Skip to content

Commit

Permalink
Shader: improve the speed (based on Lypheo/vs-placebo@09075cf)
Browse files Browse the repository at this point in the history
  • Loading branch information
@Asd-g
Asd-g committed Mar 9, 2023
1 parent 072d94d commit 4bbf726
Showing 1 changed file with 49 additions and 50 deletions.
99 changes: 49 additions & 50 deletions src/shader.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,6 @@
#endif

#include "avs_libplacebo.h"
#include "p2p_api.h"

static std::mutex mtx;

Expand All @@ -25,7 +24,6 @@ struct shader
int subw;
int subh;
std::string msg;
void* packed_dst;
};

static bool shader_do_plane(priv& p, shader& data, const pl_plane& planes_) noexcept
Expand Down Expand Up @@ -55,17 +53,15 @@ static bool shader_do_plane(priv& p, shader& data, const pl_plane& planes_) noex
pl_frame_set_chroma_location(&img, d->chromaLocation);

pl_frame out{};
out.num_planes = 1;
out.num_planes = 3;
out.repr = crpr;
out.color = csp;

for (int i{ 0 }; i < 3; ++i)
{
out.planes[i].texture = p.tex_out[0];
out.planes[i].components = p.tex_out[0]->params.format->num_components;
out.planes[i].component_mapping[0] = 0;
out.planes[i].component_mapping[1] = 1;
out.planes[i].component_mapping[2] = 2;
out.planes[i].texture = p.tex_out[i];
out.planes[i].components = 1;
out.planes[i].component_mapping[0] = i;
}

pl_render_params renderParams{};
Expand Down Expand Up @@ -98,8 +94,8 @@ static int shader_reconfig(priv& priv_, const pl_plane_data& data_, shader& d, c
for (int i{ 0 }; i < 3; ++i)
{
pl_tex_params t_r{};
t_r.w = data->width;
t_r.h = data->height;
t_r.w = data[i].width;
t_r.h = data[i].height;
t_r.format = fmt[i];
t_r.sampleable = true;
t_r.host_writable = true;
Expand All @@ -108,33 +104,32 @@ static int shader_reconfig(priv& priv_, const pl_plane_data& data_, shader& d, c
return -2;
}

pl_plane_data plane_data{};
plane_data.type = PL_FMT_UNORM;
plane_data.pixel_stride = 6;

for (int i{ 0 }; i < 3; ++i)
{
plane_data.component_map[i] = i;
plane_data.component_pad[i] = 0;
plane_data.component_size[i] = 16;
pl_plane_data plane_data{};
plane_data.type = PL_FMT_UNORM;
plane_data.component_map[0] = i;
plane_data.component_pad[0] = 0;
plane_data.component_size[0] = 16;
plane_data.pixel_stride = 2;

const pl_fmt out{ pl_plane_find_fmt(p->gpu, nullptr, &plane_data) };

pl_tex_params t_r1{};
t_r1.w = w;
t_r1.h = h;
t_r1.format = out;
t_r1.renderable = true;
t_r1.host_readable = true;

if (!pl_tex_recreate(p->gpu, &p->tex_out[i], &t_r1))
return -2;
}

const pl_fmt out{ pl_plane_find_fmt(p->gpu, nullptr, &plane_data) };

pl_tex_params t_r1{};
t_r1.w = w;
t_r1.h = h;
t_r1.format = out;
t_r1.renderable = true;
t_r1.host_readable = true;

if (!pl_tex_recreate(p->gpu, &p->tex_out[0], &t_r1))
return -2;

return 0;
}

static int shader_filter(priv& priv_, void* dst, const pl_plane_data& src_, shader& d)
static int shader_filter(priv& priv_, pl_buf* dst, const pl_plane_data& src_, shader& d, const int dst_stride)
{
priv* p{ &priv_ };
const pl_plane_data* src{ &src_ };
Expand All @@ -153,12 +148,16 @@ static int shader_filter(priv& priv_, void* dst, const pl_plane_data& src_, shad
return -2;

