Skip to content

feat: decklink support for chosing yuv output i 8-bit mode #1652

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

feat: decklink support for chosing yuv output i 8-bit mode #1652

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions src/common/memory.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -658,6 +658,11 @@ shared_ptr<T> make_shared(P0&& p0)
{
return shared_ptr<T>(std::make_shared<U>(std::forward<P0>(p0)));
}
template <typename T, typename U, typename P0, typename P1>
shared_ptr<T> make_shared(P0&& p0, P1&& p1)
{
return shared_ptr<T>(std::make_shared<U>(std::forward<P0>(p0), std::forward<P1>(p1)));
}

template <typename T, typename P0>
shared_ptr<T> make_shared(P0&& p0)
Expand Down
2 changes: 1 addition & 1 deletion src/modules/decklink/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,7 @@ project (decklink)
set(SOURCES
consumer/decklink_consumer.cpp
consumer/decklink_consumer.h
consumer/hdr_v210_strategy.cpp
consumer/v210_strategies.cpp
consumer/sdr_bgra_strategy.cpp
consumer/format_strategy.h
consumer/config.cpp
Expand Down
19 changes: 19 additions & 0 deletions src/modules/decklink/consumer/config.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -97,6 +97,25 @@ configuration parse_xml_config(const boost::property_tree::wptree& ptree,
}
config.wait_for_reference_duration = ptree.get(L"wait-for-reference-duration", config.wait_for_reference_duration);

{
auto is_8bit = channel_info.depth == common::bit_depth::bit8;
auto default_pixel_format = is_8bit ? L"rgba" : L"yuv";
auto pixel_format = ptree.get(L"pixel-format", default_pixel_format);
if (pixel_format == L"yuv") {
config.pixel_format = configuration::pixel_format_t::yuv;
} else if (pixel_format == L"rgba") {
config.pixel_format = configuration::pixel_format_t::rgba;
} else {
CASPAR_THROW_EXCEPTION(user_error() << msg_info(L"Invalid pixel format, must be rgba or yuv"));
}

if (channel_info.depth != common::bit_depth::bit8 &&
config.pixel_format == configuration::pixel_format_t::rgba) {
CASPAR_THROW_EXCEPTION(user_error()
<< msg_info(L"The decklink consumer only supports rgba output on 8-bit channels"));
}
}

config.primary = parse_output_config(ptree, format_repository);
if (config.primary.device_index == -1)
config.primary.device_index = 1;
Expand Down
7 changes: 7 additions & 0 deletions src/modules/decklink/consumer/config.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -87,6 +87,12 @@ struct configuration
disabled,
};

enum class pixel_format_t
{
rgba,
yuv,
};

bool embedded_audio = false;
keyer_t keyer = keyer_t::default_keyer;
duplex_t duplex = duplex_t::default_duplex;
Expand All @@ -95,6 +101,7 @@ struct configuration
int wait_for_reference_duration = 10; // seconds
int base_buffer_depth = 3;
bool hdr = false;
pixel_format_t pixel_format = pixel_format_t::rgba;

port_configuration primary;
std::vector<port_configuration> secondaries;
Expand Down
11 changes: 7 additions & 4 deletions src/modules/decklink/consumer/decklink_consumer.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -197,7 +197,12 @@ core::video_format_desc get_decklink_format(const port_configuration& confi

spl::shared_ptr<format_strategy> create_format_strategy(const configuration& config)
{
return config.hdr ? create_hdr_v210_strategy(config.color_space) : create_sdr_bgra_strategy();
if (config.hdr) {
return create_hdr_v210_strategy(config.color_space);
} else {
return config.pixel_format == configuration::pixel_format_t::yuv ? create_sdr_v210_strategy(config.color_space)
: create_sdr_bgra_strategy();
}
}

enum EOTF
Expand Down Expand Up @@ -1006,9 +1011,7 @@ struct decklink_consumer final : public IDeckLinkVideoOutputCallback
audio_scheduled_ += nb_samples; // TODO - what if there are too many/few samples in this frame?
}

void schedule_next_video(std::shared_ptr<void> image_data,
int nb_samples,
BMDTimeValue display_time)
void schedule_next_video(std::shared_ptr<void> image_data, int nb_samples, BMDTimeValue display_time)
{
auto fmt = format_strategy_->get_pixel_format();
auto row_bytes = format_strategy_->get_row_bytes(decklink_format_desc_.width);
Expand Down
3 changes: 2 additions & 1 deletion src/modules/decklink/consumer/format_strategy.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,7 +41,7 @@ class format_strategy

public:
format_strategy& operator=(const format_strategy&) = delete;
virtual ~format_strategy() = default;
virtual ~format_strategy() = default;

format_strategy(const format_strategy&) = delete;

