Merge 13.0 changes into 13.1 development branch

2026-01-08 10:27:49 +08:00 · 2025-06-24 12:44:35 -07:00 · 2025-06-24 12:44:35 -07:00 · 7c8fccc0d4
commit 7c8fccc0d4
parent 859debae1b 45314d7ff8
34 changed files with 241 additions and 94 deletions
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@ -77,6 +77,7 @@
            * `simpleCUFFT_2d_MGPU`
            * `simpleCUFFT_MGPU`
            * `simpleCUFFT_callback`
+* Updated toolchain for cross-compilation for Tegra QNX platforms.

 ### CUDA 12.9
 * Updated toolchain for cross-compilation for Tegra Linux platforms.
--- a/Common/helper_multiprocess.cpp
+++ b/Common/helper_multiprocess.cpp
@ -186,7 +186,7 @@ int ipcCreateSocket(ipcHandle *&handle, const char *name,
  }

  unlink(name);
-  bzero(&servaddr, sizeof(servaddr));
+  memset(&servaddr, 0, sizeof(servaddr));
  servaddr.sun_family = AF_UNIX;

  size_t len = strlen(name);
@ -220,12 +220,12 @@ int ipcOpenSocket(ipcHandle *&handle) {
    return -1;
  }

-  bzero(&cliaddr, sizeof(cliaddr));
+  memset(&cliaddr, 0, sizeof(cliaddr));
  cliaddr.sun_family = AF_UNIX;
-  char temp[10];
+  char temp[20];

  // Create unique name for the socket.
-  sprintf(temp, "%u", getpid());
+  sprintf(temp, "%s%u", SOCK_FOLDER, getpid());

  strcpy(cliaddr.sun_path, temp);
  if (bind(sock, (struct sockaddr *)&cliaddr, sizeof(cliaddr)) < 0) {
@ -326,7 +326,7 @@ int ipcSendDataToServer(ipcHandle *handle, const char *serverName,
  ssize_t sendResult;
  struct sockaddr_un serveraddr;

-  bzero(&serveraddr, sizeof(serveraddr));
+  memset(&serveraddr, 0, sizeof(serveraddr));
  serveraddr.sun_family = AF_UNIX;
  strncpy(serveraddr.sun_path, serverName, sizeof(serveraddr.sun_path) - 1);

@ -359,10 +359,10 @@ int ipcSendShareableHandle(ipcHandle *handle,
  socklen_t len = sizeof(cliaddr);

  // Construct client address to send this SHareable handle to
-  bzero(&cliaddr, sizeof(cliaddr));
+  memset(&cliaddr, 0, sizeof(cliaddr));
  cliaddr.sun_family = AF_UNIX;
-  char temp[10];
-  sprintf(temp, "%u", process);
+  char temp[20];
+  sprintf(temp, "%s%u", SOCK_FOLDER, process);
  strcpy(cliaddr.sun_path, temp);
  len = sizeof(cliaddr);

--- a/Common/helper_multiprocess.h
+++ b/Common/helper_multiprocess.h
@ -54,6 +54,13 @@
 #endif
 #include <vector>

+// Define "/tmp" as socket creating folder for QNX
+#if defined(__QNX__)
+#define SOCK_FOLDER "/tmp/" 
+#else
+#define SOCK_FOLDER ""
+#endif
+
 typedef struct sharedMemoryInfo_st {
    void *addr;
    size_t size;
--- a/README.md
+++ b/README.md
@ -135,9 +135,30 @@ $ cmake -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc -DCMAKE_LIBRARY_PATH=/usr

 ### QNX

-Note that in the current branch sample cross-compilation for QNX is not fully validated. This placeholder will be updated in the
-near future with QNX cross-compilation instructions. In the meantime, if you want to cross-compile for QNX please check out one
-of the previous tags prior to the CMake build system transition in 12.8.
+Cross-compilation for QNX with CMake is supported in the CUDA 13.0 samples release and newer. An example build for
+the Tegra Thor QNX platform might look like this:
+
+```
+$ mkdir build
+$ cd build
+
+QNX_HOST=/path/to/qnx/host \
+QNX_TARGET=/path/to/qnx/target \
+cmake .. \
+-DBUILD_TEGRA=True \
+-DCMAKE_CUDA_COMPILER=/usr/local/cuda-safe-13.0/bin/nvcc \
+-DCMAKE_TOOLCHAIN_FILE=../cmake/toolchains/toolchain-aarch64-qnx.cmake \
+-DCMAKE_LIBRARY_PATH=/usr/local/cuda-safe-13.0/thor/targets/aarch64-qnx/lib/stubs/ \
+-DCMAKE_INCLUDE_PATH=/usr/local/cuda-safe-13.0/thor/targets/aarch64-qnx/include/
+```
+
+### Forward Compatibility
+
+To build samples with new CUDA Toolkit(CUDA 13.0 or later) and UMD(Version 580 or later) and old KMD(Version 550 or earlier)，you need to set the `CMAKE_PREFIX_PATH` for using new driver library, the command might like this:
+
+```
+cmake -DCMAKE_PREFIX_PATH=/usr/local/cuda/lib64/stubs/ ..
+```

