[Bug middle-end/35509] [4.3/4.4 Regression] builtin isinf() mismatch to compile-time substitution
--- Comment #7 from dmixm at marine dot febras dot ru 2008-04-14 10:55 --- Possible the (isgreater(fabs(x), DBL_MAX) ? (signbit(x) ? -1 : 1) : 0) will be fast with common finite numbers? Question: in case of AVR target, is it possible to call extern isinf() function regardless of language dialect? Inline code is very undesirable with small memory. Thanks. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35509
[Bug target/35634] New: [avr] result of char promotion comes out of CHAR_MIN/MAX
/* The next program is aborted with avr-gcc 4.1.2, 4.2.3, 4.3.0:
result of char promotion comes out of CHAR_MIN/MAX.
Options: -W -Wall -Os
Know to work:
3.3.6, 3.4.6 - good code
4.0.4 - correct, but not the best
4.1.2, 4.2.3, 4.3.0 - without optimization only
*/
#include limits.h
void abort (void);
void exit (int);
void foo (int i)
{
static int n;
if (i CHAR_MIN || i CHAR_MAX)
abort ();
if (++n 1000)
exit (0);
}
int main ()
{
char c;
for (c = 0; ; c++) foo (c);
}
--
Summary: [avr] result of char promotion comes out of CHAR_MIN/MAX
Product: gcc
Version: 4.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: dmixm at marine dot febras dot ru
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35634
[Bug target/35506] New: [avr] 4.3.0 buid error: illegal opcode movw for mcu avr3
The building for AVR target is aborted at compilation libgcc stage with
error: Illegal opcode movw for mcu avr3. The lastest official release
of binutils is used: 2.18. The reason of error is an attempt to assemble
MOVW command for avr3 architecture which permits only classic instruction
set historically.
The workaround is to add '-mall-opcodes' avr-as option for this case.
Replace into 'config/avr/avr.h' the string:
#define ASM_SPEC %{mmcu=avr25:-mmcu=avr2;
mmcu=avr35:-mmcu=avr3;
mmcu=avr31:-mmcu=avr3;mmcu=avr51:-mmcu=avr5;mmcu=*:-mmcu=%*}
to:
#define ASM_SPEC %{mmcu=avr25:-mmcu=avr2;
mmcu=avr35:-mmcu=avr3 -mall-opcodes;
mmcu=avr31:-mmcu=avr3;mmcu=avr51:-mmcu=avr5;mmcu=*:-mmcu=%*}
After this edition the build and installation are fine.
--
Summary: [avr] 4.3.0 buid error: illegal opcode movw for mcu avr3
Product: gcc
Version: 4.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: dmixm at marine dot febras dot ru
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35506
[Bug target/35507] New: [avr] 4.3.0: size of small funcion increases from 2 to 29 words
For funcion:
long mult (long x, long y)
{
return x * y;
}
the avr-gcc 4.3.0 produces 29 words of code (-Os option):
mult:
push r14
push r15
push r16
push r17
/* prologue: function */
/* frame size = 0 */
mov r14,r18
mov r15,r19
mov r16,r20
mov r17,r21
mov r18,r22
mov r19,r23
mov r20,r24
mov r21,r25
mov r25,r17
mov r24,r16
mov r23,r15
mov r22,r14
rcall __mulsi3
mov r18,r22
mov r19,r23
mov r20,r24
mov r21,r25
mov r23,r19
mov r24,r20
mov r25,r21
/* epilogue start */
pop r17
pop r16
pop r15
pop r14
ret
The result of avr-gcc 4.1.2 is 2 words:
mult:
/* prologue: frame size=0 */
/* prologue end (size=0) */
rcall __mulsi3
/* epilogue: frame size=0 */
ret
--
Summary: [avr] 4.3.0: size of small funcion increases from 2 to
29 words
Product: gcc
Version: 4.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: dmixm at marine dot febras dot ru
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35507
[Bug target/35508] New: [avr] 4.3.0: undefined reference to `__ffshi2'
The next program is not linked (-Os option):
int ffs (int);
int main (int argc, char *argv[])
{
(void)argv;
return ffs (argc);
}
with undefined link to '__ffshi2'.
