static av_cold int init_stage(CUDAScaleContext *s, AVBufferRef *device_ctx)
{
AVBufferRef *out_ref = NULL;
AVHWFramesContext *out_ctx;
int in_sw, in_sh, out_sw, out_sh;
int ret, i;
av_pix_fmt_get_chroma_sub_sample(s->in_fmt, &in_sw, &in_sh);
av_pix_fmt_get_chroma_sub_sample(s->out_fmt, &out_sw, &out_sh);
if (!s->planes_out[0].width) {
s->planes_out[0].width = s->planes_in[0].width;
s->planes_out[0].height = s->planes_in[0].height;
}
for (i = 1; i < FF_ARRAY_ELEMS(s->planes_in); i++) {
s->planes_in[i].width = s->planes_in[0].width >> in_sw;
s->planes_in[i].height = s->planes_in[0].height >> in_sh;
s->planes_out[i].width = s->planes_out[0].width >> out_sw;
s->planes_out[i].height = s->planes_out[0].height >> out_sh;
}
out_ref = av_hwframe_ctx_alloc(device_ctx);
if (!out_ref)
return AVERROR(ENOMEM);
out_ctx = (AVHWFramesContext*)out_ref->data;
out_ctx->format = AV_PIX_FMT_CUDA;
out_ctx->sw_format = s->out_fmt;
out_ctx->width = FFALIGN(s->planes_out[0].width, 32);
out_ctx->height = FFALIGN(s->planes_out[0].height, 32);
ret = av_hwframe_ctx_init(out_ref);
if (ret < 0)
goto fail;
av_frame_unref(s->frame);
ret = av_hwframe_get_buffer(out_ref, s->frame, 0);
if (ret < 0)
goto fail;
s->frame->width = s->planes_out[0].width;
s->frame->height = s->planes_out[0].height;
av_buffer_unref(&s->frames_ctx);
s->frames_ctx = out_ref;
return 0;
fail:
av_buffer_unref(&out_ref);
return ret;
}
void ff_mjpeg_init_hvsample(AVCodecContext *avctx, int hsample[3], int vsample[3])
{
int chroma_h_shift, chroma_v_shift;
av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift,
&chroma_v_shift);
if (avctx->codec->id == AV_CODEC_ID_LJPEG &&
( avctx->pix_fmt == AV_PIX_FMT_BGR0
|| avctx->pix_fmt == AV_PIX_FMT_BGRA
|| avctx->pix_fmt == AV_PIX_FMT_BGR24)) {
vsample[0] = hsample[0] =
vsample[1] = hsample[1] =
vsample[2] = hsample[2] = 1;
} else if (avctx->pix_fmt == AV_PIX_FMT_YUV444P || avctx->pix_fmt == AV_PIX_FMT_YUVJ444P) {
vsample[0] = vsample[1] = vsample[2] = 2;
hsample[0] = hsample[1] = hsample[2] = 1;
} else {
vsample[0] = 2;
vsample[1] = 2 >> chroma_v_shift;
vsample[2] = 2 >> chroma_v_shift;
hsample[0] = 2;
hsample[1] = 2 >> chroma_h_shift;
hsample[2] = 2 >> chroma_h_shift;
}
}
static inline void copy_data(AVFrame *pic, const struct encoder_frame *frame,
int height, enum AVPixelFormat format)
{
int h_chroma_shift, v_chroma_shift;
av_pix_fmt_get_chroma_sub_sample(
format, &h_chroma_shift, &v_chroma_shift);
for (int plane = 0; plane < MAX_AV_PLANES; plane++) {
if (!frame->data[plane])
continue;
int frame_rowsize = (int)frame->linesize[plane];
int pic_rowsize = pic->linesize[plane];
int bytes = frame_rowsize < pic_rowsize ? frame_rowsize
: pic_rowsize;
int plane_height = height >> (plane ? v_chroma_shift : 0);
for (int y = 0; y < plane_height; y++) {
int pos_frame = y * frame_rowsize;
int pos_pic = y * pic_rowsize;
memcpy(pic->data[plane] + pos_pic,
frame->data[plane] + pos_frame, bytes);
}
}
}
static int map_chroma_format(enum AVPixelFormat pix_fmt)
{
int shift_h = 0, shift_v = 0;
av_pix_fmt_get_chroma_sub_sample(pix_fmt, &shift_h, &shift_v);
if (shift_h == 1 && shift_v == 1)
return cudaVideoChromaFormat_420;
else if (shift_h == 1 && shift_v == 0)
return cudaVideoChromaFormat_422;
else if (shift_h == 0 && shift_v == 0)
return cudaVideoChromaFormat_444;
return -1;
}
static int cuda_frames_init(AVHWFramesContext *ctx)
{
CUDAFramesContext *priv = ctx->internal->priv;
