To be combined to vecreduce_or (setcc X, 0, eq) in follow on patch.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
llvm/test/CodeGen/AArch64/dag-combine-setcc.ll | ||
---|---|---|
5 | Others will add related tests to this file so a more descriptive name is desirable. |
- Use more descriptive function names in test.
- Added tests for v32i1 and v64i1. I've changed the tests to take vectors of i8s rather than i1s since the arg passing codegen for larger i1 vectors is pretty horrendous and not the point of this patch.
llvm/test/CodeGen/AArch64/dag-combine-setcc.ll | ||
---|---|---|
5 | Given these are reduction tests, what's the reason for returning a vector type? I think returning i1 is more in keeping with real world usage and removes some of the weirdness seen when returning the larger vector i1 types. |
llvm/test/CodeGen/AArch64/dag-combine-setcc.ll | ||
---|---|---|
5 |
Not sure what you mean by weirdness, most of these instructions are gone in the next patch, fair point though there's no need to return vector. |
Others will add related tests to this file so a more descriptive name is desirable.