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

751 lines
24 KiB
C
Raw Normal View History

vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
#include "common/common.h"
#include "common/msg.h"
#include "ra_pl.h"
#include "utils.h"
struct ra_pl {
const struct pl_gpu *gpu;
vo_gpu: ra_pl: add timers support Added in libplacebo v60, unfortunately with some changes in design that make it a bit of an awkward fit for the way timers are used in mpv. Timer queries in libplacebo don't support "start" and "stop"-style operations, and instead are attached directly to operations. The only sane way of implementing this in the ra API is to have a single 'active timer' that gets attached to every pass, taking care to sort distinct operations into distinct pl_timer queries within that ra_timer. This design unfortunately doesn't let us have multiple 'active timers' concurrently, similar to the current such limitation in ra_gl. But it's also not a big deal.
2020-05-16 04:10:50 +00:00
struct ra_timer_pl *active_timer;
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
};
vo_gpu: ra_pl: Add getter for pl_gpu We need access to the underlying pl_gpu to make libplacebo calls from hwdecs.
2018-11-27 04:35:47 +00:00
static inline const struct pl_gpu *get_gpu(const struct ra *ra)
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
{
struct ra_pl *p = ra->priv;
return p->gpu;
}
static struct ra_fns ra_fns_pl;
vo_gpu: ra_pl: Add getter for pl_gpu We need access to the underlying pl_gpu to make libplacebo calls from hwdecs.
2018-11-27 04:35:47 +00:00
const struct pl_gpu *ra_pl_get(const struct ra *ra)
{
return ra->fns == &ra_fns_pl ? get_gpu(ra) : NULL;
}
vo_gpu: ra_pl: add timers support Added in libplacebo v60, unfortunately with some changes in design that make it a bit of an awkward fit for the way timers are used in mpv. Timer queries in libplacebo don't support "start" and "stop"-style operations, and instead are attached directly to operations. The only sane way of implementing this in the ra API is to have a single 'active timer' that gets attached to every pass, taking care to sort distinct operations into distinct pl_timer queries within that ra_timer. This design unfortunately doesn't let us have multiple 'active timers' concurrently, similar to the current such limitation in ra_gl. But it's also not a big deal.
2020-05-16 04:10:50 +00:00
static struct pl_timer *get_active_timer(const struct ra *ra);
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
struct ra *ra_create_pl(const struct pl_gpu *gpu, struct mp_log *log)
{
assert(gpu);
struct ra *ra = talloc_zero(NULL, struct ra);
ra->log = log;
ra->fns = &ra_fns_pl;
struct ra_pl *p = ra->priv = talloc_zero(ra, struct ra_pl);
p->gpu = gpu;
ra->glsl_version = gpu->glsl.version;
ra->glsl_vulkan = gpu->glsl.vulkan;
ra->glsl_es = gpu->glsl.gles;
ra->caps = RA_CAP_DIRECT_UPLOAD | RA_CAP_NESTED_ARRAY | RA_CAP_FRAGCOORD;
if (gpu->caps & PL_GPU_CAP_COMPUTE)
ra->caps |= RA_CAP_COMPUTE | RA_CAP_NUM_GROUPS;
if (gpu->caps & PL_GPU_CAP_PARALLEL_COMPUTE)
ra->caps |= RA_CAP_PARALLEL_COMPUTE;
if (gpu->caps & PL_GPU_CAP_INPUT_VARIABLES)
ra->caps |= RA_CAP_GLOBAL_UNIFORM;
if (gpu->limits.max_tex_1d_dim)
ra->caps |= RA_CAP_TEX_1D;
if (gpu->limits.max_tex_3d_dim)
ra->caps |= RA_CAP_TEX_3D;
if (gpu->limits.max_ubo_size)
ra->caps |= RA_CAP_BUF_RO;
if (gpu->limits.max_ssbo_size)
ra->caps |= RA_CAP_BUF_RW;
if (gpu->limits.min_gather_offset && gpu->limits.max_gather_offset)
ra->caps |= RA_CAP_GATHER;
// Semi-hack: assume all textures are blittable if r8 is
const struct pl_fmt *r8 = pl_find_named_fmt(gpu, "r8");
if (r8->caps & PL_FMT_CAP_BLITTABLE)
ra->caps |= RA_CAP_BLIT;
ra->max_texture_wh = gpu->limits.max_tex_2d_dim;
ra->max_shmem = gpu->limits.max_shmem_size;
ra->max_pushc_size = gpu->limits.max_pushc_size;
// Set up format wrappers
for (int i = 0; i < gpu->num_formats; i++) {
const struct pl_fmt *plfmt = gpu->formats[i];
static const enum ra_ctype fmt_type_map[PL_FMT_TYPE_COUNT] = {
[PL_FMT_UNORM] = RA_CTYPE_UNORM,
[PL_FMT_UINT] = RA_CTYPE_UINT,
[PL_FMT_FLOAT] = RA_CTYPE_FLOAT,
};
enum ra_ctype type = fmt_type_map[plfmt->type];
if (!type || !(plfmt->caps & PL_FMT_CAP_SAMPLEABLE))
continue;
struct ra_format *rafmt = talloc_zero(ra, struct ra_format);
*rafmt = (struct ra_format) {
.name = plfmt->name,
.priv = (void *) plfmt,
.ctype = type,
