Merge remote-tracking branch 'qatar/master'

* qatar/master:
  Remove h264_lowres_idct_put/add functions
  Remove snow/dwt test program
  h264: remove some disabled code
  Fix incorrect max_lowres values
  matroskadec: fix integer underflow if header length < probe length.
  cosmetics: indentation
  eac3enc: use frame exponent strategy when applicable.
  cosmetics: rename eac3dec_data.c/h to eac3_data.c/h since the tables will also be used in the E-AC-3 encoder.
  dsputil: fix ff_check_alignment()

Conflicts:
	libavcodec/Makefile
	libavcodec/h264idct_template.c

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

13 files changed, 149 insertions, 237 deletions

diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index d55baf2421..8b3d35ab38 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -127,9 +127,9 @@ OBJS-$(CONFIG_DVDSUB_ENCODER)          += dvdsubenc.o
 OBJS-$(CONFIG_DVVIDEO_DECODER)         += dv.o dvdata.o
 OBJS-$(CONFIG_DVVIDEO_ENCODER)         += dv.o dvdata.o
 OBJS-$(CONFIG_DXA_DECODER)             += dxa.o
-OBJS-$(CONFIG_EAC3_DECODER)            += eac3dec.o eac3dec_data.o
+OBJS-$(CONFIG_EAC3_DECODER)            += eac3dec.o eac3_data.o
 OBJS-$(CONFIG_EAC3_ENCODER)            += eac3enc.o ac3enc.o ac3enc_float.o \
-                                          ac3tab.o ac3.o kbdwin.o
+                                          ac3tab.o ac3.o kbdwin.o eac3_data.o
 OBJS-$(CONFIG_EACMV_DECODER)           += eacmv.o
 OBJS-$(CONFIG_EAMAD_DECODER)           += eamad.o eaidct.o mpeg12.o \
                                           mpeg12data.o mpegvideo.o  \
diff --git a/libavcodec/ac3enc.c b/libavcodec/ac3enc.c
index c90554e689..f45ed3c89b 100644
--- a/libavcodec/ac3enc.c
+++ b/libavcodec/ac3enc.c
@@ -194,6 +194,7 @@ void ff_ac3_compute_coupling_strategy(AC3EncodeContext *s)
 {
     int blk, ch;
     int got_cpl_snr;
+    int num_cpl_blocks;
 
     /* set coupling use flags for each block/channel */
     /* TODO: turn coupling on/off and adjust start band based on bit usage */
@@ -206,12 +207,14 @@ void ff_ac3_compute_coupling_strategy(AC3EncodeContext *s)
     /* enable coupling for each block if at least 2 channels have coupling
        enabled for that block */
     got_cpl_snr = 0;
+    num_cpl_blocks = 0;
     for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
         AC3Block *block = &s->blocks[blk];
         block->num_cpl_channels = 0;
         for (ch = 1; ch <= s->fbw_channels; ch++)
             block->num_cpl_channels += block->channel_in_cpl[ch];
         block->cpl_in_use = block->num_cpl_channels > 1;
+        num_cpl_blocks += block->cpl_in_use;
         if (!block->cpl_in_use) {
             block->num_cpl_channels = 0;
             for (ch = 1; ch <= s->fbw_channels; ch++)
@@ -237,6 +240,8 @@ void ff_ac3_compute_coupling_strategy(AC3EncodeContext *s)
             block->new_snr_offsets = 0;
         }
     }
+    if (!num_cpl_blocks)
+        s->cpl_on = 0;
 
     /* set bandwidth for each channel */
     for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) {
@@ -301,6 +306,9 @@ static av_cold void exponent_init(AC3EncodeContext *s)
     }
     /* LFE */
     exponent_group_tab[0][0][7] = 2;
+
+    if (CONFIG_EAC3_ENCODER && s->eac3)
+        ff_eac3_exponent_init();
 }
 
