void mp_set_pcm_codec(struct sh_stream *sh, bool sign, bool is_float, int bits,
bool is_be)
{
// This uses libavcodec pcm codec names, e.g. "pcm_u16le".
char codec[64] = "pcm_";
if (is_float) {
mp_snprintf_cat(codec, sizeof(codec), "f");
} else {
mp_snprintf_cat(codec, sizeof(codec), sign ? "s" : "u");
}
mp_snprintf_cat(codec, sizeof(codec), "%d", bits);
if (bits != 8)
mp_snprintf_cat(codec, sizeof(codec), is_be ? "be" : "le");
sh->codec = talloc_strdup(sh->audio, codec);
}
static void dump_decoder_info(struct lavc_ctx *s,
GUID *device_guids, UINT n_guids)
{
struct priv *p = s->hwdec_priv;
MP_VERBOSE(p, "%u decoder devices:\n", (unsigned)n_guids);
for (UINT i = 0; i < n_guids; i++) {
GUID *guid = &device_guids[i];
char *description = d3d_decoder_guid_to_desc(guid);
D3DFORMAT *formats = NULL;
UINT n_formats = 0;
HRESULT hr = IDirectXVideoDecoderService_GetDecoderRenderTargets(
p->decoder_service, guid, &n_formats, &formats);
if (FAILED(hr)) {
MP_ERR(p, "Failed to get render targets for decoder %s:%s\n",
description, mp_HRESULT_to_str(hr));
}
char fmts[256] = {0};
for (UINT j = 0; j < n_formats; j++) {
mp_snprintf_cat(fmts, sizeof(fmts),
" %s", mp_tag_str(formats[j]));
}
CoTaskMemFree(formats);
MP_VERBOSE(p, "%s %s\n", description, fmts);
}
}
static void list_features(int set, struct mp_log *log, int msgl, bool invert)
{
char b[1024] = {0};
for (const struct feature *f = &features[0]; f->id; f++) {
if (invert == !(f->id & set))
mp_snprintf_cat(b, sizeof(b), " [%s]", f->name);
}
mp_msg(log, msgl, "%s\n", b);
}
static void dump_decoder_info(struct lavc_ctx *s, const GUID *guid)
{
struct priv *p = s->hwdec_priv;
char fmts[256] = {0};
for (int i = 0; i < MP_ARRAY_SIZE(d3d11_formats); i++) {
const struct d3d_decoded_format *format = &d3d11_formats[i];
if (d3d11_format_supported(s, guid, format))
mp_snprintf_cat(fmts, sizeof(fmts), " %s", format->name);
}
MP_VERBOSE(p, "%s %s\n", d3d_decoder_guid_to_desc(guid), fmts);
}
static char *audio_config_to_str_buf(char *buf, size_t buf_sz, int rate,
int format, struct mp_chmap channels)
{
char ch[128];
mp_chmap_to_str_buf(ch, sizeof(ch), &channels);
char *hr_ch = mp_chmap_to_str_hr(&channels);
if (strcmp(hr_ch, ch) != 0)
mp_snprintf_cat(ch, sizeof(ch), " (%s)", hr_ch);
snprintf(buf, buf_sz, "%dHz %s %dch %s", rate,
ch, channels.num, af_fmt_to_str(format));
return buf;
}
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;
}
struct ra_layout ra_renderpass_input_layout(struct ra_renderpass_input *input)
{
size_t el_size = ra_vartype_size(input->type);
if (!el_size)
return (struct ra_layout){0};
// host data is always tightly packed
return (struct ra_layout) {
.align = 1,
.stride = el_size * input->dim_v,
.size = el_size * input->dim_v * input->dim_m,
};
}
static struct ra_renderpass_input *dup_inputs(void *ta_parent,
const struct ra_renderpass_input *inputs, int num_inputs)
{
struct ra_renderpass_input *res =
talloc_memdup(ta_parent, (void *)inputs, num_inputs * sizeof(inputs[0]));
for (int n = 0; n < num_inputs; n++)
res[n].name = talloc_strdup(res, res[n].name);
return res;
}
// Return a newly allocated deep-copy of params.
