biondizzle 3e14a25bb0 fix: don't re-init mbarrier in loop — use phase parity tracking ...

The mbarrier is initialized once before the loop with count=1.
Inside the loop: issue TMA → arrive.expect_tx → wait(phase) → flip phase.
Re-initializing the mbarrier inside the loop resets the phase, which
breaks the parity tracking and causes the wait to hang.

This matches the CUTLASS/gau-nernst pattern exactly.

2026-05-29 18:24:47 +00:00

..

attention

fix: don't re-init mbarrier in loop — use phase parity tracking

2026-05-29 18:24:47 +00:00

cache

KV Cache: schema, allocator, pools, manager, append_swa kernel

2026-05-22 00:08:38 +00:00

compressor

Restructure: cutedsl/ -> dsv4/ with proper layering

2026-05-21 17:30:44 +00:00

cuda

Clean up E2M1 dequant: use LUT approach (consultant recommendation)

2026-05-28 16:17:47 +00:00

decode

Restructure: cutedsl/ -> dsv4/ with proper layering

2026-05-21 17:30:44 +00:00

gemm

NVFP4-1.1: Mark fp4_quant.py as toolchain-blocked, clean up test files

2026-05-28 04:59:01 +00:00

indexer

Clean up E2M1 dequant: use LUT approach (consultant recommendation)

2026-05-28 16:17:47 +00:00

router

Router: Blackwell-native fused decode kernel — real CuTeDSL implementation

2026-05-21 22:04:20 +00:00

__init__.py

Restructure: cutedsl/ -> dsv4/ with proper layering

2026-05-21 17:30:44 +00:00