Merge pull request #296 from daineAMD/master

ROCm 4.1 merge staging into master

.githooks/pre-commit

@@ -29,6 +29,7 @@ fi
 
 # Change the copyright date at the top of any text files
 for file in $files; do
+    [[ -L $file ]] && continue
     echo "Processing copyright dates in $file"
     if [[ -e $file ]]; then
         /usr/bin/perl -pi -e 'INIT { exit 1 if !-f $ARGV[0] || -B $ARGV[0]; $year = (localtime)[5] + 1900 }
@@ -39,6 +40,7 @@ done
 
 # do the formatting
 for file in $files; do
+    [[ -L $file ]] && continue
     if [[ -e $file ]] && echo $file | grep -Eq '\.c$|\.h$|\.hpp$|\.cpp$|\.cl$|\.in$|\.txt$|\.yaml$|\.yml$|\.sh$|\.py$|\.pl$|\.cmake$|\.md$|\.rst$|\.groovy$|\.ini$|\.awk$|\.csv$'; then
         echo "Processing line endings in $file"
         sed -i -e 's/[[:space:]]*$//' "$file" # Remove whitespace at end of lines
@@ -58,6 +60,7 @@ done
 # if clang-format exists, run it on C/C++ files
 if command -v clang-format >/dev/null; then
     for file in $files; do
+       [[ -L $file ]] && continue
        if [[ -e $file ]] && echo $file | grep -Eq '\.c$|\.h$|\.hpp$|\.cpp$|\.cl$|\.h\.in$|\.hpp\.in$|\.cpp\.in$'; then
             echo "clang-format $file"
             clang-format -i -style=file "$file"

CHANGELOG.md

@@ -1,5 +1,17 @@
 # Change Log for hipBLAS
 
+## [hipBLAS 0.42.0 for ROCm 4.1.0]
+### Added
+- Added the following functions. All added functions include batched and strided-batched support with rocBLAS backend:
+    - axpy_ex
+    - dot_ex
+    - nrm2_ex
+    - rot_ex
+    - scal_ex
+
+### Fixed
+- Fixed complex unit test bug caused by incorrect caxpy and zaxpy function signatures
+
 ## [hipBLAS 0.40.0 for ROCm 4.0.0]
 ### Added
 - Added changelog

CMakeLists.txt

@@ -49,12 +49,18 @@ include( ROCMInstallTargets )
 include( ROCMPackageConfigHelpers )
 include( ROCMInstallSymlinks )
 
-set ( VERSION_STRING "0.40.0" )
+set ( VERSION_STRING "0.42.0" )
 rocm_setup_version( VERSION ${VERSION_STRING} )
 
+if( NOT DEFINED $ENV{HIP_PATH})
+    set( HIP_PATH "/opt/rocm/hip" )
+else( )
+    set (HIP_PATH $ENV{HIP_PATH} )
+endif( )
+
 # Append our library helper cmake path and the cmake path for hip (for convenience)
 # Users may override HIP path by specifying their own in CMAKE_MODULE_PATH
-list( APPEND CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake )
+list( APPEND CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake  ${HIP_PATH}/cmake)
 
 # NOTE:  workaround until hip cmake modules fixes symlink logic in their config files; remove when fixed
 list( APPEND CMAKE_PREFIX_PATH /opt/rocm /opt/rocm/llvm /opt/rocm/hip )
@@ -70,17 +76,24 @@ endif( )
 # BUILD_SHARED_LIBS is a cmake built-in; we make it an explicit option such that it shows in cmake-gui
 option( BUILD_SHARED_LIBS "Build hipBLAS as a shared library" ON )
 
-
-# Hip headers required of all clients; clients use hip to allocate device memory
-find_package( hip REQUIRED CONFIG PATHS ${ROCM_PATH} /opt/rocm)
-
 # 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(USE_CUDA "Look for CUDA and use that as a backend if found" ON)
 if(USE_CUDA)
     find_package( CUDA QUIET )
 endif()
 
+# Hip headers required of all clients; clients use hip to allocate device memory
+if( USE_CUDA)
+    find_package( HIP REQUIRED )
+else( )
+    find_package(hip REQUIRED CONFIG PATHS ${ROCM_PATH} /opt/rocm)
+endif( )
+
+if( USE_CUDA )
+    list( APPEND HIP_INCLUDE_DIRS "${HIP_ROOT_DIR}/include" )
+endif( )
+
 # CMake list of machine targets
 set( AMDGPU_TARGETS gfx803;gfx900;gfx906;gfx908 CACHE STRING "List of specific machine types for library to target" )
 

bump_develop_version.sh

@@ -5,11 +5,11 @@
 # - run this script in master branch
 # - after running this script merge master into develop
 
