[llvm-bugs] [Bug 51539] New: [X86] Possibly incorrect throughput for vpshufb/vpperm on bdver2 model

via llvm-bugs llvm-bugs at lists.llvm.org
Thu Aug 19 06:55:41 PDT 2021


            Bug ID: 51539
           Summary: [X86] Possibly incorrect throughput for vpshufb/vpperm
                    on bdver2 model
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Backend: X86
          Assignee: unassignedbugs at nondot.org
          Reporter: llvm-dev at redking.me.uk
                CC: craig.topper at gmail.com, lebedev.ri at gmail.com,
                    llvm-bugs at lists.llvm.org, llvm-dev at redking.me.uk,
                    pengfei.wang at intel.com, spatel+llvm at rotateright.com

I noticed this while working on the D103695 cost model helper script, the
llvm-mca reported bdver2 throughput for codegen containing vpshufb/vpperm is
often higher than expected. 

Instruction Info:
[1]: #uOps
[2]: Latency
[3]: RThroughput
[4]: MayLoad
[5]: MayStore
[6]: HasSideEffects (U)

[1]    [2]    [3]    [4]    [5]    [6]    Instructions:
 1      3     1.50                        vpshufb       %xmm0, %xmm1, %xmm2
 1      2     1.50                        vpperm        %xmm0, %xmm1, %xmm2,


All the docs I can find (AMDFam15h SoG, Agner, instlatx64 etc) indicate these
are fastpath (tp = 1.0) but just on pipe1 (xbr):


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/20210819/b66d4050/attachment.html>

More information about the llvm-bugs mailing list