add and update samples with CUDA 11.3 support

2026-10-23 21:28:17 +08:00 · 2021-04-16 11:54:26 +05:30
parent 067cb65523
commit 568b39bd5b
214 changed files with 6590 additions and 3856 deletions
--- a/Samples/conjugateGradientCudaGraphs/Makefile
+++ b/Samples/conjugateGradientCudaGraphs/Makefile
@@ -265,6 +265,12 @@ ALL_CCFLAGS += $(addprefix -Xcompiler ,$(EXTRA_CCFLAGS))

 SAMPLE_ENABLED := 1

+# This sample is not supported on QNX
+ifeq ($(TARGET_OS),qnx)
+  $(info >>> WARNING - conjugateGradientCudaGraphs is not supported on QNX - waiving sample <<<)
+  SAMPLE_ENABLED := 0
+endif
+
 ALL_LDFLAGS :=
 ALL_LDFLAGS += $(ALL_CCFLAGS)
 ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))
--- a/Samples/conjugateGradientCudaGraphs/README.md
+++ b/Samples/conjugateGradientCudaGraphs/README.md
@@ -30,7 +30,7 @@ cudaStreamBeginCapture, cudaStreamEndCapture, cudaGraphCreate, cudaGraphLaunch,

 ## Prerequisites

-Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.

 ## Build and Run
--- a/Samples/conjugateGradientCudaGraphs/conjugateGradientCudaGraphs.cu
+++ b/Samples/conjugateGradientCudaGraphs/conjugateGradientCudaGraphs.cu
@@ -25,7 +25,6 @@
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

-
 /*
 * This sample implements a conjugate gradient solver on GPU
 * using CUBLAS and CUSPARSE with CUDA Graphs
@@ -46,7 +45,6 @@
 #include <helper_cuda.h>  // helper function CUDA error checking and initialization
 #include <helper_functions.h>  // helper for shared functions common to CUDA Samples

-
 const char *sSDKname = "conjugateGradientCudaGraphs";

 #ifndef WITH_GRAPH
@@ -145,12 +143,12 @@ int main(int argc, char **argv) {
  /* Generate a random tridiagonal symmetric matrix in CSR format */
  N = 1048576;
  nz = (N - 2) * 3 + 4;
-  I = (int *)malloc(sizeof(int) * (N + 1));
-  J = (int *)malloc(sizeof(int) * nz);
-  val = (float *)malloc(sizeof(float) * nz);
+  checkCudaErrors(cudaMallocHost(&I, sizeof(int) * (N + 1)));
+  checkCudaErrors(cudaMallocHost(&J, sizeof(int) * nz));
+  checkCudaErrors(cudaMallocHost(&val, sizeof(float) * nz));
  genTridiag(I, J, val, N, nz);

-  x = (float *)malloc(sizeof(float) * N);
+  checkCudaErrors(cudaMallocHost(&x, sizeof(float) * N));
  rhs = (float *)malloc(sizeof(float) * N);

  for (int i = 0; i < N; i++) {
@@ -192,9 +190,9 @@ int main(int argc, char **argv) {

  /* Wrap raw data into cuSPARSE generic API objects */
  cusparseSpMatDescr_t matA = NULL;
-  checkCudaErrors(cusparseCreateCsr(
-      &matA, N, N, nz, d_row, d_col, d_val, CUSPARSE_INDEX_32I,
-      CUSPARSE_INDEX_32I, CUSPARSE_INDEX_BASE_ZERO, CUDA_R_32F));
+  checkCudaErrors(cusparseCreateCsr(&matA, N, N, nz, d_row, d_col, d_val,
+                                    CUSPARSE_INDEX_32I, CUSPARSE_INDEX_32I,
+                                    CUSPARSE_INDEX_BASE_ZERO, CUDA_R_32F));
  cusparseDnVecDescr_t vecx = NULL;
  checkCudaErrors(cusparseCreateDnVec(&vecx, N, d_x, CUDA_R_32F));
  cusparseDnVecDescr_t vecp = NULL;
@@ -206,7 +204,7 @@ int main(int argc, char **argv) {
  size_t bufferSize = 0;
  checkCudaErrors(cusparseSpMV_bufferSize(
      cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, matA, vecx,
-      &beta, vecAx, CUDA_R_32F, CUSPARSE_MV_ALG_DEFAULT, &bufferSize));
+      &beta, vecAx, CUDA_R_32F, CUSPARSE_SPMV_ALG_DEFAULT, &bufferSize));
  void *buffer = NULL;
  checkCudaErrors(cudaMalloc(&buffer, bufferSize));

@@ -234,9 +232,9 @@ int main(int argc, char **argv) {
  beta = 0.0;

  checkCudaErrors(cusparseSetStream(cusparseHandle, stream1));
-  checkCudaErrors(cusparseSpMV(
-    cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, matA, vecx,
-    &beta, vecAx, CUDA_R_32F, CUSPARSE_MV_ALG_DEFAULT, &buffer));
+  checkCudaErrors(cusparseSpMV(cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE,
+                               &alpha, matA, vecx, &beta, vecAx, CUDA_R_32F,
+                               CUSPARSE_SPMV_ALG_DEFAULT, buffer));

  checkCudaErrors(cublasSetStream(cublasHandle, stream1));
  checkCudaErrors(cublasSaxpy(cublasHandle, N, &alpham1, d_Ax, 1, d_r, 1));
@@ -248,9 +246,9 @@ int main(int argc, char **argv) {
  k = 1;
  // First Iteration when k=1 starts
  checkCudaErrors(cublasScopy(cublasHandle, N, d_r, 1, d_p, 1));
-  checkCudaErrors(cusparseSpMV(
-    cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, matA, vecp,
-    &beta, vecAx, CUDA_R_32F, CUSPARSE_MV_ALG_DEFAULT, &buffer));
+  checkCudaErrors(cusparseSpMV(cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE,
+                               &alpha, matA, vecp, &beta, vecAx, CUDA_R_32F,
+                               CUSPARSE_SPMV_ALG_DEFAULT, buffer));

  checkCudaErrors(cublasSdot(cublasHandle, N, d_p, 1, d_Ax, 1, d_dot));

