1(!c->yuv2packed1 && !c->yuv2packed2);
+    ((yuv2anyX_fn)inst->pfn)(c, lum_filter + sliceY * lum_fsize,
+             (const int16_t**)src0, lum_fsize, chr_filter + sliceY * chr_fsize,
+             (const int16_t**)src1, (const int16_t**)src2, chr_fsize, (const int16_t**)src3, dst, dstW, sliceY);
+
+    return 1;
+
+}
+
+int ff_init_vscale(SwsContext *c, SwsFilterDescriptor *desc, SwsSlice *src, SwsSlice *dst)
+{
+    VScalerContext *lumCtx = NULL;
+    VScalerContext *chrCtx = NULL;
+
+    if (isPlanarYUV(c->dstFormat) || (isGray(c->dstFormat) && !isALPHA(c->dstFormat))) {
+        lumCtx = av_mallocz(sizeof(VScalerContext));
+        if (!lumCtx)
+            return AVERROR(ENOMEM);
+
+
+        desc[0].process = lum_planar_vscale;
+        desc[0].instance = lumCtx;
+        desc[0].src = src;
+        desc[0].dst = dst;
+        desc[0].alpha = c->alpPixBuf != 0;
+
+        if (!isGray(c->dstFormat)) {
+            chrCtx = av_mallocz(sizeof(VScalerContext));
+            if (!chrCtx)
+                return AVERROR(ENOMEM);
+            desc[1].process = chr_planar_vscale;
+            desc[1].instance = chrCtx;
+            desc[1].src = src;
+            desc[1].dst = dst;
+        }
+    } else {
+        lumCtx = av_mallocz_array(sizeof(VScalerContext), 2);
+        if (!lumCtx)
+            return AVERROR(ENOMEM);
+        chrCtx = &lumCtx[1];
+
+        desc[0].process = c->yuv2packedX ? packed_vscale : any_vscale;
+        desc[0].instance = lumCtx;
+        desc[0].src = src;
+        desc[0].dst = dst;
+        desc[0].alpha = c->alpPixBuf != 0;
+    }
+
+    ff_init_vscale_pfn(c, c->yuv2plane1, c->yuv2planeX, c->yuv2nv12cX,
+        c->yuv2packed1, c->yuv2packed2, c->yuv2packedX, c->yuv2anyX, c->use_mmx_vfilter);
+    return 0;
+}
+
+void ff_init_vscale_pfn(SwsContext *c,
+    yuv2planar1_fn yuv2plane1,
+    yuv2planarX_fn yuv2planeX,
+    yuv2interleavedX_fn yuv2nv12cX,
+    yuv2packed1_fn yuv2packed1,
+    yuv2packed2_fn yuv2packed2,
+    yuv2packedX_fn yuv2packedX,
+    yuv2anyX_fn yuv2anyX, int use_mmx)
+{
+    VScalerContext *lumCtx = NULL;
+    VScalerContext *chrCtx = NULL;
+    int idx = c->numDesc - 1;
+
+    if (isPlanarYUV(c->dstFormat) || (isGray(c->dstFormat) && !isALPHA(c->dstFormat))) {
+        if (!isGray(c->dstFormat)) {
+            chrCtx = c->desc[idx].instance;
+
+            chrCtx->filter[0] = use_mmx ? (int16_t*)c->chrMmxFilter : c->vChrFilter;
+            chrCtx->filter_size = c->vChrFilterSize;
+            chrCtx->filter_pos = c->vChrFilterPos;
+            chrCtx->isMMX = use_mmx;
+
+            --idx;
+            if (yuv2nv12cX)               chrCtx->pfn = yuv2nv12cX;
+            else if (c->vChrFilterSize == 1) chrCtx->pfn = yuv2plane1;
+            else                             chrCtx->pfn = yuv2planeX;
+        }
+
+        lumCtx = c->desc[idx].instance;
+
+        lumCtx->filter[0] = use_mmx ? (int16_t*)c->lumMmxFilter : c->vLumFilter;
+        lumCtx->filter[1] = use_mmx ? (int16_t*)c->alpMmxFilter : c->vLumFilter;
+        lumCtx->filter_size = c->vLumFilterSize;
+        lumCtx->filter_pos = c->vLumFilterPos;
+        lumCtx->isMMX = use_mmx;
+
+        if (c->vLumFilterSize == 1) lumCtx->pfn = yuv2plane1;
+        else                        lumCtx->pfn = yuv2planeX;
+
+    } else {
+        lumCtx = c->desc[idx].instance;
+        chrCtx = &lumCtx[1];
+
+        lumCtx->filter[0] = c->vLumFilter;
+        lumCtx->filter_size = c->vLumFilterSize;
+        lumCtx->filter_pos = c->vLumFilterPos;
+
+        chrCtx->filter[0] = c->vChrFilter;
+        chrCtx->filter_size = c->vChrFilterSize;
+        chrCtx->filter_pos = c->vChrFilterPos;
+
+        lumCtx->isMMX = use_mmx;
+        chrCtx->isMMX = use_mmx;
+
+        if (yuv2packedX) {
+            if (c->yuv2packed1 && c->vLumFilterSize == 1 && c->vChrFilterSize <= 2)
+                lumCtx->pfn = yuv2packed1;
+            else if (c->yuv2packed2 && c->vLumFilterSize == 2 && c->vChrFilterSize == 2)
+                lumCtx->pfn = yuv2packed2;
+            else
+                lumCtx->pfn = yuv2packedX;
+        } else
+            lumCtx->pfn = yuv2anyX;
+    }
+}
+
+
index df29e23..fe5690d 100644 (file)
@@ -86,9 +86,9 @@ void ff_updateMMXDitherTables(SwsContext *c, int dstY, int lumBufIndex, int chrB
     const int dstH= c->dstH;
     const int flags= c->flags;
 #ifdef NEW_FILTER
-    SwsPlane *lumPlane = &c->slice[c->numSlice-1].plane[0];
-    SwsPlane *chrUPlane = &c->slice[c->numSlice-1].plane[1];
-    SwsPlane *alpPlane = &c->slice[c->numSlice-1].plane[3];
+    SwsPlane *lumPlane = &c->slice[c->numSlice-2].plane[0];
+    SwsPlane *chrUPlane = &c->slice[c->numSlice-2].plane[1];
+    SwsPlane *alpPlane = &c->slice[c->numSlice-2].plane[3];
 #else
     int16_t **lumPixBuf= c->lumPixBuf;
     int16_t **chrUPixBuf= c->chrUPixBuf;