AI/rocm-systems
2
0
Förgrening 0
Kod Ärenden Pull-förfrågningar Actions Packages Projekt Släpp Wiki Aktiviteter

Added first half of ROI support (#41)

Bläddra i källkod 
* Finished first half of adding ROI functionality to rocJPEG

* took out unnecessary .h file

* addressed review comments from Aryan

* fixed typo

* fixed documentation in code for decode_params

* fixed the spaces
This commit is contained in:
Pavel Tcherniaev
2024-07-12 12:22:49 -07:00
committad av GitHub
förälder 3f0be8387e
incheckning 6ce373b2a1
4 ändrade filer med 54 tillägg och 33 borttagningar
Download Patch File Download Diff File Expand all files Collapse all files
samples/jpegDecode/jpegdecode.cpp
+1 -1
Visa fil
@@ -100,7 +100,7 @@ int main(int argc, char **argv) {
return EXIT_FAILURE;
}
if (rocjpeg_utils.GetChannelPitchAndSizes(decode_params.output_format, subsampling, widths, heights, num_channels, output_image, channel_sizes)) {
if (rocjpeg_utils.GetChannelPitchAndSizes(decode_params, subsampling, widths, heights, num_channels, output_image, channel_sizes)) {
std::cerr << "ERROR: Failed to get the channel pitch and sizes" << std::endl;
return EXIT_FAILURE;
}
samples/jpegDecodeBatched/jpegdecodebatched.cpp
+1 -1
Visa fil
@@ -113,7 +113,7 @@ int main(int argc, char **argv) {
return EXIT_FAILURE;
}
if (rocjpeg_utils.GetChannelPitchAndSizes(decode_params.output_format, subsamplings[index], widths[index].data(), heights[index].data(), num_channels, output_images[index], channel_sizes)) {
if (rocjpeg_utils.GetChannelPitchAndSizes(decode_params, subsamplings[index], widths[index].data(), heights[index].data(), num_channels, output_images[index], channel_sizes)) {
std::cerr << "ERROR: Failed to get the channel pitch and sizes" << std::endl;
return EXIT_FAILURE;
}
samples/jpegDecodeMultiThreads/jpegdecodemultithreads.cpp
+1 -1
Visa fil
@@ -77,7 +77,7 @@ void ThreadFunction(std::vector<std::string>& jpegFiles, RocJpegHandle rocjpeg_h
return;
}
if (rocjpeg_util.GetChannelPitchAndSizes(decode_params.output_format, subsampling, widths, heights, num_channels, *output_image, channel_sizes)) {
if (rocjpeg_util.GetChannelPitchAndSizes(decode_params, subsampling, widths, heights, num_channels, *output_image, channel_sizes)) {
std::cerr << "ERROR: Failed to get the channel pitch and sizes" << std::endl;
return;
}
samples/rocjpeg_samples_utils.h
+51 -30
Visa fil
@@ -76,6 +76,7 @@ public:
* @param rocjpeg_backend The rocJPEG backend.
* @param decode_params The rocJPEG decode parameters.
* @param num_threads The number of threads.
* @param crop The crop rectangle.
* @param argc The number of command line arguments.
* @param argv The command line arguments.
*/
@@ -153,6 +154,16 @@ public:
*batch_size = atoi(argv[i]);
continue;
}
if (!strcmp(argv[i], "-crop")) {
if (++i == argc || 4 != sscanf(argv[i], "%hd,%hd,%hd,%hd", &decode_params.crop_rectangle.left, &decode_params.crop_rectangle.top, &decode_params.crop_rectangle.right, &decode_params.crop_rectangle.bottom)) {
ShowHelpAndExit("-crop");
}
if ((&decode_params.crop_rectangle.right - &decode_params.crop_rectangle.left) % 2 == 1 || (&decode_params.crop_rectangle.bottom - &decode_params.crop_rectangle.top) % 2 == 1) {
std::cout << "output crop rectangle must have width and height of even numbers" << std::endl;
exit(1);
}
continue;
}
ShowHelpAndExit(argv[i], num_threads != nullptr, batch_size != nullptr);
}
