* 3.1 doesn't bootstrap on sparcv9
@ 2002-02-22 21:18 Brad Lucier
2002-02-22 21:23 ` Alexandre Oliva
0 siblings, 1 reply; 6+ messages in thread
From: Brad Lucier @ 2002-02-22 21:18 UTC (permalink / raw)
To: gcc; +Cc: Brad Lucier
I think this has been implicit in some previous messages, but just in
case someone thinks that no-one noticed, here's the error message:
./xgcc -B./ -B/pkgs/gcc-3.1/sparcv9-sun-solaris2.8/bin/ -isystem /pkgs/gcc-3.1/sparcv9-sun-solaris2.8/include -isystem /pkgs/gcc-3.1/sparcv9-sun-solaris2.8/sys-include -O2 -DIN_GCC -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -isystem ./include -fPIC -g1 -DHAVE_GTHR_DEFAULT -DIN_LIBGCC2 -D__GCC_FLOAT_NOT_NEEDED -I. -I. -I../../gcc -I../../gcc/. -I../../gcc/config -I../../gcc/../include -DL_fixunsdfsi -c ../../gcc/libgcc2.c -o libgcc/./_fixunsdfsi.o
../../gcc/libgcc2.c: In function `__fixunsdfsi':
../../gcc/libgcc2.c:1173: unrecognizable insn:
(insn 34 33 37 (set (reg:SI 121)
(const_int -2147483648 [0xffffffff80000000])) -1 (nil)
(expr_list:REG_EQUAL (const_int -2147483648 [0xffffffff80000000])
(nil)))
../../gcc/libgcc2.c:1173: Internal compiler error in ?, at recog.c:2130
Brad
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: 3.1 doesn't bootstrap on sparcv9
2002-02-22 21:18 3.1 doesn't bootstrap on sparcv9 Brad Lucier
@ 2002-02-22 21:23 ` Alexandre Oliva
2002-02-23 22:57 ` David O'Brien
2002-02-25 11:55 ` Brad Lucier
0 siblings, 2 replies; 6+ messages in thread
From: Alexandre Oliva @ 2002-02-22 21:23 UTC (permalink / raw)
To: Brad Lucier; +Cc: gcc
On Feb 23, 2002, Brad Lucier <lucier@math.purdue.edu> wrote:
> I think this has been implicit in some previous messages, but just in
> case someone thinks that no-one noticed, here's the error message:
Fixing this is in my to-do list, but I wouldn't mind if someone beat
me to it. It's a bug that used to be hidden in the SPARC port and
that was exposed by a patch that recently added some sanity checks.
--
Alexandre Oliva Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist Professional serial bug killer
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: 3.1 doesn't bootstrap on sparcv9
2002-02-22 21:23 ` Alexandre Oliva
@ 2002-02-23 22:57 ` David O'Brien
2002-02-24 4:08 ` Alexandre Oliva
2002-02-25 11:55 ` Brad Lucier
1 sibling, 1 reply; 6+ messages in thread
From: David O'Brien @ 2002-02-23 22:57 UTC (permalink / raw)
To: Alexandre Oliva; +Cc: gcc
On Sat, Feb 23, 2002 at 02:20:25AM -0300, Alexandre Oliva wrote:
> Fixing this is in my to-do list, but I wouldn't mind if someone beat
> me to it. It's a bug that used to be hidden in the SPARC port and
> that was exposed by a patch that recently added some sanity checks.
Do you have any experimental patch that allows sparc64 to bootstrap?
--
-- David (obrien@FreeBSD.org)
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: 3.1 doesn't bootstrap on sparcv9
2002-02-23 22:57 ` David O'Brien
@ 2002-02-24 4:08 ` Alexandre Oliva
2002-02-25 14:13 ` David O'Brien
0 siblings, 1 reply; 6+ messages in thread
From: Alexandre Oliva @ 2002-02-24 4:08 UTC (permalink / raw)
To: obrien; +Cc: gcc
[-- Attachment #1: Type: text/plain, Size: 810 bytes --]
On Feb 24, 2002, "David O'Brien" <obrien@FreeBSD.org> wrote:
> On Sat, Feb 23, 2002 at 02:20:25AM -0300, Alexandre Oliva wrote:
>> Fixing this is in my to-do list, but I wouldn't mind if someone beat
>> me to it. It's a bug that used to be hidden in the SPARC port and
>> that was exposed by a patch that recently added some sanity checks.
