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

rewritten debayer kernel

Browse Source
pull/2047/head
Tici Engineer 5 years ago
parent f8745d3002
commit 09b7960874
2 changed files with 212 additions and 44 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. 136
      selfdrive/camerad/cameras/real_debayer.cl
  2. 118
      selfdrive/camerad/main.cc

136
selfdrive/camerad/cameras/real_debayer.cl
Unescape View File

@ -0,0 +1,136 @@
const __constant float3 color_correction[3] = {
// Matrix from WBraw -> sRGBD65 (normalized)
(float3)(1.18409513, -0.15969593, 0.04764678),
(float3)(-0.18071031, 1.13545802, 0.42800657),
(float3)(-0.00338482, 0.02423791, 0.52434665),
};
const __constant float3 df = (float3)(0.5, 0.5, 0.5);
float3 color_correct(float r, float g, float b) {
float3 ret = (0,0,0);
// white balance of daylight
// x /= (float3)(0.4609375, 1.0, 0.546875);
// x /= (float3)(1.0, 1.0, 1.0);
//r = max(0.0, min(1.0, x));
// fix up the colors
ret += r * color_correction[0];
ret += g * color_correction[1];
ret += b * color_correction[2];
ret = max(0.0, min(1.0, ret));
return ret;
}
uint int_from_10(const uchar * source, uint start, uint offset) {
// source: source
// start: starting address of 0
// offset: 0 - 3
uint major = (uint)source[start + offset] << 2;
uint minor = (source[start + 4] >> (2 * offset)) & 3;
return major + minor;
}
float to_normal(uint x) {
float pv = (float)(x);
const float black_level = 0;
pv = max(0.0, pv - black_level);
pv /= (1024.0f - black_level);
pv = 2 * pv / (1.0f + pv); // reinhard
return pv;
}
__kernel void debayer10(const __global uchar * in,
__global uchar * out,
__local float * cached
)
{
const int x_global = get_global_id(0);
const int y_global = get_global_id(1);
// const int globalOffset = ;
const int localRowLen = 2 + get_local_size(0); // 2 padding
const int x_local = get_local_id(0);
const int y_local = get_local_id(1);
const int localOffset = (y_local + 1) * localRowLen + x_local + 1;
// cache local pixels first
// saves memory access and avoids repeated normalization
uint globalStart_10 = y_global * FRAME_STRIDE + (5 * (x_global / 4));
uint offset_10 = x_global % 4;
uint raw_val = int_from_10(in, globalStart_10, offset_10);
cached[localOffset] = to_normal(raw_val);
// edges
if (x_global < 1 || x_global > FRAME_STRIDE - 2 || y_global < 1 || y_global > FRAME_HEIGHT - 2) {
barrier(CLK_LOCAL_MEM_FENCE);
return;
} else {
int localColOffset = -1;
int globalColOffset = -1;
// cache padding
if (x_local < 1) {
localColOffset = x_local;
globalColOffset = -1;
cached[(y_local + 1) * localRowLen + x_local] = to_normal(int_from_10(in, y_global * FRAME_STRIDE + (5 * ((x_global-1) / 4)), (offset_10 + 3) % 4));
//cached[(y_local + 1) * localRowLen + x_local] = to_normal(int_from_10(in, globalStart_10, offset_10 - 1)); // this assumes local work size is multipy of 4
} else if (x_local >= get_local_size(0) - 1) {
localColOffset = x_local + 2;
globalColOffset = 1;
cached[localOffset + 1] = to_normal(int_from_10(in, globalStart_10, offset_10 + 1));
}
if (y_local < 1) {
cached[y_local * localRowLen + x_local + 1] = to_normal(int_from_10(in, globalStart_10 - FRAME_STRIDE, offset_10));
if (localColOffset > 0) {
cached[y_local * localRowLen + localColOffset] = to_normal(int_from_10(in, (y_global-1) * FRAME_STRIDE + (5 * ((globalColOffset) / 4)), globalColOffset % 4));
}
} else if (y_local >= get_local_size(1) - 1) {
cached[(y_local + 2) * localRowLen + x_local + 1] = to_normal(int_from_10(in, globalStart_10 + FRAME_STRIDE, offset_10));
if (localColOffset > 0) {
