[GPU/OpenCL] Rebased and Changes for Addition Layer

Rebased to main Added LayerSemanticsGPU for Addition Layer Signed-off-by: yash.singh <[email protected]>
nnstreamer · Jun 6, 2024 · 9976775 · 9976775
2 parents e69d4ad + 11c096b
commit 9976775
Show file tree

Hide file tree

Showing 4 changed files with 5 additions and 15 deletions.
diff --git a/nntrainer/layers/cl_layers/blas_kernels.cpp b/nntrainer/layers/cl_layers/blas_kernels.cpp
@@ -313,7 +313,7 @@ void addition_cl(const float *input, float *res,
                                   unsigned int size, RunLayerContext &context) {
 
   bool result = false;
-  
+
   do {
     result = result =
       context.clCreateKernel(addition_cl_kernel_, context.LayerKernel::ADD,

diff --git a/nntrainer/layers/cl_layers/blas_kernels.h b/nntrainer/layers/cl_layers/blas_kernels.h
@@ -101,7 +101,6 @@ void sgemm_cl(const float *A, const float *B, float *C, unsigned int M,
               unsigned int ldb, unsigned int ldc, RunLayerContext &context);
 
 /**
-<<<<<<< HEAD
  * @brief     fp16 sgemm computation : Y = op(A)*op(B) + C,
  * where op(X) is one of X or X**T
  * @param[in] A fp16 * for Matrix A
@@ -118,7 +117,8 @@ void sgemm_cl(const float *A, const float *B, float *C, unsigned int M,
 void sgemm_cl(const __fp16 *A, const __fp16 *B, __fp16 *C, unsigned int M,
               unsigned int N, unsigned int K, unsigned int lda,
               unsigned int ldb, unsigned int ldc, RunLayerContext &context);
-=======
+
+/**
  * @brief     addition : sum of all input vectors
  * @param[in] input float * for input
  * @param[in] res float * for result/output
@@ -127,7 +127,6 @@ void sgemm_cl(const __fp16 *A, const __fp16 *B, __fp16 *C, unsigned int M,
  */
 void addition_cl(const float *input, float *res, unsigned int size,
                 RunLayerContext &context);
->>>>>>> [GPU/OpenCL] Addition Kernel added in reusable blas OpenCL kernels
 
 } // namespace nntrainer
 #endif /* __BLAS_KERNELS_H__ */
diff --git a/test/input_gen/gen_layer_tests.py b/test/input_gen/gen_layer_tests.py
@@ -889,6 +889,3 @@ def swiglu(inputs):
 
     added = K.layers.Add()
     record_single(added, [(3, 4, 3, 4), (3, 4, 3, 4)], "added_w32a32_2")
-
-    added = K.layers.Add()
-    record_single(added, [(20, 55, 50, 55), (20, 55, 50, 55)], "added_w32a32_3")
diff --git a/test/unittest/layers/unittest_layers_addition_cl.cpp b/test/unittest/layers/unittest_layers_addition_cl.cpp
@@ -26,7 +26,7 @@ auto semantic_addition_multi_gpu = LayerSemanticsParamType(
   nntrainer::AdditionLayerCL::type, {},
   LayerCreateSetPropertyOptions::AVAILABLE_FROM_APP_CONTEXT, false, 2);
 
-GTEST_PARAMETER_TEST(AdditionGPU, LayerSemantics,
+GTEST_PARAMETER_TEST(AdditionGPU, LayerSemanticsGpu,
                      ::testing::Values(semantic_addition_gpu,
                                        semantic_addition_multi_gpu));
 
@@ -40,11 +40,5 @@ auto addition_w32a32_2 = LayerGoldenTestParamType(
   "added_w32a32_2.nnlayergolden", LayerGoldenTestParamOptions::DEFAULT, "nchw",
   "fp32", "fp32");
 
-auto addition_w32a32_3 = LayerGoldenTestParamType(
-  nntrainer::createLayer<nntrainer::AdditionLayerCL>, {},
-  "20:55:50:55,20:55:50:55", "added_w32a32_3.nnlayergolden",
-  LayerGoldenTestParamOptions::DEFAULT, "nchw", "fp32", "fp32");
-
 GTEST_PARAMETER_TEST(AdditionGPU, LayerGoldenTest,
-                     ::testing::Values(addition_w32a32, addition_w32a32_2,
-                                       addition_w32a32_3));
+                     ::testing::Values(addition_w32a32, addition_w32a32_2));