In this test, I want to see whether the R1 model can read some C code and explain why the Clang autovectorizer algorithm decided not to compile SIMD instructions for the loop.
I have provided some simplified rules based on the Clang docs and the normal rules of autovectorizing algorithms.
The code is from the CPython code base.
| from itertools import combinations | |
| from collections import namedtuple | |
| Card = namedtuple("Card", ["color", "shape", "shade", "number"]) | |
| cards = [ | |
| Card("red", "diamond", "hollow", 1), | |
| Card("green", "squiggle", "solid", 3), | |
| Card("red", "oval", "striped", 3), | |
| Card("red", "oval", "hollow", 2), |
| import _testinternalcapi | |
| def get_executors(func): | |
| code = func.__code__ | |
| co_code = code.co_code | |
| executors = {} | |
| for i in range(0, len(co_code), 2): | |
| try: | |
| executors[i] = co_code[i], _testinternalcapi.get_executor(code, i) |
| """ | |
| Testing on CPython3.13a1+ | |
| Requires some recent patches from main. | |
| pip install hypercorn | |
| Have successfully run the following apps: | |
| - fastapi==0.99.0 | |
| - Flask | |
| """ |
Talk notes
| (1111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111*()) |
| # Delete all forks that haven't been updated since 2020 | |
| gh auth refresh -h github.com -s delete_repo | |
| gh search repos \ | |
| --owner tonybaloney \ | |
| --updated="<2020-01-01" \ | |
| --include-forks=only \ | |
| --limit 100 \ | |
| --json url \ | |
| --jq ".[] .url" \ | xargs -I {} gh repo delete {} --confirm |
Links from the AIMQ workshop talk
| git clone --depth=1 https://github.com/microsoft/WSL2-Linux-kernel.git | |
| sudo apt install -y build-essential flex bison libssl-dev libelf-dev libbabeltrace-dev libunwind-dev libdw-dev binutils-dev libiberty-dev | |
| cd WSL2-Linux-kernel/tools/perf/ | |
| make -j2 -s && make install |