HazyResearch / ThunderKittens

Tile primitives for speedy kernels
MIT License
1.65k stars 69 forks source link

h100.cu(97): error: "wait" is ambiguous #54

Open ziyuhuang123 opened 2 months ago

ziyuhuang123 commented 2 months ago

firstly the env.src is incorrect, and I modify it to:

# So that you can see the python packages from the tests
export PYTHONPATH=${PYTHONPATH}:$PWD/include/common/pyutils
export THUNDERKITTENS_ROOT=${PWD}/include

But I get error:

/home/zyhuang/temp_can/ThunderKittens-main/include/common/base_types.cuh(110): error: namespace "std" has no member "bit_cast"
      static __attribute__((device)) inline constexpr half neg_infty() { return std::bit_cast<__half>(uint16_t(0xFC00)); }
                                                                                     ^

/home/zyhuang/temp_can/ThunderKittens-main/include/common/base_types.cuh(110): error: type name is not allowed
      static __attribute__((device)) inline constexpr half neg_infty() { return std::bit_cast<__half>(uint16_t(0xFC00)); }
                                                                                              ^

h100.cu(97): error: "wait" is ambiguous
                  wait(compute_done[tic], (kv_idx/2)%2);
ziyuhuang123 commented 2 months ago

By the way, this is kernel attn/h100.