cached[(y_local + 2) * localRowLen + localColOffset] = to_normal(int_from_10(in, (y_global+1) * FRAME_STRIDE + (5 * ((globalColOffset) / 4)), globalColOffset % 4));
}
}
// sync
barrier(CLK_LOCAL_MEM_FENCE);
// perform debayer
float r;
float g;
float b;
if (x_global % 2 == 0) {
if (y_global % 2 == 0) { // G1
r = (cached[localOffset - 1] + cached[localOffset + 1]) / 2.0f;
g = cached[localOffset];
b = (cached[localOffset - get_local_size(0) - 2] + cached[localOffset + get_local_size(0) + 2]) / 2.0f;
} else { // B
r = (cached[localOffset - get_local_size(0) - 3] + cached[localOffset - get_local_size(0) - 1] + cached[localOffset + get_local_size(0) + 1] + cached[localOffset + get_local_size(0) + 3]) / 4.0f;
g = (cached[localOffset - get_local_size(0) - 2] + cached[localOffset + get_local_size(0) + 2] + cached[localOffset - 1] + cached[localOffset + 1]) / 4.0f;
b = cached[localOffset];
}
} else {
if (y_global % 2 == 0) { // R
r = cached[localOffset];
g = (cached[localOffset - get_local_size(0) - 2] + cached[localOffset + get_local_size(0) + 2] + cached[localOffset - 1] + cached[localOffset + 1]) / 4.0f;
b = (cached[localOffset - get_local_size(0) - 3] + cached[localOffset - get_local_size(0) - 1] + cached[localOffset + get_local_size(0) + 1] + cached[localOffset + get_local_size(0) + 3]) / 4.0f;
} else { // G2
r = (cached[localOffset - get_local_size(0) - 2] + cached[localOffset + get_local_size(0) + 2]) / 2.0f;
g = cached[localOffset];
b = (cached[localOffset - 1] + cached[localOffset + 1]) / 2.0f;
}
}
float3 rgb = color_correct(r, g, b);
// rgb = srgb_gamma(rgb);
// BGR output
out[3 * x_global + 3 * y_global * RGB_WIDTH + 0] = (uchar)(255.0f * rgb.z);
out[3 * x_global + 3 * y_global * RGB_WIDTH + 1] = (uchar)(255.0f * rgb.y);
out[3 * x_global + 3 * y_global * RGB_WIDTH + 2] = (uchar)(255.0f * rgb.x);
}
}

118
selfdrive/camerad/main.cc
Unescape View File

@ -34,6 +34,7 @@
#include "cereal/gen/cpp/log.capnp.h" #include "cereal/gen/cpp/log.capnp.h"
#define UI_BUF_COUNT 4 #define UI_BUF_COUNT 4
#define DEBAYER_LOCAL_WORKSIZE 16
#define YUV_COUNT 40 #define YUV_COUNT 40
#define MAX_CLIENTS 5 #define MAX_CLIENTS 5
@ -79,6 +80,9 @@ struct VisionState {
cl_kernel krnl_debayer_rear; cl_kernel krnl_debayer_rear;
cl_kernel krnl_debayer_front; cl_kernel krnl_debayer_front;
cl_kernel krnl_debayer_wide; cl_kernel krnl_debayer_wide;
int debayer_cl_localMemSize;
size_t debayer_cl_globalWorkSize[2];
size_t debayer_cl_localWorkSize[2];
cl_program prg_rgb_laplacian; cl_program prg_rgb_laplacian;
cl_kernel krnl_rgb_laplacian; cl_kernel krnl_rgb_laplacian;
@ -212,14 +216,20 @@ void* frontview_thread(void *arg) {
assert(err == 0); assert(err == 0);
err = clSetKernelArg(s->krnl_debayer_front, 1, sizeof(cl_mem), &s->rgb_front_bufs_cl[rgb_idx]); err = clSetKernelArg(s->krnl_debayer_front, 1, sizeof(cl_mem), &s->rgb_front_bufs_cl[rgb_idx]);
assert(err == 0); assert(err == 0);
#ifdef QCOM2
err = clSetKernelArg(s->krnl_debayer_front, 2, s->debayer_cl_localMemSize, 0);
assert(err == 0);
err = clEnqueueNDRangeKernel(q, s->krnl_debayer_front, 2, NULL,
s->debayer_cl_globalWorkSize, s->debayer_cl_localWorkSize, 0, 0, &debayer_event);
#else
float digital_gain = 1.0; float digital_gain = 1.0;
err = clSetKernelArg(s->krnl_debayer_front, 2, sizeof(float), &digital_gain); err = clSetKernelArg(s->krnl_debayer_front, 2, sizeof(float), &digital_gain);
assert(err == 0); assert(err == 0);
const size_t debayer_work_size = s->rgb_front_height; // doesn't divide evenly, is this okay? const size_t debayer_work_size = s->rgb_front_height; // doesn't divide evenly, is this okay?
