[SYCLomatic #1847] Add test for the migration of cublas[Set|Get]Vector[Async]_64 and cublas[Set|Get]Matrix[Async]_64 (#676)

Signed-off-by: Jiang, Zhiwei <[email protected]>

Author: zhiweij1

features/feature_case/cublas/cublas_64.cu

@@ -54,6 +54,18 @@ void foo() {
   int64_t incx;
   int64_t incy;
 
+  int64_t elemSize;
+  cudaStream_t stream;
+  status = cublasSetVector_64(n, elemSize, A_s, incx, C_s, incy);
+  status = cublasGetVector_64(n, elemSize, A_s, incx, C_s, incy);
+  status = cublasSetVectorAsync_64(n, elemSize, A_s, incx, C_s, incy, stream);
+  status = cublasGetVectorAsync_64(n, elemSize, A_s, incx, C_s, incy, stream);
+
+  status = cublasSetMatrix_64(m, n, elemSize, A_s, lda, C_s, ldb);
+  status = cublasGetMatrix_64(m, n, elemSize, A_s, lda, C_s, ldb);
+  status = cublasSetMatrixAsync_64(m, n, elemSize, A_s, lda, C_s, ldb, stream);
+  status = cublasGetMatrixAsync_64(m, n, elemSize, A_s, lda, C_s, ldb, stream);
+
   status = cublasIsamax_64(handle, n, A_s, lda, &result);
   status = cublasIdamax_64(handle, n, A_d, lda, &result);
   status = cublasIcamax_64(handle, n, A_c, lda, &result);

help_function/help_function.xml

@@ -70,6 +70,7 @@
     <test testName="blas_utils_get_value_usm" configFile="config/TEMPLATE_help_function_blas_usm.xml" splitGroup="double" />
     <test testName="blas_utils_parameter_wrapper_buf" configFile="config/TEMPLATE_help_function_blas_usmnone.xml" splitGroup="double"/>
     <test testName="blas_utils_parameter_wrapper_usm" configFile="config/TEMPLATE_help_function_blas_usm.xml" splitGroup="double"/>
+    <test testName="blas_utils_matrix_mem_copy_test" configFile="config/TEMPLATE_help_function_skip_cuda_backend.xml" />
     <test testName="util_cast_value_test" configFile="config/TEMPLATE_help_function_skip_double.xml"  splitGroup="double" />
     <test testName="util_fast_length_test" configFile="config/TEMPLATE_help_function_skip_cuda_backend.xml" />
     <test testName="util_make_index_sequence_test" configFile="config/TEMPLATE_help_function_skip_cuda_backend.xml" />

help_function/src/blas_utils_matrix_mem_copy_test.cpp

@@ -0,0 +1,94 @@
+// ===------- blas_utils_matrix_mem_copy_test.cpp ---------- *- C++ -* ---=== //
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+// ===--------------------------------------------------------------------=== //
+
+#include <sycl/sycl.hpp>
+#include <dpct/dpct.hpp>
+#include <dpct/blas_utils.hpp>
+
+const std::int64_t M = 3;
+const std::int64_t N = 2;
+
+void matrix_mem_copy_test_1() {
+  dpct::device_ext &dev_ct1 = dpct::get_current_device();
+  sycl::queue &q_ct1 = dev_ct1.default_queue();
+  float *devPtrA;
+  devPtrA = (float *)sycl::malloc_device(M * N * sizeof(float), q_ct1);
+  float host_a[6] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f};
+  float host_b[6] = {-1.0f, -1.0f, -1.0f, -1.0f, -1.0f, -1.0f};
+  float host_c[6] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f};
+
+  dpct::blas::matrix_mem_copy(devPtrA, host_a, M, M, M, N, sizeof(float),
+                              dpct::automatic, q_ct1, true);
+  dpct::blas::matrix_mem_copy(host_b, devPtrA, M, M, M, N, sizeof(float));
+
+  for (int i = 0; i < M * N; i++) {
+    if (fabs(host_b[i] - host_c[i]) > 1e-5) {
+      printf("matrix_mem_copy_test_1.1 failed\n");
+      exit(-1);
+    }
+  }
+
+  // Because to_ld == from_ld, matrix_mem_copy just do one copy.
+  // All padding data is also copied except the last padding.
+  float host_d[6] = {-2.0f, -2.0f, -2.0f, -2.0f, -2.0f, -2.0f};
+  float host_e[6] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, -2.0f};
+  dpct::blas::matrix_mem_copy(host_d, devPtrA, M /*to_ld*/, M /*from_ld*/,
+                              M - 1 /*rows*/, N /*cols*/, sizeof(float));
+
+  for (int i = 0; i < M * N; i++) {
+    if (fabs(host_d[i] - host_e[i]) > 1e-5) {
+      printf("matrix_mem_copy_test_1.2 failed\n");
+      exit(-1);
+    }
+  }
+
+  sycl::free(devPtrA, q_ct1);
+}
+
+void matrix_mem_copy_test_2() {
+  dpct::device_ext &dev_ct1 = dpct::get_current_device();
+  sycl::queue &q_ct1 = dev_ct1.default_queue();
+
+  float *devPtrA;
+  devPtrA = (float *)sycl::malloc_device(M * N * sizeof(float), q_ct1);
+  float host_a[6] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f};
+  float host_b[6] = {-1.0f, -1.0f, -1.0f, -1.0f, -1.0f, -1.0f};
+  float host_c[6] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f};
+
+  dpct::blas::matrix_mem_copy(devPtrA, host_a, M, M, M, N, sizeof(float),
+                              dpct::automatic, q_ct1, true);
+  dpct::blas::matrix_mem_copy(host_b, devPtrA, M, M, M, N, sizeof(float));
+
+  for (int i = 0; i < M * N; i++) {
+    if (fabs(host_b[i] - host_c[i]) > 1e-5) {
+      printf("matrix_mem_copy_test_2.1 failed\n");
+      exit(-1);
+    }
+  }
+
+  float host_d[4] = {-2.0f, -2.0f, -2.0f, -2.0f};
+  float host_e[4] = {1.0f, 2.0f, 4.0f, 5.0f};
+  dpct::blas::matrix_mem_copy(host_d, devPtrA, M - 1 /*to_ld*/, M /*from_ld*/,
+                              M - 1 /*rows*/, N /*cols*/, sizeof(float));
+
+  for (int i = 0; i < (M - 1) * N; i++) {
+    if (fabs(host_d[i] - host_e[i]) > 1e-5) {
+      printf("matrix_mem_copy_test_2.2 failed\n");
+      exit(-1);
+    }
+  }
+
+  sycl::free(devPtrA, q_ct1);
+}
+
+int main() {
+  matrix_mem_copy_test_1();
+  matrix_mem_copy_test_2();
+  printf("matrix_mem_copy_test passed\n");
+  return 0;
+}