Skip to content

Commit

Permalink
cherry pick to add documentation (#94)
Browse files Browse the repository at this point in the history
* Prepare documentation (#57)

* Updated the readme file to point to the official documentation

* first commit of new documentation structure

* split up and reconfigured Using rocJpeg

* Updated the readme file to point to the official documentation

* first commit of new documentation structure

* split up and reconfigured Using rocJpeg

* fixed missing space

* All changes asked for in the PR except the Destroy Handles part.

* separated creating and destroying handles; added a destroying handles section; added links to hipmalloc() and hipfree()

* removed the installation and build sections so that it doesn't conflict/go out of sync with the official doc

* putting readme back the way it was, minus documentation section

* changed some wording around hipFree()

* Fix a typo in README/Docs (#66)

* Edited the readme and reorganized the toc (#69)

* Edited the readme so that it doesn't duplicate information in the official online docs

* rearranged the toc to match the new structure

* moved Docker section

---------

Co-authored-by: Aryan Salmanpour <[email protected]>
  • Loading branch information
spolifroni-amd and AryanSalmanpour authored Dec 5, 2024
1 parent 66ee736 commit a3df5df
Show file tree
Hide file tree
Showing 12 changed files with 524 additions and 941 deletions.
185 changes: 12 additions & 173 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -4,158 +4,21 @@

rocJPEG is a high performance JPEG decode SDK for AMD GPUs. Using the rocJPEG API, you can access the JPEG decoding features available on your GPU.

## Supported JPEG chroma subsampling
>[!Note]
>The published documentation that includes installation and build instructions is available at [rocJPEG](https://rocm.docs.amd.com/projects/rocJPEG/en/latest/) in an organized, easy-to-read format, with search and a table of contents. The documentation source files reside in the `docs` folder of this repository. As with all ROCm projects, the documentation is open source. For more information on contributing to the documentation, see [Contribute to ROCm documentation](https://rocm.docs.amd.com/en/latest/contribute/contributing.html)
* YUV 4:4:4
* YUV 4:4:0
* YUV 4:2:2
* YUV 4:2:0
* YUV 4:0:0
# Installation folders

## Prerequisites
* Libraries are located in: `/opt/rocm/lib`
* Header files are located in: `/opt/rocm/include/rocjpeg`
* Samples are located in: `/opt/rocm/share/rocjpeg`
* Documentation is located in: `/opt/rocm/share/doc/rocjpeg`

* Linux distribution
* Ubuntu - `20.04` / `22.04` / `24.04`
* RHEL - `8` / `9`
* SLES - `15-SP5`
# Samples

* [ROCm supported hardware](https://rocm.docs.amd.com/projects/install-on-linux/en/latest/reference/system-requirements.html)
Samples that decode JPEG images are available [under `samples` in this repository](samples/). Refer to the individual folders to build and run the samples.

> [!IMPORTANT]
> `gfx908` or higher GPU required
* Install ROCm `6.3.0` or later with [amdgpu-install](https://rocm.docs.amd.com/projects/install-on-linux/en/latest/how-to/amdgpu-install.html): Required usecase - rocm

> [!IMPORTANT]
> `sudo amdgpu-install --usecase=rocm`
* Video Acceleration API (VA-API) Version `2.16.0+` - `Libva` is an implementation for VA-API
```shell
sudo apt install libva-amdgpu-dev
```
> [!NOTE]
> RPM Packages for `RHEL`/`SLES` - `libva-amdgpu-devel`
* AMD VA Drivers
```shell
sudo apt install libva2-amdgpu libva-amdgpu-drm2 libva-amdgpu-wayland2 libva-amdgpu-x11-2 mesa-amdgpu-va-drivers
```
> [!NOTE]
> RPM Packages for `RHEL`/`SLES` - `libva-amdgpu mesa-amdgpu-va-drivers`
* CMake `3.5` or later

```shell
sudo apt install cmake
```

* pkg-config

```shell
sudo apt install pkg-config
```

> [!IMPORTANT]
>
> * If using Ubuntu 22.04, you must install `libstdc++-12-dev`
>
> ```shell
> sudo apt install libstdc++-12-dev
> ```
>
>[!NOTE]
>
> * All package installs are shown with the `apt` package manager. Use the appropriate package manager for your operating system.
> * To install rocJPEG with minimum requirements, follow the [quick-start](./docs/install/quick-start.rst) instructions
### Prerequisites setup script for Linux
For your convenience, we provide the setup script,
[rocJPEG-setup.py](rocJPEG-setup.py) which installs all required dependencies. Run this script only once.
**Usage:**
```shell
python rocJPEG-setup.py --rocm_path [ ROCm Installation Path - optional (default:/opt/rocm)]
```
**NOTE:** This script only needs to be executed once.

