Improve TheRock build and optional dependency handling

CMakeLists.txt

@@ -177,7 +177,7 @@ find_package(rocjpeg 1.0.0 QUIET)
 find_package(pybind11 REQUIRED)
 find_package(DLPACK REQUIRED)
 find_package(FFmpeg QUIET)
-set(ROCPYDECODE_USE_FFMPEG ${FFMPEG_FOUND})
+set(ROCPYDECODE_USE_FFMPEG FALSE)
 
 set(ROCPYDECODE_DEPENDENCY_READY FALSE)
 set(ROCPYJPEG_DEPENDENCY_READY FALSE)
@@ -220,14 +220,25 @@ if(rocdecode_FOUND)
     set(include ${include_base})
     set(sources ${sources_base})
 
+    set(ROCDECODE_FFMPEG_UTILS_DIR "${ROCM_PATH}/share/rocdecode/utils/ffmpegvideodecode")
+    file(GLOB ROCDECODE_FFMPEG_UTILS_SOURCES "${ROCDECODE_FFMPEG_UTILS_DIR}/*.cpp")
+    if(FFMPEG_FOUND
+       AND EXISTS "${ROCDECODE_FFMPEG_UTILS_DIR}/ffmpeg_video_dec.h"
+       AND ROCDECODE_FFMPEG_UTILS_SOURCES)
+        set(ROCPYDECODE_USE_FFMPEG TRUE)
+    endif()
+
     if(ROCPYDECODE_USE_FFMPEG)
         # Demux + CPU backend (FFmpeg-dependent)
-        file(GLOB sources_ffmpeg src/rocdecode/roc_pyvideodemuxer.cpp src/rocdecode/roc_pyvideodecodecpu.cpp src/rocdecode/roc_pydecode_test.cpp ${ROCM_PATH}/share/rocdecode/utils/ffmpegvideodecode/*.cpp)
-        file(GLOB include_ffmpeg ${ROCM_PATH}/share/rocdecode/utils/ffmpegvideodecode/*.h)
+        file(GLOB sources_ffmpeg src/rocdecode/roc_pyvideodemuxer.cpp src/rocdecode/roc_pyvideodecodecpu.cpp src/rocdecode/roc_pydecode_test.cpp ${ROCDECODE_FFMPEG_UTILS_SOURCES})
+        file(GLOB include_ffmpeg ${ROCDECODE_FFMPEG_UTILS_DIR}/*.h)
         list(APPEND include ${include_ffmpeg})
         list(APPEND sources ${sources_ffmpeg})
         include_directories(${AVUTIL_INCLUDE_DIR} ${AVCODEC_INCLUDE_DIR} ${AVFORMAT_INCLUDE_DIR})
         set(LINK_LIBRARY_LIST ${LINK_LIBRARY_LIST} ${FFMPEG_LIBRARIES})
+    elseif(FFMPEG_FOUND)
+        message(STATUS "FFmpeg found, but rocDecode FFmpeg utilities were not found under ${ROCDECODE_FFMPEG_UTILS_DIR}; building rocPyDecode without demux/CPU backend (GPU decode only).")
+        set(ROCPYDECODE_HOST_LIBRARY_FOUND FALSE)
     else()
         message(STATUS "FFmpeg not found; building rocPyDecode without demux/CPU backend (GPU decode only).")
         set(ROCPYDECODE_HOST_LIBRARY_FOUND FALSE)
@@ -435,16 +446,16 @@ endif()
 if(ROCPYDECODE_DEPENDENCY_READY)
     set(ROCDECODE_VIDEO_DIR "${ROCM_PATH}/share/rocdecode/video")
     if(IS_DIRECTORY "${ROCDECODE_VIDEO_DIR}")
-        file(MAKE_DIRECTORY "${CMAKE_SOURCE_DIR}/data/videos")
+        file(MAKE_DIRECTORY "${CMAKE_BINARY_DIR}/data/videos")
         file(GLOB ROCDECODE_VIDEO_FILES "${ROCDECODE_VIDEO_DIR}/*")
         foreach(VID_FILE IN LISTS ROCDECODE_VIDEO_FILES)
             get_filename_component(VID_NAME "${VID_FILE}" NAME)
-            set(VID_LINK "${CMAKE_SOURCE_DIR}/data/videos/${VID_NAME}")
+            set(VID_LINK "${CMAKE_BINARY_DIR}/data/videos/${VID_NAME}")
             if(NOT EXISTS "${VID_LINK}")
                 execute_process(COMMAND ${CMAKE_COMMAND} -E create_symlink "${VID_FILE}" "${VID_LINK}")
             endif()
         endforeach()
-        message("-- ${BoldBlue}Ensured rocdecode sample videos are linked into ${CMAKE_SOURCE_DIR}/data/videos${ColourReset}")
+        message("-- ${BoldBlue}Ensured rocdecode sample videos are linked into ${CMAKE_BINARY_DIR}/data/videos${ColourReset}")
     endif()
 endif()
 

