[llvm-bugs] [Bug 39473] New: Merge consecutive load insertions into a vector where possible

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=39473

            Bug ID: 39473
           Summary: Merge consecutive load insertions into a vector where
                    possible
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Common Code Generator Code
          Assignee: unassignedbugs at nondot.org
          Reporter: llvm-dev at redking.me.uk
                CC: andrea.dibiagio at gmail.com, filcab at gmail.com,
                    llvm-bugs at lists.llvm.org, spatel+llvm at rotateright.com

#include <x86intrin.h>

__m128i load_00123456(const unsigned short *data) {
  return _mm_setr_epi16(data[0], data[0], data[1], data[2], data[3], data[4],
data[5], data[6]);
}

-O3 -march=btver2

_Z13load_00123456PKt: # @_Z13load_00123456PKt
  movzwl (%rdi), %eax
  vmovd %eax, %xmm0
  vpinsrw $1, %eax, %xmm0, %xmm0
  vpinsrw $2, 2(%rdi), %xmm0, %xmm0
  vpinsrw $3, 4(%rdi), %xmm0, %xmm0
  vpinsrw $4, 6(%rdi), %xmm0, %xmm0
  vpinsrw $5, 8(%rdi), %xmm0, %xmm0
  vpinsrw $6, 10(%rdi), %xmm0, %xmm0
  vpinsrw $7, 12(%rdi), %xmm0, %xmm0
  retq

Many of the loads/insertions could be merged to something like:

_Z19load_00123456_mergePKt: # @_Z19load_00123456_mergePKt
  movzwl (%rdi), %eax
  vmovd %eax, %xmm0
  vpshuflw $224, %xmm0, %xmm0 # xmm0 = xmm0[0,0,2,3,4,5,6,7]
  vpshufd $0, %xmm0, %xmm0 # xmm0 = xmm0[0,0,0,0]
  vpinsrd $1, 2(%rdi), %xmm0, %xmm0
  vpinsrq $1, 6(%rdi), %xmm0, %xmm0
  retq

https://godbolt.org/z/-HLpsE

-- 
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/20181029/361f26af/attachment.html>