bool DXVA2Decoder::CreateDecoder(void)
{
if (!m_service || !m_context.surface)
return false;
HRESULT hr = IDirectXVideoDecoderService_CreateVideoDecoder(
m_service, m_input, &m_format,
const_cast<DXVA2_ConfigPictureDecode*>(m_context.cfg),
m_context.surface, m_context.surface_count, &m_context.decoder);
if (FAILED(hr))
return false;
LOG(VB_PLAYBACK, LOG_INFO, LOC +
QString("Created decoder: %1->%2x%3 (%4 surfaces)")
.arg(toString(m_codec_id)).arg(m_width).arg(m_height)
.arg(m_context.surface_count));
return true;
}
/**
* It creates a DXVA2 decoder using the given video format
*/
static int DxCreateVideoDecoder(vlc_va_t *va, int codec_id, const video_format_t *fmt, bool b_threading)
{
VLC_UNUSED(b_threading);
vlc_va_sys_t *p_sys = va->sys;
directx_sys_t *sys = &va->sys->dx_sys;
if (FAILED(IDirectXVideoDecoderService_CreateSurface((IDirectXVideoDecoderService*) sys->d3ddec,
sys->surface_width,
sys->surface_height,
sys->surface_count - 1,
p_sys->render,
D3DPOOL_DEFAULT,
0,
DXVA2_VideoDecoderRenderTarget,
(LPDIRECT3DSURFACE9*) sys->hw_surface,
NULL))) {
msg_Err(va, "IDirectXVideoAccelerationService_CreateSurface failed");
sys->surface_count = 0;
return VLC_EGENERIC;
}
msg_Dbg(va, "IDirectXVideoAccelerationService_CreateSurface succeed with %d surfaces (%dx%d)",
sys->surface_count, sys->surface_width, sys->surface_height);
/* */
DXVA2_VideoDesc dsc;
ZeroMemory(&dsc, sizeof(dsc));
dsc.SampleWidth = fmt->i_width;
dsc.SampleHeight = fmt->i_height;
dsc.Format = p_sys->render;
if (fmt->i_frame_rate > 0 && fmt->i_frame_rate_base > 0) {
dsc.InputSampleFreq.Numerator = fmt->i_frame_rate;
dsc.InputSampleFreq.Denominator = fmt->i_frame_rate_base;
} else {
dsc.InputSampleFreq.Numerator = 0;
dsc.InputSampleFreq.Denominator = 0;
}
dsc.OutputFrameFreq = dsc.InputSampleFreq;
dsc.UABProtectionLevel = FALSE;
dsc.Reserved = 0;
/* FIXME I am unsure we can let unknown everywhere */
DXVA2_ExtendedFormat *ext = &dsc.SampleFormat;
ext->SampleFormat = 0;//DXVA2_SampleUnknown;
ext->VideoChromaSubsampling = 0;//DXVA2_VideoChromaSubsampling_Unknown;
ext->NominalRange = 0;//DXVA2_NominalRange_Unknown;
ext->VideoTransferMatrix = 0;//DXVA2_VideoTransferMatrix_Unknown;
ext->VideoLighting = 0;//DXVA2_VideoLighting_Unknown;
ext->VideoPrimaries = 0;//DXVA2_VideoPrimaries_Unknown;
ext->VideoTransferFunction = 0;//DXVA2_VideoTransFunc_Unknown;
/* List all configurations available for the decoder */
UINT cfg_count = 0;
DXVA2_ConfigPictureDecode *cfg_list = NULL;
if (FAILED(IDirectXVideoDecoderService_GetDecoderConfigurations((IDirectXVideoDecoderService*) sys->d3ddec,
&sys->input,
&dsc,
NULL,
&cfg_count,
&cfg_list))) {
msg_Err(va, "IDirectXVideoDecoderService_GetDecoderConfigurations failed");
return VLC_EGENERIC;
}
msg_Dbg(va, "we got %d decoder configurations", cfg_count);
/* Select the best decoder configuration */
int cfg_score = 0;
for (unsigned i = 0; i < cfg_count; i++) {
const DXVA2_ConfigPictureDecode *cfg = &cfg_list[i];
/* */
msg_Dbg(va, "configuration[%d] ConfigBitstreamRaw %d",
i, cfg->ConfigBitstreamRaw);
/* */
int score;
if (cfg->ConfigBitstreamRaw == 1)
score = 1;
else if (codec_id == AV_CODEC_ID_H264 && cfg->ConfigBitstreamRaw == 2)
score = 2;
else
continue;
if (IsEqualGUID(&cfg->guidConfigBitstreamEncryption, &DXVA_NoEncrypt))
score += 16;
if (cfg_score < score) {
p_sys->cfg = *cfg;
cfg_score = score;
}
}
CoTaskMemFree(cfg_list);
if (cfg_score <= 0) {
msg_Err(va, "Failed to find a supported decoder configuration");
return VLC_EGENERIC;
}
/* Create the decoder */
IDirectXVideoDecoder *decoder;
if (FAILED(IDirectXVideoDecoderService_CreateVideoDecoder((IDirectXVideoDecoderService*) sys->d3ddec,
&sys->input,
&dsc,
&p_sys->cfg,
(LPDIRECT3DSURFACE9*) sys->hw_surface,
sys->surface_count,
&decoder))) {
msg_Err(va, "IDirectXVideoDecoderService_CreateVideoDecoder failed");
return VLC_EGENERIC;
}
sys->decoder = (IUnknown*) decoder;
msg_Dbg(va, "IDirectXVideoDecoderService_CreateVideoDecoder succeed");
return VLC_SUCCESS;
}
/**
* It creates a DXVA2 decoder using the given video format
*/
static int DxCreateVideoDecoder(vlc_va_dxva2_t *va,
int codec_id, const video_format_t *fmt)
{
/* */
msg_Dbg(va->log, "DxCreateVideoDecoder id %d %dx%d",
codec_id, fmt->i_width, fmt->i_height);
va->width = fmt->i_width;
va->height = fmt->i_height;
/* Allocates all surfaces needed for the decoder */
va->surface_width = (fmt->i_width + 15) & ~15;