 ## Running All Samples as Tests

--- 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 ${CUDA_KERNEL_SOURCE}
--- a/Samples/0_Introduction/matrixMulDynlinkJIT/CMakeLists.txt
+++ b/Samples/0_Introduction/matrixMulDynlinkJIT/CMakeLists.txt
@ -31,12 +31,17 @@ set_target_properties(matrixMulDynlinkJIT PROPERTIES
    CUDA_SEPARABLE_COMPILATION ON
    POSITION_INDEPENDENT_CODE OFF
 )
-set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -no-pie")
+
+# Only add -no-pie for GCC or Clang
+if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU" OR CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
+    set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -no-pie")
+endif()
+
 target_link_libraries(matrixMulDynlinkJIT PUBLIC
    CUDA::cudart
    CUDA::cuda_driver
 )

-if(UNIX)
+if(${CMAKE_SYSTEM_NAME} STREQUAL "Linux")
    target_link_libraries(matrixMulDynlinkJIT PUBLIC dl)
 endif()
--- a/Samples/0_Introduction/matrixMul_nvrtc/CMakeLists.txt
+++ b/Samples/0_Introduction/matrixMul_nvrtc/CMakeLists.txt
@ -32,23 +32,26 @@ target_link_libraries(matrixMul_nvrtc PRIVATE
    CUDA::cuda_driver
 )

+# The primary directory of CUDAToolkit_INCLUDE_DIRS is the CUDA Toolkit's include directory for finding the header files.
+list(GET CUDAToolkit_INCLUDE_DIRS 0 CUDA_INCLUDE_DIR)
+
 # Copy clock_kernel.cu to the output directory
 add_custom_command(TARGET matrixMul_nvrtc POST_BUILD
    COMMAND ${CMAKE_COMMAND} -E copy_if_different
-    ${CMAKE_CURRENT_SOURCE_DIR}/matrixMul_kernel.cu ${CUDAToolkit_BIN_DIR}/../include/cooperative_groups.h ${CMAKE_CURRENT_BINARY_DIR}
+    ${CMAKE_CURRENT_SOURCE_DIR}/matrixMul_kernel.cu ${CUDA_INCLUDE_DIR}/cooperative_groups.h ${CMAKE_CURRENT_BINARY_DIR}
 )

 add_custom_command(TARGET matrixMul_nvrtc POST_BUILD
    COMMAND ${CMAKE_COMMAND} -E copy_directory
-    ${CUDAToolkit_BIN_DIR}/../include/cooperative_groups ${CMAKE_CURRENT_BINARY_DIR}/cooperative_groups
+    ${CUDA_INCLUDE_DIR}/cooperative_groups ${CMAKE_CURRENT_BINARY_DIR}/cooperative_groups
 )

 add_custom_command(TARGET matrixMul_nvrtc POST_BUILD
    COMMAND ${CMAKE_COMMAND} -E copy_directory
-    ${CUDAToolkit_BIN_DIR}/../include/cccl/nv ${CMAKE_CURRENT_BINARY_DIR}/nv
+    ${CUDA_INCLUDE_DIR}/cccl/nv ${CMAKE_CURRENT_BINARY_DIR}/nv
 )

 add_custom_command(TARGET matrixMul_nvrtc POST_BUILD
    COMMAND ${CMAKE_COMMAND} -E copy_directory
-    ${CUDAToolkit_BIN_DIR}/../include/cccl/cuda ${CMAKE_CURRENT_BINARY_DIR}/cuda
+    ${CUDA_INCLUDE_DIR}/cccl/cuda ${CMAKE_CURRENT_BINARY_DIR}/cuda
 )
--- a/Samples/0_Introduction/simpleAWBarrier/CMakeLists.txt
+++ b/Samples/0_Introduction/simpleAWBarrier/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 "simpleAWBarrier is not supported on QNX")
+    return()
+endif()
+
 # Source file
 # Add target for simpleAWBarrier
 add_executable(simpleAWBarrier simpleAWBarrier.cu)
--- 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
@ -37,7 +37,7 @@ target_link_libraries(threadMigration PUBLIC
    CUDA::cuda_driver
 )

