Merge commit 'ea031b75fb50c59196ccb654c9d143fb75365da9'

* commit 'ea031b75fb50c59196ccb654c9d143fb75365da9':
  hqx: Move DSP related code to a separate file

Conflicts:
	libavcodec/hqx.c

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

1 files changed, 33 insertions, 136 deletions

diff --git a/libavcodec/hqx.c b/libavcodec/hqx.c
index 2a77444d81..8c568f2439 100644
--- a/libavcodec/hqx.c
+++ b/libavcodec/hqx.c
@@ -28,6 +28,7 @@
 #include "internal.h"
 
 #include "hqx.h"
+#include "hqxdsp.h"
 
 /* HQX has four modes - 422, 444, 422alpha and 444alpha - all 12-bit */
 enum HQXFormat {
@@ -81,115 +82,7 @@ static const uint8_t hqx_quant_chroma[64] = {
     44,  91,  96, 197, 203, 209, 232, 246,
 };
 
-static inline void idct_col(int16_t *blk, const uint8_t *quant)
-{
-    int t0, t1, t2, t3, t4, t5, t6, t7, t8, t9, tA, tB, tC, tD, tE, tF;
-    int t10, t11, t12, t13;
-    int s0, s1, s2, s3, s4, s5, s6, s7;
-
-    s0 = (int) blk[0 * 8] * quant[0 * 8];
-    s1 = (int) blk[1 * 8] * quant[1 * 8];
-    s2 = (int) blk[2 * 8] * quant[2 * 8];
-    s3 = (int) blk[3 * 8] * quant[3 * 8];
-    s4 = (int) blk[4 * 8] * quant[4 * 8];
-    s5 = (int) blk[5 * 8] * quant[5 * 8];
-    s6 = (int) blk[6 * 8] * quant[6 * 8];
-    s7 = (int) blk[7 * 8] * quant[7 * 8];
-
-    t0  =  (s3 * 19266 + s5 * 12873) >> 15;
-    t1  =  (s5 * 19266 - s3 * 12873) >> 15;
-    t2  = ((s7 * 4520  + s1 * 22725) >> 15) - t0;
-    t3  = ((s1 * 4520  - s7 * 22725) >> 15) - t1;
-    t4  = t0 * 2 + t2;
-    t5  = t1 * 2 + t3;
-    t6  = t2 - t3;
-    t7  = t3 * 2 + t6;
-    t8  = (t6 * 11585) >> 14;
-    t9  = (t7 * 11585) >> 14;
-    tA  = (s2 * 8867 - s6 * 21407) >> 14;
-    tB  = (s6 * 8867 + s2 * 21407) >> 14;
-    tC  = (s0 >> 1) - (s4 >> 1);
-    tD  = (s4 >> 1) * 2 + tC;
-    tE  = tC - (tA >> 1);
-    tF  = tD - (tB >> 1);
-    t10 = tF - t5;
-    t11 = tE - t8;
-    t12 = tE + (tA >> 1) * 2 - t9;
-    t13 = tF + (tB >> 1) * 2 - t4;
-
-    blk[0 * 8] = t13 + t4 * 2;
-    blk[1 * 8] = t12 + t9 * 2;
-    blk[2 * 8] = t11 + t8 * 2;
-    blk[3 * 8] = t10 + t5 * 2;
-    blk[4 * 8] = t10;
-    blk[5 * 8] = t11;
-    blk[6 * 8] = t12;
-    blk[7 * 8] = t13;
-}
-
-static inline void idct_row(int16_t *blk)
-{
-    int t0, t1, t2, t3, t4, t5, t6, t7, t8, t9, tA, tB, tC, tD, tE, tF;
-    int t10, t11, t12, t13;
-
-    t0  =  (blk[3] * 19266 + blk[5] * 12873) >> 14;
-    t1  =  (blk[5] * 19266 - blk[3] * 12873) >> 14;
-    t2  = ((blk[7] * 4520  + blk[1] * 22725) >> 14) - t0;
-    t3  = ((blk[1] * 4520  - blk[7] * 22725) >> 14) - t1;
-    t4  = t0 * 2 + t2;
-    t5  = t1 * 2 + t3;
-    t6  = t2 - t3;
-    t7  = t3 * 2 + t6;
-    t8  = (t6 * 11585) >> 14;
-    t9  = (t7 * 11585) >> 14;
-    tA  = (blk[2] * 8867 - blk[6] * 21407) >> 14;
-    tB  = (blk[6] * 8867 + blk[2] * 21407) >> 14;
-    tC  = blk[0] - blk[4];
-    tD  = blk[4] * 2 + tC;
-    tE  = tC - tA;
-    tF  = tD - tB;
-    t10 = tF - t5;
-    t11 = tE - t8;
-    t12 = tE + tA * 2 - t9;
-    t13 = tF + tB * 2 - t4;
-
-    blk[0] = (t13 + t4 * 2 + 4) >> 3;
-    blk[1] = (t12 + t9 * 2 + 4) >> 3;
-    blk[2] = (t11 + t8 * 2 + 4) >> 3;
-    blk[3] = (t10 + t5 * 2 + 4) >> 3;
-    blk[4] = (t10          + 4) >> 3;
-    blk[5] = (t11          + 4) >> 3;
-    blk[6] = (t12          + 4) >> 3;
-    blk[7] = (t13          + 4) >> 3;
-}
-
-static void hqx_idct(int16_t *block, const uint8_t *quant)
-{
-    int i;
-
-    for (i = 0; i < 8; i++)
-        idct_col(block + i, quant + i);
-    for (i = 0; i < 8; i++)
-        idct_row(block + i * 8);
-}
-
-static void hqx_idct_put(uint16_t *dst, ptrdiff_t stride,
-                         int16_t *block, const uint8_t *quant)
-{
-    int i, j;
-
-    hqx_idct(block, quant);
-
-    for (i = 0; i < 8; i++) {
-        for (j = 0; j < 8; j++) {
-            int v = av_clip_uintp2(block[j + i * 8] + 0x800, 12);
-            dst[j] = (v << 4) | (v >> 8);
-        }
-        dst += stride >> 1;
-    }
-}
-
-static inline void put_blocks(AVFrame *pic, int plane,
+static inline void put_blocks(HQXContext *ctx, AVFrame *pic, int plane,
                               int x, int y, int ilace,
                               int16_t *block0, int16_t *block1,
                               const uint8_t *quant)
@@ -198,9 +91,10 @@ static inline void put_blocks(AVFrame *pic, int plane,
     int lsize = pic->linesize[plane];
     uint8_t *p = pic->data[plane] + x * 2;
 
