move CMP0104 to the top, fix issue in win cuda11

* otherwise, windows cmake still gives CMP0104 warning * generic api is available from cuda10.1 (expect for win), cuda11 (all)
ginkgo-project · Aug 7, 2020 · 880e1d1 · 880e1d1
1 parent c7d32a6
commit 880e1d1
Show file tree

Hide file tree

Showing 5 changed files with 36 additions and 29 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -5,6 +5,12 @@ if (CMAKE_VERSION VERSION_GREATER_EQUAL 3.12)
     cmake_policy(SET CMP0074 NEW)
 endif()
 
+# Let CAS handle the CUDA architecture flags (for now)
+# Windows still gives CMP0104 warning if putting it in cuda.
+if (CMAKE_VERSION VERSION_GREATER_EQUAL 3.18)
+    cmake_policy(SET CMP0104 OLD)
+endif()
+
 project(Ginkgo LANGUAGES C CXX VERSION 1.2.0 DESCRIPTION "A numerical linear algebra library targeting many-core architectures")
 set(Ginkgo_VERSION_TAG "develop")
 set(PROJECT_VERSION_TAG ${Ginkgo_VERSION_TAG})

diff --git a/benchmark/utils/cuda_linops.hpp b/benchmark/utils/cuda_linops.hpp
@@ -494,8 +494,9 @@ class CuspHybrid
 #endif
 
 
-#if defined(CUDA_VERSION) && (CUDA_VERSION >= 10010) && \
-    !(defined(_WIN32) || defined(__CYGWIN__))
+#if defined(CUDA_VERSION) &&  \
+    (CUDA_VERSION >= 11000 || \
+     ((CUDA_VERSION >= 10010) && !(defined(_WIN32) || defined(__CYGWIN__))))
 
 
 template <typename ValueType>
@@ -690,8 +691,8 @@ class CuspGenericCoo
 };
 
 
-#endif  // defined(CUDA_VERSION) && (CUDA_VERSION >= 10010) &&
-        // !(defined(_WIN32) || defined(__CYGWIN__))
+#endif  // defined(CUDA_VERSION) && (CUDA_VERSION >= 11000 || ((CUDA_VERSION >=
+        // 10010) && !(defined(_WIN32) || defined(__CYGWIN__))))
 
 
 }  // namespace detail
@@ -706,8 +707,9 @@ using cusp_csrmm = detail::CuspCsrmm<>;
 #endif
 
 
-#if defined(CUDA_VERSION) && (CUDA_VERSION >= 10010) && \
-    !(defined(_WIN32) || defined(__CYGWIN__))
+#if defined(CUDA_VERSION) &&  \
+    (CUDA_VERSION >= 11000 || \
+     ((CUDA_VERSION >= 10010) && !(defined(_WIN32) || defined(__CYGWIN__))))
 
 
 using cusp_gcsr = detail::CuspGenericCsr<>;
@@ -716,8 +718,8 @@ using cusp_gcsr2 =
 using cusp_gcoo = detail::CuspGenericCoo<>;
 
 
-#endif  // defined(CUDA_VERSION) && (CUDA_VERSION >= 10010) &&
-        // !(defined(_WIN32) || defined(__CYGWIN__))
+#endif  // defined(CUDA_VERSION) && (CUDA_VERSION >= 11000 || ((CUDA_VERSION >=
+        // 10010) && !(defined(_WIN32) || defined(__CYGWIN__))))
 
 
 #if defined(CUDA_VERSION) && (CUDA_VERSION < 11000)

