NASP/cuda-samples
85
0
Fork 0
mirror of https://github.com/NVIDIA/cuda-samples.git synced 2026-10-23 21:28:17 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add and update samples for cuda 11.0 support

Browse Source
This commit is contained in:
Mahesh Doijade
2020-05-18 22:22:06 +05:30
parent 6be514679b
commit 4f6e02970b
410 changed files with 21920 additions and 3864 deletions
Download Patch File Download Diff File Expand all files Collapse all files

68
Samples/memMapIPCDrv/Makefile
View File

@@ -72,9 +72,9 @@ endif
# architecture
HOST_ARCH := $(shell uname -m)
TARGET_ARCH ?= $(HOST_ARCH)
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le armv7l))
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le))
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le))
TARGET_SIZE := 64
else ifneq (,$(filter $(TARGET_ARCH),armv7l))
TARGET_SIZE := 32
@@ -85,8 +85,17 @@ ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l))
else
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
endif
# sbsa and aarch64 systems look similar. Need to differentiate them at host level for now.
ifeq ($(HOST_ARCH),aarch64)
ifeq ($(CUDA_PATH)/targets/sbsa-linux,$(shell ls -1d $(CUDA_PATH)/targets/sbsa-linux))
HOST_ARCH := sbsa
TARGET_ARCH := sbsa
endif
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-ppc64le))
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-sbsa x86_64-ppc64le))
$(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
endif
endif
@@ -141,6 +150,8 @@ else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
else ifeq ($(TARGET_OS), android)
HOST_COMPILER ?= aarch64-linux-android-clang++
endif
else ifeq ($(TARGET_ARCH),sbsa)
HOST_COMPILER ?= aarch64-linux-gnu-g++
else ifeq ($(TARGET_ARCH),ppc64le)
HOST_COMPILER ?= powerpc64le-linux-gnu-g++
endif
@@ -185,19 +196,27 @@ ifneq ($(TARGET_ARCH),$(HOST_ARCH))
CCFLAGS += --sysroot=$(TARGET_FS)
endif
LDFLAGS += --sysroot=$(TARGET_FS)
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L $(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L $(TARGET_FS)/usr/lib/aarch64-linux-gnu
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L$(TARGET_FS)/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/lib/aarch64-linux-gnu -L$(TARGET_FS)/lib/aarch64-linux-gnu
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L$(TARGET_FS)/usr/lib
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L$(TARGET_FS)/usr/lib/aarch64-linux-gnu
LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
CCFLAGS += -isystem=$(TARGET_FS)/usr/include
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu
CCFLAGS += -isystem=$(TARGET_FS)/usr/include -I$(TARGET_FS)/usr/include
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu -I$(TARGET_FS)/usr/include/aarch64-linux-gnu
endif
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM -I/usr/include/aarch64-qnx-gnu
LDFLAGS += -lsocket
LDFLAGS += -rpath=/usr/lib/aarch64-qnx-gnu -L/usr/lib/aarch64-qnx-gnu
ifneq ($(TARGET_FS),)
LDFLAGS += -rpath=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
LDFLAGS += -rpath=$(TARGET_FS)/usr/libnvidia -L $(TARGET_FS)/usr/libnvidia
endif
ifdef TARGET_OVERRIDE # cuda toolkit targets override
NVCCFLAGS += -target-dir $(TARGET_OVERRIDE)
endif
endif
endif
ifeq ($(TARGET_OS),qnx)
CCFLAGS += -DWIN_INTERFACE_CUSTOM
LDFLAGS += -lsocket
endif
# Install directory of different arch
@@ -206,6 +225,8 @@ ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),sbsa-linux)
CUDA_INSTALL_TARGET_DIR = targets/sbsa-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
@@ -254,6 +275,12 @@ ifeq ($(TARGET_ARCH),aarch64)
SAMPLE_ENABLED := 0
endif
# This sample is not supported on sbsa
ifeq ($(TARGET_ARCH),sbsa)
$(info >>> WARNING - memMapIPCDrv is not supported on sbsa - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))
@@ -275,8 +302,8 @@ ifeq ($(GENCODE_FLAGS),)
$(foreach sm,$(SMS),$(eval GENCODE_FLAGS += -gencode arch=compute_$(sm),code=sm_$(sm)))
ifeq ($(SMS),)
# Generate PTX code from SM 30
GENCODE_FLAGS += -gencode arch=compute_30,code=compute_30
# Generate PTX code from SM 35
GENCODE_FLAGS += -gencode arch=compute_35,code=compute_35
endif
# Generate PTX code from the highest SM architecture in $(SMS) to guarantee forward-compatibility
@@ -302,6 +329,10 @@ else
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/aarch64-linux/lib/stubs
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),sbsa-linux)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/sbsa-linux/lib/stubs
endif
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/armv7-linux-androideabi/lib/stubs
endif
@@ -316,12 +347,19 @@ else
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/aarch64-qnx/lib/stubs
ifdef TARGET_OVERRIDE
CUDA_SEARCH_PATH := $(CUDA_PATH)/targets/$(TARGET_OVERRIDE)/lib/stubs
endif
endif
ifeq ($(TARGET_ARCH),ppc64le)
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/ppc64le-linux/lib/stubs
endif
ifeq ($(HOST_ARCH),ppc64le)
CUDA_SEARCH_PATH += $(CUDA_PATH)/lib64/stubs
endif
CUDALIB ?= $(shell find -L $(CUDA_SEARCH_PATH) -maxdepth 1 -name libcuda.so 2> /dev/null)
ifeq ("$(CUDALIB)","")
$(info >>> WARNING - libcuda.so not found, CUDA Driver is not installed. Please re-install the driver. <<<)

