| Issue |
242368
|
| Summary |
[LV] Nested loop vectorization miscompilation
|
| Labels |
miscompilation,
vectorizers,
llvm:transforms
|
| Assignees |
|
| Reporter |
SavchenkoValeriy
|
https://llvm.godbolt.org/z/65P48P55e
When LoopVectorize hoists the memory runtime check out of the outer loop, it widens each pointer range to start at the first outer iteration's value (a = -8). That start wraps around the address space, so the unsigned a <u b_end comparison is false and the check misses the overlap between a and b in the iteration that actually runs (y = 1).
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs