return -1;
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->slice_size, ctx->m.mb_height*sizeof(uint32_t), fail);
+ FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->slice_offs, ctx->m.mb_height*sizeof(uint32_t), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->mb_bits, ctx->m.mb_num *sizeof(uint16_t), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->mb_qscale, ctx->m.mb_num *sizeof(uint8_t) , fail);
ctx->frame.pict_type = FF_I_TYPE;
ctx->m.avctx->coded_frame = &ctx->frame;
- if (avctx->thread_count > MAX_THREADS || (avctx->thread_count > ctx->m.mb_height)) {
+ if (avctx->thread_count > MAX_THREADS) {
av_log(avctx, AV_LOG_ERROR, "too many threads\n");
return -1;
}
memcpy(ctx->thread[i], ctx, sizeof(DNXHDEncContext));
}
- for (i = 0; i < avctx->thread_count; i++) {
- ctx->thread[i]->m.start_mb_y = (ctx->m.mb_height*(i ) + avctx->thread_count/2) / avctx->thread_count;
- ctx->thread[i]->m.end_mb_y = (ctx->m.mb_height*(i+1) + avctx->thread_count/2) / avctx->thread_count;
- }
-
return 0;
fail: //for FF_ALLOCZ_OR_GOTO
return -1;
}
}
-static int dnxhd_calc_bits_thread(AVCodecContext *avctx, void *arg)
+static int dnxhd_calc_bits_thread(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)
{
- DNXHDEncContext *ctx = *(void**)arg;
- int mb_y, mb_x;
- int qscale = ctx->thread[0]->qscale;
-
- for (mb_y = ctx->m.start_mb_y; mb_y < ctx->m.end_mb_y; mb_y++) {
- ctx->m.last_dc[0] =
- ctx->m.last_dc[1] =
- ctx->m.last_dc[2] = 1024;
-
- for (mb_x = 0; mb_x < ctx->m.mb_width; mb_x++) {
- unsigned mb = mb_y * ctx->m.mb_width + mb_x;
- int ssd = 0;
- int ac_bits = 0;
- int dc_bits = 0;
- int i;
-
- dnxhd_get_blocks(ctx, mb_x, mb_y);
-
- for (i = 0; i < 8; i++) {
- DECLARE_ALIGNED_16(DCTELEM, block[64]);
- DCTELEM *src_block = ctx->blocks[i];
- int overflow, nbits, diff, last_index;
- int n = dnxhd_switch_matrix(ctx, i);
-
- memcpy(block, src_block, sizeof(block));
- last_index = ctx->m.dct_quantize((MpegEncContext*)ctx, block, i, qscale, &overflow);
- ac_bits += dnxhd_calc_ac_bits(ctx, block, last_index);
-
- diff = block[0] - ctx->m.last_dc[n];
- if (diff < 0) nbits = av_log2_16bit(-2*diff);
- else nbits = av_log2_16bit( 2*diff);
- dc_bits += ctx->cid_table->dc_bits[nbits] + nbits;
-
- ctx->m.last_dc[n] = block[0];
-
- if (avctx->mb_decision == FF_MB_DECISION_RD || !RC_VARIANCE) {
- dnxhd_unquantize_c(ctx, block, i, qscale, last_index);
- ctx->m.dsp.idct(block);
- ssd += dnxhd_ssd_block(block, src_block);
- }
+ DNXHDEncContext *ctx = avctx->priv_data;
+ int mb_y = jobnr, mb_x;
+ int qscale = ctx->qscale;
+ ctx = ctx->thread[threadnr];
+
+ ctx->m.last_dc[0] =
+ ctx->m.last_dc[1] =
+ ctx->m.last_dc[2] = 1024;
+
+ for (mb_x = 0; mb_x < ctx->m.mb_width; mb_x++) {
+ unsigned mb = mb_y * ctx->m.mb_width + mb_x;
+ int ssd = 0;
+ int ac_bits = 0;
+ int dc_bits = 0;
+ int i;
+
+ dnxhd_get_blocks(ctx, mb_x, mb_y);
+
+ for (i = 0; i < 8; i++) {
+ DECLARE_ALIGNED_16(DCTELEM, block[64]);
+ DCTELEM *src_block = ctx->blocks[i];
+ int overflow, nbits, diff, last_index;
+ int n = dnxhd_switch_matrix(ctx, i);
+
+ memcpy(block, src_block, sizeof(block));
+ last_index = ctx->m.dct_quantize((MpegEncContext*)ctx, block, i, qscale, &overflow);
+ ac_bits += dnxhd_calc_ac_bits(ctx, block, last_index);
+
+ diff = block[0] - ctx->m.last_dc[n];
+ if (diff < 0) nbits = av_log2_16bit(-2*diff);
+ else nbits = av_log2_16bit( 2*diff);
