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 vdpau_frames_init(AVHWFramesContext *ctx)
{
VDPAUDeviceContext *device_priv = ctx->device_ctx->internal->priv;
VDPAUFramesContext *priv = ctx->internal->priv;
int i;
switch (ctx->sw_format) {
case AV_PIX_FMT_YUV420P: priv->chroma_type = VDP_CHROMA_TYPE_420; break;
case AV_PIX_FMT_YUV422P: priv->chroma_type = VDP_CHROMA_TYPE_422; break;
case AV_PIX_FMT_YUV444P: priv->chroma_type = VDP_CHROMA_TYPE_444; break;
default:
av_log(ctx, AV_LOG_ERROR, "Unsupported data layout: %s\n",
av_get_pix_fmt_name(ctx->sw_format));
return AVERROR(ENOSYS);
}
for (i = 0; i < FF_ARRAY_ELEMS(vdpau_pix_fmts); i++) {
if (vdpau_pix_fmts[i].chroma_type == priv->chroma_type) {
priv->chroma_idx = i;
priv->pix_fmts = device_priv->pix_fmts[i];
priv->nb_pix_fmts = device_priv->nb_pix_fmts[i];
break;
}
}
if (!priv->pix_fmts) {
av_log(ctx, AV_LOG_ERROR, "Unsupported chroma type: %d\n", priv->chroma_type);
return AVERROR(ENOSYS);
}
if (!ctx->pool) {
ctx->internal->pool_internal = av_buffer_pool_init2(sizeof(VdpVideoSurface), ctx,
vdpau_pool_alloc, NULL);
if (!ctx->internal->pool_internal)
return AVERROR(ENOMEM);
}
priv->get_data = device_priv->get_data;
priv->put_data = device_priv->put_data;
return 0;
}
static int qsv_init_pool(AVHWFramesContext *ctx, uint32_t fourcc)
{
QSVFramesContext *s = ctx->internal->priv;
AVQSVFramesContext *frames_hwctx = ctx->hwctx;
int i, ret = 0;
if (ctx->initial_pool_size <= 0) {
av_log(ctx, AV_LOG_ERROR, "QSV requires a fixed frame pool size\n");
return AVERROR(EINVAL);
}
s->surfaces_internal = av_mallocz_array(ctx->initial_pool_size,
sizeof(*s->surfaces_internal));
if (!s->surfaces_internal)
return AVERROR(ENOMEM);
for (i = 0; i < ctx->initial_pool_size; i++) {
ret = qsv_init_surface(ctx, &s->surfaces_internal[i]);
if (ret < 0)
return ret;
}
if (!(frames_hwctx->frame_type & MFX_MEMTYPE_OPAQUE_FRAME)) {
ret = qsv_init_child_ctx(ctx);
if (ret < 0)
return ret;
}
ctx->internal->pool_internal = av_buffer_pool_init2(sizeof(mfxFrameSurface1),
ctx, qsv_pool_alloc, NULL);
if (!ctx->internal->pool_internal)
return AVERROR(ENOMEM);
frames_hwctx->surfaces = s->surfaces_internal;
frames_hwctx->nb_surfaces = ctx->initial_pool_size;
return 0;
}
static int dxva2_init_pool(AVHWFramesContext *ctx)
{
AVDXVA2FramesContext *frames_hwctx = ctx->hwctx;
AVDXVA2DeviceContext *device_hwctx = ctx->device_ctx->hwctx;
DXVA2FramesContext *s = ctx->internal->priv;
int decode = (frames_hwctx->surface_type == DXVA2_VideoDecoderRenderTarget);
int i;
HRESULT hr;
if (ctx->initial_pool_size <= 0)
return 0;
hr = IDirect3DDeviceManager9_OpenDeviceHandle(device_hwctx->devmgr, &s->device_handle);
