abrown opened PR #1762 from ternary-imm8
to master
:
The InsertLane format has an ordering (
value().imm().value()
) and immediate name ("lane"
) that make it awkward to use for other instructions. This changes the ordering (value().value().imm()
) and uses the default name ("imm"
) throughout the codebase.
abrown requested iximeow for a review on PR #1762.
abrown updated PR #1762 from ternary-imm8
to master
:
The InsertLane format has an ordering (
value().imm().value()
) and immediate name ("lane"
) that make it awkward to use for other instructions. This changes the ordering (value().value().imm()
) and uses the default name ("imm"
) throughout the codebase.
iximeow submitted PR Review.
iximeow submitted PR Review.
iximeow created PR Review Comment:
I'm surprised to see we don't use
0x6b
/0x69
-encodedIMUL
, whereTernaryImm8
would be a better-fitting format in fact (it's a three-operand encoding in the formimul rrr, r/m, imm8
, orimm32
for the 32/64-bit forms). Do you know what other instructions are referenced here? A read through the legalizer seems to suggest this isn't actually true.
iximeow submitted PR Review.
iximeow created PR Review Comment:
answer:
x86_pshufd
where the immediate is not a lane selector, it's a mask of how to shuffle (where there is no encoding for a type-invalid mask that we would want to check in this arm)and this comment will become more true in the future with additional instructions taking mask immediates.
abrown closed without merge PR #1762.
Last updated: Dec 23 2024 at 13:07 UTC