Bug 5133216: Add QNX tooltrain and cross build support

2025-08-23 15:10:31 +08:00 · 2025-05-29 16:57:28 +08:00 · 2025-05-29 16:57:28 +08:00 · 49307463b5
commit 49307463b5
parent 9c0d5aaae6
16 changed files with 113 additions and 72 deletions
--- a/Samples/0_Introduction/UnifiedMemoryStreams/CMakeLists.txt
+++ b/Samples/0_Introduction/UnifiedMemoryStreams/CMakeLists.txt
@ -19,6 +19,12 @@ endif()
 # Include directories and libraries
 include_directories(../../../Common)

+# This sample is not supported on QNX
+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    message(STATUS "Will not build sample ${PROJECT_NAME} - not supported on QNX")
+    return()
+endif()
+
 # Source file
 if(CMAKE_GENERATOR MATCHES "Visual Studio")
    find_package(OpenMP REQUIRED C CXX)
--- a/Samples/0_Introduction/matrixMulDrv/CMakeLists.txt
+++ b/Samples/0_Introduction/matrixMulDrv/CMakeLists.txt
@ -46,6 +46,14 @@ foreach(arch ${CMAKE_CUDA_ARCHITECTURES})
    list(APPEND GENCODE_FLAGS "-gencode=arch=compute_${arch},code=sm_${arch}")
 endforeach()

+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    set(INCLUDES_LIST)
+    foreach(dir ${CUDAToolkit_INCLUDE_DIRS})
+        list(APPEND INCLUDES_LIST "-I${dir}") 
+    endforeach()
+    string(JOIN " " INCLUDES "${INCLUDES_LIST}")
+endif()
+
 add_custom_command(
    OUTPUT ${CUDA_FATBIN_FILE}
    COMMAND ${CMAKE_CUDA_COMPILER} ${INCLUDES} ${ALL_CCFLAGS} -Wno-deprecated-gpu-targets  ${GENCODE_FLAGS} -o ${CUDA_FATBIN_FILE} -fatbin ${CUDA_KERNEL_SOURCE}
--- a/Samples/0_Introduction/matrixMulDynlinkJIT/CMakeLists.txt
+++ b/Samples/0_Introduction/matrixMulDynlinkJIT/CMakeLists.txt
@ -37,6 +37,6 @@ target_link_libraries(matrixMulDynlinkJIT PUBLIC
    CUDA::cuda_driver
 )

-if(UNIX)
+if(${CMAKE_SYSTEM_NAME} STREQUAL "Linux")
    target_link_libraries(matrixMulDynlinkJIT PUBLIC dl)
 endif()
--- a/Samples/0_Introduction/simpleDrvRuntime/CMakeLists.txt
+++ b/Samples/0_Introduction/simpleDrvRuntime/CMakeLists.txt
@ -48,6 +48,14 @@ foreach(arch ${CMAKE_CUDA_ARCHITECTURES})
    list(APPEND GENCODE_FLAGS "-gencode=arch=compute_${arch},code=sm_${arch}")
 endforeach()

+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    set(INCLUDES_LIST)
+    foreach(dir ${CUDAToolkit_INCLUDE_DIRS})
+        list(APPEND INCLUDES_LIST "-I${dir}") 
+    endforeach()
+    string(JOIN " " INCLUDES "${INCLUDES_LIST}")
+endif()
+
 add_custom_command(
    OUTPUT ${CUDA_FATBIN_FILE}
    COMMAND ${CMAKE_CUDA_COMPILER} ${INCLUDES} ${ALL_CCFLAGS} -Wno-deprecated-gpu-targets  ${GENCODE_FLAGS} -o ${CUDA_FATBIN_FILE} -fatbin ${CUDA_KERNEL_SOURCE}
--- a/Samples/0_Introduction/simpleTextureDrv/CMakeLists.txt
+++ b/Samples/0_Introduction/simpleTextureDrv/CMakeLists.txt
@ -45,6 +45,14 @@ foreach(arch ${CMAKE_CUDA_ARCHITECTURES})
    list(APPEND GENCODE_FLAGS "-gencode=arch=compute_${arch},code=sm_${arch}")
 endforeach()

