One less copy

commaai · Sep 26, 2024 · 27c89a0 · 27c89a0
1 parent 27fac49
commit 27c89a0
Show file tree

Hide file tree

Showing 3 changed files with 11 additions and 10 deletions.
diff --git a/selfdrive/modeld/modeld.py b/selfdrive/modeld/modeld.py
@@ -116,9 +116,6 @@ def run(self, buf: VisionBuf, wbuf: VisionBuf, transform: np.ndarray, transform_
       tensor_inputs['input_imgs'] = Tensor.from_blob(rawbuf_ptr, IMG_INPUT_SHAPE, dtype=dtypes.uint8, device='QCOM')
     else:
       tensor_inputs['input_imgs'] = Tensor(self.frame.buffer_from_cl(input_imgs_cl)).reshape(IMG_INPUT_SHAPE)
-    a = tensor_inputs['input_imgs'].numpy().flatten()
-
-    print(a[:10], a[MODEL_FRAME_SIZE:MODEL_FRAME_SIZE+10])
 
     if wbuf is not None:
       big_input_imgs_cl = self.wide_frame.prepare(wbuf, transform_wide.flatten(), None)

diff --git a/selfdrive/modeld/models/commonmodel.cc b/selfdrive/modeld/models/commonmodel.cc
@@ -14,20 +14,23 @@ ModelFrame::ModelFrame(cl_device_id device_id, cl_context context) {
   y_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, MODEL_WIDTH * MODEL_HEIGHT, NULL, &err));
   u_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, (MODEL_WIDTH / 2) * (MODEL_HEIGHT / 2), NULL, &err));
   v_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, (MODEL_WIDTH / 2) * (MODEL_HEIGHT / 2), NULL, &err));
-  net_input_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, MODEL_FRAME_SIZE * sizeof(uint8_t), NULL, &err));
+
+  cl_buffer_region region;
+  region.origin = MODEL_FRAME_SIZE * sizeof(uint8_t);
+  region.size = MODEL_FRAME_SIZE * sizeof(uint8_t);
+  latest_frame_cl = CL_CHECK_ERR(clCreateSubBuffer(input_frames_cl, CL_MEM_READ_WRITE, CL_BUFFER_CREATE_TYPE_REGION, &region, &err));
 
   transform_init(&transform, context, device_id);
   loadyuv_init(&loadyuv, context, device_id, MODEL_WIDTH, MODEL_HEIGHT);
 }
 
 cl_mem* ModelFrame::prepare(cl_mem yuv_cl, int frame_width, int frame_height, int frame_stride, int frame_uv_offset, const mat3 &projection, cl_mem *output) {
+  CL_CHECK(clEnqueueCopyBuffer(q, latest_frame_cl, input_frames_cl, 0, 0, MODEL_FRAME_SIZE * sizeof(uint8_t), 0, nullptr, nullptr));
+
   transform_queue(&this->transform, q,
                   yuv_cl, frame_width, frame_height, frame_stride, frame_uv_offset,
                   y_cl, u_cl, v_cl, MODEL_WIDTH, MODEL_HEIGHT, projection);
-
-  loadyuv_queue(&loadyuv, q, y_cl, u_cl, v_cl, net_input_cl);
-  CL_CHECK(clEnqueueCopyBuffer(q, input_frames_cl, input_frames_cl, MODEL_FRAME_SIZE * sizeof(uint8_t), 0, MODEL_FRAME_SIZE * sizeof(uint8_t), 0, nullptr, nullptr));
-  CL_CHECK(clEnqueueCopyBuffer(q, net_input_cl, input_frames_cl, 0,  MODEL_FRAME_SIZE * sizeof(uint8_t), MODEL_FRAME_SIZE * sizeof(uint8_t), 0, nullptr, nullptr));
+  loadyuv_queue(&loadyuv, q, y_cl, u_cl, v_cl, latest_frame_cl);
 
   // NOTE: Since thneed is using a different command queue, this clFinish is needed to ensure the image is ready.
   clFinish(q);
@@ -43,7 +46,8 @@ uint8_t* ModelFrame::buffer_from_cl(cl_mem *in_frames) {
 ModelFrame::~ModelFrame() {
   transform_destroy(&transform);
   loadyuv_destroy(&loadyuv);
-  CL_CHECK(clReleaseMemObject(net_input_cl));
+  CL_CHECK(clReleaseMemObject(latest_frame_cl));
+  CL_CHECK(clReleaseMemObject(input_frames_cl));
   CL_CHECK(clReleaseMemObject(v_cl));
   CL_CHECK(clReleaseMemObject(u_cl));
   CL_CHECK(clReleaseMemObject(y_cl));

diff --git a/selfdrive/modeld/models/commonmodel.h b/selfdrive/modeld/models/commonmodel.h
@@ -32,6 +32,6 @@ class ModelFrame {
   Transform transform;
   LoadYUVState loadyuv;
   cl_command_queue q;
-  cl_mem y_cl, u_cl, v_cl, net_input_cl, input_frames_cl;
+  cl_mem y_cl, u_cl, v_cl, latest_frame_cl, input_frames_cl;
   std::unique_ptr<uint8_t[]> input_frames;
 };