Merge commit '9b30f8dd8fa5bef5f16904cb98745b4a58f8f776'

* commit '9b30f8dd8fa5bef5f16904cb98745b4a58f8f776':
  h264: remove the svq3-specific code

Merged-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>

3 files changed, 86 insertions, 127 deletions

diff --git a/libavcodec/h264.c b/libavcodec/h264.c
index a85953ed46..5a0f0bb0f4 100644
--- a/libavcodec/h264.c
+++ b/libavcodec/h264.c
@@ -48,7 +48,6 @@
 #include "mpegutils.h"
 #include "profiles.h"
 #include "rectangle.h"
-#include "svq3.h"
 #include "thread.h"
 #include "vdpau_compat.h"
 
diff --git a/libavcodec/h264_mb.c b/libavcodec/h264_mb.c
index 8791e2e182..3c5c93207d 100644
--- a/libavcodec/h264_mb.c
+++ b/libavcodec/h264_mb.c
@@ -33,7 +33,6 @@
 #include "avcodec.h"
 #include "h264.h"
 #include "qpeldsp.h"
-#include "svq3.h"
 #include "thread.h"
 
 static inline int get_lowest_part_list_y(H264SliceContext *sl,
@@ -612,8 +611,7 @@ static av_always_inline void dctcoef_set(int16_t *mb, int high_bit_depth,
 
 static av_always_inline void hl_decode_mb_predict_luma(const H264Context *h,
                                                        H264SliceContext *sl,
-                                                       int mb_type, int is_h264,
-                                                       int simple,
+                                                       int mb_type, int simple,
                                                        int transform_bypass,
                                                        int pixel_shift,
                                                        const int *block_offset,
@@ -693,48 +691,40 @@ static av_always_inline void hl_decode_mb_predict_luma(const H264Context *h,
                     h->hpc.pred4x4[dir](ptr, topright, linesize);
                     nnz = sl->non_zero_count_cache[scan8[i + p * 16]];
                     if (nnz) {
-                        if (is_h264) {
-                            if (nnz == 1 && dctcoef_get(sl->mb, pixel_shift, i * 16 + p * 256))
-                                idct_dc_add(ptr, sl->mb + (i * 16 + p * 256 << pixel_shift), linesize);
-                            else
-                                idct_add(ptr, sl->mb + (i * 16 + p * 256 << pixel_shift), linesize);
-                        } else if (CONFIG_SVQ3_DECODER)
-                            ff_svq3_add_idct_c(ptr, sl->mb + i * 16 + p * 256, linesize, qscale, 0);
+                        if (nnz == 1 && dctcoef_get(sl->mb, pixel_shift, i * 16 + p * 256))
+                            idct_dc_add(ptr, sl->mb + (i * 16 + p * 256 << pixel_shift), linesize);
+                        else
+                            idct_add(ptr, sl->mb + (i * 16 + p * 256 << pixel_shift), linesize);
                     }
                 }
             }
         }
     } else {
         h->hpc.pred16x16[sl->intra16x16_pred_mode](dest_y, linesize);
-        if (is_h264) {
-            if (sl->non_zero_count_cache[scan8[LUMA_DC_BLOCK_INDEX + p]]) {
-                if (!transform_bypass)
-                    h->h264dsp.h264_luma_dc_dequant_idct(sl->mb + (p * 256 << pixel_shift),
-                                                         sl->mb_luma_dc[p],
-                                                         h->dequant4_coeff[p][qscale][0]);
-                else {
-                    static const uint8_t dc_mapping[16] = {
-                         0 * 16,  1 * 16,  4 * 16,  5 * 16,
-                         2 * 16,  3 * 16,  6 * 16,  7 * 16,
-                         8 * 16,  9 * 16, 12 * 16, 13 * 16,
-                        10 * 16, 11 * 16, 14 * 16, 15 * 16
-                    };
-                    for (i = 0; i < 16; i++)
-                        dctcoef_set(sl->mb + (p * 256 << pixel_shift),
-                                    pixel_shift, dc_mapping[i],
-                                    dctcoef_get(sl->mb_luma_dc[p],
-                                                pixel_shift, i));
-                }
+        if (sl->non_zero_count_cache[scan8[LUMA_DC_BLOCK_INDEX + p]]) {
+            if (!transform_bypass)
+                h->h264dsp.h264_luma_dc_dequant_idct(sl->mb + (p * 256 << pixel_shift),
+                                                     sl->mb_luma_dc[p],
+                                                     h->dequant4_coeff[p][qscale][0]);
+            else {
+                static const uint8_t dc_mapping[16] = {
+                     0 * 16,  1 * 16,  4 * 16,  5 * 16,
+                     2 * 16,  3 * 16,  6 * 16,  7 * 16,
+                     8 * 16,  9 * 16, 12 * 16, 13 * 16,
+                    10 * 16, 11 * 16, 14 * 16, 15 * 16
+                };
+                for (i = 0; i < 16; i++)
+                    dctcoef_set(sl->mb + (p * 256 << pixel_shift),
+                                pixel_shift, dc_mapping[i],
+                                dctcoef_get(sl->mb_luma_dc[p],
+                                            pixel_shift, i));
             }
-        } else if (CONFIG_SVQ3_DECODER)
-            ff_svq3_luma_dc_dequant_idct_c(sl->mb + p * 256,
-                                           sl->mb_luma_dc[p], qscale);
+        }
     }
 }
 