if (FAILED(hr)) {
av_log(ctx, AV_LOG_ERROR, "Failed to open device handle\n");
return AVERROR_UNKNOWN;
}
hr = IDirect3DDeviceManager9_GetVideoService(device_hwctx->devmgr,
s->device_handle,
decode ? &video_decoder_service : &video_processor_service,
(void **)&s->service);
if (FAILED(hr)) {
av_log(ctx, AV_LOG_ERROR, "Failed to create the video service\n");
return AVERROR_UNKNOWN;
}
for (i = 0; i < FF_ARRAY_ELEMS(supported_formats); i++) {
if (ctx->sw_format == supported_formats[i].pix_fmt) {
s->format = supported_formats[i].d3d_format;
break;
}
}
if (i == FF_ARRAY_ELEMS(supported_formats)) {
av_log(ctx, AV_LOG_ERROR, "Unsupported pixel format: %s\n",
av_get_pix_fmt_name(ctx->sw_format));
return AVERROR(EINVAL);
}
s->surfaces_internal = av_mallocz_array(ctx->initial_pool_size,
sizeof(*s->surfaces_internal));
if (!s->surfaces_internal)
return AVERROR(ENOMEM);
hr = IDirectXVideoAccelerationService_CreateSurface(s->service,
ctx->width, ctx->height,
ctx->initial_pool_size - 1,
s->format, D3DPOOL_DEFAULT, 0,
frames_hwctx->surface_type,
s->surfaces_internal, NULL);
if (FAILED(hr)) {
av_log(ctx, AV_LOG_ERROR, "Could not create the surfaces\n");
return AVERROR_UNKNOWN;
}
ctx->internal->pool_internal = av_buffer_pool_init2(sizeof(*s->surfaces_internal),
ctx, dxva2_pool_alloc, NULL);
if (!ctx->internal->pool_internal)
return AVERROR(ENOMEM);
frames_hwctx->surfaces = s->surfaces_internal;
frames_hwctx->nb_surfaces = ctx->initial_pool_size;
return 0;
}
int ff_nvdec_decode_init(AVCodecContext *avctx)
{
NVDECContext *ctx = avctx->internal->hwaccel_priv_data;
NVDECFramePool *pool;
AVHWFramesContext *frames_ctx;
const AVPixFmtDescriptor *sw_desc;
CUVIDDECODECREATEINFO params = { 0 };
int cuvid_codec_type, cuvid_chroma_format;
int ret = 0;
sw_desc = av_pix_fmt_desc_get(avctx->sw_pix_fmt);
if (!sw_desc)
return AVERROR_BUG;
cuvid_codec_type = map_avcodec_id(avctx->codec_id);
if (cuvid_codec_type < 0) {
av_log(avctx, AV_LOG_ERROR, "Unsupported codec ID\n");
return AVERROR_BUG;
}
cuvid_chroma_format = map_chroma_format(avctx->sw_pix_fmt);
if (cuvid_chroma_format < 0) {
av_log(avctx, AV_LOG_ERROR, "Unsupported chroma format\n");
return AVERROR(ENOSYS);
}
if (!avctx->hw_frames_ctx) {
ret = ff_decode_get_hw_frames_ctx(avctx, AV_HWDEVICE_TYPE_CUDA);
if (ret < 0)
return ret;
}
frames_ctx = (AVHWFramesContext*)avctx->hw_frames_ctx->data;
params.ulWidth = avctx->coded_width;
params.ulHeight = avctx->coded_height;
params.ulTargetWidth = avctx->coded_width;
params.ulTargetHeight = avctx->coded_height;
params.bitDepthMinus8 = sw_desc->comp[0].depth - 8;
params.OutputFormat = params.bitDepthMinus8 ?