}
@@ -265,7 +276,7 @@ public:
* This function gets the channel pitch and sizes based on the specified output format, chroma subsampling,
* output image, and channel sizes.
*
* @param output_format The output format.
* @param decode_params The decode parameters that specify the output format and crop rectangle.
* @param subsampling The chroma subsampling.
* @param widths The array to store the channel widths.
* @param heights The array to store the channel heights.
@@ -274,37 +285,46 @@ public:
* @param channel_sizes The array to store the channel sizes.
* @return The channel pitch.
*/
int GetChannelPitchAndSizes(RocJpegOutputFormat output_format, RocJpegChromaSubsampling subsampling, uint32_t *widths, uint32_t *heights,
int GetChannelPitchAndSizes(RocJpegDecodeParams decode_params, RocJpegChromaSubsampling subsampling, uint32_t *widths, uint32_t *heights,
uint32_t &num_channels, RocJpegImage &output_image, uint32_t *channel_sizes) {
switch (output_format) {
bool is_roi_valid = false;
uint32_t roi_width;
uint32_t roi_height;
roi_width = decode_params.crop_rectangle.right - decode_params.crop_rectangle.left;
roi_height = decode_params.crop_rectangle.bottom - decode_params.crop_rectangle.top;
if (roi_width > 0 && roi_height > 0 && roi_width <= widths[0] && roi_height <= heights[0]) {
is_roi_valid = true;
}
switch (decode_params.output_format) {
case ROCJPEG_OUTPUT_NATIVE:
switch (subsampling) {
case ROCJPEG_CSS_444:
num_channels = 3;
output_image.pitch[2] = output_image.pitch[1] = output_image.pitch[0] = widths[0];
channel_sizes[2] = channel_sizes[1] = channel_sizes[0] = output_image.pitch[0] * heights[0];
output_image.pitch[2] = output_image.pitch[1] = output_image.pitch[0] = is_roi_valid ? roi_width : widths[0];
channel_sizes[2] = channel_sizes[1] = channel_sizes[0] = output_image.pitch[0] * (is_roi_valid ? roi_height : heights[0]);
break;
case ROCJPEG_CSS_440:
num_channels = 3;
output_image.pitch[2] = output_image.pitch[1] = output_image.pitch[0] = widths[0];
channel_sizes[0] = output_image.pitch[0] * heights[0];
channel_sizes[2] = channel_sizes[1] = output_image.pitch[0] * (heights[0] >> 1);
output_image.pitch[2] = output_image.pitch[1] = output_image.pitch[0] = is_roi_valid ? roi_width : widths[0];
channel_sizes[0] = output_image.pitch[0] * (is_roi_valid ? roi_height : heights[0]);
channel_sizes[2] = channel_sizes[1] = output_image.pitch[0] * ((is_roi_valid ? roi_height : heights[0]) >> 1);
break;
case ROCJPEG_CSS_422:
num_channels = 1;
output_image.pitch[0] = widths[0] * 2;
channel_sizes[0] = output_image.pitch[0] * heights[0];
output_image.pitch[0] = (is_roi_valid ? roi_width : widths[0]) * 2;
channel_sizes[0] = output_image.pitch[0] * (is_roi_valid ? roi_height : heights[0]);
break;
case ROCJPEG_CSS_420:
num_channels = 2;
output_image.pitch[1] = output_image.pitch[0] = widths[0];
channel_sizes[0] = output_image.pitch[0] * heights[0];
channel_sizes[1] = output_image.pitch[1] * (heights[0] >> 1);
output_image.pitch[1] = output_image.pitch[0] = is_roi_valid ? roi_width : widths[0];
channel_sizes[0] = output_image.pitch[0] * (is_roi_valid ? roi_height : heights[0]);
channel_sizes[1] = output_image.pitch[1] * ((is_roi_valid ? roi_height : heights[0]) >> 1);
break;
case ROCJPEG_CSS_400:
num_channels = 1;
output_image.pitch[0] = widths[0];
channel_sizes[0] = output_image.pitch[0] * heights[0];