 static av_always_inline void hl_decode_mb_idct_luma(const H264Context *h, H264SliceContext *sl,
-                                                    int mb_type,
-                                                    int is_h264, int simple,
+                                                    int mb_type, int simple,
                                                     int transform_bypass,
                                                     int pixel_shift,
                                                     const int *block_offset,
@@ -745,60 +735,50 @@ static av_always_inline void hl_decode_mb_idct_luma(const H264Context *h, H264Sl
     int i;
     block_offset += 16 * p;
     if (!IS_INTRA4x4(mb_type)) {
-        if (is_h264) {
-            if (IS_INTRA16x16(mb_type)) {
-                if (transform_bypass) {
-                    if (h->sps.profile_idc == 244 &&
-                        (sl->intra16x16_pred_mode == VERT_PRED8x8 ||
-                         sl->intra16x16_pred_mode == HOR_PRED8x8)) {
-                        h->hpc.pred16x16_add[sl->intra16x16_pred_mode](dest_y, block_offset,
-                                                                      sl->mb + (p * 256 << pixel_shift),
-                                                                      linesize);
-                    } else {
-                        for (i = 0; i < 16; i++)
-                            if (sl->non_zero_count_cache[scan8[i + p * 16]] ||
-                                dctcoef_get(sl->mb, pixel_shift, i * 16 + p * 256))
-                                h->h264dsp.h264_add_pixels4_clear(dest_y + block_offset[i],
-                                                                  sl->mb + (i * 16 + p * 256 << pixel_shift),
-                                                                  linesize);
-                    }
-                } else {
-                    h->h264dsp.h264_idct_add16intra(dest_y, block_offset,
-                                                    sl->mb + (p * 256 << pixel_shift),
-                                                    linesize,
-                                                    sl->non_zero_count_cache + p * 5 * 8);
-                }
-            } else if (sl->cbp & 15) {
-                if (transform_bypass) {
-                    const int di = IS_8x8DCT(mb_type) ? 4 : 1;
-                    idct_add = IS_8x8DCT(mb_type) ? h->h264dsp.h264_add_pixels8_clear
-                                                  : h->h264dsp.h264_add_pixels4_clear;
-                    for (i = 0; i < 16; i += di)
-                        if (sl->non_zero_count_cache[scan8[i + p * 16]])
-                            idct_add(dest_y + block_offset[i],
-                                     sl->mb + (i * 16 + p * 256 << pixel_shift),
-                                     linesize);
+        if (IS_INTRA16x16(mb_type)) {
+            if (transform_bypass) {
+                if (h->sps.profile_idc == 244 &&
+                    (sl->intra16x16_pred_mode == VERT_PRED8x8 ||
+                     sl->intra16x16_pred_mode == HOR_PRED8x8)) {
+                    h->hpc.pred16x16_add[sl->intra16x16_pred_mode](dest_y, block_offset,
+                                                                   sl->mb + (p * 256 << pixel_shift),
+                                                                   linesize);
                 } else {
-                    if (IS_8x8DCT(mb_type))
-                        h->h264dsp.h264_idct8_add4(dest_y, block_offset,
-                                                   sl->mb + (p * 256 << pixel_shift),
-                                                   linesize,
-                                                   sl->non_zero_count_cache + p * 5 * 8);
-                    else
-                        h->h264dsp.h264_idct_add16(dest_y, block_offset,
-                                                   sl->mb + (p * 256 << pixel_shift),
-                                                   linesize,
-                                                   sl->non_zero_count_cache + p * 5 * 8);
+                    for (i = 0; i < 16; i++)
+                        if (sl->non_zero_count_cache[scan8[i + p * 16]] ||
+                            dctcoef_get(sl->mb, pixel_shift, i * 16 + p * 256))
+                            h->h264dsp.h264_add_pixels4_clear(dest_y + block_offset[i],
+                                                              sl->mb + (i * 16 + p * 256 << pixel_shift),
+                                                              linesize);
                 }
+            } else {
+                h->h264dsp.h264_idct_add16intra(dest_y, block_offset,
+                                                sl->mb + (p * 256 << pixel_shift),
+                                                linesize,
+                                                sl->non_zero_count_cache + p * 5 * 8);
+            }
+        } else if (sl->cbp & 15) {
+            if (transform_bypass) {
+                const int di = IS_8x8DCT(mb_type) ? 4 : 1;
+                idct_add = IS_8x8DCT(mb_type) ? h->h264dsp.h264_add_pixels8_clear
+                    : h->h264dsp.h264_add_pixels4_clear;
+                for (i = 0; i < 16; i += di)
+                    if (sl->non_zero_count_cache[scan8[i + p * 16]])
+                        idct_add(dest_y + block_offset[i],
+                                 sl->mb + (i * 16 + p * 256 << pixel_shift),
+                                 linesize);
+            } else {
+                if (IS_8x8DCT(mb_type))
+                    h->h264dsp.h264_idct8_add4(dest_y, block_offset,
+                                               sl->mb + (p * 256 << pixel_shift),
+                                               linesize,
+                                               sl->non_zero_count_cache + p * 5 * 8);
+                else
+                    h->h264dsp.h264_idct_add16(dest_y, block_offset,
+                                               sl->mb + (p * 256 << pixel_shift),
+                                               linesize,
+                                               sl->non_zero_count_cache + p * 5 * 8);
             }
-        } else if (CONFIG_SVQ3_DECODER) {
-            for (i = 0; i < 16; i++)
-                if (sl->non_zero_count_cache[scan8[i + p * 16]] || sl->mb[i * 16 + p * 256]) {
-                    // FIXME benchmark weird rule, & below
-                    uint8_t *const ptr = dest_y + block_offset[i];
-                    ff_svq3_add_idct_c(ptr, sl->mb + i * 16 + p * 256, linesize,
-                                       sl->qscale, IS_INTRA(mb_type) ? 1 : 0);
-                }
         }
     }
 }
diff --git a/libavcodec/h264_mb_template.c b/libavcodec/h264_mb_template.c
index 54420b9022..f7f8a93fd5 100644
--- a/libavcodec/h264_mb_template.c
+++ b/libavcodec/h264_mb_template.c
@@ -19,8 +19,6 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
-#include "svq3.h"
-
 #undef FUNC
 #undef PIXEL_SHIFT
 
