Claude 45161610f0 fix: thread_local reentrant guard — prevent cudaMalloc infinite recursion ...

CUB CachingDeviceAllocator::DeviceAllocate calls cudaMalloc internally
on cache miss. Without a guard, our intercepted cudaMalloc recurses
into DeviceAllocate → cudaMalloc → DeviceAllocate → segfault.

thread_local g_in_allocator flag detects reentrant calls and forwards
them directly to the real cudaMalloc/cudaFree via dlsym(RTLD_NEXT).

2026-08-13 10:37:32 +00:00

..

include

feat: CCCL CachingDeviceAllocator preload — 完整依赖链 288 files

2026-08-13 09:26:41 +00:00

build_cccl_preload.sh

fix: use cccl_preload::cub namespace — CUB_WRAPPED_NAMESPACE requires it

2026-08-13 10:36:27 +00:00

cccl_allocator_preload.cu

fix: thread_local reentrant guard — prevent cudaMalloc infinite recursion

2026-08-13 10:37:32 +00:00

verify_preload.py

feat: CCCL CachingDeviceAllocator LD_PRELOAD — bypass CoreX expandable_segments ASSERT

2026-08-13 09:53:42 +00:00