Repository navigation
Fix checks to accommodate register wraparounds in multi-reg ops - #101430
Conversation
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch |
|
@a74nh @kunalspathak @dotnet/arm64-contrib |
amanasifkhalid
left a comment
There was a problem hiding this comment.
LGTM assuming CI passes. Thanks!
|
@SwapnilGaikwad - did we end up assigning |
Yup, we are using more local variables in The 'genGenerateCode' phase from the JIT dump |
|
Thanks @SwapnilGaikwad . I see it now: mvni v26.4s, #0
mvni v27.4s, #0
mov v28.16b, v26.16b
mov v29.16b, v27.16b
mov v30.16b, v9.16b
mov v31.16b, v10.16b
tbl v26.16b, {v28.16b, v29.16b, v30.16b, v31.16b}, v17.16b
mvni v27.4s, #0
mvni v28.4s, #0
mov v29.16b, v27.16b
mov v30.16b, v28.16b
mov v31.16b, v9.16b
mov v0.16b, v10.16b
tbl v27.16b, {v29.16b, v30.16b, v31.16b, v0.16b}, v18.16b
mvni v28.4s, #0
mvni v29.4s, #0
mov v30.16b, v28.16b
mov v31.16b, v29.16b
mov v0.16b, v9.16b
mov v1.16b, v10.16b
tbl v28.16b, {v30.16b, v31.16b, v0.16b, v1.16b}, v19.16b |
All those Not sure if this is tiered compilation or not - if lower tier then doesn't really matter. |
|
Yes, this is with |
How can I get such assembly dump without hitting the asserts? With |
I ported your fix and tried to get disassembly just to see the instruction sequence. |
Fixes issue #101070