I've backported this to GCC8 too since it had the same failures:
The testcase for PR62178 has been failing for a while due to the pass
conditions being too tight, resulting in failures with -mcmodel=tiny:
ldr q2, [x0], 124
ld1r {v1.4s}, [x1], 4
cmp x0, x2
mla v0.4s, v2.4s, v1.4s
bne .L7
-mcmodel=small generates the slightly different:
ldr q1, [x0], 124
ldr s2, [x1, 4]!
cmp x0, x2
mla v0.4s, v1.4s, v2.s[0]
bne .L7
This is due to Combine merging a DUP instruction with either a load
or MLA - we can't force it to prefer one over the other. However the
generated vector loop is fast either way since it generates MLA and
merges the DUP either with a load or MLA. So relax the conditions
slightly and check we still generate MLA and there is no DUP or FMOV.
The testcase now passes - committed as obvious.
ChangeLog
2019-01-09 Wilco Dijkstra <wdijk...@arm.com>
testsuite/
* gcc.target/aarch64/pr62178.c: Relax scan-assembler checks.
--- gcc/testsuite/gcc.target/aarch64/pr62178.c (revision 266178)
+++ gcc/testsuite/gcc.target/aarch64/pr62178.c (working copy)
@@ -18,5 +18,5 @@
/* { dg-final { scan-assembler "ldr\\tq\[0-9\]+, \\\[x\[0-9\]+\\\], \[0-9\]+"
} } */
/* { dg-final { scan-assembler "mla\\tv\[0-9\]+\.4s, v\[0-9\]+\.4s, v\[0-9\]+"
} } */
-/* { dg-final { scan-assembler-not { dup } } } */
-/* { dg-final { scan-assembler-not { fmov } } } */
+/* { dg-final { scan-assembler-not {dup} } } */
+/* { dg-final { scan-assembler-not {fmov} } } */