4
Samples/memMapIPCDrv/README.md
View File

@@ -10,7 +10,7 @@ CUDA Driver API, cuMemMap IPC, MMAP
## Supported SM Architectures
[SM 3.0 ](https://developer.nvidia.com/cuda-gpus) [SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus)
[SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus) [SM 8.0 ](https://developer.nvidia.com/cuda-gpus)
## Supported OSes
@@ -30,7 +30,7 @@ cuModuleLoad, cuModuleLoadDataEx, cuModuleGetFunction, cuLaunchKernel, cuMemcpyD
## Prerequisites
Download and install the [CUDA Toolkit 10.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.0](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

2
Samples/memMapIPCDrv/memMapIPCDrv_vs2012.vcxproj
View File

@@ -62,7 +62,7 @@
<OutputFile>$(OutDir)/memMapIPCDrv.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,compute_30;</CodeGeneration>
<CodeGeneration>compute_35,compute_35;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>

2
Samples/memMapIPCDrv/memMapIPCDrv_vs2013.vcxproj
View File

@@ -62,7 +62,7 @@
<OutputFile>$(OutDir)/memMapIPCDrv.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,compute_30;</CodeGeneration>
<CodeGeneration>compute_35,compute_35;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>

2
Samples/memMapIPCDrv/memMapIPCDrv_vs2015.vcxproj
View File

@@ -62,7 +62,7 @@
<OutputFile>$(OutDir)/memMapIPCDrv.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,compute_30;</CodeGeneration>
<CodeGeneration>compute_35,compute_35;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>

2
Samples/memMapIPCDrv/memMapIPCDrv_vs2017.vcxproj
View File

@@ -67,7 +67,7 @@
<OutputFile>$(OutDir)/memMapIPCDrv.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,compute_30;</CodeGeneration>
<CodeGeneration>compute_35,compute_35;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>

2
Samples/memMapIPCDrv/memMapIPCDrv_vs2019.vcxproj
View File

@@ -63,7 +63,7 @@
<OutputFile>$(OutDir)/memMapIPCDrv.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_30,compute_30;</CodeGeneration>
<CodeGeneration>compute_35,compute_35;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>

