Handle more than 64 registers - Part 5 #103188
Conversation
ghost
added
the
area-CodeGen-coreclr
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch |
|
@dotnet/jit-contrib @jakobbotsch PTAL |
| // passed in is preserved by the validator and take care to get the | ||
| // target from the register for the call (even in debug mode). | ||
| static_assert_no_msg((RBM_VALIDATE_INDIRECT_CALL_TRASH & (1 << REG_VALIDATE_INDIRECT_CALL_ADDR)) == 0); | ||
| static_assert_no_msg((RBM_VALIDATE_INDIRECT_CALL_TRASH & regMaskTP(1 << REG_VALIDATE_INDIRECT_CALL_ADDR)) == |
There was a problem hiding this comment.
We can make IsRegNumInMask constexpr to make this a bit more natural (feel free to do as part of a follow-up).
There was a problem hiding this comment.
I tried doing that, but as a cascade effect I need to make genSingleTypeRegSet, getRegForType(), etc. as constexpr which I am not sure is worth doing for just an assert. Will pass this on for now.
| hasPC = (imm & SRBM_PC) != 0; | ||
| hasLR = (imm & SRBM_LR) != 0; | ||
| imm &= ~(SRBM_PC | SRBM_LR); |
There was a problem hiding this comment.
I guess this means there is (and was) an invariant on the exact indices defined for these registers?
| static constexpr regMaskTP CreateFromRegNum(regNumber reg, regMaskSmall mask) | ||
| { |
There was a problem hiding this comment.
Is mask needed as a parameter? Can't we compute it here as 1 << reg?
There was a problem hiding this comment.
predicate registers starts from 64 thru 80 and the mask of them start with 0x0. I can make it work, but just want to rely on the mask that we already have in register files and how we use them currently.
There was a problem hiding this comment.
For predicate registers it would just be (1 << (reg - 64)) under the if, I suppose.
There was a problem hiding this comment.
Sure, I don't have much preference on which way or the other, but will keep this in mind during follow up.
jakobbotsch
left a comment
jakobbotsch
left a comment
There was a problem hiding this comment.
Nice, this looks great to me. Most changes look mechanical.
|
/rerun |
|
@dotnet-policy-service rerun |
1 similar comment
|
@dotnet-policy-service rerun |
2 participants
RBM_REG*toregMaskTPSRBM_REG*that can be used at places where we just needSingleTypeRegSetregMaskTPfor registers at killRefPosition