struct ra_renderpass_params *ra_renderpass_params_copy(void *ta_parent,
const struct ra_renderpass_params *params)
{
struct ra_renderpass_params *res = talloc_ptrtype(ta_parent, res);
*res = *params;
res->inputs = dup_inputs(res, res->inputs, res->num_inputs);
res->vertex_attribs =
dup_inputs(res, res->vertex_attribs, res->num_vertex_attribs);
res->cached_program = bstrdup(res, res->cached_program);
res->vertex_shader = talloc_strdup(res, res->vertex_shader);
res->frag_shader = talloc_strdup(res, res->frag_shader);
res->compute_shader = talloc_strdup(res, res->compute_shader);
return res;
};
struct glsl_fmt {
enum ra_ctype ctype;
int num_components;
int component_depth[4];
const char *glsl_format;
};
// List taken from the GLSL specification, sans snorm and sint formats
static const struct glsl_fmt ra_glsl_fmts[] = {
{RA_CTYPE_FLOAT, 1, {16}, "r16f"},
{RA_CTYPE_FLOAT, 1, {32}, "r32f"},
{RA_CTYPE_FLOAT, 2, {16, 16}, "rg16f"},
{RA_CTYPE_FLOAT, 2, {32, 32}, "rg32f"},
{RA_CTYPE_FLOAT, 4, {16, 16, 16, 16}, "rgba16f"},
{RA_CTYPE_FLOAT, 4, {32, 32, 32, 32}, "rgba32f"},
{RA_CTYPE_FLOAT, 3, {11, 11, 10}, "r11f_g11f_b10f"},
{RA_CTYPE_UNORM, 1, {8}, "r8"},
{RA_CTYPE_UNORM, 1, {16}, "r16"},
{RA_CTYPE_UNORM, 2, {8, 8}, "rg8"},
{RA_CTYPE_UNORM, 2, {16, 16}, "rg16"},
{RA_CTYPE_UNORM, 4, {8, 8, 8, 8}, "rgba8"},
{RA_CTYPE_UNORM, 4, {16, 16, 16, 16}, "rgba16"},
{RA_CTYPE_UNORM, 4, {10, 10, 10, 2}, "rgb10_a2"},
{RA_CTYPE_UINT, 1, {8}, "r8ui"},
{RA_CTYPE_UINT, 1, {16}, "r16ui"},
{RA_CTYPE_UINT, 1, {32}, "r32ui"},
{RA_CTYPE_UINT, 2, {8, 8}, "rg8ui"},
{RA_CTYPE_UINT, 2, {16, 16}, "rg16ui"},
{RA_CTYPE_UINT, 2, {32, 32}, "rg32ui"},
{RA_CTYPE_UINT, 4, {8, 8, 8, 8}, "rgba8ui"},
{RA_CTYPE_UINT, 4, {16, 16, 16, 16}, "rgba16ui"},
{RA_CTYPE_UINT, 4, {32, 32, 32, 32}, "rgba32ui"},
{RA_CTYPE_UINT, 4, {10, 10, 10, 2}, "rgb10_a2ui"},
};
const char *ra_fmt_glsl_format(const struct ra_format *fmt)
{
for (int n = 0; n < MP_ARRAY_SIZE(ra_glsl_fmts); n++) {
const struct glsl_fmt *gfmt = &ra_glsl_fmts[n];
if (fmt->ctype != gfmt->ctype)
continue;
if (fmt->num_components != gfmt->num_components)
continue;
for (int i = 0; i < fmt->num_components; i++) {
if (fmt->component_depth[i] != gfmt->component_depth[i])
goto next_fmt;
}
return gfmt->glsl_format;
next_fmt: ; // equivalent to `continue`
}
return NULL;
}
// Return whether this is a tightly packed format with no external padding and
// with the same bit size/depth in all components, and the shader returns
// components in the same order as in memory.
static bool ra_format_is_regular(const struct ra_format *fmt)
{
if (!fmt->pixel_size || !fmt->num_components || !fmt->ordered)
return false;
for (int n = 1; n < fmt->num_components; n++) {
if (fmt->component_size[n] != fmt->component_size[0] ||
fmt->component_depth[n] != fmt->component_depth[0])
return false;
}
if (fmt->component_size[0] * fmt->num_components != fmt->pixel_size * 8)
return false;
return true;
}
// Return a regular filterable format using RA_CTYPE_UNORM.
