3togo

Results 52 comments of 3togo

The revised version will still fail. How to avoid memory accumulation? ``` use arrayfire::*; // use std::mem; fn main() { let device = get_device(); for i in 0..1000 { let...

use load_image_native instead of load_image could avoid the problem

Actually load_image_native will still cause memory accumulation but much slower than that of load_image. How to free the accumulated memory to avoid the program crashing?

adding sm_61 could not fix the problem compile arrayfire git version c++ in debug mode could avoid this problem

the same code if compiling using opencl, it will work without errors.

this commit, 18028c090be521414a9e0c4495c2c29b1d4436f1 caused the bug.

The bug persists, but I have been unable to create a minimal program that reproduces the error. The GDB backtrace initially pointed to af::min as the source of the crash....

1. export AF_JIT_KERNEL_TRACE=stdout does not produce any extra information. 2. It only occurs for the CUDA backend. OpenCL does not have any problems.