https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125794
Bug ID: 125794
Summary: aarch64: incorrect SVE assembly for loop
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: kristerw at gcc dot gnu.org
Blocks: 118443
Target Milestone: ---
Target: aarch64-unknown-linux-gnu
The test case below fails when compiled for AArch64 with -O2 -march=armv9.5-a:
typedef unsigned V __attribute__((__vector_size__ (32)));
V x;
int
main ()
{
x[5] = 1;
for (unsigned i = 0; i < 8; i++)
if (x[i] != (i == 5))
__builtin_abort ();
return 0;
}
The vectorized check for elements x[4] through x[7] is incorrectly emitted as:
mov z31.d, #4294967296
ldr q30, [x0, 16]
cmpne p7.s, p7/z, z30.s, z31.s
b.any .L2
The `mov` instruction replicates the 64-bit constant across the entire SVE
vector, which makes the comparison check that both x[5] and x[7] are equal to
1.
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118443
[Bug 118443] [Meta bug] Bugs triggered by and blocking more smtgcc testing