docker/rocpydecode-with-therock.dockerfile

@@ -0,0 +1,72 @@
+FROM ubuntu:24.04
+
+ARG DEBIAN_FRONTEND=noninteractive
+ARG ROCM_SERIES=10.1
+ARG GPU_TARGET=gfx1250
+ARG ROCM_BUILD=20260821-32431166035
+ARG ROCM_PACKAGE_VERSION=10.1.0~20260821-32431166035
+ARG ALLOW_UNSIGNED_ROCM_REPOSITORY=false
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+        ca-certificates \
+        gnupg \
+        wget \
+    && rm -rf /var/lib/apt/lists/* \
+    && mkdir -p /etc/apt/keyrings \
+    && wget -qO- https://repo.amd.com/rocm/packages-multi-arch/gpg/rocm.gpg \
+        | gpg --dearmor -o /etc/apt/keyrings/amdrocm.gpg
+
+RUN if [ "${ALLOW_UNSIGNED_ROCM_REPOSITORY}" = "true" ]; then \
+        repository_options="trusted=yes"; \
+    elif [ "${ALLOW_UNSIGNED_ROCM_REPOSITORY}" = "false" ]; then \
+        repository_options="signed-by=/etc/apt/keyrings/amdrocm.gpg"; \
+    else \
+        echo "ALLOW_UNSIGNED_ROCM_REPOSITORY must be true or false" >&2; \
+        exit 2; \
+    fi \
+    && echo "deb [${repository_options}] https://rocm.nightlies.amd.com/packages-multi-arch/deb/${ROCM_BUILD} stable main" \
+        > /etc/apt/sources.list.d/rocm-nightly.list \
+    && apt-get update \
+    && apt-get install -y --no-install-recommends \
+        "amdrocm-core-sdk${ROCM_SERIES}-${GPU_TARGET}=${ROCM_PACKAGE_VERSION}" \
+        "amdrocm-decode-dev${ROCM_SERIES}=${ROCM_PACKAGE_VERSION}" \
+        "amdrocm-decode-test${ROCM_SERIES}=${ROCM_PACKAGE_VERSION}" \
+        "amdrocm-jpeg-dev${ROCM_SERIES}=${ROCM_PACKAGE_VERSION}" \
+        "amdrocm-jpeg-test${ROCM_SERIES}=${ROCM_PACKAGE_VERSION}" \
+        build-essential \
+        cmake \
+        git \
+        libavcodec-dev \
+        libavformat-dev \
+        libavutil-dev \
+        libdlpack-dev \
+        libnuma-dev \
+        libswscale-dev \
+        ninja-build \
+        pkg-config \
+        python3.12 \
+        python3.12-dev \
+        python3.12-venv \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN python3.12 -m venv /opt/venv \
+    && /opt/venv/bin/python -m pip install --no-cache-dir --upgrade pip setuptools wheel \
+    && /opt/venv/bin/python -m pip install --no-cache-dir numpy pybind11 pytest
+
+ENV VIRTUAL_ENV=/opt/venv
+ENV ROCM_PATH=/opt/rocm
+ENV ROCM_HOME=/opt/rocm
+ENV PATH=/opt/venv/bin:/opt/rocm/bin:/opt/rocm/lib/llvm/bin:${PATH}
+ENV CMAKE_PREFIX_PATH=/opt/rocm/lib/cmake:/opt/venv/lib/python3.12/site-packages/pybind11/share/cmake/pybind11
+ENV LD_LIBRARY_PATH=/opt/rocm/lib
+
+RUN hipconfig --full \
+    && test -d /opt/rocm/share/rocdecode/utils \
+    && test -d /opt/rocm/share/rocdecode/test \
+    && test -d /opt/rocm/share/rocdecode/video \
+    && test -d /opt/rocm/share/rocjpeg/images
+
+WORKDIR /workspace/EssamWork/ROCPYDECODE
+
+CMD ["bash"]