 
@@ -342,8 +350,15 @@ static void compute_exp_strategy(AC3EncodeContext *s)
         exp_strategy[0] = EXP_NEW;
         exp += AC3_MAX_COEFS;
         for (blk = 1; blk < AC3_MAX_BLOCKS; blk++, exp += AC3_MAX_COEFS) {
-            if ((ch == CPL_CH && (!s->blocks[blk].cpl_in_use || !s->blocks[blk-1].cpl_in_use)) ||
-                (ch  > CPL_CH && (s->blocks[blk].channel_in_cpl[ch] != s->blocks[blk-1].channel_in_cpl[ch]))) {
+            if (ch == CPL_CH) {
+                if (!s->blocks[blk-1].cpl_in_use) {
+                    exp_strategy[blk] = EXP_NEW;
+                    continue;
+                } else if (!s->blocks[blk].cpl_in_use) {
+                    exp_strategy[blk] = EXP_REUSE;
+                    continue;
+                }
+            } else if (s->blocks[blk].channel_in_cpl[ch] != s->blocks[blk-1].channel_in_cpl[ch]) {
                 exp_strategy[blk] = EXP_NEW;
                 continue;
             }
@@ -377,6 +392,10 @@ static void compute_exp_strategy(AC3EncodeContext *s)
         for (blk = 1; blk < AC3_MAX_BLOCKS; blk++)
             s->exp_strategy[ch][blk] = EXP_REUSE;
     }
+
+    /* for E-AC-3, determine frame exponent strategy */
+    if (CONFIG_EAC3_ENCODER && s->eac3)
+        ff_eac3_get_frame_exp_strategy(s);
 }
 
 
@@ -611,8 +630,12 @@ static void count_frame_bits_fixed(AC3EncodeContext *s)
         frame_bits += 2;
         frame_bits += 10;
         /* exponent strategy */
-        for (blk = 0; blk < AC3_MAX_BLOCKS; blk++)
-            frame_bits += 2 * s->fbw_channels + s->lfe_on;
+        if (s->use_frame_exp_strategy)
+            frame_bits += 5 * s->fbw_channels;
+        else
+            frame_bits += AC3_MAX_BLOCKS * 2 * s->fbw_channels;
+        if (s->lfe_on)
+            frame_bits += AC3_MAX_BLOCKS;
         /* converter exponent strategy */
         frame_bits += s->fbw_channels * 5;
         /* snr offsets */
@@ -735,8 +758,14 @@ static void count_frame_bits(AC3EncodeContext *s)
             }
         }
         /* coupling exponent strategy */
-        for (blk = 0; blk < AC3_MAX_BLOCKS; blk++)
-            frame_bits += 2 * s->blocks[blk].cpl_in_use;
+        if (s->cpl_on) {
+            if (s->use_frame_exp_strategy) {
+                frame_bits += 5 * s->cpl_on;
+            } else {
+                for (blk = 0; blk < AC3_MAX_BLOCKS; blk++)
+                    frame_bits += 2 * s->blocks[blk].cpl_in_use;
+            }
+        }
     } else {
         if (opt->audio_production_info)
             frame_bits += 7;
diff --git a/libavcodec/ac3enc.h b/libavcodec/ac3enc.h
index e3b4ba3133..f00f1cf190 100644
--- a/libavcodec/ac3enc.h
+++ b/libavcodec/ac3enc.h
@@ -218,6 +218,8 @@ typedef struct AC3EncodeContext {
     uint8_t *cpl_coord_mant_buffer;
 
     uint8_t exp_strategy[AC3_MAX_CHANNELS][AC3_MAX_BLOCKS]; ///< exponent strategies
+    uint8_t frame_exp_strategy[AC3_MAX_CHANNELS];           ///< frame exp strategy index
+    int use_frame_exp_strategy;                             ///< indicates use of frame exp strategy
     uint8_t exp_ref_block[AC3_MAX_CHANNELS][AC3_MAX_BLOCKS]; ///< reference blocks for EXP_REUSE
     uint8_t *ref_bap     [AC3_MAX_CHANNELS][AC3_MAX_BLOCKS]; ///< bit allocation pointers (bap)
     int ref_bap_set;                                         ///< indicates if ref_bap pointers have been set
diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index cb951709a5..26e23cd84b 100644
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -2811,9 +2811,9 @@ av_cold void dsputil_static_init(void)
 
 int ff_check_alignment(void){
     static int did_fail=0;
-    LOCAL_ALIGNED_16(int, aligned);
+    LOCAL_ALIGNED_16(int, aligned, [4]);
 
