Skip to contentSkip to conversation

GCC -O2 failure for mipsel

DISCUSSION Fuxin Zhang 8 participants 15 messages May 1, 2003 – May 9, 2003 MIPS

lore
Share & tools
lore
#0 lore raw

Hello,
I've met a case where mipsel-linux-gcc -O2 fails,for both 2.96 and the fresh new 3.2.3. Maybe someone can tell me what's wrong.

I've reduced the problem to the test case below,compile it with mipsel-linux-gcc -O2(FROM H.J.Lu's redhat miniport,all version, and 3.2.3 is tested too)

#define PUT_CODE(x,code) ((x)->code = (code))
union test_union {
struct test *t;
int a;
};

struct test {
unsigned short code;
union test_union u[1];
};

char memory[2000];

struct test *test_alloc(int code)
{
struct test *t;
int length=sizeof(struct test);

t = (struct test*)memory;
length = (sizeof(struct test) - sizeof(union test_union)-1)/sizeof(int); for (;length>=0;length--)
((int*)t)[length] = 0;

PUT_CODE(t,code);

return t;
}

int main()
{
struct test *t;

t = test_alloc(4);

printf("t->code=%d\n",t->code);
}

Problem happens in test_alloc:

00400890 <test_alloc>:
400890: 3c1c0fc0 lui gp,0xfc0
400894: 279c77b0 addiu gp,gp,30640
400898: 0399e021 addu gp,gp,t9
40089c: 8f828054 lw v0,-32684(gp)
4008a0: 8f818054 lw at,-32684(gp)
4008a4: 00000000 nop
4008a8: a4240000 sh a0,0(at)
4008ac: 03e00008 jr ra
4008b0: ac400000 sw zero,0(v0)
--->the last sw is wrong,it should not be reorder to run later than 4008a8.

And gcc experts will find this is a simplfied case of rtx_alloc:). Yes,the failure
shows up when i try to compile SPEC CPU2000 176.gcc with -O2. The const0_rtx 's code will be set to zero like above.

Thank you in advance.

Regards

#1 lore raw
Quoted 4 lines · Fuxin Zhang

Hello,
I've met a case where mipsel-linux-gcc -O2 fails,for both 2.96 and the fresh new 3.2.3. Maybe someone can tell me what's wrong.

Your code is incorrect.

Quoted 18 lines

I've reduced the problem to the test case below,compile it with mipsel-linux-gcc -O2(FROM H.J.Lu's redhat miniport,all version, and 3.2.3 is tested too)

#define PUT_CODE(x,code) ((x)->code = (code))
union test_union {
struct test *t;
int a;
};

struct test {
unsigned short code;
union test_union u[1];
};

char memory[2000];

Quoted 8 lines

struct test *test_alloc(int code)
{
struct test *t;
int length=sizeof(struct test);

t = (struct test*)memory;
length = (sizeof(struct test) - sizeof(union test_union)-1)/sizeof(int); for (;length>=0;length--)

This is the errant line:

Quoted 1 line

((int*)t)[length] = 0;

You have declared t as a pointer to struct test, but you're using it as a pointer to int. If you look at Pointers, Section 6.2.2.3 in ISO 9899-1990 you'll see that this results in undefined behaviour.

-fno-strict-aliasing should generate the code you want, but it's better to fix your source. If you want to use a pointer as a different type, put it in a union.

Andrew.

#2 lore raw

Thanks, -fno-strict-aliasing works.
--The actual code can't be changed: because it is part of spec cpu2000:)

Quoted 52 lines · Andrew Haley
Quoted 4 lines · Fuxin Zhang

Hello,
I've met a case where mipsel-linux-gcc -O2 fails,for both 2.96 and the fresh new 3.2.3. Maybe someone can tell me what's wrong.

Your code is incorrect.

Quoted 18 lines

I've reduced the problem to the test case below,compile it with mipsel-linux-gcc -O2(FROM H.J.Lu's redhat miniport,all version, and 3.2.3 is tested too)

#define PUT_CODE(x,code) ((x)->code = (code))
union test_union {
struct test *t;
int a;
};

struct test {
unsigned short code;
union test_union u[1];
};

char memory[2000];

Quoted 8 lines

struct test *test_alloc(int code)
{
struct test *t;
int length=sizeof(struct test);

t = (struct test*)memory;
length = (sizeof(struct test) - sizeof(union test_union)-1)/sizeof(int); for (;length>=0;length--)

This is the errant line:

Quoted 1 line

((int*)t)[length] = 0;

You have declared t as a pointer to struct test, but you're using it as a pointer to int. If you look at Pointers, Section 6.2.2.3 in ISO 9899-1990 you'll see that this results in undefined behaviour.