+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    set(INCLUDES_LIST)
+    foreach(dir ${CUDAToolkit_INCLUDE_DIRS})
+        list(APPEND INCLUDES_LIST "-I${dir}") 
+    endforeach()
+    string(JOIN " " INCLUDES "${INCLUDES_LIST}")
+endif()
+
 add_custom_command(
    OUTPUT ${CUDA_FATBIN_FILE}
    COMMAND ${CMAKE_CUDA_COMPILER} ${INCLUDES} ${ALL_CCFLAGS} -Wno-deprecated-gpu-targets  ${GENCODE_FLAGS} -o ${CUDA_FATBIN_FILE} -fatbin ${CUDA_KERNEL_SOURCE}
--- a/Samples/0_Introduction/vectorAddDrv/CMakeLists.txt
+++ b/Samples/0_Introduction/vectorAddDrv/CMakeLists.txt
@ -45,6 +45,14 @@ foreach(arch ${CMAKE_CUDA_ARCHITECTURES})
    list(APPEND GENCODE_FLAGS "-gencode=arch=compute_${arch},code=sm_${arch}")
 endforeach()

+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    set(INCLUDES_LIST)
+    foreach(dir ${CUDAToolkit_INCLUDE_DIRS})
+        list(APPEND INCLUDES_LIST "-I${dir}") 
+    endforeach()
+    string(JOIN " " INCLUDES "${INCLUDES_LIST}")
+endif()
+
 add_custom_command(
    OUTPUT ${CUDA_FATBIN_FILE}
    COMMAND ${CMAKE_CUDA_COMPILER} ${INCLUDES} ${ALL_CCFLAGS} -Wno-deprecated-gpu-targets  ${GENCODE_FLAGS} -o ${CUDA_FATBIN_FILE} -fatbin ${CUDA_KERNEL_SOURCE}
--- a/Samples/2_Concepts_and_Techniques/threadMigration/CMakeLists.txt
+++ b/Samples/2_Concepts_and_Techniques/threadMigration/CMakeLists.txt
@ -50,6 +50,14 @@ foreach(arch ${CMAKE_CUDA_ARCHITECTURES})
    list(APPEND GENCODE_FLAGS "-gencode=arch=compute_${arch},code=sm_${arch}")
 endforeach()

+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    set(INCLUDES_LIST)
+    foreach(dir ${CUDAToolkit_INCLUDE_DIRS})
+        list(APPEND INCLUDES_LIST "-I${dir}") 
+    endforeach()
+    string(JOIN " " INCLUDES "${INCLUDES_LIST}")
+endif()
+
 add_custom_command(
    OUTPUT ${CUDA_FATBIN_FILE}
    COMMAND ${CMAKE_CUDA_COMPILER} ${INCLUDES} ${ALL_CCFLAGS} -Wno-deprecated-gpu-targets  ${GENCODE_FLAGS} -o ${CUDA_FATBIN_FILE} -fatbin ${CUDA_KERNEL_SOURCE}
--- a/Samples/3_CUDA_Features/bf16TensorCoreGemm/CMakeLists.txt
+++ b/Samples/3_CUDA_Features/bf16TensorCoreGemm/CMakeLists.txt
@ -18,6 +18,12 @@ endif()
 # Include directories and libraries
 include_directories(../../../Common)

+# This sample is not supported on QNX
+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    message(STATUS "Will not build sample ${PROJECT_NAME} - not supported on QNX")
+    return()
+endif()
+
 # Source file
 # Add target for bf16TensorCoreGemm
 add_executable(bf16TensorCoreGemm bf16TensorCoreGemm.cu)
--- a/Samples/3_CUDA_Features/dmmaTensorCoreGemm/CMakeLists.txt
+++ b/Samples/3_CUDA_Features/dmmaTensorCoreGemm/CMakeLists.txt
@ -17,6 +17,12 @@ else()
    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()

+# This sample is not supported on QNX
+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    message(STATUS "Will not build sample ${PROJECT_NAME} - not supported on QNX")
+    return()
+endif()
+
 # Include directories and libraries
 include_directories(../../../Common)

--- a/Samples/3_CUDA_Features/globalToShmemAsyncCopy/CMakeLists.txt
+++ b/Samples/3_CUDA_Features/globalToShmemAsyncCopy/CMakeLists.txt
@ -20,6 +20,12 @@ endif()
 # Include directories and libraries
 include_directories(../../../Common)

+# This sample is not supported on QNX
+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    message(STATUS "Will not build sample ${PROJECT_NAME} - not supported on QNX")
+    return()
+endif()
+
 # Source file
 # Add target for globalToShmemAsyncCopy
 add_executable(globalToShmemAsyncCopy globalToShmemAsyncCopy.cu)
--- a/Samples/3_CUDA_Features/tf32TensorCoreGemm/CMakeLists.txt
+++ b/Samples/3_CUDA_Features/tf32TensorCoreGemm/CMakeLists.txt
@ -20,6 +20,12 @@ endif()
 # Include directories and libraries
 include_directories(../../../Common)