const size_t debayer_local_work_size = 128; const size_t debayer_local_work_size = 128;
err = clEnqueueNDRangeKernel(q, s->krnl_debayer_front, 1, NULL, err = clEnqueueNDRangeKernel(q, s->krnl_debayer_front, 1, NULL,
&debayer_work_size, &debayer_local_work_size, 0, 0, &debayer_event); &debayer_work_size, &debayer_local_work_size, 0, 0, &debayer_event);
#endif
assert(err == 0); assert(err == 0);
} else { } else {
assert(s->rgb_front_buf_size >= s->cameras.front.frame_size); assert(s->rgb_front_buf_size >= s->cameras.front.frame_size);
@ -251,7 +261,7 @@ void* frontview_thread(void *arg) {
} }
#ifdef NOSCREEN #ifdef NOSCREEN
if (frame_data.frame_id % 2 == 0) { if (frame_data.frame_id % 4 == 2) {
sendrgb(&s->cameras, (uint8_t*) s->rgb_front_bufs[rgb_idx].addr, s->rgb_front_bufs[rgb_idx].len, 2); sendrgb(&s->cameras, (uint8_t*) s->rgb_front_bufs[rgb_idx].addr, s->rgb_front_bufs[rgb_idx].len, 2);
} }
#endif #endif
@ -316,10 +326,10 @@ void* frontview_thread(void *arg) {
x_end = s->rhd_front ? s->rgb_front_width * 2 / 5:s->rgb_front_width; x_end = s->rhd_front ? s->rgb_front_width * 2 / 5:s->rgb_front_width;
} }
#ifdef QCOM2 #ifdef QCOM2
x_start = 0.2*s->rgb_front_width; x_start = 0.1*s->rgb_front_width;
x_end = 0.8*s->rgb_front_width; x_end = 0.9*s->rgb_front_width;
y_start = 0.2*s->rgb_front_height; y_start = 0.1*s->rgb_front_height;
y_end = 0.8*s->rgb_front_height; y_end = 0.9*s->rgb_front_height;
#endif #endif
uint32_t lum_binning[256] = {0,}; uint32_t lum_binning[256] = {0,};
for (int y = y_start; y < y_end; ++y) { for (int y = y_start; y < y_end; ++y) {
@ -447,17 +457,14 @@ void* wideview_thread(void *arg) {
cl_event debayer_event; cl_event debayer_event;
if (s->cameras.wide.ci.bayer) { if (s->cameras.wide.ci.bayer) {
err = clSetKernelArg(s->krnl_debayer_rear, 0, sizeof(cl_mem), &s->wide_camera_bufs_cl[buf_idx]); // TODO: add separate cl krl err = clSetKernelArg(s->krnl_debayer_wide, 0, sizeof(cl_mem), &s->wide_camera_bufs_cl[buf_idx]);
assert(err == 0); assert(err == 0);
err = clSetKernelArg(s->krnl_debayer_rear, 1, sizeof(cl_mem), &s->rgb_wide_bufs_cl[rgb_idx]); err = clSetKernelArg(s->krnl_debayer_wide, 1, sizeof(cl_mem), &s->rgb_wide_bufs_cl[rgb_idx]);
assert(err == 0); assert(err == 0);
err = clSetKernelArg(s->krnl_debayer_rear, 2, sizeof(float), &s->cameras.wide.digital_gain); err = clSetKernelArg(s->krnl_debayer_wide, 2, s->debayer_cl_localMemSize, 0);
assert(err == 0); assert(err == 0);
err = clEnqueueNDRangeKernel(q, s->krnl_debayer_wide, 2, NULL,
const size_t debayer_work_size = s->rgb_height; // doesn't divide evenly, is this okay? s->debayer_cl_globalWorkSize, s->debayer_cl_localWorkSize, 0, 0, &debayer_event);
const size_t debayer_local_work_size = 128;
err = clEnqueueNDRangeKernel(q, s->krnl_debayer_rear, 1, NULL,
&debayer_work_size, &debayer_local_work_size, 0, 0, &debayer_event);
assert(err == 0); assert(err == 0);
} else { } else {
assert(s->rgb_wide_buf_size >= s->frame_size); assert(s->rgb_wide_buf_size >= s->frame_size);
