[Bug target/51509] Inefficient neon intrinsic code sequence

[linux at carewolf dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=51509

Allan Jensen <linux at carewolf dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |linux at carewolf dot com

--- Comment #6 from Allan Jensen <linux at carewolf dot com> ---
I have run into a similar problem with vld3 and vst4.

uint8x16x3_t tmp = vld3q_u8(src);
vst4q_u8((uint8_t *)dst, {tmp.val[2], tmp.val[1], tmp.val[0], fullVector});

produces:
  70:   4cdf4061        ld3     {v1.16b-v3.16b}, [x3], #48
  74:   4e083c04        mov     x4, v0.d[0]
  78:   4e183c05        mov     x5, v0.d[1]
  7c:   6f000400        mvni    v0.4s, #0x0
  80:   4e083c4a        mov     x10, v2.d[0]
  84:   4e183c4b        mov     x11, v2.d[1]
  88:   aa0403e2        mov     x2, x4
  8c:   aa0503e1        mov     x1, x5
  90:   4e083c24        mov     x4, v1.d[0]
  94:   4e183c25        mov     x5, v1.d[1]
  98:   a90007e2        stp     x2, x1, [sp]
  9c:   3d800fe0        str     q0, [sp,#48]
  a0:   a9012fea        stp     x10, x11, [sp,#16]
  a4:   aa0403e6        mov     x6, x4
  a8:   a90217e6        stp     x6, x5, [sp,#32]
  ac:   4c4023e0        ld1     {v0.16b-v3.16b}, [sp]
  b0:   4c9f0000        st4     {v0.16b-v3.16b}, [x0], #64


But if I add -fno-split-wide-types it compiles to:
  68:   4cdf4064        ld3     {v4.16b-v6.16b}, [x3], #48
  6c:   4f000400        movi    v0.4s, #0x0
  70:   6f000403        mvni    v3.4s, #0x0
  74:   4ea51ca1        mov     v1.16b, v5.16b
  78:   4ea41c82        mov     v2.16b, v4.16b
  7c:   4c9f0000        st4     {v0.16b-v3.16b}, [x0], #64

This happens with both 4.9 and 5.1 that I have tried.