// Download planes
pl_tex_transfer_params ttr1{};
ttr1.tex = p->tex_out[0];
ttr1.ptr = dst;
for (int i{ 0 }; i < 3; ++i)
{
pl_tex_transfer_params ttr1{};
ttr1.tex = p->tex_out[i];
ttr1.row_pitch = dst_stride;
ttr1.buf = dst[i];

if (!pl_tex_download(p->gpu, &ttr1))
return -3;
if (!pl_tex_download(p->gpu, &ttr1))
return -3;
}

return 0;
}
Expand Down Expand Up @@ -187,7 +186,12 @@ static AVS_VideoFrame* AVSC_CC shader_get_frame(AVS_FilterInfo* fi, int n)
}

constexpr int planes[3]{ AVS_PLANAR_Y, AVS_PLANAR_U, AVS_PLANAR_V };
const int dst_stride = ((avs_get_pitch(dst)) + (d->vf->gpu->limits.align_tex_xfer_pitch) - 1) & ~((d->vf->gpu->limits.align_tex_xfer_pitch) - 1);
pl_plane_data pl[3]{};
pl_buf dst_buf[3]{};
pl_buf_params buf_params{};
buf_params.size = dst_stride * fi->vi.height;
buf_params.host_mapped = true;

for (int i{ 0 }; i < 3; ++i)
{
Expand All @@ -200,6 +204,8 @@ static AVS_VideoFrame* AVSC_CC shader_get_frame(AVS_FilterInfo* fi, int n)
pl[i].component_size[0] = 16;
pl[i].component_pad[0] = 0;
pl[i].component_map[0] = i;

dst_buf[i] = pl_buf_create(d->vf->gpu, &buf_params);
}

{
Expand All @@ -208,7 +214,7 @@ static AVS_VideoFrame* AVSC_CC shader_get_frame(AVS_FilterInfo* fi, int n)
const int reconf{ shader_reconfig(*d->vf.get(), *pl, *d, fi->vi.width, fi->vi.height) };
if (reconf == 0)
{
const int filt{ shader_filter(*d->vf.get(), d->packed_dst, *pl, *d) };
const int filt{ shader_filter(*d->vf.get(), dst_buf, *pl, *d, dst_stride) };

if (filt)
{
Expand All @@ -235,27 +241,22 @@ static AVS_VideoFrame* AVSC_CC shader_get_frame(AVS_FilterInfo* fi, int n)
avs_release_video_frame(src);
avs_release_video_frame(dst);

for (int i{ 0 }; i < 3; ++i)
pl_buf_destroy(d->vf->gpu, &dst_buf[i]);

fi->error = ErrorText;

return nullptr;
}
else
{
p2p_buffer_param pack_params{};
pack_params.width = fi->vi.width;
pack_params.height = fi->vi.height;
pack_params.packing = p2p_bgr48_le;
pack_params.src[0] = d->packed_dst;
pack_params.src_stride[0] = fi->vi.width * 2 * 3;

for (int k = 0; k < 3; ++k)
for (int i{ 0 }; i < 3; ++i)
{
pack_params.dst[k] = avs_get_write_ptr_p(dst, planes[k]);
pack_params.dst_stride[k] = avs_get_pitch_p(dst, planes[k]);
while (pl_buf_poll(d->vf->gpu, dst_buf[i], 0));
memcpy(avs_get_write_ptr_p(dst, planes[i]), dst_buf[i]->data, dst_buf[i]->params.size);
pl_buf_destroy(d->vf->gpu, &dst_buf[i]);
}

p2p_unpack_frame(&pack_params, 0);

avs_release_video_frame(src);

return dst;
Expand All @@ -266,7 +267,6 @@ static void AVSC_CC free_shader(AVS_FilterInfo* fi)
{
shader* d{ reinterpret_cast<shader*>(fi->user_data) };

operator delete (d->packed_dst);
pl_mpv_user_shader_destroy(&d->shader);
avs_libplacebo_uninit(std::move(d->vf));
delete d;
Expand Down Expand Up @@ -460,7 +460,6 @@ AVS_Value AVSC_CC create_shader(AVS_ScriptEnvironment* env, AVS_Value args, void
params->subw = avs_get_plane_width_subsampling(&fi->vi, AVS_PLANAR_U);
params->subh = avs_get_plane_height_subsampling(&fi->vi, AVS_PLANAR_U);

params->packed_dst = operator new(fi->vi.width * fi->vi.height * 2 * 3);
fi->vi.pixel_type = AVS_CS_YUV444P16;

AVS_Value v{ avs_new_value_clip(clip) };
Expand Down

0 comments on commit 4bbf726

Please sign in to comment.