[llvm-bugs] [Bug 38030] New: [MCSchedModel] Variant class support in computeInstrLatency and getReciprocalThroughput

via llvm-bugs llvm-bugs at lists.llvm.org
Tue Jul 3 04:15:57 PDT 2018


            Bug ID: 38030
           Summary: [MCSchedModel] Variant class support in
                    computeInstrLatency and getReciprocalThroughput
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: MC
          Assignee: unassignedbugs at nondot.org
          Reporter: llvm-dev at redking.me.uk
                CC: andrea.dibiagio at gmail.com, atrick at apple.com,
                    clement.courbet at gmail.com, gchatelet at google.com,
                    llvm-bugs at lists.llvm.org
            Blocks: 36550

Both MCSchedModel::computeInstrLatency and
MCSchedModel::getReciprocalThroughput are missing support for variant classes
from some versions of the methods.

They either assert if MCSchedClassDesc.isVariant() or don't take it into
account at all (and give some very odd latency/throughput numbers...). This
appears to be at least in part to them not always having access to a MCInst
instance which prevents them resolving with resolveVariantSchedClass.

Some of these methods may be not in use and can be removed, while for others we
need some way to resolve the variant to a default (worst case?) sched class.

Referenced Bugs:

[Bug 36550] Derive TTI instruction costs from scheduling models
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20180703/7ae246a0/attachment.html>

More information about the llvm-bugs mailing list