[Beignet] [patch v3 2/2] add benckmark for copy data from buffer to image.

[xionghu.luo at intel.com]

From: Luo Xionghu <xionghu.luo at intel.com>

v2: use random input data; update comments.
v3: change the image attribute to __write_only.
Signed-off-by: Luo Xionghu <xionghu.luo at intel.com>
---
 benchmark/CMakeLists.txt                           |  1 +
 benchmark/benchmark_copy_buffer_to_image.cpp       | 66 ++++++++++++++++++++++
 .../cl_internal_copy_buffer_to_image_2d_align16.cl |  2 +-
 3 files changed, 68 insertions(+), 1 deletion(-)
 create mode 100644 benchmark/benchmark_copy_buffer_to_image.cpp

diff --git a/benchmark/CMakeLists.txt b/benchmark/CMakeLists.txt
index 7bd61ee..3e43a21 100644
--- a/benchmark/CMakeLists.txt
+++ b/benchmark/CMakeLists.txt
@@ -15,6 +15,7 @@ set (benchmark_sources
   benchmark_use_host_ptr_buffer.cpp
   benchmark_read_buffer.cpp
   benchmark_read_image.cpp
+  benchmark_copy_buffer_to_image.cpp
   benchmark_copy_image_to_buffer.cpp)
 
 
diff --git a/benchmark/benchmark_copy_buffer_to_image.cpp b/benchmark/benchmark_copy_buffer_to_image.cpp
new file mode 100644
index 0000000..2177cfe
--- /dev/null
+++ b/benchmark/benchmark_copy_buffer_to_image.cpp
@@ -0,0 +1,66 @@
+#include <string.h>
+#include "utests/utest_helper.hpp"
+#include <sys/time.h>
+
+#define IMAGE_BPP 2
+
+double benchmark_copy_buffer_to_image(void)
+{
+  struct timeval start,stop;
+  const size_t w = 960 * 4;
+  const size_t h = 540 * 4;
+  const size_t sz = IMAGE_BPP * w * h;
+  cl_image_format format;
+  cl_image_desc desc;
+
+  memset(&desc, 0x0, sizeof(cl_image_desc));
+  memset(&format, 0x0, sizeof(cl_image_format));
+
+  // Setup image and buffer
+  buf_data[0] = (unsigned short*) malloc(sz);
+  for (uint32_t i = 0; i < w*h; ++i) {
+    ((unsigned short*)buf_data[0])[i] = (rand() & 0xffff);
+  }
+
+  format.image_channel_order = CL_R;
+  format.image_channel_data_type = CL_UNSIGNED_INT16;
+  desc.image_type = CL_MEM_OBJECT_IMAGE2D;
+  desc.image_width = w;
+  desc.image_height = h;
+  desc.image_row_pitch = 0;
+  OCL_CREATE_BUFFER(buf[0], CL_MEM_COPY_HOST_PTR, sz, buf_data[0]);
+  OCL_CREATE_IMAGE(buf[1], 0, &format, &desc, NULL);
+
+  /*copy buffer to image*/
+  size_t origin[3] = {0, 0, 0};
+  size_t region[3] = {w, h, 1};
+
+  OCL_CALL (clEnqueueCopyBufferToImage, queue, buf[0], buf[1], 0, origin, region,
+            0, NULL, NULL);
+  OCL_FINISH();
+  OCL_MAP_BUFFER_GTT(1);
+  /*check result*/
+  for (uint32_t j = 0; j < h; ++j)
+    for (uint32_t i = 0; i < w; i++)
+    {
+      OCL_ASSERT(((unsigned short*)buf_data[0])[j * w + i] == ((unsigned short*)buf_data[1])[j * w + i]);
+    }
+  OCL_UNMAP_BUFFER_GTT(1);
+  gettimeofday(&start,0);
+
+  for (uint32_t i=0; i<100; i++) {
+    OCL_CALL (clEnqueueCopyBufferToImage, queue, buf[0], buf[1], 0, origin, region,
+            0, NULL, NULL);
+  }
+  OCL_FINISH();
+
+  gettimeofday(&stop,0);
+  free(buf_data[0]);
+  buf_data[0] = NULL;
+
+  double elapsed = time_subtract(&stop, &start, 0);
+
+  return BANDWIDTH(sz * 100, elapsed);
+}
+
+MAKE_BENCHMARK_FROM_FUNCTION(benchmark_copy_buffer_to_image);
diff --git a/src/kernels/cl_internal_copy_buffer_to_image_2d_align16.cl b/src/kernels/cl_internal_copy_buffer_to_image_2d_align16.cl
index e4cef73..5b32cd5 100644
--- a/src/kernels/cl_internal_copy_buffer_to_image_2d_align16.cl
+++ b/src/kernels/cl_internal_copy_buffer_to_image_2d_align16.cl
@@ -1,4 +1,4 @@
-kernel void __cl_copy_buffer_to_image_2d_align16(__read_only image2d_t image, global uint4* buffer,
+kernel void __cl_copy_buffer_to_image_2d_align16(__write_only image2d_t image, global uint4* buffer,
                                         unsigned int region0, unsigned int region1, unsigned int region2,
                                         unsigned int dst_origin0, unsigned int dst_origin1, unsigned int dst_origin2,
                                         unsigned int src_offset)
-- 
1.9.1