Merge pull request #268 from daineAMD/master

Merge develop into master for ROCm 3.10 feature complete

.jenkins/multicompiler.groovy

@@ -98,6 +98,5 @@ ci: {
     String clangBuildCommand = './install.sh -c --compiler=clang++'
 
     setupCI(urlJobName, jobNameList, hostBuildCommand, runCI, 'g++')
-    setupCI(urlJobName, jobNameList, clangBuildCommand, runCI, 'clang++')
     setupCI(urlJobName, jobNameList, hipClangBuildCommand, runCI, 'hip-clang')
 }

CMakeLists.txt

@@ -22,6 +22,10 @@ if( NOT DEFINED CMAKE_CONFIGURATION_TYPES AND NOT DEFINED CMAKE_BUILD_TYPE )
   set( CMAKE_BUILD_TYPE Release CACHE STRING "Choose the type of build, options are: None Debug Release RelWithDebInfo MinSizeRel." )
 endif()
 
+if ( NOT DEFINED CMAKE_Fortran_COMPILER AND NOT DEFINED ENV{FC} )
+  set( CMAKE_Fortran_COMPILER  "gfortran" )
+endif()
+
 project( hipblas LANGUAGES CXX Fortran )
 
 # This finds the rocm-cmake project, and installs it if not found
@@ -45,7 +49,7 @@ include( ROCMInstallTargets )
 include( ROCMPackageConfigHelpers )
 include( ROCMInstallSymlinks )
 
-set ( VERSION_STRING "0.36.0" )
+set ( VERSION_STRING "0.38.0" )
 rocm_setup_version( VERSION ${VERSION_STRING} )
 
 # Append our library helper cmake path and the cmake path for hip (for convenience)
@@ -84,8 +88,8 @@ message("amdhip64: ${LIBAMDHIP64_LIBRARY}")
 
 # Quietly look for CUDA, but if not found it's not an error
 # The presense of hip is not sufficient to determine if we want a rocm or cuda backend
-option(TRY_CUDA "Look for CUDA and use that as a backend if found" ON)
-if(TRY_CUDA)
+option(USE_CUDA "Look for CUDA and use that as a backend if found" ON)
+if(USE_CUDA)
     find_package( CUDA QUIET )
 endif()
 

bump_develop_version.sh

@@ -5,7 +5,7 @@
 # - run this script in master branch
 # - after running this script merge master into develop
 
-OLD_HIPBLAS_VERSION="0.36.0"
-NEW_HIPBLAS_VERSION="0.37.0"
+OLD_HIPBLAS_VERSION="0.38.0"
+NEW_HIPBLAS_VERSION="0.39.0"
 
 sed -i "s/${OLD_HIPBLAS_VERSION}/${NEW_HIPBLAS_VERSION}/g" CMakeLists.txt

bump_master_version.sh

@@ -6,7 +6,7 @@
 # - after running this script and merging develop into master, run bump_develop_version.sh in master and
 #   merge master into develop
 
-OLD_HIPBLAS_VERSION="0.35.0"
-NEW_HIPBLAS_VERSION="0.36.0"
+OLD_HIPBLAS_VERSION="0.37.0"
+NEW_HIPBLAS_VERSION="0.38.0"
 
 sed -i "s/${OLD_HIPBLAS_VERSION}/${NEW_HIPBLAS_VERSION}/g" CMakeLists.txt

clients/gtest/CMakeLists.txt

@@ -42,6 +42,7 @@ set(hipblas_test_source
   set_get_pointer_mode_gtest.cpp
   set_get_vector_gtest.cpp
   set_get_matrix_gtest.cpp
+  set_get_atomics_mode_gtest.cpp
   blas1_gtest.cpp
   gbmv_gtest.cpp
   gbmv_batched_gtest.cpp