-    if((intptr_t)&aligned & 15){
+    if((intptr_t)aligned & 15){
         if(!did_fail){
 #if HAVE_MMX || HAVE_ALTIVEC
             av_log(NULL, AV_LOG_ERROR,
@@ -2851,29 +2851,8 @@ av_cold void dsputil_init(DSPContext* c, AVCodecContext *avctx)
 #endif //CONFIG_ENCODERS
 
     if(avctx->lowres==1){
-        if(avctx->idct_algo==FF_IDCT_INT || avctx->idct_algo==FF_IDCT_AUTO || !CONFIG_H264_DECODER){
-            c->idct_put= ff_jref_idct4_put;
-            c->idct_add= ff_jref_idct4_add;
-        }else{
-            if (avctx->codec_id != CODEC_ID_H264) {
-                c->idct_put= ff_h264_lowres_idct_put_8_c;
-                c->idct_add= ff_h264_lowres_idct_add_8_c;
-            } else {
-                switch (avctx->bits_per_raw_sample) {
-                    case 9:
-                        c->idct_put= ff_h264_lowres_idct_put_9_c;
-                        c->idct_add= ff_h264_lowres_idct_add_9_c;
-                        break;
-                    case 10:
-                        c->idct_put= ff_h264_lowres_idct_put_10_c;
-                        c->idct_add= ff_h264_lowres_idct_add_10_c;
-                        break;
-                    default:
-                        c->idct_put= ff_h264_lowres_idct_put_8_c;
-                        c->idct_add= ff_h264_lowres_idct_add_8_c;
-                }
-            }
-        }
+        c->idct_put= ff_jref_idct4_put;
+        c->idct_add= ff_jref_idct4_add;
         c->idct    = j_rev_dct4;
         c->idct_permutation_type= FF_NO_IDCT_PERM;
     }else if(avctx->lowres==2){
diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
index 2579bea91b..cf0fa7274a 100644
--- a/libavcodec/dsputil.h
+++ b/libavcodec/dsputil.h
@@ -58,8 +58,6 @@ void ff_h264_idct8_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride)
 void ff_h264_idct_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
 void ff_h264_idct8_dc_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
 void ff_h264_idct_dc_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
-void ff_h264_lowres_idct_add_ ## depth ## _c(uint8_t *dst, int stride, DCTELEM *block);\
-void ff_h264_lowres_idct_put_ ## depth ## _c(uint8_t *dst, int stride, DCTELEM *block);\
 void ff_h264_idct_add16_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
 void ff_h264_idct_add16intra_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
 void ff_h264_idct8_add4_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
diff --git a/libavcodec/eac3dec_data.c b/libavcodec/eac3_data.c
index 031702e9e9..b159e1682f 100644
--- a/libavcodec/eac3dec_data.c
+++ b/libavcodec/eac3_data.c
@@ -1,5 +1,5 @@
 /*
- * E-AC-3 decoder tables
+ * E-AC-3 tables
  * Copyright (c) 2007 Bartlomiej Wolowiec <bartek.wolowiec@gmail.com>
  *
  * This file is part of FFmpeg.
@@ -24,7 +24,7 @@
  * Tables taken directly from the E-AC-3 spec.
  */
 
-#include "eac3dec_data.h"
+#include "eac3_data.h"
 #include "ac3.h"
 
 const uint8_t ff_eac3_bits_vs_hebap[20] = {
diff --git a/libavcodec/eac3dec_data.h b/libavcodec/eac3_data.h
index 133183398f..10a67f16d2 100644
--- a/libavcodec/eac3dec_data.h
+++ b/libavcodec/eac3_data.h
@@ -1,5 +1,5 @@
 /*
- * E-AC-3 decoder tables
+ * E-AC-3 tables
  * Copyright (c) 2007 Bartlomiej Wolowiec <bartek.wolowiec@gmail.com>
  *
  * This file is part of FFmpeg.
@@ -19,8 +19,8 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
-#ifndef AVCODEC_EAC3DEC_DATA_H
-#define AVCODEC_EAC3DEC_DATA_H
+#ifndef AVCODEC_EAC3_DATA_H
+#define AVCODEC_EAC3_DATA_H
 
 #include <stdint.h>
 
@@ -33,4 +33,4 @@ extern const int16_t (* const ff_eac3_mantissa_vq[8])[6];
 extern const uint8_t ff_eac3_frm_expstr[32][6];
 extern const float   ff_eac3_spx_atten_tab[32][3];
 
