Remove superfluous cudaStreamSynchronize() calls

YOUSIKI · Feb 11, 2022 · f02efe7 · f02efe7
1 parent d829746
commit f02efe7
Show file tree

Hide file tree

Showing 2 changed files with 0 additions and 2 deletions.
diff --git a/include/tiny-cuda-nn/optimizers/shampoo.h b/include/tiny-cuda-nn/optimizers/shampoo.h
@@ -644,7 +644,6 @@ class ShampooOptimizer : public Optimizer<T> {
 			});
 
 			CUDA_CHECK_THROW(cudaMemcpyAsync(delta.data(), sum_tmp, n_matrices * sizeof(ROOT_TYPE), cudaMemcpyDeviceToHost, stream));
-			CUDA_CHECK_THROW(cudaStreamSynchronize(stream));
 
 			if (std::any_of(std::begin(delta), std::end(delta), [](ROOT_TYPE v) { return !std::isfinite(v); })) {
 				std::cout << "Failed to converge! " << delta[0] << std::endl;

diff --git a/include/tiny-cuda-nn/reduce_sum.h b/include/tiny-cuda-nn/reduce_sum.h
@@ -149,7 +149,6 @@ float reduce_sum(T* device_pointer, F fun, uint32_t n_elements, cudaStream_t str
 
 	float sum;
 	CUDA_CHECK_THROW(cudaMemcpyAsync(&sum, workspace_data, sizeof(float), cudaMemcpyDeviceToHost, stream));
-	CUDA_CHECK_THROW(cudaStreamSynchronize(stream));
 	return sum;
 }