Continued from http://reviews.llvm.org/D14874
Details
Details
Diff Detail
Diff Detail
Event Timeline
Comment Actions
LGTM with a few nits
lib/Target/Mips/AsmParser/MipsAsmParser.cpp | ||
---|---|---|
3060 | Shouldn't the `hasShortDelaySlot(Inst.getOpcode())` part be based on the properties of CFC1 rather than Inst? Also, it's not really a delay slot, it's a hazard. hasShortDelaySlot() will always return false for CFC1 and the trunc macros. | |
3064 | Likewise | |
3069 | Likewise | |
test/MC/Mips/mips64/valid.s | ||
288–289 | Indentation | |
test/MC/Mips/mips64r2/valid.s | ||
314–315 | Indentation | |
test/MC/Mips/mips64r3/valid.s | ||
314–315 | Indentation | |
test/MC/Mips/mips64r5/valid.s | ||
315–316 | Indentation |
Shouldn't the `hasShortDelaySlot(Inst.getOpcode())` part be based on the properties of CFC1 rather than Inst? Also, it's not really a delay slot, it's a hazard. hasShortDelaySlot() will always return false for CFC1 and the trunc macros.
I think we should just hard code it to true/false (with a comment explaining why) until we represent hazards in tablegen