.ordered = pl_fmt_is_ordered(plfmt),
.num_components = plfmt->num_components,
.pixel_size = plfmt->texel_size,
.linear_filter = plfmt->caps & PL_FMT_CAP_LINEAR,
.renderable = plfmt->caps & PL_FMT_CAP_RENDERABLE,
video/out/gpu: Add a `storable` flag to ra_format While `ra` supports the concept of a texture as a storage destination, it does not support the concept of a texture format being usable for a storage texture. This can lead to us attempting to create a texture from an incompatible format, with undefined results. So, let's introduce an explicit format flag for storage and use it. In `ra_pl` we can simply reflect the `storable` flag. For GL and D3D, we'll need to write some new code to do the compatibility checks. I'm not going to do it here because it's not a regression; we were already implicitly assuming all formats were storable. Fixes #6657
2019-06-13 19:30:18 +00:00
.storable = plfmt->caps & PL_FMT_CAP_STORABLE,
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
.glsl_format = plfmt->glsl_format,
};
for (int c = 0; c < plfmt->num_components; c++) {
rafmt->component_size[c] = plfmt->host_bits[c];
rafmt->component_depth[c] = plfmt->component_depth[c];
}
MP_TARRAY_APPEND(ra, ra->formats, ra->num_formats, rafmt);
}
return ra;
}
static void destroy_ra_pl(struct ra *ra)
{
talloc_free(ra);
}
static struct ra_format *map_fmt(struct ra *ra, const struct pl_fmt *plfmt)
{
for (int i = 0; i < ra->num_formats; i++) {
if (ra->formats[i]->priv == plfmt)
return ra->formats[i];
}
MP_ERR(ra, "Failed mapping pl_fmt '%s' to ra_fmt?\n", plfmt->name);
return NULL;
}
bool mppl_wrap_tex(struct ra *ra, const struct pl_tex *pltex,
struct ra_tex *out_tex)
{
if (!pltex)
return false;
*out_tex = (struct ra_tex) {
.params = {
.dimensions = pl_tex_params_dimension(pltex->params),
.w = pltex->params.w,
.h = pltex->params.h,
.d = pltex->params.d,
.format = map_fmt(ra, pltex->params.format),
.render_src = pltex->params.sampleable,
.render_dst = pltex->params.renderable,
.storage_dst = pltex->params.storable,
.blit_src = pltex->params.blit_src,
.blit_dst = pltex->params.blit_dst,
.host_mutable = pltex->params.host_writable,
.downloadable = pltex->params.host_readable,
vo_gpu: placebo: update for upstream API changes The concept of sample/address modes was moved from `pl_tex` to `pl_desc_binding`. The `pl_tex_blit()` function also underwent an API change.
2020-12-03 07:25:23 +00:00
#if PL_API_VER >= 103
// These don't exist upstream, so just pick something reasonable
.src_linear = pltex->params.format->caps & PL_FMT_CAP_LINEAR,
.src_repeat = false,
#else
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
.src_linear = pltex->params.sample_mode == PL_TEX_SAMPLE_LINEAR,
.src_repeat = pltex->params.address_mode == PL_TEX_ADDRESS_REPEAT,
vo_gpu: placebo: update for upstream API changes The concept of sample/address modes was moved from `pl_tex` to `pl_desc_binding`. The `pl_tex_blit()` function also underwent an API change.
2020-12-03 07:25:23 +00:00
#endif
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
},
.priv = (void *) pltex,
};
return !!out_tex->params.format;
}
static struct ra_tex *tex_create_pl(struct ra *ra,
const struct ra_tex_params *params)
{
const struct pl_gpu *gpu = get_gpu(ra);
const struct pl_tex *pltex = pl_tex_create(gpu, &(struct pl_tex_params) {
.w = params->w,
.h = params->dimensions >= 2 ? params->h : 0,
.d = params->dimensions >= 3 ? params->d : 0,
.format = params->format->priv,
.sampleable = params->render_src,
.renderable = params->render_dst,
.storable = params->storage_dst,
.blit_src = params->blit_src,
.blit_dst = params->blit_dst || params->render_dst,
.host_writable = params->host_mutable,
.host_readable = params->downloadable,
vo_gpu: placebo: update for upstream API changes The concept of sample/address modes was moved from `pl_tex` to `pl_desc_binding`. The `pl_tex_blit()` function also underwent an API change.
2020-12-03 07:25:23 +00:00
#if PL_API_VER < 103
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
.sample_mode = params->src_linear ? PL_TEX_SAMPLE_LINEAR
: PL_TEX_SAMPLE_NEAREST,
.address_mode = params->src_repeat ? PL_TEX_ADDRESS_REPEAT
: PL_TEX_ADDRESS_CLAMP,
vo_gpu: placebo: update for upstream API changes The concept of sample/address modes was moved from `pl_tex` to `pl_desc_binding`. The `pl_tex_blit()` function also underwent an API change.