-#endif /* AVCODEC_EAC3DEC_DATA_H */
+#endif /* AVCODEC_EAC3_DATA_H */
diff --git a/libavcodec/eac3dec.c b/libavcodec/eac3dec.c
index 40f571ffc7..ad240e4341 100644
--- a/libavcodec/eac3dec.c
+++ b/libavcodec/eac3dec.c
@@ -51,7 +51,7 @@
 #include "ac3_parser.h"
 #include "ac3dec.h"
 #include "ac3dec_data.h"
-#include "eac3dec_data.h"
+#include "eac3_data.h"
 
 /** gain adaptive quantization mode */
 typedef enum {
diff --git a/libavcodec/eac3enc.c b/libavcodec/eac3enc.c
index 0433db49d7..09fa80fcb9 100644
--- a/libavcodec/eac3enc.c
+++ b/libavcodec/eac3enc.c
@@ -27,6 +27,7 @@
 #define CONFIG_AC3ENC_FLOAT 1
 #include "ac3enc.h"
 #include "eac3enc.h"
+#include "eac3_data.h"
 
 
 #define AC3ENC_TYPE AC3ENC_TYPE_EAC3
@@ -35,6 +36,51 @@ static const AVClass eac3enc_class = { "E-AC-3 Encoder", av_default_item_name,
                                        eac3_options, LIBAVUTIL_VERSION_INT };
 
 
+/**
+ * LUT for finding a matching frame exponent strategy index from a set of
+ * exponent strategies for a single channel across all 6 blocks.
+ */
+static int8_t eac3_frame_expstr_index_tab[3][4][4][4][4][4];
+
+
+void ff_eac3_exponent_init(void)
+{
+    int i;
+
+    memset(eac3_frame_expstr_index_tab, -1, sizeof(eac3_frame_expstr_index_tab));
+    for (i = 0; i < 32; i++) {
+        eac3_frame_expstr_index_tab[ff_eac3_frm_expstr[i][0]-1]
+                                   [ff_eac3_frm_expstr[i][1]]
+                                   [ff_eac3_frm_expstr[i][2]]
+                                   [ff_eac3_frm_expstr[i][3]]
+                                   [ff_eac3_frm_expstr[i][4]]
+                                   [ff_eac3_frm_expstr[i][5]] = i;
+    }
+}
+
+
+void ff_eac3_get_frame_exp_strategy(AC3EncodeContext *s)
+{
+    int ch;
+
+    s->use_frame_exp_strategy = 1;
+    for (ch = !s->cpl_on; ch <= s->fbw_channels; ch++) {
+        int expstr = eac3_frame_expstr_index_tab[s->exp_strategy[ch][0]-1]
+                                                [s->exp_strategy[ch][1]]
+                                                [s->exp_strategy[ch][2]]
+                                                [s->exp_strategy[ch][3]]
+                                                [s->exp_strategy[ch][4]]
+                                                [s->exp_strategy[ch][5]];
+        if (expstr < 0) {
+            s->use_frame_exp_strategy = 0;
+            break;
+        }
+        s->frame_exp_strategy[ch] = expstr;
+    }
+}
+
+
+
 void ff_eac3_set_cpl_states(AC3EncodeContext *s)
 {
     int ch, blk;
@@ -98,7 +144,7 @@ void ff_eac3_output_frame_header(AC3EncodeContext *s)
     put_bits(&s->pb, 1, 0);                         /* no additional bit stream info */
 