+ dc_bits += ctx->cid_table->dc_bits[nbits] + nbits;
+
+ ctx->m.last_dc[n] = block[0];
+
+ if (avctx->mb_decision == FF_MB_DECISION_RD || !RC_VARIANCE) {
+ dnxhd_unquantize_c(ctx, block, i, qscale, last_index);
+ ctx->m.dsp.idct(block);
+ ssd += dnxhd_ssd_block(block, src_block);
}
- ctx->mb_rc[qscale][mb].ssd = ssd;
- ctx->mb_rc[qscale][mb].bits = ac_bits+dc_bits+12+8*ctx->vlc_bits[0];
}
+ ctx->mb_rc[qscale][mb].ssd = ssd;
+ ctx->mb_rc[qscale][mb].bits = ac_bits+dc_bits+12+8*ctx->vlc_bits[0];
}
return 0;
}
-static int dnxhd_encode_thread(AVCodecContext *avctx, void *arg)
+static int dnxhd_encode_thread(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)
{
- DNXHDEncContext *ctx = *(void**)arg;
- int mb_y, mb_x;
-
- for (mb_y = ctx->m.start_mb_y; mb_y < ctx->m.end_mb_y; mb_y++) {
- ctx->m.last_dc[0] =
- ctx->m.last_dc[1] =
- ctx->m.last_dc[2] = 1024;
- for (mb_x = 0; mb_x < ctx->m.mb_width; mb_x++) {
- unsigned mb = mb_y * ctx->m.mb_width + mb_x;
- int qscale = ctx->mb_qscale[mb];
- int i;
-
- put_bits(&ctx->m.pb, 12, qscale<<1);
-
- dnxhd_get_blocks(ctx, mb_x, mb_y);
-
- for (i = 0; i < 8; i++) {
- DCTELEM *block = ctx->blocks[i];
- int last_index, overflow;
- int n = dnxhd_switch_matrix(ctx, i);
- last_index = ctx->m.dct_quantize((MpegEncContext*)ctx, block, i, qscale, &overflow);
- //START_TIMER;
- dnxhd_encode_block(ctx, block, last_index, n);
- //STOP_TIMER("encode_block");
- }
+ DNXHDEncContext *ctx = avctx->priv_data;
+ int mb_y = jobnr, mb_x;
+ ctx = ctx->thread[threadnr];
+ init_put_bits(&ctx->m.pb, (uint8_t *)arg + 640 + ctx->slice_offs[jobnr], ctx->slice_size[jobnr]);
+
+ ctx->m.last_dc[0] =
+ ctx->m.last_dc[1] =
+ ctx->m.last_dc[2] = 1024;
+ for (mb_x = 0; mb_x < ctx->m.mb_width; mb_x++) {
+ unsigned mb = mb_y * ctx->m.mb_width + mb_x;
+ int qscale = ctx->mb_qscale[mb];
+ int i;
+
+ put_bits(&ctx->m.pb, 12, qscale<<1);
+
+ dnxhd_get_blocks(ctx, mb_x, mb_y);
+
+ for (i = 0; i < 8; i++) {
+ DCTELEM *block = ctx->blocks[i];
+ int last_index, overflow;
+ int n = dnxhd_switch_matrix(ctx, i);
+ last_index = ctx->m.dct_quantize((MpegEncContext*)ctx, block, i, qscale, &overflow);
+ //START_TIMER;
+ dnxhd_encode_block(ctx, block, last_index, n);
+ //STOP_TIMER("encode_block");
}
- if (put_bits_count(&ctx->m.pb)&31)
- put_bits(&ctx->m.pb, 32-(put_bits_count(&ctx->m.pb)&31), 0);
}
+ if (put_bits_count(&ctx->m.pb)&31)
+ put_bits(&ctx->m.pb, 32-(put_bits_count(&ctx->m.pb)&31), 0);
flush_put_bits(&ctx->m.pb);
return 0;
}
-static void dnxhd_setup_threads_slices(DNXHDEncContext *ctx, uint8_t *buf)
+static void dnxhd_setup_threads_slices(DNXHDEncContext *ctx)
{
int mb_y, mb_x;
- int i, offset = 0;
- for (i = 0; i < ctx->m.avctx->thread_count; i++) {
- int thread_size = 0;
- for (mb_y = ctx->thread[i]->m.start_mb_y; mb_y < ctx->thread[i]->m.end_mb_y; mb_y++) {
+ int offset = 0;
+ for (mb_y = 0; mb_y < ctx->m.mb_height; mb_y++) {
+ int thread_size;
+ ctx->slice_offs[mb_y] = offset;
ctx->slice_size[mb_y] = 0;
for (mb_x = 0; mb_x < ctx->m.mb_width; mb_x++) {
unsigned mb = mb_y * ctx->m.mb_width + mb_x;
}
ctx->slice_size[mb_y] = (ctx->slice_size[mb_y]+31)&~31;
ctx->slice_size[mb_y] >>= 3;
- thread_size += ctx->slice_size[mb_y];
- }
- init_put_bits(&ctx->thread[i]->m.pb, buf + 640 + offset, thread_size);
+ thread_size = ctx->slice_size[mb_y];
offset += thread_size;
}
}
-static int dnxhd_mb_var_thread(AVCodecContext *avctx, void *arg)
