From: Roland McGrath <roland@hack.frob.com>
To: Joseph Myers <joseph@codesourcery.com>
Cc: libc-ports@sourceware.org
Subject: [PATCH roland/arm-avoid-pc] ARM: Support avoiding pc as destination register.
Date: Mon, 11 Mar 2013 20:54:00 -0000 [thread overview]
Message-ID: <20130311205410.2A99B2C08D@topped-with-meat.com> (raw)
Updated for current trunk. Tested as before:
On arm-linux-gnueabihf, I tested that this doesn't change the object
code at all. I also tested the changes by hacking the local copy of
arm-features.h to define ARM_ALWAYS_BX and verifying that there are no
regressions (no failures at all) when running 'make check subdirs=string'.
I'd be grateful for any suggestions to improve the efficiency of the
code in the ARM_ALWAYS_BX case. The extra push/pop for the scratch
register seems unavoidable without reworking the whole function in some
way more complicated than I wanted to think about. But maybe ARM
experts have better ideas.
OK?
Thanks,
Roland
ports/ChangeLog.arm
2013-03-11 Roland McGrath <roland@hack.frob.com>
* sysdeps/arm/arm-features.h: Add comment about ARM_ALWAYS_BX.
* sysdeps/arm/memcpy.S: Include <arm-features.h>.
[ARM_ALWAYS_BX]: Avoid pc as destination.
* sysdeps/arm/memmove.S: Likewise.
--- a/ports/sysdeps/arm/arm-features.h
+++ b/ports/sysdeps/arm/arm-features.h
@@ -36,4 +36,8 @@
at runtime (or that we never care about its state) and so need not
be checked for. */
+/* A more-specific arm-features.h file may define ARM_ALWAYS_BX to indicate
+ that instructions using pc as a destination register must never be used,
+ so a "bx" (or "blx") instruction is always required. */
+
#endif /* arm-features.h */
--- a/ports/sysdeps/arm/memcpy.S
+++ b/ports/sysdeps/arm/memcpy.S
@@ -20,6 +20,7 @@
/* Thumb requires excessive IT insns here. */
#define NO_THUMB
#include <sysdep.h>
+#include <arm-features.h>
/*
* Data preload for architectures that support it (ARM V5TE and above)
@@ -88,7 +89,12 @@ ENTRY(memcpy)
CALGN( bcs 2f )
CALGN( adr r4, 6f )
CALGN( subs r2, r2, r3 ) @ C gets set
+#ifndef ARM_ALWAYS_BX
CALGN( add pc, r4, ip )
+#else
+ CALGN( add r4, r4, ip )
+ CALGN( bx r4 )
+#endif
PLD( pld [r1, #0] )
2: PLD( subs r2, r2, #96 )
@@ -107,8 +113,16 @@ ENTRY(memcpy)
5: ands ip, r2, #28
rsb ip, ip, #32
+#ifndef ARM_ALWAYS_BX
addne pc, pc, ip @ C is always clear here
b 7f
+#else
+ beq 7f
+ push {r10}
+ cfi_adjust_cfa_offset (4)
+ add r10, pc, ip
+ bx r10
+#endif
6: nop
ldr r3, [r1], #4
ldr r4, [r1], #4
@@ -118,8 +132,13 @@ ENTRY(memcpy)
ldr r8, [r1], #4
ldr lr, [r1], #4
+#ifndef ARM_ALWAYS_BX
add pc, pc, ip
nop
+#else
+ add r10, pc, ip
+ bx r10
+#endif
nop
str r3, [r0], #4
str r4, [r0], #4
@@ -129,6 +148,11 @@ ENTRY(memcpy)
str r8, [r0], #4
str lr, [r0], #4
+#ifdef ARM_ALWAYS_BX