2020-12-03 07:25:23 +00:00
#endif
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
.initial_data = params->initial_data,
});
struct ra_tex *ratex = talloc_ptrtype(NULL, ratex);
if (!mppl_wrap_tex(ra, pltex, ratex)) {
pl_tex_destroy(gpu, &pltex);
talloc_free(ratex);
return NULL;
}
vo_gpu: placebo: keep track of texture sample mode This fixes an issue where dithering was effectively broken on libplacebo versions >= 103 because the dither texture was being sampled with edge-clamped rather than repeating semantics.
2021-03-21 16:18:20 +00:00
// Keep track of these, so we can correctly bind them later
ratex->params.src_repeat = params->src_repeat;
ratex->params.src_linear = params->src_linear;
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
return ratex;
}
static void tex_destroy_pl(struct ra *ra, struct ra_tex *tex)
{
if (!tex)
return;
pl_tex_destroy(get_gpu(ra), (const struct pl_tex **) &tex->priv);
talloc_free(tex);
}
static bool tex_upload_pl(struct ra *ra, const struct ra_tex_upload_params *params)
{
vo_gpu: placebo: add fallback code for stride mismatch For cases in which the requirements of the GPU API prevent directly uploading a texture with a given stride, we need to fix the stride manually in host memory. This incurs an extra memcpy, but there's not much we can do about it. (Even in `ra_gl` land, the driver will just hide this memcpy from the user) Note: This code could be done better. It could only copy as many texels as needed, and it could pick a stride that's a multiple of `gpu->limits.align_tex_xfer_stride` for better performance. Patches welcome (tm) Fixes #7759
2020-06-16 00:19:13 +00:00
const struct pl_gpu *gpu = get_gpu(ra);
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
const struct pl_tex *tex = params->tex->priv;
struct pl_tex_transfer_params pl_params = {
.tex = tex,
.buf = params->buf ? params->buf->priv : NULL,
.buf_offset = params->buf_offset,
.ptr = (void *) params->src,
vo_gpu: ra_pl: add timers support Added in libplacebo v60, unfortunately with some changes in design that make it a bit of an awkward fit for the way timers are used in mpv. Timer queries in libplacebo don't support "start" and "stop"-style operations, and instead are attached directly to operations. The only sane way of implementing this in the ra API is to have a single 'active timer' that gets attached to every pass, taking care to sort distinct operations into distinct pl_timer queries within that ra_timer. This design unfortunately doesn't let us have multiple 'active timers' concurrently, similar to the current such limitation in ra_gl. But it's also not a big deal.
2020-05-16 04:10:50 +00:00
.timer = get_active_timer(ra),
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
};
vo_gpu: placebo: add fallback code for stride mismatch For cases in which the requirements of the GPU API prevent directly uploading a texture with a given stride, we need to fix the stride manually in host memory. This incurs an extra memcpy, but there's not much we can do about it. (Even in `ra_gl` land, the driver will just hide this memcpy from the user) Note: This code could be done better. It could only copy as many texels as needed, and it could pick a stride that's a multiple of `gpu->limits.align_tex_xfer_stride` for better performance. Patches welcome (tm) Fixes #7759
2020-06-16 00:19:13 +00:00
const struct pl_buf *staging = NULL;
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
if (params->tex->params.dimensions == 2) {
vo_gpu: placebo: add fallback code for stride mismatch For cases in which the requirements of the GPU API prevent directly uploading a texture with a given stride, we need to fix the stride manually in host memory. This incurs an extra memcpy, but there's not much we can do about it. (Even in `ra_gl` land, the driver will just hide this memcpy from the user) Note: This code could be done better. It could only copy as many texels as needed, and it could pick a stride that's a multiple of `gpu->limits.align_tex_xfer_stride` for better performance. Patches welcome (tm) Fixes #7759
2020-06-16 00:19:13 +00:00
size_t texel_size = tex->params.format->texel_size;
pl_params.stride_w = params->stride / texel_size;
size_t stride = pl_params.stride_w * texel_size;
int lines = tex->params.h;
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
if (params->rc) {
pl_params.rc = (struct pl_rect3d) {
.x0 = params->rc->x0, .x1 = params->rc->x1,
.y0 = params->rc->y0, .y1 = params->rc->y1,
};
vo_gpu: placebo: add fallback code for stride mismatch For cases in which the requirements of the GPU API prevent directly uploading a texture with a given stride, we need to fix the stride manually in host memory. This incurs an extra memcpy, but there's not much we can do about it. (Even in `ra_gl` land, the driver will just hide this memcpy from the user) Note: This code could be done better. It could only copy as many texels as needed, and it could pick a stride that's a multiple of `gpu->limits.align_tex_xfer_stride` for better performance. Patches welcome (tm) Fixes #7759
2020-06-16 00:19:13 +00:00
lines = pl_rect_h(pl_params.rc);
}
if (stride != params->stride) {
// Fall back to uploading via a staging buffer prepared in CPU
staging = pl_buf_create(gpu, &(struct pl_buf_params) {
.type = PL_BUF_TEX_TRANSFER,
.size = lines * stride,
.memory_type = PL_BUF_MEM_HOST,