+# This sample is not supported on QNX
+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    message(STATUS "Will not build sample ${PROJECT_NAME} - not supported on QNX")
+    return()
+endif()
+
 # Source file
 # Add target for tf32TensorCoreGemm
 add_executable(tf32TensorCoreGemm tf32TensorCoreGemm.cu)
--- a/Samples/4_CUDA_Libraries/conjugateGradientMultiBlockCG/CMakeLists.txt
+++ b/Samples/4_CUDA_Libraries/conjugateGradientMultiBlockCG/CMakeLists.txt
@ -21,6 +21,12 @@ endif()
 # Include directories and libraries
 include_directories(../../../Common)

+# This sample is not supported on QNX
+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    message(STATUS "Will not build sample ${PROJECT_NAME} - not supported on QNX")
+    return()
+endif()
+
 # Source file
 # Add target for conjugateGradientMultiBlockCG
 add_executable(conjugateGradientMultiBlockCG conjugateGradientMultiBlockCG.cu)
--- a/Samples/4_CUDA_Libraries/conjugateGradientMultiDeviceCG/CMakeLists.txt
+++ b/Samples/4_CUDA_Libraries/conjugateGradientMultiDeviceCG/CMakeLists.txt
@ -20,6 +20,12 @@ endif()
 # Include directories and libraries
 include_directories(../../../Common)

+# This sample is not supported on QNX
+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    message(STATUS "Will not build sample ${PROJECT_NAME} - not supported on QNX")
+    return()
+endif()
+
 # Source file
 # Add target for conjugateGradientMultiDeviceCG
 add_executable(conjugateGradientMultiDeviceCG conjugateGradientMultiDeviceCG.cu)
--- a/Samples/4_CUDA_Libraries/jitLto/CMakeLists.txt
+++ b/Samples/4_CUDA_Libraries/jitLto/CMakeLists.txt
@ -19,6 +19,12 @@ endif()
 # Include directories and libraries
 include_directories(../../../Common)

+# This sample is not supported on QNX
+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    message(STATUS "Will not build sample ${PROJECT_NAME} - not supported on QNX")
+    return()
+endif()
+
 # Source file
 # Add target for jitLto
 add_executable(jitLto jitLto.cpp)
--- a/Samples/4_CUDA_Libraries/watershedSegmentationNPP/CMakeLists.txt
+++ b/Samples/4_CUDA_Libraries/watershedSegmentationNPP/CMakeLists.txt
@ -19,6 +19,12 @@ endif()
 # Include directories and libraries
 include_directories(../../../Common)

+# This sample is not supported on QNX
+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    message(STATUS "Will not build sample ${PROJECT_NAME} - not supported on QNX")
+    return()
+endif()
+
 # Source file
 # Add target for watershedSegmentationNPP
 add_executable(watershedSegmentationNPP watershedSegmentationNPP.cpp)
--- a/cmake/toolchains/toolchain-aarch64-qnx.cmake
+++ b/cmake/toolchains/toolchain-aarch64-qnx.cmake
@ -1,80 +1,27 @@
-#==============================================================================
-# Toolchain file for cross-compiling to aarch64 QNX
-#==============================================================================
-
-# Cross-compiling, so tell CMake that we are not building for the host system
 set(CMAKE_SYSTEM_NAME QNX)
-
-# Target processor architecture
 set(CMAKE_SYSTEM_PROCESSOR aarch64)

-#------------------------------------------------------------------------------
-# QNX host and target come from environment
-# Adjust these or hard-code paths as needed:
-#
-#    set(QNX_HOST "/path/to/qnx/host")      # e.g. /qnx/qnx710/host/linux/x86_64
-#    set(QNX_TARGET "/path/to/qnx/target")  # e.g. /qnx/qnx710/target/qnx7
-#
-# You can also pass them on the cmake command line:
-#    cmake -D QNX_HOST=/path/to/qnx/host \
-#          -D QNX_TARGET=/path/to/qnx/target \
-#          -DCMAKE_TOOLCHAIN_FILE=toolchain-aarch64-qnx.cmake ..
-#------------------------------------------------------------------------------
+set(QNX_HOST $ENV{QNX_HOST})
+set(QNX_TARGET $ENV{QNX_TARGET})

