Merge pull request PaddlePaddle#26 from mthreads/fix_automicadd

Fix automicadd
hanhaowen-mt · Aug 8, 2023 · fe6699b · fe6699b
2 parents 6a656f5 + 51284d9
commit fe6699b
Show file tree

Hide file tree

Showing 4 changed files with 5 additions and 10 deletions.
diff --git a/paddle/fluid/framework/details/op_handle_base.cc b/paddle/fluid/framework/details/op_handle_base.cc
@@ -34,7 +34,7 @@ OpHandleBase::~OpHandleBase() PADDLE_MAY_THROW {
 #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP) || defined(PADDLE_WITH_MUSA)
   for (auto &ev : events_) {
     if (ev.second) {
-#if defined(PADDLE_WITH_HIP)
+#ifdef PADDLE_WITH_HIP
       PADDLE_ENFORCE_GPU_SUCCESS(hipEventDestroy(ev.second));
 #elif defined(PADDLE_WITH_MUSA)
       PADDLE_ENFORCE_GPU_SUCCESS(musaEventDestroy(ev.second));

diff --git a/paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.cc b/paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.cc
@@ -11,6 +11,7 @@
 // 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.
+
 #include "paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.h"
 
 #include <sstream>

diff --git a/paddle/phi/backends/gpu/musa/musa_device_function.h b/paddle/phi/backends/gpu/musa/musa_device_function.h
@@ -14,8 +14,8 @@ limitations under the License. */
 
 #pragma once
 
+#define PADDLE_CUDA_FP16
 // NOTE(): support float16 to half in header file.
-#define PADDLE_MUSA_FP16
 #include "paddle/phi/common/bfloat16.h"
 #include "paddle/phi/common/complex.h"
 #include "paddle/phi/common/float16.h"
@@ -66,7 +66,7 @@ __forceinline__ __device__ phi::dtype::float16 CudaShuffleDownSync(
 template <>
 __forceinline__ __device__ phi::dtype::bfloat16 CudaShuffleDownSync(
     unsigned mask, phi::dtype::bfloat16 val, int delta, int width) {
-#if defined(PADDLE_MUSA_BF16)
+#if defined(PADDLE_MUSA_BF16) && defined(__MUSA_ARCH__) && __MUSA_ARCH__ >= 220
   return phi::dtype::bfloat16(__shfl_down_sync(
       mask, val.to_mt_bfloat16(), static_cast<unsigned>(delta), width));
 #else

diff --git a/paddle/phi/kernels/funcs/eigen/CMakeLists.txt b/paddle/phi/kernels/funcs/eigen/CMakeLists.txt
@@ -2,15 +2,9 @@ file(
   GLOB EIGEN_CC_SOURCES
   RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}"
   "*.cc")
-# TODO(@caizhi): compile all cu files
 file(
   GLOB EIGEN_CU_SOURCES
   RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}"
-  "s*.cu"
-  "p*.cu"
-  "b*.cu"
-  "c*.cu"
-  "e*.cu"
-  "l*.cu")
+  "*.cu")
 
 collect_srcs(kernels_srcs SRCS ${EIGEN_CC_SOURCES} ${EIGEN_CU_SOURCES})