* frv gas: align code with nops
@ 2004-05-06 2:57 Alexandre Oliva
2004-05-06 11:25 ` Nick Clifton
0 siblings, 1 reply; 2+ messages in thread
From: Alexandre Oliva @ 2004-05-06 2:57 UTC (permalink / raw)
To: binutils
[-- Attachment #1: Type: text/plain, Size: 209 bytes --]
We currently use the all-zero bit pattern to pad code for alignment,
but that's bad, for the reasons explained in the comments below. Ok
to install? Ok for 2.15? Tested on i686-pc-linux-gnu-x-frv-uclinux.
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: gas-frv-align-code-with-nops.patch --]
[-- Type: text/x-patch, Size: 2349 bytes --]
Index: gas/ChangeLog
from Alexandre Oliva <aoliva@redhat.com>
* config/tc-frv.h (MAX_MEM_FOR_RS_ALIGN_CODE): New.
(HANDLE_ALIGN): New.
Index: gas/config/tc-frv.h
===================================================================
RCS file: /cvs/uberbaum/gas/config/tc-frv.h,v
retrieving revision 1.5
diff -u -p -r1.5 tc-frv.h
--- gas/config/tc-frv.h 6 May 2004 02:46:45 -0000 1.5
+++ gas/config/tc-frv.h 6 May 2004 02:56:17 -0000
@@ -86,3 +86,44 @@ extern long md_pcrel_from_section PARAMS
for any relocations that pic won't support. */
#define tc_frob_file() frv_frob_file ()
extern void frv_frob_file PARAMS ((void));
+
+/* We don't want 0x00 for code alignment because this generates `add.p
+ gr0, gr0, gr0' patterns. Although it's fine as a nop instruction,
+ it has the VLIW packing bit set, which means if you have a bunch of
+ them in a row and attempt to execute them, you'll exceed the VLIW
+ capacity and fail. This also gets GDB confused sometimes, because
+ it won't set breakpoints in instructions other than the first of a
+ VLIW pack, so you used to be unable to set a breakpoint in the
+ initial instruction of a function that followed such
+ alignment-introduced instructions.
+
+ We could have arranged to emit `nop' instructions (0x80880000),
+ maybe even VLIW-pack sequences of nop instructions as much as
+ possible for the selected machine type, just in case the alignment
+ code actually happens to run, but this is probably too much effort
+ for little gain. This code is not meant to be run anyway, so just
+ emit nops. */
+#define MAX_MEM_FOR_RS_ALIGN_CODE (3 + 4)
+#define HANDLE_ALIGN(FRAGP) do \
+ if ((FRAGP)->fr_type == rs_align_code) \
+ { \
+ valueT count = ((FRAGP)->fr_next->fr_address \
+ - ((FRAGP)->fr_address + (FRAGP)->fr_fix)); \
+ unsigned char *dest = (FRAGP)->fr_literal + (FRAGP)->fr_fix; \
+ if ((count & 3) != 0) \
+ { \
+ memset (dest, 0, (count & 3)); \
+ (FRAGP)->fr_fix += (count & 3); \
+ dest += (count & 3); \
+ count -= (count & 3); \
+ } \
+ if (count) \
+ { \
+ (FRAGP)->fr_var = 4; \
+ *dest++ = 0x80; \
+ *dest++ = 0x88; \
+ *dest++ = 0x00; \
+ *dest++ = 0x00; \
+ } \
+ } \
+ while (0)
[-- Attachment #3: Type: text/plain, Size: 188 bytes --]
--
Alexandre Oliva http://www.ic.unicamp.br/~oliva/
Red Hat Compiler Engineer aoliva@{redhat.com, gcc.gnu.org}
Free Software Evangelist oliva@{lsd.ic.unicamp.br, gnu.org}
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: frv gas: align code with nops
2004-05-06 2:57 frv gas: align code with nops Alexandre Oliva
@ 2004-05-06 11:25 ` Nick Clifton
0 siblings, 0 replies; 2+ messages in thread
From: Nick Clifton @ 2004-05-06 11:25 UTC (permalink / raw)
To: Alexandre Oliva; +Cc: binutils
Hi Alex,
>We currently use the all-zero bit pattern to pad code for alignment,
>but that's bad, for the reasons explained in the comments below. Ok
>to install? Ok for 2.15? Tested on i686-pc-linux-gnu-x-frv-uclinux.
>
>
Approved (for mainline). Please apply.
Cheers
Nick
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2004-05-06 11:25 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2004-05-06 2:57 frv gas: align code with nops Alexandre Oliva
2004-05-06 11:25 ` Nick Clifton
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).