[ROCm] Fix the experimental HIP backend for ROCm 7.x on AMD GPUs#484

Merged

stotko merged 10 commits into

stotko:masterfrom

jeffdaily:moat-port

Jul 8, 2026

Commits on Jul 8, 2026

[ROCm] Fix the experimental HIP backend for ROCm 7.x on AMD GPUs

jeffdaily
authored and
stotko
committed
[ROCm] Apply clang-format to ported wave-lock critical sections

jeffdaily
authored and
stotko
committed
[ROCm] Fold small device reductions on the host instead of a custom kernel

jeffdaily
authored and
stotko
committed
[ROCm] Tie host-fold reduction threshold to rocPRIM single-block cutoff

jeffdaily
authored and
stotko
committed
[ROCm] Move warp-convergent execution into the backend namespace dispatch

jeffdaily
authored and
stotko
committed
[ROCm] Use __activemask and threadIdx for warp-convergent lane election

jeffdaily
authored and
stotko
committed
[ROCm] Correct the rationale for the HIP language source markings

jeffdaily
authored and
stotko
committed
[ROCm] Include user-facing execution.h; drop redundant comments

jeffdaily
authored and
stotko
committed
[ROCm] Guard execution_detail.h include for device compilers only

jeffdaily
authored and
stotko
committed
[ROCm] Use IS_DEVICE_COMPILED guard; rename execution_detail.cuh

jeffdaily
authored and
stotko
committed