.host_mapped = true,
});
if (!staging)
return false;
const uint8_t *src = params->buf ? params->buf->data : params->src;
assert(src);
for (int y = 0; y < lines; y++)
memcpy(staging->data + y * stride, src + y * params->stride, stride);
pl_params.ptr = NULL;
pl_params.buf = staging;
pl_params.buf_offset = 0;
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
}
}
vo_gpu: placebo: add fallback code for stride mismatch For cases in which the requirements of the GPU API prevent directly uploading a texture with a given stride, we need to fix the stride manually in host memory. This incurs an extra memcpy, but there's not much we can do about it. (Even in `ra_gl` land, the driver will just hide this memcpy from the user) Note: This code could be done better. It could only copy as many texels as needed, and it could pick a stride that's a multiple of `gpu->limits.align_tex_xfer_stride` for better performance. Patches welcome (tm) Fixes #7759
2020-06-16 00:19:13 +00:00
bool ok = pl_tex_upload(gpu, &pl_params);
pl_buf_destroy(gpu, &staging);
return ok;
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
}
static bool tex_download_pl(struct ra *ra, struct ra_tex_download_params *params)
{
const struct pl_tex *tex = params->tex->priv;
vo_gpu: placebo: add fallback code for stride mismatch For cases in which the requirements of the GPU API prevent directly uploading a texture with a given stride, we need to fix the stride manually in host memory. This incurs an extra memcpy, but there's not much we can do about it. (Even in `ra_gl` land, the driver will just hide this memcpy from the user) Note: This code could be done better. It could only copy as many texels as needed, and it could pick a stride that's a multiple of `gpu->limits.align_tex_xfer_stride` for better performance. Patches welcome (tm) Fixes #7759
2020-06-16 00:19:13 +00:00
size_t texel_size = tex->params.format->texel_size;
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
struct pl_tex_transfer_params pl_params = {
.tex = tex,
.ptr = params->dst,
vo_gpu: placebo: add fallback code for stride mismatch For cases in which the requirements of the GPU API prevent directly uploading a texture with a given stride, we need to fix the stride manually in host memory. This incurs an extra memcpy, but there's not much we can do about it. (Even in `ra_gl` land, the driver will just hide this memcpy from the user) Note: This code could be done better. It could only copy as many texels as needed, and it could pick a stride that's a multiple of `gpu->limits.align_tex_xfer_stride` for better performance. Patches welcome (tm) Fixes #7759
2020-06-16 00:19:13 +00:00
.stride_w = params->stride / texel_size,
vo_gpu: ra_pl: add timers support Added in libplacebo v60, unfortunately with some changes in design that make it a bit of an awkward fit for the way timers are used in mpv. Timer queries in libplacebo don't support "start" and "stop"-style operations, and instead are attached directly to operations. The only sane way of implementing this in the ra API is to have a single 'active timer' that gets attached to every pass, taking care to sort distinct operations into distinct pl_timer queries within that ra_timer. This design unfortunately doesn't let us have multiple 'active timers' concurrently, similar to the current such limitation in ra_gl. But it's also not a big deal.
2020-05-16 04:10:50 +00:00
.timer = get_active_timer(ra),
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
};
vo_gpu: placebo: add fallback code for stride mismatch For cases in which the requirements of the GPU API prevent directly uploading a texture with a given stride, we need to fix the stride manually in host memory. This incurs an extra memcpy, but there's not much we can do about it. (Even in `ra_gl` land, the driver will just hide this memcpy from the user) Note: This code could be done better. It could only copy as many texels as needed, and it could pick a stride that's a multiple of `gpu->limits.align_tex_xfer_stride` for better performance. Patches welcome (tm) Fixes #7759
2020-06-16 00:19:13 +00:00
uint8_t *staging = NULL;
size_t stride = pl_params.stride_w * texel_size;
if (stride != params->stride) {
staging = talloc_size(NULL, tex->params.h * stride);
pl_params.ptr = staging;
}
bool ok = pl_tex_download(get_gpu(ra), &pl_params);
if (ok && staging) {
for (int y = 0; y < tex->params.h; y++) {
memcpy((uint8_t *) params->dst + y * params->stride,
staging + y * stride,
stride);
}
}
talloc_free(staging);
return ok;
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
}
static struct ra_buf *buf_create_pl(struct ra *ra,
const struct ra_buf_params *params)
{
static const enum pl_buf_type buf_type[] = {
[RA_BUF_TYPE_TEX_UPLOAD] = PL_BUF_TEX_TRANSFER,
[RA_BUF_TYPE_SHADER_STORAGE] = PL_BUF_STORAGE,
[RA_BUF_TYPE_UNIFORM] = PL_BUF_UNIFORM,
[RA_BUF_TYPE_SHARED_MEMORY] = 0,
};
vo_gpu: libplacebo: require v2.72.0 It's about a year old, and packaged pretty much everywhere that bothers to package libplacebo at all. Older versions are only a thing on LTS, which will probably also use older mpv so it works out. Starting with v2.72.0, libplacebo validates all of its parameters internally and turns them into function failures. Doing it twice is awfully redundant, so we can drop the parameter validation. Also allows us to drop some preprocessor macros.