clients/gtest/blas1_gtest.cpp

@@ -2356,11 +2356,11 @@ TEST_P(blas1_gtest, amin_strided_batched_float_complex)
 // so each elment in xxx_range is a avector,
 // ValuesIn take each element (a vector) and combine them and feed them to test_p
 // The combinations are  { N, {alpha, beta}, {incx, incy} }
-INSTANTIATE_TEST_CASE_P(hipblasBlas1,
-                        blas1_gtest,
-                        Combine(ValuesIn(N_range),
-                                ValuesIn(alpha_beta_range),
-                                ValuesIn(incx_incy_range),
-                                ValuesIn(stride_scale_range),
-                                ValuesIn(batch_count_range),
-                                ValuesIn(is_fortran)));
+INSTANTIATE_TEST_SUITE_P(hipblasBlas1,
+                         blas1_gtest,
+                         Combine(ValuesIn(N_range),
+                                 ValuesIn(alpha_beta_range),
+                                 ValuesIn(incx_incy_range),
+                                 ValuesIn(stride_scale_range),
+                                 ValuesIn(batch_count_range),
+                                 ValuesIn(is_fortran)));

clients/gtest/dgmm_gtest.cpp

@@ -274,10 +274,10 @@ TEST_P(dgmm_gtest, dgmm_strided_batched_gtest_float_complex)
 // ValuesIn take each element (a vector) and combine them and feed them to test_p
 // The combinations are  { {M}, {incx,incy} {alpha, alphai, beta, betai}, {transA}, {stride_scale}, {batch_count} }
 
-INSTANTIATE_TEST_CASE_P(hipblasDgmm,
-                        dgmm_gtest,
-                        Combine(ValuesIn(matrix_size_range),
-                                ValuesIn(side_range),
-                                ValuesIn(stride_scale_range),
-                                ValuesIn(batch_count_range),
-                                ValuesIn(is_fortran)));
+INSTANTIATE_TEST_SUITE_P(hipblasDgmm,
+                         dgmm_gtest,
+                         Combine(ValuesIn(matrix_size_range),
+                                 ValuesIn(side_range),
+                                 ValuesIn(stride_scale_range),
+                                 ValuesIn(batch_count_range),
+                                 ValuesIn(is_fortran)));

clients/gtest/gbmv_batched_gtest.cpp

@@ -191,11 +191,11 @@ TEST_P(gbmv_gtest_batched, gbmv_gtest_float_complex)
 // ValuesIn take each element (a vector) and combine them and feed them to test_p
 // The combinations are  { {M, N, lda}, {incx,incy} {alpha, beta}, {transA}, {batch_count} }
 
-INSTANTIATE_TEST_CASE_P(hipblasGbmvBatched,
-                        gbmv_gtest_batched,
-                        Combine(ValuesIn(matrix_size_range),
-                                ValuesIn(incx_incy_range),
-                                ValuesIn(alpha_beta_range),
-                                ValuesIn(transA_range),
-                                ValuesIn(batch_count_range),
-                                ValuesIn(is_fortran)));
+INSTANTIATE_TEST_SUITE_P(hipblasGbmvBatched,
+                         gbmv_gtest_batched,
+                         Combine(ValuesIn(matrix_size_range),
+                                 ValuesIn(incx_incy_range),
+                                 ValuesIn(alpha_beta_range),
+                                 ValuesIn(transA_range),
+                                 ValuesIn(batch_count_range),
+                                 ValuesIn(is_fortran)));

clients/gtest/gbmv_gtest.cpp

@@ -193,10 +193,10 @@ TEST_P(gbmv_gtest, gbmv_gtest_float_complex)
 // ValuesIn take each element (a vector) and combine them and feed them to test_p
 // The combinations are  { {M, N, lda}, {incx,incy} {alpha, beta}, {transA} }
 
