| Issue |
176481
|
| Summary |
Wasm SIMD shuffle "don't care" that maps efficiently to SSE and NEON?
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
juj
|
When doing WebAssembly SIMD `wasm_i32x4_shuffle()`s (using i32x4 as an example), the developer is pressured to concern themselves about efficient lowering to SSE and NEON; given that Wasm SIMD is a pseudo SIMD API after all.
Sometimes in shuffles I have some amount of flexibility/freedom to pick arbitrary shuffle patterns.
Does there exist a mechanism, where the programmer could declare a "don't care" shuffle lane?
Then the WebAssembly backend could pick the most optimal value, that it knows will lower the best to the fewest SSE and NEON shuffle instructions sequence. (e.g. a shufps or an unpack?)
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