@ -475,7 +482,7 @@ void* wideview_thread(void *arg) {
visionbuf_sync(&s->rgb_wide_bufs[rgb_idx], VISIONBUF_SYNC_FROM_DEVICE); visionbuf_sync(&s->rgb_wide_bufs[rgb_idx], VISIONBUF_SYNC_FROM_DEVICE);
#ifdef NOSCREEN #ifdef NOSCREEN
if (frame_data.frame_id % 2 == 0) { if (frame_data.frame_id % 4 == 0) {
sendrgb(&s->cameras, (uint8_t*) s->rgb_wide_bufs[rgb_idx].addr, s->rgb_wide_bufs[rgb_idx].len, 1); sendrgb(&s->cameras, (uint8_t*) s->rgb_wide_bufs[rgb_idx].addr, s->rgb_wide_bufs[rgb_idx].len, 1);
} }
#endif #endif
@ -532,17 +539,17 @@ void* wideview_thread(void *arg) {
auto words = capnp::messageToFlatArray(msg); auto words = capnp::messageToFlatArray(msg);
auto bytes = words.asBytes(); auto bytes = words.asBytes();
s->frame_sock->send((char*)bytes.begin(), bytes.size()); s->wide_frame_sock->send((char*)bytes.begin(), bytes.size());
} }
} }
tbuffer_dispatch(&s->ui_wide_tb, ui_idx); tbuffer_dispatch(&s->ui_wide_tb, ui_idx);
// auto exposure over big box // auto exposure over big box
const int exposure_x = 2;//290; const int exposure_x = 120*2;//290;
const int exposure_y = 200;//282 + 40; // TODO: fix this? should not use med imo const int exposure_y = 180*2;//282 + 40; // TODO: fix this? should not use med imo
const int exposure_height = 250;// 314; const int exposure_height = 180*2;// 314;
const int exposure_width = 450;//560; const int exposure_width = 720*2;//560;
if (cnt % 3 == 0) { if (cnt % 3 == 0) {
// find median box luminance for AE // find median box luminance for AE
uint32_t lum_binning[256] = {0,}; uint32_t lum_binning[256] = {0,};
@ -615,19 +622,26 @@ void* processing_thread(void *arg) {
int ui_idx = tbuffer_select(&s->ui_tb); int ui_idx = tbuffer_select(&s->ui_tb);
int rgb_idx = ui_idx; int rgb_idx = ui_idx;
// double t10 = millis_since_boot();
cl_event debayer_event; cl_event debayer_event;
if (s->cameras.rear.ci.bayer) { if (s->cameras.rear.ci.bayer) {
err = clSetKernelArg(s->krnl_debayer_rear, 0, sizeof(cl_mem), &s->camera_bufs_cl[buf_idx]); err = clSetKernelArg(s->krnl_debayer_rear, 0, sizeof(cl_mem), &s->camera_bufs_cl[buf_idx]);
assert(err == 0); assert(err == 0);
err = clSetKernelArg(s->krnl_debayer_rear, 1, sizeof(cl_mem), &s->rgb_bufs_cl[rgb_idx]); err = clSetKernelArg(s->krnl_debayer_rear, 1, sizeof(cl_mem), &s->rgb_bufs_cl[rgb_idx]);
assert(err == 0); assert(err == 0);
#ifdef QCOM2
err = clSetKernelArg(s->krnl_debayer_rear, 2, s->debayer_cl_localMemSize, 0);
assert(err == 0);
err = clEnqueueNDRangeKernel(q, s->krnl_debayer_rear, 2, NULL,
s->debayer_cl_globalWorkSize, s->debayer_cl_localWorkSize, 0, 0, &debayer_event);
#else
err = clSetKernelArg(s->krnl_debayer_rear, 2, sizeof(float), &s->cameras.rear.digital_gain); err = clSetKernelArg(s->krnl_debayer_rear, 2, sizeof(float), &s->cameras.rear.digital_gain);
assert(err == 0); assert(err == 0);
const size_t debayer_work_size = s->rgb_height; // doesn't divide evenly, is this okay? const size_t debayer_work_size = s->rgb_height; // doesn't divide evenly, is this okay?
