Add support for rocJpegDecodeBatched API - part 2 of 2 (#39)
* Modify the batchDecode submission * Add support for grouping the jpeg streams based on their chroma_subsampling, width, and height * Modify the RocJpegVappiMemPoolEntry to use vectors for va_surface_ids and hip_interops * Add a function to get the current vcn jpeg spec * Add support to get hip interop mem for the batched submission * switch to std::mutext * Updatet the rocJPEG version * use std::find to find the surface_id * correct the vcn_jpeg_spec_ table for gfx942-mi300a/mi300x * Fix typos * pass the decode_params to the SubmitDecode function
Этот коммит содержится в:
Aryan Salmanpour
коммит произвёл
GitHub
GitHub
родитель
d9df1e04b5
Коммит
3f0be8387e
@@ -102,7 +102,7 @@ RocJpegStatus RocJpegDecoder::InitializeDecoder() {
|
||||
* @return The status of the JPEG decoding operation.
|
||||
*/
|
||||
RocJpegStatus RocJpegDecoder::Decode(RocJpegStreamHandle jpeg_stream_handle, const RocJpegDecodeParams *decode_params, RocJpegImage *destination) {
|
||||
std::lock_guard<std::recursive_mutex> lock(mutex_);
|
||||
std::lock_guard<std::mutex> lock(mutex_);
|
||||
RocJpegStatus rocjpeg_status = ROCJPEG_STATUS_SUCCESS;
|
||||
if (jpeg_stream_handle == nullptr || decode_params == nullptr || destination == nullptr) {
|
||||
return ROCJPEG_STATUS_INVALID_PARAMETER;
|
||||
@@ -111,7 +111,7 @@ RocJpegStatus RocJpegDecoder::Decode(RocJpegStreamHandle jpeg_stream_handle, con
|
||||
const JpegStreamParameters *jpeg_stream_params = rocjpeg_stream_handle->rocjpeg_stream->GetJpegStreamParameters();
|
||||
|
||||
VASurfaceID current_surface_id;
|
||||
CHECK_ROCJPEG(jpeg_vaapi_decoder_.SubmitDecode(jpeg_stream_params, current_surface_id, decode_params->output_format));
|
||||
CHECK_ROCJPEG(jpeg_vaapi_decoder_.SubmitDecode(jpeg_stream_params, current_surface_id, decode_params));
|
||||
|
||||
HipInteropDeviceMem hip_interop_dev_mem = {};
|
||||
CHECK_ROCJPEG(jpeg_vaapi_decoder_.SyncSurface(current_surface_id));
|
||||
@@ -171,10 +171,79 @@ RocJpegStatus RocJpegDecoder::Decode(RocJpegStreamHandle jpeg_stream_handle, con
|
||||
* @return A RocJpegStatus value indicating the success or failure of the decoding operation.
|
||||
*/
|
||||
RocJpegStatus RocJpegDecoder::DecodeBatched(RocJpegStreamHandle *jpeg_streams, int batch_size, const RocJpegDecodeParams *decode_params, RocJpegImage *destinations) {
|
||||
std::lock_guard<std::recursive_mutex> lock(mutex_);
|
||||
for(int i = 0; i < batch_size; i++) {
|
||||
CHECK_ROCJPEG(Decode(jpeg_streams[i], decode_params, &destinations[i]));
|
||||
std::lock_guard<std::mutex> lock(mutex_);
|
||||
if (jpeg_streams == nullptr || decode_params == nullptr || destinations == nullptr) {
|
||||
return ROCJPEG_STATUS_INVALID_PARAMETER;
|
||||
}
|
||||
|
||||
std::vector<VASurfaceID> current_surface_ids;
|
||||
std::vector<JpegStreamParameters> jpeg_streams_params;
|
||||
current_surface_ids.resize(batch_size);
|
||||
jpeg_streams_params.resize(batch_size);
|
||||
VcnJpegSpec current_vcn_jpeg_spec = jpeg_vaapi_decoder_.GetCurrentVcnJpegSpec();
|
||||
|
||||
for (int i = 0; i < batch_size; i += current_vcn_jpeg_spec.num_jpeg_cores) {
|
||||
int batch_end = std::min(i + static_cast<int>(current_vcn_jpeg_spec.num_jpeg_cores), batch_size);
|
||||
int current_batch_size = batch_end - i;
|
||||
|
||||
for (int j = i; j < batch_end; j++) {
|
||||
auto rocjpeg_stream_handle = static_cast<RocJpegStreamParserHandle*>(jpeg_streams[j]);
|
||||
const JpegStreamParameters *jpeg_stream_params = rocjpeg_stream_handle->rocjpeg_stream->GetJpegStreamParameters();
|
||||
jpeg_streams_params[j] = std::move(*jpeg_stream_params);
|
||||
}
|
||||
|
||||
CHECK_ROCJPEG(jpeg_vaapi_decoder_.SubmitDecodeBatched(jpeg_streams_params.data() + i, current_batch_size, decode_params, current_surface_ids.data() + i));
|
||||
|
||||
