forked from OSchip/llvm-project
[RISCV] Assert initial load/store SEW is the EEW
In D127983, I had flipped from using the computed EEW to using the SEW value pulled from the VSETVLI when checking compatibility. This wasn't intentional, though thankfully it appears to be a non-functional difference. The new code does make a unchecked assumption that the initial SEW operand on the load/store is the EEW. This patch clarifies the assumption, and adds an assert to make sure this remains true. Differential Revision: https://reviews.llvm.org/D128085
This commit is contained in:
parent
1c92e06ded
commit
820e84e050
|
@ -408,6 +408,8 @@ static DemandedFields getDemanded(const MachineInstr &MI) {
|
|||
// They instead demand the ratio of the two which is used in computing
|
||||
// EMUL, but which allows us the flexibility to change SEW and LMUL
|
||||
// provided we don't change the ratio.
|
||||
// Note: We assume that the instructions initial SEW is the EEW encoded
|
||||
// in the opcode. This is asserted when constructing the VSETVLIInfo.
|
||||
if (getEEWForLoadStore(MI)) {
|
||||
Res.SEW = false;
|
||||
Res.LMUL = false;
|
||||
|
@ -885,6 +887,11 @@ static VSETVLIInfo computeInfoForInstr(const MachineInstr &MI, uint64_t TSFlags,
|
|||
} else {
|
||||
InstrInfo.setAVLReg(RISCV::NoRegister);
|
||||
}
|
||||
#ifndef NDEBUG
|
||||
if (Optional<unsigned> EEW = getEEWForLoadStore(MI)) {
|
||||
assert(SEW == EEW && "Initial SEW doesn't match expected EEW");
|
||||
}
|
||||
#endif
|
||||
InstrInfo.setVTYPE(VLMul, SEW, TailAgnostic, MaskAgnostic);
|
||||
|
||||
return InstrInfo;
|
||||
|
|
Loading…
Reference in New Issue