-    hqx_idct_put((uint16_t *)(p + y * lsize), lsize * fields, block0, quant);
-    hqx_idct_put((uint16_t *)(p + (y + (ilace ? 1 : 8)) * lsize),
-                 lsize * fields, block1, quant);
+    ctx->hqxdsp.idct_put((uint16_t *)(p + y * lsize),
+                         lsize * fields, block0, quant);
+    ctx->hqxdsp.idct_put((uint16_t *)(p + (y + (ilace ? 1 : 8)) * lsize),
+                         lsize * fields, block1, quant);
 }
 
 static inline void hqx_get_ac(GetBitContext *gb, const HQXAC *ac,
@@ -285,10 +179,10 @@ static int hqx_decode_422(HQXContext *ctx, HQXSliceData * slice_data, AVFrame *p
             return ret;
     }
 
-    put_blocks(pic, 0, x     , y, flag, slice_data->block[0], slice_data->block[2], hqx_quant_luma);
-    put_blocks(pic, 0, x + 8 , y, flag, slice_data->block[1], slice_data->block[3], hqx_quant_luma);
-    put_blocks(pic, 2, x >> 1, y, flag, slice_data->block[4], slice_data->block[5], hqx_quant_chroma);
-    put_blocks(pic, 1, x >> 1, y, flag, slice_data->block[6], slice_data->block[7], hqx_quant_chroma);
+    put_blocks(ctx, pic, 0, x     , y, flag, slice_data->block[0], slice_data->block[2], hqx_quant_luma);
+    put_blocks(ctx, pic, 0, x + 8 , y, flag, slice_data->block[1], slice_data->block[3], hqx_quant_luma);
+    put_blocks(ctx, pic, 2, x >> 1, y, flag, slice_data->block[4], slice_data->block[5], hqx_quant_chroma);
+    put_blocks(ctx, pic, 1, x >> 1, y, flag, slice_data->block[6], slice_data->block[7], hqx_quant_chroma);
 
     return 0;
 }
@@ -332,12 +226,12 @@ static int hqx_decode_422a(HQXContext *ctx, HQXSliceData * slice_data, AVFrame *
         }
     }
 
-    put_blocks(pic, 3, x,      y, flag, slice_data->block[ 0], slice_data->block[ 2], hqx_quant_luma);
-    put_blocks(pic, 3, x + 8,  y, flag, slice_data->block[ 1], slice_data->block[ 3], hqx_quant_luma);
-    put_blocks(pic, 0, x,      y, flag, slice_data->block[ 4], slice_data->block[ 6], hqx_quant_luma);
-    put_blocks(pic, 0, x + 8,  y, flag, slice_data->block[ 5], slice_data->block[ 7], hqx_quant_luma);
-    put_blocks(pic, 2, x >> 1, y, flag, slice_data->block[ 8], slice_data->block[ 9], hqx_quant_chroma);
-    put_blocks(pic, 1, x >> 1, y, flag, slice_data->block[10], slice_data->block[11], hqx_quant_chroma);
+    put_blocks(ctx, pic, 3, x,      y, flag, slice_data->block[ 0], slice_data->block[ 2], hqx_quant_luma);
+    put_blocks(ctx, pic, 3, x + 8,  y, flag, slice_data->block[ 1], slice_data->block[ 3], hqx_quant_luma);
+    put_blocks(ctx, pic, 0, x,      y, flag, slice_data->block[ 4], slice_data->block[ 6], hqx_quant_luma);
+    put_blocks(ctx, pic, 0, x + 8,  y, flag, slice_data->block[ 5], slice_data->block[ 7], hqx_quant_luma);
+    put_blocks(ctx, pic, 2, x >> 1, y, flag, slice_data->block[ 8], slice_data->block[ 9], hqx_quant_chroma);
+    put_blocks(ctx, pic, 1, x >> 1, y, flag, slice_data->block[10], slice_data->block[11], hqx_quant_chroma);
 
     return 0;
 }
@@ -367,12 +261,12 @@ static int hqx_decode_444(HQXContext *ctx, HQXSliceData * slice_data, AVFrame *p
             return ret;
     }
 
