trunk/373ac598ff432bea957f4885acf8a9bea16c2880: Enable Sleef in the NNC LLVM codegen on aarch64 (#198002) (#198002)
- PyTorch: 1416 events in the last 90 days
- PyTorch: 1398th Release in the last 90 days
- Previous: earlier the same day · trunk/80e7e60b9808464040ce4b9b40adeb38a79e8ca4
What happened
Summary: NNC's LLVM backend only ever reached Sleef on x86. Two things blocked aarch64, and either one alone was sufficient: intrinsic_symbols.cpp registers the Sleef_* symbols with the JIT inside a #if !defined(_MSC_VER) && defined(__x86_64__) block, so on aarch64 the JIT knew about no vector math symbols at all. getSimdFunction() gated vectorized lowering on the target feature string containing +avx , which is nev…
Summary assembled by rule from the sources below