2021-02-16 13:36:39 +00:00
const struct pl_buf *plbuf = pl_buf_create(get_gpu(ra), &(struct pl_buf_params) {
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
.type = buf_type[params->type],
.size = params->size,
.host_mapped = params->host_mapped,
.host_writable = params->host_mutable,
.initial_data = params->initial_data,
});
if (!plbuf)
return NULL;
struct ra_buf *rabuf = talloc_ptrtype(NULL, rabuf);
*rabuf = (struct ra_buf) {
.params = *params,
.data = plbuf->data,
.priv = (void *) plbuf,
};
rabuf->params.initial_data = NULL;
return rabuf;
}
static void buf_destroy_pl(struct ra *ra, struct ra_buf *buf)
{
if (!buf)
return;
pl_buf_destroy(get_gpu(ra), (const struct pl_buf **) &buf->priv);
talloc_free(buf);
}
static void buf_update_pl(struct ra *ra, struct ra_buf *buf, ptrdiff_t offset,
const void *data, size_t size)
{
pl_buf_write(get_gpu(ra), buf->priv, offset, data, size);
}
static bool buf_poll_pl(struct ra *ra, struct ra_buf *buf)
{
return !pl_buf_poll(get_gpu(ra), buf->priv, 0);
}
static void clear_pl(struct ra *ra, struct ra_tex *dst, float color[4],
struct mp_rect *scissor)
{
// TODO: implement scissor clearing by bltting a 1x1 tex instead
pl_tex_clear(get_gpu(ra), dst->priv, color);
}
static void blit_pl(struct ra *ra, struct ra_tex *dst, struct ra_tex *src,
struct mp_rect *dst_rc, struct mp_rect *src_rc)
{
struct pl_rect3d plsrc = {0}, pldst = {0};
if (src_rc) {
plsrc.x0 = MPMIN(MPMAX(src_rc->x0, 0), src->params.w);
plsrc.y0 = MPMIN(MPMAX(src_rc->y0, 0), src->params.h);
plsrc.x1 = MPMIN(MPMAX(src_rc->x1, 0), src->params.w);
plsrc.y1 = MPMIN(MPMAX(src_rc->y1, 0), src->params.h);
}
if (dst_rc) {
pldst.x0 = MPMIN(MPMAX(dst_rc->x0, 0), dst->params.w);
pldst.y0 = MPMIN(MPMAX(dst_rc->y0, 0), dst->params.h);
pldst.x1 = MPMIN(MPMAX(dst_rc->x1, 0), dst->params.w);
pldst.y1 = MPMIN(MPMAX(dst_rc->y1, 0), dst->params.h);
}
vo_gpu: placebo: update for upstream API changes The concept of sample/address modes was moved from `pl_tex` to `pl_desc_binding`. The `pl_tex_blit()` function also underwent an API change.
2020-12-03 07:25:23 +00:00
#if PL_API_VER >= 103
pl_tex_blit(get_gpu(ra), &(struct pl_tex_blit_params) {
.src = src->priv,
.dst = dst->priv,
.src_rc = plsrc,
.dst_rc = pldst,
.sample_mode = src->params.src_linear ? PL_TEX_SAMPLE_LINEAR
: PL_TEX_SAMPLE_NEAREST,
});
#else
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
pl_tex_blit(get_gpu(ra), dst->priv, src->priv, pldst, plsrc);
vo_gpu: placebo: update for upstream API changes The concept of sample/address modes was moved from `pl_tex` to `pl_desc_binding`. The `pl_tex_blit()` function also underwent an API change.
2020-12-03 07:25:23 +00:00
#endif
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
}
static const enum pl_var_type var_type[RA_VARTYPE_COUNT] = {
[RA_VARTYPE_INT] = PL_VAR_SINT,
[RA_VARTYPE_FLOAT] = PL_VAR_FLOAT,
};
static const enum pl_desc_type desc_type[RA_VARTYPE_COUNT] = {
[RA_VARTYPE_TEX] = PL_DESC_SAMPLED_TEX,
[RA_VARTYPE_IMG_W] = PL_DESC_STORAGE_IMG,
[RA_VARTYPE_BUF_RO] = PL_DESC_BUF_UNIFORM,
[RA_VARTYPE_BUF_RW] = PL_DESC_BUF_STORAGE,
};
static const enum pl_fmt_type fmt_type[RA_VARTYPE_COUNT] = {
[RA_VARTYPE_INT] = PL_FMT_SINT,
[RA_VARTYPE_FLOAT] = PL_FMT_FLOAT,
[RA_VARTYPE_BYTE_UNORM] = PL_FMT_UNORM,
};
static const size_t var_size[RA_VARTYPE_COUNT] = {
[RA_VARTYPE_INT] = sizeof(int),
[RA_VARTYPE_FLOAT] = sizeof(float),
[RA_VARTYPE_BYTE_UNORM] = sizeof(uint8_t),
};
static struct ra_layout uniform_layout_pl(struct ra_renderpass_input *inp)
{
// To get the alignment requirements, we try laying this out with
// an offset of 1 and then see where it ends up. This will always be
// the minimum alignment requirement.