samples/rocdecode/videodecoderaw.py

@@ -138,20 +138,37 @@ def feed_packet(buf, is_eos=False):
         return False
 
     data = memoryview(Path(input_path).read_bytes())
-    sps = pps = None
+    vps = sps = pps = None
     for nal in _annexb_slices(data):
         # nal header byte after start code
         sc_len = 4 if nal[:4] == b"\x00\x00\x00\x01" else 3
-        nal_type = nal[sc_len] & 0x1F
-        if nal_type == 7:  # SPS
-            sps = bytes(nal)
-            continue
-        if nal_type == 8:  # PPS
-            pps = bytes(nal)
-            continue
-        if nal_type in (1, 5):  # non-IDR / IDR slice -> frame
-            # prepend latest SPS/PPS so each frame is self-contained
+        if codec == dectypes.rocDecVideoCodec_HEVC:
+            nal_type = (nal[sc_len] >> 1) & 0x3F
+            if nal_type == 32:  # VPS
+                vps = bytes(nal)
+                continue
+            if nal_type == 33:  # SPS
+                sps = bytes(nal)
+                continue
+            if nal_type == 34:  # PPS
+                pps = bytes(nal)
+                continue
+            is_slice = nal_type <= 31
+        else:
+            nal_type = nal[sc_len] & 0x1F
+            if nal_type == 7:  # SPS
+                sps = bytes(nal)
+                continue
+            if nal_type == 8:  # PPS
+                pps = bytes(nal)
+                continue
+            is_slice = nal_type in (1, 5)  # non-IDR / IDR slice
+
+        if is_slice:
+            # Prepend the latest parameter sets so each frame is self-contained.
             parts = []
+            if vps:
+                parts.append(vps)
             if sps:
                 parts.append(sps)
             if pps:
@@ -173,7 +190,7 @@ def feed_packet(buf, is_eos=False):
             f"({fps:.2f} fps)."
         )
     else:
-        print("info: No frames decoded.")
+        raise RuntimeError("No frames decoded.")
 
 
 def main():

src/common/roc_pybuffer.cpp

@@ -123,7 +123,8 @@ const DLTensor &BufferInterface::dlTensor() const {
 }
 
 void BufferInterface::ExportToPython(py::module &m) {
-    py::class_<BufferInterface, std::shared_ptr<BufferInterface>>(m, "BufferInterface", py::dynamic_attr())
+    py::class_<BufferInterface, std::shared_ptr<BufferInterface>>(
+        m, "BufferInterface", py::module_local(), py::dynamic_attr())
         .def_property_readonly("shape", &BufferInterface::shape, "Get the shape of the buffer as an array")
         .def_property_readonly("strides", &BufferInterface::strides, "Get the strides of the buffer")
         .def_property_readonly("dtype", &BufferInterface::dtype, "Get the data type of the buffer")