[llvm-bugs] [Bug 38522] New: Inefficient code generation for v16i8 vertical less-equal

via llvm-bugs llvm-bugs at lists.llvm.org
Fri Aug 10 10:14:43 PDT 2018


https://bugs.llvm.org/show_bug.cgi?id=38522

            Bug ID: 38522
           Summary: Inefficient code generation for v16i8 vertical
                    less-equal
           Product: new-bugs
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: gonzalobg88 at gmail.com
                CC: chandlerc at gmail.com, hfinkel at anl.gov,
                    llvm-bugs at lists.llvm.org, llvm-dev at redking.me.uk,
                    spatel+llvm at rotateright.com

The following Rust program:

extern crate packed_simd;
use packed_simd::*;
pub fn le_i8x16(x: i8x16, y: i8x16) -> bool {
    x.le(y).all()
}

when compiled with AVX2 and O3 (RUSTFLAGS="-C target-feature=+avx2") generates
the following LLVM-IR (https://godbolt.org/g/R2zra8) and assembly:

declare i32 @llvm.x86.sse41.ptestc(<2 x i64>, <2 x i64>);
define zeroext i1 @le_i8x16(<16 x i8>*  %x, <16 x i8>* %y) {
start:
%0 = load <16 x i8>, <16 x i8>* %x, align 16
%1 = load <16 x i8>, <16 x i8>* %y, align 16
%2 = icmp sle <16 x i8> %0, %1
%3 = sext <16 x i1> %2 to <16 x i8>
%4 = bitcast <16 x i8> %3 to <2 x i64>
%5 = tail call i32 @llvm.x86.sse41.ptestc(<2 x i64> %4, <2 x i64> <i64 -1, i64
-1>)
%6 = icmp eq i32 %5, 1
ret i1 %6
}

packed_simd_iter::le_i8x16:
 push    rbp
 mov     rbp, rsp
 vmovdqa xmm0, xmmword, ptr, [rdi]
 vpcmpgtb xmm0, xmm0, xmmword, ptr, [rsi]
 vpcmpeqd xmm1, xmm1, xmm1
 vpxor   xmm0, xmm0, xmm1
 vptest  xmm0, xmm1
 setb    al
 pop     rbp
 ret

Note that Rust generates IR for a vertical <= vector-vector operation which
returns a mask, and then IR for the all reduction.

The assembly we would expect is:

    pushq       %rbp
    movq        %rsp, %rbp
    vmovdqa     (%rdi), %xmm0
    vpcmpgtb    (%rsi), %xmm0, %xmm0
    vptest      %xmm0, %xmm0
    sete        %al
    popq        %rbp
    retq

-- 
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/20180810/0be3f8c4/attachment-0001.html>


More information about the llvm-bugs mailing list