@@ -290,9 +288,9 @@ int main(int argc, char **argv) {

  checkCudaErrors(
      cusparseSetPointerMode(cusparseHandle, CUSPARSE_POINTER_MODE_HOST));
-  checkCudaErrors(cusparseSpMV(
-    cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, matA, vecp,
-    &beta, vecAx, CUDA_R_32F, CUSPARSE_MV_ALG_DEFAULT, &buffer));
+  checkCudaErrors(cusparseSpMV(cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE,
+                               &alpha, matA, vecp, &beta, vecAx, CUDA_R_32F,
+                               CUSPARSE_SPMV_ALG_DEFAULT, buffer));

  checkCudaErrors(cudaMemsetAsync(d_dot, 0, sizeof(float), stream1));
  checkCudaErrors(cublasSdot(cublasHandle, N, d_p, 1, d_Ax, 1, d_dot));
@@ -335,8 +333,8 @@ int main(int argc, char **argv) {
    checkCudaErrors(cublasSaxpy(cublasHandle, N, &alpha, d_r, 1, d_p, 1));

    checkCudaErrors(cusparseSpMV(
-      cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, matA, vecp,
-      &beta, vecAx, CUDA_R_32F, CUSPARSE_MV_ALG_DEFAULT, &buffer));
+        cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, matA, vecp,
+        &beta, vecAx, CUDA_R_32F, CUSPARSE_SPMV_ALG_DEFAULT, buffer));

    cublasSetPointerMode(cublasHandle, CUBLAS_POINTER_MODE_DEVICE);
    checkCudaErrors(cublasSdot(cublasHandle, N, d_p, 1, d_Ax, 1, d_dot));
@@ -395,23 +393,31 @@ int main(int argc, char **argv) {
  cusparseDestroy(cusparseHandle);
  cublasDestroy(cublasHandle);

-  if (matA       ) { checkCudaErrors(cusparseDestroySpMat(matA)); }
-  if (vecx       ) { checkCudaErrors(cusparseDestroyDnVec(vecx)); }
-  if (vecAx      ) { checkCudaErrors(cusparseDestroyDnVec(vecAx)); }
-  if (vecp       ) { checkCudaErrors(cusparseDestroyDnVec(vecp)); }
+  if (matA) {
+    checkCudaErrors(cusparseDestroySpMat(matA));
+  }
+  if (vecx) {
+    checkCudaErrors(cusparseDestroyDnVec(vecx));
+  }
+  if (vecAx) {
+    checkCudaErrors(cusparseDestroyDnVec(vecAx));
+  }
+  if (vecp) {
+    checkCudaErrors(cusparseDestroyDnVec(vecp));
+  }

-  free(I);
-  free(J);
-  free(val);
-  free(x);
+  checkCudaErrors(cudaFreeHost(I));
+  checkCudaErrors(cudaFreeHost(J));
+  checkCudaErrors(cudaFreeHost(val));
+  checkCudaErrors(cudaFreeHost(x));
  free(rhs);
-  cudaFree(d_col);
-  cudaFree(d_row);
-  cudaFree(d_val);
-  cudaFree(d_x);
-  cudaFree(d_r);
-  cudaFree(d_p);
-  cudaFree(d_Ax);
+  checkCudaErrors(cudaFree(d_col));
+  checkCudaErrors(cudaFree(d_row));
+  checkCudaErrors(cudaFree(d_val));
+  checkCudaErrors(cudaFree(d_x));
+  checkCudaErrors(cudaFree(d_r));
+  checkCudaErrors(cudaFree(d_p));
+  checkCudaErrors(cudaFree(d_Ax));

  printf("Test Summary:  Error amount = %f\n", err);
  exit((k <= max_iter) ? 0 : 1);
--- a/Samples/conjugateGradientCudaGraphs/conjugateGradientCudaGraphs_vs2017.vcxproj
+++ b/Samples/conjugateGradientCudaGraphs/conjugateGradientCudaGraphs_vs2017.vcxproj
@@ -38,7 +38,7 @@
  </PropertyGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
  <ImportGroup Label="ExtensionSettings">
-    <Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
+    <Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
  </ImportGroup>
  <ImportGroup Label="PropertySheets">
    <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -108,6 +108,6 @@
  </ItemGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
  <ImportGroup Label="ExtensionTargets">
-    <Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
+    <Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
  </ImportGroup>
 </Project>
--- a/Samples/conjugateGradientCudaGraphs/conjugateGradientCudaGraphs_vs2019.vcxproj
+++ b/Samples/conjugateGradientCudaGraphs/conjugateGradientCudaGraphs_vs2019.vcxproj
@@ -34,7 +34,7 @@
  </PropertyGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
  <ImportGroup Label="ExtensionSettings">
-    <Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
+    <Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
  </ImportGroup>
  <ImportGroup Label="PropertySheets">
    <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -104,6 +104,6 @@
  </ItemGroup>
  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
  <ImportGroup Label="ExtensionTargets">
-    <Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
+    <Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
  </ImportGroup>
 </Project>