-if(UNIX)
+if(UNIX AND NOT ${CMAKE_SYSTEM_NAME} STREQUAL "QNX")
    target_link_libraries(threadMigration PUBLIC pthread)
 endif()

@ -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/cdpAdvancedQuicksort/CMakeLists.txt
+++ b/Samples/3_CUDA_Features/cdpAdvancedQuicksort/CMakeLists.txt
@ -9,8 +9,9 @@ find_package(CUDAToolkit REQUIRED)
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)

 # The aarch64/sbsa_aarch64 CUDA toolkit are support on Tegra since 13.0, so need to check which version of the toolkit is installed
-string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-linux" _aarch64_ctk)
-if(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64" AND NOT _aarch64_ctk EQUAL -1)
+string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-linux" _aarch64_linux_ctk)
+string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-qnx" _aarch64_qnx_ctk)
+if(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64" AND (NOT _aarch64_linux_ctk EQUAL -1 OR NOT _aarch64_qnx_ctk EQUAL -1))
    set(CMAKE_CUDA_ARCHITECTURES 87 110)
 else()
    set(CMAKE_CUDA_ARCHITECTURES 75 80 86 89 90 100 110 120)
--- a/Samples/3_CUDA_Features/cdpBezierTessellation/CMakeLists.txt
+++ b/Samples/3_CUDA_Features/cdpBezierTessellation/CMakeLists.txt
@ -9,8 +9,9 @@ find_package(CUDAToolkit REQUIRED)
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)

 # The aarch64/sbsa_aarch64 CUDA toolkit are support on Tegra since 13.0, so need to check which version of the toolkit is installed
-string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-linux" _aarch64_ctk)
-if(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64" AND NOT _aarch64_ctk EQUAL -1)
+string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-linux" _aarch64_linux_ctk)
+string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-qnx" _aarch64_qnx_ctk)
+if(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64" AND (NOT _aarch64_linux_ctk EQUAL -1 OR NOT _aarch64_qnx_ctk EQUAL -1))
    set(CMAKE_CUDA_ARCHITECTURES 87 110)
 else()
    set(CMAKE_CUDA_ARCHITECTURES 75 80 86 89 90 100 110 120)
--- a/Samples/3_CUDA_Features/cdpQuadtree/CMakeLists.txt
+++ b/Samples/3_CUDA_Features/cdpQuadtree/CMakeLists.txt
@ -9,8 +9,9 @@ find_package(CUDAToolkit REQUIRED)
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)

 # The aarch64/sbsa_aarch64 CUDA toolkit are support on Tegra since 13.0, so need to check which version of the toolkit is installed
-string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-linux" _aarch64_ctk)
-if(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64" AND NOT _aarch64_ctk EQUAL -1)
+string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-linux" _aarch64_linux_ctk)
+string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-qnx" _aarch64_qnx_ctk)
+if(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64" AND (NOT _aarch64_linux_ctk EQUAL -1 OR NOT _aarch64_qnx_ctk EQUAL -1))
    set(CMAKE_CUDA_ARCHITECTURES 87 110)
 else()
    set(CMAKE_CUDA_ARCHITECTURES 75 80 86 89 90 100 110 120)
--- a/Samples/3_CUDA_Features/cdpSimplePrint/CMakeLists.txt
+++ b/Samples/3_CUDA_Features/cdpSimplePrint/CMakeLists.txt
@ -9,8 +9,9 @@ find_package(CUDAToolkit REQUIRED)
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)

 # The aarch64/sbsa_aarch64 CUDA toolkit are support on Tegra since 13.0, so need to check which version of the toolkit is installed
-string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-linux" _aarch64_ctk)
-if(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64" AND NOT _aarch64_ctk EQUAL -1)
+string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-linux" _aarch64_linux_ctk)
+string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-qnx" _aarch64_qnx_ctk)
+if(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64" AND (NOT _aarch64_linux_ctk EQUAL -1 OR NOT _aarch64_qnx_ctk EQUAL -1))
    set(CMAKE_CUDA_ARCHITECTURES 87 110)
 else()
    set(CMAKE_CUDA_ARCHITECTURES 75 80 86 89 90 100 110 120)
--- a/Samples/3_CUDA_Features/cdpSimpleQuicksort/CMakeLists.txt
+++ b/Samples/3_CUDA_Features/cdpSimpleQuicksort/CMakeLists.txt
@ -9,8 +9,9 @@ find_package(CUDAToolkit REQUIRED)
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)