Expand All @@ -57,6 +57,7 @@ class format_strategy
};

spl::shared_ptr<format_strategy> create_sdr_bgra_strategy();
spl::shared_ptr<format_strategy> create_sdr_v210_strategy(core::color_space colorspace);
spl::shared_ptr<format_strategy> create_hdr_v210_strategy(core::color_space colorspace);

}} // namespace caspar::decklink
107 changes: 73 additions & 34 deletions ...s/decklink/consumer/hdr_v210_strategy.cpp → ...les/decklink/consumer/v210_strategies.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,8 @@
#include <simde/x86/avx2.h>
#endif

#include <type_traits>

#include "format_strategy.h"

#include <common/memshfl.h>
Expand Down Expand Up @@ -153,15 +155,17 @@ inline void pack_v210_avx2(__m256i luma[6], __m256i chroma[6], __m128i** v210_de
}
}

template <typename T = uint16_t>
struct ARGBPixel
{
uint16_t R;
uint16_t G;
uint16_t B;
uint16_t A;
T R;
T G;
T B;
T A;
};

void pack_v210(const ARGBPixel* src, const std::vector<int32_t>& color_matrix, uint32_t* dest, int num_pixels)
template <typename T>
void pack_v210(const ARGBPixel<T>* src, const std::vector<int32_t>& color_matrix, uint32_t* dest, int num_pixels)
{
auto write_v210 = [dest, index = 0, shift = 0](uint32_t val) mutable {
dest[index] |= ((val & 0x3FF) << shift);
Expand Down Expand Up @@ -202,9 +206,9 @@ void pack_v210(const ARGBPixel* src, const std::vector<int32_t>& color_matrix, u
}
}

class hdr_v210_strategy
class v210_strategy
: public format_strategy
, std::enable_shared_from_this<hdr_v210_strategy>
, std::enable_shared_from_this<v210_strategy>
{
std::vector<float> bt709{0.212639005871510,
0.715168678767756,
Expand All @@ -227,13 +231,15 @@ class hdr_v210_strategy

std::vector<int32_t> color_matrix;
__m128i black_batch;
uint8_t bpc;

public:
explicit hdr_v210_strategy(core::color_space color_space)
explicit v210_strategy(core::color_space color_space, uint8_t bpc)
: color_matrix(create_int_matrix(color_space == core::color_space::bt2020 ? bt2020 : bt709))
, bpc(bpc)
{
// setup black batch (6 pixels of black, encoded as v210)
ARGBPixel black[6];
ARGBPixel<> black[6];
memset(black, 0, sizeof(black));
memset(&black_batch, 0, sizeof(__m128i));
pack_v210(black, color_matrix, reinterpret_cast<uint32_t*>(&black_batch), 6);
Expand All @@ -248,33 +254,47 @@ class hdr_v210_strategy
auto size = get_row_bytes(format_desc.width) * format_desc.height;
return create_aligned_buffer(size, 128);
}

std::shared_ptr<void> convert_frame_for_port(const core::video_format_desc& channel_format_desc,
const core::video_format_desc& decklink_format_desc,
const port_configuration& config,
const core::const_frame& frame1,
const core::const_frame& frame2,
BMDFieldDominance field_dominance) override
{
return bpc == 1 ? do_convert_frame_for_port<uint8_t>(
channel_format_desc, decklink_format_desc, config, frame1, frame2, field_dominance)
: do_convert_frame_for_port<uint16_t>(
channel_format_desc, decklink_format_desc, config, frame1, frame2, field_dominance);
}

private:
template <typename T>
std::shared_ptr<void> do_convert_frame_for_port(const core::video_format_desc& channel_format_desc,
const core::video_format_desc& decklink_format_desc,
const port_configuration& config,
const core::const_frame& frame1,
const core::const_frame& frame2,
BMDFieldDominance field_dominance)
{
std::shared_ptr<void> image_data = allocate_frame_data(decklink_format_desc);

if (field_dominance != bmdProgressiveFrame) {
convert_frame(channel_format_desc,
decklink_format_desc,
config,
image_data,
field_dominance == bmdUpperFieldFirst,
frame1);

convert_frame(channel_format_desc,
decklink_format_desc,
config,
image_data,
field_dominance != bmdUpperFieldFirst,
frame2);
convert_frame<T>(channel_format_desc,
decklink_format_desc,
config,
image_data,
field_dominance == bmdUpperFieldFirst,
frame1);

convert_frame<T>(channel_format_desc,
decklink_format_desc,
config,
image_data,
field_dominance != bmdUpperFieldFirst,
frame2);

