We faced an issue that a register copy was triggered between FPR and VRF registers that caused failures. We fixed that in this PR: #8086
As part of that PR, I added an assertion to assert if register copy was triggered to copy FPR to VRF. For VRF to FPR case, it is not feasible to have such assertion because FPRs overlap with VRFs and we can get false failure if registers are in the overlapping range.
I open this issue to investigate if we can have correct register kind on real registers in the overlapping range or if we can use the virtual register instead of the real register to get the register kind which always have the correct register type.
We faced an issue that a register copy was triggered between FPR and VRF registers that caused failures. We fixed that in this PR: #8086
As part of that PR, I added an assertion to assert if register copy was triggered to copy FPR to VRF. For VRF to FPR case, it is not feasible to have such assertion because FPRs overlap with VRFs and we can get false failure if registers are in the overlapping range.
I open this issue to investigate if we can have correct register kind on real registers in the overlapping range or if we can use the virtual register instead of the real register to get the register kind which always have the correct register type.