| format PE64 GUI 6.2 at 0x1234_56780000 | |
| include 'mywin.inc' | |
| GWL_EXSTYLE := -20 | |
| TaskbarDemo: | |
| iterate msg, WM_INITDIALOG,WM_COMMAND,WM_CLOSE | |
| cmp edx,msg | |
| jz .msg | |
| end iterate |
| import torch | |
| import torch._inductor.config | |
| import time | |
| torch._inductor.config.triton.cudagraphs = False | |
| torch.set_float32_matmul_precision('high') | |
| def bench(f, name=None, iters=100, warmup=5, display=True, profile=False): | |
| for _ in range(warmup): | |
| f() |
Example of how to capture CPU counters with ETW on Windows, perf on Linux or kperf on Apple.
Using ETW needs somewhat recently updated Windows 10 or 11. Not sure about exact version.
Currently tested on:
| ; We can make a global "#pragma once" mechanism if we assume all files have | |
| ; a unique path. Non-cannonical path are fine as long as it's always the same | |
| ; form. Be mindful of file organization if you're going to use this. | |
| ; Primary benefits: | |
| ; + files are not loaded multiple times | |
| ; + no change to file needed (i.e. if/endif wrapper) | |
| ;macro onetime?: line& | |
| ; local name | |
| ; match any rest,line |
| // $ cc -o checker checker.c | |
| // $ ./checker | |
| // $ ffmpeg -i output-%02d.ppm -r 60 output.mp4 | |
| #include <stdio.h> | |
| int main() | |
| { | |
| char buf[256]; | |
| for (int i = 0; i < 60; ++i) { | |
| snprintf(buf, sizeof(buf), "output-%02d.ppm", i); |