support unittests

src/fastertransformer/utils/gemm.cc

@@ -26,7 +26,7 @@ Gemm::Gemm(IAllocator* allocator, cudaStream_t stream, std::string config_file)
     stream_    = stream;
     mutex_     = new std::mutex();  // mutex per process
     check_cuda_error(cublasCreate(&cublas_handle_));
-    check_cuda_error(cublasLtCreate(&cublaslt_handle_));
+    // check_cuda_error(cublasLtCreate(&cublaslt_handle_));
     check_cuda_error(cublasSetStream(cublas_handle_, stream));
 
     if (allocator_ != nullptr) {
@@ -41,7 +41,7 @@ Gemm::~Gemm()
         allocator_->free((void**)(&workspace_));
         allocator_ = nullptr;
     }
-    cublasLtDestroy(cublaslt_handle_);
+    // cublasLtDestroy(cublaslt_handle_);
     cublasDestroy(cublas_handle_);
     delete cublas_algo_map_;
     delete mutex_;
@@ -248,7 +248,8 @@ void Gemm::gemm(const GemmOp   transa,
     mutex_->lock();
     // Use cublas as default in FP32 and cublasLt as default in FP16
     bool is_fp16_compute_type = compute_type_ == TYPE_FP16;
-    bool using_cublasLt       = Atype == TYPE_FP16;
+    // bool using_cublasLt       = Atype == TYPE_FP16;
+    bool using_cublasLt       = (Atype == TYPE_FP16) ? false : false;
     int  batch_count          = 1;
 
     half        h_alpha = (half)alpha;

tests/unittests/CMakeLists.txt

@@ -19,6 +19,10 @@ FetchContent_Declare(
   googletest
   GIT_REPOSITORY https://github.com/google/googletest.git
   GIT_TAG release-1.12.1
+  # URL /path/to/local/googletest-release-1.12.1.zip
+  # URL_HASH SHA256=24564e3b712d3eb30ac9a85d92f7d720f60cc0173730ac166f27dda7fed76cb2
+  # export C_INCLUDE_PATH=/path/to/local/googletest-release-1.12.1/googletest/include${C_INCLUDE_PATH:+:${C_INCLUDE_PATH}}
+  # export CPLUS_INCLUDE_PATH=/path/to/local/googletest-release-1.12.1/include${CPLUS_INCLUDE_PATH:+:${CPLUS_INCLUDE_PATH}}
 )
 add_definitions(-DTORCH_CUDA=1)
 

tests/unittests/test_context_decoder_layer.cu

@@ -74,7 +74,7 @@ bool test_context_sharing(const std::string& weight_dir, const std::string& data
     cublasLtHandle_t cublaslt_handle;
     check_cuda_error(cudaStreamCreate(&stream));
     check_cuda_error(cublasCreate(&cublas_handle));
-    check_cuda_error(cublasLtCreate(&cublaslt_handle));
+    // check_cuda_error(cublasLtCreate(&cublaslt_handle));
     check_cuda_error(cublasSetStream(cublas_handle, stream));
 
     cublasAlgoMap cublas_algo_map(GEMM_CONFIG);

tests/unittests/test_gemm.cu

@@ -378,7 +378,7 @@ void testGemmConsistencyMatmul(size_t m, size_t n, size_t k) {
     cublasHandle_t cublas_handle;
     cublasLtHandle_t cublaslt_handle;
     check_cuda_error(cublasCreate(&cublas_handle));
-    check_cuda_error(cublasLtCreate(&cublaslt_handle));
+    // check_cuda_error(cublasLtCreate(&cublaslt_handle));
     check_cuda_error(cublasSetStream(cublas_handle, stream));
     cublasAlgoMap cublas_algo_map(GEMM_CONFIG);
     std::mutex* cublas_wrapper_mutex = new std::mutex();
@@ -436,7 +436,7 @@ void testGemmConsistencyMatmul(size_t m, size_t n, size_t k) {
     }
 
     delete cublas_wrapper_mutex;
-    check_cuda_error(cublasLtDestroy(cublaslt_handle));
+    // check_cuda_error(cublasLtDestroy(cublaslt_handle));
     check_cuda_error(cublasDestroy(cublas_handle));
     check_cuda_error(cudaStreamDestroy(stream));
 }
@@ -494,7 +494,7 @@ void testGemmConsistencyBatchedMatmul(size_t m, size_t n, size_t k) {
     cublasHandle_t cublas_handle;
     cublasLtHandle_t cublaslt_handle;
     check_cuda_error(cublasCreate(&cublas_handle));
-    check_cuda_error(cublasLtCreate(&cublaslt_handle));
+    // check_cuda_error(cublasLtCreate(&cublaslt_handle));
     check_cuda_error(cublasSetStream(cublas_handle, stream));
     cublasAlgoMap cublas_algo_map(GEMM_CONFIG);
     std::mutex* cublas_wrapper_mutex = new std::mutex();
@@ -569,7 +569,7 @@ void testGemmConsistencyBatchedMatmul(size_t m, size_t n, size_t k) {
     c_tensors.clear();
     expecteds.clear();
     delete cublas_wrapper_mutex;
-    check_cuda_error(cublasLtDestroy(cublaslt_handle));
+    // check_cuda_error(cublasLtDestroy(cublaslt_handle));
     check_cuda_error(cublasDestroy(cublas_handle));
     check_cuda_error(cudaStreamDestroy(stream));
 }
@@ -594,7 +594,7 @@ void testGemmConsistencyStridedBatchedMatmul(size_t batch_size, size_t m, size_t
     cublasHandle_t cublas_handle;
     cublasLtHandle_t cublaslt_handle;
     check_cuda_error(cublasCreate(&cublas_handle));
-    check_cuda_error(cublasLtCreate(&cublaslt_handle));
+    // check_cuda_error(cublasLtCreate(&cublaslt_handle));
     check_cuda_error(cublasSetStream(cublas_handle, stream));
     cublasAlgoMap cublas_algo_map(GEMM_CONFIG);
     std::mutex* cublas_wrapper_mutex = new std::mutex();
@@ -683,7 +683,7 @@ void testGemmConsistencyStridedBatchedMatmul(size_t batch_size, size_t m, size_t
     }
 
     delete cublas_wrapper_mutex;
-    check_cuda_error(cublasLtDestroy(cublaslt_handle));
+    // check_cuda_error(cublasLtDestroy(cublaslt_handle));
     check_cuda_error(cublasDestroy(cublas_handle));
     check_cuda_error(cudaStreamDestroy(stream));
 }
@@ -779,7 +779,7 @@ void testSpGemmConsistencyMatmul(size_t m, size_t n, size_t k) {
     cublasHandle_t cublas_handle;
     cublasLtHandle_t cublaslt_handle;
     check_cuda_error(cublasCreate(&cublas_handle));
-    check_cuda_error(cublasLtCreate(&cublaslt_handle));
+    // check_cuda_error(cublasLtCreate(&cublaslt_handle));
     check_cuda_error(cublasSetStream(cublas_handle, stream));
     cublasAlgoMap cublas_algo_map(GEMM_CONFIG);
     std::mutex* cublas_wrapper_mutex = new std::mutex();
@@ -844,7 +844,7 @@ void testSpGemmConsistencyMatmul(size_t m, size_t n, size_t k) {
     }
 
     delete cublas_wrapper_mutex;
-    check_cuda_error(cublasLtDestroy(cublaslt_handle));
+    // check_cuda_error(cublasLtDestroy(cublaslt_handle));
     check_cuda_error(cublasDestroy(cublas_handle));
     check_cuda_error(cudaStreamDestroy(stream));
 }