using script.tcl to build a prj in vivado hls 2018.3 error

[milozzzyj]

i run like readme.md

there is an error:

[onioncc]

Hi, can you try to debug using the generated executable under the build/bin dictionary to locate the exact crash point?

Best,
Cong

[haihengcao]

I've met the exact problem in 2019.2 Vivado HLS;Can't see any concrete error either

[haihengcao]

it turns out that the stack size is not large enough.
https://forums.xilinx.com/t5/Vivado/XAPP1167-%E8%87%AA%E5%B8%A6%E4%BE%8B%E7%A8%8Bremap%E5%B7%A5%E7%A8%8B-%E6%8A%A5%E9%94%99/td-p/890685
增加栈的大小。具体方式为将Vivado HLS的Project Settings-->Simulation-->Linker Flags设置为-Wl,--stack,10485670。但是需要注意的是这种方法可能会因为机器没有足够的内存而无效。

Solved.

[weiren2]

The default stack size on Windows is usually smaller than Linux (typically 8MB). This could explain why the code runs fine (assume it was developed in a Linux environment) but not on my Windows machine.

I think the large local array debug (on line 400) is likely to cause stack overflow.

SkyNet/FPGA/HLS/tb.cc

Lines 395 to 402 in 007e445

	std::ifstream ifs_image_raw_g(img_g, std::ios::in | std::ios::binary);
	ifs_image_raw_g.read((char*)(image_raw_g_burst), 3*162*2*322*2*sizeof(uint8));
	float predict_box[4][5];
	int constant[4][3]; //conf_j, conf_m, conf_n
	float debug[10][32][44][84];
	float box[4] = {1.4940052559648322, 2.3598481287086823, 4.0113013115312155, 5.760873975661669};

As mentioned in #2 (comment), you can increase stack size to a larger value (at least a little more than 10x32x44x84 floats or ~4.5MB).

If you don't want to increase stack size, you can try modify the local array to static (static float debug[10][32][44][84];) so it won't use stack to hold the array any more. (I'm not sure what the debug array does but the enclosing function test_one_group is only called once in the testbench so making it static should be safe).