Have you checked if all other Scalars.size() calls are correct?
Fri, Nov 25
@deadalnix Please can you rebase?
Thu, Nov 24
Still not a tuning flag
Wed, Nov 23
About to go on PTO - I will try to get these changes applied soon.
Have you confirmed that this passes the whole of check-llvm? Sometime these cost changes have effects in transforms/codegen you don't expect
Tue, Nov 22
LGTM - these often end up with missing CHECKS for some RUN combos (and the update scripts sometime fail to warn you) but I can't see any
we don't yet have codesize/latency/sizelatency variants of the shuffle cost tests - these need adding first - it should be as simple as copying the core analysis/costmodel/x86/shuffle-* test files and testing for each cost kind (look at something like the arith-int* files for reference)
Mon, Nov 21
@marcauberer reverse ping - what's happening with this patch please?
LGTM - any other comments?
Sun, Nov 20
please can you add a patch summary
@kazu Out of interest - how were you able to identify these?
Remove "NumReadAdvanceEntries == 0" test code
Sat, Nov 19
Fri, Nov 18
Thu, Nov 17
@sdardis Any thoughts on the MIPS changes?
LGTM (with the test precommit) - cheers
Agreed - as it is this patch puts even more strain on the DAG to load combine back to legality.
Wed, Nov 16
Tue, Nov 15
LGTM - cheers
I think the FIXME would be more useful if it was added to the code than forgotten in a commit message?
Mon, Nov 14
Thanks - almost there! Please can you tidyup the patch summary?
please can you recreate the diffs with context?