OpenSourceJoker/video2x
3
0
Fork 0
mirror of https://github.com/k4yt3x/video2x.git synced 2025-01-15 04:08:52 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

perf(*): change multiversioning target clones to x86-64-v3 and x86-64-v4

Browse Source 
Signed-off-by: k4yt3x <i@k4yt3x.com>
This commit is contained in:
k4yt3x 2024-12-30 00:00:00 +00:00
parent 7665cd217c
commit a72c094588
No known key found for this signature in database
5 changed files with 12 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
CMakeLists.txt
View File

@ -1,4 +1,4 @@
cmake_minimum_required(VERSION 3.10)
cmake_minimum_required(VERSION 3.10 FATAL_ERROR)
project(video2x VERSION 6.3.1 LANGUAGES CXX)
# The FindBoost module is removed in CMake 3.30
@ -346,10 +346,10 @@ target_include_directories(libvideo2x PRIVATE
)
# Set the compile options for the shared library
if(MSVC)
target_compile_options(libvideo2x PRIVATE $<$<CONFIG:Debug>:/Zi /DDEBUG>)
else()
target_compile_options(libvideo2x PRIVATE -fPIC $<$<CONFIG:Debug>:-g -DDEBUG>)
if(CMAKE_CXX_COMPILER_ID STREQUAL "MSVC")
target_compile_options(libvideo2x PRIVATE $<$<CONFIG:Debug>:/Zi /Od /MDd>)
elseif(CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang")
target_compile_options(libvideo2x PRIVATE -fPIC $<$<CONFIG:Debug>:-g -O0>)
endif()
# Define the paths to the shared libraries

2
src/avutils.cpp
View File

@ -135,7 +135,7 @@ AVPixelFormat get_encoder_default_pix_fmt(const AVCodec *encoder, AVPixelFormat
return best_pix_fmt;
}
[[gnu::target_clones("default", "avx2", "avx512f")]]
[[gnu::target_clones("arch=x86-64-v4", "arch=x86-64-v3", "default")]]
float get_frame_diff(AVFrame *frame1, AVFrame *frame2) {
if (!frame1 || !frame2) {
logger()->error("Invalid frame(s) provided for comparison");

6
src/conversions.cpp
View File

@ -11,7 +11,7 @@ namespace video2x {
namespace conversions {
// Convert AVFrame format
[[gnu::target_clones("default", "avx2", "avx512f")]]
[[gnu::target_clones("arch=x86-64-v4", "arch=x86-64-v3", "default")]]
AVFrame *convert_avframe_pix_fmt(AVFrame *src_frame, AVPixelFormat pix_fmt) {
AVFrame *dst_frame = av_frame_alloc();
if (dst_frame == nullptr) {
@ -68,7 +68,7 @@ AVFrame *convert_avframe_pix_fmt(AVFrame *src_frame, AVPixelFormat pix_fmt) {
}
// Convert AVFrame to ncnn::Mat by copying the data
[[gnu::target_clones("default", "avx2", "avx512f")]]
[[gnu::target_clones("arch=x86-64-v4", "arch=x86-64-v3", "default")]]
ncnn::Mat avframe_to_ncnn_mat(AVFrame *frame) {
AVFrame *converted_frame = nullptr;
@ -108,7 +108,7 @@ ncnn::Mat avframe_to_ncnn_mat(AVFrame *frame) {
}
// Convert ncnn::Mat to AVFrame with a specified pixel format (this part is unchanged)
[[gnu::target_clones("default", "avx2", "avx512f")]]
[[gnu::target_clones("arch=x86-64-v4", "arch=x86-64-v3", "default")]]
AVFrame *ncnn_mat_to_avframe(const ncnn::Mat &mat, AVPixelFormat pix_fmt) {
int ret;

4
src/encoder.cpp
View File

@ -254,7 +254,7 @@ int Encoder::init(
return 0;
}
[[gnu::target_clones("default", "avx2", "avx512f")]]
[[gnu::target_clones("arch=x86-64-v4", "arch=x86-64-v3", "default")]]
int Encoder::write_frame(AVFrame *frame, int64_t frame_idx) {
AVFrame *converted_frame = nullptr;
int ret;
@ -326,7 +326,7 @@ int Encoder::write_frame(AVFrame *frame, int64_t frame_idx) {
return 0;
}
[[gnu::target_clones("default", "avx2", "avx512f")]]
[[gnu::target_clones("arch=x86-64-v4", "arch=x86-64-v3", "default")]]
int Encoder::flush() {
int ret;
AVPacket *enc_pkt = av_packet_alloc();

2
src/libvideo2x.cpp
View File

@ -28,7 +28,7 @@ VideoProcessor::VideoProcessor(
hw_device_type_(hw_device_type),
benchmark_(benchmark) {}
[[gnu::target_clones("default", "avx2", "avx512f")]]
[[gnu::target_clones("arch=x86-64-v4", "arch=x86-64-v3", "default")]]
int VideoProcessor::process(
const std::filesystem::path in_fname,
const std::filesystem::path out_fname