avcodec/mpeg12enc: Basic support for encoding non even QPs for -non_linear_quant 1

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

4 files changed, 38 insertions, 38 deletions

diff --git a/libavcodec/mpeg12enc.c b/libavcodec/mpeg12enc.c
index 6f87117058..b88e2e6b8e 100644
--- a/libavcodec/mpeg12enc.c
+++ b/libavcodec/mpeg12enc.c
@@ -42,12 +42,6 @@
 #include "mpegutils.h"
 #include "mpegvideo.h"
 
-static const int8_t inv_non_linear_qscale[] = {
-    0, 2, 4, 6, 8, 9, 10, 11, 12, 13, 14, 15, 16,
-   -1,17,-1,18,-1,19, -1, 20, -1, 21, -1, 22, -1,
-   23,-1,24,-1,-1,-1
-};
-
 static const uint8_t svcd_scan_offset_placeholder[] = {
     0x10, 0x0E, 0x00, 0x80, 0x81, 0x00, 0x80,
     0x81, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
@@ -402,13 +396,7 @@ static inline void encode_mb_skip_run(MpegEncContext *s, int run)
 
 static av_always_inline void put_qscale(MpegEncContext *s)
 {
-    if (s->q_scale_type) {
-        int qp = inv_non_linear_qscale[s->qscale];
-        av_assert2(s->qscale >= 1 && qp > 0);
-        put_bits(&s->pb, 5, qp);
-    } else {
-        put_bits(&s->pb, 5, s->qscale);
-    }
+    put_bits(&s->pb, 5, s->qscale);
 }
 
 void ff_mpeg1_encode_slice_header(MpegEncContext *s)
diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c
index 7c0e9c43e5..96634ec7b5 100644
--- a/libavcodec/mpegvideo.c
+++ b/libavcodec/mpegvideo.c
@@ -114,7 +114,8 @@ static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
     int i, level, nCoeffs;
     const uint16_t *quant_matrix;
 
-    qscale <<= 1;
+    if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
+    else                 qscale <<= 1;
 
     if(s->alternate_scan) nCoeffs= 63;
     else nCoeffs= s->block_last_index[n];
@@ -144,7 +145,8 @@ static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
     const uint16_t *quant_matrix;
     int sum=-1;
 
-    qscale <<= 1;
+    if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
+    else                 qscale <<= 1;
 
     if(s->alternate_scan) nCoeffs= 63;
     else nCoeffs= s->block_last_index[n];
@@ -177,7 +179,8 @@ static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
     const uint16_t *quant_matrix;
     int sum=-1;
 
-    qscale <<= 1;
+    if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
+    else                 qscale <<= 1;
 
     if(s->alternate_scan) nCoeffs= 63;
     else nCoeffs= s->block_last_index[n];
diff --git a/libavcodec/mpegvideo_enc.c b/libavcodec/mpegvideo_enc.c
index 65f6e68299..8aca1d0b3a 100644
--- a/libavcodec/mpegvideo_enc.c
+++ b/libavcodec/mpegvideo_enc.c
@@ -97,6 +97,11 @@ void ff_convert_matrix(MpegEncContext *s, int (*qmat)[64],
 
     for (qscale = qmin; qscale <= qmax; qscale++) {
         int i;
+        int qscale2;
+
+        if (s->q_scale_type) qscale2 = ff_mpeg2_non_linear_qscale[qscale];
+        else                 qscale2 = qscale << 1;
+
         if (fdsp->fdct == ff_jpeg_fdct_islow_8  ||
 #if CONFIG_FAANDCT
             fdsp->fdct == ff_faandct            ||
@@ -104,40 +109,40 @@ void ff_convert_matrix(MpegEncContext *s, int (*qmat)[64],
             fdsp->fdct == ff_jpeg_fdct_islow_10) {
             for (i = 0; i < 64; i++) {
                 const int j = s->idsp.idct_permutation[i];
-                int64_t den = (int64_t) qscale * quant_matrix[j];
+                int64_t den = (int64_t) qscale2 * quant_matrix[j];
                 /* 16 <= qscale * quant_matrix[i] <= 7905
                  * Assume x = ff_aanscales[i] * qscale * quant_matrix[i]
                  *             19952 <=              x  <= 249205026
                  * (1 << 36) / 19952 >= (1 << 36) / (x) >= (1 << 36) / 249205026
                  *           3444240 >= (1 << 36) / (x) >= 275 */
 
-                qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / den);
+                qmat[qscale][i] = (int)((UINT64_C(2) << QMAT_SHIFT) / den);
             }
         } else if (fdsp->fdct == ff_fdct_ifast) {
             for (i = 0; i < 64; i++) {
                 const int j = s->idsp.idct_permutation[i];
-                int64_t den = ff_aanscales[i] * (int64_t) qscale * quant_matrix[j];
+                int64_t den = ff_aanscales[i] * (int64_t) qscale2 * quant_matrix[j];
                 /* 16 <= qscale * quant_matrix[i] <= 7905
                  * Assume x = ff_aanscales[i] * qscale * quant_matrix[i]
                  *             19952 <=              x  <= 249205026
                  * (1 << 36) / 19952 >= (1 << 36) / (x) >= (1 << 36) / 249205026
                  *           3444240 >= (1 << 36) / (x) >= 275 */
 
