[llvm-bugs] [Bug 43146] New: Compiler fails to optimize big endian load since clang 8.0

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=43146

            Bug ID: 43146
           Summary: Compiler fails to optimize big endian load since clang
                    8.0
           Product: clang
           Version: 8.0
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: LLVM Codegen
          Assignee: unassignedclangbugs at nondot.org
          Reporter: mvels at google.com
                CC: llvm-bugs at lists.llvm.org, neeilans at live.com,
                    richard-llvm at metafoo.co.uk

The compiler fails to recognize a BigEndian load opportunity since clang 7

Consider:

struct BESize {
    unsigned char c1;
    unsigned char c2;
    unsigned char c3;
    unsigned char c4;
    unsigned char c5;
    unsigned char c6;
    unsigned char c7;
    unsigned char c8;

    uint64_t Load() const {
        return 
            (static_cast<uint64_t>(c1) << 56) | 
            (static_cast<uint64_t>(c2) << 48) | 
            (static_cast<uint64_t>(c3) << 40) | 
            (static_cast<uint64_t>(c4) << 32) | 
            (static_cast<uint64_t>(c5) << 24) | 
            (static_cast<uint64_t>(c6) << 16) | 
            (static_cast<uint64_t>(c7) <<  8) | 
            static_cast<uint64_t>(c8);
    }
};

Compiling this with clang 7.0 -O3 -march=haswell, it creates proper movbe code:
As per https://gcc.godbolt.org/z/YklsAm

uint64_t load_be2(const BESize* p) {
    return p->Load();
}

-->

load_be2(BESize const*):                   # @load_be2(BESize const*)
        movbe   rax, qword ptr [rdi]
        ret


Starting with clang 8.0, it generates a bunch of avx code, see
https://gcc.godbolt.org/z/7rfh9J


clang 8.0 still recognizes it is a Big Endian load, as running without
-march=haswell, it creates:

load_be2(BESize const*):                   # @load_be2(BESize const*)
        mov     rax, qword ptr [rdi]
        bswap   rax
        ret


Is this by design? I would guess the movbe is desirable over the AVX code?

-- 
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/20190828/decbebd4/attachment.html>