va->surface_height = (fmt->i_height + 15) & ~15;
int surface_count;
switch (codec_id) {
case AV_CODEC_ID_H264:
surface_count = 16 + 1 + 2;
break;
default:
surface_count = 2 + 1;
break;
}
if (surface_count > VA_DXVA2_MAX_SURFACE_COUNT)
return VLC_EGENERIC;
va->surface_count = surface_count;
if (FAILED(IDirectXVideoDecoderService_CreateSurface(va->vs,
va->surface_width,
va->surface_height,
va->surface_count - 1,
va->render,
D3DPOOL_DEFAULT,
0,
DXVA2_VideoDecoderRenderTarget,
va->hw_surface,
NULL))) {
msg_Err(va->log, "IDirectXVideoAccelerationService_CreateSurface failed");
va->surface_count = 0;
return VLC_EGENERIC;
}
for (unsigned i = 0; i < va->surface_count; i++) {
vlc_va_surface_t *surface = &va->surface[i];
surface->d3d = va->hw_surface[i];
surface->refcount = 0;
surface->order = 0;
}
msg_Dbg(va->log, "IDirectXVideoAccelerationService_CreateSurface succeed with %d surfaces (%dx%d)",
va->surface_count, fmt->i_width, fmt->i_height);
/* */
DXVA2_VideoDesc dsc;
ZeroMemory(&dsc, sizeof(dsc));
dsc.SampleWidth = fmt->i_width;
dsc.SampleHeight = fmt->i_height;
dsc.Format = va->render;
if (fmt->i_frame_rate > 0 && fmt->i_frame_rate_base > 0) {
dsc.InputSampleFreq.Numerator = fmt->i_frame_rate;
dsc.InputSampleFreq.Denominator = fmt->i_frame_rate_base;
} else {
dsc.InputSampleFreq.Numerator = 0;
dsc.InputSampleFreq.Denominator = 0;
}
dsc.OutputFrameFreq = dsc.InputSampleFreq;
dsc.UABProtectionLevel = FALSE;
dsc.Reserved = 0;
/* FIXME I am unsure we can let unknown everywhere */
DXVA2_ExtendedFormat *ext = &dsc.SampleFormat;
ext->SampleFormat = 0;//DXVA2_SampleUnknown;
ext->VideoChromaSubsampling = 0;//DXVA2_VideoChromaSubsampling_Unknown;
ext->NominalRange = 0;//DXVA2_NominalRange_Unknown;
ext->VideoTransferMatrix = 0;//DXVA2_VideoTransferMatrix_Unknown;
ext->VideoLighting = 0;//DXVA2_VideoLighting_Unknown;
ext->VideoPrimaries = 0;//DXVA2_VideoPrimaries_Unknown;
ext->VideoTransferFunction = 0;//DXVA2_VideoTransFunc_Unknown;
/* List all configurations available for the decoder */
UINT cfg_count = 0;
DXVA2_ConfigPictureDecode *cfg_list = NULL;
if (FAILED(IDirectXVideoDecoderService_GetDecoderConfigurations(va->vs,
&va->input,
&dsc,
NULL,
&cfg_count,
&cfg_list))) {
msg_Err(va->log, "IDirectXVideoDecoderService_GetDecoderConfigurations failed");
return VLC_EGENERIC;
}
msg_Dbg(va->log, "we got %d decoder configurations", cfg_count);
/* Select the best decoder configuration */
int cfg_score = 0;
for (unsigned i = 0; i < cfg_count; i++) {
const DXVA2_ConfigPictureDecode *cfg = &cfg_list[i];
/* */
msg_Dbg(va->log, "configuration[%d] ConfigBitstreamRaw %d",
i, cfg->ConfigBitstreamRaw);
/* */
int score;
if (cfg->ConfigBitstreamRaw == 1)
score = 1;
else if (codec_id == AV_CODEC_ID_H264 && cfg->ConfigBitstreamRaw == 2)
score = 2;
else
continue;
if (IsEqualGUID(&cfg->guidConfigBitstreamEncryption, &DXVA_NoEncrypt))
score += 16;
if (cfg_score < score) {
va->cfg = *cfg;
cfg_score = score;
}
}
CoTaskMemFree(cfg_list);
if (cfg_score <= 0) {
msg_Err(va->log, "Failed to find a supported decoder configuration");
return VLC_EGENERIC;
}
/* Create the decoder */
IDirectXVideoDecoder *decoder;
if (FAILED(IDirectXVideoDecoderService_CreateVideoDecoder(va->vs,
&va->input,
&dsc,
&va->cfg,
va->hw_surface,
va->surface_count,
&decoder))) {
msg_Err(va->log, "IDirectXVideoDecoderService_CreateVideoDecoder failed");
return VLC_EGENERIC;
}
va->decoder = decoder;
msg_Dbg(va->log, "IDirectXVideoDecoderService_CreateVideoDecoder succeed");
return VLC_SUCCESS;
}
static int dxva2_create_decoder(AVCodecContext *s)
{
HwAccelContext *hac = s->opaque;
DXVA2Context *ctx = hac->hwaccel_ctx;
struct dxva_context *dxva_ctx = s->hwaccel_context;
GUID *guid_list = NULL;
unsigned guid_count = 0, i, j;
GUID device_guid = GUID_NULL;
D3DFORMAT target_format = 0;
DXVA2_VideoDesc desc = { 0 };
DXVA2_ConfigPictureDecode config;
HRESULT hr;
int surface_alignment;
int ret;
hr = IDirectXVideoDecoderService_GetDecoderDeviceGuids(ctx->decoder_service, &guid_count, &guid_list);
if (FAILED(hr)) {
av_log(NULL, loglevel, "Failed to retrieve decoder device GUIDs\n");
goto fail;
}
for (i = 0; dxva2_modes[i].guid; i++) {
D3DFORMAT *target_list = NULL;
unsigned target_count = 0;
const dxva2_mode *mode = &dxva2_modes[i];
if (mode->codec != s->codec_id)
continue;
for (j = 0; j < guid_count; j++) {
if (IsEqualGUID(mode->guid, &guid_list[j]))
