soc模式下使用bmcv_image_draw_lines绘制线条太慢了

[RaechelliaYao]

No description provided.

[RaechelliaYao]

代码如下 使用tpu绘制速度比cpu慢了好多个量级

static bm_status_t bmcv_draw_line(bm_handle_t handle, u64 base_addr, int pixel_stride, int row_stride, int height, int width, bmcv_point_t start_point, bmcv_point_t end_point, u8 val)
{
	// Bresenham 直线算法
	int dx = abs(end_point.x - start_point.x);
	int dy = abs(end_point.y - start_point.y);
	int sx, sy;

	if (start_point.x < end_point.x) {
		sx = 1;
	}
	else {
		sx = -1;
	}

	if (start_point.y < end_point.y) {
		sy = 1;
	}
	else {
		sy = -1;
	}

	int err = dx - dy;
	int x = start_point.x;
	int y = start_point.y;

	while (true) {
		// 绘制当前点
		if (x >= 0 && x < width && y >= 0 && y < height) {
			u64 offset = x * pixel_stride + y * row_stride;
			base_addr = base_addr + offset;
			bm_api_memset_byte_t api = { base_addr, 1, 1, 1, 1, 1, 1, 1, pixel_stride, val };

			if (BM_SUCCESS != bm_send_api(handle, (bm_api_id_t)177, (uint8_t*)&api, sizeof(api)))
			{
				std::cout << "draw line send api error\r\n" << std::endl;
				return BM_ERR_FAILURE;
			}
		}

		if (x == end_point.x && y == end_point.y) {
			break;
		}

		int e2 = 2 * err;
		if (e2 > -dy) {
			err = err - dy;
			x = x + sx;
		}
		if (e2 < dx) {
			err = err + dx;
			y = y + sy;
		}
	}

	if (BM_SUCCESS != bm_sync_api(handle)) {
		std::cout << "draw line sync api error\r\n" << std::endl;
		return BM_ERR_FAILURE;
	}

	return BM_SUCCESS;
}

[lingwei2023]

代码如下 使用tpu绘制速度比cpu慢了好多个量级

static bm_status_t bmcv_draw_line(bm_handle_t handle, u64 base_addr, int pixel_stride, int row_stride, int height, int width, bmcv_point_t start_point, bmcv_point_t end_point, u8 val)
{
	// Bresenham 直线算法
	int dx = abs(end_point.x - start_point.x);
	int dy = abs(end_point.y - start_point.y);
	int sx, sy;

	if (start_point.x < end_point.x) {
		sx = 1;
	}
	else {
		sx = -1;
	}

	if (start_point.y < end_point.y) {
		sy = 1;
	}
	else {
		sy = -1;
	}

	int err = dx - dy;
	int x = start_point.x;
	int y = start_point.y;

	while (true) {
		// 绘制当前点
		if (x >= 0 && x < width && y >= 0 && y < height) {
			u64 offset = x * pixel_stride + y * row_stride;
			base_addr = base_addr + offset;
			bm_api_memset_byte_t api = { base_addr, 1, 1, 1, 1, 1, 1, 1, pixel_stride, val };

			if (BM_SUCCESS != bm_send_api(handle, (bm_api_id_t)177, (uint8_t*)&api, sizeof(api)))
			{
				std::cout << "draw line send api error\r\n" << std::endl;
				return BM_ERR_FAILURE;
			}
		}

		if (x == end_point.x && y == end_point.y) {
			break;
		}

		int e2 = 2 * err;
		if (e2 > -dy) {
			err = err - dy;
			x = x + sx;
		}
		if (e2 < dx) {
			err = err + dx;
			y = y + sy;
		}
	}

	if (BM_SUCCESS != bm_sync_api(handle)) {
		std::cout << "draw line sync api error\r\n" << std::endl;
		return BM_ERR_FAILURE;
	}

	return BM_SUCCESS;
}

您好，首先感谢您使用这个接口，并提出意见，宝贵的意见是我们优化的动力。
从代码看，您使用的是1684 下 tpu 画框的接口，在实现画线的功能。建议您尝试已封装好的接口 bmcv_image_draw_rectangle 来实现画直线的功能。 我们内部测试，画一条1024*1024的直线，耗时在400us左右。

[RaechelliaYao]

@lingwei2023 你好，画一条1024*1024的直线，耗时确实是在400us左右，但是如果需要画斜线需要如何处理呢