-INSTANTIATE_TEST_CASE_P(hipblasGbmv,
-                        gbmv_gtest,
-                        Combine(ValuesIn(matrix_size_range),
-                                ValuesIn(incx_incy_range),
-                                ValuesIn(alpha_beta_range),
-                                ValuesIn(transA_range),
-                                ValuesIn(is_fortran)));
+INSTANTIATE_TEST_SUITE_P(hipblasGbmv,
+                         gbmv_gtest,
+                         Combine(ValuesIn(matrix_size_range),
+                                 ValuesIn(incx_incy_range),
+                                 ValuesIn(alpha_beta_range),
+                                 ValuesIn(transA_range),
+                                 ValuesIn(is_fortran)));

clients/gtest/gbmv_strided_batched_gtest.cpp

@@ -205,12 +205,12 @@ TEST_P(gbmv_gtest_strided_batched, gbmv_gtest_float_complex)
 // ValuesIn take each element (a vector) and combine them and feed them to test_p
 // The combinations are  { {M, N, lda}, {incx,incy} {stride_scale}, {alpha, beta}, {transA}, {batch_count} }
 
-INSTANTIATE_TEST_CASE_P(hipblasGbmvStridedBatched,
-                        gbmv_gtest_strided_batched,
-                        Combine(ValuesIn(matrix_size_range),
-                                ValuesIn(incx_incy_range),
-                                ValuesIn(stride_scale_range),
-                                ValuesIn(alpha_beta_range),
-                                ValuesIn(transA_range),
-                                ValuesIn(batch_count_range),
-                                ValuesIn(is_fortran)));
+INSTANTIATE_TEST_SUITE_P(hipblasGbmvStridedBatched,
+                         gbmv_gtest_strided_batched,
+                         Combine(ValuesIn(matrix_size_range),
+                                 ValuesIn(incx_incy_range),
+                                 ValuesIn(stride_scale_range),
+                                 ValuesIn(alpha_beta_range),
+                                 ValuesIn(transA_range),
+                                 ValuesIn(batch_count_range),
+                                 ValuesIn(is_fortran)));

clients/gtest/geam_gtest.cpp

@@ -301,11 +301,11 @@ TEST_P(geam_gtest, geam_strided_batched_gtest_double_complex)
 // This function mainly test the scope of alpha_beta, transA_transB,.the scope of matrix_size_range
 // is small
 
-INSTANTIATE_TEST_CASE_P(hipblasGeam_scalar_transpose,
-                        geam_gtest,
-                        Combine(ValuesIn(matrix_size_range),
-                                ValuesIn(alpha_beta_range),
-                                ValuesIn(transA_transB_range),
-                                ValuesIn(stride_scale_range),
-                                ValuesIn(batch_count_range),
-                                ValuesIn(is_fortran)));
+INSTANTIATE_TEST_SUITE_P(hipblasGeam_scalar_transpose,
+                         geam_gtest,
+                         Combine(ValuesIn(matrix_size_range),
+                                 ValuesIn(alpha_beta_range),
+                                 ValuesIn(transA_transB_range),
+                                 ValuesIn(stride_scale_range),
+                                 ValuesIn(batch_count_range),
+                                 ValuesIn(is_fortran)));

clients/gtest/gemm_batched_gtest.cpp

@@ -305,10 +305,10 @@ TEST_P(gemm_batched_gtest, hipblasDoubleComplex)
 // The combinations are  { {M, N, K, lda, ldb, ldc}, {alpha, beta}, {transA, transB}, {batch_count}
 // }
 
-INSTANTIATE_TEST_CASE_P(hipblasGemmBatched,
-                        gemm_batched_gtest,
-                        Combine(ValuesIn(matrix_size_range),
-                                ValuesIn(alpha_beta_range),
-                                ValuesIn(transA_transB_range),
-                                ValuesIn(batch_count_range),
-                                ValuesIn(is_fortran)));
+INSTANTIATE_TEST_SUITE_P(hipblasGemmBatched,
+                         gemm_batched_gtest,
+                         Combine(ValuesIn(matrix_size_range),
+                                 ValuesIn(alpha_beta_range),
+                                 ValuesIn(transA_transB_range),
+                                 ValuesIn(batch_count_range),
+                                 ValuesIn(is_fortran)));