Skip to content

Clarify MathDx usage with load_inline#517

Draft
msaroufim wants to merge 1 commit into
mainfrom
agent/enable-mathdx-headers
Draft

Clarify MathDx usage with load_inline#517
msaroufim wants to merge 1 commit into
mainfrom
agent/enable-mathdx-headers

Conversation

@msaroufim

@msaroufim msaroufim commented Jul 20, 2026

Copy link
Copy Markdown
Member

Summary

Clarify the existing Modal runner guidance: MathDx headers are already installed and exported through CPLUS_INCLUDE_PATH, so load_inline users only need to put #include <cublasdx.hpp> in cuda_sources rather than cpp_sources.

Why

load_inline compiles cpp_sources with the host C++ compiler and cuda_sources with NVCC. cuBLASDx belongs in the CUDA compilation unit; no additional Kernelbot image setting or submission compile flag is required.

Impact

Comment-only change in modal_runner.py. No dependency, image, example, or test changes, and no Modal image was deployed.

Validation

  • uv run --extra dev ruff check src/runners/modal_runner.py
  • git diff --check

@msaroufim
msaroufim force-pushed the agent/enable-mathdx-headers branch from 9f57e4f to 176b12c Compare July 20, 2026 18:35
@msaroufim msaroufim changed the title Document MathDx headers for Modal submissions Document MathDx load_inline flags Jul 20, 2026
@msaroufim
msaroufim force-pushed the agent/enable-mathdx-headers branch from 176b12c to 8744ef1 Compare July 20, 2026 18:37
@msaroufim msaroufim changed the title Document MathDx load_inline flags Clarify MathDx usage with load_inline Jul 20, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant