https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128447
Bug ID: 128447
Summary: Another related values case for ifcvt.cc
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: rtl-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: law at gcc dot gnu.org
Blocks: 26163
Target Milestone: ---
Target: riscv
int
f (int a, int b, int c)
{
return c ? (a << 4) + b : (a + b) << 4;
}
Compiled with -O2 -march=rv64gcbv_zicond generates:
addw a5,a0,a1 # 35 [c=8 l=4] addsi3_extended/0
slliw a0,a0,4 # 37 [c=8 l=4] ashlsi3_extend
addw t1,a0,a1 # 38 [c=8 l=4] addsi3_extended/0
slliw t0,a5,4 # 36 [c=8 l=4] ashlsi3_extend
czero.nez t2,t0,a2 # 41 [c=4 l=4] *czero.nez.didi
czero.eqz a1,t1,a2 # 40 [c=4 l=4] *czero.eqz.didi
add a0,t2,a1 # 30 [c=4 l=4] *adddi3/0
Which is a generalized conditional move. But we can do better.
What's different across the arms is whether or not we select b or b << 4. SO
we could instead generate something like this:
slliw a5,a0,4
addi a4,zero,4
czero.nez a4,a4,a2
sllw a1,a1,a4
addw a0,a5,a1
This shows up 10-12m times across x264 in related forms.
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=26163
[Bug 26163] [meta-bug] missed optimization in SPEC (2026, 2k17, 2k and 2k6 and
95)