> Do you have any experimental patch that allows sparc64 to bootstrap?
I've just now completed the following, totally untested patch (well,
not really, it got to the point of completing the build of cc1 in
stage1, so syntax errors are ruled out), that I've started
bootstrapping on sparc-sun-solaris2.7 and sparcv9-sun-solaris2.7. No
ChangeLog entry yet, since this is not a patch submission, rather just
a token that I've finally got around to working on it :-)
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: sparcv9-const.patch --]
[-- Type: text/x-patch, Size: 6339 bytes --]
Index: gcc/config/sparc/sparc.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/config/sparc/sparc.c,v
retrieving revision 1.184
diff -u -p -r1.184 sparc.c
--- gcc/config/sparc/sparc.c 2002/02/22 18:43:54 1.184
+++ gcc/config/sparc/sparc.c 2002/02/24 11:56:14
@@ -1004,17 +1004,11 @@ const64_operand (op, mode)
int
const64_high_operand (op, mode)
rtx op;
- enum machine_mode mode ATTRIBUTE_UNUSED;
+ enum machine_mode mode;
{
return ((GET_CODE (op) == CONST_INT
&& (INTVAL (op) & ~(HOST_WIDE_INT)0x3ff) != 0
- && SPARC_SETHI_P (INTVAL (op))
-#if HOST_BITS_PER_WIDE_INT != 64
- /* Must be positive on non-64bit host else the
- optimizer is fooled into thinking that sethi
- sign extends, even though it does not. */
- && INTVAL (op) >= 0
-#endif
+ && SPARC_SETHI_P (INTVAL (op) & GET_MODE_MASK (mode))
)
|| (GET_CODE (op) == CONST_DOUBLE
&& CONST_DOUBLE_HIGH (op) == 0
@@ -1227,12 +1221,7 @@ input_operand (op, mode)
variants when we are working in DImode and !arch64. */
if (GET_MODE_CLASS (mode) == MODE_INT
&& ((GET_CODE (op) == CONST_INT
- && ((SPARC_SETHI_P (INTVAL (op))
- && (! TARGET_ARCH64
- || (INTVAL (op) >= 0)
- || mode == SImode
- || mode == HImode
- || mode == QImode))
+ && (SPARC_SETHI_P (INTVAL (op) & GET_MODE_MASK (mode))
|| SPARC_SIMM13_P (INTVAL (op))
|| (mode == DImode
&& ! TARGET_ARCH64)))
@@ -1311,7 +1300,7 @@ sparc_emit_set_const32 (op0, op1)
{
HOST_WIDE_INT value = INTVAL (op1);
- if (SPARC_SETHI_P (value)
+ if (SPARC_SETHI_P (value & GET_MODE_MASK (mode))
|| SPARC_SIMM13_P (value))
abort ();
}
Index: gcc/config/sparc/sparc.h
===================================================================
RCS file: /cvs/gcc/egcs/gcc/config/sparc/sparc.h,v
retrieving revision 1.160
diff -u -p -r1.160 sparc.h
--- gcc/config/sparc/sparc.h 2002/02/22 18:43:54 1.160
+++ gcc/config/sparc/sparc.h 2002/02/24 11:56:15
@@ -1393,7 +1393,8 @@ extern const char leaf_reg_remap[];
`J' is used for the range which is just zero (since that is R0).
`K' is used for constants which can be loaded with a single sethi insn.
`L' is used for the range of constants supported by the movcc insns.