@@ -51,8 +49,6 @@ static av_noinline void FUNC(hl_decode_mb)(const H264Context *h, H264SliceContex
     int i, j;
     const int *block_offset = &h->block_offset[0];
     const int transform_bypass = !SIMPLE && (sl->qscale == 0 && h->sps.transform_bypass);
-    /* is_h264 should always be true if SVQ3 is disabled. */
-    const int is_h264 = !CONFIG_SVQ3_DECODER || SIMPLE || h->avctx->codec_id == AV_CODEC_ID_H264;
     void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
     const int block_h   = 16 >> h->chroma_y_shift;
     const int chroma422 = CHROMA422(h);
@@ -164,14 +160,14 @@ static av_noinline void FUNC(hl_decode_mb)(const H264Context *h, H264SliceContex
                 h->hpc.pred8x8[sl->chroma_pred_mode](dest_cr, uvlinesize);
             }
 
-            hl_decode_mb_predict_luma(h, sl, mb_type, is_h264, SIMPLE,
+            hl_decode_mb_predict_luma(h, sl, mb_type, SIMPLE,
                                       transform_bypass, PIXEL_SHIFT,
                                       block_offset, linesize, dest_y, 0);
 
             if (sl->deblocking_filter)
                 xchg_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
                                uvlinesize, 0, 0, SIMPLE, PIXEL_SHIFT);
-        } else if (is_h264) {
+        } else {
             if (chroma422) {
                 FUNC(hl_motion_422)(h, sl, dest_y, dest_cb, dest_cr,
                               h->qpel_put, h->h264chroma.put_h264_chroma_pixels_tab,
@@ -187,7 +183,7 @@ static av_noinline void FUNC(hl_decode_mb)(const H264Context *h, H264SliceContex
             }
         }
 