The used libc library (Avr-libc project) contains the ffs() funcion.
--
Summary: [avr] 4.3.0: undefined reference to `__ffshi2'
Product: gcc
Version: 4.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: dmixm at marine dot febras dot ru
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35508
[Bug target/35509] New: [avr] 4.3.0: builtin isinf() mismatch to compile-time substitution
The next program is aborted in execution (avr-gcc 4.3.0, -Os):
int isinf (double);
void abort (void);
volatile double x = -1.0/0.0;
int main ()
{
if (isinf (x) != isinf (-__builtin_inf ()))
abort ();
return 0;
}
The second comparison argument is evaluated at compile time: -1 value
is substituted. This is traditional result for isinf(-Inf). The first
call of isinf() is replaced to inline code, which evaluates the -Inf
to +1 value.
The early avr-gcc versions are fine: the extern isinf() function was called.
--
Summary: [avr] 4.3.0: builtin isinf() mismatch to compile-time
substitution
Product: gcc
Version: 4.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: dmixm at marine dot febras dot ru
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35509
[Bug target/11180] [avr-gcc] Optimization decrease performance of struct assignment.
--- Comment #14 from dmixm at marine dot febras dot ru 2007-09-16 03:22
---
It was:
(define_insn movdi
[(set (match_operand:DI 0 nonimmediate_operand )
(match_operand:DI 1 general_operand ))]
#)
(define_split
[(set (match_operand:DI 0 nonimmediate_operand )
(match_operand:DI 1 general_operand ))]
reload_completed
[(set (match_dup 2) (match_dup 4))
(set (match_dup 3) (match_dup 5))]
{
operands[2] = simplify_gen_subreg (SImode, operands[0], DImode, 0);
operands[3] = simplify_gen_subreg (SImode, operands[0], DImode, 4);
operands[4] = simplify_gen_subreg (SImode, operands[1], DImode, 0);
operands[5] = simplify_gen_subreg (SImode, operands[1], DImode, 4);
})
Alas, this elementary addition has appeared erroneous.
The following program leads to emergency end of compilation:
int main ()
{
volatile long long x = 0x0102030405060708LL;
if (x != 0x0102030405060708LL)
exit (__LINE__);
exit (0);
}
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11180
[Bug target/11180] [avr-gcc] Optimization decrease performance of struct assignment.
--- Comment #12 from dmixm at marine dot febras dot ru 2007-09-09 21:59 --- Andy Hutchinson wrote (comment #6) that addition a 'movdi' instruction improves the result. I have try to add a very simple 'movdi' (which split into 2 SImode instuctions). In result: -O0 -- 85 words, -O1,2,3,s -- 50 words. Version is 4.2.1 -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11180
[Bug target/11180] [avr-gcc] Optimization decrease performance of struct assignment.
--- Comment #10 from dmixm at marine dot febras dot ru 2007-07-27 01:24 --- Yes, results are: avr-gcc-3.3.6: O0 -- 75, O1,O2,O3,Os -- 79 avr-gcc-4.2.1: O0 -- 109, O1,O2,O3,Os -- 79 The mistake is corrected? It is possible to tell and so as now application of keys of optimization shortens a code. However, correction does not improve quality of optimization, it only worsens not optimized code. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11180
[Bug target/30289] New: avr-gcc: builtin memset(): wrong code
Thanks to Tom [EMAIL PROTECTED].
Program:
void * memset (void *buf, int val, unsigned int n);
void foo (void)
{
memset ((char *)0x4000, 0, 0x4000);
}
is compiled into wrong code, where pointer and counter use 1 register pair:
ldi r30,lo8(16384)
ldi r31,hi8(16384)
st Z+,__zero_reg__
sbiw r30,1
brne .-6
Compiler: 4.1.1
Options: -W -Wall -Os -S
I have try a set of versions:
avr-gcc 3.3.5 - OK, loop counter is separate register pair
avr-gcc 3.4.4, 4.0.2, 4.1.1 - wrong code.
