This patch adds a MVT named i64x8 needed for lowering inline asssembly operands which materialize a sequence of eight general purpose registers.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Unit Tests
Event Timeline
The only change from the last revision is that getTypeForEVT() now returns i512 for MVT::i64x8 instead of [8 x i64].
LGTM.
It's a little unfortunate we need a dedicated valuetype just for usage in inline asm, but I don't see a good alternative; at least, not without substantially rewriting the inline asm handling.
clang-format: please reformat the code
4 diff lines are omitted. See full path.