sadmen
/
jyoung
mirror of https://github.com/jyoung8607/openpilot.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Refactor ModelFrame struct to class (#20005)

Browse Source 
* refactor modelframe

* remove DUMP_YUV

* use constructor/destructor instead of init/free

* OMG...should be clEnqueueReadBuffer

* rebase master

* rebase master

* fix build

* small cleanup

Co-authored-by: Adeeb Shihadeh <adeebshihadeh@gmail.com>
pull/69/head
Dean Lee 4 years ago committed by GitHub
parent 9ebfc2ba8d
commit 8da538c18c
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 54 additions and 75 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 62
      selfdrive/modeld/models/commonmodel.cc
  2. 32
      selfdrive/modeld/models/commonmodel.h
  3. 31
      selfdrive/modeld/models/driving.cc
  4. 4
      selfdrive/modeld/models/driving.h

62
selfdrive/modeld/models/commonmodel.cc
Unescape View File

@ -3,51 +3,45 @@
#include <algorithm> #include <algorithm>
#include <cassert> #include <cassert>
#include <cmath> #include <cmath>
#include <cstring>
#include "selfdrive/common/clutil.h" #include "selfdrive/common/clutil.h"
#include "selfdrive/common/mat.h" #include "selfdrive/common/mat.h"
#include "selfdrive/common/timing.h" #include "selfdrive/common/timing.h"
void frame_init(ModelFrame* frame, int width, int height, ModelFrame::ModelFrame(cl_device_id device_id, cl_context context) {
cl_device_id device_id, cl_context context) { input_frames = std::make_unique<float[]>(buf_size);
transform_init(&frame->transform, context, device_id);
frame->width = width;
frame->height = height;
frame->y_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, (size_t)width*height, NULL, &err)); q = CL_CHECK_ERR(clCreateCommandQueue(context, device_id, 0, &err));
frame->u_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, (size_t)(width/2)*(height/2), NULL, &err)); y_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, MODEL_WIDTH * MODEL_HEIGHT, NULL, &err));
frame->v_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, (size_t)(width/2)*(height/2), NULL, &err)); u_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, (MODEL_WIDTH / 2) * (MODEL_HEIGHT / 2), NULL, &err));
frame->net_input_size = ((width*height*3)/2)*sizeof(float); v_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, (MODEL_WIDTH / 2) * (MODEL_HEIGHT / 2), NULL, &err));
frame->net_input = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, net_input_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, MODEL_FRAME_SIZE * sizeof(float), NULL, &err));
frame->net_input_size, (void*)NULL, &err));
loadyuv_init(&frame->loadyuv, context, device_id, width, height); transform_init(&transform, context, device_id);
loadyuv_init(&loadyuv, context, device_id, MODEL_WIDTH, MODEL_HEIGHT);
} }
float *frame_prepare(ModelFrame* frame, cl_command_queue q, float* ModelFrame::prepare(cl_mem yuv_cl, int frame_width, int frame_height, const mat3 &transform) {
cl_mem yuv_cl, int width, int height, transform_queue(&this->transform, q,
const mat3 &transform) { yuv_cl, frame_width, frame_height,
transform_queue(&frame->transform, q, y_cl, u_cl, v_cl, MODEL_WIDTH, MODEL_HEIGHT, transform);
yuv_cl, width, height, loadyuv_queue(&loadyuv, q, y_cl, u_cl, v_cl, net_input_cl);
frame->y_cl, frame->u_cl, frame->v_cl,
frame->width, frame->height, std::memmove(&input_frames[0], &input_frames[MODEL_FRAME_SIZE], sizeof(float) * MODEL_FRAME_SIZE);
transform); clEnqueueReadBuffer(q, net_input_cl, CL_TRUE, 0, MODEL_FRAME_SIZE * sizeof(float), &input_frames[MODEL_FRAME_SIZE], 0, nullptr, nullptr);
loadyuv_queue(&frame->loadyuv, q,
frame->y_cl, frame->u_cl, frame->v_cl,
frame->net_input);
float *net_input_buf = (float *)CL_CHECK_ERR(clEnqueueMapBuffer(q, frame->net_input, CL_TRUE,
CL_MAP_READ, 0, frame->net_input_size,
0, NULL, NULL, &err));
clFinish(q); clFinish(q);
return net_input_buf; return &input_frames[0];
} }
void frame_free(ModelFrame* frame) { ModelFrame::~ModelFrame() {
transform_destroy(&frame->transform); transform_destroy(&transform);
loadyuv_destroy(&frame->loadyuv); loadyuv_destroy(&loadyuv);
CL_CHECK(clReleaseMemObject(frame->net_input)); CL_CHECK(clReleaseMemObject(net_input_cl));
CL_CHECK(clReleaseMemObject(frame->v_cl)); CL_CHECK(clReleaseMemObject(v_cl));
CL_CHECK(clReleaseMemObject(frame->u_cl)); CL_CHECK(clReleaseMemObject(u_cl));
CL_CHECK(clReleaseMemObject(frame->y_cl)); CL_CHECK(clReleaseMemObject(y_cl));
CL_CHECK(clReleaseCommandQueue(q));
} }
void softmax(const float* input, float* output, size_t len) { void softmax(const float* input, float* output, size_t len) {

