NASP/cuda-samples
85
0
Fork 0
mirror of https://github.com/NVIDIA/cuda-samples.git synced 2026-10-23 13:18:16 +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

BIN
Common/FreeImage/lib/linux/sbsa/libfreeimage.a Normal file
View File

Binary file not shown.

16
Common/helper_cuda.h
View File

@@ -663,6 +663,7 @@ inline int _ConvertSMVer2Cores(int major, int minor) {
{0x70, 64},
{0x72, 64},
{0x75, 64},
{0x80, 64},
{-1, -1}};
int index = 0;
@@ -707,6 +708,7 @@ inline const char* _ConvertSMVer2ArchName(int major, int minor) {
{0x70, "Volta"},
{0x72, "Xavier"},
{0x75, "Turing"},
{0x80, "Ampere"},
{-1, "Graphics Device"}};
int index = 0;
@@ -817,7 +819,19 @@ inline int gpuGetMaxGflopsDeviceId() {
}
int multiProcessorCount = 0, clockRate = 0;
checkCudaErrors(cudaDeviceGetAttribute(&multiProcessorCount, cudaDevAttrMultiProcessorCount, current_device));
checkCudaErrors(cudaDeviceGetAttribute(&clockRate, cudaDevAttrClockRate, current_device));
cudaError_t result = cudaDeviceGetAttribute(&clockRate, cudaDevAttrClockRate, current_device);
if (result != cudaSuccess) {
// If cudaDevAttrClockRate attribute is not supported we
// set clockRate as 1, to consider GPU with most SMs and CUDA Cores.
if(result == cudaErrorInvalidValue) {
clockRate = 1;
}
else {
fprintf(stderr, "CUDA error at %s:%d code=%d(%s) \n", __FILE__, __LINE__,
static_cast<unsigned int>(result), _cudaGetErrorEnum(result));
exit(EXIT_FAILURE);
}
}
uint64_t compute_perf = (uint64_t)multiProcessorCount * sm_per_multiproc * clockRate;
if (compute_perf > max_compute_perf) {

35
Common/helper_multiprocess.cpp
View File

@@ -475,38 +475,19 @@ int ipcSendData(HANDLE mailslot, const void *data, size_t sz) {
}
int ipcRecvData(ipcHandle *handle, void *data, size_t sz) {
DWORD cbMessage, cMessage, cbRead;
BOOL fResult;
DWORD cbRead = 0;
cbMessage = cMessage = cbRead = 0;
HANDLE mailslot = handle->hMailslot[0];
pollMailSlot:
fResult = GetMailslotInfo(mailslot, (LPDWORD)NULL, &cbMessage, &cMessage,
(LPDWORD)NULL);
if (!fResult) {
printf("IPC failure: GetMailslotInfo failed with %d.\n", GetLastError());
if (!ReadFile(handle->hMailslot[0], data, (DWORD)sz, &cbRead, NULL)) {
printf("IPC failure: ReadFile failed with %d.\n", GetLastError());
return -1;
}
if (cbMessage == MAILSLOT_NO_MESSAGE) {
goto pollMailSlot;
if (sz != (size_t)cbRead) {
printf(
"IPC failure: ReadFile didn't receive the expected number of bytes\n");
return -1;
}
while (cMessage != 0) {
fResult = ReadFile(mailslot, data, (DWORD)sz, &cbRead, NULL);
if (!fResult) {
printf("IPC failure: ReadFile failed with %d.\n", GetLastError());
return -1;
}
fResult = GetMailslotInfo(mailslot, (LPDWORD)NULL, &cbMessage, &cMessage,
(LPDWORD)NULL);
if (!fResult) {
printf("IPC failure: GetMailslotInfo failed (%d)\n", GetLastError());
return -1;
}
}
return 0;
}
@@ -530,7 +511,7 @@ int ipcSendShareableHandles(
printf("IPC failure: DuplicateHandle failed (%d)\n", GetLastError());
return -1;
}
checkIpcErrors(ipcSendData(handle->hMailslot[i], &hDup, sizeof(HANDLE)));
checkIpcErrors(ipcSendData(handle->hMailslot[i], &hDup, sizeof(hDup)));
}
CloseHandle(hProcess);
}

42
Common/nvrtc_helper.h
View File

@@ -68,7 +68,37 @@ void compileFileToPTX(char *filename, int argc, char **argv, char **ptxResult,
int numCompileOptions = 0;
char *compileParams[1];
char *compileParams[2];
int major = 0, minor = 0;
char deviceName[256];
// Picks the best CUDA device available
CUdevice cuDevice = findCudaDeviceDRV(argc, (const char **)argv);
// get compute capabilities and the devicename
checkCudaErrors(cuDeviceGetAttribute(
&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice));
checkCudaErrors(cuDeviceGetAttribute(
&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice));
{
// Compile for the GPU arch on which are going to run cuda kernel.
std::string compileOptions;
compileOptions = "--gpu-architecture=compute_";
compileParams[numCompileOptions] = reinterpret_cast<char *>(
malloc(sizeof(char) * (compileOptions.length() + 10)));
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
sprintf_s(compileParams[numCompileOptions], sizeof(char) * (compileOptions.length() + 10),
"%s%d%d", compileOptions.c_str(), major, minor);
#else
snprintf(compileParams[numCompileOptions], compileOptions.size() + 10, "%s%d%d",
compileOptions.c_str(), major, minor);
#endif
}
numCompileOptions++;
if (requiresCGheaders) {
std::string compileOptions;
@@ -92,13 +122,13 @@ void compileFileToPTX(char *filename, int argc, char **argv, char **ptxResult,
argv[0]);
}
compileOptions += path.c_str();
compileParams[0] = reinterpret_cast<char *>(
compileParams[numCompileOptions] = reinterpret_cast<char *>(
malloc(sizeof(char) * (compileOptions.length() + 1)));
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
sprintf_s(compileParams[0], sizeof(char) * (compileOptions.length() + 1),
sprintf_s(compileParams[numCompileOptions], sizeof(char) * (compileOptions.length() + 1),
"%s", compileOptions.c_str());
#else
snprintf(compileParams[0], compileOptions.size(), "%s",
snprintf(compileParams[numCompileOptions], compileOptions.size(), "%s",
compileOptions.c_str());
#endif
numCompileOptions++;
@@ -137,7 +167,9 @@ void compileFileToPTX(char *filename, int argc, char **argv, char **ptxResult,
*ptxResult = ptx;
*ptxResultSize = ptxSize;
if (requiresCGheaders) free(compileParams[0]);
for (int i = 0; i < numCompileOptions; i++) {
free(compileParams[i]);
}
}
CUmodule loadPTX(char *ptx, int argc, char **argv) {