Re: committed] correct memcmp expansion of constant representations containing embedded nuls (PR 95189)

["H.J. Lu" <hjl.tools@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2221 bytes --]

On Thu, Jul 23, 2020 at 4:50 PM Martin Sebor <msebor@gmail.com> wrote:
>
> On 7/23/20 2:18 PM, H.J. Lu wrote:
> > On Thu, Jul 23, 2020 at 1:14 PM Martin Sebor via Gcc-patches
> > <gcc-patches@gcc.gnu.org> wrote:
> >>
> >> On 7/22/20 2:23 AM, Rainer Orth wrote:
> >>> Hi Martin,
> >>>
> >>>> I have committed this change in r11-2231 after Jeff approved it
> >>>> off list last Thursday.
> >>>
> >>> the new gcc.target/i386/memcpy-pr95886.c test FAILs on 32-bit x86
> >>> (i386-pc-solaris2.11):
> >>>
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 1976943448883713" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 576467370915332609" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578431098682540545" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695685198337" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695685198337" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752110593" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752306689" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752307200" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752307201" 2
> >>
> >> Thanks for letting me know.  The test looks for patterns that are
> >> apparently LP64-specific so I restricted it to just that data model.
> >>
> >
> > Shouldn't it also work for x32?
>
> I would expect the optimization to work on any target that does
> the piecemeal copy, including x32.  But you probably meant if
> the test should pass as is on x32.  Possibly yes, but I didn't
> check.
>
> A better test would exercise the solution at least on all three
> i386 targets.  It could probably be done fairly simply by reducing
> the sizes of the arrays, or by using the large arrays only on LP64
> or whatever makes the difference and figuring out the right magic
> target selector to use in the dg- directives.

Here is the patch I am checking in.

-- 
H.J.

[-- Attachment #2: 0001-Restrict-PR-middle-end-95886-x86-test-to-ia32.patch --]
[-- Type: text/x-patch, Size: 1120 bytes --]

From 23daf7386b07c78b79223a00a85c8b002a5a67e3 Mon Sep 17 00:00:00 2001
From: "H.J. Lu" <hjl.tools@gmail.com>
Date: Thu, 23 Jul 2020 19:14:06 -0700
Subject: [PATCH] Restrict PR middle-end/95886 x86 test to !ia32

Since gcc.target/i386/memcpy-pr95886.c requires 64-bit register, restrict
it to !ia32.

	PR middle-end/95886
	* gcc.target/i386/memcpy-pr95886.c: Restrict test to !ia32.
---
 gcc/testsuite/gcc.target/i386/memcpy-pr95886.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/gcc/testsuite/gcc.target/i386/memcpy-pr95886.c b/gcc/testsuite/gcc.target/i386/memcpy-pr95886.c
index ddfdfd2bc78..0699146296a 100644
--- a/gcc/testsuite/gcc.target/i386/memcpy-pr95886.c
+++ b/gcc/testsuite/gcc.target/i386/memcpy-pr95886.c
@@ -1,7 +1,6 @@
 /* PR middle-end/95886 - suboptimal memcpy with embedded zero bytes
-   { dg-do compile }
-   { dg-options "-O2 -Wall -fdump-rtl-expand" }
-   { dg-require-effective-target lp64 } */
+   { dg-do compile { target { ! ia32 } } }
+   { dg-options "-O2 -Wall -fdump-rtl-expand" } */
 
 const char a1234567890[10] = { 1, 2, 3, 4, 5, 6, 7, 8, 9 };
 
-- 
2.26.2