32
selfdrive/modeld/models/commonmodel.h
Unescape View File

@ -3,6 +3,8 @@
#include <cfloat> #include <cfloat>
#include <cstdlib> #include <cstdlib>
#include <memory>
#define CL_USE_DEPRECATED_OPENCL_1_2_APIS #define CL_USE_DEPRECATED_OPENCL_1_2_APIS
#ifdef __APPLE__ #ifdef __APPLE__
#include <OpenCL/cl.h> #include <OpenCL/cl.h>
@ -14,24 +16,28 @@
#include "selfdrive/modeld/transforms/loadyuv.h" #include "selfdrive/modeld/transforms/loadyuv.h"
#include "selfdrive/modeld/transforms/transform.h" #include "selfdrive/modeld/transforms/transform.h"
constexpr int MODEL_WIDTH = 512;
constexpr int MODEL_HEIGHT = 256;
constexpr int MODEL_FRAME_SIZE = MODEL_WIDTH * MODEL_HEIGHT * 3 / 2;
const bool send_raw_pred = getenv("SEND_RAW_PRED") != NULL; const bool send_raw_pred = getenv("SEND_RAW_PRED") != NULL;
void softmax(const float* input, float* output, size_t len); void softmax(const float* input, float* output, size_t len);
float softplus(float input); float softplus(float input);
float sigmoid(float input); float sigmoid(float input);
typedef struct ModelFrame { class ModelFrame {
public:
ModelFrame(cl_device_id device_id, cl_context context);
~ModelFrame();
float* prepare(cl_mem yuv_cl, int width, int height, const mat3& transform);
const int buf_size = MODEL_FRAME_SIZE * 2;
private:
Transform transform; Transform transform;
int width, height;
cl_mem y_cl, u_cl, v_cl;
LoadYUVState loadyuv; LoadYUVState loadyuv;
cl_mem net_input; cl_command_queue q;
size_t net_input_size; cl_mem y_cl, u_cl, v_cl, net_input_cl;
} ModelFrame; std::unique_ptr<float[]> input_frames;
};
void frame_init(ModelFrame* frame, int width, int height,
cl_device_id device_id, cl_context context);
float *frame_prepare(ModelFrame* frame, cl_command_queue q,
cl_mem yuv_cl, int width, int height,
const mat3 &transform);
void frame_free(ModelFrame* frame);

