No test case as none of the in-tree targets with GlobalISel support has this condition.
Details
Diff Detail
Event Timeline
LGTM with nitpicks
| lib/CodeGen/GlobalISel/LegalizerHelper.cpp | ||
|---|---|---|
| 574 | Period at the end of sentence | |
| 577 | I would call the variable X and Y or change the names in the comment to RHS and LHS to have both the code and the comment to match. | |
| 580 | FNEG could be illegal and thus it may be legalized into FSUB. | |
| lib/CodeGen/GlobalISel/LegalizerHelper.cpp | ||
|---|---|---|
| 580 | My feeling is that it's probably better not to try to define generically how G_FSUB should be lowered if the target doesn't support it. | |
| lib/CodeGen/GlobalISel/LegalizerHelper.cpp | ||
|---|---|---|
| 580 | I believe it has value to be in the generic code. FWIW, this is how SDISel does is. (See ~LegalizeDAG.cpp:3236) | |
| lib/CodeGen/GlobalISel/LegalizerHelper.cpp | ||
|---|---|---|
| 580 | That's what I thought. We can simply return UnableToLegalize for that case. | |
| lib/CodeGen/GlobalISel/LegalizerHelper.cpp | ||
|---|---|---|
| 580 | Agreed. | |
Period at the end of sentence