const struct ra_format *ra_find_unorm_format(struct ra *ra,
int bytes_per_component,
int n_components)
{
for (int n = 0; n < ra->num_formats; n++) {
const struct ra_format *fmt = ra->formats[n];
if (fmt->ctype == RA_CTYPE_UNORM && fmt->num_components == n_components &&
fmt->pixel_size == bytes_per_component * n_components &&
fmt->component_depth[0] == bytes_per_component * 8 &&
fmt->linear_filter && ra_format_is_regular(fmt))
return fmt;
}
return NULL;
}
// Return a regular format using RA_CTYPE_UINT.
const struct ra_format *ra_find_uint_format(struct ra *ra,
int bytes_per_component,
int n_components)
{
for (int n = 0; n < ra->num_formats; n++) {
const struct ra_format *fmt = ra->formats[n];
if (fmt->ctype == RA_CTYPE_UINT && fmt->num_components == n_components &&
fmt->pixel_size == bytes_per_component * n_components &&
fmt->component_depth[0] == bytes_per_component * 8 &&
ra_format_is_regular(fmt))
return fmt;
}
return NULL;
}
// Find a float format of any precision that matches the C type of the same
// size for upload.
// May drop bits from the mantissa (such as selecting float16 even if
// bytes_per_component == 32); prefers possibly faster formats first.
static const struct ra_format *ra_find_float_format(struct ra *ra,
int bytes_per_component,
int n_components)
{
// Assumes ra_format are ordered by performance.
// The >=16 check is to avoid catching fringe formats.
for (int n = 0; n < ra->num_formats; n++) {
const struct ra_format *fmt = ra->formats[n];
if (fmt->ctype == RA_CTYPE_FLOAT && fmt->num_components == n_components &&
fmt->pixel_size == bytes_per_component * n_components &&
fmt->component_depth[0] >= 16 &&
fmt->linear_filter && ra_format_is_regular(fmt))
return fmt;
}
return NULL;
}
// Return a filterable regular format that uses at least float16 internally, and
// uses a normal C float for transfer on the CPU side. (This is just so we don't
// need 32->16 bit conversion on CPU, which would be messy.)
const struct ra_format *ra_find_float16_format(struct ra *ra, int n_components)
{
return ra_find_float_format(ra, sizeof(float), n_components);
}
const struct ra_format *ra_find_named_format(struct ra *ra, const char *name)
{
for (int n = 0; n < ra->num_formats; n++) {
const struct ra_format *fmt = ra->formats[n];
if (strcmp(fmt->name, name) == 0)
return fmt;
}
return NULL;
}
// Like ra_find_unorm_format(), but if no fixed point format is available,
// return an unsigned integer format.
static const struct ra_format *find_plane_format(struct ra *ra, int bytes,
int n_channels,
enum mp_component_type ctype)
{
switch (ctype) {
case MP_COMPONENT_TYPE_UINT: {
const struct ra_format *f = ra_find_unorm_format(ra, bytes, n_channels);
if (f)
return f;
return ra_find_uint_format(ra, bytes, n_channels);
}
case MP_COMPONENT_TYPE_FLOAT:
return ra_find_float_format(ra, bytes, n_channels);
default: return NULL;
}
}
// Put a mapping of imgfmt to texture formats into *out. Basically it selects
// the correct texture formats needed to represent an imgfmt in a shader, with
// textures using the same memory organization as on the CPU.
// Each plane is represented by a texture, and each texture has a RGBA
// component order. out->components describes the meaning of them.
// May return integer formats for >8 bit formats, if the driver has no
// normalized 16 bit formats.
// Returns false (and *out is not touched) if no format found.
bool ra_get_imgfmt_desc(struct ra *ra, int imgfmt, struct ra_imgfmt_desc *out)
{
struct ra_imgfmt_desc res = {0};
struct mp_regular_imgfmt regfmt;
if (mp_get_regular_imgfmt(®fmt, imgfmt)) {
enum ra_ctype ctype = RA_CTYPE_UNKNOWN;
res.num_planes = regfmt.num_planes;
res.component_bits = regfmt.component_size * 8;
res.component_pad = regfmt.component_pad;
for (int n = 0; n < regfmt.num_planes; n++) {
struct mp_regular_imgfmt_plane *plane = ®fmt.planes[n];
res.planes[n] = find_plane_format(ra, regfmt.component_size,
plane->num_components,
regfmt.component_type);
if (!res.planes[n])
return false;
for (int i = 0; i < plane->num_components; i++)
res.components[n][i] = plane->components[i];
// Dropping LSBs when shifting will lead to dropped MSBs.
