[llvm-bugs] [Bug 32433] New: [X86] Failure to use HADDPS for partial register result

[via llvm-bugs]

http://bugs.llvm.org/show_bug.cgi?id=32433

            Bug ID: 32433
           Summary: [X86] Failure to use HADDPS for partial register
                    result
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Backend: X86
          Assignee: unassignedbugs at nondot.org
          Reporter: llvm-dev at redking.me.uk
                CC: llvm-bugs at lists.llvm.org, spatel+llvm at rotateright.com

While the 256-bit horizontal pair sums work fine (both on btver2 and btver1),
the 128-bit version completely fails:

#include <x86intrin.h>
void sum_pairs_128(__m128 f, float *p) {
  p[0] = f[0] + f[1];
  p[1] = f[2] + f[3];
}
void sum_pairs_256(__m256 f, float *p) {
  p[0] = f[0] + f[1];
  p[1] = f[2] + f[3];
  p[2] = f[4] + f[5];
  p[3] = f[6] + f[7];
}

clang -O3 -march=btver2

sum_pairs_128(float __vector(4), float*):
        vmovshdup       %xmm0, %xmm1    # xmm1 = xmm0[1,1,3,3]
        vaddss  %xmm1, %xmm0, %xmm1
        vmovss  %xmm1, (%rdi)
        vpermilpd       $1, %xmm0, %xmm1 # xmm1 = xmm0[1,0]
        vpermilps       $231, %xmm0, %xmm0 # xmm0 = xmm0[3,1,2,3]
        vaddss  %xmm0, %xmm1, %xmm0
        vmovss  %xmm0, 4(%rdi)
        retq

sum_pairs_256(float __vector(8), float*):
        vextractf128    $1, %ymm0, %xmm1
        vhaddps %xmm1, %xmm0, %xmm0
        vmovups %xmm0, (%rdi)
        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/20170327/d5652a71/attachment.html>