-fno-strict-aliasing should generate the code you want, but it's better to fix your source. If you want to use a pointer as a different type, put it in a union.

Andrew.

#3 lore raw
Quoted 2 lines · Fuxin Zhang

Thanks, -fno-strict-aliasing works.
--The actual code can't be changed: because it is part of spec cpu2000:)

Perhaps SPEC need to have ISO C explained to them...

Andrew.

#4 lore raw
Quoted 5 lines · Andrew Haley
Quoted 2 lines · Fuxin Zhang

Thanks, -fno-strict-aliasing works.
--The actual code can't be changed: because it is part of spec cpu2000:)

Perhaps SPEC need to have ISO C explained to them...

It's just there so you can't turn on aliasing for reasonable base options. It's all a conspiracy, I tell you...

greg

#5 lore raw

Problem of cross-mipsel-compiler GLIBC-2.3.X

In reply to Greg Lindahl

Dear All,
I want to make a cross-compilered glibc-2.3.x and I get the source from
ftp.gun.org. GCC version is 3.2.3, binutils is 2.13.2.1. The step is as following:

1. Try to build binutils
2. Try to make static GCC
3. Try to make glibc -----> Then it is failed

Is there anybody know what's going on or somebody had successfully to build the crossed glibc-2.3.x?

Thanks and best regs,

#6 lore raw

Re: Problem of cross-mipsel-compiler GLIBC-2.3.X

Quoted 11 lines · smills_ho

Dear All,
I want to make a cross-compilered glibc-2.3.x and I get the source from
ftp.gun.org. GCC version is 3.2.3, binutils is 2.13.2.1. The step is as following:

1. Try to build binutils
2. Try to make static GCC
3. Try to make glibc -----> Then it is failed

Is there anybody know what's going on or somebody had successfully to build the crossed glibc-2.3.x?

A host cross host compiler for linux systems is a little more involved than this :)

However, I don't know where you went wrong since you really didn't provide much in the way of information as to what you did or where it failed.

-eric

Signature · 1 line
Eric Christopher <[email protected]>
#7 lore raw

Re: Problem of cross-mipsel-compiler GLIBC-2.3.X

Oddly enough, I followed these basic steps and wound up with a working cross-compiler from sparc64 -> mipseb (Sun Blade 100 to mips big-endian). Gcc nor glibc gave me any issues.....However, when I tried the same exact steps on i686, glibc complained about libgcc not being available, among other things. It's got me baffled, but I'm not exactly complaining. Currently, it's gcc-3.2.3 (propolice patched) + glibc-2.3.2 + binutils-2.13.90.0.20, which it'll get rebuilt for the new binutils 2.14.

--Kumba

Quoted 24 lines · Eric Christopher
Quoted 11 lines · smills_ho

Dear All,
I want to make a cross-compilered glibc-2.3.x and I get the source from
ftp.gun.org. GCC version is 3.2.3, binutils is 2.13.2.1. The step is as following:

1. Try to build binutils
2. Try to make static GCC
3. Try to make glibc -----> Then it is failed

Is there anybody know what's going on or somebody had successfully to build the crossed glibc-2.3.x?

A host cross host compiler for linux systems is a little more involved than this :)

However, I don't know where you went wrong since you really didn't provide much in the way of information as to what you did or where it failed.

-eric

#12 lore raw

Re: Problem of cross-mipsel-compiler GLIBC-2.3.X

In reply to Eric Christopher

Dear Eric,
I follow the step that provide by Bradley
http://www.ltc.com/~brad/mips/mipsel-linux-cross-toolchain-building.txt

Thanks and best regards

----- Original Message -----
From: "Eric Christopher" <[email protected]>
To: "smills_ho" <[email protected]>
Cc: "Linux/MIPS Development" <[email protected]>; <[email protected]>
Sent: Friday, May 09, 2003 3:21 PM
Subject: Re: Problem of cross-mipsel-compiler GLIBC-2.3.X

Quoted 3 lines · smills_ho
Quoted 2 lines

Dear All,
I want to make a cross-compilered glibc-2.3.x and I get the source

from

Quoted 8 lines
Quoted 8 lines

ftp.gun.org. GCC version is 3.2.3, binutils is 2.13.2.1. The step is as following:

1. Try to build binutils
2. Try to make static GCC
3. Try to make glibc -----> Then it is failed

Is there anybody know what's going on or somebody had successfully to

build

Quoted 14 lines
Quoted 1 line

the crossed glibc-2.3.x?

