RepoMirrors
/
mpv
mirror of https://github.com/mpv-player/mpv
1
0
Fork 0
Code Issues Releases Wiki Activity

vo_gpu_next: switch to unpooled hwdec mapping 

This makes use of the new frame acquire/release callbacks to hold on to
hwdec images only as long as necessary. This should greatly improve the
smoothness/efficiency of hwdec interop, by not holding on to them for
longer than needed.

This also avoids the need to pool hwdec mappers altogether.

Should fix #10067 as well, since frames are now only mapped when we
actually use them.
This commit is contained in:
Niklas Haas 2022-04-09 23:58:59 +02:00 committed by Niklas Haas
parent 2cafe5137f
commit 26a3a06861
5 changed files with 92 additions and 71 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
meson.build
View File

@ -965,14 +965,14 @@ if libplacebo.found()
sources += files('video/out/placebo/ra_pl.c',
'video/out/placebo/utils.c')
pl_api_ver = libplacebo.version().split('.')[1]
if pl_api_ver.version_compare('>=199')
if pl_api_ver.version_compare('>=202')
features += 'libplacebo-next'
libplacebo_next = true
message('libplacebo v4.199+ found! Enabling vo_gpu_next.')
message('libplacebo v4.202+ found! Enabling vo_gpu_next.')
sources += files('video/out/vo_gpu_next.c',
'video/out/gpu_next/context.c')
else
message('libplacebo v4.199+ not found! Disabling vo_gpu_next.')
message('libplacebo v4.202+ not found! Disabling vo_gpu_next.')
endif
endif

2
video/out/gpu/hwdec.c
View File

@ -122,7 +122,7 @@ bool ra_hwdec_test_format(struct ra_hwdec *hwdec, int imgfmt)
}
struct ra_hwdec_mapper *ra_hwdec_mapper_create(struct ra_hwdec *hwdec,
struct mp_image_params *params)
const struct mp_image_params *params)
{
assert(ra_hwdec_test_format(hwdec, params->imgfmt));

2
video/out/gpu/hwdec.h
View File

@ -138,7 +138,7 @@ void ra_hwdec_uninit(struct ra_hwdec *hwdec);
bool ra_hwdec_test_format(struct ra_hwdec *hwdec, int imgfmt);
struct ra_hwdec_mapper *ra_hwdec_mapper_create(struct ra_hwdec *hwdec,
struct mp_image_params *params);
const struct mp_image_params *params);
void ra_hwdec_mapper_free(struct ra_hwdec_mapper **mapper);
void ra_hwdec_mapper_unmap(struct ra_hwdec_mapper *mapper);
int ra_hwdec_mapper_map(struct ra_hwdec_mapper *mapper, struct mp_image *img);

