[Mlir-commits] [mlir] [MLIR][XeGPU][VectorToXeGPU] Lower vector.load/store/transfer_read/transfer_write to new offsets syntax (PR #162095)
Adam Siemieniuk
llvmlistbot at llvm.org
Tue Oct 21 02:14:25 PDT 2025
================
@@ -121,6 +121,39 @@ isValidGatherScatterParams(Type maskTy, VectorType valueTy,
return success();
}
+// Verify that number of offsets matches either the source rank or the tdesc
----------------
adam-smnk wrote:
> I think we should simply check that offsets can't have lower rank than tdesc rank. It shoud be fine if it is larger than tdesc rank.
Sound reasonable. Just please clearly define how offsets are interpreted.
For example, memref is 4D, descriptor is 2D and there are 3 offsets. I guess they'd apply to the innermost dimensions.
https://github.com/llvm/llvm-project/pull/162095
More information about the Mlir-commits
mailing list