[Bug target/55987] Redundant constant emitted
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55987 --- Comment #2 from Tilman Sauerbeck 2013-02-15 16:50:52 UTC --- Created attachment 29470 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29470 dejagnu testcase Simplified dejagnu test case.
[Bug target/56110] Sub-optimal code: unnecessary CMP after AND
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56110
--- Comment #1 from Tilman Sauerbeck 2013-02-16
16:49:34 UTC ---
Changing the literal in the test function so that it fits in 8 bits makes gcc
go with the TST instruction instead of AND+CMP:
unsigned f2 (unsigned x, unsigned m)
{
if (m & 0x80)
x >>= 8;
return x;
}
=>
tstr1, #128
movner0, r0, lsr #8
bxlr
So I guess I shouldn't ask why gcc generates AND+CMP instead of ANDS, but why
it chooses not to use TST.
[Bug target/56110] Sub-optimal code: unnecessary CMP after AND
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56110 --- Comment #2 from Tilman Sauerbeck 2013-03-16 09:01:17 UTC --- Created attachment 29677 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29677 WIP patch I'm not really sure if this patch is good or bad. The discussion on the ML died off eventually (maybe because I asked one too many silly questions). Anyway, without guidance I won't be able to finish this, but I want to post my WIP here at least.
[Bug target/59689] New: Code bloat introduced by pointer arithmetic
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59689
Bug ID: 59689
Summary: Code bloat introduced by pointer arithmetic
Product: gcc
Version: 4.8.2
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: til...@code-monkey.de
The C snippet below contains two lines which AFAICS should be equivalent, but
depending on which of the two is used, gcc will either generate "good" (few
instructions) or "bad" (more instructions) code.
$ armv5tel-softfloat-linux-gnueabi-gcc -O2 t.c -S -DGOOD -o good.S -std=gnu99
$ armv5tel-softfloat-linux-gnueabi-gcc -O2 t.c -S -DBAD -o bad.S -std=gnu99
$ wc -l good.S bad.S
132 good.S
206 bad.S
$
#define M(c) \
do { \
c = *(++ptr); \
if (!flag && f2()) \
flag = 1; \
if (!flag) { \
f3 (c); \
} \
} while (0)
unsigned f2 (void);
void f3 (unsigned c);
void f()
{
const int size = 512;
char *ptr, line[size];
unsigned flag = 0;
#ifdef GOOD
ptr = &line[0] - 1; // !!
#elif BAD
ptr = line; ptr--; // !!
#else
# error need either GOOD or BAD
#endif
for (;;) {
unsigned c;
M (c); M (c); M (c); M (c);
M (c); M (c); M (c); M (c);
M (c); M (c); M (c); M (c);
M (c); M (c); M (c); M (c);
}
}
[Bug target/59689] Code bloat introduced by pointer arithmetic
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59689 --- Comment #2 from Tilman Sauerbeck --- Sigh, I wondered about that but wasn't sure. Sorry about the noise.
[Bug pending/55987] New: Redundant constant emitted
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55987
Bug #: 55987
Summary: Redundant constant emitted
Classification: Unclassified
Product: gcc
Version: 4.7.2
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: pending
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: til...@code-monkey.de
Target: arm
For this code
uint32_t add (uint32_t x, uint32_t y)
{
uint32_t a, b;
a = (x & 0x7f7f7f7f) + (y & 0x7f7f7f7f);
b = (x ^ y) & ~0x7f7f7f7f;
return a ^ b;
}
gcc 4.7.2 generates the following machine code for ARMv5:
$ gcc -Wall -O2 -mtune=arm9tdmi -march=armv5te -S add.c
ldrr2, .L2 @ .word 2139062143
ldrr3, .L2+4 @ .word -2139062144
eorip, r1, r0
andr1, r1, r2
andr2, r0, r2
andr3, ip, r3
addr0, r1, r2
eorr0, r3, r0
bxlr
gcc missed that we can use bic to clear the low bits, which means
we don't need that 2nd constant:
ldrr2, .L2 @ .word 2139062143
eorip, r1, r0
bicip, ip, r2 @ use bic instead of and
andr1, r1, r2
andr0, r0, r2
addr0, r0, r1
eorr0, r0, ip
bxlr
[Bug target/56096] New: Bad code generated for conditional shift
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56096
Bug #: 56096
Summary: Bad code generated for conditional shift
Classification: Unclassified
Product: gcc
Version: 4.7.2
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: til...@code-monkey.de
Compiling this snippet
unsigned f1 (unsigned x, unsigned m)
{
x >>= ((m & 0x008080) ? 8 : 0);
return x;
}
with gcc 4.7.2 gives this code for ARMv5:
$ armv5tel-softfloat-linux-gnueabi-gcc -O2 -S -o- f.c
[...]
ldrr3, .L4
andr3, r1, r3
cmpr3, #0
movner3, #8 @ XXX
moveqr3, #0 @ XXX
movr0, r0, lsr r3 @ XXX
bxlr
[...]
Those three mov instructions are clearly sub-optimal.
Replacing the ternary operator with an if-statement gives the expected code
sequence:
unsigned f1 (unsigned x, unsigned m)
{
if (m & 0x008080)
x >>= 8;
return x;
}
->
ldrr3, .L6
andr3, r1, r3
cmpr3, #0
movner0, r0, lsr #8
bxlr
ie we saved two mov instructions.
[Bug target/56110] New: Sub-optimal code: unnecessary CMP after AND
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56110
Bug #: 56110
Summary: Sub-optimal code: unnecessary CMP after AND
Classification: Unclassified
Product: gcc
Version: 4.7.2
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: til...@code-monkey.de
Target: arm
Compiling this code
unsigned f1 (unsigned x, unsigned m)
{
if (m & 0x008080)
x >>= 8;
return x;
}
with gcc 4.7.2 gives this code for ARMv5:
$ armv5tel-softfloat-linux-gnueabi-gcc -O2 -S -o- f1.c
[...]
ldrr3, .L6
andr3, r1, r3 @ XXX
cmpr3, #0 @ XXX
movner0, r0, lsr #8
bxlr
[...]
AFAICS we could get rid of the CMP if we used ANDS instead of AND:
ldrr3, .L6
andsr3, r1, r3
movner0, r0, lsr #8
bxlr
