Windows platform adapts to cuda12.6 (PaddlePaddle#71286)

* win-cuda12.6 fix

* glog

* glog

* del glog

Author: swgu98

cmake/external/glog.cmake

@@ -14,6 +14,8 @@
 
 include(ExternalProject)
 
+add_definitions(-DGLOG_NO_ABBREVIATED_SEVERITIES)
+
 set(GLOG_PREFIX_DIR ${THIRD_PARTY_PATH}/glog)
 set(GLOG_INSTALL_DIR ${THIRD_PARTY_PATH}/install/glog)
 set(GLOG_INCLUDE_DIR

paddle/common/enforce.h

@@ -37,8 +37,6 @@
 #include <execinfo.h>
 #endif
 
-// msvc conflict logging with windows.h
-#define GLOG_NO_ABBREVIATED_SEVERITIES
 #include "paddle/utils/string/printf.h"
 #include "paddle/utils/string/to_string.h"
 #include "paddle/utils/variant.h"

paddle/fluid/framework/io/shell.cc

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include <array>
-#define GLOG_NO_ABBREVIATED_SEVERITIES  // msvc conflict logging with windows.h
+
 #include "paddle/fluid/framework/io/shell.h"
 
 #include "paddle/common/enforce.h"

paddle/fluid/framework/op_registry.h

@@ -24,8 +24,7 @@ limitations under the License. */
 #include <unordered_map>
 #include <unordered_set>
 
-#define GLOG_NO_ABBREVIATED_SEVERITIES  // msvc conflict logging with windows.h
-#include "glog/logging.h"               // For VLOG()
+#include "glog/logging.h"  // For VLOG()
 #include "paddle/common/flags.h"
 #include "paddle/common/macros.h"
 #include "paddle/fluid/framework/attribute.h"

paddle/fluid/platform/enforce.h

@@ -59,7 +59,6 @@ limitations under the License. */
 #include <execinfo.h>
 #endif
 
-#define GLOG_NO_ABBREVIATED_SEVERITIES  // msvc conflict logging with windows.h
 #include "glog/logging.h"
 #include "paddle/common/errors.h"
 #include "paddle/common/flags.h"

paddle/phi/common/port.h

@@ -17,8 +17,6 @@
 #include <string>
 #include "paddle/utils/test_macros.h"
 
-#define GLOG_NO_ABBREVIATED_SEVERITIES  // msvc conflict logging with windows.h
-
 #if !defined(_WIN32)
 #include <dlfcn.h>  // dladdr
 #include <sys/time.h>

paddle/phi/core/memory/allocation/system_allocator.cc

@@ -11,7 +11,6 @@ distributed under the License is distributed on an "AS IS" BASIS,
 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
-#define GLOG_NO_ABBREVIATED_SEVERITIES
 
 #include "paddle/phi/core/memory/allocation/system_allocator.h"
 

paddle/phi/core/platform/profiler/cpu_utilization.cc

@@ -12,7 +12,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#define GLOG_NO_ABBREVIATED_SEVERITIES  // msvc conflict logging with windows.h
 #include "paddle/phi/core/platform/profiler/cpu_utilization.h"
 #include <array>
 #include "glog/logging.h"

paddle/phi/kernels/funcs/shuffle_batch.cu.h

@@ -149,8 +149,9 @@ struct write_output_op_fixed {
   std::uint64_t m;
   InputIterT in;
   OutputIterT out;
-  // flag contains inclusive scan of valid keys
-  // perform gather using valid keys
+// flag contains inclusive scan of valid keys
+// perform gather using valid keys
+#if CUDA_VERSION < 12060
   __thrust_exec_check_disable__ __host__ __device__ std::size_t operator()(
       key_flag_tuple_fixed x) {
     if (x.key < m) {
@@ -159,6 +160,16 @@ struct write_output_op_fixed {
     }
     return 0;  // Discarded
   }
+#else
+  _CCCL_EXEC_CHECK_DISABLE
+  _CCCL_HOST_DEVICE std::size_t operator()(key_flag_tuple_fixed x) {
+    if (x.key < m) {
+      // -1 because inclusive scan
+      out[x.flag - 1] = in[x.key];
+    }
+    return 0;  // Discarded
+  }
+#endif
 };
 
 template <typename ExecutionPolicy,

test/cpp/fluid/platform/bfloat16_test.cc

@@ -13,7 +13,6 @@ limitations under the License. */
 
 #include "paddle/phi/kernels/funcs/eigen/extensions.h"
 
-#define GLOG_NO_ABBREVIATED_SEVERITIES  // msvc conflict logging with windows.h
 #include "gtest/gtest.h"
 #include "paddle/fluid/framework/lod_tensor.h"
 #include "paddle/fluid/platform/enforce.h"