struct pl_var_layout layout = pl_buf_uniform_layout(1, &(struct pl_var) {
.name = inp->name,
.type = var_type[inp->type],
.dim_v = inp->dim_v,
.dim_m = inp->dim_m,
.dim_a = 1,
});
return (struct ra_layout) {
.align = layout.offset,
.stride = layout.stride,
.size = layout.size,
};
}
static struct ra_layout push_constant_layout_pl(struct ra_renderpass_input *inp)
{
struct pl_var_layout layout = pl_push_constant_layout(1, &(struct pl_var) {
.name = inp->name,
.type = var_type[inp->type],
.dim_v = inp->dim_v,
.dim_m = inp->dim_m,
.dim_a = 1,
});
return (struct ra_layout) {
.align = layout.offset,
.stride = layout.stride,
.size = layout.size,
};
}
static int desc_namespace_pl(struct ra *ra, enum ra_vartype type)
{
return pl_desc_namespace(get_gpu(ra), desc_type[type]);
}
struct pass_priv {
const struct pl_pass *pl_pass;
uint16_t *inp_index; // index translation map
// Space to hold the descriptor bindings and variable updates
struct pl_desc_binding *binds;
struct pl_var_update *varups;
int num_varups;
};
static struct ra_renderpass *renderpass_create_pl(struct ra *ra,
const struct ra_renderpass_params *params)
{
void *tmp = talloc_new(NULL);
const struct pl_gpu *gpu = get_gpu(ra);
struct ra_renderpass *pass = NULL;
static const enum pl_pass_type pass_type[] = {
[RA_RENDERPASS_TYPE_RASTER] = PL_PASS_RASTER,
[RA_RENDERPASS_TYPE_COMPUTE] = PL_PASS_COMPUTE,
};
struct pl_var *vars = NULL;
struct pl_desc *descs = NULL;
int num_vars = 0, num_descs = 0;
struct pass_priv *priv = talloc_ptrtype(tmp, priv);
priv->inp_index = talloc_zero_array(priv, uint16_t, params->num_inputs);
for (int i = 0; i < params->num_inputs; i++) {
const struct ra_renderpass_input *inp = &params->inputs[i];
if (var_type[inp->type]) {
priv->inp_index[i] = num_vars;
MP_TARRAY_APPEND(tmp, vars, num_vars, (struct pl_var) {
.name = inp->name,
.type = var_type[inp->type],
.dim_v = inp->dim_v,
.dim_m = inp->dim_m,
.dim_a = 1,
});
} else if (desc_type[inp->type]) {
priv->inp_index[i] = num_descs;
MP_TARRAY_APPEND(tmp, descs, num_descs, (struct pl_desc) {
.name = inp->name,
.type = desc_type[inp->type],
.binding = inp->binding,
.access = inp->type == RA_VARTYPE_IMG_W ? PL_DESC_ACCESS_WRITEONLY
: inp->type == RA_VARTYPE_BUF_RW ? PL_DESC_ACCESS_READWRITE
: PL_DESC_ACCESS_READONLY,
});
}
}
// Allocate space to store the bindings map persistently
priv->binds = talloc_zero_array(priv, struct pl_desc_binding, num_descs);
struct pl_pass_params pl_params = {
.type = pass_type[params->type],
.variables = vars,
.num_variables = num_vars,
.descriptors = descs,
.num_descriptors = num_descs,
.push_constants_size = params->push_constants_size,
.glsl_shader = params->type == RA_RENDERPASS_TYPE_COMPUTE
? params->compute_shader
: params->frag_shader,
.cached_program = params->cached_program.start,
.cached_program_len = params->cached_program.len,
};
struct pl_blend_params blend_params;
if (params->type == RA_RENDERPASS_TYPE_RASTER) {
pl_params.vertex_shader = params->vertex_shader;
pl_params.vertex_type = PL_PRIM_TRIANGLE_LIST;
pl_params.vertex_stride = params->vertex_stride;
pl_params.target_dummy.params.format = params->target_format->priv;
pl_params.load_target = !params->invalidate_target;
if (params->enable_blend) {
pl_params.blend_params = &blend_params;
blend_params = (struct pl_blend_params) {
// Same enum order as ra_blend
.src_rgb = (enum ra_blend) params->blend_src_rgb,
.dst_rgb = (enum ra_blend) params->blend_dst_rgb,
.src_alpha = (enum ra_blend) params->blend_src_alpha,
.dst_alpha = (enum ra_blend) params->blend_dst_alpha,
};
}
for (int i = 0; i < params->num_vertex_attribs; i++) {
const struct ra_renderpass_input *inp = &params->vertex_attribs[i];
struct pl_vertex_attrib attrib = {
.name = inp->name,
.offset = inp->offset,
.location = i,
.fmt = pl_find_fmt(gpu, fmt_type[inp->type], inp->dim_v, 0,
var_size[inp->type] * 8, PL_FMT_CAP_VERTEX),
};
if (!attrib.fmt) {
MP_ERR(ra, "Failed mapping vertex attrib '%s' to pl_fmt?\n",
inp->name);
goto error;
}
MP_TARRAY_APPEND(tmp, pl_params.vertex_attribs,
pl_params.num_vertex_attribs, attrib);
}
}
priv->pl_pass = pl_pass_create(gpu, &pl_params);
if (!priv->pl_pass)
goto error;
pass = talloc_ptrtype(NULL, pass);
*pass = (struct ra_renderpass) {
.params = *ra_renderpass_params_copy(pass, params),