const size_t debayer_local_work_size = 128; const size_t debayer_local_work_size = 128;
err = clEnqueueNDRangeKernel(q, s->krnl_debayer_rear, 1, NULL, err = clEnqueueNDRangeKernel(q, s->krnl_debayer_rear, 1, NULL,
&debayer_work_size, &debayer_local_work_size, 0, 0, &debayer_event); &debayer_work_size, &debayer_local_work_size, 0, 0, &debayer_event);
#endif
assert(err == 0); assert(err == 0);
} else { } else {
assert(s->rgb_buf_size >= s->frame_size); assert(s->rgb_buf_size >= s->frame_size);
@ -636,16 +650,18 @@ void* processing_thread(void *arg) {
0, 0, s->rgb_buf_size, 0, 0, &debayer_event); 0, 0, s->rgb_buf_size, 0, 0, &debayer_event);
assert(err == 0); assert(err == 0);
} }
clWaitForEvents(1, &debayer_event); clWaitForEvents(1, &debayer_event);
clReleaseEvent(debayer_event); clReleaseEvent(debayer_event);
// double t11 = millis_since_boot();
// printf("kernel time: %f ms\n", t11 - t10);
tbuffer_release(&s->cameras.rear.camera_tb, buf_idx); tbuffer_release(&s->cameras.rear.camera_tb, buf_idx);
visionbuf_sync(&s->rgb_bufs[rgb_idx], VISIONBUF_SYNC_FROM_DEVICE); visionbuf_sync(&s->rgb_bufs[rgb_idx], VISIONBUF_SYNC_FROM_DEVICE);
#ifdef NOSCREEN #ifdef NOSCREEN
if (frame_data.frame_id % 2 == 1) { if (frame_data.frame_id % 4 == 1) {
sendrgb(&s->cameras, (uint8_t*) s->rgb_bufs[rgb_idx].addr, s->rgb_bufs[rgb_idx].len, 0); sendrgb(&s->cameras, (uint8_t*) s->rgb_bufs[rgb_idx].addr, s->rgb_bufs[rgb_idx].len, 0);
} }
#endif #endif
@ -856,10 +872,10 @@ void* processing_thread(void *arg) {
tbuffer_dispatch(&s->ui_tb, ui_idx); tbuffer_dispatch(&s->ui_tb, ui_idx);
// auto exposure over big box // auto exposure over big box
const int exposure_x = 2; const int exposure_x = 120*2;
const int exposure_y = 200;// + 40; const int exposure_y = 180*2;// + 40;
const int exposure_height = 250;// 314; const int exposure_height = 180*2;// 314;
const int exposure_width = 450;//560; const int exposure_width = 720*2;//560;
if (cnt % 3 == 0) { if (cnt % 3 == 0) {
// find median box luminance for AE // find median box luminance for AE
uint32_t lum_binning[256] = {0,}; uint32_t lum_binning[256] = {0,};
@ -1222,8 +1238,8 @@ cl_program build_debayer_program(VisionState *s,
int frame_width, int frame_height, int frame_stride, int frame_width, int frame_height, int frame_stride,
int rgb_width, int rgb_height, int rgb_stride, int rgb_width, int rgb_height, int rgb_stride,
int bayer_flip, int hdr) { int bayer_flip, int hdr) {
assert(rgb_width == frame_width/2); assert(rgb_width == frame_width);
assert(rgb_height == frame_height/2); assert(rgb_height == frame_height);
#ifdef QCOM2 #ifdef QCOM2
int dnew = 1; int dnew = 1;
@ -1240,7 +1256,11 @@ cl_program build_debayer_program(VisionState *s,
frame_width, frame_height, frame_stride, frame_width, frame_height, frame_stride,
rgb_width, rgb_height, rgb_stride, rgb_width, rgb_height, rgb_stride,