31
selfdrive/modeld/models/driving.cc
Unescape View File

@ -17,10 +17,6 @@ constexpr int OTHER_META_SIZE = 32;
constexpr int NUM_META_INTERVALS = 5; constexpr int NUM_META_INTERVALS = 5;
constexpr int META_STRIDE = 6; constexpr int META_STRIDE = 6;
constexpr int MODEL_WIDTH = 512;
constexpr int MODEL_HEIGHT = 256;
constexpr int MODEL_FRAME_SIZE = MODEL_WIDTH * MODEL_HEIGHT * 3 / 2;
constexpr int PLAN_MHP_N = 5; constexpr int PLAN_MHP_N = 5;
constexpr int PLAN_MHP_COLUMNS = 15; constexpr int PLAN_MHP_COLUMNS = 15;
constexpr int PLAN_MHP_VALS = 15*33; constexpr int PLAN_MHP_VALS = 15*33;
@ -60,8 +56,7 @@ float prev_brake_3ms2_probs[3] = {0,0,0};
// #define DUMP_YUV // #define DUMP_YUV
void model_init(ModelState* s, cl_device_id device_id, cl_context context) { void model_init(ModelState* s, cl_device_id device_id, cl_context context) {
frame_init(&s->frame, MODEL_WIDTH, MODEL_HEIGHT, device_id, context); s->frame = new ModelFrame(device_id, context);
s->input_frames = std::make_unique<float[]>(MODEL_FRAME_SIZE * 2);
constexpr int output_size = OUTPUT_SIZE + TEMPORAL_SIZE; constexpr int output_size = OUTPUT_SIZE + TEMPORAL_SIZE;
s->output.resize(output_size); s->output.resize(output_size);
@ -85,8 +80,6 @@ void model_init(ModelState* s, cl_device_id device_id, cl_context context) {
s->traffic_convention[idx] = 1.0; s->traffic_convention[idx] = 1.0;
s->m->addTrafficConvention(s->traffic_convention, TRAFFIC_CONVENTION_LEN); s->m->addTrafficConvention(s->traffic_convention, TRAFFIC_CONVENTION_LEN);
#endif #endif
s->q = CL_CHECK_ERR(clCreateCommandQueue(context, device_id, 0, &err));
} }
ModelDataRaw model_eval_frame(ModelState* s, cl_mem yuv_cl, int width, int height, ModelDataRaw model_eval_frame(ModelState* s, cl_mem yuv_cl, int width, int height,
@ -108,19 +101,8 @@ ModelDataRaw model_eval_frame(ModelState* s, cl_mem yuv_cl, int width, int heigh
//for (int i = 0; i < OUTPUT_SIZE + TEMPORAL_SIZE; i++) { printf("%f ", s->output[i]); } printf("\n"); //for (int i = 0; i < OUTPUT_SIZE + TEMPORAL_SIZE; i++) { printf("%f ", s->output[i]); } printf("\n");
float *new_frame_buf = frame_prepare(&s->frame, s->q, yuv_cl, width, height, transform); auto net_input_buf = s->frame->prepare(yuv_cl, width, height, transform);
memmove(&s->input_frames[0], &s->input_frames[MODEL_FRAME_SIZE], sizeof(float)*MODEL_FRAME_SIZE); s->m->execute(net_input_buf, s->frame->buf_size);
memmove(&s->input_frames[MODEL_FRAME_SIZE], new_frame_buf, sizeof(float)*MODEL_FRAME_SIZE);
s->m->execute(&s->input_frames[0], MODEL_FRAME_SIZE*2);
#ifdef DUMP_YUV
FILE *dump_yuv_file = fopen("/sdcard/dump.yuv", "wb");
fwrite(new_frame_buf, MODEL_HEIGHT*MODEL_WIDTH*3/2, sizeof(float), dump_yuv_file);
fclose(dump_yuv_file);
assert(1==2);
#endif
clEnqueueUnmapMemObject(s->q, s->frame.net_input, (void*)new_frame_buf, 0, NULL, NULL);
// net outputs // net outputs
ModelDataRaw net_outputs; ModelDataRaw net_outputs;
@ -136,8 +118,7 @@ ModelDataRaw model_eval_frame(ModelState* s, cl_mem yuv_cl, int width, int heigh
} }
void model_free(ModelState* s) { void model_free(ModelState* s) {
frame_free(&s->frame); delete s->frame;
CL_CHECK(clReleaseCommandQueue(s->q));
} }
static const float *get_best_data(const float *data, int size, int group_size, int offset) { static const float *get_best_data(const float *data, int size, int group_size, int offset) {
@ -203,8 +184,8 @@ void fill_meta(cereal::ModelDataV2::MetaData::Builder meta, const float *meta_da
fill_sigmoid(&meta_data[DESIRE_LEN+5], brake_4ms2_sigmoid, NUM_META_INTERVALS, META_STRIDE); fill_sigmoid(&meta_data[DESIRE_LEN+5], brake_4ms2_sigmoid, NUM_META_INTERVALS, META_STRIDE);
fill_sigmoid(&meta_data[DESIRE_LEN+6], brake_5ms2_sigmoid, NUM_META_INTERVALS, META_STRIDE); fill_sigmoid(&meta_data[DESIRE_LEN+6], brake_5ms2_sigmoid, NUM_META_INTERVALS, META_STRIDE);
memmove(prev_brake_5ms2_probs, &prev_brake_5ms2_probs[1], 4*sizeof(float)); std::memmove(prev_brake_5ms2_probs, &prev_brake_5ms2_probs[1], 4*sizeof(float));
memmove(prev_brake_3ms2_probs, &prev_brake_3ms2_probs[1], 2*sizeof(float)); std::memmove(prev_brake_3ms2_probs, &prev_brake_3ms2_probs[1], 2*sizeof(float));
prev_brake_5ms2_probs[4] = brake_5ms2_sigmoid[0]; prev_brake_5ms2_probs[4] = brake_5ms2_sigmoid[0];
prev_brake_3ms2_probs[2] = brake_3ms2_sigmoid[0]; prev_brake_3ms2_probs[2] = brake_3ms2_sigmoid[0];

4
selfdrive/modeld/models/driving.h
Unescape View File

@ -32,11 +32,9 @@ struct ModelDataRaw {
}; };
typedef struct ModelState { typedef struct ModelState {
ModelFrame frame; ModelFrame *frame;
std::vector<float> output; std::vector<float> output;
std::unique_ptr<float[]> input_frames;
std::unique_ptr<RunModel> m; std::unique_ptr<RunModel> m;
cl_command_queue q;
#ifdef DESIRE #ifdef DESIRE
float prev_desire[DESIRE_LEN] = {}; float prev_desire[DESIRE_LEN] = {};
float pulse_desire[DESIRE_LEN] = {}; float pulse_desire[DESIRE_LEN] = {};

Write Preview
Loading…
Cancel
Save