-    put_blocks(pic, 0, x,     y, flag, slice_data->block[0], slice_data->block[ 2], hqx_quant_luma);
-    put_blocks(pic, 0, x + 8, y, flag, slice_data->block[1], slice_data->block[ 3], hqx_quant_luma);
-    put_blocks(pic, 2, x,     y, flag, slice_data->block[4], slice_data->block[ 6], hqx_quant_chroma);
-    put_blocks(pic, 2, x + 8, y, flag, slice_data->block[5], slice_data->block[ 7], hqx_quant_chroma);
-    put_blocks(pic, 1, x,     y, flag, slice_data->block[8], slice_data->block[10], hqx_quant_chroma);
-    put_blocks(pic, 1, x + 8, y, flag, slice_data->block[9], slice_data->block[11], hqx_quant_chroma);
+    put_blocks(ctx, pic, 0, x,     y, flag, slice_data->block[0], slice_data->block[ 2], hqx_quant_luma);
+    put_blocks(ctx, pic, 0, x + 8, y, flag, slice_data->block[1], slice_data->block[ 3], hqx_quant_luma);
+    put_blocks(ctx, pic, 2, x,     y, flag, slice_data->block[4], slice_data->block[ 6], hqx_quant_chroma);
+    put_blocks(ctx, pic, 2, x + 8, y, flag, slice_data->block[5], slice_data->block[ 7], hqx_quant_chroma);
+    put_blocks(ctx, pic, 1, x,     y, flag, slice_data->block[8], slice_data->block[10], hqx_quant_chroma);
+    put_blocks(ctx, pic, 1, x + 8, y, flag, slice_data->block[9], slice_data->block[11], hqx_quant_chroma);
 
     return 0;
 }
@@ -413,14 +307,14 @@ static int hqx_decode_444a(HQXContext *ctx, HQXSliceData * slice_data, AVFrame *
         }
     }
 
-    put_blocks(pic, 3, x,     y, flag, slice_data->block[ 0], slice_data->block[ 2], hqx_quant_luma);
-    put_blocks(pic, 3, x + 8, y, flag, slice_data->block[ 1], slice_data->block[ 3], hqx_quant_luma);
-    put_blocks(pic, 0, x,     y, flag, slice_data->block[ 4], slice_data->block[ 6], hqx_quant_luma);
-    put_blocks(pic, 0, x + 8, y, flag, slice_data->block[ 5], slice_data->block[ 7], hqx_quant_luma);
-    put_blocks(pic, 2, x,     y, flag, slice_data->block[ 8], slice_data->block[10], hqx_quant_chroma);
-    put_blocks(pic, 2, x + 8, y, flag, slice_data->block[ 9], slice_data->block[11], hqx_quant_chroma);
-    put_blocks(pic, 1, x,     y, flag, slice_data->block[12], slice_data->block[14], hqx_quant_chroma);
-    put_blocks(pic, 1, x + 8, y, flag, slice_data->block[13], slice_data->block[15], hqx_quant_chroma);
+    put_blocks(ctx, pic, 3, x,     y, flag, slice_data->block[ 0], slice_data->block[ 2], hqx_quant_luma);
+    put_blocks(ctx, pic, 3, x + 8, y, flag, slice_data->block[ 1], slice_data->block[ 3], hqx_quant_luma);
+    put_blocks(ctx, pic, 0, x,     y, flag, slice_data->block[ 4], slice_data->block[ 6], hqx_quant_luma);
+    put_blocks(ctx, pic, 0, x + 8, y, flag, slice_data->block[ 5], slice_data->block[ 7], hqx_quant_luma);
+    put_blocks(ctx, pic, 2, x,     y, flag, slice_data->block[ 8], slice_data->block[10], hqx_quant_chroma);
+    put_blocks(ctx, pic, 2, x + 8, y, flag, slice_data->block[ 9], slice_data->block[11], hqx_quant_chroma);
+    put_blocks(ctx, pic, 1, x,     y, flag, slice_data->block[12], slice_data->block[14], hqx_quant_chroma);
+    put_blocks(ctx, pic, 1, x + 8, y, flag, slice_data->block[13], slice_data->block[15], hqx_quant_chroma);
 
     return 0;
 }
@@ -646,6 +540,9 @@ static av_cold int hqx_decode_init(AVCodecContext *avctx)
     int ret = ff_hqx_init_vlcs(ctx);
     if (ret < 0)
         hqx_decode_close(avctx);
+
+    ff_hqxdsp_init(&ctx->hqxdsp);
+
     return ret;
 }