cudaVideoSurfaceFormat_P016 : cudaVideoSurfaceFormat_NV12;
params.CodecType = cuvid_codec_type;
params.ChromaFormat = cuvid_chroma_format;
params.ulNumDecodeSurfaces = frames_ctx->initial_pool_size;
params.ulNumOutputSurfaces = frames_ctx->initial_pool_size;
ret = nvdec_decoder_create(&ctx->decoder_ref, frames_ctx->device_ref, ¶ms, avctx);
if (ret < 0) {
if (params.ulNumDecodeSurfaces > 32) {
av_log(avctx, AV_LOG_WARNING, "Using more than 32 (%d) decode surfaces might cause nvdec to fail.\n",
(int)params.ulNumDecodeSurfaces);
av_log(avctx, AV_LOG_WARNING, "Try lowering the amount of threads. Using %d right now.\n",
avctx->thread_count);
}
return ret;
}
pool = av_mallocz(sizeof(*pool));
if (!pool) {
ret = AVERROR(ENOMEM);
goto fail;
}
pool->dpb_size = frames_ctx->initial_pool_size;
ctx->decoder_pool = av_buffer_pool_init2(sizeof(int), pool,
nvdec_decoder_frame_alloc, av_free);
if (!ctx->decoder_pool) {
ret = AVERROR(ENOMEM);
goto fail;
}
return 0;
fail:
ff_nvdec_decode_uninit(avctx);
return ret;
}
static int qsv_init_pool(AVHWFramesContext *ctx, uint32_t fourcc)
{
QSVFramesContext *s = ctx->internal->priv;
AVQSVFramesContext *frames_hwctx = ctx->hwctx;
const AVPixFmtDescriptor *desc;
int i, ret = 0;
desc = av_pix_fmt_desc_get(ctx->sw_format);
if (!desc)
return AVERROR_BUG;
if (ctx->initial_pool_size <= 0) {
av_log(ctx, AV_LOG_ERROR, "QSV requires a fixed frame pool size\n");
return AVERROR(EINVAL);
}
s->surfaces_internal = av_mallocz_array(ctx->initial_pool_size,
sizeof(*s->surfaces_internal));
if (!s->surfaces_internal)
return AVERROR(ENOMEM);
for (i = 0; i < ctx->initial_pool_size; i++) {
mfxFrameSurface1 *surf = &s->surfaces_internal[i];
surf->Info.BitDepthLuma = desc->comp[0].depth;
surf->Info.BitDepthChroma = desc->comp[0].depth;
surf->Info.Shift = desc->comp[0].depth > 8;
if (desc->log2_chroma_w && desc->log2_chroma_h)
surf->Info.ChromaFormat = MFX_CHROMAFORMAT_YUV420;
else if (desc->log2_chroma_w)
surf->Info.ChromaFormat = MFX_CHROMAFORMAT_YUV422;
else
surf->Info.ChromaFormat = MFX_CHROMAFORMAT_YUV444;
surf->Info.FourCC = fourcc;
surf->Info.Width = ctx->width;
surf->Info.CropW = ctx->width;
surf->Info.Height = ctx->height;
surf->Info.CropH = ctx->height;
surf->Info.FrameRateExtN = 25;
surf->Info.FrameRateExtD = 1;
}
if (!(frames_hwctx->frame_type & MFX_MEMTYPE_OPAQUE_FRAME)) {
ret = qsv_init_child_ctx(ctx);
if (ret < 0)
return ret;
}
ctx->internal->pool_internal = av_buffer_pool_init2(sizeof(mfxFrameSurface1),
ctx, qsv_pool_alloc, NULL);
if (!ctx->internal->pool_internal)
return AVERROR(ENOMEM);
frames_hwctx->surfaces = s->surfaces_internal;
frames_hwctx->nb_surfaces = ctx->initial_pool_size;
return 0;
}
static int vaapi_frames_init(AVHWFramesContext *hwfc)
{
AVVAAPIFramesContext *avfc = hwfc->hwctx;
VAAPIFramesContext *ctx = hwfc->internal->priv;
AVVAAPIDeviceContext *hwctx = hwfc->device_ctx->hwctx;
VAImageFormat *expected_format;
AVBufferRef *test_surface = NULL;
VASurfaceID test_surface_id;
VAImage test_image;
VAStatus vas;
int err, i;
unsigned int fourcc, rt_format;
for (i = 0; i < FF_ARRAY_ELEMS(vaapi_format_map); i++) {
if (vaapi_format_map[i].pix_fmt == hwfc->sw_format) {
fourcc = vaapi_format_map[i].fourcc;
rt_format = vaapi_format_map[i].rt_format;
break;
}
}
if (i >= FF_ARRAY_ELEMS(vaapi_format_map)) {
av_log(hwfc, AV_LOG_ERROR, "Unsupported format: %s.\n",
av_get_pix_fmt_name(hwfc->sw_format));
return AVERROR(EINVAL);
}
if (!hwfc->pool) {
int need_memory_type = 1, need_pixel_format = 1;
for (i = 0; i < avfc->nb_attributes; i++) {
if (ctx->attributes[i].type == VASurfaceAttribMemoryType)
need_memory_type = 0;
if (ctx->attributes[i].type == VASurfaceAttribPixelFormat)
need_pixel_format = 0;
}
ctx->nb_attributes =
avfc->nb_attributes + need_memory_type + need_pixel_format;
ctx->attributes = av_malloc(ctx->nb_attributes *
sizeof(*ctx->attributes));
if (!ctx->attributes) {
err = AVERROR(ENOMEM);
goto fail;
}
for (i = 0; i < avfc->nb_attributes; i++)
ctx->attributes[i] = avfc->attributes[i];
if (need_memory_type) {
ctx->attributes[i++] = (VASurfaceAttrib) {
.type = VASurfaceAttribMemoryType,
.flags = VA_SURFACE_ATTRIB_SETTABLE,
.value.type = VAGenericValueTypeInteger,
.value.value.i = VA_SURFACE_ATTRIB_MEM_TYPE_VA,
};
}
if (need_pixel_format) {
ctx->attributes[i++] = (VASurfaceAttrib) {
.type = VASurfaceAttribPixelFormat,
.flags = VA_SURFACE_ATTRIB_SETTABLE,
.value.type = VAGenericValueTypeInteger,
.value.value.i = fourcc,
};
}
av_assert0(i == ctx->nb_attributes);
ctx->rt_format = rt_format;
if (hwfc->initial_pool_size > 0) {
// This pool will be usable as a render target, so we need to store
// all of the surface IDs somewhere that vaCreateContext() calls
// will be able to access them.