30
Samples/memMapIPCDrv/memMapIpc.cpp
View File

@@ -302,7 +302,7 @@ static void memMapGetDeviceFunction(char **argv) {
cuModuleGetFunction(&_memMapIpc_kernel, cuModule, "memMapIpc_kernel"));
}
static void childProcess(int id, char **argv) {
static void childProcess(int devId, int id, char **argv) {
volatile shmStruct *shm = NULL;
sharedMemoryInfo info;
ipcHandle *ipcChildHandle = NULL;
@@ -329,7 +329,7 @@ static void childProcess(int id, char **argv) {
CUstream stream;
int multiProcessorCount;
checkCudaErrors(cuDeviceGet(&device, id));
checkCudaErrors(cuDeviceGet(&device, devId));
checkCudaErrors(cuCtxCreate(&ctx, 0, device));
checkCudaErrors(cuStreamCreate(&stream, CU_STREAM_NON_BLOCKING));
@@ -350,7 +350,7 @@ static void childProcess(int id, char **argv) {
// Import the memory allocations shared by the parent with us and map them in
// our address space.
memMapImportAndMapMemory(d_ptr, DATA_BUF_SIZE, shHandle, id);
memMapImportAndMapMemory(d_ptr, DATA_BUF_SIZE, shHandle, devId);
// Since we have imported allocations shared by the parent with us, we can
// close all the ShareableHandles.
@@ -424,7 +424,6 @@ static void parentProcess(char *app) {
checkCudaErrors(cuDeviceGetCount(&devCount));
std::vector<CUdevice> devices(devCount);
std::vector<CUcontext> ctxs(devCount);
if (sharedMemoryCreate(shmName, sizeof(*shm), &info) != 0) {
printf("Failed to create shared memory slab\n");
@@ -436,9 +435,11 @@ static void parentProcess(char *app) {
for (i = 0; i < devCount; i++) {
checkCudaErrors(cuDeviceGet(&devices[i], i));
checkCudaErrors(cuCtxCreate(&ctxs[i], 0, devices[i]));
}
std::vector<CUcontext> ctxs;
std::vector<unsigned char> selectedDevices;
// Pick all the devices that can access each other's memory for this test
// Keep in mind that CUDA has minimal support for fork() without a
// corresponding exec() in the child process, but in this case our
@@ -500,6 +501,10 @@ static void parentProcess(char *app) {
}
}
if (allPeers) {
CUcontext ctx;
checkCudaErrors(cuCtxCreate(&ctx, 0, devices[i]));
ctxs.push_back(ctx);
// Enable peers here. This isn't necessary for IPC, but it will
// setup the peers for the device. For systems that only allow 8
// peers per GPU at a time, this acts to remove devices from CanAccessPeer
@@ -509,6 +514,7 @@ static void parentProcess(char *app) {
checkCudaErrors(cuCtxSetCurrent(ctxs[j]));
checkCudaErrors(cuCtxEnablePeerAccess(ctxs[i], 0));
}
selectedDevices.push_back(i);
nprocesses++;
if (nprocesses >= MAX_DEVICES) {
break;
@@ -521,13 +527,17 @@ static void parentProcess(char *app) {
}
}
for (int i = 0; i < ctxs.size(); ++i) {
checkCudaErrors(cuCtxDestroy(ctxs[i]));
};
if (nprocesses == 0) {
printf("No CUDA devices support IPC\n");
exit(EXIT_WAIVED);
}
shm->nprocesses = nprocesses;
unsigned char firstSelectedDevice = 0;
unsigned char firstSelectedDevice = selectedDevices[0];
std::vector<ShareableHandle> shHandles(nprocesses);
std::vector<CUmemGenericAllocationHandle> allocationHandles(nprocesses);
@@ -540,10 +550,12 @@ static void parentProcess(char *app) {
// Launch the child processes!
for (i = 0; i < nprocesses; i++) {
char devIdx[10];
char *const args[] = {app, devIdx, NULL};
char procIdx[10];
char *const args[] = {app, devIdx, procIdx, NULL};
Process process;
SPRINTF(devIdx, "%d", i);
SPRINTF(devIdx, "%d", selectedDevices[i]);
SPRINTF(procIdx, "%d", i);
if (spawnProcess(&process, app, args)) {
printf("Failed to create process\n");
@@ -593,7 +605,7 @@ int main(int argc, char **argv) {
if (argc == 1) {
parentProcess(argv[0]);
} else {
childProcess(atoi(argv[1]), argv);
childProcess(atoi(argv[1]), atoi(argv[2]), argv);
}
return EXIT_SUCCESS;
#endif