https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128777

            Bug ID: 128777
           Summary: [17 Regression] ICE in fmax intrinsics in aarch64
                    since r17-6220-gacefdec15b7
           Product: gcc
           Version: 17.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tnfchris at gcc dot gnu.org
  Target Milestone: ---
            Target: aarch64*

#include <arm_neon.h>

float32x4_t aom_gcc17_vmax_zero_repro(float32x4_t x) {
  return vmaxq_f32(x, vdupq_n_f32(0));
}

with -O3

ICEs with

dup.c: In function 'aom_gcc17_vmax_zero_repro':
dup.c:5:1: error: unrecognizable insn:
    5 | }
      | ^
(insn 6 3 7 2 (set (reg:V4SF 103)
        (unspec:V4SF [
                (reg/v:V4SF 102 [ x ])
                (const_vector:V4SF [
                        (const_double:SF 0.0 [0x0.0p+0]) repeated x4
                    ])
            ] UNSPEC_FMAX)) "dup.c":4:10 -1
     (nil))
during RTL pass: vregs
dup.c:5:1: internal compiler error: in extract_insn, at recog.cc:2955

Reply via email to