-        hl_decode_mb_idct_luma(h, sl, mb_type, is_h264, SIMPLE, transform_bypass,
+        hl_decode_mb_idct_luma(h, sl, mb_type, SIMPLE, transform_bypass,
                                PIXEL_SHIFT, block_offset, linesize, dest_y, 0);
 
         if ((SIMPLE || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) &&
@@ -225,39 +221,23 @@ static av_noinline void FUNC(hl_decode_mb)(const H264Context *h, H264SliceContex
                     }
                 }
             } else {
-                if (is_h264) {
-                    int qp[2];
-                    if (chroma422) {
-                        qp[0] = sl->chroma_qp[0] + 3;
-                        qp[1] = sl->chroma_qp[1] + 3;
-                    } else {
-                        qp[0] = sl->chroma_qp[0];
-                        qp[1] = sl->chroma_qp[1];
-                    }
-                    if (sl->non_zero_count_cache[scan8[CHROMA_DC_BLOCK_INDEX + 0]])
-                        h->h264dsp.h264_chroma_dc_dequant_idct(sl->mb + (16 * 16 * 1 << PIXEL_SHIFT),
-                                                               h->dequant4_coeff[IS_INTRA(mb_type) ? 1 : 4][qp[0]][0]);
-                    if (sl->non_zero_count_cache[scan8[CHROMA_DC_BLOCK_INDEX + 1]])
-                        h->h264dsp.h264_chroma_dc_dequant_idct(sl->mb + (16 * 16 * 2 << PIXEL_SHIFT),
-                                                               h->dequant4_coeff[IS_INTRA(mb_type) ? 2 : 5][qp[1]][0]);
-                    h->h264dsp.h264_idct_add8(dest, block_offset,
-                                              sl->mb, uvlinesize,
-                                              sl->non_zero_count_cache);
-                } else if (CONFIG_SVQ3_DECODER) {
-                    h->h264dsp.h264_chroma_dc_dequant_idct(sl->mb + 16 * 16 * 1,
-                                                           h->dequant4_coeff[IS_INTRA(mb_type) ? 1 : 4][sl->chroma_qp[0]][0]);
-                    h->h264dsp.h264_chroma_dc_dequant_idct(sl->mb + 16 * 16 * 2,
-                                                           h->dequant4_coeff[IS_INTRA(mb_type) ? 2 : 5][sl->chroma_qp[1]][0]);
-                    for (j = 1; j < 3; j++) {
-                        for (i = j * 16; i < j * 16 + 4; i++)
-                            if (sl->non_zero_count_cache[scan8[i]] || sl->mb[i * 16]) {
-                                uint8_t *const ptr = dest[j - 1] + block_offset[i];
-                                ff_svq3_add_idct_c(ptr, sl->mb + i * 16,
-                                                   uvlinesize,
-                                                   ff_h264_chroma_qp[0][sl->qscale + 12] - 12, 2);
-                            }
-                    }
+                int qp[2];
+                if (chroma422) {
+                    qp[0] = sl->chroma_qp[0] + 3;
+                    qp[1] = sl->chroma_qp[1] + 3;
+                } else {
+                    qp[0] = sl->chroma_qp[0];
+                    qp[1] = sl->chroma_qp[1];
                 }
+                if (sl->non_zero_count_cache[scan8[CHROMA_DC_BLOCK_INDEX + 0]])
+                    h->h264dsp.h264_chroma_dc_dequant_idct(sl->mb + (16 * 16 * 1 << PIXEL_SHIFT),
+                                                           h->dequant4_coeff[IS_INTRA(mb_type) ? 1 : 4][qp[0]][0]);
+                if (sl->non_zero_count_cache[scan8[CHROMA_DC_BLOCK_INDEX + 1]])
+                    h->h264dsp.h264_chroma_dc_dequant_idct(sl->mb + (16 * 16 * 2 << PIXEL_SHIFT),
+                                                           h->dequant4_coeff[IS_INTRA(mb_type) ? 2 : 5][qp[1]][0]);
+                h->h264dsp.h264_idct_add8(dest, block_offset,
+                                          sl->mb, uvlinesize,
+                                          sl->non_zero_count_cache);
             }
         }
     }
@@ -344,7 +324,7 @@ static av_noinline void FUNC(hl_decode_mb_444)(const H264Context *h, H264SliceCo
                                linesize, 1, 1, SIMPLE, PIXEL_SHIFT);
 
             for (p = 0; p < plane_count; p++)
-                hl_decode_mb_predict_luma(h, sl, mb_type, 1, SIMPLE,
+                hl_decode_mb_predict_luma(h, sl, mb_type, SIMPLE,
                                           transform_bypass, PIXEL_SHIFT,
                                           block_offset, linesize, dest[p], p);
 
@@ -360,7 +340,7 @@ static av_noinline void FUNC(hl_decode_mb_444)(const H264Context *h, H264SliceCo
         }
 
         for (p = 0; p < plane_count; p++)
-            hl_decode_mb_idct_luma(h, sl, mb_type, 1, SIMPLE, transform_bypass,
+            hl_decode_mb_idct_luma(h, sl, mb_type, SIMPLE, transform_bypass,
                                    PIXEL_SHIFT, block_offset, linesize,
                                    dest[p], p);
     }