break;
}
if (j == guid_count)
continue;
hr = IDirectXVideoDecoderService_GetDecoderRenderTargets(ctx->decoder_service, mode->guid, &target_count, &target_list);
if (FAILED(hr)) {
continue;
}
for (j = 0; j < target_count; j++) {
const D3DFORMAT format = target_list[j];
if (format == MKTAG('N','V','1','2')) {
target_format = format;
break;
}
}
CoTaskMemFree(target_list);
if (target_format) {
device_guid = *mode->guid;
break;
}
}
CoTaskMemFree(guid_list);
if (IsEqualGUID(&device_guid, &GUID_NULL)) {
av_log(NULL, loglevel, "No decoder device for codec found\n");
goto fail;
}
desc.SampleWidth = s->coded_width;
desc.SampleHeight = s->coded_height;
desc.Format = target_format;
ret = dxva2_get_decoder_configuration(s, &device_guid, &desc, &config);
if (ret < 0) {
goto fail;
}
/* decoding MPEG-2 requires additional alignment on some Intel GPUs,
but it causes issues for H.264 on certain AMD GPUs..... */
if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO)
surface_alignment = 32;
/* the HEVC DXVA2 spec asks for 128 pixel aligned surfaces to ensure
all coding features have enough room to work with */
else if (s->codec_id == AV_CODEC_ID_HEVC)
surface_alignment = 128;
else
surface_alignment = 16;
/* 4 base work surfaces */
ctx->num_surfaces = 4;
/* add surfaces based on number of possible refs */
if (s->codec_id == AV_CODEC_ID_H264 || s->codec_id == AV_CODEC_ID_HEVC)
ctx->num_surfaces += 16;
else
ctx->num_surfaces += 2;
/* add extra surfaces for frame threading */
if (s->active_thread_type & FF_THREAD_FRAME)
ctx->num_surfaces += s->thread_count;
ctx->surfaces = av_mallocz(ctx->num_surfaces * sizeof(*ctx->surfaces));
ctx->surface_infos = av_mallocz(ctx->num_surfaces * sizeof(*ctx->surface_infos));
if (!ctx->surfaces || !ctx->surface_infos) {
av_log(NULL, loglevel, "Unable to allocate surface arrays\n");
goto fail;
}
hr = IDirectXVideoDecoderService_CreateSurface(ctx->decoder_service,
FFALIGN(s->coded_width, surface_alignment),
FFALIGN(s->coded_height, surface_alignment),
ctx->num_surfaces - 1,
target_format, D3DPOOL_DEFAULT, 0,
DXVA2_VideoDecoderRenderTarget,
ctx->surfaces, NULL);
if (FAILED(hr)) {
av_log(NULL, loglevel, "Failed to create %d video surfaces\n", ctx->num_surfaces);
goto fail;
}
hr = IDirectXVideoDecoderService_CreateVideoDecoder(ctx->decoder_service, &device_guid,
&desc, &config, ctx->surfaces,
ctx->num_surfaces, &ctx->decoder);
if (FAILED(hr)) {
av_log(NULL, loglevel, "Failed to create DXVA2 video decoder\n");
goto fail;
}
ctx->decoder_guid = device_guid;
ctx->decoder_config = config;
dxva_ctx->cfg = &ctx->decoder_config;
dxva_ctx->decoder = ctx->decoder;
dxva_ctx->surface = ctx->surfaces;
dxva_ctx->surface_count = ctx->num_surfaces;
if (IsEqualGUID(&ctx->decoder_guid, &DXVADDI_Intel_ModeH264_E))
dxva_ctx->workaround |= FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO;
return 0;
fail:
dxva2_destroy_decoder(s);
return AVERROR(EINVAL);
}
static int dxva2_create_decoder(struct lavc_ctx *s, int w, int h,
enum AVCodecID codec_id, int profile)
{
DXVA2Context *ctx = s->hwdec_priv;
struct dxva_context *dxva_ctx = s->avctx->hwaccel_context;
GUID *guid_list = NULL;
unsigned guid_count = 0, i, j;
GUID device_guid = GUID_NULL;
D3DFORMAT target_format = 0;
DXVA2_VideoDesc desc = { 0 };
DXVA2_ConfigPictureDecode config;
HRESULT hr;
int surface_alignment;
int ret;
hr = IDirectXVideoDecoderService_GetDecoderDeviceGuids(ctx->decoder_service, &guid_count, &guid_list);
if (FAILED(hr)) {
MP_ERR(ctx, "Failed to retrieve decoder device GUIDs\n");
goto fail;
}
// dump all decoder info
MP_VERBOSE(ctx, "%d decoder devices:\n", (int)guid_count);
for (j = 0; j < guid_count; j++) {
GUID *guid = &guid_list[j];
const char *name = "<unknown>";
for (i = 0; dxva2_modes[i].guid; i++) {
if (IsEqualGUID(dxva2_modes[i].guid, guid))
name = dxva2_modes[i].name;
}
D3DFORMAT *target_list = NULL;
unsigned target_count = 0;
hr = IDirectXVideoDecoderService_GetDecoderRenderTargets(ctx->decoder_service, guid, &target_count, &target_list);
if (FAILED(hr))
continue;
char fmts[256] = {0};
for (i = 0; i < target_count; i++)
mp_snprintf_cat(fmts, sizeof(fmts), " %s", mp_tag_str(target_list[i]));
CoTaskMemFree(target_list);
MP_VERBOSE(ctx, "%s %s %s\n", mp_GUID_to_str(guid), name, fmts);
}
// find a suitable decoder
for (i = 0; dxva2_modes[i].guid; i++) {
D3DFORMAT *target_list = NULL;
