polish some details

PaddlePaddle · Oct 26, 2021 · 558a848 · 558a848
1 parent 5fb285c
commit 558a848
Show file tree

Hide file tree

Showing 5 changed files with 2 additions and 9 deletions.
diff --git a/paddle/fluid/operators/CMakeLists.txt b/paddle/fluid/operators/CMakeLists.txt
@@ -80,7 +80,6 @@ if(WITH_UNITY_BUILD)
 endif()
 
 set(OP_HEADER_DEPS ${OP_HEADER_DEPS} pten)
-#set(OP_HEADER_DEPS ${OP_HEADER_DEPS} pten_utils)
 register_operators(EXCLUDES
 py_layer_op py_func_op warpctc_op dgc_op load_combine_op lstm_op run_program_op eye_op 
         recurrent_op save_combine_op sparse_attention_op sync_batch_norm_op spectral_op ${OP_MKL_DEPS} DEPS ${OP_HEADER_DEPS})

diff --git a/paddle/pten/core/tensor_meta.h b/paddle/pten/core/tensor_meta.h
@@ -78,7 +78,7 @@ inline bool DenseTensorMeta::valid() const noexcept {
   bool valid{true};
   valid = valid && (type != DataType::UNDEFINED);
   valid = valid && (layout != DataLayout::UNDEFINED);
-  valid = valid && (is_scalar || product(dims));
+  valid = valid && (is_scalar || product(dims) >= 0);
   return valid;
 }
 

diff --git a/paddle/pten/kernels/cuda/math.cu b/paddle/pten/kernels/cuda/math.cu
@@ -60,8 +60,6 @@ void Sign(const CUDAContext& dev_ctx, const DenseTensor& x, DenseTensor* out) {
 
 template <typename T>
 void Mean(const CUDAContext& dev_ctx, const DenseTensor& x, DenseTensor* out) {
-  VLOG(1) << "chenweihang: call new pt mean kernel.";
-  // eigen::Mean<CUDAContext, T>(dev_ctx, x, out);
   auto size_prob = x.numel();
   const T* x_data = x.data<T>();
   T* out_data = out->mutable_data<T>();

diff --git a/paddle/pten/kernels/functions/eigen/sign.h b/paddle/pten/kernels/functions/eigen/sign.h
@@ -25,11 +25,7 @@ namespace eigen {
 
 template <typename DevCtx, typename T>
 void Sign(const DevCtx& dev_ctx, const DenseTensor& x, DenseTensor* out) {
-  VLOG(1) << "enter module::Sign";
-  // out->mutable_data<T>(x.place());
   out->mutable_data<T>();
-
-  VLOG(1) << "module::Sign, calc by eigen.";
   // TODO(chenweihang): if we design new tensor, we should support
   // the low-level calc functor use new tensor as input,
   // which may be a big project!

diff --git a/python/paddle/fluid/tests/unittests/test_fuse_all_reduce_pass.py b/python/paddle/fluid/tests/unittests/test_fuse_all_reduce_pass.py
@@ -83,7 +83,7 @@ def _decorate_compare_fused_all_reduce(self, model, use_device):
             use_device,
             init_feed_dict=init_data,
             optimizer=self.optimizer,
-            fuse_all_optimizer_ops=True)
+            fuse_all_optimizer_ops=False)
 
     def test_simple_fc_with_fuse_all_reduce(self):
         self._decorate_compare_fused_all_reduce(simple_fc_net, DeviceType.CUDA)