bayer_flip, hdr, dnew); bayer_flip, hdr, dnew);
#ifdef QCOM2
return CLU_LOAD_FROM_FILE(s->context, s->device_id, "cameras/real_debayer.cl", args);
#else
return CLU_LOAD_FROM_FILE(s->context, s->device_id, "cameras/debayer.cl", args); return CLU_LOAD_FROM_FILE(s->context, s->device_id, "cameras/debayer.cl", args);
#endif
} }
cl_program build_conv_program(VisionState *s, cl_program build_conv_program(VisionState *s,
@ -1305,7 +1325,7 @@ void init_buffers(VisionState *s) {
for (int i=0; i<FRAME_BUF_COUNT; i++) { for (int i=0; i<FRAME_BUF_COUNT; i++) {
s->camera_bufs[i] = visionbuf_allocate_cl(s->frame_size, s->device_id, s->context, s->camera_bufs[i] = visionbuf_allocate_cl(s->frame_size, s->device_id, s->context,
&s->camera_bufs_cl[i]); &s->camera_bufs_cl[i]);
#ifndef QCOM2 #ifdef QCOM
// TODO: make lengths correct // TODO: make lengths correct
s->focus_bufs[i] = visionbuf_allocate(0xb80); s->focus_bufs[i] = visionbuf_allocate(0xb80);
s->stats_bufs[i] = visionbuf_allocate(0xb80); s->stats_bufs[i] = visionbuf_allocate(0xb80);
@ -1326,8 +1346,13 @@ void init_buffers(VisionState *s) {
// processing buffers // processing buffers
if (s->cameras.rear.ci.bayer) { if (s->cameras.rear.ci.bayer) {
s->rgb_width = s->frame_width/2; #ifdef QCOM2
s->rgb_height = s->frame_height/2; s->rgb_width = s->frame_width;
s->rgb_height = s->frame_height;
#else
s->rgb_width = s->frame_width / 2;
s->rgb_height = s->frame_height / 2;
#endif
} else { } else {
s->rgb_width = s->frame_width; s->rgb_width = s->frame_width;
s->rgb_height = s->frame_height; s->rgb_height = s->frame_height;
@ -1345,21 +1370,20 @@ void init_buffers(VisionState *s) {
//assert(s->cameras.front.ci.bayer); //assert(s->cameras.front.ci.bayer);
if (s->cameras.front.ci.bayer) { if (s->cameras.front.ci.bayer) {
s->rgb_front_width = s->cameras.front.ci.frame_width/2; #ifdef QCOM2
s->rgb_front_height = s->cameras.front.ci.frame_height/2;
} else {
s->rgb_front_width = s->cameras.front.ci.frame_width; s->rgb_front_width = s->cameras.front.ci.frame_width;
s->rgb_front_height = s->cameras.front.ci.frame_height; s->rgb_front_height = s->cameras.front.ci.frame_height;
} #else
s->rgb_front_width = s->cameras.front.ci.frame_width / 2;
if (s->cameras.wide.ci.bayer) { s->rgb_front_height = s->cameras.front.ci.frame_height / 2;
s->rgb_wide_width = s->cameras.wide.ci.frame_width/2; #endif
s->rgb_wide_height = s->cameras.wide.ci.frame_height/2;
} else { } else {
s->rgb_wide_width = s->cameras.wide.ci.frame_width; s->rgb_front_width = s->cameras.front.ci.frame_width;
s->rgb_wide_height = s->cameras.wide.ci.frame_height; s->rgb_front_height = s->cameras.front.ci.frame_height;
} }
s->rgb_wide_width = s->cameras.wide.ci.frame_width;
s->rgb_wide_height = s->cameras.wide.ci.frame_height;
for (int i=0; i<UI_BUF_COUNT; i++) { for (int i=0; i<UI_BUF_COUNT; i++) {