unsigned target_count = 0;
const dxva2_mode *mode = &dxva2_modes[i];
if (mode->codec != codec_id)
continue;
for (j = 0; j < guid_count; j++) {
if (IsEqualGUID(mode->guid, &guid_list[j]))
break;
}
if (j == guid_count)
continue;
hr = IDirectXVideoDecoderService_GetDecoderRenderTargets(ctx->decoder_service, mode->guid, &target_count, &target_list);
if (FAILED(hr)) {
continue;
}
for (j = 0; j < target_count; j++) {
const D3DFORMAT format = target_list[j];
if (format == MKTAG('N','V','1','2')) {
target_format = format;
break;
}
}
CoTaskMemFree(target_list);
if (target_format) {
device_guid = *mode->guid;
break;
}
}
CoTaskMemFree(guid_list);
if (IsEqualGUID(&device_guid, &GUID_NULL)) {
MP_ERR(ctx, "No decoder device for codec found\n");
goto fail;
}
desc.SampleWidth = w;
desc.SampleHeight = h;
desc.Format = target_format;
ret = dxva2_get_decoder_configuration(s, codec_id, &device_guid, &desc, &config);
if (ret < 0) {
goto fail;
}
/* decoding MPEG-2 requires additional alignment on some Intel GPUs,
but it causes issues for H.264 on certain AMD GPUs..... */
if (codec_id == AV_CODEC_ID_MPEG2VIDEO)
surface_alignment = 32;
/* the HEVC DXVA2 spec asks for 128 pixel aligned surfaces to ensure
all coding features have enough room to work with */
else if (codec_id == AV_CODEC_ID_HEVC)
surface_alignment = 128;
else
surface_alignment = 16;
/* 4 base work surfaces */
ctx->num_surfaces = 4 + ADDTIONAL_SURFACES;
/* add surfaces based on number of possible refs */
if (codec_id == AV_CODEC_ID_H264 || codec_id == AV_CODEC_ID_HEVC)
ctx->num_surfaces += 16;
else
ctx->num_surfaces += 2;
ctx->surfaces = av_mallocz(ctx->num_surfaces * sizeof(*ctx->surfaces));
ctx->surface_infos = av_mallocz(ctx->num_surfaces * sizeof(*ctx->surface_infos));
if (!ctx->surfaces || !ctx->surface_infos) {
MP_ERR(ctx, "Unable to allocate surface arrays\n");
goto fail;
}
hr = IDirectXVideoDecoderService_CreateSurface(ctx->decoder_service,
FFALIGN(w, surface_alignment),
FFALIGN(h, surface_alignment),
ctx->num_surfaces - 1,
target_format, D3DPOOL_DEFAULT, 0,
DXVA2_VideoDecoderRenderTarget,
ctx->surfaces, NULL);
if (FAILED(hr)) {
MP_ERR(ctx, "Failed to create %d video surfaces\n", ctx->num_surfaces);
goto fail;
}
hr = IDirectXVideoDecoderService_CreateVideoDecoder(ctx->decoder_service, &device_guid,
&desc, &config, ctx->surfaces,
ctx->num_surfaces, &ctx->decoder);
if (FAILED(hr)) {
MP_ERR(ctx, "Failed to create DXVA2 video decoder\n");
goto fail;
}
ctx->decoder_guid = device_guid;
ctx->decoder_config = config;
dxva_ctx->cfg = &ctx->decoder_config;
dxva_ctx->decoder = ctx->decoder;
dxva_ctx->surface = ctx->surfaces;
dxva_ctx->surface_count = ctx->num_surfaces;
if (IsEqualGUID(&ctx->decoder_guid, &DXVADDI_Intel_ModeH264_E))
dxva_ctx->workaround |= FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO;
return 0;
fail:
dxva2_destroy_decoder(s);
return -1;
}
static void test_decoder_resolution(IDirectXVideoDecoderService *service, REFGUID const guid, D3DFORMAT format,
UINT width, UINT height, UINT surface_count)
{
HRESULT hr;
DXVA2_VideoDesc desc;
UINT count;
DXVA2_ConfigPictureDecode *configs;
DXVA2_ConfigPictureDecode config;
IDirect3DSurface9 **surfaces;
IDirectXVideoDecoder *decoder;
UINT i;
trace("Analysing buffer sizes for: %s (%u x %u)\n", convert_decoderguid_to_str(guid), width, height);
memset(&desc, 0, sizeof(desc));
desc.SampleWidth = width;
desc.SampleHeight = height;
desc.Format = format;
desc.InputSampleFreq.Numerator = 25;
desc.InputSampleFreq.Denominator = 1;
desc.OutputFrameFreq.Numerator = 25;
desc.OutputFrameFreq.Denominator = 1;
if (0) /* crashes on windows */
{
hr = IDirectXVideoDecoderService_GetDecoderConfigurations(service, guid, &desc, NULL, NULL, NULL);
ok(hr == E_INVALIDARG, "Expected E_INVALIDARG, got %x\n", hr);
hr = IDirectXVideoDecoderService_GetDecoderConfigurations(service, guid, &desc, NULL, &count, NULL);
ok(hr == E_INVALIDARG, "Expected E_INVALIDARG, got %x\n", hr);
hr = IDirectXVideoDecoderService_GetDecoderConfigurations(service, guid, &desc, NULL, NULL, &configs);
ok(hr == E_INVALIDARG, "Expected E_INVALIDARG, got %x\n", hr);
}
hr = IDirectXVideoDecoderService_GetDecoderConfigurations(service, &GUID_NULL, &desc, NULL, &count, &configs);
ok(hr == D3DERR_INVALIDCALL, "Expected D3DERR_INVALIDCALL, got %x\n", hr);
hr = IDirectXVideoDecoderService_GetDecoderConfigurations(service, guid, &desc, NULL, &count, &configs);