A host cross host compiler for linux systems is a little more involved than this :)

However, I don't know where you went wrong since you really didn't provide much in the way of information as to what you did or where it failed.

-eric

--
Eric Christopher <[email protected]>

#8 lore raw

Re: Problem of cross-mipsel-compiler GLIBC-2.3.X

In reply to Kumba

Dear Kumba,

    Should we try this binutils-2.13.90.0.20?

We try the version binutils-2.13.90.0.18 (Debian used) and it is failed on cross-gcc step :-(

Thanks and best regards,

----- Original Message -----
From: "Kumba" <[email protected]>
To: <[email protected]>
Sent: Friday, May 09, 2003 3:29 PM
Subject: Re: Problem of cross-mipsel-compiler GLIBC-2.3.X

Quoted 9 lines

Oddly enough, I followed these basic steps and wound up with a working cross-compiler from sparc64 -> mipseb (Sun Blade 100 to mips big-endian). Gcc nor glibc gave me any issues.....However, when I tried the same exact steps on i686, glibc complained about libgcc not being available, among other things. It's got me baffled, but I'm not exactly complaining. Currently, it's gcc-3.2.3 (propolice patched) + glibc-2.3.2 + binutils-2.13.90.0.20, which it'll get rebuilt for the new binutils 2.14.

Quoted 9 lines

--Kumba

Quoted 4 lines · Eric Christopher
Quoted 2 lines · smills_ho

Dear All,
I want to make a cross-compilered glibc-2.3.x and I get the source

from

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

ftp.gun.org. GCC version is 3.2.3, binutils is 2.13.2.1. The step is as following:

1. Try to build binutils
2. Try to make static GCC
3. Try to make glibc -----> Then it is failed

Is there anybody know what's going on or somebody had successfully to

build

Quoted 14 lines
Quoted 12 lines
Quoted 1 line

the crossed glibc-2.3.x?

A host cross host compiler for linux systems is a little more involved than this :)

However, I don't know where you went wrong since you really didn't provide much in the way of information as to what you did or where it failed.

-eric

#9 lore raw

Re: Problem of cross-mipsel-compiler GLIBC-2.3.X

I'm not quite sure. I don't see how trying it could hurt, maybe it
will work, maybe it won't. Although, 2.14.90.0.1 was just released as well, you might try that, or even CVS Head, as I heard that includes last-minute mips fixes. I'm fairly new to the whole cross-compiler thing myself, and recently discovered the HOWTO mentioned in your previous mail, which I'm looking at to try and get an i686->mips cross-compiler going. For now though, my sparc->mips cross-compiler works, so it's what I'll rely on the most.

--Kumba

Quoted 66 lines · smills_ho

Dear Kumba,

    Should we try this binutils-2.13.90.0.20?

We try the version binutils-2.13.90.0.18 (Debian used) and it is failed on cross-gcc step :-(

Thanks and best regards,

----- Original Message -----
From: "Kumba" <[email protected]>
To: <[email protected]>
Sent: Friday, May 09, 2003 3:29 PM
Subject: Re: Problem of cross-mipsel-compiler GLIBC-2.3.X

Quoted 9 lines

Oddly enough, I followed these basic steps and wound up with a working cross-compiler from sparc64 -> mipseb (Sun Blade 100 to mips big-endian). Gcc nor glibc gave me any issues.....However, when I tried the same exact steps on i686, glibc complained about libgcc not being available, among other things. It's got me baffled, but I'm not exactly complaining. Currently, it's gcc-3.2.3 (propolice patched) + glibc-2.3.2 + binutils-2.13.90.0.20, which it'll get rebuilt for the new binutils 2.14.