output_image.pitch[0] = is_roi_valid ? roi_width : widths[0];
channel_sizes[0] = output_image.pitch[0] * (is_roi_valid ? roi_height : heights[0]);
break;
default:
std::cout << "Unknown chroma subsampling!" << std::endl;
@@ -314,32 +334,32 @@ public:
case ROCJPEG_OUTPUT_YUV_PLANAR:
if (subsampling == ROCJPEG_CSS_400) {
num_channels = 1;
output_image.pitch[0] = widths[0];
channel_sizes[0] = output_image.pitch[0] * heights[0];
output_image.pitch[0] = is_roi_valid ? roi_width : widths[0];
channel_sizes[0] = output_image.pitch[0] * (is_roi_valid ? roi_height : heights[0]);
} else {
num_channels = 3;
output_image.pitch[0] = widths[0];
output_image.pitch[1] = widths[1];
output_image.pitch[2] = widths[2];
channel_sizes[0] = output_image.pitch[0] * heights[0];
channel_sizes[1] = output_image.pitch[1] * heights[1];
channel_sizes[2] = output_image.pitch[2] * heights[2];
output_image.pitch[0] = is_roi_valid ? roi_width : widths[0];
output_image.pitch[1] = is_roi_valid ? roi_width : widths[1];
output_image.pitch[2] = is_roi_valid ? roi_width : widths[2];
channel_sizes[0] = output_image.pitch[0] * (is_roi_valid ? roi_height : heights[0]);
channel_sizes[1] = output_image.pitch[1] * (is_roi_valid ? roi_height : heights[1]);
channel_sizes[2] = output_image.pitch[2] * (is_roi_valid ? roi_height : heights[2]);
}
break;
case ROCJPEG_OUTPUT_Y:
num_channels = 1;
output_image.pitch[0] = widths[0];
channel_sizes[0] = output_image.pitch[0] * heights[0];
output_image.pitch[0] = is_roi_valid ? roi_width : widths[0];
channel_sizes[0] = output_image.pitch[0] * (is_roi_valid ? roi_height : heights[0]);
break;
case ROCJPEG_OUTPUT_RGB:
num_channels = 1;
output_image.pitch[0] = widths[0] * 3;
channel_sizes[0] = output_image.pitch[0] * heights[0];
output_image.pitch[0] = (is_roi_valid ? roi_width : widths[0]) * 3;
channel_sizes[0] = output_image.pitch[0] * (is_roi_valid ? roi_height : heights[0]);
break;
case ROCJPEG_OUTPUT_RGB_PLANAR:
num_channels = 3;
output_image.pitch[2] = output_image.pitch[1] = output_image.pitch[0] = widths[0];
channel_sizes[2] = channel_sizes[1] = channel_sizes[0] = output_image.pitch[0] * heights[0];
output_image.pitch[2] = output_image.pitch[1] = output_image.pitch[0] = is_roi_valid ? roi_width : widths[0];
channel_sizes[2] = channel_sizes[1] = channel_sizes[0] = output_image.pitch[0] * (is_roi_valid ? roi_height : heights[0]);
break;
default:
std::cout << "Unknown output format!" << std::endl;
@@ -593,6 +613,7 @@ private:
" 1 for hybrid JPEG decoding using CPU and GPU HIP kernels (currently not supported)) [optional - default: 0]\n"
"-fmt [output format] - select rocJPEG output format for decoding, one of the [native, yuv, y, rgb, rgb_planar] - [optional - default: native]\n"
"-o [output path] - path to an output file or a path to a directory - write decoded images to a file or directory based on selected output format - [optional]\n"
"-crop -crop [crop rectangle] - crop rectangle for output in a comma-separated format: left,top,right,bottom - [optional]\n"
"-d [device id] - specify the GPU device id for the desired device (use 0 for the first device, 1 for the second device, and so on) [optional - default: 0]\n";
if (show_threads) {
std::cout << "-t [threads] - number of threads for parallel JPEG decoding - [optional - default: 2]\n";
@@ -603,4 +624,4 @@ private:
exit(0);
}
};
#endif //ROC_JPEG_SAMPLES_COMMON
#endif //ROC_JPEG_SAMPLES_COMMON