VisionImg img = visionimg_alloc_rgb24(s->rgb_front_width, s->rgb_front_height, &s->rgb_front_bufs[i]); VisionImg img = visionimg_alloc_rgb24(s->rgb_front_width, s->rgb_front_height, &s->rgb_front_bufs[i]);
@ -1463,7 +1487,13 @@ void init_buffers(VisionState *s) {
s->krnl_debayer_wide = clCreateKernel(s->prg_debayer_wide, "debayer10", &err); s->krnl_debayer_wide = clCreateKernel(s->prg_debayer_wide, "debayer10", &err);
assert(err == 0); assert(err == 0);
} }
s->debayer_cl_localMemSize = (DEBAYER_LOCAL_WORKSIZE + 2 * (3 / 2)) * (DEBAYER_LOCAL_WORKSIZE + 2 * (3 / 2)) * sizeof(float);
s->debayer_cl_globalWorkSize[0] = s->rgb_width;
s->debayer_cl_globalWorkSize[1] = s->rgb_height;
s->debayer_cl_localWorkSize[0] = DEBAYER_LOCAL_WORKSIZE;
s->debayer_cl_localWorkSize[1] = DEBAYER_LOCAL_WORKSIZE;
#ifdef QCOM
s->prg_rgb_laplacian = build_conv_program(s, s->rgb_width/NUM_SEGMENTS_X, s->rgb_height/NUM_SEGMENTS_Y, s->prg_rgb_laplacian = build_conv_program(s, s->rgb_width/NUM_SEGMENTS_X, s->rgb_height/NUM_SEGMENTS_Y,
3); 3);
s->krnl_rgb_laplacian = clCreateKernel(s->prg_rgb_laplacian, "rgb2gray_conv2d", &err); s->krnl_rgb_laplacian = clCreateKernel(s->prg_rgb_laplacian, "rgb2gray_conv2d", &err);
@ -1482,6 +1512,7 @@ void init_buffers(VisionState *s) {
s->conv_cl_localWorkSize[1] = CONV_LOCAL_WORKSIZE; s->conv_cl_localWorkSize[1] = CONV_LOCAL_WORKSIZE;
for (int i=0; i<(ROI_X_MAX-ROI_X_MIN+1)*(ROI_Y_MAX-ROI_Y_MIN+1); i++) {s->lapres[i] = 16160;} for (int i=0; i<(ROI_X_MAX-ROI_X_MIN+1)*(ROI_Y_MAX-ROI_Y_MIN+1); i++) {s->lapres[i] = 16160;}
#endif
rgb_to_yuv_init(&s->rgb_to_yuv_state, s->context, s->device_id, s->yuv_width, s->yuv_height, s->rgb_stride); rgb_to_yuv_init(&s->rgb_to_yuv_state, s->context, s->device_id, s->yuv_width, s->yuv_height, s->rgb_stride);
rgb_to_yuv_init(&s->front_rgb_to_yuv_state, s->context, s->device_id, s->yuv_front_width, s->yuv_front_height, s->rgb_front_stride); rgb_to_yuv_init(&s->front_rgb_to_yuv_state, s->context, s->device_id, s->yuv_front_width, s->yuv_front_height, s->rgb_front_stride);
@ -1492,8 +1523,10 @@ void free_buffers(VisionState *s) {
// free bufs // free bufs
for (int i=0; i<FRAME_BUF_COUNT; i++) { for (int i=0; i<FRAME_BUF_COUNT; i++) {
visionbuf_free(&s->camera_bufs[i]); visionbuf_free(&s->camera_bufs[i]);
#ifdef QCOM
visionbuf_free(&s->focus_bufs[i]); visionbuf_free(&s->focus_bufs[i]);
visionbuf_free(&s->stats_bufs[i]); visionbuf_free(&s->stats_bufs[i]);
#endif
} }
for (int i=0; i<FRAME_BUF_COUNT; i++) { for (int i=0; i<FRAME_BUF_COUNT; i++) {
@ -1504,7 +1537,6 @@ void free_buffers(VisionState *s) {
visionbuf_free(&s->wide_camera_bufs[i]); visionbuf_free(&s->wide_camera_bufs[i]);
} }
for (int i=0; i<UI_BUF_COUNT; i++) { for (int i=0; i<UI_BUF_COUNT; i++) {
visionbuf_free(&s->rgb_bufs[i]); visionbuf_free(&s->rgb_bufs[i]);
} }

Write Preview
Loading…
Cancel
Save