     /* frame header */
-    put_bits(&s->pb, 1, 1);                         /* exponent strategy syntax = each block */
+    put_bits(&s->pb, 1, !s->use_frame_exp_strategy);/* exponent strategy syntax */
     put_bits(&s->pb, 1, 0);                         /* aht enabled = no */
     put_bits(&s->pb, 2, 0);                         /* snr offset strategy = 1 */
     put_bits(&s->pb, 1, 0);                         /* transient pre-noise processing enabled = no */
@@ -120,16 +166,25 @@ void ff_eac3_output_frame_header(AC3EncodeContext *s)
         }
     }
     /* exponent strategy */
-    for (blk = 0; blk < AC3_MAX_BLOCKS; blk++)
-        for (ch = !s->blocks[blk].cpl_in_use; ch <= s->fbw_channels; ch++)
-            put_bits(&s->pb, 2, s->exp_strategy[ch][blk]);
+    if (s->use_frame_exp_strategy) {
+        for (ch = !s->cpl_on; ch <= s->fbw_channels; ch++)
+            put_bits(&s->pb, 5, s->frame_exp_strategy[ch]);
+    } else {
+        for (blk = 0; blk < AC3_MAX_BLOCKS; blk++)
+            for (ch = !s->blocks[blk].cpl_in_use; ch <= s->fbw_channels; ch++)
+                put_bits(&s->pb, 2, s->exp_strategy[ch][blk]);
+    }
     if (s->lfe_on) {
         for (blk = 0; blk < AC3_MAX_BLOCKS; blk++)
             put_bits(&s->pb, 1, s->exp_strategy[s->lfe_channel][blk]);
     }
     /* E-AC-3 to AC-3 converter exponent strategy (unfortunately not optional...) */
-    for (ch = 1; ch <= s->fbw_channels; ch++)
-        put_bits(&s->pb, 5, 0);
+    for (ch = 1; ch <= s->fbw_channels; ch++) {
+        if (s->use_frame_exp_strategy)
+            put_bits(&s->pb, 5, s->frame_exp_strategy[ch]);
+        else
+            put_bits(&s->pb, 5, 0);
+    }
     /* snr offsets */
     put_bits(&s->pb, 6, s->coarse_snr_offset);
     put_bits(&s->pb, 4, s->fine_snr_offset[1]);
diff --git a/libavcodec/eac3enc.h b/libavcodec/eac3enc.h
index eacb8cf164..a92a24c2fa 100644
--- a/libavcodec/eac3enc.h
+++ b/libavcodec/eac3enc.h
@@ -30,6 +30,16 @@
 #include "ac3enc.h"
 
 /**
+ * Initialize E-AC-3 exponent tables.
+ */
+void ff_eac3_exponent_init(void);
+
+/**
+ * Determine frame exponent strategy use and indices.
+ */
+void ff_eac3_get_frame_exp_strategy(AC3EncodeContext *s);
+
+/**
  * Set coupling states.
  * This determines whether certain flags must be written to the bitstream or
  * whether they will be implicitly already known by the decoder.
diff --git a/libavcodec/h264.c b/libavcodec/h264.c
index 835b29dbcc..fe3cd93c0f 100644
--- a/libavcodec/h264.c
+++ b/libavcodec/h264.c
@@ -3463,52 +3463,6 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
             }
         }
     }
-
-#if 0
-    for(;s->mb_y < s->mb_height; s->mb_y++){
-        for(;s->mb_x < s->mb_width; s->mb_x++){
-            int ret= decode_mb(h);
-
-            ff_h264_hl_decode_mb(h);
-
-            if(ret<0){
-                av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
-                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
-
-                return -1;
-            }
-
-            if(++s->mb_x >= s->mb_width){
-                s->mb_x=0;
-                if(++s->mb_y >= s->mb_height){
-                    if(get_bits_count(s->gb) == s->gb.size_in_bits){
-                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
-
-                        return 0;
-                    }else{
-                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
-
-                        return -1;
-                    }
-                }
-            }
-
-            if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
-                if(get_bits_count(s->gb) == s->gb.size_in_bits){
-                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
-
-                    return 0;
-                }else{
-                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
-
-                    return -1;
-                }
-            }
-        }
-        s->mb_x=0;
-        ff_draw_horiz_band(s, 16*s->mb_y, 16);
-    }
-#endif
 }
 
 /**
@@ -3976,109 +3930,6 @@ int main(void){
         STOP_TIMER("get_se_golomb");
     }
 
