From e5f6ba2121995613d628a0ee19dce62cb08986a1 Mon Sep 17 00:00:00 2001
From: yangfei <yangfei11@baidu.com>
Date: Mon, 15 Oct 2018 19:49:16 +0800
Subject: [PATCH] imp feed op

---
 src/framework/cl/cl_tensor.h            |  4 ++--
 src/operators/kernel/cl/feed_kernel.cpp | 10 ++--------
 2 files changed, 4 insertions(+), 10 deletions(-)

diff --git a/src/framework/cl/cl_tensor.h b/src/framework/cl/cl_tensor.h
index e4d840f832..c38091dd39 100644
--- a/src/framework/cl/cl_tensor.h
+++ b/src/framework/cl/cl_tensor.h
@@ -37,10 +37,10 @@ class CLTensor : TensorBase {
   }
 
   template <typename T>
-  inline T *mutable_with_data(void *data) {
+  inline T mutable_with_data(void *data) {
     int64_t size = numel() * sizeof(float);
     holder_.reset(new PlaceholderImpl(size, data, typeid(T), context_));
-    return reinterpret_cast<T *>(
+    return reinterpret_cast<T>(
         reinterpret_cast<void *>(reinterpret_cast<uintptr_t>(holder_->ptr())));
   }
 
diff --git a/src/operators/kernel/cl/feed_kernel.cpp b/src/operators/kernel/cl/feed_kernel.cpp
index 93bb371eba..d12dcdbdce 100644
--- a/src/operators/kernel/cl/feed_kernel.cpp
+++ b/src/operators/kernel/cl/feed_kernel.cpp
@@ -32,16 +32,13 @@ void FeedKernel<GPU_CL, float>::Compute(const FeedParam<GPU_CL> &param) {
   const Tensor *input = param.InputX();
   const float *input_data = input->data<float>();
   int numel = input->numel();
-  DLOG << "numel = " << numel;
   cl_mem cl_image = output->GetCLImage();
   int height = output->dims()[2];
   int width = output->dims()[3];
-  DLOG << output->dims();
   CLTensor input_cl_tensor(this->cl_helper_.CLContext());
   input_cl_tensor.Resize(input->dims());
-  cl_mem *inputBuffer =
+  cl_mem inputBuffer =
       input_cl_tensor.mutable_with_data<cl_mem>((void *)input_data);
-  DLOG << "yangfei";
 
   status = clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&inputBuffer);
   CL_CHECK_ERRORS(status);
@@ -53,21 +50,18 @@ void FeedKernel<GPU_CL, float>::Compute(const FeedParam<GPU_CL> &param) {
   CL_CHECK_ERRORS(status);
 
   size_t global_work_size[2] = {height, width};
-  DLOG << "yangfei";
   status = clEnqueueNDRangeKernel(this->cl_helper_.CLCommandQueue(), kernel, 2,
                                   NULL, global_work_size, NULL, 0, NULL, NULL);
   CL_CHECK_ERRORS(status);
 
   int len = 4 * 224 * 224;
   half *out = new half[len];
-  DLOG << "yangfei";
   cl_command_queue commandQueue = this->cl_helper_.CLCommandQueue();
   size_t origin[3] = {0, 0, 0};
   size_t region[3] = {height, width, 1};
   clEnqueueReadImage(commandQueue, cl_image, CL_TRUE, origin, region, 0, 0, out,
                      0, NULL, NULL);
-  DLOG << "yangfei";
-  for (int i = 0; i < 100; i++) DLOG << out[i];
+  for (int i = 0; i < numel; i++) DLOG << Half2Float(out[i]);
 }
 
 template class FeedKernel<GPU_CL, float>;
-- 
GitLab