On 02/03/2021 23:37, Ilya Leoshkevich wrote:
> Bootstrapped and regtested on x86_64-redhat-linux, ppc64le-redhat-linux
> and s390x-redhat-linux. Ok for master?
>
>
>
> Commit efb6bc55a93a ("fwprop: Allow (subreg (mem)) simplifications")
> introduced a check that was supposed to look at the propagated def's
> number of uses. It uses insn_info::num_uses (), which in reality
> returns the number of uses def's insn has. The whole change therefore
> works only by accident.
>
> Fix by looking at def_info's uses instead of insn_info's uses. This
> requires passing around def_info instead of insn_info.
>
If building glibc on s390x with -march=z14, there is currently an extra
"unneeded" stackframe in e.g. sqrtl, llroundl, lroundl, llrintl, llrint
and roundevenl.
0000000000000000 <__ieee754_sqrtl>:
0: b3 c1 00 4f ldgr %f4,%r15
4: e3 f0 ff 50 ff 71 lay %r15,-176(%r15)
a: e3 00 f0 a8 00 20 cg %r0,168(%r15)
# The long double argument is loaded ...
10: e7 00 30 00 30 06 vl %v0,0(%r3),3
# ... stored to extra "unneeded" stackframe ...
16: e7 00 f0 a0 30 0e vst %v0,160(%r15),3
# ... and loaded from there into a floating-point-pair
1c: 68 00 f0 a0 ld %f0,160(%r15)
20: 68 20 f0 a8 ld %f2,168(%r15)
24: b3 16 00 00 sqxbr %f0,%f0
28: 60 00 20 00 std %f0,0(%r2)
2c: 60 20 20 08 std %f2,8(%r2)
30: b3 cd 00 f4 lgdr %r15,%f4
34: 07 fe br %r14
With this patch, there is no stackframe in those functions anymore as it
was if build with gcc 10.
0000000000000000 <__ieee754_sqrtl>:
0: 68 00 30 00 ld %f0,0(%r3)
4: 68 20 30 08 ld %f2,8(%r3)
8: b3 16 00 00 sqxbr %f0,%f0
c: 60 00 20 00 std %f0,0(%r2)
10: 60 20 20 08 std %f2,8(%r2)
14: 07 fe br %r14
Thanks,
Stefan