## Installation instructions

The installation process uses the following steps:

* [ROCm-supported hardware](https://rocm.docs.amd.com/projects/install-on-linux/en/latest/reference/system-requirements.html) install verification

* Install ROCm `6.3.0` or later with [amdgpu-install](https://rocm.docs.amd.com/projects/install-on-linux/en/latest/how-to/amdgpu-install.html) with `--usecase=rocm`

* Use either [Package install](#package-install) or [Source install](#source-install) as described below.

### Package install

Install rocJPEG runtime, development, and test packages.

* Runtime package - `rocjpeg` only provides the rocjpeg library `librocjpeg.so`
* Development package - `rocjpeg-dev`/`rocjpeg-devel` provides the library, header files, and samples
* Test package - `rocjpeg-test` provides CTest to verify installation

#### Ubuntu

```shell
sudo apt install rocjpeg rocjpeg-dev rocjpeg-test
```

#### RHEL

```shell
sudo yum install rocjpeg rocjpeg-devel rocjpeg-test
```

#### SLES

```shell
sudo zypper install rocjpeg rocjpeg-devel rocjpeg-test
```

>[!NOTE]
> Package install auto installs all dependencies.
### Source install

```shell
git clone https://github.com/ROCm/rocJPEG.git
cd rocJPEG
mkdir build && cd build
cmake ../
make -j8
sudo make install
```

#### Run tests

```shell
make test
```

**NOTE:** run tests with verbose option `make test ARGS="-VV"`

#### Make package

```shell
sudo make package
```

## Verify installation

The installer will copy

* Libraries into `/opt/rocm/lib`
* Header files into `/opt/rocm/include/rocjpeg`
* Samples folder into `/opt/rocm/share/rocjpeg`
* Documents folder into `/opt/rocm/share/doc/rocjpeg`

### Using sample application
## Verifying your installation using samples and tests

To verify your installation using a sample application, run:

Expand All @@ -166,8 +29,6 @@ make -j8
./jpegdecode -i /opt/rocm/share/rocjpeg/images/mug_420.jpg
```

### Using test package

To verify your installation using the `rocjpeg-test` package, run:

```shell
Expand All @@ -176,28 +37,6 @@ cmake /opt/rocm/share/rocjpeg/test/
ctest -VV
```

## Samples

The tool provides a few samples to decode JPEG images [here](samples/). Please refer to the individual folders to build and run the samples.
You can access samples to decode your images in our
[GitHub repository](https://github.com/ROCm/rocJPEG/tree/develop/samples). Refer to the
individual folders to build and run the samples.

## Docker

You can find rocJPEG Docker containers in our
[GitHub repository](https://github.com/ROCm/rocJPEG/tree/develop/docker).

## Documentation

Run the following code to build our documentation locally.