int aligned_width = FFALIGN(ctx->width, CUDA_FRAME_ALIGNMENT);
int i;
for (i = 0; i < FF_ARRAY_ELEMS(supported_formats); i++) {
if (ctx->sw_format == supported_formats[i])
break;
}
if (i == FF_ARRAY_ELEMS(supported_formats)) {
av_log(ctx, AV_LOG_ERROR, "Pixel format '%s' is not supported\n",
av_get_pix_fmt_name(ctx->sw_format));
return AVERROR(ENOSYS);
}
av_pix_fmt_get_chroma_sub_sample(ctx->sw_format, &priv->shift_width, &priv->shift_height);
if (!ctx->pool) {
int size;
switch (ctx->sw_format) {
case AV_PIX_FMT_NV12:
case AV_PIX_FMT_YUV420P:
size = aligned_width * ctx->height * 3 / 2;
break;
case AV_PIX_FMT_YUV444P:
case AV_PIX_FMT_P010:
case AV_PIX_FMT_P016:
size = aligned_width * ctx->height * 3;
break;
case AV_PIX_FMT_YUV444P16:
size = aligned_width * ctx->height * 6;
break;
default:
av_log(ctx, AV_LOG_ERROR, "BUG: Pixel format missing from size calculation.");
return AVERROR_BUG;
}
ctx->internal->pool_internal = av_buffer_pool_init2(size, ctx, cuda_pool_alloc, NULL);
if (!ctx->internal->pool_internal)
return AVERROR(ENOMEM);
}
return 0;
}
static int get_vpx_chroma_subsampling(AVFormatContext *s,
enum AVPixelFormat pixel_format,
enum AVChromaLocation chroma_location)
{
int chroma_w, chroma_h;
if (av_pix_fmt_get_chroma_sub_sample(pixel_format, &chroma_w, &chroma_h) == 0) {
if (chroma_w == 1 && chroma_h == 1) {
return (chroma_location == AVCHROMA_LOC_LEFT)
? VPX_SUBSAMPLING_420_VERTICAL
: VPX_SUBSAMPLING_420_COLLOCATED_WITH_LUMA;
} else if (chroma_w == 1 && chroma_h == 0) {
return VPX_SUBSAMPLING_422;
} else if (chroma_w == 0 && chroma_h == 0) {
return VPX_SUBSAMPLING_444;
}
}
av_log(s, AV_LOG_ERROR, "Unsupported pixel format (%d)\n", pixel_format);
return -1;
}
void video_effect_show_title_safe(struct raw_frame_ref dest)
{
int chroma_shift_horiz, chroma_shift_vert;
av_pix_fmt_get_chroma_sub_sample(dest.pix_fmt,
&chroma_shift_horiz, &chroma_shift_vert);
unsigned width = FRAME_WIDTH;
unsigned height = dest.height;
unsigned border_horiz = (FRAME_WIDTH + 5) / 10;
unsigned border_vert = (dest.height + 5) / 10;
unsigned bias = luma_bias;
// Darken the non-title-safe area
for (int plane = 0; plane != 3; ++plane)
{
if (plane == 1)
{
width >>= chroma_shift_horiz;
border_horiz >>= chroma_shift_horiz;
height >>= chroma_shift_vert;
border_vert >>= chroma_shift_vert;
bias = chroma_bias;
}
for (unsigned y = 0; y != height; ++y)
{
uint8_t * p, * end;
// Do left border
p = dest.planes.data[plane] + dest.planes.linesize[plane] * y;
end = p + border_horiz;
while (p != end)
*p = (*p + bias) / 2, ++p;
end = p + width - border_horiz;
if (y >= border_vert && y < height - border_vert)
// Skip to right border
p += width - 2 * border_horiz;
// else continue across top border or bottom border
while (p != end)
*p = (*p + bias) / 2, ++p;
}
}
static int amv_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
const AVFrame *pic_arg, int *got_packet)
{
MpegEncContext *s = avctx->priv_data;
AVFrame *pic;
int i, ret;
int chroma_h_shift, chroma_v_shift;
av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
//CODEC_FLAG_EMU_EDGE have to be cleared
if(s->avctx->flags & CODEC_FLAG_EMU_EDGE)
return AVERROR(EINVAL);
if ((avctx->height & 15) && avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL) {
av_log(avctx, AV_LOG_ERROR,
"Heights which are not a multiple of 16 might fail with some decoders, "
"use vstrict=-1 / -strict -1 to use %d anyway.\n", avctx->height);
