| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
| |
Do it in 2 passes: 1. gather variables, 2. step thru RPO transforming
loads/stores to each variable. More elegant solution to f9e3a52eaa.
More efficient since all variable usages are visited in CFG order, which
matters due to block sealing.
|
| |
|
|
|
| |
Move the alloca to the beginning of the block and do a 2nd pass on those
odd cases.
|
| |
|
|
|
|
|
|
|
| |
mem2reg was assuming alloca's appeared before all their uses. When this
didn't happen it broke the logic re. sealed blocks. Normally this
doesn't happen in most natural code. Even with gotos, RPO would assure
this in most cases, the exception I found is a goto that jumps into a
loop past a variable declaration that is earlier in the loop. Then even
RPO would keep the alloca itself past the first uses, breaking things.
|
| |
|
|
|
| |
The motivation is for aarch64 msub/madd instrs, for isel to produce. But
it should be useful for other things too.
|
| |
|
|
| |
Looks nicer
|
| | |
|
| |
|