- `M' is used for the range of constants supported by the movrcc insns. */
+ `M' is used for the range of constants supported by the movrcc insns.
+ `N' is like K, but for constants wider than 32 bits. */
#define SPARC_SIMM10_P(X) ((unsigned HOST_WIDE_INT) (X) + 0x200 < 0x400)
#define SPARC_SIMM11_P(X) ((unsigned HOST_WIDE_INT) (X) + 0x400 < 0x800)
@@ -1405,15 +1406,18 @@ extern const char leaf_reg_remap[];
#define SMALL_INT32(X) (SPARC_SIMM13_P (trunc_int_for_mode \
(INTVAL (X), SImode)))
#define SPARC_SETHI_P(X) \
-(((unsigned HOST_WIDE_INT) (X) & \
- (TARGET_ARCH64 ? ~(unsigned HOST_WIDE_INT) 0xfffffc00 : 0x3ff)) == 0)
+ (((unsigned HOST_WIDE_INT) (X) \
+ & ((unsigned HOST_WIDE_INT) 0x3ff - GET_MODE_MASK (SImode) - 1)) == 0)
+#define SPARC_SETHI32_P(X) \
+ (SPARC_SETHI_P ((unsigned HOST_WIDE_INT) (X) & GET_MODE_MASK (SImode)))
#define CONST_OK_FOR_LETTER_P(VALUE, C) \
((C) == 'I' ? SPARC_SIMM13_P (VALUE) \
: (C) == 'J' ? (VALUE) == 0 \
- : (C) == 'K' ? SPARC_SETHI_P (VALUE) \
+ : (C) == 'K' ? SPARC_SETHI32_P (VALUE) \
: (C) == 'L' ? SPARC_SIMM11_P (VALUE) \
: (C) == 'M' ? SPARC_SIMM10_P (VALUE) \
+ : (C) == 'N' ? SPARC_SETHI_P (VALUE) \
: 0)
/* Similar, but for floating constants, and defining letters G and H.
Index: gcc/config/sparc/sparc.md
===================================================================
RCS file: /cvs/gcc/egcs/gcc/config/sparc/sparc.md,v
retrieving revision 1.147
diff -u -p -r1.147 sparc.md
--- gcc/config/sparc/sparc.md 2002/02/22 18:43:54 1.147
+++ gcc/config/sparc/sparc.md 2002/02/24 11:56:20
@@ -2515,7 +2515,7 @@
(define_insn "*movdi_insn_sp64_novis"
[(set (match_operand:DI 0 "nonimmediate_operand" "=r,r,r,r,m,?e,?e,?m")
- (match_operand:DI 1 "input_operand" "rI,K,J,m,rJ,e,m,e"))]
+ (match_operand:DI 1 "input_operand" "rI,N,J,m,rJ,e,m,e"))]
"TARGET_ARCH64 && ! TARGET_VIS
&& (register_operand (operands[0], DImode)
|| reg_or_0_operand (operands[1], DImode))"
@@ -2533,7 +2533,7 @@
(define_insn "*movdi_insn_sp64_vis"
[(set (match_operand:DI 0 "nonimmediate_operand" "=r,r,r,r,m,?e,?e,?m,b")
- (match_operand:DI 1 "input_operand" "rI,K,J,m,rJ,e,m,e,J"))]
+ (match_operand:DI 1 "input_operand" "rI,N,J,m,rJ,e,m,e,J"))]
"TARGET_ARCH64 && TARGET_VIS &&
(register_operand (operands[0], DImode)
|| reg_or_0_operand (operands[1], DImode))"
@@ -2791,7 +2791,7 @@
/* Slick... but this trick loses if this subreg constant part
can be done in one insn. */
if (CONST_DOUBLE_LOW (operands[1]) == CONST_DOUBLE_HIGH (operands[1])
- && !(SPARC_SETHI_P (CONST_DOUBLE_HIGH (operands[1]))
+ && !(SPARC_SETHI32_P (CONST_DOUBLE_HIGH (operands[1]))
|| SPARC_SIMM13_P (CONST_DOUBLE_HIGH (operands[1]))))
{
emit_insn (gen_movsi (gen_lowpart (SImode, operands[0]),
@@ -3403,7 +3403,7 @@
/* Slick... but this trick loses if this subreg constant part
can be done in one insn. */
if (l[1] == l[0]
- && !(SPARC_SETHI_P (l[0])
+ && !(SPARC_SETHI32_P (l[0])
|| SPARC_SIMM13_P (l[0])))
{
emit_insn (gen_movsi (gen_lowpart (SImode, operands[0]),