av_log(avctx, AV_LOG_WARNING, "If you have a device that plays AMV videos, please test if videos "
"with such heights work with it and report your findings to [email protected]\n");
return AVERROR_EXPERIMENTAL;
}
pic = av_frame_clone(pic_arg);
if (!pic)
return AVERROR(ENOMEM);
//picture should be flipped upside-down
for(i=0; i < 3; i++) {
int vsample = i ? 2 >> chroma_v_shift : 2;
pic->data[i] += pic->linesize[i] * (vsample * s->height / V_MAX - 1);
pic->linesize[i] *= -1;
}
ret = ff_mpv_encode_picture(avctx, pkt, pic, got_packet);
av_frame_free(&pic);
return ret;
}
static av_cold int encode_init(AVCodecContext* avc_context)
{
th_info t_info;
th_comment t_comment;
ogg_packet o_packet;
unsigned int offset;
TheoraContext *h = avc_context->priv_data;
uint32_t gop_size = avc_context->gop_size;
/* Set up the theora_info struct */
th_info_init(&t_info);
t_info.frame_width = FFALIGN(avc_context->width, 16);
t_info.frame_height = FFALIGN(avc_context->height, 16);
t_info.pic_width = avc_context->width;
t_info.pic_height = avc_context->height;
t_info.pic_x = 0;
t_info.pic_y = 0;
/* Swap numerator and denominator as time_base in AVCodecContext gives the
* time period between frames, but theora_info needs the framerate. */
t_info.fps_numerator = avc_context->time_base.den;
t_info.fps_denominator = avc_context->time_base.num;
if (avc_context->sample_aspect_ratio.num) {
t_info.aspect_numerator = avc_context->sample_aspect_ratio.num;
t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
} else {
t_info.aspect_numerator = 1;
t_info.aspect_denominator = 1;
}
if (avc_context->color_primaries == AVCOL_PRI_BT470M)
t_info.colorspace = TH_CS_ITU_REC_470M;
else if (avc_context->color_primaries == AVCOL_PRI_BT470BG)
t_info.colorspace = TH_CS_ITU_REC_470BG;
else
t_info.colorspace = TH_CS_UNSPECIFIED;
if (avc_context->pix_fmt == AV_PIX_FMT_YUV420P)
t_info.pixel_fmt = TH_PF_420;
else if (avc_context->pix_fmt == AV_PIX_FMT_YUV422P)
t_info.pixel_fmt = TH_PF_422;
else if (avc_context->pix_fmt == AV_PIX_FMT_YUV444P)
t_info.pixel_fmt = TH_PF_444;
else {
av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
return -1;
}
av_pix_fmt_get_chroma_sub_sample(avc_context->pix_fmt,
&h->uv_hshift, &h->uv_vshift);
if (avc_context->flags & CODEC_FLAG_QSCALE) {
/* to be constant with the libvorbis implementation, clip global_quality to 0 - 10
Theora accepts a quality parameter p, which is:
* 0 <= p <=63
* an int value
*/
t_info.quality = av_clipf(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
t_info.target_bitrate = 0;
} else {
t_info.target_bitrate = avc_context->bit_rate;
t_info.quality = 0;
}
/* Now initialise libtheora */
h->t_state = th_encode_alloc(&t_info);
if (!h->t_state) {
av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
return -1;
}
h->keyframe_mask = (1 << t_info.keyframe_granule_shift) - 1;
/* Clear up theora_info struct */
th_info_clear(&t_info);
if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
&gop_size, sizeof(gop_size))) {
av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
return -1;
}
// need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
if (avc_context->flags & CODEC_FLAG_PASS1) {
if (get_stats(avc_context, 0))
return -1;
} else if (avc_context->flags & CODEC_FLAG_PASS2) {
if (submit_stats(avc_context))
return -1;
}
/*
Output first header packet consisting of theora
header, comment, and tables.
Each one is prefixed with a 16bit size, then they
are concatenated together into libavcodec's extradata.
