Hi, I'm Sung-Woo Choi from South Korea.
I'm a Computer Science student at UNIST.
I'm interested in compilers, systems, and GPUs.
My work focuses on LLVM-based analysis, low-level performance, and correctness issues in systems software.
swc0317@unist.ac.kr / swchoi@nshc.net
-
Merged pull request #828 in the
exaloop/codonrepository addressed the issue of handling cases where no concrete GPU kernels are realized. The patch ensures that no GPU module is created, no PTX blob is emitted, and the__codon_ptx__()function is patched to null, effectively streamlining the GPU preparation process. -
Opened pull request #826 introduces a Python GPU DSL decorator implementation. This implementation adds a
codon.gpudecorator and includes GPU-specific source rewriting, allowing Python functions to be transformed into Codon GPU source code. The pull request is currently open and has not yet been merged. -
A comment on the open issue related to pull request #826 noted that test cases in
codon/test/transform/kernels.codonpass successfully when all kernels are rewritten in the Python DSL, indicating progress in the implementation. -
The pull request #826 includes significant changes with 318 additions and 2 deletions across 7 files, reflecting the complexity of the new GPU DSL implementation.
- 💪 Opened PR #828 in exaloop/codon
- 🗣 Commented on #826 in exaloop/codon
- 💪 Opened PR #826 in exaloop/codon
- 💪 Opened PR #821 in exaloop/codon
- 🗣 Commented on #819 in exaloop/codon
- 💪 Opened PR #819 in exaloop/codon
- ❗ Opened issue #818 in exaloop/codon
- 🎉 Merged PR #804 in exaloop/codon
- 🗣 Commented on #804 in exaloop/codon
- ❗ Opened issue #811 in exaloop/codon
