From 3498434bccaf65fdb8cf59ccc07d2a6900d38188 Mon Sep 17 00:00:00 2001
From: Yancey1989 <yancey1989@gmail.com>
Date: Sat, 10 Feb 2018 10:17:40 +0800
Subject: [PATCH] fix ci

---
 paddle/operators/concat_op.h      | 10 ++++++----
 paddle/operators/split_op.h       |  5 +++--
 paddle/operators/strided_memcpy.h |  4 ++--
 3 files changed, 11 insertions(+), 8 deletions(-)

diff --git a/paddle/operators/concat_op.h b/paddle/operators/concat_op.h
index 2ee9912a3b4..ae10b81ea7e 100644
--- a/paddle/operators/concat_op.h
+++ b/paddle/operators/concat_op.h
@@ -37,8 +37,9 @@ class ConcatKernel : public framework::OpKernel<T> {
     size_t output_offset = 0;
     for (auto* in : ins) {
       auto in_stride = framework::stride_numel(in->dims());
-      StridedNumelCopyWithAxis<T>(ctx, axis, out->data<T>() + output_offset,
-                                  out_stride, in->data<T>(), in_stride);
+      StridedNumelCopyWithAxis<T>(ctx.device_context(), axis,
+                                  out->data<T>() + output_offset, out_stride,
+                                  in->data<T>(), in_stride);
       output_offset += in_stride[axis];
     }
   }
@@ -57,8 +58,9 @@ class ConcatGradKernel : public framework::OpKernel<T> {
     for (auto& out : outs) {
       out->mutable_data<T>(ctx.GetPlace());
       auto out_stride = framework::stride_numel(out->dims());
-      StridedNumelCopyWithAxis<T>(ctx, axis, out->data<T>(), out_stride,
-                                  in->data<T>() + input_offset, in_stride);
+      StridedNumelCopyWithAxis<T>(ctx.device_context(), axis, out->data<T>(),
+                                  out_stride, in->data<T>() + input_offset,
+                                  in_stride);
       input_offset += out_stride[axis];
     }
   }
diff --git a/paddle/operators/split_op.h b/paddle/operators/split_op.h
index e239c9cf303..b956808ef92 100644
--- a/paddle/operators/split_op.h
+++ b/paddle/operators/split_op.h
@@ -37,8 +37,9 @@ class SplitOpKernel : public framework::OpKernel<T> {
     for (auto& out : outs) {
       out->mutable_data<T>(ctx.GetPlace());
       auto out_stride = framework::stride_numel(out->dims());
-      StridedNumelCopyWithAxis<T>(ctx, axis, out->data<T>(), out_stride,
-                                  in->data<T>() + input_offset, in_stride);
+      StridedNumelCopyWithAxis<T>(ctx.device_context(), axis, out->data<T>(),
+                                  out_stride, in->data<T>() + input_offset,
+                                  in_stride);
       input_offset += out_stride[axis];
     }
   }
diff --git a/paddle/operators/strided_memcpy.h b/paddle/operators/strided_memcpy.h
index 49795db91d5..ddecfd76ddd 100644
--- a/paddle/operators/strided_memcpy.h
+++ b/paddle/operators/strided_memcpy.h
@@ -50,7 +50,7 @@ inline void StridedMemcpy(const platform::DeviceContext& dev_ctx, const T* src,
 // NOTE: The src and dst tensor should have the same elements
 // except the specified axis.
 template <typename T>
-inline void StridedNumelCopyWithAxis(const framework::ExecutionContext& ctx,
+inline void StridedNumelCopyWithAxis(const platform::DeviceContext& ctx,
                                      int64_t axis, T* dst,
                                      const framework::DDim& dst_stride_numel,
                                      const T* src,
@@ -88,7 +88,7 @@ inline void StridedNumelCopyWithAxis(const framework::ExecutionContext& ctx,
       auto& gpu_place = boost::get<platform::CUDAPlace>(place);
       auto& cuda_ctx =
           reinterpret_cast<const platform::CUDADeviceContext&>(ctx);
-      memory::Copy(cpu_place, dst + i * dst_after, cpu_place,
+      memory::Copy(gpu_place, dst + i * dst_after, gpu_place,
                    src + i * src_after, sizeof(T) * src_after,
                    cuda_ctx.stream());
 #else
-- 
GitLab