} else {
convert_frame(channel_format_desc, decklink_format_desc, config, image_data, true, frame1);
convert_frame<T>(channel_format_desc, decklink_format_desc, config, image_data, true, frame1);
}

if (config.key_only) {
Expand All @@ -284,7 +304,7 @@ class hdr_v210_strategy
return image_data;
}

private:
template <typename T>
void convert_frame(const core::video_format_desc& channel_format_desc,
const core::video_format_desc& decklink_format_desc,
const port_configuration& config,
Expand Down Expand Up @@ -337,7 +357,7 @@ class hdr_v210_strategy
continue;
}

auto src = reinterpret_cast<const ARGBPixel*>(frame.image_data(0).data()) +
auto src = reinterpret_cast<const ARGBPixel<T>*>(frame.image_data(0).data()) +
(src_y * channel_format_desc.width + config.src_x);
// Pack pixels in batches of 48
for (int batch_index_x = 0; batch_index_x < fullspeed_x_batches; batch_index_x++) {
Expand All @@ -348,12 +368,26 @@ class hdr_v210_strategy

__m256i zero = _mm256_setzero_si256();
for (int packet_index = 0; packet_index < 6; packet_index++) {
__m256i p0123 = _mm256_loadu_si256(pixeldata + packet_index * 2);
__m256i p4567 = _mm256_loadu_si256(pixeldata + packet_index * 2 + 1);

// shift down to 10 bit precision
p0123 = _mm256_srli_epi16(p0123, 6);
p4567 = _mm256_srli_epi16(p4567, 6);
__m256i p0123, p4567;
if constexpr (std::is_same<T, uint16_t>()) {
p0123 = _mm256_loadu_si256(pixeldata + packet_index * 2);
p4567 = _mm256_loadu_si256(pixeldata + packet_index * 2 + 1);

// shift down to 10 bit precision
p0123 = _mm256_srli_epi16(p0123, 6);
p4567 = _mm256_srli_epi16(p4567, 6);
} else if constexpr (std::is_same<T, uint8_t>()) {
auto p01234567 = _mm256_loadu_si256(pixeldata + packet_index);
auto p01452367 = _mm256_permute4x64_epi64(p01234567, 0b11011000);

p0123 = _mm256_unpacklo_epi8(p01452367, zero);
p4567 = _mm256_unpackhi_epi8(p01452367, zero);

p0123 = _mm256_slli_epi16(p0123, 2);
p4567 = _mm256_slli_epi16(p4567, 2);
} else {
static_assert(!std::is_same<T, T>(), "Unsupported template type for v210 conversion");
}

// unpack 16 bit values to 32 bit registers, padding with zeros
__m256i pixel_pairs[4];
Expand Down Expand Up @@ -388,7 +422,7 @@ class hdr_v210_strategy
// // pad the rest with black pixels to fill the v210 packet
int last_x_pixels = rest_x_pixels - rest_x_6pixels;
if (last_x_pixels > 0) {
ARGBPixel pixels[6];
ARGBPixel<T> pixels[6];
memset(pixels, 0, sizeof(pixels));
memcpy(pixels, src, last_x_pixels * 8);
pack_v210(pixels, color_matrix, reinterpret_cast<uint32_t*>(v210_dest), 6);
Expand All @@ -410,9 +444,14 @@ class hdr_v210_strategy
}
};

spl::shared_ptr<format_strategy> create_sdr_v210_strategy(core::color_space color_space)
{
return spl::make_shared<format_strategy, v210_strategy>(color_space, 1);
}

spl::shared_ptr<format_strategy> create_hdr_v210_strategy(core::color_space color_space)
{
return spl::make_shared<format_strategy, hdr_v210_strategy>(color_space);
return spl::make_shared<format_strategy, v210_strategy>(color_space, 2);
}

}} // namespace caspar::decklink
1 change: 1 addition & 0 deletions src/shell/casparcg.config
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,6 +90,7 @@
<keyer>external [external|external_separate_device|internal|default]</keyer>
<key-only>false [true|false]</key-only>
<buffer-depth>3 [1..]</buffer-depth>
<pixel-format>[yuv|rgba](default is rgba for 8bit channels, yuv for high bit-depth channels)</pixel-format>
<video-mode>(Run the decklink at a different video-mode. Note: the framerate must match that of the channel)</video-mode>
<subregion>
<src-x>0 (x offset into the channel)</src-x>
Expand Down