```shell
cd docs
pip3 install -r sphinx/requirements.txt
python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en . _build/html
```
# Docker

For more information on documentation builds, refer to the
[Building documentation](https://rocm.docs.amd.com/en/latest/contribute/building.html)
page.
You can find rocJPEG Docker containers [under `develop/docker` in this repository](https://github.com/ROCm/rocJPEG/tree/develop/docker).
151 changes: 151 additions & 0 deletions docs/how-to/rocjpeg-decoding-a-jpeg-stream.rst
Original file line number Diff line number Diff line change
@@ -0,0 +1,151 @@
.. meta::
:description: decoding a jpeg stream with rocJPEG
:keywords: rocJPEG, ROCm, API, documentation, decoding, jpeg


********************************************************************
Decoding a JPEG stream with rocJPEG
********************************************************************

rocJPEG provides two functions, ``rocJpegDecode()`` and ``rocJpegDecodeBatched()``, for decoding JPEG image.

.. code:: cpp
RocJpegStatus rocJpegDecode(
RocJpegHandle handle,
RocJpegStreamHandle jpeg_stream_handle,
const RocJpegDecodeParams *decode_params,
RocJpegImage *destination);
RocJpegStatus rocJpegDecodeBatched(
RocJpegHandle handle,
RocJpegStreamHandle *jpeg_stream_handles,
int batch_size,
const RocJpegDecodeParams *decode_params,
RocJpegImage *destinations);
``rocJpegDecode()`` is used for decoding single images and ``rocJpegDecodeBatched()`` is used for decoding batches of JPEG images. ``rocJpegDecode()`` and ``rocJpegDecodeBatched()`` copy decoded images to a ``RocJpegImage`` struct.

.. code:: cpp
typedef struct {
uint8_t* channel[ROCJPEG_MAX_COMPONENT];
uint32_t pitch[ROCJPEG_MAX_COMPONENT];
} RocJpegImage;
``rocJpegDecodeBatched()`` behaves the same way as ``rocJpegDecode()`` except that ``rocJpegDecodeBatched()`` takes an array of stream handles and an array of decode parameters as input, decodes the batch of JPEG images, and stores the decoded images in an output array of destination images.

``rocJpegDecodeBatched()`` is suited for use on ASICs with multiple JPEG cores and is more efficient than multiple calls to ``rocJpegDecode()``. Choosing a batch size that is a multiple of available JPEG cores is recommended.

Memory has to be allocate to each channel of ``RocJpegImage``, including every channel of every ``RocJpegImage`` in the destination image array passed to ``rocJpegDecodeBatched()``. Use |hipmalloc|_ to allocate memory.

.. |hipmalloc| replace:: ``hipMalloc()``
.. _hipmalloc: https://rocm.docs.amd.com/projects/HIP/en/latest/how-to/virtual_memory.html

For example:

.. code:: cpp
// Allocate device memory for the decoded output image
RocJpegImage output_image = {};
RocJpegDecodeParams decode_params = {};
decode_params.output_format = ROCJPEG_OUTPUT_NATIVE;
// For this sample assuming the input image has a YUV420 chroma subsampling.
// For YUV420 subsampling, the native decoded output image would be NV12 (i.e., the rocJPegDecode API copies Y to first channel and UV (interleaved) to second channel of RocJpegImage)
output_image.pitch[1] = output_image.pitch[0] = widths[0];
hipError_t hip_status;
hip_status = hipMalloc(&output_image.channel[0], output_image.pitch[0] * heights[0]);
if (hip_status != hipSuccess) {
std::cerr << "Failed to allocate device memory for the first channel" << std::endl;
rocJpegStreamDestroy(rocjpeg_stream_handle);
rocJpegDestroy(handle);
return EXIT_FAILURE;
}
hip_status = hipMalloc(&output_image.channel[1], output_image.pitch[1] * (heights[0] >> 1));
if (hip_status != hipSuccess) {
std::cerr << "Failed to allocate device memory for the second channel" << std::endl;
hipFree((void *)output_image.channel[0]);
rocJpegStreamDestroy(rocjpeg_stream_handle);
rocJpegDestroy(handle);
return EXIT_FAILURE;
}
// Decode the JPEG stream
status = rocJpegDecode(handle, rocjpeg_stream_handle, &decode_params, &output_image);
if (status != ROCJPEG_STATUS_SUCCESS) {
std::cerr << "Failed to decode JPEG stream with error code: " << rocJpegGetErrorName(status) << std::endl;
hipFree((void *)output_image.channel[0]);
hipFree((void *)output_image.channel[1]);
rocJpegStreamDestroy(rocjpeg_stream_handle);
rocJpegDestroy(handle);
return EXIT_FAILURE;
}
The behaviors of ``rocJpegDecode()`` and ``rocJpegDecodeBatched()`` depend on ``RocJpegOutputFormat`` and ``RocJpegDecodeParms``.

``RocJpegOutputFormat`` specifies the output format to be used to decode the JPEG image. It can be set to any one of these output formats:

.. csv-table::
:header: "Output format", "Meaning"

"ROCJPEG_OUTPUT_NATIVE", "Return native unchanged decoded YUV image from the VCN JPEG deocder."