-OLD_HIPBLAS_VERSION="0.40.0"
-NEW_HIPBLAS_VERSION="0.41.0"
+OLD_HIPBLAS_VERSION="0.42.0"
+NEW_HIPBLAS_VERSION="0.43.0"
 
-OLD_MINIMUM_ROCBLAS_VERSION="2.34.0"
-NEW_MINIMUM_ROCBLAS_VERSION="2.35.0"
+OLD_MINIMUM_ROCBLAS_VERSION="2.36.0"
+NEW_MINIMUM_ROCBLAS_VERSION="2.37.0"
 
 OLD_MINIMUM_ROCSOLVER_VERSION="3.10.0"
 NEW_MINIMUM_ROCSOLVER_VERSION="3.11.0"

bump_master_version.sh

@@ -6,11 +6,11 @@
 # - after running this script and merging develop into master, run bump_develop_version.sh in master and
 #   merge master into develop
 
-OLD_HIPBLAS_VERSION="0.39.0"
-NEW_HIPBLAS_VERSION="0.40.0"
+OLD_HIPBLAS_VERSION="0.41.0"
+NEW_HIPBLAS_VERSION="0.42.0"
 
-OLD_MINIMUM_ROCBLAS_VERSION="2.33.0"
-NEW_MINIMUM_ROCBLAS_VERSION="2.34.0"
+OLD_MINIMUM_ROCBLAS_VERSION="2.35.0"
+NEW_MINIMUM_ROCBLAS_VERSION="2.36.0"
 
 OLD_MINIMUM_ROCSOLVER_VERSION="3.10.0"
 NEW_MINIMUM_ROCSOLVER_VERSION="3.11.0"

clients/benchmarks/CMakeLists.txt

@@ -20,6 +20,9 @@ if( NOT TARGET hipblas )
   endif( )
 endif( )
 
+find_package( GTest REQUIRED )
+include_directories(${GTEST_INCLUDE_DIRS})
+
 set( hipblas_benchmark_common
       ../common/utility.cpp
       ../common/cblas_interface.cpp

clients/benchmarks/client.cpp

@@ -1073,32 +1073,32 @@ try
          "Leading dimension of matrix D, is only applicable to BLAS-EX ")
 
         ("stride_a",
-         value<hipblas_int>(&arg.stride_a)->default_value(128*128),
+         value<hipblasStride>(&arg.stride_a)->default_value(128*128),
          "Specific stride of strided_batched matrix A, is only applicable to strided batched"
          "BLAS-2 and BLAS-3: second dimension * leading dimension.")
 
         ("stride_b",
-         value<hipblas_int>(&arg.stride_b)->default_value(128*128),
+         value<hipblasStride>(&arg.stride_b)->default_value(128*128),
          "Specific stride of strided_batched matrix B, is only applicable to strided batched"
          "BLAS-2 and BLAS-3: second dimension * leading dimension.")
 
         ("stride_c",
-         value<hipblas_int>(&arg.stride_c)->default_value(128*128),
+         value<hipblasStride>(&arg.stride_c)->default_value(128*128),
          "Specific stride of strided_batched matrix C, is only applicable to strided batched"
          "BLAS-2 and BLAS-3: second dimension * leading dimension.")
 
         ("stride_d",
-         value<hipblas_int>(&arg.stride_d)->default_value(128*128),
+         value<hipblasStride>(&arg.stride_d)->default_value(128*128),
          "Specific stride of strided_batched matrix D, is only applicable to strided batched"
          "BLAS_EX: second dimension * leading dimension.")
 
         ("stride_x",
-         value<hipblas_int>(&arg.stride_x)->default_value(128),
+         value<hipblasStride>(&arg.stride_x)->default_value(128),
          "Specific stride of strided_batched vector x, is only applicable to strided batched"
          "BLAS_2: second dimension.")
 
         ("stride_y",
-         value<hipblas_int>(&arg.stride_y)->default_value(128),
+         value<hipblasStride>(&arg.stride_y)->default_value(128),
          "Specific stride of strided_batched vector y, is only applicable to strided batched"
          "BLAS_2: leading dimension.")