if (res.component_bits > res.planes[n]->component_depth[0] &&
res.component_pad < 0)
return false;
// Renderer restriction, but actually an unwanted corner case.
if (ctype != RA_CTYPE_UNKNOWN && ctype != res.planes[n]->ctype)
return false;
ctype = res.planes[n]->ctype;
}
res.chroma_w = regfmt.chroma_w;
res.chroma_h = regfmt.chroma_h;
goto supported;
}
for (int n = 0; n < ra->num_formats; n++) {
if (imgfmt && ra->formats[n]->special_imgfmt == imgfmt) {
res = *ra->formats[n]->special_imgfmt_desc;
goto supported;
}
}
// Unsupported format
return false;
supported:
*out = res;
return true;
}
void ra_dump_tex_formats(struct ra *ra, int msgl)
{
if (!mp_msg_test(ra->log, msgl))
return;
MP_MSG(ra, msgl, "Texture formats:\n");
MP_MSG(ra, msgl, " NAME COMP*TYPE SIZE DEPTH PER COMP.\n");
for (int n = 0; n < ra->num_formats; n++) {
const struct ra_format *fmt = ra->formats[n];
const char *ctype = "unknown";
switch (fmt->ctype) {
case RA_CTYPE_UNORM: ctype = "unorm"; break;
case RA_CTYPE_UINT: ctype = "uint "; break;
case RA_CTYPE_FLOAT: ctype = "float"; break;
}
char cl[40] = "";
for (int i = 0; i < fmt->num_components; i++) {
mp_snprintf_cat(cl, sizeof(cl), "%s%d", i ? " " : "",
fmt->component_size[i]);
if (fmt->component_size[i] != fmt->component_depth[i])
mp_snprintf_cat(cl, sizeof(cl), "/%d", fmt->component_depth[i]);
}
MP_MSG(ra, msgl, " %-10s %d*%s %3dB %s %s %s {%s}\n", fmt->name,
fmt->num_components, ctype, fmt->pixel_size,
fmt->luminance_alpha ? "LA" : " ",
fmt->linear_filter ? "LF" : " ",
fmt->renderable ? "CR" : " ", cl);
}
MP_MSG(ra, msgl, " LA = LUMINANCE_ALPHA hack format\n");
MP_MSG(ra, msgl, " LF = linear filterable\n");
MP_MSG(ra, msgl, " CR = can be used for render targets\n");
}
void ra_dump_imgfmt_desc(struct ra *ra, const struct ra_imgfmt_desc *desc,
int msgl)
{
char pl[80] = "";
char pf[80] = "";
for (int n = 0; n < desc->num_planes; n++) {
if (n > 0) {
mp_snprintf_cat(pl, sizeof(pl), "/");
mp_snprintf_cat(pf, sizeof(pf), "/");
}
char t[5] = {0};
for (int i = 0; i < 4; i++)
t[i] = "_rgba"[desc->components[n][i]];
for (int i = 3; i > 0 && t[i] == '_'; i--)
t[i] = '\0';
mp_snprintf_cat(pl, sizeof(pl), "%s", t);
mp_snprintf_cat(pf, sizeof(pf), "%s", desc->planes[n]->name);
}
MP_MSG(ra, msgl, "%d planes %dx%d %d/%d [%s] (%s)\n",
desc->num_planes, desc->chroma_w, desc->chroma_h,
desc->component_bits, desc->component_pad, pf, pl);
}
void ra_dump_img_formats(struct ra *ra, int msgl)
{
if (!mp_msg_test(ra->log, msgl))
return;
MP_MSG(ra, msgl, "Image formats:\n");
for (int imgfmt = IMGFMT_START; imgfmt < IMGFMT_END; imgfmt++) {
const char *name = mp_imgfmt_to_name(imgfmt);
if (strcmp(name, "unknown") == 0)
continue;
MP_MSG(ra, msgl, " %s", name);
struct ra_imgfmt_desc desc;
if (ra_get_imgfmt_desc(ra, imgfmt, &desc)) {
MP_MSG(ra, msgl, " => ");
ra_dump_imgfmt_desc(ra, &desc, msgl);
} else {
MP_MSG(ra, msgl, "\n");
}
}
}