ok(!hr, "Failed to get decoder congiruation for: %s\n", convert_decoderguid_to_str(guid));
if (hr) return;
if (!count)
{
skip("Decoder for %s does not support a single decoder configuration? Skipping test\n",
convert_decoderguid_to_str(guid));
return;
}
memcpy(&config, &configs[0], sizeof(config));
CoTaskMemFree(configs);
surfaces = (void*)HeapAlloc(GetProcessHeap(), 0, surface_count * sizeof(IDirect3DSurface9));
if (!surfaces)
{
skip("Failed to allocate memory for surfaces\n");
return;
}
hr = IDirectXVideoDecoderService_CreateSurface(service, width, height, surface_count-1, format, D3DPOOL_DEFAULT, 0,
DXVA2_VideoDecoderRenderTarget, surfaces, NULL);
ok(!hr, "Failed to create surfaces: %x\n", hr);
if (hr)
{
HeapFree(GetProcessHeap(), 0, surfaces);
return;
}
hr = IDirectXVideoDecoderService_CreateVideoDecoder(service, guid, &desc, &configs[0], surfaces, surface_count, &decoder);
ok(!hr, "Failed to create decoder for %s: %x\n", convert_decoderguid_to_str(guid), hr);
if (!hr)
{
test_decoding(decoder, guid, surfaces);
IDirectXVideoDecoder_Release(decoder);
}
for (i = 0; i < surface_count; i++)
IDirect3DSurface9_Release(surfaces[i]);
HeapFree(GetProcessHeap(), 0, surfaces);
}
static int dxva2_init_decoder(struct lavc_ctx *s, int w, int h)
{
HRESULT hr;
int ret = -1;
struct priv *p = s->hwdec_priv;
TA_FREEP(&p->decoder_pool);
int n_surfaces = hwdec_get_max_refs(s) + ADDITIONAL_SURFACES;
IDirect3DSurface9 **surfaces = NULL;
IDirectXVideoDecoder *decoder = NULL;
void *tmp = talloc_new(NULL);
UINT n_guids;
GUID *device_guids;
hr = IDirectXVideoDecoderService_GetDecoderDeviceGuids(
p->decoder_service, &n_guids, &device_guids);
if (FAILED(hr)) {
MP_ERR(p, "Failed to retrieve decoder device GUIDs: %s\n",
mp_HRESULT_to_str(hr));
goto done;
}
dump_decoder_info(s, device_guids, n_guids);
struct d3d_decoder_fmt fmt =
d3d_select_decoder_mode(s, device_guids, n_guids,
d3d9_formats, MP_ARRAY_SIZE(d3d9_formats),
dxva2_format_supported);
CoTaskMemFree(device_guids);
if (!fmt.format) {
MP_ERR(p, "Failed to find a suitable decoder\n");
goto done;
}
p->mpfmt_decoded = fmt.format->mpfmt;
struct mp_image_pool *decoder_pool =
talloc_steal(tmp, mp_image_pool_new(n_surfaces));
DXVA2_ConfigPictureDecode *decoder_config =
talloc_zero(decoder_pool, DXVA2_ConfigPictureDecode);
int w_align = w, h_align = h;
d3d_surface_align(s, &w_align, &h_align);
DXVA2_VideoDesc video_desc ={
.SampleWidth = w,
.SampleHeight = h,
.Format = fmt.format->dxfmt,
};
UINT n_configs = 0;
DXVA2_ConfigPictureDecode *configs = NULL;
hr = IDirectXVideoDecoderService_GetDecoderConfigurations(
p->decoder_service, fmt.guid, &video_desc, NULL,
&n_configs, &configs);
if (FAILED(hr)) {
MP_ERR(p, "Unable to retrieve decoder configurations: %s\n",
mp_HRESULT_to_str(hr));
goto done;
}
unsigned max_score = 0;
for (UINT i = 0; i < n_configs; i++) {
unsigned score = d3d_decoder_config_score(
s, &configs[i].guidConfigBitstreamEncryption,
configs[i].ConfigBitstreamRaw);
if (score > max_score) {
max_score = score;
*decoder_config = configs[i];
}
}
CoTaskMemFree(configs);
if (!max_score) {
MP_ERR(p, "Failed to find a suitable decoder configuration\n");
goto done;
}
surfaces = talloc_zero_array(decoder_pool, IDirect3DSurface9*, n_surfaces);
hr = IDirectXVideoDecoderService_CreateSurface(
p->decoder_service,
w_align, h_align,
n_surfaces - 1, fmt.format->dxfmt, D3DPOOL_DEFAULT, 0,
DXVA2_VideoDecoderRenderTarget, surfaces, NULL);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create %d video surfaces: %s\n",
n_surfaces, mp_HRESULT_to_str(hr));
goto done;
}
hr = IDirectXVideoDecoderService_CreateVideoDecoder(
p->decoder_service, fmt.guid, &video_desc, decoder_config,
surfaces, n_surfaces, &decoder);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create DXVA2 video decoder: %s\n",
mp_HRESULT_to_str(hr));
goto done;
}
for (int i = 0; i < n_surfaces; i++) {
struct mp_image *img = dxva2_new_ref(decoder, surfaces[i], w, h);
if (!img) {
MP_ERR(p, "Failed to create DXVA2 image\n");
goto done;
}
mp_image_pool_add(decoder_pool, img); // transferred to pool
}
// Pass required information on to ffmpeg.
struct dxva_context *dxva_ctx = s->avctx->hwaccel_context;
dxva_ctx->cfg = decoder_config;
dxva_ctx->decoder = decoder;
dxva_ctx->surface_count = n_surfaces;
dxva_ctx->surface = surfaces;
dxva_ctx->workaround = is_clearvideo(fmt.guid) ?
FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO : 0;
p->decoder_pool = talloc_steal(NULL, decoder_pool);
ret = 0;
done:
// On success, `p->decoder_pool` mp_images still hold refs to `surfaces` and
// `decoder`, so the pointers in the ffmpeg `dxva_context` strcture remain
// valid for the lifetime of the pool.
if (surfaces) {
for (int i = 0; i < n_surfaces; i++)
IDirect3DSurface9_Release(surfaces[i]);
}
if (decoder)
IDirectXVideoDecoder_Release(decoder);
talloc_free(tmp);
return ret;
}
static void destroy_device(struct lavc_ctx *s)
{
struct priv *p = s->hwdec_priv;
if (p->device)
IDirect3DDevice9_Release(p->device);
if (p->d3d9)
IDirect3D9_Release(p->d3d9);
}
static bool create_device(struct lavc_ctx *s)
{
struct priv *p = s->hwdec_priv;
d3d_load_dlls();
if (!d3d9_dll) {
MP_ERR(p, "Failed to load D3D9 library\n");
return false;
}
HRESULT (WINAPI *Direct3DCreate9Ex)(UINT, IDirect3D9Ex **) =
(void *)GetProcAddress(d3d9_dll, "Direct3DCreate9Ex");
if (!Direct3DCreate9Ex) {
MP_ERR(p, "Failed to locate Direct3DCreate9Ex\n");
return false;
}
IDirect3D9Ex *d3d9ex = NULL;
HRESULT hr = Direct3DCreate9Ex(D3D_SDK_VERSION, &d3d9ex);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create IDirect3D9Ex object\n");
return false;
}
UINT adapter = D3DADAPTER_DEFAULT;
D3DDISPLAYMODEEX modeex = {0};
IDirect3D9Ex_GetAdapterDisplayModeEx(d3d9ex, adapter, &modeex, NULL);
D3DPRESENT_PARAMETERS present_params = {
.Windowed = TRUE,
.BackBufferWidth = 640,
.BackBufferHeight = 480,
.BackBufferCount = 0,
.BackBufferFormat = modeex.Format,
.SwapEffect = D3DSWAPEFFECT_DISCARD,
.Flags = D3DPRESENTFLAG_VIDEO,
};
IDirect3DDevice9Ex *exdev = NULL;
hr = IDirect3D9Ex_CreateDeviceEx(d3d9ex, adapter,
D3DDEVTYPE_HAL,
GetShellWindow(),
D3DCREATE_SOFTWARE_VERTEXPROCESSING |
D3DCREATE_MULTITHREADED |
D3DCREATE_FPU_PRESERVE,
&present_params,
NULL,
&exdev);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create Direct3D device: %s\n",
mp_HRESULT_to_str(hr));
IDirect3D9_Release(d3d9ex);
return false;
}
p->d3d9 = (IDirect3D9 *)d3d9ex;
p->device = (IDirect3DDevice9 *)exdev;
return true;
}
static void dxva2_uninit(struct lavc_ctx *s)
{
struct priv *p = s->hwdec_priv;
if (!p)
return;
av_freep(&s->avctx->hwaccel_context);
talloc_free(p->decoder_pool);
if (p->decoder_service)
IDirectXVideoDecoderService_Release(p->decoder_service);
if (p->device_manager && p->device_handle != INVALID_HANDLE_VALUE)
IDirect3DDeviceManager9_CloseDeviceHandle(p->device_manager, p->device_handle);
if (p->device_manager)
IDirect3DDeviceManager9_Release(p->device_manager);
destroy_device(s);
TA_FREEP(&s->hwdec_priv);
}
static int dxva2_init(struct lavc_ctx *s)
{
HRESULT hr;
struct priv *p = talloc_zero(NULL, struct priv);
if (!p)
return -1;
s->hwdec_priv = p;
p->device_handle = INVALID_HANDLE_VALUE;
p->log = mp_log_new(s, s->log, "dxva2");
if (s->hwdec->type == HWDEC_DXVA2_COPY) {
mp_check_gpu_memcpy(p->log, NULL);
p->sw_pool = talloc_steal(p, mp_image_pool_new(17));
}
p->device = hwdec_devices_load(s->hwdec_devs, s->hwdec->type);
if (p->device) {
IDirect3D9_AddRef(p->device);
MP_VERBOSE(p, "Using VO-supplied device %p.\n", p->device);
} else if (s->hwdec->type == HWDEC_DXVA2) {
MP_ERR(p, "No Direct3D device provided for native dxva2 decoding\n");
goto fail;
} else {
if (!create_device(s))
goto fail;
}
d3d_load_dlls();
if (!dxva2_dll) {
MP_ERR(p, "Failed to load DXVA2 library\n");
goto fail;
}
HRESULT (WINAPI *CreateDeviceManager9)(UINT *, IDirect3DDeviceManager9 **) =
(void *)GetProcAddress(dxva2_dll, "DXVA2CreateDirect3DDeviceManager9");
if (!CreateDeviceManager9) {
MP_ERR(p, "Failed to locate DXVA2CreateDirect3DDeviceManager9\n");
goto fail;
}
unsigned reset_token = 0;
hr = CreateDeviceManager9(&reset_token, &p->device_manager);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create Direct3D device manager: %s\n",
mp_HRESULT_to_str(hr));
goto fail;
}
hr = IDirect3DDeviceManager9_ResetDevice(p->device_manager,
p->device, reset_token);
if (FAILED(hr)) {
MP_ERR(p, "Failed to bind Direct3D device to device manager: %s\n",
mp_HRESULT_to_str(hr));
goto fail;
}
hr = IDirect3DDeviceManager9_OpenDeviceHandle(p->device_manager,
&p->device_handle);
if (FAILED(hr)) {
MP_ERR(p, "Failed to open device handle: %s\n",
mp_HRESULT_to_str(hr));
goto fail;
}
hr = IDirect3DDeviceManager9_GetVideoService(
p->device_manager, p->device_handle, &IID_IDirectXVideoDecoderService,
(void **)&p->decoder_service);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create IDirectXVideoDecoderService: %s\n",
mp_HRESULT_to_str(hr));
goto fail;
}
s->avctx->hwaccel_context = av_mallocz(sizeof(struct dxva_context));
if (!s->avctx->hwaccel_context)
goto fail;
return 0;
fail:
dxva2_uninit(s);
return -1;
}
static int dxva2_probe(struct lavc_ctx *ctx, struct vd_lavc_hwdec *hwdec,
const char *codec)
{
// dxva2-copy can do without external context; dxva2 requires it.
