[Beignet] [PATCH v2 5/6] Implement TILE_Y large image in clEnqueueReadImage.

[yan.wang at linux.intel.com]

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

It will fail to copy data from TILE_Y large image to buffer by memcpy.
Use clEnqueueCopyImageToBuffer to do this on GPU side.

Signed-off-by: Yan Wang <yan.wang at linux.intel.com>
---
 src/cl_api_mem.c | 53 +++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 53 insertions(+)

diff --git a/src/cl_api_mem.c b/src/cl_api_mem.c
index 7c6699d..8cbddf4 100644
--- a/src/cl_api_mem.c
+++ b/src/cl_api_mem.c
@@ -1724,6 +1724,54 @@ clEnqueueMapImage(cl_command_queue command_queue,
   return mem_ptr;
 }
 
+static cl_int
+clEnqueueReadImageByKernel(cl_command_queue command_queue,
+                   cl_mem mem,
+                   cl_bool blocking_read,
+                   const size_t *porigin,
+                   const size_t *pregion,
+                   size_t row_pitch,
+                   size_t slice_pitch,
+                   void *ptr,
+                   cl_uint num_events_in_wait_list,
+                   const cl_event *event_wait_list,
+                   cl_event *event)
+{
+  cl_int err = CL_SUCCESS;
+  struct _cl_mem_image *image = NULL;
+  size_t region[3];
+  size_t origin[3];
+
+  image = cl_mem_image(mem);
+
+  err = check_image_region(image, pregion, region);
+  if (err != CL_SUCCESS)
+    return err;
+
+  err = check_image_origin(image, porigin, origin);
+  if (err != CL_SUCCESS)
+    return err;
+
+  if (mem->tmp_ker_buf)
+    clReleaseMemObject(mem->tmp_ker_buf);
+
+  mem->tmp_ker_buf = clCreateBuffer(command_queue->ctx, CL_MEM_ALLOC_HOST_PTR, mem->size, NULL, &err);
+  if (mem->tmp_ker_buf == NULL || err != CL_SUCCESS) {
+    mem->tmp_ker_buf = NULL;
+    return err;
+  }
+
+  err = clEnqueueCopyImageToBuffer(command_queue, mem, mem->tmp_ker_buf, origin, region,
+    0, 0, NULL, NULL);
+  if (err != CL_SUCCESS) {
+    clReleaseMemObject(mem->tmp_ker_buf);
+    return err;
+  }
+
+  return clEnqueueReadBuffer(command_queue, mem->tmp_ker_buf, blocking_read, 0, mem->size, ptr,
+    num_events_in_wait_list, event_wait_list, event);
+}
+
 cl_int
 clEnqueueReadImage(cl_command_queue command_queue,
                    cl_mem mem,
@@ -1809,6 +1857,11 @@ clEnqueueReadImage(cl_command_queue command_queue,
       break;
     }
 
+    if (mem->is_ker_copy) {
+      return clEnqueueReadImageByKernel(command_queue, mem, blocking_read, origin,
+        region, row_pitch, slice_pitch, ptr, num_events_in_wait_list, event_wait_list, event);
+    }
+
     err = cl_event_check_waitlist(num_events_in_wait_list, event_wait_list,
                                   event, command_queue->ctx);
     if (err != CL_SUCCESS) {
-- 
2.7.4