[Beignet] [PATCH v3 7/7] Optimize clEnqueueWriteImageByKernel and clEnqueuReadImageByKernel.

[yan.wang at linux.intel.com]

From: Yan Wang <yan.wang at linux.intel.com>

1. Only copy the data by origin and region defined.
2. Add clFinish to guarantee the kernel copying is finished when blocking writing.

Signed-off-by: Yan Wang <yan.wang at linux.intel.com>
---
 src/cl_api_mem.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

diff --git a/src/cl_api_mem.c b/src/cl_api_mem.c
index a450e19..46672b7 100644
--- a/src/cl_api_mem.c
+++ b/src/cl_api_mem.c
@@ -1843,8 +1843,9 @@ clEnqueueReadImageByKernel(cl_command_queue command_queue,
   if (image->tmp_ker_buf)
     clReleaseMemObject(image->tmp_ker_buf);
 
+  size_t buf_size = region[0] * region[1] * region[2] * image->bpp;
   image->tmp_ker_buf = clCreateBuffer(command_queue->ctx, CL_MEM_ALLOC_HOST_PTR,
-    mem->size, NULL, &err);
+    buf_size, NULL, &err);
   if (image->tmp_ker_buf == NULL || err != CL_SUCCESS) {
     image->tmp_ker_buf = NULL;
     return err;
@@ -1859,7 +1860,7 @@ clEnqueueReadImageByKernel(cl_command_queue command_queue,
   }
 
   return clEnqueueReadBuffer(command_queue, image->tmp_ker_buf, blocking_read, 0,
-    mem->size, ptr, num_events_in_wait_list, event_wait_list, event);
+    buf_size, ptr, num_events_in_wait_list, event_wait_list, event);
 }
 
 cl_int
@@ -2050,14 +2051,20 @@ clEnqueueWriteImageByKernel(cl_command_queue command_queue,
   if (image->tmp_ker_buf)
     clReleaseMemObject(image->tmp_ker_buf);
 
-  image->tmp_ker_buf = clCreateBuffer(command_queue->ctx, CL_MEM_USE_HOST_PTR, mem->size, (void*)ptr, &err);
+  size_t buf_size = region[0] * region[1] * region[2] * image->bpp;
+  image->tmp_ker_buf = clCreateBuffer(command_queue->ctx, CL_MEM_USE_HOST_PTR, buf_size, (void*)ptr, &err);
   if (image->tmp_ker_buf == NULL || err != CL_SUCCESS) {
     image->tmp_ker_buf = NULL;
     return err;
   }
 
-  return clEnqueueCopyBufferToImage(command_queue, image->tmp_ker_buf, mem, 0, origin, region,
+  err = clEnqueueCopyBufferToImage(command_queue, image->tmp_ker_buf, mem, 0, origin, region,
     num_events_in_wait_list, event_wait_list, event);
+
+  if (blocking_write)
+    err = clFinish(command_queue);
+
+  return err;
 }
 
 cl_int
-- 
2.7.4