[Bug tree-optimization/127910] New: full O3 vectorizer cost model slows down SpecCPU'26 zstd

[andi-gcc at firstfloor dot org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127910

            Bug ID: 127910
           Summary: full O3 vectorizer cost model slows down SpecCPU'26
                    zstd
           Product: gcc
           Version: 16.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: andi-gcc at firstfloor dot org
  Target Milestone: ---

Created attachment 65849
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=65849&action=edit
reduced reproducer

Disabling the vectorizer (over -O3) or setting the cost model to cheap speeds
up SpecCPU'26 zstd slightly on a Intel PantherLake/Pcore. seen with 16.0 and
trunk.

Attached is an (AI generated) reduced reproducer.┘

With cheap this loop is not vectorized.

 if (input < loop_limit) {
        size_t diff = load_word(match) ^ load_word(input);
        if (diff != 0)
            return common_bytes(diff);
        input += sizeof(size_t);
        match += sizeof(size_t);
        while (input < loop_limit) {
            diff = load_word(match) ^ load_word(input);
            if (diff == 0) {
                input += sizeof(size_t);
                match += sizeof(size_t);
                continue;
            }
            input += common_bytes(diff);
            return (size_t)(input - start);
        }
    }