-                qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) / den);
+                qmat[qscale][i] = (int)((UINT64_C(2) << (QMAT_SHIFT + 14)) / den);
             }
         } else {
             for (i = 0; i < 64; i++) {
                 const int j = s->idsp.idct_permutation[i];
-                int64_t den = (int64_t) qscale * quant_matrix[j];
+                int64_t den = (int64_t) qscale2 * quant_matrix[j];
                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
                  * Assume x = qscale * quant_matrix[i]
                  * So             16 <=              x  <= 7905
                  * so (1 << 19) / 16 >= (1 << 19) / (x) >= (1 << 19) / 7905
                  * so          32768 >= (1 << 19) / (x) >= 67 */
-                qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / den);
+                qmat[qscale][i] = (int)((UINT64_C(2) << QMAT_SHIFT) / den);
                 //qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) /
                 //                    (qscale * quant_matrix[i]);
-                qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / den;
+                qmat16[qscale][0][i] = (2 << QMAT_SHIFT_MMX) / den;
 
                 if (qmat16[qscale][0][i] == 0 ||
                     qmat16[qscale][0][i] == 128 * 256)
@@ -167,22 +172,19 @@ void ff_convert_matrix(MpegEncContext *s, int (*qmat)[64],
 
 static inline void update_qscale(MpegEncContext *s)
 {
-    if (s->q_scale_type == 1) {
+    if (s->q_scale_type == 1 && 0) {
         int i;
         int bestdiff=INT_MAX;
         int best = 1;
-        static const uint8_t non_linear_qscale[] = {
-            1,2,3,4,5,6,7,8,9,10,11,12,14,16,18,20,22,24,26,28
-        };
-
-        for (i = 0 ; i<FF_ARRAY_ELEMS(non_linear_qscale); i++) {
-            int diff = FFABS((non_linear_qscale[i]<<(FF_LAMBDA_SHIFT + 7)) - (int)s->lambda * 139);
-            if (non_linear_qscale[i] < s->avctx->qmin ||
-                (non_linear_qscale[i] > s->avctx->qmax && !s->vbv_ignore_qmax))
+
+        for (i = 0 ; i<FF_ARRAY_ELEMS(ff_mpeg2_non_linear_qscale); i++) {
+            int diff = FFABS((ff_mpeg2_non_linear_qscale[i]<<(FF_LAMBDA_SHIFT + 6)) - (int)s->lambda * 139);
+            if (ff_mpeg2_non_linear_qscale[i] < s->avctx->qmin ||
+                (ff_mpeg2_non_linear_qscale[i] > s->avctx->qmax && !s->vbv_ignore_qmax))
                 continue;
             if (diff < bestdiff) {
                 bestdiff = diff;
-                best = non_linear_qscale[i];
+                best = i;
             }
         }
         s->qscale = best;
@@ -3854,6 +3856,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
     uint8_t * length;
     uint8_t * last_length;
     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
+    int mpeg2_qscale;
 
     s->fdsp.fdct(block);
 
@@ -3862,6 +3865,9 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
     qmul= qscale*16;
     qadd= ((qscale-1)|1)*8;
 
+    if (s->q_scale_type) mpeg2_qscale = ff_mpeg2_non_linear_qscale[qscale];
+    else                 mpeg2_qscale = qscale << 1;
+
     if (s->mb_intra) {
         int q;
         if (!s->h263_aic) {
@@ -3978,10 +3984,10 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
             }else{ //MPEG1
                 j = s->idsp.idct_permutation[scantable[i]]; // FIXME: optimize
                 if(s->mb_intra){
-                        unquant_coeff = (int)(  alevel  * qscale * matrix[j]) >> 3;
+                        unquant_coeff = (int)(  alevel  * mpeg2_qscale * matrix[j]) >> 4;
                         unquant_coeff =   (unquant_coeff - 1) | 1;
                 }else{
-                        unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) matrix[j])) >> 4;
+                        unquant_coeff = (((  alevel  << 1) + 1) * mpeg2_qscale * ((int) matrix[j])) >> 5;
                         unquant_coeff =   (unquant_coeff - 1) | 1;
                 }
                 unquant_coeff<<= 3;
@@ -4097,7 +4103,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
             if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
                     unquant_coeff= (alevel*qmul + qadd)>>3;
             }else{ //MPEG1
-                    unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) matrix[0])) >> 4;
+                    unquant_coeff = (((  alevel  << 1) + 1) * mpeg2_qscale * ((int) matrix[0])) >> 5;
                     unquant_coeff =   (unquant_coeff - 1) | 1;
             }
             unquant_coeff = (unquant_coeff + 4) >> 3;
diff --git a/libavcodec/x86/mpegvideo.c b/libavcodec/x86/mpegvideo.c
index 3df6b864c5..af47422cf2 100644
--- a/libavcodec/x86/mpegvideo.c
+++ b/libavcodec/x86/mpegvideo.c
@@ -25,6 +25,7 @@
 #include "libavutil/x86/cpu.h"
 #include "libavcodec/avcodec.h"
 #include "libavcodec/mpegvideo.h"
+#include "libavcodec/mpegvideodata.h"
 
 #if HAVE_MMX_INLINE
 
@@ -308,7 +309,8 @@ static void dct_unquantize_mpeg2_intra_mmx(MpegEncContext *s,
 
     av_assert2(s->block_last_index[n]>=0);
 
-    qscale <<= 1;
+    if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
+    else                 qscale <<= 1;
 
     if(s->alternate_scan) nCoeffs= 63; //FIXME
     else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
@@ -375,7 +377,8 @@ static void dct_unquantize_mpeg2_inter_mmx(MpegEncContext *s,
 
     av_assert2(s->block_last_index[n]>=0);
 
-    qscale <<= 1;
+    if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
+    else                 qscale <<= 1;
 
     if(s->alternate_scan) nCoeffs= 63; //FIXME
     else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];