149
video/out/vo_gpu_next.c
View File

@ -90,10 +90,7 @@ struct priv {
struct ra_ctx *ra_ctx;
struct gpu_ctx *context;
struct ra_hwdec_ctx hwdec_ctx;
// Pooled/cached mappers, for performance
struct ra_hwdec_mapper **hwdec_mappers;
int num_hwdec_mappers;
struct ra_hwdec_mapper *hwdec_mapper;
// Allocated DR buffers
pthread_mutex_t dr_lock;
@ -323,7 +320,7 @@ struct frame_priv {
struct vo *vo;
struct osd_state subs;
uint64_t osd_sync;
struct ra_hwdec_mapper *hwdec_mapper;
struct ra_hwdec *hwdec;
};
static int plane_data_from_imgfmt(struct pl_plane_data out_data[4],
@ -459,12 +456,31 @@ static struct pl_color_space get_mpi_csp(struct vo *vo, struct mp_image *mpi)
return csp;
}
// For RAs not based on ra_pl, this creates a new pl_tex wrapper
static pl_tex hwdec_get_tex(struct frame_priv *fp, int n)
static bool hwdec_reconfig(struct priv *p, struct ra_hwdec *hwdec,
const struct mp_image_params *par)
{
struct priv *p = fp->vo->priv;
struct ra_tex *ratex = fp->hwdec_mapper->tex[n];
struct ra *ra = fp->hwdec_mapper->ra;
if (p->hwdec_mapper) {
if (mp_image_params_equal(par, &p->hwdec_mapper->src_params)) {
return p->hwdec_mapper;
} else {
ra_hwdec_mapper_free(&p->hwdec_mapper);
}
}
p->hwdec_mapper = ra_hwdec_mapper_create(hwdec, par);
if (!p->hwdec_mapper) {
MP_ERR(p, "Initializing texture for hardware decoding failed.\n");
return NULL;
}
return p->hwdec_mapper;
}
// For RAs not based on ra_pl, this creates a new pl_tex wrapper
static pl_tex hwdec_get_tex(struct priv *p, int n)
{
struct ra_tex *ratex = p->hwdec_mapper->tex[n];
struct ra *ra = p->hwdec_mapper->ra;
if (ra_pl_get(ra))
return (pl_tex) ratex->priv;
@ -502,35 +518,38 @@ static pl_tex hwdec_get_tex(struct frame_priv *fp, int n)
return false;
}
static void discard_frame(const struct pl_source_frame *src)
static bool hwdec_acquire(pl_gpu gpu, struct pl_frame *frame)
{
struct mp_image *mpi = src->frame_data;
talloc_free(mpi);
}
static void unmap_frame(pl_gpu gpu, struct pl_frame *frame,
const struct pl_source_frame *src)
{
struct mp_image *mpi = src->frame_data;
struct mp_image *mpi = frame->user_data;
struct frame_priv *fp = mpi->priv;
struct priv *p = fp->vo->priv;
if (fp->hwdec_mapper) {
// Clean up after wrapped plane textures
if (!ra_pl_get(fp->hwdec_mapper->ra)) {
for (int n = 0; n < frame->num_planes; n++)
pl_tex_destroy(p->gpu, &frame->planes[n].texture);
}
if (!hwdec_reconfig(p, fp->hwdec, &mpi->params))
return false;
ra_hwdec_mapper_unmap(fp->hwdec_mapper);
MP_TARRAY_APPEND(p, p->hwdec_mappers, p->num_hwdec_mappers, fp->hwdec_mapper);
fp->hwdec_mapper = NULL;
if (ra_hwdec_mapper_map(p->hwdec_mapper, mpi) < 0) {
MP_ERR(p, "Mapping hardware decoded surface failed.\n");
return false;
}
for (int i = 0; i < MP_ARRAY_SIZE(fp->subs.entries); i++) {
pl_tex tex = fp->subs.entries[i].tex;
if (tex)
MP_TARRAY_APPEND(p, p->sub_tex, p->num_sub_tex, tex);
for (int n = 0; n < frame->num_planes; n++) {
if (!(frame->planes[n].texture = hwdec_get_tex(p, n)))
return false;
}
talloc_free(mpi);
return true;
}
static void hwdec_release(pl_gpu gpu, struct pl_frame *frame)
{
struct mp_image *mpi = frame->user_data;
struct frame_priv *fp = mpi->priv;
struct priv *p = fp->vo->priv;
if (!ra_pl_get(p->hwdec_mapper->ra)) {
for (int n = 0; n < frame->num_planes; n++)
pl_tex_destroy(p->gpu, &frame->planes[n].texture);
}
ra_hwdec_mapper_unmap(p->hwdec_mapper);
}
static bool map_frame(pl_gpu gpu, pl_tex *tex, const struct pl_source_frame *src,
@ -542,31 +561,18 @@ static bool map_frame(pl_gpu gpu, pl_tex *tex, const struct pl_source_frame *src
struct vo *vo = fp->vo;
struct priv *p = vo->priv;
struct ra_hwdec *hwdec = ra_hwdec_get(&p->hwdec_ctx, mpi->imgfmt);