Possible, the reason is in absence of '' in clobber statement 'avr.md':
...
(use (match_operand:HI 1 register_operand !w,d))
(use (match_operand:HI 2 const_int_operand n,n))
(clobber (match_scratch:HI 3 =0,0))
(clobber (match_scratch:HI 4 =1,1))]
...
--
Summary: avr-gcc: builtin memset(): wrong code
Product: gcc
Version: 4.1.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: dmixm at marine dot febras dot ru
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30289
[Bug target/30243] New: avr-gcc 4.1.1: signbit() causes an internal compiler error
avr-gcc (GCC) 4.1.1
Compilation of next program (-W -Wall -Os -S):
extern int signbit (double);
int foo (double x)
{
return signbit (x);
}
causes an compilation abort with message:
internal compiler error: in gen_lowpart_general, at rtlhooks.c:51
Older versions are work:
3.3.5, 3.4.4 - call of external signbit() function
4.0.2- inline-code
--
Summary: avr-gcc 4.1.1: signbit() causes an internal compiler
error
Product: gcc
Version: 4.1.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: dmixm at marine dot febras dot ru
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30243
[Bug target/19684] New: avr-gcc 4.0 (and 3.3.4): wrong size in asm comment
Comment about function `foo' size:
27/* function foo size 10 (9) */
show wrong value. (Compare it with next function.)
Compiler: avr-gcc (GCC) 4.0.0 20050109 (experimental)
Options: -W -Wall -Os -Wa,-ahl
The same error (for `foo') in 3.3.4 .
Certainly, it is nonsense. But, maybe, this display
of more serious mistake?
Program:
int foo (int x)
{
int i;
for (i= 10; i; i--)
x *= 2;
return x;
}
int foo_with_true_size (int x)
{
x += 100;
do {
x *= 2;
} while (x 0);
return x;
}
Listing:
GAS LISTING /tmp/ccJGSrye.s page 1
1.file foo.c
2.arch avr2
3__SREG__ = 0x3f
4__SP_H__ = 0x3e
5__SP_L__ = 0x3d
6__tmp_reg__ = 0
7__zero_reg__ = 1
8.global __do_copy_data
9.global __do_clear_bss
10.text
11.global foo
12.type foo, @function
13foo:
14/* prologue: frame size=0 */
15/* prologue end (size=0) */
16 2AE0 ldi r18,lo8(10)
17 0002 30E0 ldi r19,hi8(10)
18.L2:
19 0004 880F add r24,r24
20 0006 991F adc r25,r25
21 0008 2150 subi r18,lo8(-(-1))
22 000a 3040 sbci r19,hi8(-(-1))
23 000c D9F7 brne .L2
24/* epilogue: frame size=0 */
25 000e 0895 ret
26/* epilogue end (size=1) */
27/* function foo size 10 (9) */
28.size foo, .-foo
29.global foo_with_true_size
30.type foo_with_true_size, @function
31foo_with_true_size:
32/* prologue: frame size=0 */
33/* prologue end (size=0) */
34 0010 8C59 subi r24,lo8(-(100))
35 0012 9F4F sbci r25,hi8(-(100))
36.L8:
37 0014 880F add r24,r24
38 0016 991F adc r25,r25
39 0018 1816 cp __zero_reg__,r24
40 001a 1906 cpc __zero_reg__,r25
41 001c DCF3 brlt .L8
42/* epilogue: frame size=0 */
43 001e 0895 ret
44/* epilogue end (size=1) */
45/* function foo_with_true_size size 8 (7) */
46.size foo_with_true_size,
.-foo_with_true_size
47/* File foo.c: code 18 = 0x0012 ( 16), prologues
0, epilogues 2 */
--
Summary: avr-gcc 4.0 (and 3.3.4): wrong size in asm comment
Product: gcc
Version: 4.0.0
Status: UNCONFIRMED
Severity: normal
Priority: P2
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: dmixm at marine dot febras dot ru