-#if 0
-    printf("testing 4x4 (I)DCT\n");
-
-    DCTELEM block[16];
-    uint8_t src[16], ref[16];
-    uint64_t error= 0, max_error=0;
-
-    for(i=0; i<COUNT; i++){
-        int j;
-//        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
-        for(j=0; j<16; j++){
-            ref[j]= random()%255;
-            src[j]= random()%255;
-        }
-
-        h264_diff_dct_c(block, src, ref, 4);
-
-        //normalize
-        for(j=0; j<16; j++){
-//            printf("%d ", block[j]);
-            block[j]= block[j]*4;
-            if(j&1) block[j]= (block[j]*4 + 2)/5;
-            if(j&4) block[j]= (block[j]*4 + 2)/5;
-        }
-//        printf("\n");
-
-        h->h264dsp.h264_idct_add(ref, block, 4);
-/*        for(j=0; j<16; j++){
-            printf("%d ", ref[j]);
-        }
-        printf("\n");*/
-
-        for(j=0; j<16; j++){
-            int diff= FFABS(src[j] - ref[j]);
-
-            error+= diff*diff;
-            max_error= FFMAX(max_error, diff);
-        }
-    }
-    printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
-    printf("testing quantizer\n");
-    for(qp=0; qp<52; qp++){
-        for(i=0; i<16; i++)
-            src1_block[i]= src2_block[i]= random()%255;
-
-    }
-    printf("Testing NAL layer\n");
-
-    uint8_t bitstream[COUNT];
-    uint8_t nal[COUNT*2];
-    H264Context h;
-    memset(&h, 0, sizeof(H264Context));
-
-    for(i=0; i<COUNT; i++){
-        int zeros= i;
-        int nal_length;
-        int consumed;
-        int out_length;
-        uint8_t *out;
-        int j;
-
-        for(j=0; j<COUNT; j++){
-            bitstream[j]= (random() % 255) + 1;
-        }
-
-        for(j=0; j<zeros; j++){
-            int pos= random() % COUNT;
-            while(bitstream[pos] == 0){
-                pos++;
-                pos %= COUNT;
-            }
-            bitstream[pos]=0;
-        }
-
-        START_TIMER
-
-        nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
-        if(nal_length<0){
-            printf("encoding failed\n");
-            return -1;
-        }
-
-        out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
-
-        STOP_TIMER("NAL")
-
-        if(out_length != COUNT){
-            printf("incorrect length %d %d\n", out_length, COUNT);
-            return -1;
-        }
-
-        if(consumed != nal_length){
-            printf("incorrect consumed length %d %d\n", nal_length, consumed);
-            return -1;
-        }
-
-        if(memcmp(bitstream, out, COUNT)){
-            printf("mismatch\n");
-            return -1;
-        }
-    }
-#endif
-
     printf("Testing RBSP\n");
 
 
diff --git a/libavcodec/h264idct_template.c b/libavcodec/h264idct_template.c
index 94d073ef30..3b386695a2 100644
--- a/libavcodec/h264idct_template.c
+++ b/libavcodec/h264idct_template.c
@@ -46,57 +46,46 @@ static const uint8_t scan8[16*3]={
 };
 #endif
 
-static av_always_inline void FUNCC(idct_internal)(uint8_t *p_dst, DCTELEM *p_block, int stride, int block_stride, int shift, int add){
+void FUNCC(ff_h264_idct_add)(uint8_t *_dst, DCTELEM *_block, int stride)
+{
     int i;
     INIT_CLIP
-    pixel *dst = (pixel*)p_dst;
-    dctcoef *block = (dctcoef*)p_block;
+    pixel *dst = (pixel*)_dst;
+    dctcoef *block = (dctcoef*)_block;
     stride >>= sizeof(pixel)-1;
 
-    block[0] += 1<<(shift-1);
+    block[0] += 1 << 5;
 
     for(i=0; i<4; i++){
-        const int z0=  block[i + block_stride*0]     +  block[i + block_stride*2];
-        const int z1=  block[i + block_stride*0]     -  block[i + block_stride*2];
-        const int z2= (block[i + block_stride*1]>>1) -  block[i + block_stride*3];
-        const int z3=  block[i + block_stride*1]     + (block[i + block_stride*3]>>1);
-
-        block[i + block_stride*0]= z0 + z3;
-        block[i + block_stride*1]= z1 + z2;
-        block[i + block_stride*2]= z1 - z2;
-        block[i + block_stride*3]= z0 - z3;
+        const int z0=  block[i + 4*0]     +  block[i + 4*2];
+        const int z1=  block[i + 4*0]     -  block[i + 4*2];
+        const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
+        const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
+
+        block[i + 4*0]= z0 + z3;
+        block[i + 4*1]= z1 + z2;
+        block[i + 4*2]= z1 - z2;
+        block[i + 4*3]= z0 - z3;
     }
 