Index: gcc/doc/md.texi
===================================================================
RCS file: /cvs/gcc/egcs/gcc/doc/md.texi,v
retrieving revision 1.35
diff -u -p -r1.35 md.texi
--- gcc/doc/md.texi 2002/02/23 12:59:08 1.35
+++ gcc/doc/md.texi 2002/02/24 11:56:34
@@ -1859,6 +1859,17 @@ Zero
32-bit constant with the low 12 bits clear (a constant that can be
loaded with the @code{sethi} instruction)
+@item L
+A constant in the range supported by @code{movcc} instructions
+
+@item M
+A constant in the range supported by @code{movrcc} instructions
+
+@item N
+Same as @samp{K}, except that it verifies that bits that are not in the
+lower 32-bits range are all zero. Must be used instead of @samp{K} for
+modes wider than @code{SImode}
+
@item G
Floating-point zero
[-- Attachment #3: Type: text/plain, Size: 289 bytes --]
--
Alexandre Oliva Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist Professional serial bug killer
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: 3.1 doesn't bootstrap on sparcv9
2002-02-22 21:23 ` Alexandre Oliva
2002-02-23 22:57 ` David O'Brien
@ 2002-02-25 11:55 ` Brad Lucier
1 sibling, 0 replies; 6+ messages in thread
From: Brad Lucier @ 2002-02-25 11:55 UTC (permalink / raw)
To: Alexandre Oliva; +Cc: Brad Lucier, gcc
>
> On Feb 23, 2002, Brad Lucier <lucier@math.purdue.edu> wrote:
>
> > I think this has been implicit in some previous messages, but just in
> > case someone thinks that no-one noticed, here's the error message:
>
> Fixing this is in my to-do list, but I wouldn't mind if someone beat
> me to it. It's a bug that used to be hidden in the SPARC port and
> that was exposed by a patch that recently added some sanity checks.
Thank you for the patch, I was able to bootstrap and regtest on
sparcv9-sun-solaris2.8.
Unfortunately, there were quite a number of new regressions, see especially
the gcc tests in
http://gcc.gnu.org/ml/gcc-testresults/2002-02/msg00540.html
versus
http://gcc.gnu.org/ml/gcc-testresults/2002-02/msg00421.html
Brad
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: 3.1 doesn't bootstrap on sparcv9
2002-02-24 4:08 ` Alexandre Oliva
@ 2002-02-25 14:13 ` David O'Brien
0 siblings, 0 replies; 6+ messages in thread
From: David O'Brien @ 2002-02-25 14:13 UTC (permalink / raw)
To: Alexandre Oliva; +Cc: gcc
I was able to bootstrap both a cross and hosted compiler FreeBSD/sparc64
compiler on FreeBSD/i386 after your commit of this patch.
(ie, x86box$ configure --target=sparc64-unknown-freebsd5 and
x86box$ configure --target=sparc64-unknown-freebsd5 --host=sparc64-unknown-freebsd5)
Thanks!!
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2002-02-25 21:28 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2002-02-22 21:18 3.1 doesn't bootstrap on sparcv9 Brad Lucier
2002-02-22 21:23 ` Alexandre Oliva
2002-02-23 22:57 ` David O'Brien
2002-02-24 4:08 ` Alexandre Oliva
2002-02-25 14:13 ` David O'Brien
2002-02-25 11:55 ` Brad Lucier
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).