-#----------------------------------------------------------------------------
-# C/C++ Compilers from QNX
-#----------------------------------------------------------------------------
-find_program(QNX_QCC   NAMES qcc   PATHS "${QNX_HOST}/usr/bin")
-find_program(QNX_QPLUS NAMES q++   PATHS "${QNX_HOST}/usr/bin")
+message(STATUS "QNX_HOST = ${QNX_HOST}")
+message(STATUS "QNX_TARGET = ${QNX_TARGET}")

-if(NOT QNX_QCC OR NOT QNX_QPLUS)
-    message(FATAL_ERROR "Could not find qcc or q++ in QNX_HOST=${QNX_HOST}/usr/bin")
-endif()
+set(CMAKE_C_COMPILER ${QNX_HOST}/usr/bin/qcc)
+set(CMAKE_CXX_COMPILER ${QNX_HOST}/usr/bin/q++)

-set(CMAKE_C_COMPILER   "${QNX_QCC}")
-set(CMAKE_CXX_COMPILER "${QNX_QPLUS}")
+set(CMAKE_C_COMPILER_TARGET aarch64)
+set(CMAKE_CXX_COMPILER_TARGET aarch64)

-#----------------------------------------------------------------------------
-# Sysroot (if you want CMake to know the default sysroot)
-#----------------------------------------------------------------------------
-# This is optional, but convenient if the QNX headers/libraries must be found:
-#----------------------------------------------------------------------------
-if(DEFINED QNX_TARGET)
-    set(CMAKE_SYSROOT "${QNX_TARGET}")
-endif()
+set(CMAKE_CUDA_HOST_COMPILER ${CMAKE_CXX_COMPILER} CACHE STRING "" FORCE)
+set(CMAKE_CUDA_COMPILER_ID_TEST_FLAGS_FIRST "-nodlink -L${CUDA_ROOT}/lib64 -L${CUDA_ROOT}/lib -I${CUDA_ROOT}/include")

-#----------------------------------------------------------------------------
-# Additional preprocessor definitions & include paths
-#----------------------------------------------------------------------------
-add_compile_options(
-    -D_QNX_SOURCE
-    -DWIN_INTERFACE_CUSTOM
-)
+set(CMAKE_C_FLAGS " \"-V${__qnx_gcc_ver},gcc_ntoaarch64le\"")
+set(CMAKE_CXX_FLAGS " \"-V${__qnx_gcc_ver},gcc_ntoaarch64le\"")
+set(CMAKE_CUDA_FLAGS " --qpp-config=${__qnx_gcc_ver},gcc_ntoaarch64le")
+set(AUTOMAGIC_NVCC_FLAGS --qpp-config=${__qnx_gcc_ver},gcc_ntoaarch64le CACHE STRING "automagic feature detection flags for cross build")
+add_link_options("-V${__qnx_gcc_ver},gcc_ntoaarch64le")

-# Add an include path to /usr/include/aarch64-qnx-gnu:
-include_directories("/usr/include/aarch64-qnx-gnu")
-
-#----------------------------------------------------------------------------
-# Linker flags
-#----------------------------------------------------------------------------
-set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -L/usr/lib/aarch64-qnx-gnu")
-
-# Because the Makefile also adds -Wl,-rpath-link,/usr/lib/aarch64-qnx-gnu:
-set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -Wl,-rpath-link,/usr/lib/aarch64-qnx-gnu")
-
-# If you have a “target filesystem” (TARGET_FS) to link with:
-#   -L$(TARGET_FS)/usr/lib
-#   -L$(TARGET_FS)/usr/libnvidia
-# etc., you can optionally extend the link flags.  For example:
-#
-# if(DEFINED TARGET_FS)
-#     set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} \
-#         -L${TARGET_FS}/usr/lib -Wl,-rpath-link,${TARGET_FS}/usr/lib \
-#         -L${TARGET_FS}/usr/libnvidia -Wl,-rpath-link,${TARGET_FS}/usr/libnvidia")
-#     include_directories("${TARGET_FS}/../include")
-# endif()
-
-# If you need to link additional libraries, e.g. -lslog2 under certain conditions:
-#   list(APPEND EXTRA_LIBS "slog2")
-#   ...
-#----------------------------------------------------------------------------
+set(CROSS_COMPILE_FOR_QNX ON CACHE BOOL "Cross compiling for QNX platforms")
+string(APPEND CMAKE_CXX_FLAGS " -D_QNX_SOURCE")
+string(APPEND CMAKE_CUDA_FLAGS " -D_QNX_SOURCE")