+static int dnxhd_mb_var_thread(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)
{
- DNXHDEncContext *ctx = *(void**)arg;
- int mb_y, mb_x;
- for (mb_y = ctx->m.start_mb_y; mb_y < ctx->m.end_mb_y; mb_y++) {
- for (mb_x = 0; mb_x < ctx->m.mb_width; mb_x++) {
- unsigned mb = mb_y * ctx->m.mb_width + mb_x;
- uint8_t *pix = ctx->thread[0]->src[0] + ((mb_y<<4) * ctx->m.linesize) + (mb_x<<4);
- int sum = ctx->m.dsp.pix_sum(pix, ctx->m.linesize);
- int varc = (ctx->m.dsp.pix_norm1(pix, ctx->m.linesize) - (((unsigned)(sum*sum))>>8)+128)>>8;
- ctx->mb_cmp[mb].value = varc;
- ctx->mb_cmp[mb].mb = mb;
- }
+ DNXHDEncContext *ctx = avctx->priv_data;
+ int mb_y = jobnr, mb_x;
+ ctx = ctx->thread[threadnr];
+ for (mb_x = 0; mb_x < ctx->m.mb_width; mb_x++) {
+ unsigned mb = mb_y * ctx->m.mb_width + mb_x;
+ uint8_t *pix = ctx->thread[0]->src[0] + ((mb_y<<4) * ctx->m.linesize) + (mb_x<<4);
+ int sum = ctx->m.dsp.pix_sum(pix, ctx->m.linesize);
+ int varc = (ctx->m.dsp.pix_norm1(pix, ctx->m.linesize) - (((unsigned)(sum*sum))>>8)+128)>>8;
+ ctx->mb_cmp[mb].value = varc;
+ ctx->mb_cmp[mb].mb = mb;
}
return 0;
}
for (q = 1; q < avctx->qmax; q++) {
ctx->qscale = q;
- avctx->execute(avctx, dnxhd_calc_bits_thread, &ctx->thread[0], NULL, avctx->thread_count, sizeof(void*));
+ avctx->execute2(avctx, dnxhd_calc_bits_thread, NULL, NULL, ctx->m.mb_height);
}
up_step = down_step = 2<<LAMBDA_FRAC_BITS;
lambda = ctx->lambda;
last_higher = FFMAX(lambda, last_higher);
if (last_lower != INT_MAX)
lambda = (lambda+last_lower)>>1;
+ else if ((int64_t)lambda + up_step > INT_MAX)
+ return -1;
else
lambda += up_step;
- up_step *= 5;
+ up_step = FFMIN((int64_t)up_step*5, INT_MAX);
down_step = 1<<LAMBDA_FRAC_BITS;
}
}
bits = 0;
ctx->qscale = qscale;
// XXX avoid recalculating bits
- ctx->m.avctx->execute(ctx->m.avctx, dnxhd_calc_bits_thread, &ctx->thread[0], NULL, ctx->m.avctx->thread_count, sizeof(void*));
+ ctx->m.avctx->execute2(ctx->m.avctx, dnxhd_calc_bits_thread, NULL, NULL, ctx->m.mb_height);
for (y = 0; y < ctx->m.mb_height; y++) {
for (x = 0; x < ctx->m.mb_width; x++)
bits += ctx->mb_rc[qscale][y*ctx->m.mb_width+x].bits;
}
if (!ret) {
if (RC_VARIANCE)
- avctx->execute(avctx, dnxhd_mb_var_thread, &ctx->thread[0], NULL, avctx->thread_count, sizeof(void*));
+ avctx->execute2(avctx, dnxhd_mb_var_thread, NULL, NULL, ctx->m.mb_height);
radix_sort(ctx->mb_cmp, ctx->m.mb_num);
for (x = 0; x < ctx->m.mb_num && max_bits > ctx->frame_bits; x++) {
int mb = ctx->mb_cmp[x].mb;
else
ret = dnxhd_encode_fast(avctx, ctx);
if (ret < 0) {
- av_log(avctx, AV_LOG_ERROR, "picture could not fit ratecontrol constraints\n");
+ av_log(avctx, AV_LOG_ERROR,
+ "picture could not fit ratecontrol constraints, increase qmax\n");
return -1;
}
- dnxhd_setup_threads_slices(ctx, buf);
+ dnxhd_setup_threads_slices(ctx);
offset = 0;
for (i = 0; i < ctx->m.mb_height; i++) {
assert(!(ctx->slice_size[i] & 3));
}
- avctx->execute(avctx, dnxhd_encode_thread, &ctx->thread[0], NULL, avctx->thread_count, sizeof(void*));
+ avctx->execute2(avctx, dnxhd_encode_thread, buf, NULL, ctx->m.mb_height);
assert(640 + offset + 4 <= ctx->cid_table->coding_unit_size);
memset(buf + 640 + offset, 0, ctx->cid_table->coding_unit_size - 4 - offset - 640);
av_freep(&ctx->mb_rc);
av_freep(&ctx->mb_cmp);
av_freep(&ctx->slice_size);
+ av_freep(&ctx->slice_offs);
av_freep(&ctx->qmatrix_c);
av_freep(&ctx->qmatrix_l);
projects / ffmpeg / blobdiff