for (int k = 0; k < current_batch_size; k++) {
|
||||
HipInteropDeviceMem hip_interop_dev_mem = {};
|
||||
VASurfaceID current_surface_id = *(current_surface_ids.data() + k + i);
|
||||
const JpegStreamParameters *jpeg_stream_params = jpeg_streams_params.data() + k + i;
|
||||
CHECK_ROCJPEG(jpeg_vaapi_decoder_.SyncSurface(current_surface_id));
|
||||
CHECK_ROCJPEG(jpeg_vaapi_decoder_.GetHipInteropMem(current_surface_id, hip_interop_dev_mem));
|
||||
|
||||
uint16_t chroma_height = 0;
|
||||
|
||||
switch (decode_params->output_format) {
|
||||
case ROCJPEG_OUTPUT_NATIVE:
|
||||
// Copy the native decoded output buffers from interop memory directly to the destination buffers
|
||||
CHECK_ROCJPEG(GetChromaHeight(hip_interop_dev_mem.surface_format, jpeg_stream_params->picture_parameter_buffer.picture_height, chroma_height));
|
||||
// Copy Luma (first channel) for any surface format
|
||||
CHECK_ROCJPEG(CopyChannel(hip_interop_dev_mem, jpeg_stream_params->picture_parameter_buffer.picture_height, 0, &destinations[k + i]));
|
||||
if (hip_interop_dev_mem.surface_format == VA_FOURCC_NV12) {
|
||||
// Copy the second channel (UV interleaved) for NV12
|
||||
CHECK_ROCJPEG(CopyChannel(hip_interop_dev_mem, chroma_height, 1, &destinations[k + i]));
|
||||
} else if (hip_interop_dev_mem.surface_format == VA_FOURCC_444P ||
|
||||
hip_interop_dev_mem.surface_format == VA_FOURCC_422V) {
|
||||
// Copy the second and third channels for YUV444 and YUV440 (i.e., YUV422V)
|
||||
CHECK_ROCJPEG(CopyChannel(hip_interop_dev_mem, chroma_height, 1, &destinations[k + i]));
|
||||
CHECK_ROCJPEG(CopyChannel(hip_interop_dev_mem, chroma_height, 2, &destinations[k + i]));
|
||||
}
|
||||
break;
|
||||
case ROCJPEG_OUTPUT_YUV_PLANAR:
|
||||
CHECK_ROCJPEG(GetChromaHeight(hip_interop_dev_mem.surface_format, jpeg_stream_params->picture_parameter_buffer.picture_height, chroma_height));
|
||||
CHECK_ROCJPEG(GetPlanarYUVOutputFormat(hip_interop_dev_mem, jpeg_stream_params->picture_parameter_buffer.picture_width,
|
||||
jpeg_stream_params->picture_parameter_buffer.picture_height, chroma_height, &destinations[k + i]));
|
||||
break;
|
||||
case ROCJPEG_OUTPUT_Y:
|
||||
CHECK_ROCJPEG(GetYOutputFormat(hip_interop_dev_mem, jpeg_stream_params->picture_parameter_buffer.picture_width,
|
||||
jpeg_stream_params->picture_parameter_buffer.picture_height, &destinations[k + i]));
|
||||
break;
|
||||
case ROCJPEG_OUTPUT_RGB:
|
||||
CHECK_ROCJPEG(ColorConvertToRGB(hip_interop_dev_mem, jpeg_stream_params->picture_parameter_buffer.picture_width,
|
||||
jpeg_stream_params->picture_parameter_buffer.picture_height, &destinations[k + i]));
|
||||
break;
|
||||
case ROCJPEG_OUTPUT_RGB_PLANAR:
|
||||
CHECK_ROCJPEG(ColorConvertToRGBPlanar(hip_interop_dev_mem, jpeg_stream_params->picture_parameter_buffer.picture_width,
|
||||
jpeg_stream_params->picture_parameter_buffer.picture_height, &destinations[k + i]));
|
||||
break;
|
||||
default:
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
CHECK_HIP(hipStreamSynchronize(hip_stream_));
|
||||
return ROCJPEG_STATUS_SUCCESS;
|
||||
}
|
||||
/**
|
||||
@@ -192,7 +261,7 @@ RocJpegStatus RocJpegDecoder::DecodeBatched(RocJpegStreamHandle *jpeg_streams, i
|
||||
* or ROCJPEG_STATUS_INVALID_PARAMETER if any of the input parameters are invalid.
|
||||
*/
|
||||
RocJpegStatus RocJpegDecoder::GetImageInfo(RocJpegStreamHandle jpeg_stream_handle, uint8_t *num_components, RocJpegChromaSubsampling *subsampling, uint32_t *widths, uint32_t *heights){
|
||||
std::lock_guard<std::recursive_mutex> lock(mutex_);
|
||||
std::lock_guard<std::mutex> lock(mutex_);
|
||||
if (jpeg_stream_handle == nullptr || num_components == nullptr || subsampling == nullptr || widths == nullptr || heights == nullptr) {
|
||||
return ROCJPEG_STATUS_INVALID_PARAMETER;
|
||||
}
|
||||
|
||||
Ссылка в новой задаче
Block a user