CC: gcc-bugs at gcc dot gnu dot org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19684
[Bug target/19686] New: avr-gcc 4.0: loop performance decrease (extra conversion)
Program:
int foo (int x)
{
int i= x;
do {
x *= 2;
} while (--i);
return x;
}
avr-gcc 4.0 (20050116) with `-Os' option:
foo:
/* prologue: frame size=0 */
/* prologue end (size=0) */
ldi r18,lo8(0)
ldi r19,hi8(0)
clr r20
clr r21
sub r20,r24
sbc r21,r25
.L2:
add r24,r24
adc r25,r25
subi r18,lo8(-(-1))
sbci r19,hi8(-(-1))
cp r18,r20
cpc r19,r21
brne .L2
/* epilogue: frame size=0 */
ret
avr-gcc 3.3.4:
foo:
/* prologue: frame size=0 */
/* prologue end (size=0) */
mov r18,r24
mov r19,r25
.L2:
add r24,r24
adc r25,r25
subi r18,lo8(-(-1))
sbci r19,hi8(-(-1))
brne .L2
/* epilogue: frame size=0 */
ret
--
Summary: avr-gcc 4.0: loop performance decrease (extra
conversion)
Product: gcc
Version: 4.0.0
Status: UNCONFIRMED
Severity: normal
Priority: P2
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: dmixm at marine dot febras dot ru
CC: gcc-bugs at gcc dot gnu dot org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19686
[Bug rtl-optimization/19597] New: avr-gcc 4.0, multiplication by constant, very long code
avr-gcc 4.0 (20050116) replaces multiplication by a constant to a combination
of more simple
operations. Thus it does not pay attention to efficiency of such replacement.
For example:
int foo (int x) { return 24011 * x; }
produce 49 words (-mmcu=atmega8 -Os).
Older GCC branches -- only 11 words in this case.
--
Summary: avr-gcc 4.0, multiplication by constant, very long code
Product: gcc
Version: 4.0.0
Status: UNCONFIRMED
Severity: normal
Priority: P2
Component: rtl-optimization
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: dmixm at marine dot febras dot ru
CC: gcc-bugs at gcc dot gnu dot org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19597
[Bug middle-end/18424] 3.4.3 ~6x+ performance regression vs 3.3.1, constant trees not being computed.
--- Additional Comments From dmixm at marine dot febras dot ru 2004-11-16
23:58 ---
In March, 2004 Richard Sandiford has offered a patch for elimination of
this problem. See: http://gcc.gnu.org/ml/gcc/2004-03/msg01456.html This
patch modifies function do_jump (a file dojump.c). This change now is
present at a branch 4.0, but does not enter in 3.4.x. I have tried to
apply this patch to 3.4.3. It has earned only after change of a line
if (TREE_CODE (TREE_OPERAND (exp, 0)) == RSHIFT_EXPR
...
to
tree arg_nops = TREE_OPERAND (exp, 0); /* and below the same subst. */
STRIP_NOPS (arg_nops);
if (TREE_CODE (arg_nops) == RSHIFT_EXPR
...
and only for function foo_i.
foo_i() before patch:
...
mov r24,r25
ldi r25,6
1: lsr r24
dec r25
brne 1b
sbrs r24,0
rjmp .L2
foo_i() after patch:
...
sbrs r25,6
rjmp .L8
But foo_ll (shift loop with count 62!) and foo_l have remained on old -
through shift of the left argument.
Source file:
~~~
int foo_ll (long long x)
{
return (x 0x4000LL) ? 1 : 3;
}
int foo_l (long x)
{
return (x 0x4000) ? 5 : 7;
}
int foo_i (int x)
{
return (x 0x4000) ? 9 : 11;
}
int foo_c (char x)
{
return (x 0x40) ? 13 : 15;
}
P.S. The code for foo_c was and remains beautiful due to work
`gcc/combine.c' .
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18424
