Bo Qiao
Bo Qiao
Hi, would you mind update the full code s.t. we can also run some tests locally?
Hi @mzy2240, there are a few findings I can share regarding to the 1000x performance diff ;/. I believe you were comparing the performance of an after-compilation Numba version with...
@mzy2240, how much performance degradation did you observe for the corrected code? (BTW, the imag part seems do not need any freeze, only the real part is needed?). Regarding to...
> I observed a 10X degradation and you are right, only the real part needs freeze. (so the actual performance difference for the same function between taichi and numba will...
Hi, first sorry for the late reply. Regarding to the performance difference, I just ran the previous code with and without "freeze dS_dVm[k]". On my local PC with i9-11900K there...
@yuanming-hu @masahi It turns out a bit difficult to support the new reduce warp intrinsics at this moment. For example, `__reduce_add_sync (i32)` will need to be mapped to `redux.sync.add.s32`. This...
Update: Since we are approaching v1.1.0 release, I would like to draw an intermediate summary on this issue. Thanks to our contributors, the list of warp-level intrinsics has been fully...
@YuCrazing can you give an example of how you would like to see this information?
Double checked on Ubuntu 20.04 for building Taichi C-API lib, the following build commands worked for me: Prerequisite (mainly for build with llvm): - Use clang-10 instead of gcc: -...