[Bug middle-end/83022] New: malloc & memset -> calloc is not always an optimization

[nathan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83022

            Bug ID: 83022
           Summary: malloc & memset -> calloc is not always an
                    optimization
           Product: gcc
           Version: 8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: nathan at gcc dot gnu.org
  Target Milestone: ---

Created attachment 42623
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=42623&action=edit
exemplar

We like to optimize malloc followed by memset into a calloc call.  Even when
the memset is conditional.  That's well formed, but pessimizes, and noticeable
when the size is large and we do unnecessary clearing.

The attached example, compiled on x86_64 with -O results in:
_Z1mmb:
        movl    $1, %esi
        jmp     calloc

But, it causes a noticeable performance regression, as 'c' is false
sufficiently often and 's' is large sufficiently often.