[llvm-bugs] [Bug 51591] New: Failure to vectorize none-of 4xbyte

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=51591

            Bug ID: 51591
           Summary: Failure to vectorize none-of 4xbyte
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Scalar Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: lebedev.ri at gmail.com
                CC: llvm-bugs at lists.llvm.org

https://godbolt.org/z/4WcY9aW5r
https://alive2.llvm.org/ce/z/RgJ439

We need to either/or do a better job of canonicalizing this in instcombine,
or teach SLPVectorizer about this.

SLP: Analyzing blocks in _Z3fooPKh.
SLP:  bundle:   %cmp.i.i.i.i.i.i = icmp eq i8 %2, -1
SLP:  initialize schedule region to   %cmp.i.i.i.i.i.i = icmp eq i8 %2, -1
SLP:  extend schedule region end to   %cmp.i.i59.i.i.i.i = icmp eq i64 %3,
65280
SLP:  extend schedule region end to   %cmp.i.i60.i.i.i.i = icmp eq i64 %4,
16711680
SLP:  extend schedule region end to   %cmp.i.i61.i.i.i.i = icmp eq i64 %5,
4278190080
SLP: try schedule bundle [  %cmp.i.i.i.i.i.i = icmp eq i8 %2, -1; 
%cmp.i.i59.i.i.i.i = icmp eq i64 %3, 65280;  %cmp.i.i60.i.i.i.i = icmp eq i64
%4, 16711680;  %cmp.i.i61.i.i.i.i = icmp eq i64 %5, 4278190080] in block entry
SLP:       update deps of [  %cmp.i.i.i.i.i.i = icmp eq i8 %2, -1; 
%cmp.i.i59.i.i.i.i = icmp eq i64 %3, 65280;  %cmp.i.i60.i.i.i.i = icmp eq i64
%4, 16711680;  %cmp.i.i61.i.i.i.i = icmp eq i64 %5, 4278190080]
SLP:       update deps of /   %cmp.i.i59.i.i.i.i = icmp eq i64 %3, 65280
SLP:       update deps of /   %cmp.i.i60.i.i.i.i = icmp eq i64 %4, 16711680
SLP:       update deps of /   %cmp.i.i61.i.i.i.i = icmp eq i64 %5, 4278190080
SLP:       update deps of   %or.cond27 = select i1 %or.cond, i1 true, i1
%cmp.i.i60.i.i.i.i
SLP:     gets ready on update:   %or.cond27 = select i1 %or.cond, i1 true, i1
%cmp.i.i60.i.i.i.i
SLP:       update deps of   %or.cond = or i1 %cmp.i.i.i.i.i.i,
%cmp.i.i59.i.i.i.i
SLP:   schedule   %or.cond27 = select i1 %or.cond, i1 true, i1
%cmp.i.i60.i.i.i.i
SLP:    gets ready (def):   %or.cond = or i1 %cmp.i.i.i.i.i.i,
%cmp.i.i59.i.i.i.i
SLP:   schedule   %or.cond = or i1 %cmp.i.i.i.i.i.i, %cmp.i.i59.i.i.i.i
SLP:    gets ready (def): [  %cmp.i.i.i.i.i.i = icmp eq i8 %2, -1; 
%cmp.i.i59.i.i.i.i = icmp eq i64 %3, 65280;  %cmp.i.i60.i.i.i.i = icmp eq i64
%4, 16711680;  %cmp.i.i61.i.i.i.i = icmp eq i64 %5, 4278190080]
SLP: We are able to schedule this bundle.
SLP:  cancel scheduling of [  %cmp.i.i.i.i.i.i = icmp eq i8 %2, -1; 
%cmp.i.i59.i.i.i.i = icmp eq i64 %3, 65280;  %cmp.i.i60.i.i.i.i = icmp eq i64
%4, 16711680;  %cmp.i.i61.i.i.i.i = icmp eq i64 %5, 4278190080]
SLP: Gathering cmp with different predicate.
SLP: Check whether the tree with height 1 is fully vectorizable .
SLP: Trying to vectorize a list of length = 2.
SLP: Analyzing 2 operations 
SLP:  bundle:   %cmp.i.i.i.i.i.i = icmp eq i8 %2, -1
SLP:  initialize schedule region to   %cmp.i.i.i.i.i.i = icmp eq i8 %2, -1
SLP:  extend schedule region end to   %cmp.i.i59.i.i.i.i = icmp eq i64 %3,
65280
SLP: try schedule bundle [  %cmp.i.i.i.i.i.i = icmp eq i8 %2, -1; 
%cmp.i.i59.i.i.i.i = icmp eq i64 %3, 65280] in block entry
SLP:       update deps of [  %cmp.i.i.i.i.i.i = icmp eq i8 %2, -1; 
%cmp.i.i59.i.i.i.i = icmp eq i64 %3, 65280]
SLP:       update deps of /   %cmp.i.i59.i.i.i.i = icmp eq i64 %3, 65280
SLP:     gets ready on update:   %cmp.i.i.i.i.i.i = icmp eq i8 %2, -1
SLP: We are able to schedule this bundle.
SLP:  cancel scheduling of [  %cmp.i.i.i.i.i.i = icmp eq i8 %2, -1; 
%cmp.i.i59.i.i.i.i = icmp eq i64 %3, 65280]
SLP: Gathering cmp with different predicate.
SLP: Check whether the tree with height 1 is fully vectorizable .

-- 
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/20210823/78f986eb/attachment.html>