 # The aarch64/sbsa_aarch64 CUDA toolkit are support on Tegra since 13.0, so need to check which version of the toolkit is installed
-string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-linux" _aarch64_ctk)
-if(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64" AND NOT _aarch64_ctk EQUAL -1)
+string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-linux" _aarch64_linux_ctk)
+string(FIND "${CUDAToolkit_INCLUDE_DIRS}" "aarch64-qnx" _aarch64_qnx_ctk)
+if(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64" AND (NOT _aarch64_linux_ctk EQUAL -1 OR NOT _aarch64_qnx_ctk EQUAL -1))
    set(CMAKE_CUDA_ARCHITECTURES 87 110)
 else()
    set(CMAKE_CUDA_ARCHITECTURES 75 80 86 89 90 100 110 120)
--- a/Samples/3_CUDA_Features/dmmaTensorCoreGemm/CMakeLists.txt
+++ b/Samples/3_CUDA_Features/dmmaTensorCoreGemm/CMakeLists.txt
@ -17,6 +17,12 @@ else()
    set(CMAKE_CUDA_FLAGS "${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/memMapIPCDrv/CMakeLists.txt
+++ b/Samples/3_CUDA_Features/memMapIPCDrv/CMakeLists.txt
@ -37,13 +37,25 @@ target_link_libraries(memMapIPCDrv PUBLIC
    CUDA::cuda_driver
 )

-if(UNIX)
+if(UNIX AND NOT ${CMAKE_SYSTEM_NAME} STREQUAL "QNX")
    target_link_libraries(memMapIPCDrv PUBLIC rt)
 endif()

+if(${CMAKE_SYSTEM_NAME} STREQUAL "QNX")
+    target_link_libraries(memMapIPCDrv PUBLIC socket)
+endif()
+
 set(CUDA_PTX_FILE "${CMAKE_CURRENT_BINARY_DIR}/memMapIpc_kernel64.ptx")
 set(CUDA_KERNEL_SOURCE "${CMAKE_CURRENT_SOURCE_DIR}/memMapIpc_kernel.cu")

+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_PTX_FILE}
    COMMAND ${CMAKE_CUDA_COMPILER} ${INCLUDES} ${ALL_CCFLAGS} -Wno-deprecated-gpu-targets -o ${CUDA_PTX_FILE} -ptx ${CUDA_KERNEL_SOURCE}
--- a/Samples/3_CUDA_Features/ptxjit/CMakeLists.txt
+++ b/Samples/3_CUDA_Features/ptxjit/CMakeLists.txt
@ -41,6 +41,14 @@ target_link_libraries(ptxjit PUBLIC
 set(CUDA_PTX_FILE "${CMAKE_CURRENT_BINARY_DIR}/ptxjit_kernel64.ptx")
 set(CUDA_KERNEL_SOURCE "${CMAKE_CURRENT_SOURCE_DIR}/ptxjit_kernel.cu")

+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_PTX_FILE}
    COMMAND ${CMAKE_CUDA_COMPILER} ${INCLUDES} ${ALL_CCFLAGS} -Wno-deprecated-gpu-targets -o ${CUDA_PTX_FILE} -ptx ${CUDA_KERNEL_SOURCE}
--- 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 ${PROJECT_NAME} - not supported on QNX")
+    return()
+endif()
+
 # Source file
 # Add target for tf32TensorCoreGemm
 add_executable(tf32TensorCoreGemm tf32TensorCoreGemm.cu)
--- a/Samples/4_CUDA_Libraries/conjugateGradientCudaGraphs/CMakeLists.txt
+++ b/Samples/4_CUDA_Libraries/conjugateGradientCudaGraphs/CMakeLists.txt
@ -4,6 +4,11 @@ list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/Module

 project(conjugateGradientCudaGraphs LANGUAGES CUDA CXX)

+# Disable response file for libraries on QNX as qcc does not support lib paths with double quotes
+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    set(CMAKE_CUDA_USE_RESPONSE_FILE_FOR_LIBRARIES OFF)
+endif()
+
 find_package(CUDAToolkit REQUIRED)

