RE: [PATCH] arm: Fix ICE due to infinite splitting [PR109800]

[Kyrylo Tkachov <Kyrylo.Tkachov@arm.com>]

Hi Alex,

> -----Original Message-----
> From: Alex Coplan <Alex.Coplan@arm.com>
> Sent: Thursday, May 11, 2023 12:15 PM
> To: gcc-patches@gcc.gnu.org
> Cc: nickc@redhat.com; Richard Earnshaw <Richard.Earnshaw@arm.com>;
> Ramana Radhakrishnan <ramana.gcc@gmail.com>; Kyrylo Tkachov
> <Kyrylo.Tkachov@arm.com>
> Subject: [PATCH] arm: Fix ICE due to infinite splitting [PR109800]
> 
> Hi,
> 
> In r11-966-g9a182ef9ee011935d827ab5c6c9a7cd8e22257d8 we introduce a
> simplification to emit_move_insn that attempts to simplify moves of the
> form:
> 
> (set (subreg:M1 (reg:M2 ...)) (constant C))
> 
> where M1 and M2 are of equal mode size. That is problematic for the splitter
> vfp.md:no_literal_pool_df_immediate in the arm backend, which tries to pun
> an
> lvalue DFmode pseudo into DImode and assign a constant to it with
> emit_move_insn, as the new transformation simply undoes this, and we end
> up
> splitting indefinitely.
> 
> This patch changes things around in the arm backend so that we use a
> DImode temporary (instead of DFmode) and first load the DImode constant
> into the pseudo, and then pun the pseudo into DFmode as an rvalue in a
> reg -> reg move. I believe this should be semantically equivalent but
> avoids the pathalogical behaviour seen in the PR.
> 
> Bootstrapped/regtested on arm-linux-gnueabihf, regtested on
> arm-none-eabi and armeb-none-eabi.
> 
> OK for trunk and backports?

Ok but the testcase...

> 
> Thanks,
> Alex
> 
> gcc/ChangeLog:
> 
> 	PR target/109800
> 	* config/arm/arm.md (movdf): Generate temporary pseudo in
> DImode
> 	instead of DFmode.
> 	* config/arm/vfp.md (no_literal_pool_df_immediate): Rather than
> punning an
> 	lvalue DFmode pseudo into DImode, use a DImode pseudo and pun it
> into
> 	DFmode as an rvalue.
> 
> gcc/testsuite/ChangeLog:
> 
> 	PR target/109800
> 	* gcc.target/arm/pr109800.c: New test.

diff --git a/gcc/testsuite/gcc.target/arm/pr109800.c b/gcc/testsuite/gcc.target/arm/pr109800.c
new file mode 100644
index 00000000000..71d1ede13dd
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/pr109800.c
@@ -0,0 +1,3 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -march=armv7-m -mfloat-abi=hard -mfpu=fpv4-sp-d16 -mbig-endian -mpure-code" } */
+double f() { return 5.0; }

... The arm testsuite options are kinda hard to get right with all the effective targets and multilibs and such hardcoded abi and march options tend to break in some target.
I suggest you put this testcase in gcc.target/arm/pure-code and add a dg-skip-if to skip the test if the multilib options specify a different float-abi.

Thanks,
Kyrill