On Tue, 2 Sep 2025, Jakub Jelinek wrote:
> On Tue, Sep 02, 2025 at 11:34:59AM +0200, Richard Biener wrote:
> >
> > Oops, now also to the list.
>
> If so, it needs to punt on anything that uses something more than once.
> So, alg_add_factor and alg_sub_factor are always bad (so return NULL),
> because they do use accumulator both in the PLUS/MINUS and LSHIFT.
> alg_add_t2_m/alg_sub_t2_m/alg_shift if synth_shift_p
> and alg.log[i], all in addition to what it has been doing before.
Like the following?
Thanks,
Richard.
>From 651c377e4f63ff234f034bfaa2633686203ed713 Mon Sep 17 00:00:00 2001
From: Richard Biener <rguent...@suse.de>
Date: Tue, 2 Sep 2025 10:16:28 +0200
Subject: [PATCH] tree-optimization/121753 - ICE with pattern breaking
reduction constraints
To: gcc-patches@gcc.gnu.org
The recent change to vect_synth_mult_by_constant missed to handle
the synth_shift_p case for alg_shift, so we still changed c * 4
to c + c + c + c. The following also amends alg_add_t2_m, alg_sub_t2_m,
alg_add_factor and alg_sub_factor appropriately.
PR tree-optimization/121753
* tree-vect-patterns.cc (vect_synth_mult_by_constant): Properly
bail when synth_shift_p and an alg_shift use. Handle other
problematic cases.
---
gcc/tree-vect-patterns.cc | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/gcc/tree-vect-patterns.cc b/gcc/tree-vect-patterns.cc
index 64a49cccc24..d0bf2f9e799 100644
--- a/gcc/tree-vect-patterns.cc
+++ b/gcc/tree-vect-patterns.cc
@@ -4329,7 +4329,14 @@ vect_synth_mult_by_constant (vec_info *vinfo, tree op,
tree val,
case alg_add_t2_m:
case alg_sub_t2_m:
op_uses++;
+ /* Fallthru. */
+ case alg_shift:
+ if (synth_shift_p && alg.log[i])
+ return NULL;
break;
+ case alg_add_factor:
+ case alg_sub_factor:
+ return NULL;
default:
break;
}
--
2.43.0