     for(i=0; i<4; i++){
-        const int z0=  block[0 + block_stride*i]     +  block[2 + block_stride*i];
-        const int z1=  block[0 + block_stride*i]     -  block[2 + block_stride*i];
-        const int z2= (block[1 + block_stride*i]>>1) -  block[3 + block_stride*i];
-        const int z3=  block[1 + block_stride*i]     + (block[3 + block_stride*i]>>1);
-
-        dst[i + 0*stride]= CLIP(add*dst[i + 0*stride] + ((z0 + z3) >> shift));
-        dst[i + 1*stride]= CLIP(add*dst[i + 1*stride] + ((z1 + z2) >> shift));
-        dst[i + 2*stride]= CLIP(add*dst[i + 2*stride] + ((z1 - z2) >> shift));
-        dst[i + 3*stride]= CLIP(add*dst[i + 3*stride] + ((z0 - z3) >> shift));
+        const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
+        const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
+        const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
+        const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
+
+        dst[i + 0*stride]= CLIP(dst[i + 0*stride] + ((z0 + z3) >> 6));
+        dst[i + 1*stride]= CLIP(dst[i + 1*stride] + ((z1 + z2) >> 6));
+        dst[i + 2*stride]= CLIP(dst[i + 2*stride] + ((z1 - z2) >> 6));
+        dst[i + 3*stride]= CLIP(dst[i + 3*stride] + ((z0 - z3) >> 6));
     }
 }
 
-void FUNCC(ff_h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride){
-    FUNCC(idct_internal)(dst, block, stride, 4, 6, 1);
-}
-
-void FUNCC(ff_h264_lowres_idct_add)(uint8_t *dst, int stride, DCTELEM *block){
-    FUNCC(idct_internal)(dst, block, stride, 8, 3, 1);
-}
-
-void FUNCC(ff_h264_lowres_idct_put)(uint8_t *dst, int stride, DCTELEM *block){
-    FUNCC(idct_internal)(dst, block, stride, 8, 3, 0);
-}
-
-void FUNCC(ff_h264_idct8_add)(uint8_t *p_dst, DCTELEM *p_block, int stride){
+void FUNCC(ff_h264_idct8_add)(uint8_t *_dst, DCTELEM *_block, int stride){
     int i;
     INIT_CLIP
-    pixel *dst = (pixel*)p_dst;
-    dctcoef *block = (dctcoef*)p_block;
+    pixel *dst = (pixel*)_dst;
+    dctcoef *block = (dctcoef*)_block;
     stride >>= sizeof(pixel)-1;
 
     block[0] += 32;
@@ -200,7 +189,7 @@ void FUNCC(ff_h264_idct_add16)(uint8_t *dst, const int *block_offset, DCTELEM *b
         int nnz = nnzc[ scan8[i] ];
         if(nnz){
             if(nnz==1 && ((dctcoef*)block)[i*16]) FUNCC(ff_h264_idct_dc_add)(dst + block_offset[i], block + i*16*sizeof(pixel), stride);
-            else                                  FUNCC(idct_internal      )(dst + block_offset[i], block + i*16*sizeof(pixel), stride, 4, 6, 1);
+            else                                  FUNCC(ff_h264_idct_add   )(dst + block_offset[i], block + i*16*sizeof(pixel), stride);
         }
     }
 }
@@ -208,7 +197,7 @@ void FUNCC(ff_h264_idct_add16)(uint8_t *dst, const int *block_offset, DCTELEM *b
 void FUNCC(ff_h264_idct_add16intra)(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[15*8]){
     int i;
     for(i=0; i<16; i++){
-        if(nnzc[ scan8[i] ])             FUNCC(idct_internal      )(dst + block_offset[i], block + i*16*sizeof(pixel), stride, 4, 6, 1);
+        if(nnzc[ scan8[i] ])             FUNCC(ff_h264_idct_add   )(dst + block_offset[i], block + i*16*sizeof(pixel), stride);
         else if(((dctcoef*)block)[i*16]) FUNCC(ff_h264_idct_dc_add)(dst + block_offset[i], block + i*16*sizeof(pixel), stride);
     }
 }
diff --git a/libavcodec/kgv1dec.c b/libavcodec/kgv1dec.c
index c364cfc593..3e3922774e 100644
--- a/libavcodec/kgv1dec.c
+++ b/libavcodec/kgv1dec.c
@@ -174,6 +174,5 @@ AVCodec ff_kgv1_decoder = {
     NULL,
     decode_end,
     decode_frame,
-    .max_lowres = 1,
     .long_name = NULL_IF_CONFIG_SMALL("Kega Game Video"),
 };