+ pop {r10}
+ cfi_adjust_cfa_offset (-4)
+#endif
+
CALGN( bcs 2b )
7: pop {r5 - r8}
@@ -146,7 +170,8 @@ ENTRY(memcpy)
strcsb r4, [r0], #1
strcsb ip, [r0]
-#if defined (__ARM_ARCH_4T__) && defined(__THUMB_INTERWORK__)
+#if ((defined (__ARM_ARCH_4T__) && defined(__THUMB_INTERWORK__)) \
+ || defined (ARM_ALWAYS_BX))
pop {r0, r4, lr}
cfi_adjust_cfa_offset (-12)
cfi_restore (r4)
--- a/ports/sysdeps/arm/memmove.S
+++ b/ports/sysdeps/arm/memmove.S
@@ -20,6 +20,7 @@
/* Thumb requires excessive IT insns here. */
#define NO_THUMB
#include <sysdep.h>
+#include <arm-features.h>
/*
* Data preload for architectures that support it (ARM V5TE and above)
@@ -104,7 +105,12 @@ ENTRY(memmove)
CALGN( bcs 2f )
CALGN( adr r4, 6f )
CALGN( subs r2, r2, ip ) @ C is set here
+#ifndef ARM_ALWAYS_BX
CALGN( add pc, r4, ip )
+#else
+ CALGN( add r4, r4, ip )
+ CALGN( bx r4 )
+#endif
PLD( pld [r1, #-4] )
2: PLD( subs r2, r2, #96 )
@@ -123,8 +129,16 @@ ENTRY(memmove)
5: ands ip, r2, #28
rsb ip, ip, #32
+#ifndef ARM_ALWAYS_BX
addne pc, pc, ip @ C is always clear here
b 7f
+#else
+ beq 7f
+ push {r10}
+ cfi_adjust_cfa_offset (4)
+ add r10, pc, ip
+ bx r10
+#endif
6: nop
ldr r3, [r1, #-4]!
ldr r4, [r1, #-4]!
@@ -134,8 +148,13 @@ ENTRY(memmove)
ldr r8, [r1, #-4]!
ldr lr, [r1, #-4]!
+#ifndef ARM_ALWAYS_BX
add pc, pc, ip
nop
+#else
+ add r10, pc, ip
+ bx r10
+#endif
nop
str r3, [r0, #-4]!
str r4, [r0, #-4]!
@@ -145,6 +164,11 @@ ENTRY(memmove)
str r8, [r0, #-4]!
str lr, [r0, #-4]!
+#ifdef ARM_ALWAYS_BX
+ pop {r10}
+ cfi_adjust_cfa_offset (-4)
+#endif
+
CALGN( bcs 2b )
7: pop {r5 - r8}
@@ -162,7 +186,8 @@ ENTRY(memmove)
strcsb r4, [r0, #-1]!
strcsb ip, [r0, #-1]
-#if defined (__ARM_ARCH_4T__) && defined (__THUMB_INTERWORK__)
+#if ((defined (__ARM_ARCH_4T__) && defined (__THUMB_INTERWORK__)) \
+ || defined (ARM_ALWAYS_BX))
pop {r0, r4, lr}
cfi_adjust_cfa_offset (-12)
cfi_restore (r4)
next reply other threads:[~2013-03-11 20:54 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2013-03-11 20:54 Roland McGrath [this message]
2013-03-11 20:55 ` [PATCH roland/arm-avoid-pc] ARM_BX_ALIGN_LOG2 Roland McGrath
2013-03-12 23:40 ` Joseph S. Myers
2013-03-13 19:42 ` Roland McGrath
2013-03-12 23:35 ` [PATCH roland/arm-avoid-pc] ARM: Support avoiding pc as destination register Joseph S. Myers
2013-03-13 16:50 ` Roland McGrath
-- strict thread matches above, loose matches on Subject: below --
2013-03-04 19:08 Roland McGrath
2013-03-09 2:21 ` Joseph S. Myers
2013-03-11 16:42 ` Roland McGrath
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20130311205410.2A99B2C08D@topped-with-meat.com \
--to=roland@hack.frob.com \
--cc=joseph@codesourcery.com \
--cc=libc-ports@sourceware.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).