if (hwdec->type == HWDEC_DXVA2) {
if (!hwdec_devices_load(ctx->hwdec_devs, HWDEC_DXVA2))
return HWDEC_ERR_NO_CTX;
} else {
hwdec_devices_load(ctx->hwdec_devs, HWDEC_DXVA2_COPY);
}
return d3d_probe_codec(codec);
}
const struct vd_lavc_hwdec mp_vd_lavc_dxva2 = {
.type = HWDEC_DXVA2,
.image_format = IMGFMT_DXVA2,
.probe = dxva2_probe,
.init = dxva2_init,
.uninit = dxva2_uninit,
.init_decoder = dxva2_init_decoder,
.allocate_image = dxva2_allocate_image,
};
const struct vd_lavc_hwdec mp_vd_lavc_dxva2_copy = {
.type = HWDEC_DXVA2_COPY,
.copying = true,
.image_format = IMGFMT_DXVA2,
.probe = dxva2_probe,
.init = dxva2_init,
.uninit = dxva2_uninit,
.init_decoder = dxva2_init_decoder,
.allocate_image = dxva2_allocate_image,
.process_image = dxva2_retrieve_image,
.delay_queue = HWDEC_DELAY_QUEUE_COUNT,
};
/**
* It creates a DXVA2 decoder using the given video format
*/
static int hb_dx_create_video_decoder( hb_va_dxva2_t *dxva2, int codec_id, const hb_title_t* fmt )
{
dxva2->width = fmt->width;
dxva2->height = fmt->height;
dxva2->surface_width = (fmt->width + 15) & ~15;
dxva2->surface_height = (fmt->height + 15) & ~15;
switch( codec_id )
{
case AV_CODEC_ID_H264:
dxva2->surface_count = 16 + 1;
break;
default:
dxva2->surface_count = 2 + 1;
break;
}
LPDIRECT3DSURFACE9 surface_list[VA_DXVA2_MAX_SURFACE_COUNT];
if( FAILED( IDirectXVideoDecoderService_CreateSurface( dxva2->vs,
dxva2->surface_width,
dxva2->surface_height,
dxva2->surface_count - 1,
dxva2->render,
D3DPOOL_DEFAULT,
0,
DXVA2_VideoDecoderRenderTarget,
surface_list, NULL )))
{
hb_log( "dxva2:IDirectXVideoAccelerationService_CreateSurface failed" );
dxva2->surface_count = 0;
return HB_WORK_ERROR;
}
unsigned i;
for( i = 0; i<dxva2->surface_count; i++ )
{
hb_va_surface_t *surface = &dxva2->surface[i];
surface->d3d = surface_list[i];
surface->refcount = 0;
surface->order = 0;
}
hb_log( "dxva2:CreateSurface succeed with %d, fmt (%dx%d) surfaces (%dx%d)", dxva2->surface_count,
fmt->width, fmt->height, dxva2->surface_width, dxva2->surface_height );
DXVA2_VideoDesc dsc;
memset( &dsc, 0, sizeof(dsc));
dsc.SampleWidth = fmt->width;
dsc.SampleHeight = fmt->height;
dsc.Format = dxva2->render;
if( fmt->rate> 0 && fmt->rate_base> 0 )
{
dsc.InputSampleFreq.Numerator = fmt->rate;
dsc.InputSampleFreq.Denominator = fmt->rate_base;
}
else
{
dsc.InputSampleFreq.Numerator = 0;
dsc.InputSampleFreq.Denominator = 0;
}
dsc.OutputFrameFreq = dsc.InputSampleFreq;
dsc.UABProtectionLevel = FALSE;
dsc.Reserved = 0;
/* FIXME I am unsure we can let unknown everywhere */
DXVA2_ExtendedFormat *ext = &dsc.SampleFormat;
ext->SampleFormat = 0; //DXVA2_SampleUnknown;
ext->VideoChromaSubsampling = 0; //DXVA2_VideoChromaSubsampling_Unknown;
ext->NominalRange = 0; //DXVA2_NominalRange_Unknown;
ext->VideoTransferMatrix = 0; //DXVA2_VideoTransferMatrix_Unknown;
ext->VideoLighting = 0; //DXVA2_VideoLighting_Unknown;
ext->VideoPrimaries = 0; //DXVA2_VideoPrimaries_Unknown;
ext->VideoTransferFunction = 0; //DXVA2_VideoTransFunc_Unknown;
/* List all configurations available for the decoder */
UINT cfg_count = 0;
DXVA2_ConfigPictureDecode *cfg_list = NULL;
if( FAILED( IDirectXVideoDecoderService_GetDecoderConfigurations( dxva2->vs, &dxva2->input, &dsc, NULL, &cfg_count, &cfg_list )))
{
hb_log( "dxva2:IDirectXVideoDecoderService_GetDecoderConfigurations failed" );
return HB_WORK_ERROR;
}
hb_log( "dxva2:we got %d decoder configurations", cfg_count );
/* Select the best decoder configuration */
int cfg_score = 0;
for( i = 0; i < cfg_count; i++ )
{
const DXVA2_ConfigPictureDecode *cfg = &cfg_list[i];
hb_log( "dxva2:configuration[%d] ConfigBitstreamRaw %d", i, cfg->ConfigBitstreamRaw );
int score;
if( cfg->ConfigBitstreamRaw == 1 )
score = 1;
else if( codec_id == AV_CODEC_ID_H264 && cfg->ConfigBitstreamRaw == 2 )
score = 2;
else
continue;
if( IsEqualGUID( &cfg->guidConfigBitstreamEncryption, &DXVA_NoEncrypt ))
score += 16;
if( cfg_score < score )
{
dxva2->cfg = *cfg;
cfg_score = score;
}
}
//my_release(cfg_list);
if( cfg_score <= 0 )
{
hb_log( "dxva2:Failed to find a supported decoder configuration" );
return HB_WORK_ERROR;
}
/* Create the decoder */
IDirectXVideoDecoder *decoder;
if( FAILED( IDirectXVideoDecoderService_CreateVideoDecoder( dxva2->vs, &dxva2->input, &dsc, &dxva2->cfg, surface_list, dxva2->surface_count, &decoder )))
{
hb_log( "dxva2:IDirectXVideoDecoderService_CreateVideoDecoder failed" );
return HB_WORK_ERROR;
}
dxva2->decoder = decoder;
hb_log( "dxva2:IDirectXVideoDecoderService_CreateVideoDecoder succeed" );
return HB_WORK_OK;
}
static int dxva2_create_decoder(AVCodecContext *s)
{
InputStream *ist = s->opaque;
int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
DXVA2Context *ctx = ist->hwaccel_ctx;
struct dxva_context *dxva_ctx = s->hwaccel_context;
GUID *guid_list = NULL;
unsigned guid_count = 0, i, j;
GUID device_guid = GUID_NULL;
const D3DFORMAT surface_format = s->sw_pix_fmt == AV_PIX_FMT_YUV420P10 ?