 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
--- 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 ${PROJECT_NAME} - not supported on QNX")
+    return()
+endif()
+
 # Source file
 # Add target for jitLto
 add_executable(jitLto jitLto.cpp)
--- a/Samples/4_CUDA_Libraries/simpleCUFFT/CMakeLists.txt
+++ b/Samples/4_CUDA_Libraries/simpleCUFFT/CMakeLists.txt
@ -4,6 +4,11 @@ list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/Module

 project(simpleCUFFT LANGUAGES CUDA)

+# Disable response file for libraries on QNX as qcc does not support lib paths with double quotes
+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    set(CMAKE_CUDA_USE_RESPONSE_FILE_FOR_LIBRARIES OFF)
+endif()
+
 find_package(CUDAToolkit REQUIRED)

 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
--- a/Samples/4_CUDA_Libraries/simpleCUFFT_2d_MGPU/CMakeLists.txt
+++ b/Samples/4_CUDA_Libraries/simpleCUFFT_2d_MGPU/CMakeLists.txt
@ -4,6 +4,11 @@ list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/Module

 project(simpleCUFFT_2d_MGPU LANGUAGES CUDA)

+# Disable response file for libraries on QNX as qcc does not support lib paths with double quotes
+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    set(CMAKE_CUDA_USE_RESPONSE_FILE_FOR_LIBRARIES OFF)
+endif()
+
 find_package(CUDAToolkit REQUIRED)

 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
--- a/Samples/4_CUDA_Libraries/simpleCUFFT_MGPU/CMakeLists.txt
+++ b/Samples/4_CUDA_Libraries/simpleCUFFT_MGPU/CMakeLists.txt
@ -4,6 +4,11 @@ list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/Module

 project(simpleCUFFT_MGPU LANGUAGES CUDA)

+# Disable response file for libraries on QNX as qcc does not support lib paths with double quotes
+if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
+    set(CMAKE_CUDA_USE_RESPONSE_FILE_FOR_LIBRARIES OFF)
+endif()
+
 find_package(CUDAToolkit REQUIRED)

 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
--- 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/Samples/7_libNVVM/README.md
+++ b/Samples/7_libNVVM/README.md
@ -58,6 +58,14 @@ Alternatively, we provide a Makefile that will automatically build these
 samples on Linux as part of the toplevel cuda-samples build.  Windows users
 should build manually via utils/built.bat or Visual Studio's CMake integration.

+To build and run the libNVVM samples with CUDA Toolkits 13.0 or newer, user-mode
+driver 580 and newer, old kernel-mode drivers (version 550 or earlier)，youm must
+update the CMake invocation in utils/build.sh or build.bat by adding
+`CMAKE_PREFIX_PATH` with a stubs path as follows:
+```
+-DCMAKE_PREFIX_PATH=/usr/local/cuda/lib64/stubs/
+```
+
 A Note About the cuda-c-linking Sample
 --------------------------------------

--- a/cmake/toolchains/toolchain-aarch64-qnx.cmake
+++ b/cmake/toolchains/toolchain-aarch64-qnx.cmake
@ -1,29 +1,13 @@
-#==============================================================================
-# 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 ..
-#------------------------------------------------------------------------------
+# Need to set the QNX_HOST and QNX_TARGET environment variables
+set(QNX_HOST $ENV{QNX_HOST})
+set(QNX_TARGET $ENV{QNX_TARGET})
+
+message(STATUS "QNX_HOST = ${QNX_HOST}")
+message(STATUS "QNX_TARGET = ${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")

@ -31,50 +15,23 @@ 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   "${QNX_QCC}")
-set(CMAKE_CXX_COMPILER "${QNX_QPLUS}")
+# Specify the cross-compilers
+set(CMAKE_C_COMPILER ${QNX_QCC})
+set(CMAKE_CXX_COMPILER ${QNX_QPLUS})

-#----------------------------------------------------------------------------
-# 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_C_COMPILER_TARGET aarch64)
+set(CMAKE_CXX_COMPILER_TARGET aarch64)

-#----------------------------------------------------------------------------
-# Additional preprocessor definitions & include paths
-#----------------------------------------------------------------------------
-add_compile_options(
-    -D_QNX_SOURCE
-    -DWIN_INTERFACE_CUSTOM
-)
+# Set compiler flags
+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")

-# Add an include path to /usr/include/aarch64-qnx-gnu:
-include_directories("/usr/include/aarch64-qnx-gnu")
+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")

-#----------------------------------------------------------------------------
-# 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")