diff --git a/benchmark/utils/formats.hpp b/benchmark/utils/formats.hpp
@@ -109,17 +109,18 @@ std::string format_description =
     "cusp_coo: is an alias of cusp_gcoo.\n"
 #endif
     "cusp_csrex: benchmark CuSPARSE with the cusparseXcsrmvEx function."
-#if defined(CUDA_VERSION) && (CUDA_VERSION >= 10010) && \
-    !(defined(_WIN32) || defined(__CYGWIN__))
+#if defined(CUDA_VERSION) &&  \
+    (CUDA_VERSION >= 11000 || \
+     ((CUDA_VERSION >= 10010) && !(defined(_WIN32) || defined(__CYGWIN__))))
     "\n"
     "cusp_gcsr: benchmark CuSPARSE with the generic csr with default "
     "algorithm.\n"
     "cusp_gcsr2: benchmark CuSPARSE with the generic csr with "
     "CUSPARSE_CSRMV_ALG2.\n"
     "cusp_gcoo: benchmark CuSPARSE with the generic coo with default "
     "algorithm.\n"
-#endif  // defined(CUDA_VERSION) && (CUDA_VERSION >= 10010) &&
-        // !(defined(_WIN32) || defined(__CYGWIN__))
+#endif  // defined(CUDA_VERSION) && (CUDA_VERSION >= 11000 || ((CUDA_VERSION >=
+        // 10010) && !(defined(_WIN32) || defined(__CYGWIN__))))
 #endif  // HAS_CUDA
 #ifdef HAS_HIP
     "\n"
@@ -212,13 +213,14 @@ const std::map<std::string, std::function<std::unique_ptr<gko::LinOp>(
         {"cusp_coo", read_matrix_from_data<cusp_gcoo>},
 #endif
         {"cusp_csrex", read_matrix_from_data<cusp_csrex>},
-#if defined(CUDA_VERSION) && (CUDA_VERSION >= 10010) && \
-    !(defined(_WIN32) || defined(__CYGWIN__))
+#if defined(CUDA_VERSION) &&  \
+    (CUDA_VERSION >= 11000 || \
+     ((CUDA_VERSION >= 10010) && !(defined(_WIN32) || defined(__CYGWIN__))))
         {"cusp_gcsr", read_matrix_from_data<cusp_gcsr>},
         {"cusp_gcsr2", read_matrix_from_data<cusp_gcsr2>},
         {"cusp_gcoo", read_matrix_from_data<cusp_gcoo>},
-#endif  // defined(CUDA_VERSION) && (CUDA_VERSION >= 10010) &&
-        // !(defined(_WIN32) || defined(__CYGWIN__))
+#endif  // defined(CUDA_VERSION) && (CUDA_VERSION >= 11000 || ((CUDA_VERSION >=
+        // 10010) && !(defined(_WIN32) || defined(__CYGWIN__))))
 #endif  // HAS_CUDA
 #ifdef HAS_HIP
         {"hipsp_csr", read_matrix_from_data<hipsp_csr>},

diff --git a/cuda/CMakeLists.txt b/cuda/CMakeLists.txt
@@ -151,11 +151,6 @@ target_link_libraries(ginkgo_cuda PRIVATE ${CUDA_RUNTIME_LIBS} ${CUBLAS} ${CUSPA
 # Need to link against ginkgo_hip for the `raw_copy_to(HipExecutor ...)` method
 target_link_libraries(ginkgo_cuda PUBLIC ginkgo_hip)
 
-# Let CAS handle the CUDA architecture flags (for now)
-if (CMAKE_VERSION VERSION_GREATER_EQUAL 3.18)
-    cmake_policy(SET CMP0104 OLD)
-endif()
-
 cas_target_cuda_architectures(ginkgo_cuda
     ARCHITECTURES ${GINKGO_CUDA_ARCHITECTURES}
     UNSUPPORTED "20" "21")

diff --git a/cuda/base/types.hpp b/cuda/base/types.hpp
@@ -198,8 +198,9 @@ constexpr cudaDataType_t cuda_data_type_impl<uint8>()
 }
 
 
-#if defined(CUDA_VERSION) && (CUDA_VERSION >= 10010) && \
-    !(defined(_WIN32) || defined(__CYGWIN__))
+#if defined(CUDA_VERSION) &&  \
+    (CUDA_VERSION >= 11000 || \
+     ((CUDA_VERSION >= 10010) && !(defined(_WIN32) || defined(__CYGWIN__))))
 
 
 template <typename T>
@@ -221,8 +222,8 @@ constexpr cusparseIndexType_t cusparse_index_type_impl<int64>()
 }
 
 
-#endif  // defined(CUDA_VERSION) && (CUDA_VERSION >= 10010) &&
-        // !(defined(_WIN32) || defined(__CYGWIN__))
+#endif  // defined(CUDA_VERSION) && (CUDA_VERSION >= 11000 || ((CUDA_VERSION >=
+        // 10010) && !(defined(_WIN32) || defined(__CYGWIN__))))
 
 
 }  // namespace detail
@@ -243,8 +244,9 @@ constexpr cudaDataType_t cuda_data_type()
 }
 
 
-#if defined(CUDA_VERSION) && (CUDA_VERSION >= 10010) && \
-    !(defined(_WIN32) || defined(__CYGWIN__))
+#if defined(CUDA_VERSION) &&  \
+    (CUDA_VERSION >= 11000 || \
+     ((CUDA_VERSION >= 10010) && !(defined(_WIN32) || defined(__CYGWIN__))))
 
 
 /**
@@ -262,8 +264,8 @@ constexpr cusparseIndexType_t cusparse_index_type()
 }
 
 
-#endif  // defined(CUDA_VERSION) && (CUDA_VERSION >= 10010) &&
-        // !(defined(_WIN32) || defined(__CYGWIN__))
+#endif  // defined(CUDA_VERSION) && (CUDA_VERSION >= 11000 || ((CUDA_VERSION >=
+        // 10010) && !(defined(_WIN32) || defined(__CYGWIN__))))
 
 
 /**