https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127267
Bug ID: 127267
Summary: [17 Regression] AOP changes causing execution failure
and poor code on iq2000-elf
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: law at gcc dot gnu.org
Target Milestone: ---
This change:
commit fb21f1686938d6cca13fa7efd9f94830a7a6ef04 (HEAD)
Author: Roger Sayle <[email protected]>
Date: Mon Sep 7 16:53:07 2026 +0100
middle-end: Improve bitreverse RTL expansion with AOP.
This patch is a revision of my previous patch to use add_optab in
the expansion of bitreverse and store_fixed_bitfield which improves
code generation on x86, cris, riscv, parisc, sh and possibly other
targets.
https://gcc.gnu.org/pipermail/gcc-patches/2026-August/728199.html
https://gcc.gnu.org/pipermail/gcc-patches/2026-August/726901.html
[ ... ]
Triggered:
Tests that now fail, but worked before (1 tests):
iq2000-sim: gcc: gcc.c-torture/execute/20040705-1.c -O1 execution test
Note carefully -O1. It doesn't fail at -O2.
If we look at the diffs, they're primarily of this nature:
*************** fn1_1:
*** 102,108 ****
addu %4,%4,%5
sll %4,%4,26
andoui %2,%2,0x3ff
! or %2,%2,%4
j %31
sw %2,8(%3)
--- 102,108 ----
addu %4,%4,%5
sll %4,%4,26
andoui %2,%2,0x3ff
! addu %2,%2,%4
j %31
sw %2,8(%3)
But in main, which presumably has inlined copies of some functions we see:
*************** main:
*** 2988,2995 ****
sw %16,0(%29)
lui %16,%hi(b) # high
addiu %16,%16,%lo(b) # low
! lui %2,%hi(-851543506) # high
! addiu %2,%2,%lo(-851543506) # low
sw %2,8(%16)
lui %3,%hi(c) # high
addiu %3,%3,%lo(c) # low
--- 2988,3007 ----
sw %16,0(%29)
lui %16,%hi(b) # high
addiu %16,%16,%lo(b) # low
! lw %2,8(%16)
! nop
! andoui %2,%2,0x3ff
! lui %3,0xcc00 # -872415232
! ori %3,%3,0
! addu %2,%2,%3
! lui %3,%hi(-67076097) # high
! addiu %3,%3,%lo(-67076097) # low
! and %2,%2,%3
! lui %3,0x13e # 20840448
! ori %3,%3,0
! addu %2,%2,%3
! andoi %2,%2,0x8000
! addiu %2,%2,31278
sw %2,8(%16)
lui %3,%hi(c) # high
addiu %3,%3,%lo(c) # low
There's a few instances like that. I don't know what inlined function they
correspond to. The fact that it's an execution failure is quite a surprise,
though I guess we've never really due into iq2000 testresults in recent
history, so there may well be a correctness bug lurking in there.
Anyway, not working on this. But wanted to get it filed so it's not forgotten.
jeff