"ROCJPEG_OUTPUT_YUV_PLANAR", "Return in the YUV planar format."
"ROCJPEG_OUTPUT_Y", "Return the Y component only."
"ROCJPEG_OUTPUT_RGB", "Convert to interleaved RGB."
"ROCJPEG_OUTPUT_RGB_PLANAR", "Convert to planar RGB."

``RocJpegOutputFormat`` is a member of the ``RocJpegDecodeParams`` struct. ``RocJpegDecodeParams`` defines the output format, crop rectangle, and target dimensions to use when decoding the image.

.. code:: cpp
typedef struct {
RocJpegOutputFormat output_format; /**< Output data format. See RocJpegOutputFormat for description. */
struct {
int16_t left; /**< Left coordinate of the crop rectangle. */
int16_t top; /**< Top coordinate of the crop rectangle. */
int16_t right; /**< Right coordinate of the crop rectangle. */
int16_t bottom; /**< Bottom coordinate of the crop rectangle. */
} crop_rectangle; /**< Defines the region of interest (ROI) to be copied into the RocJpegImage output buffers. */
struct {
uint32_t width; /**< Target width of the picture to be resized. */
uint32_t height; /**< Target height of the picture to be resized. */
} target_dimension; /**< (future use) Defines the target width and height of the picture to be resized. Both should be even.
If specified, allocate the RocJpegImage buffers based on these dimensions. */
} RocJpegDecodeParams;
For example, consider a situation where ``RocJpegOutputFormat`` is set to ``ROCJPEG_OUTPUT_NATIVE``. Based on the chroma subsampling of the input image, ``rocJpegDecode()`` does one of the following:

* For ``ROCJPEG_CSS_444`` and ``ROCJPEG_CSS_440``: writes Y, U, and V to the first, second, and third channels of ``RocJpegImage``.
* For ``ROCJPEG_CSS_422``: writes YUYV (packed) to the first channel of ``RocJpegImage``.
* For ``ROCJPEG_CSS_420``: writes Y to the first channel and UV (interleaved) to the second channel of ``RocJpegImage``.
* For ``ROCJPEG_CSS_400``: writes Y to the first channel of ``RocJpegImage``.

If ``RocJpegOutputFormat`` is set to ``ROCJPEG_OUTPUT_Y`` or ``ROCJPEG_OUTPUT_RGB``, then ``rocJpegDecode()`` copies the output to the first channel of ``RocJpegImage``.

If ``RocJpegOutputFormat`` is set to ``ROCJPEG_OUTPUT_YUV_PLANAR`` or ``ROCJPEG_OUTPUT_RGB_PLANAR``, the data is written to the corresponding channels of the ``RocJpegImage`` destination structure.

The destination images must be large enough to store the output.

Use |rocjpegimageinfo|_ to extract information and calculate the required memory sizes for the destination image following these guidelines:.

.. |rocjpegimageinfo| replace:: ``rocJpegGetImageInfo()``
.. _rocjpegimageinfo: ./rocjpeg-retrieve-image-info.html

.. csv-table::
:header: "Output format", "Chroma subsampling", "Minimum size of destination.pitch[c]", "Minimum size of destination.channel[c]"

"ROCJPEG_OUTPUT_NATIVE", "ROCJPEG_CSS_444", "destination.pitch[c] = widths[c] for c = 0, 1, 2", "destination.channel[c] = destination.pitch[c] * heights[0] for c = 0, 1, 2"
"ROCJPEG_OUTPUT_NATIVE", "ROCJPEG_CSS_440", "destination.pitch[c] = widths[c] for c = 0, 1, 2", "destination.channel[0] = destination.pitch[0] * heights[0], destination.channel[c] = destination.pitch[c] * heights[0] / 2 for c = 1, 2"
"ROCJPEG_OUTPUT_NATIVE", "ROCJPEG_CSS_422", "destination.pitch[0] = widths[0] * 2", "destination.channel[0] = destination.pitch[0] * heights[0]"
"ROCJPEG_OUTPUT_NATIVE", "ROCJPEG_CSS_420", "destination.pitch[1] = destination.pitch[0] = widths[0]", "destination.channel[0] = destination.pitch[0] * heights[0], destination.channel[1] = destination.pitch[1] * (heights[0] >> 1)"
"ROCJPEG_OUTPUT_NATIVE", "ROCJPEG_CSS_400", "destination.pitch[0] = widths[0]", "destination.channel[0] = destination.pitch[0] * heights[0]"
"ROCJPEG_OUTPUT_YUV_PLANAR", "ROCJPEG_CSS_444, ROCJPEG_CSS_440, ROCJPEG_CSS_422, ROCJPEG_CSS_420", "destination.pitch[c] = widths[c] for c = 0, 1, 2", "destination.channel[c] = destination.pitch[c] * heights[c] for c = 0, 1, 2"
"ROCJPEG_OUTPUT_YUV_PLANAR", "ROCJPEG_CSS_400", "destination.pitch[0] = widths[0]", "destination.channel[0] = destination.pitch[0] * heights[0]"
"ROCJPEG_OUTPUT_Y", "Any of the supported chroma subsampling", "destination.pitch[0] = widths[0]", "destination.channel[0] = destination.pitch[0] * heights[0]"
"ROCJPEG_OUTPUT_RGB", "Any of the supported chroma subsampling", "destination.pitch[0] = widths[0] * 3", "destination.channel[0] = destination.pitch[0] * heights[0]"
"ROCJPEG_OUTPUT_RGB_PLANAR", "Any of the supported chroma subsampling", "destination.pitch[c] = widths[c] for c = 0, 1, 2", "destination.channel[c] = destination.pitch[c] * heights[c] for c = 0, 1, 2"

Loading

0 comments on commit a3df5df

Please sign in to comment.