vitoplantamura
commented
1 year ago hi,
I don't think memory would be an issue, as SD 1.5 requires 1.1GB of RAM in OnnxStream in its least conservative configuration.
I've tried the latest version of OnnxStream on my RPI Zero 2 and in Windows, and it works, so this isn't a problem introduced with the latest commit.
My guess is that it could be an incompatibility between XNNPACK and the Amlogic S905X.
Unfortunately I can do little without having the SBC in my hands.
This is a great guide that I have in my bookmarks:
https://jvns.ca/blog/2018/04/28/debugging-a-segfault-on-linux/
Let me know if you need help,
Thanks, Vito
rhalai
Hi, I'm getting a segmentation fault when attempting to run. I'm using a SweetPotato (Raspberry Pi clone), running Raspbian Bullseye 64-bit OS. This device has 2GB of DDR4 memory.
I was able to compile XNNPACK successfully and without errors following your instructions
I was able to compile your OnnxStream code - this gives an information/warning by appears to compile:
$ cmake --build . --config Release [ 33%] Building CXX object CMakeFiles/sd.dir/sd.cpp.o [ 66%] Building CXX object CMakeFiles/sd.dir/onnxstream.cpp.o In file included from /usr/include/c++/10/bits/stl_algobase.h:64, from /usr/include/c++/10/vector:60, from /home/frenchfry/projects/XNNPACK/OnnxStream/src/onnxstream.h:4, from /home/frenchfry/projects/XNNPACK/OnnxStream/src/onnxstream.cpp:1: /usr/include/c++/10/bits/stl_pair.h: In instantiation of ‘constexpr std::pair<typename std::__strip_reference_wrapper<typename std::decay<_Tp>::type>::__type, typename std::__strip_reference_wrapper<typename std::decay<_Tp2>::type>::__type> std::make_pair(_T1&&, _T2&&) [with _T1 = float&; _T2 = float&; typename std::__strip_reference_wrapper<typename std::decay<_Tp2>::type>::__type = float; typename std::__strip_reference_wrapper<typename std::decay<_Tp>::type>::__type = float]’: /home/frenchfry/projects/XNNPACK/OnnxStream/src/onnxstream.cpp:2085:60: required from here /usr/include/c++/10/bits/stl_pair.h:567:5: note: parameter passing for argument of type ‘std::pair<float, float>’ when C++17 is enabled changed to match C++14 in GCC 10.1 567 | make_pair(_T1&& __x, _T2&& __y) | ^~~~~~~~~ [100%] Linking CXX executable sd [100%] Built target sdBut, when I run, I get a segmentation fault within a few seconds:
$ ./sd --rpi --steps 3 --rpi-lowmem ----------------[start]------------------ positive_prompt: a photo of an astronaut riding a horse on mars negative_prompt: ugly, blurry output_png_path: ./result.png steps: 3 ----------------[prompt]------------------ Segmentation faultI added the command line switch to print operations and it shows the following: ` $ ./sd --rpi --steps 3 --rpi-lowmem --ops-printf ----------------[start]------------------ positive_prompt: a photo of an astronaut riding a horse on mars negative_prompt: ugly, blurry output_png_path: ./result.png steps: 3 ----------------[prompt]------------------
0) Reshape (Reshape_113)
1) Gather (Gather_114)
2) Add (Add_116)
3) ReduceMean (ReduceMean_123)
4) Sub (Sub_124)
5) Pow (Pow_126)
6) ReduceMean (ReduceMean_127)
7) Add (Add_129)
8) Sqrt (Sqrt_130)
9) Div (Div_131)
10) Mul (Mul_132)
11) Add (Add_133)
12) MatMul (MatMul_134)
13) Add (Add_135)
14) Mul (Mul_137)
15) MatMul (MatMul_138)
16) Add (Add_139)
17) Reshape (Reshape_140)
18) Transpose (Transpose_141)
19) MatMul (MatMul_142)
20) Add (Add_143)
21) Reshape (Reshape_144)
22) Transpose (Transpose_145)
23) Reshape (Reshape_146)
24) Transpose (Transpose_147)
25) Reshape (Reshape_148)
26) Reshape (Reshape_149)
27) Reshape (Reshape_150)
28) Transpose (Transpose_151)
29) MatMul (MatMul_152)
30) Reshape (Reshape_153)
31) Add (Add_154)
32) Reshape (Reshape_155)
33) Softmax (Softmax_156)
34) MatMul (MatMul_157)
35) Reshape (Reshape_158)
36) Transpose (Transpose_159)
37) Reshape (Reshape_160)
38) MatMul (MatMul_161)
39) Add (Add_162)
40) Add (Add_163)
41) ReduceMean (ReduceMean_164)
42) Sub (Sub_165)
43) Pow (Pow_167)
44) ReduceMean (ReduceMean_168)
45) Add (Add_170)
46) Sqrt (Sqrt_171)
47) Div (Div_172)
48) Mul (Mul_173)
49) Add (Add_174)
50) MatMul (MatMul_175)
51) Add (Add_176)
52) Mul (Mul_178)
53) Sigmoid (Sigmoid_179)
54) Mul (Mul_180)
55) MatMul (MatMul_181)
Segmentation fault `
I have tried with --rpi-lowmem set and unset (same results). I know --lowmem was meant to be <500MB, but not sure how much memory is required if this switch is not set.
I have tried re-compiling OnnxStream with -DMAX_SPEED=ON and -DMAX_SPEED=OFF (same results).
Any ideas on how do debug?