update README and INSTALL

INSTALL.md

@@ -81,6 +81,14 @@ Ginkgo adds the following additional switches to control what is being built:
     this option see the
     [`ARCHITECTURES` specification list](https://github.com/ginkgo-project/CudaArchitectureSelector/blob/master/CudaArchitectureSelector.cmake#L58)
     section in the documentation of the CudaArchitectureSelector CMake module.
+* `-DGINKGO_WINDOWS_SHARED_LIBRARY_RELPATH=<path>` where <path> is a relative
+    path with `PROJECT_BINARY_DIR`. Users must to add the absoulte path
+    (`PROJECT_BINARY_DIR`/`GINKGO_WINDOWS_SHARED_LIBRARY_RELPATH`) into the
+    environment variable PATH when building shared libraries and executable
+    program, default is `windows_shared_library`.
+* `-DGINKGO_CHECK_PATH={ON, OFF}` checks the environment variable PATH is valid.
+    It is valid only when building shared libraries and executable program,
+    default is `ON`
 
 For example, to build everything (in debug mode), use:
 

README.md

@@ -54,10 +54,21 @@ following:
 
 ### Windows
 
-Windows is currently not supported, but we are working on porting the library
-there. If you are interested in helping us with this effort, feel free to
-contact one of the developers. (The library itself doesn't use any non-standard
-C++ features, so most of the effort here is in modifying the build system.)
+The prequirement needs to be verified
+*   _cmake 3.9+_
+*   C++11 compliant 64-bits compiler:
+    *   _MinGW : gcc 5.3+, 6.3+, 7.3+, 8.1+_
+    *   _CygWin : gcc 5.3+, 6.3+, 7.3+, 8.1+_
+    *   _Microsoft Visual Studio : VS 2017 15.7+_
+
+The Ginkgo CUDA module has the following __additional__ requirements:
+
+*   _CUDA 9.0+_
+*   _Microsoft Visual Stutio_
+*   Any host compiler restrictions your version of CUDA may impose also apply
+    here. For the newest CUDA version, this information can be found in the
+    [CUDA installation guide for Windows](https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html)
+
 
 __NOTE:__ Some restrictions will also apply on the version of C and C++ standard
 libraries installed on the system. This needs further investigation.

cuda/base/math.hpp

@@ -99,10 +99,11 @@ __device__ GKO_INLINE std::complex<double> one<std::complex<double>>()
 // as a __device__ function, so it results in a compiler error.
 // Here, `isfinite` is written by hand, which might not be as performant as the
 // intrinsic function from CUDA, but it compiles and works.
-#if defined(__CUDA_ARCH__) && (defined(_MSC_VER) ||                    \
-    (defined(__CUDACC_VER_MAJOR__) && defined(__CUDACC_VER_MINOR__) && \
-    (__CUDACC_VER_MAJOR__ * 1000 + __CUDACC_VER_MINOR__) < 9002 &&     \
-    (defined(__clang__) || defined(__ICC) || defined(__ICL))))
+#if defined(__CUDA_ARCH__) &&                                           \
+    (defined(_MSC_VER) ||                                               \
+     (defined(__CUDACC_VER_MAJOR__) && defined(__CUDACC_VER_MINOR__) && \
+      (__CUDACC_VER_MAJOR__ * 1000 + __CUDACC_VER_MINOR__) < 9002 &&    \
+      (defined(__clang__) || defined(__ICC) || defined(__ICL))))
 
 
 namespace detail {

reference/test/matrix/hybrid_kernels.cpp

@@ -96,7 +96,7 @@ class Hybrid : public ::testing::Test {
         auto v = m->get_const_values();
         auto c = m->get_const_col_idxs();
         auto r = m->get_const_row_ptrs();
-        
+
         ASSERT_EQ(m->get_size(), gko::dim<2>(2, 3));
         ASSERT_EQ(m->get_num_stored_elements(), 4);
         EXPECT_EQ(r[0], 0);