From 2b6ca3ac7d86d4a06f0b22fbf3589aa344c676e2 Mon Sep 17 00:00:00 2001
From: Michal Sojka <michal.sojka@cvut.cz>
Date: Fri, 5 Oct 2018 00:29:50 +0200
Subject: [PATCH] Add cudaStreamSynchronize after FFT

This is just a temporary solution, it should be possible to get rid of
CPU/GPU synchronization at most places.
---
 src/fft_cufft.cpp | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/src/fft_cufft.cpp b/src/fft_cufft.cpp
index 61bf1c9..e551eaa 100644
--- a/src/fft_cufft.cpp
+++ b/src/fft_cufft.cpp
@@ -92,6 +92,7 @@ void cuFFT::forward_window(MatScaleFeats &feat, ComplexMat &complex_result, MatS
     else
         cudaErrorCheck(cufftExecR2C(plan_fw_all_scales, temp_data, complex_result.get_dev_data()));
 #endif
+    CudaSafeCall(cudaStreamSynchronize(cudaStreamPerThread));
 }
 
 void cuFFT::inverse(ComplexMat &complex_input, MatScales &real_result)
@@ -110,6 +111,7 @@ void cuFFT::inverse(ComplexMat &complex_input, MatScales &real_result)
 #endif
     // TODO: Investigate whether this scalling is needed or not
     cudaErrorCheck(cublasSscal(cublas, real_result.total(), &alpha, out, 1));
+    CudaSafeCall(cudaStreamSynchronize(cudaStreamPerThread));
 }
 
 cuFFT::~cuFFT()
-- 
2.39.2