Quoted 11 lines

--Kumba

Quoted 5 lines · Eric Christopher
Quoted 2 lines · smills_ho

Dear All,
I want to make a cross-compilered glibc-2.3.x and I get the source

from

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

ftp.gun.org. GCC version is 3.2.3, binutils is 2.13.2.1. The step is as following:

1. Try to build binutils
2. Try to make static GCC
3. Try to make glibc -----> Then it is failed

Is there anybody know what's going on or somebody had successfully to

build

Quoted 14 lines
Quoted 12 lines
Quoted 1 line

the crossed glibc-2.3.x?

A host cross host compiler for linux systems is a little more involved than this :)

However, I don't know where you went wrong since you really didn't provide much in the way of information as to what you did or where it failed.

-eric

#10 lore raw

Re: Problem of cross-mipsel-compiler GLIBC-2.3.X

Dear Kumba,

After try several versions of binutils and glibc , it works in crossing
XFree86 4.2.1.
binutils 2.13.2.1 + gcc 3.2.3 + glibc 2.3.2
and some patchs come from debian's packages ( binutils 2.13.90.0.18 , gcc-3.2.1,glibc-2.3.1)

Thanks and best regards,

Smills_ho

----- Original Message -----
From: "Kumba" <[email protected]>
To: <[email protected]>
Sent: Friday, May 09, 2003 4:30 PM
Subject: Re: Problem of cross-mipsel-compiler GLIBC-2.3.X

Quoted 10 lines

I'm not quite sure. I don't see how trying it could hurt, maybe it will work, maybe it won't. Although, 2.14.90.0.1 was just released as well, you might try that, or even CVS Head, as I heard that includes last-minute mips fixes. I'm fairly new to the whole cross-compiler thing myself, and recently discovered the HOWTO mentioned in your previous mail, which I'm looking at to try and get an i686->mips cross-compiler going. For now though, my sparc->mips cross-compiler works, so it's what I'll rely on the most.

Quoted 7 lines

--Kumba

Quoted 3 lines · smills_ho

Dear Kumba,

    Should we try this binutils-2.13.90.0.20?

We try the version binutils-2.13.90.0.18 (Debian used) and it is failed

on

Quoted 30 lines
Quoted 30 lines

cross-gcc step :-(

Thanks and best regards,

----- Original Message -----
From: "Kumba" <[email protected]>
To: <[email protected]>
Sent: Friday, May 09, 2003 3:29 PM
Subject: Re: Problem of cross-mipsel-compiler GLIBC-2.3.X

Quoted 18 lines

Oddly enough, I followed these basic steps and wound up with a working cross-compiler from sparc64 -> mipseb (Sun Blade 100 to mips big-endian). Gcc nor glibc gave me any issues.....However, when I tried the same exact steps on i686, glibc complained about libgcc not being available, among other things. It's got me baffled, but I'm not exactly complaining. Currently, it's gcc-3.2.3 (propolice patched) + glibc-2.3.2 + binutils-2.13.90.0.20, which it'll get rebuilt for the new binutils 2.14.

--Kumba

Quoted 3 lines · Eric Christopher

On Thu, 2003-05-08 at 19:05, smills_ho wrote:

Quoted 6 lines
Quoted 6 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

Dear All,
I want to make a cross-compilered glibc-2.3.x and I get the source

from

Quoted 1 line
Quoted 1 line
Quoted 1 line

ftp.gun.org. GCC version is 3.2.3, binutils is 2.13.2.1. The step is

as

Quoted 29 lines
Quoted 27 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

following:

1. Try to build binutils
2. Try to make static GCC
3. Try to make glibc -----> Then it is failed

Is there anybody know what's going on or somebody had successfully to

build

Quoted 14 lines
Quoted 12 lines
Quoted 1 line

the crossed glibc-2.3.x?

A host cross host compiler for linux systems is a little more involved than this :)

However, I don't know where you went wrong since you really didn't provide much in the way of information as to what you did or where it failed.

-eric

#11 lore raw

Re: Problem of cross-mipsel-compiler GLIBC-2.3.X

In reply to Kumba
Quoted 3 lines · Kumba

will work, maybe it won't. Although, 2.14.90.0.1 was just released as well, you might try that, or even CVS Head, as I heard that includes last-minute mips fixes. I'm fairly new to the whole cross-compiler

I needed 2.14.90.0.1 to get the glibc testcases build right. Without them I had lot's of unresolded "__libc_global_ctors" among others which looked like a symbol visibility problem.
Regards,
-- Guido