News for hackers. Stuff that matters.

A comment on Meta LLM Compiler: neural optimizer and disassembler. Parent comment

Re:Meta LLM Compiler: neural optimizer and disassembler by a_t48 on Friday June 28, 2024 @07:19PM (#40826379)
NVCC CUDA builds were nondeterministic last time I checked, it made certain things (trying to get very clever with generating patches) difficult. This was also hampered by certain libraries (maybe GTSAM?) wanting to write __DATE__ somewhere in the build output, creating endlessly changing builds.
[ Parent | on HN ]
Re:Meta LLM Compiler: neural optimizer and disassembler by sigmoid10 on Saturday June 29, 2024 @05:10AM (#40828876)
In parallel computing you run into nondeterminism pretty quickly anyways - especially with CUDA because of undetermined execution order and floating point accuracy.
[ Parent | on HN ]
Re:Meta LLM Compiler: neural optimizer and disassembler by a_t48 on Saturday June 29, 2024 @03:51PM (#40832951)
Yes, at runtime. Compiling CUDA doesn’t require a GPU, though, and doesn’t really use “parallel computing”. I think CUDA via clang gets this right and will produce the same build every time - it was purely an NVCC issue.
[ Parent | on HN ]