.priv = talloc_steal(pass, priv),
};
pass->params.cached_program = (struct bstr) {
.start = (void *) priv->pl_pass->params.cached_program,
.len = priv->pl_pass->params.cached_program_len,
};
// fall through
error:
talloc_free(tmp);
return pass;
}
static void renderpass_destroy_pl(struct ra *ra, struct ra_renderpass *pass)
{
if (!pass)
return;
struct pass_priv *priv = pass->priv;
pl_pass_destroy(get_gpu(ra), (const struct pl_pass **) &priv->pl_pass);
talloc_free(pass);
}
static void renderpass_run_pl(struct ra *ra,
const struct ra_renderpass_run_params *params)
{
struct pass_priv *p = params->pass->priv;
p->num_varups = 0;
for (int i = 0; i < params->num_values; i++) {
const struct ra_renderpass_input_val *val = &params->values[i];
const struct ra_renderpass_input *inp = &params->pass->params.inputs[i];
if (var_type[inp->type]) {
MP_TARRAY_APPEND(p, p->varups, p->num_varups, (struct pl_var_update) {
.index = p->inp_index[val->index],
.data = val->data,
});
} else {
struct pl_desc_binding bind;
switch (inp->type) {
case RA_VARTYPE_TEX:
vo_gpu: placebo: update for upstream API changes The concept of sample/address modes was moved from `pl_tex` to `pl_desc_binding`. The `pl_tex_blit()` function also underwent an API change.
2020-12-03 07:25:23 +00:00
case RA_VARTYPE_IMG_W: {
struct ra_tex *tex = *((struct ra_tex **) val->data);
bind.object = tex->priv;
#if PL_API_VER >= 103
bind.sample_mode = tex->params.src_linear ? PL_TEX_SAMPLE_LINEAR
: PL_TEX_SAMPLE_NEAREST;
bind.address_mode = tex->params.src_repeat ? PL_TEX_ADDRESS_REPEAT
: PL_TEX_ADDRESS_CLAMP;
#endif
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
break;
vo_gpu: placebo: update for upstream API changes The concept of sample/address modes was moved from `pl_tex` to `pl_desc_binding`. The `pl_tex_blit()` function also underwent an API change.
2020-12-03 07:25:23 +00:00
}
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
case RA_VARTYPE_BUF_RO:
case RA_VARTYPE_BUF_RW:
bind.object = (* (struct ra_buf **) val->data)->priv;
break;
default: abort();
};
p->binds[p->inp_index[val->index]] = bind;
};
}
struct pl_pass_run_params pl_params = {
.pass = p->pl_pass,
.var_updates = p->varups,
.num_var_updates = p->num_varups,
.desc_bindings = p->binds,
.push_constants = params->push_constants,
vo_gpu: ra_pl: add timers support Added in libplacebo v60, unfortunately with some changes in design that make it a bit of an awkward fit for the way timers are used in mpv. Timer queries in libplacebo don't support "start" and "stop"-style operations, and instead are attached directly to operations. The only sane way of implementing this in the ra API is to have a single 'active timer' that gets attached to every pass, taking care to sort distinct operations into distinct pl_timer queries within that ra_timer. This design unfortunately doesn't let us have multiple 'active timers' concurrently, similar to the current such limitation in ra_gl. But it's also not a big deal.
2020-05-16 04:10:50 +00:00
#if PL_API_VER >= 60
.timer = get_active_timer(ra),
#endif
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
};
if (p->pl_pass->params.type == PL_PASS_RASTER) {
pl_params.target = params->target->priv;
pl_params.viewport = mp_rect2d_to_pl(params->viewport);
pl_params.scissors = mp_rect2d_to_pl(params->scissors);
pl_params.vertex_data = params->vertex_data;
pl_params.vertex_count = params->vertex_count;
} else {
for (int i = 0; i < MP_ARRAY_SIZE(pl_params.compute_groups); i++)
pl_params.compute_groups[i] = params->compute_groups[i];
}
pl_pass_run(get_gpu(ra), &pl_params);
}
vo_gpu: ra_pl: add timers support Added in libplacebo v60, unfortunately with some changes in design that make it a bit of an awkward fit for the way timers are used in mpv. Timer queries in libplacebo don't support "start" and "stop"-style operations, and instead are attached directly to operations. The only sane way of implementing this in the ra API is to have a single 'active timer' that gets attached to every pass, taking care to sort distinct operations into distinct pl_timer queries within that ra_timer. This design unfortunately doesn't let us have multiple 'active timers' concurrently, similar to the current such limitation in ra_gl. But it's also not a big deal.
