mirror of
https://github.com/sunnypilot/sunnypilot.git
synced 2026-03-01 14:53:56 +08:00
7b5a4fbb03
* squash * bump tg * bump tg * debump tinygrad * bump tinygrad * bump tg * Skip init iteration * fixes * cleanups * skip first test sample * typos * linter unhappy * update cpu usage * OPENCL just zeros for now * imports * Try printing * Runs again, but slower * unused import * Allow more buffer with tg and all on gpu * bump tinygrad * seems ok * stricter timings for driving looser for dm * try llvm * check nvidia * More timeout for now * make test pass * Revert "try llvm" This reverts commit ef136e478320101fea262bae3579e558da991902. * small fixes * whitespace * revert test timeout * No model runners * Always CPU always fast * No onnx runtime GPU * more cores * cleanup * Is this faster * Is this faster * at least runs * FP32 is faster than 16 * fix deps * whitespace * comment --------- Co-authored-by: Adeeb Shihadeh <adeebshihadeh@gmail.com>
59 lines
2.7 KiB
C++
59 lines
2.7 KiB
C++
#include "selfdrive/modeld/models/commonmodel.h"
|
|
|
|
#include <cassert>
|
|
#include <cmath>
|
|
#include <cstring>
|
|
|
|
#include "common/clutil.h"
|
|
|
|
ModelFrame::ModelFrame(cl_device_id device_id, cl_context context) {
|
|
input_frames = std::make_unique<uint8_t[]>(buf_size);
|
|
input_frames_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, buf_size, NULL, &err));
|
|
|
|
q = CL_CHECK_ERR(clCreateCommandQueue(context, device_id, 0, &err));
|
|
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));
|
|
img_buffer_20hz_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, 5*frame_size_bytes, NULL, &err));
|
|
region.origin = 4 * frame_size_bytes;
|
|
region.size = frame_size_bytes;
|
|
last_img_cl = CL_CHECK_ERR(clCreateSubBuffer(img_buffer_20hz_cl, CL_MEM_READ_WRITE, CL_BUFFER_CREATE_TYPE_REGION, ®ion, &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) {
|
|
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);
|
|
|
|
for (int i = 0; i < 4; i++) {
|
|
CL_CHECK(clEnqueueCopyBuffer(q, img_buffer_20hz_cl, img_buffer_20hz_cl, (i+1)*frame_size_bytes, i*frame_size_bytes, frame_size_bytes, 0, nullptr, nullptr));
|
|
}
|
|
loadyuv_queue(&loadyuv, q, y_cl, u_cl, v_cl, last_img_cl);
|
|
|
|
copy_queue(&loadyuv, q, img_buffer_20hz_cl, input_frames_cl, 0, 0, frame_size_bytes);
|
|
copy_queue(&loadyuv, q, last_img_cl, input_frames_cl, 0, frame_size_bytes, frame_size_bytes);
|
|
|
|
// NOTE: Since thneed is using a different command queue, this clFinish is needed to ensure the image is ready.
|
|
clFinish(q);
|
|
return &input_frames_cl;
|
|
}
|
|
|
|
uint8_t* ModelFrame::buffer_from_cl(cl_mem *in_frames) {
|
|
CL_CHECK(clEnqueueReadBuffer(q, *in_frames, CL_TRUE, 0, MODEL_FRAME_SIZE * 2 * sizeof(uint8_t), &input_frames[0], 0, nullptr, nullptr));
|
|
clFinish(q);
|
|
return &input_frames[0];
|
|
}
|
|
|
|
ModelFrame::~ModelFrame() {
|
|
transform_destroy(&transform);
|
|
loadyuv_destroy(&loadyuv);
|
|
CL_CHECK(clReleaseMemObject(img_buffer_20hz_cl));
|
|
CL_CHECK(clReleaseMemObject(last_img_cl));
|
|
CL_CHECK(clReleaseMemObject(v_cl));
|
|
CL_CHECK(clReleaseMemObject(u_cl));
|
|
CL_CHECK(clReleaseMemObject(y_cl));
|
|
CL_CHECK(clReleaseCommandQueue(q));
|
|
} |