MKTAG('P', '0', '1', '0') : MKTAG('N', 'V', '1', '2');
D3DFORMAT target_format = 0;
DXVA2_VideoDesc desc = { 0 };
DXVA2_ConfigPictureDecode config;
HRESULT hr;
int surface_alignment, num_surfaces;
int ret;
AVDXVA2FramesContext *frames_hwctx;
AVHWFramesContext *frames_ctx;
hr = IDirectXVideoDecoderService_GetDecoderDeviceGuids(ctx->decoder_service, &guid_count, &guid_list);
if (FAILED(hr)) {
av_log(NULL, loglevel, "Failed to retrieve decoder device GUIDs\n");
goto fail;
}
for (i = 0; dxva2_modes[i].guid; i++) {
D3DFORMAT *target_list = NULL;
unsigned target_count = 0;
const dxva2_mode *mode = &dxva2_modes[i];
if (mode->codec != s->codec_id)
continue;
for (j = 0; j < guid_count; j++) {
if (IsEqualGUID(mode->guid, &guid_list[j]))
break;
}
if (j == guid_count)
continue;
hr = IDirectXVideoDecoderService_GetDecoderRenderTargets(ctx->decoder_service, mode->guid, &target_count, &target_list);
if (FAILED(hr)) {
continue;
}
for (j = 0; j < target_count; j++) {
const D3DFORMAT format = target_list[j];
if (format == surface_format) {
target_format = format;
break;
}
}
CoTaskMemFree(target_list);
if (target_format) {
device_guid = *mode->guid;
break;
}
}
CoTaskMemFree(guid_list);
if (IsEqualGUID(&device_guid, &GUID_NULL)) {
av_log(NULL, loglevel, "No decoder device for codec found\n");
goto fail;
}
desc.SampleWidth = s->coded_width;
desc.SampleHeight = s->coded_height;
desc.Format = target_format;
ret = dxva2_get_decoder_configuration(s, &device_guid, &desc, &config);
if (ret < 0) {
goto fail;
}
/* decoding MPEG-2 requires additional alignment on some Intel GPUs,
but it causes issues for H.264 on certain AMD GPUs..... */
if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO)
surface_alignment = 32;
/* the HEVC DXVA2 spec asks for 128 pixel aligned surfaces to ensure
all coding features have enough room to work with */
else if (s->codec_id == AV_CODEC_ID_HEVC)
surface_alignment = 128;
else
surface_alignment = 16;
/* 4 base work surfaces */
num_surfaces = 4;
/* add surfaces based on number of possible refs */
if (s->codec_id == AV_CODEC_ID_H264 || s->codec_id == AV_CODEC_ID_HEVC)
num_surfaces += 16;
else
num_surfaces += 2;
/* add extra surfaces for frame threading */
if (s->active_thread_type & FF_THREAD_FRAME)
num_surfaces += s->thread_count;
ctx->hw_frames_ctx = av_hwframe_ctx_alloc(ctx->hw_device_ctx);
if (!ctx->hw_frames_ctx)
goto fail;
frames_ctx = (AVHWFramesContext*)ctx->hw_frames_ctx->data;
frames_hwctx = frames_ctx->hwctx;
frames_ctx->format = AV_PIX_FMT_DXVA2_VLD;
frames_ctx->sw_format = s->sw_pix_fmt == AV_PIX_FMT_YUV420P10 ?
AV_PIX_FMT_P010 : AV_PIX_FMT_NV12;
frames_ctx->width = FFALIGN(s->coded_width, surface_alignment);
frames_ctx->height = FFALIGN(s->coded_height, surface_alignment);
frames_ctx->initial_pool_size = num_surfaces;
frames_hwctx->surface_type = DXVA2_VideoDecoderRenderTarget;
ret = av_hwframe_ctx_init(ctx->hw_frames_ctx);
if (ret < 0) {
av_log(NULL, loglevel, "Failed to initialize the HW frames context\n");
goto fail;
}
hr = IDirectXVideoDecoderService_CreateVideoDecoder(ctx->decoder_service, &device_guid,
&desc, &config, frames_hwctx->surfaces,
frames_hwctx->nb_surfaces, &frames_hwctx->decoder_to_release);
if (FAILED(hr)) {
av_log(NULL, loglevel, "Failed to create DXVA2 video decoder\n");
goto fail;
}
ctx->decoder_guid = device_guid;
ctx->decoder_config = config;
dxva_ctx->cfg = &ctx->decoder_config;
dxva_ctx->decoder = frames_hwctx->decoder_to_release;
dxva_ctx->surface = frames_hwctx->surfaces;
dxva_ctx->surface_count = frames_hwctx->nb_surfaces;
if (IsEqualGUID(&ctx->decoder_guid, &DXVADDI_Intel_ModeH264_E))
dxva_ctx->workaround |= FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO;
return 0;
fail:
av_buffer_unref(&ctx->hw_frames_ctx);
return AVERROR(EINVAL);
}