avfc->nb_surfaces = 0;
avfc->surface_ids = av_malloc(hwfc->initial_pool_size *
sizeof(*avfc->surface_ids));
if (!avfc->surface_ids) {
err = AVERROR(ENOMEM);
goto fail;
}
} else {
// This pool allows dynamic sizing, and will not be usable as a
// render target.
avfc->nb_surfaces = 0;
avfc->surface_ids = NULL;
}
hwfc->internal->pool_internal =
av_buffer_pool_init2(sizeof(VASurfaceID), hwfc,
&vaapi_pool_alloc, NULL);
if (!hwfc->internal->pool_internal) {
av_log(hwfc, AV_LOG_ERROR, "Failed to create VAAPI surface pool.\n");
err = AVERROR(ENOMEM);
goto fail;
}
}
// Allocate a single surface to test whether vaDeriveImage() is going
// to work for the specific configuration.
if (hwfc->pool) {
test_surface = av_buffer_pool_get(hwfc->pool);
if (!test_surface) {
av_log(hwfc, AV_LOG_ERROR, "Unable to allocate a surface from "
"user-configured buffer pool.\n");
err = AVERROR(ENOMEM);
goto fail;
}
} else {
test_surface = av_buffer_pool_get(hwfc->internal->pool_internal);
if (!test_surface) {
av_log(hwfc, AV_LOG_ERROR, "Unable to allocate a surface from "
"internal buffer pool.\n");
err = AVERROR(ENOMEM);
goto fail;
}
}
test_surface_id = (VASurfaceID)(uintptr_t)test_surface->data;
ctx->derive_works = 0;
err = vaapi_get_image_format(hwfc->device_ctx,
hwfc->sw_format, &expected_format);
if (err == 0) {
vas = vaDeriveImage(hwctx->display, test_surface_id, &test_image);
if (vas == VA_STATUS_SUCCESS) {
if (expected_format->fourcc == test_image.format.fourcc) {
av_log(hwfc, AV_LOG_DEBUG, "Direct mapping possible.\n");
ctx->derive_works = 1;
} else {
av_log(hwfc, AV_LOG_DEBUG, "Direct mapping disabled: "
"derived image format %08x does not match "
"expected format %08x.\n",
expected_format->fourcc, test_image.format.fourcc);
}
vaDestroyImage(hwctx->display, test_image.image_id);
} else {
av_log(hwfc, AV_LOG_DEBUG, "Direct mapping disabled: "
"deriving image does not work: "
"%d (%s).\n", vas, vaErrorStr(vas));
}
} else {
av_log(hwfc, AV_LOG_DEBUG, "Direct mapping disabled: "
"image format is not supported.\n");
}
av_buffer_unref(&test_surface);
return 0;
fail:
av_buffer_unref(&test_surface);
av_freep(&avfc->surface_ids);
av_freep(&ctx->attributes);
return err;
}
static void vaapi_frames_uninit(AVHWFramesContext *hwfc)
{
AVVAAPIFramesContext *avfc = hwfc->hwctx;
VAAPIFramesContext *ctx = hwfc->internal->priv;
av_freep(&avfc->surface_ids);
av_freep(&ctx->attributes);
}
