[#6018] gcc-4.3 ICEs with {interrupt,exception,nmi}_handler function attributes in copyprop_hardreg_forward_1, at regrename.c:1787

Submitted By: Mike Frysinger

Open Date

2010-05-05 02:17:11 Close Date

2010-07-29 11:20:59

Priority:

Medium High Assignee:

Vivi Li

Steve Kilbane

Board:

N/A Silicon Revision:

Resolution:

Fixed Fixed In Release:

N/A

Processor:

ALL

Host Operating System:

toolchain rev.:

kernel rev.:

State:

Closed Found In Release:

2009R1.1_RC2

Is this bug repeatable?:

yes

Summary: gcc-4.3 ICEs with {interrupt,exception,nmi}_handler function attributes in copyprop_hardreg_forward_1, at regrename.c:1787

Details:

gcc-4.1.2 compiles this code just fine. replacing "nmi" with "interrupt" or "exception" results in same thing.

$ cat test.c

__attribute__((nmi_handler)) void evt_nmi(void) {}

$ bfin-uclinux-gcc test.c -c

test.c: In function ‘evt_nmi’:

test.c:1: error: insn does not satisfy its constraints:

(insn 18 17 19 2 test.c:3 (set (reg:BI 34 CC)

(const_int 1 [0x1])) 12 {movbi} (nil))

test.c:1: internal compiler error: in extract_constrain_insn_cached, at recog.c:1918

Please submit a full bug report,

with preprocessed source if appropriate.

See <URL:http://blackfin.uclinux.org/gf/project/toolchain/tracker> for instructions.

Follow-ups

--- Steve Kilbane 2010-05-28 10:57:24

I believe I have a fix for this. I'll run some tests over the weekend.

Patch:

Index: bfin.md

===================================================================

--- bfin.md (revision 4603)

+++ bfin.md (working copy)

@@ -540,7 +540,7 @@

})

(define_insn "movbi"

- [(set (match_operand:BI 0 "nonimmediate_operand"

"=x,x,d,md,C,d,C,P1")

+ [(set (match_operand:BI 0 "nonimmediate_operand"

"=x,x,d,md,C,d,C,C")

(match_operand:BI 1 "general_operand"

"x,xKs3,md,d,d,C,P0,P1"))]

--- Robin Getz 2010-06-05 20:41:13

Works for me.

rgetz@curly:~/tmp$ cat foo.c

__attribute__((nmi_handler)) void evt_nmi(void) {}

__attribute__((interrupt_handler)) void evt_interrupt(void) {}

__attribute__((exception_handler)) void evt_exception(void) {}

rgetz@curly:~/tmp$ ~/blackfin/trunk/gcc-elf/bin/bfin-elf-gcc -mcpu=bf537-0.3

-O2 -g foo.c -c -o foo.o

00000000 <_evt_nmi>:

__attribute__((nmi_handler)) void evt_nmi(void) {}

0: 66 01 [--SP] = ASTAT;

2: 00 e8 00 00 LINK 0x0; /* (0) */

6: 01 e8 00 00 UNLINK;

a: 26 01 ASTAT = [SP++];

c: 13 00 RTN;

...

00000010 <_evt_interrupt>:

__attribute__((interrupt_handler)) void evt_interrupt(void) {}

10: 66 01 [--SP] = ASTAT;

12: 00 e8 00 00 LINK 0x0; /* (0) */

16: 01 e8 00 00 UNLINK;

1a: 26 01 ASTAT = [SP++];

1c: 11 00 RTI;

...

00000020 <_evt_exception>:

__attribute__((exception_handler)) void evt_exception(void) {}

20: a6 6f SP += -0xc; /* (-12) */

22: 66 01 [--SP] = ASTAT;

24: 00 e8 00 00 LINK 0x0; /* (0) */

28: 01 e8 00 00 UNLINK;

2c: 26 01 ASTAT = [SP++];

2e: 66 6c SP += 0xc; /* ( 12) */

30: 12 00 RTX;

Marking as fixed.

Steve: -- you should be able to add something to

gcc-4.3/gcc/testsuite/gcc.target/bfin

to test with the elf compiler:

which has something like:

/* { dg-do compile { target bfin-elf-* } } */

/* { dg-options "-O0" "-O1" "-O2" "-O3"

} */

/* { dg-final { scan-assembler "RTI" } } */

/* { dg-final { scan-assembler "RTE" } } */

/* { dg-final { scan-assembler "RTX" } } */

I think that should to the trick to make sure we notice this if it breaks

again.

Feel free to close things when you add the testcase.

-Robin

--- Sonic Zhang 2010-06-07 04:24:03

Hi Steven,

Can you add a new test case for this bug to test suite under gcc?

--- Steve Kilbane 2010-07-29 04:11:54

Added tests a while back:

hdlr_except-1.c

hdlr_intr-1.c

hdlr_intr-2.c

hdlr_nmi-1.c

--- Mike Frysinger 2010-07-29 11:20:59

sounds good ... thanks !

Files

Changes

Commits

Dependencies

Duplicates

Associations