*/
offset = 0;
/* Headers */
th_comment_init(&t_comment);
while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
if (concatenate_packet(&offset, avc_context, &o_packet))
return -1;
th_comment_clear(&t_comment);
/* Set up the output AVFrame */
avc_context->coded_frame = av_frame_alloc();
return 0;
}
static int yuv4_write_packet(AVFormatContext *s, AVPacket *pkt)
{
AVStream *st = s->streams[pkt->stream_index];
AVIOContext *pb = s->pb;
AVFrame *frame;
int* first_pkt = s->priv_data;
int width, height, h_chroma_shift, v_chroma_shift;
int i;
char buf2[Y4M_LINE_MAX + 1];
uint8_t *ptr, *ptr1, *ptr2;
frame = (AVFrame *)pkt->data;
/* for the first packet we have to output the header as well */
if (*first_pkt) {
*first_pkt = 0;
if (yuv4_generate_header(s, buf2) < 0) {
av_log(s, AV_LOG_ERROR,
"Error. YUV4MPEG stream header write failed.\n");
return AVERROR(EIO);
} else {
avio_write(pb, buf2, strlen(buf2));
}
}
/* construct frame header */
avio_printf(s->pb, "%s\n", Y4M_FRAME_MAGIC);
width = st->codec->width;
height = st->codec->height;
ptr = frame->data[0];
switch (st->codec->pix_fmt) {
case AV_PIX_FMT_GRAY8:
case AV_PIX_FMT_YUV411P:
case AV_PIX_FMT_YUV420P:
case AV_PIX_FMT_YUV422P:
case AV_PIX_FMT_YUV444P:
break;
case AV_PIX_FMT_GRAY16:
case AV_PIX_FMT_YUV420P9:
case AV_PIX_FMT_YUV422P9:
case AV_PIX_FMT_YUV444P9:
case AV_PIX_FMT_YUV420P10:
case AV_PIX_FMT_YUV422P10:
case AV_PIX_FMT_YUV444P10:
case AV_PIX_FMT_YUV420P12:
case AV_PIX_FMT_YUV422P12:
case AV_PIX_FMT_YUV444P12:
case AV_PIX_FMT_YUV420P14:
case AV_PIX_FMT_YUV422P14:
case AV_PIX_FMT_YUV444P14:
case AV_PIX_FMT_YUV420P16:
case AV_PIX_FMT_YUV422P16:
case AV_PIX_FMT_YUV444P16:
width *= 2;
break;
default:
av_log(s, AV_LOG_ERROR, "The pixel format '%s' is not supported.\n",
av_get_pix_fmt_name(st->codec->pix_fmt));
return AVERROR(EINVAL);
}
for (i = 0; i < height; i++) {
avio_write(pb, ptr, width);
ptr += frame->linesize[0];
}
if (st->codec->pix_fmt != AV_PIX_FMT_GRAY8 &&
st->codec->pix_fmt != AV_PIX_FMT_GRAY16) {
// Adjust for smaller Cb and Cr planes
av_pix_fmt_get_chroma_sub_sample(st->codec->pix_fmt, &h_chroma_shift,
&v_chroma_shift);
width = FF_CEIL_RSHIFT(width, h_chroma_shift);
height = FF_CEIL_RSHIFT(height, v_chroma_shift);
ptr1 = frame->data[1];
ptr2 = frame->data[2];
for (i = 0; i < height; i++) { /* Cb */
avio_write(pb, ptr1, width);
ptr1 += frame->linesize[1];
}
for (i = 0; i < height; i++) { /* Cr */
avio_write(pb, ptr2, width);
ptr2 += frame->linesize[2];
}
}
return 0;
}
static int yuv4_write_packet(AVFormatContext *s, AVPacket *pkt)
{
AVStream *st = s->streams[pkt->stream_index];
AVIOContext *pb = s->pb;
AVFrame *frame;
int* first_pkt = s->priv_data;
int width, height, h_chroma_shift, v_chroma_shift;
int i;
char buf2[Y4M_LINE_MAX + 1];
char buf1[20];
uint8_t *ptr, *ptr1, *ptr2;
frame = (AVFrame *)pkt->data;
/* for the first packet we have to output the header as well */
if (*first_pkt) {
*first_pkt = 0;
if (yuv4_generate_header(s, buf2) < 0) {
av_log(s, AV_LOG_ERROR,
"Error. YUV4MPEG stream header write failed.\n");
return AVERROR(EIO);
} else {
avio_write(pb, buf2, strlen(buf2));
}
}
/* construct frame header */
snprintf(buf1, sizeof(buf1), "%s\n", Y4M_FRAME_MAGIC);
avio_write(pb, buf1, strlen(buf1));
width = st->codec->width;
height = st->codec->height;
ptr = frame->data[0];
for (i = 0; i < height; i++) {
avio_write(pb, ptr, width);
ptr += frame->linesize[0];
}
if (st->codec->pix_fmt != AV_PIX_FMT_GRAY8) {
// Adjust for smaller Cb and Cr planes
av_pix_fmt_get_chroma_sub_sample(st->codec->pix_fmt, &h_chroma_shift,
&v_chroma_shift);
// Shift right, rounding up
width = AV_CEIL_RSHIFT(width, h_chroma_shift);
height = AV_CEIL_RSHIFT(height, v_chroma_shift);
ptr1 = frame->data[1];
ptr2 = frame->data[2];
for (i = 0; i < height; i++) { /* Cb */
avio_write(pb, ptr1, width);
ptr1 += frame->linesize[1];
}
for (i = 0; i < height; i++) { /* Cr */
avio_write(pb, ptr2, width);
ptr2 += frame->linesize[2];
}
}
return 0;
}
