131 lines
3.7 KiB
C++
131 lines
3.7 KiB
C++
#include <emscripten/bind.h>
|
|
#include <emscripten/val.h>
|
|
|
|
#include "lib/jxl/base/thread_pool_internal.h"
|
|
#include "lib/jxl/enc_file.h"
|
|
#include "lib/jxl/external_image.h"
|
|
|
|
using namespace emscripten;
|
|
|
|
thread_local const val Uint8Array = val::global("Uint8Array");
|
|
|
|
struct JXLOptions {
|
|
// 1 = slowest
|
|
// 7 = fastest
|
|
int speed;
|
|
float quality;
|
|
bool progressive;
|
|
int epf;
|
|
int nearLossless;
|
|
bool lossyPalette;
|
|
};
|
|
|
|
val encode(std::string image, int width, int height, JXLOptions options) {
|
|
jxl::CompressParams cparams;
|
|
jxl::PassesEncoderState passes_enc_state;
|
|
jxl::CodecInOut io;
|
|
jxl::PaddedBytes bytes;
|
|
jxl::ImageBundle* main = &io.Main();
|
|
jxl::ThreadPoolInternal* pool_ptr = nullptr;
|
|
#ifdef __EMSCRIPTEN_PTHREADS__
|
|
jxl::ThreadPoolInternal pool;
|
|
pool_ptr = &pool;
|
|
#endif
|
|
|
|
cparams.epf = options.epf;
|
|
cparams.speed_tier = static_cast<jxl::SpeedTier>(options.speed);
|
|
cparams.near_lossless = options.nearLossless;
|
|
|
|
if (options.lossyPalette) {
|
|
cparams.lossy_palette = true;
|
|
cparams.palette_colors = 0;
|
|
cparams.options.predictor = jxl::Predictor::Zero;
|
|
}
|
|
|
|
// Reduce memory usage of tree learning for lossless data.
|
|
// TODO(veluca93): this is a mitigation for excessive memory usage in the JXL encoder.
|
|
float megapixels = width * height * 0.000001;
|
|
if (megapixels > 8) {
|
|
cparams.options.nb_repeats = 0.1;
|
|
} else if (megapixels > 4) {
|
|
cparams.options.nb_repeats = 0.3;
|
|
} else {
|
|
// default is OK.
|
|
}
|
|
|
|
float quality = options.quality;
|
|
|
|
// Quality settings roughly match libjpeg qualities.
|
|
if (quality < 7 || quality == 100) {
|
|
cparams.modular_mode = true;
|
|
// Internal modular quality to roughly match VarDCT size.
|
|
cparams.quality_pair.first = cparams.quality_pair.second =
|
|
std::min(35 + (quality - 7) * 3.0f, 100.0f);
|
|
} else {
|
|
cparams.modular_mode = false;
|
|
if (quality >= 30) {
|
|
cparams.butteraugli_distance = 0.1 + (100 - quality) * 0.09;
|
|
} else {
|
|
cparams.butteraugli_distance = 6.4 + pow(2.5, (30 - quality) / 5.0f) / 6.25f;
|
|
}
|
|
}
|
|
|
|
if (options.progressive) {
|
|
cparams.qprogressive_mode = true;
|
|
cparams.responsive = 1;
|
|
if (!cparams.modular_mode) {
|
|
cparams.progressive_dc = 1;
|
|
}
|
|
}
|
|
|
|
if (cparams.modular_mode) {
|
|
if (cparams.quality_pair.first != 100 || cparams.quality_pair.second != 100) {
|
|
cparams.color_transform = jxl::ColorTransform::kXYB;
|
|
} else {
|
|
cparams.color_transform = jxl::ColorTransform::kNone;
|
|
}
|
|
}
|
|
|
|
if (cparams.near_lossless) {
|
|
// Near-lossless assumes -R 0
|
|
cparams.responsive = 0;
|
|
cparams.modular_mode = true;
|
|
}
|
|
|
|
io.metadata.m.SetAlphaBits(8);
|
|
if (!io.metadata.size.Set(width, height)) {
|
|
return val::null();
|
|
}
|
|
|
|
uint8_t* inBuffer = (uint8_t*)image.c_str();
|
|
|
|
auto result = jxl::ConvertImage(
|
|
jxl::Span<const uint8_t>(reinterpret_cast<const uint8_t*>(image.data()), image.size()), width,
|
|
height, jxl::ColorEncoding::SRGB(/*is_gray=*/false), /*has_alpha=*/true,
|
|
/*alpha_is_premultiplied=*/false, /*bits_per_alpha=*/8, /*bits_per_sample=*/8,
|
|
/*big_endian=*/false, /*flipped_y=*/false, pool_ptr, main);
|
|
|
|
if (!result) {
|
|
return val::null();
|
|
}
|
|
|
|
auto js_result = val::null();
|
|
if (EncodeFile(cparams, &io, &passes_enc_state, &bytes, /*aux=*/nullptr, pool_ptr)) {
|
|
js_result = Uint8Array.new_(typed_memory_view(bytes.size(), bytes.data()));
|
|
}
|
|
|
|
return js_result;
|
|
}
|
|
|
|
EMSCRIPTEN_BINDINGS(my_module) {
|
|
value_object<JXLOptions>("JXLOptions")
|
|
.field("speed", &JXLOptions::speed)
|
|
.field("quality", &JXLOptions::quality)
|
|
.field("progressive", &JXLOptions::progressive)
|
|
.field("nearLossless", &JXLOptions::nearLossless)
|
|
.field("lossyPalette", &JXLOptions::lossyPalette)
|
|
.field("epf", &JXLOptions::epf);
|
|
|
|
function("encode", &encode);
|
|
}
|