if (hwdec) {
if (MP_TARRAY_POP(p->hwdec_mappers, p->num_hwdec_mappers, &fp->hwdec_mapper)) {
if (!mp_image_params_equal(&mpi->params, &fp->hwdec_mapper->src_params))
ra_hwdec_mapper_free(&fp->hwdec_mapper);
}
if (!fp->hwdec_mapper) {
fp->hwdec_mapper = ra_hwdec_mapper_create(hwdec, &mpi->params);
if (!fp->hwdec_mapper) {
MP_ERR(p, "Initializing texture for hardware decoding failed.\n");
talloc_free(mpi);
return false;
}
}
if (ra_hwdec_mapper_map(fp->hwdec_mapper, mpi) < 0) {
MP_ERR(p, "Mapping hardware decoded surface failed.\n");
MP_TARRAY_APPEND(p, p->hwdec_mappers, p->num_hwdec_mappers, fp->hwdec_mapper);
fp->hwdec_mapper = NULL;
fp->hwdec = ra_hwdec_get(&p->hwdec_ctx, mpi->imgfmt);
if (fp->hwdec) {
// Note: We don't actually need the mapper to map the frame yet, we
// only reconfig the mapper here (potentially creating it) to access
// `dst_params`. In practice, though, this should not matter unless the
// image format changes mid-stream.
if (!hwdec_reconfig(p, fp->hwdec, &mpi->params)) {
talloc_free(mpi);
return false;
}
par = &fp->hwdec_mapper->dst_params;
par = &p->hwdec_mapper->dst_params;
}
*frame = (struct pl_frame) {
@ -601,18 +607,14 @@ static bool map_frame(pl_gpu gpu, pl_tex *tex, const struct pl_source_frame *src
default: break;
}
if (hwdec) {
if (fp->hwdec) {
struct mp_imgfmt_desc desc = mp_imgfmt_get_desc(par->imgfmt);
frame->acquire = hwdec_acquire;
frame->release = hwdec_release;
frame->num_planes = desc.num_planes;
for (int n = 0; n < frame->num_planes; n++) {
struct pl_plane *plane = &frame->planes[n];
plane->texture = hwdec_get_tex(fp, n);
if (!plane->texture) {
unmap_frame(gpu, frame, src);
return false;
}
int *map = plane->component_mapping;
for (int c = 0; c < mp_imgfmt_desc_get_num_comps(&desc); c++) {
if (desc.comps[c].plane != n)
@ -659,7 +661,7 @@ static bool map_frame(pl_gpu gpu, pl_tex *tex, const struct pl_source_frame *src
if (!pl_upload_plane(gpu, plane, &tex[n], &data[n])) {
MP_ERR(vo, "Failed uploading frame!\n");
talloc_free(data[n].priv);
unmap_frame(gpu, frame, src);
talloc_free(mpi);
return false;
}
}
@ -704,6 +706,26 @@ static bool map_frame(pl_gpu gpu, pl_tex *tex, const struct pl_source_frame *src
return true;
}
static void unmap_frame(pl_gpu gpu, struct pl_frame *frame,
const struct pl_source_frame *src)
{
struct mp_image *mpi = src->frame_data;
struct frame_priv *fp = mpi->priv;
struct priv *p = fp->vo->priv;
for (int i = 0; i < MP_ARRAY_SIZE(fp->subs.entries); i++) {
pl_tex tex = fp->subs.entries[i].tex;
if (tex)
MP_TARRAY_APPEND(p, p->sub_tex, p->num_sub_tex, tex);
}
talloc_free(mpi);
}
static void discard_frame(const struct pl_source_frame *src)
{
struct mp_image *mpi = src->frame_data;
talloc_free(mpi);
}
static void info_callback(void *priv, const struct pl_render_info *info)
{
struct vo *vo = priv;
@ -1284,8 +1306,7 @@ static void uninit(struct vo *vo)
pl_mpv_user_shader_destroy(&p->user_hooks[i].hook);
if (vo->hwdec_devs) {
for (int n = 0; n < p->num_hwdec_mappers; n++)
ra_hwdec_mapper_free(&p->hwdec_mappers[n]);
ra_hwdec_mapper_free(&p->hwdec_mapper);
ra_hwdec_ctx_uninit(&p->hwdec_ctx);
hwdec_devices_set_loader(vo->hwdec_devs, NULL, NULL);
hwdec_devices_destroy(vo->hwdec_devs);

4
wscript
View File

@ -741,9 +741,9 @@ video_output_features = [
'func': check_pkg_config('libplacebo >= 4.157.0'),
}, {
'name': 'libplacebo-next',
'desc': 'libplacebo v4.199+, needed for vo_gpu_next',
'desc': 'libplacebo v4.202+, needed for vo_gpu_next',
'deps': 'libplacebo',
'func': check_preprocessor('libplacebo/config.h', 'PL_API_VER >= 199',
'func': check_preprocessor('libplacebo/config.h', 'PL_API_VER >= 202',
use='libplacebo'),
}, {
'name': '--vulkan',