From 7fc23489243b14b38ea4a4b7789db5fe376b95f0 Mon Sep 17 00:00:00 2001
From: Philippe <ptillet@g.harvard.edu>
Date: Thu, 5 Feb 2015 23:42:31 -0500
Subject: [PATCH] Fixed CUDA benchmark

---
 bench/CMakeLists.txt           |  2 +-
 bench/blas.cpp                 | 20 ++++++++++----------
 cmake_modules/FindOpenCL.cmake |  5 ++---
 3 files changed, 13 insertions(+), 14 deletions(-)

diff --git a/bench/CMakeLists.txt b/bench/CMakeLists.txt
index 950089f64..5b75b011d 100644
--- a/bench/CMakeLists.txt
+++ b/bench/CMakeLists.txt
@@ -37,7 +37,7 @@ string(REPLACE ";" " " BLAS_DEF_STR "${BLAS_DEF}")
 foreach(PROG blas overhead)
     include_directories(${CMAKE_CURRENT_SOURCE_DIR})
     if(CUDA_FOUND)
-        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} " ${BLAS_DEF_STR}")
+        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} " ${BLAS_DEF_STR} -std=c++11")
         set(CUPROG ${CMAKE_CURRENT_BINARY_DIR}/${PROG}.cu)
         file(COPY ${PROG}.cpp DESTINATION ${CMAKE_CURRENT_BINARY_DIR})
 	file(RENAME ${CMAKE_CURRENT_BINARY_DIR}/${PROG}.cpp ${CUPROG})
diff --git a/bench/blas.cpp b/bench/blas.cpp
index 112bf2bc0..970d770e8 100644
--- a/bench/blas.cpp
+++ b/bench/blas.cpp
@@ -53,20 +53,20 @@ void bench(ad::numeric_type dtype)
   {\
   std::vector<long> times;\
   double total_time = 0;\
-  double time;\
+  float time;\
+  cudaEvent_t start, stop;\
+  cudaEventCreate(&start);\
+  cudaEventCreate(&stop);\
   while(total_time*1e-3 < 1e-1){\
-    cudaEvent_t start, stop;\
-    cudaEventCreate(&start);\
-    cudaEventCreate(&stop);\
-    cudaEventRecord(start);\
+    cudaEventRecord(start,0);\
     OP;\
-    cudaEventRecord(stop);\
-    cudaEventSynchronize();\
+    cudaEventRecord(stop,0);\
+    cudaEventSynchronize(stop);\
     cudaEventElapsedTime(&time, start, stop);\
-    times.push_back(time);\
+    times.push_back(time*1e6);\
     total_time+=time;\
   }\
-  double t = 1e-6*median(times);\
+  double t = median(times);\
   std::cout << " " << PERF << std::flush;\
   }
 
@@ -98,7 +98,7 @@ void bench(ad::numeric_type dtype)
     T *cux, *cuy;
     cudaMalloc((void**) &cux, N * sizeof(T));
     cudaMalloc((void**) &cuy, N * sizeof(T));
-    BENCHMARK(cublasSaxpy(N, 2, cux, 1, cuy, 1), 3*N*dtsize/t)
+    BENCHMARK_CUDA(cublasSaxpy(N, 2, cux, 1, cuy, 1), 3*N*dtsize/t)
     cudaFree(cux);
     cudaFree(cuy);
 #endif
diff --git a/cmake_modules/FindOpenCL.cmake b/cmake_modules/FindOpenCL.cmake
index 88b5c0874..10cf9b5c1 100644
--- a/cmake_modules/FindOpenCL.cmake
+++ b/cmake_modules/FindOpenCL.cmake
@@ -1,9 +1,8 @@
 file(GLOB AMDAPPSDK_ROOT /opt/AMDAPPSDK*)
 
 find_package(CUDA QUIET)
-find_path(OPENCL_INCLUDE_DIR CL/cl.hpp HINTS ${AMDAPPSDK_ROOT}/include/ ${CUDA_SDK_ROOT_DIR}/include)
-find_library(OPENCL_LIBRARIES NAMES OpenCL HINTS ${AMDAPPSDK_ROOT}/lib/x86_64/ ${CUDA_SDK_ROOT_DIR}/lib64)
+find_library(OPENCL_LIBRARIES NAMES OpenCL HINTS ${AMDAPPSDK_ROOT}/lib/x86_64/ ${CUDA_TOOLKIT_ROOT_DIR}/targets/x86_64-linux/lib/)
 
 include(FindPackageHandleStandardArgs)
-find_package_handle_standard_args(OpenCL  DEFAULT_MSG OPENCL_LIBRARIES OPENCL_INCLUDE_DIR)
+find_package_handle_standard_args(OpenCL  DEFAULT_MSG OPENCL_LIBRARIES)
 mark_as_advanced(OpenCL)