2020-05-16 04:10:50 +00:00
struct ra_timer_pl {
// Because libpplacebo only supports one operation per timer, we need
// to use multiple pl_timers to sum up multiple passes/transfers
struct pl_timer **timers;
int num_timers;
int idx_timers;
};
static ra_timer *timer_create_pl(struct ra *ra)
{
struct ra_timer_pl *t = talloc_zero(ra, struct ra_timer_pl);
return t;
}
static void timer_destroy_pl(struct ra *ra, ra_timer *timer)
{
const struct pl_gpu *gpu = get_gpu(ra);
struct ra_timer_pl *t = timer;
for (int i = 0; i < t->num_timers; i++)
pl_timer_destroy(gpu, &t->timers[i]);
talloc_free(t);
}
static void timer_start_pl(struct ra *ra, ra_timer *timer)
{
struct ra_pl *p = ra->priv;
struct ra_timer_pl *t = timer;
// There's nothing easy we can do in this case, since libplacebo only
// supports one timer object per operation; so just ignore "inner" timers
// when the user is nesting different timer queries
if (p->active_timer)
return;
p->active_timer = t;
t->idx_timers = 0;
}
static uint64_t timer_stop_pl(struct ra *ra, ra_timer *timer)
{
struct ra_pl *p = ra->priv;
struct ra_timer_pl *t = timer;
if (p->active_timer != t)
return 0;
p->active_timer = NULL;
// Sum up all of the active results
uint64_t res = 0;
for (int i = 0; i < t->idx_timers; i++)
res += pl_timer_query(p->gpu, t->timers[i]);
return res;
}
static struct pl_timer *get_active_timer(const struct ra *ra)
{
struct ra_pl *p = ra->priv;
if (!p->active_timer)
return NULL;
struct ra_timer_pl *t = p->active_timer;
if (t->idx_timers == t->num_timers)
MP_TARRAY_APPEND(t, t->timers, t->num_timers, pl_timer_create(p->gpu));
return t->timers[t->idx_timers++];
}
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
static struct ra_fns ra_fns_pl = {
.destroy = destroy_ra_pl,
.tex_create = tex_create_pl,
.tex_destroy = tex_destroy_pl,
.tex_upload = tex_upload_pl,
.tex_download = tex_download_pl,
.buf_create = buf_create_pl,
.buf_destroy = buf_destroy_pl,
.buf_update = buf_update_pl,
.buf_poll = buf_poll_pl,
.clear = clear_pl,
.blit = blit_pl,
.uniform_layout = uniform_layout_pl,
.push_constant_layout = push_constant_layout_pl,
.desc_namespace = desc_namespace_pl,
.renderpass_create = renderpass_create_pl,
.renderpass_destroy = renderpass_destroy_pl,
.renderpass_run = renderpass_run_pl,
vo_gpu: ra_pl: add timers support Added in libplacebo v60, unfortunately with some changes in design that make it a bit of an awkward fit for the way timers are used in mpv. Timer queries in libplacebo don't support "start" and "stop"-style operations, and instead are attached directly to operations. The only sane way of implementing this in the ra API is to have a single 'active timer' that gets attached to every pass, taking care to sort distinct operations into distinct pl_timer queries within that ra_timer. This design unfortunately doesn't let us have multiple 'active timers' concurrently, similar to the current such limitation in ra_gl. But it's also not a big deal.
2020-05-16 04:10:50 +00:00
.timer_create = timer_create_pl,
.timer_destroy = timer_destroy_pl,
.timer_start = timer_start_pl,
.timer_stop = timer_stop_pl,
vo_gpu: vulkan: use libplacebo instead This commit rips out the entire mpv vulkan implementation in favor of exposing lightweight wrappers on top of libplacebo instead, which provides much of the same except in a more up-to-date and polished form. This (finally) unifies the code base between mpv and libplacebo, which is something I've been hoping to do for a long time. Note: The ra_pl wrappers are abstract enough from the actual libplacebo device type that we can in theory re-use them for other devices like d3d11 or even opengl in the future, so I moved them to a separate directory for the time being. However, the rest of the code is still vulkan-specific, so I've kept the "vulkan" naming and file paths, rather than introducing a new `--gpu-api` type. (Which would have been ended up with significantly more code duplicaiton) Plus, the code and functionality is similar enough that for most users this should just be a straight-up drop-in replacement. Note: This commit excludes some changes; specifically, the updates to context_win and hwdec_cuda are deferred to separate commits for authorship reasons.
2018-11-10 11:53:33 +00:00
};