[Bug tree-optimization/17884] New: asm 'volatile' is not honored as documented

[Bug tree-optimization/17884] New: asm 'volatile' is not honored as documented

[fjahanian at apple dot com]

In the following test case compiled with -O1 the add instruction is found *outside* the protective
"mtfsf". 'volatile' keyword added to 'asm' was meant to prevent such cases. gcc-3.3 behaves
correctly. Is this a know problem? Is there a PR already for it?

double foo( )
{
        double sum;
        double fpenv;
        double DDD, FFF;

        sum = DDD + FFF;
        asm volatile("mtfsf 255,%0" : : "f" (fpenv));
        return sum;
}

_foo:
        mtfsf 255,f0
        fadd f1,f1,f0
        blr

-- 
           Summary: asm 'volatile' is not honored as documented
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: fjahanian at apple dot com
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: powerpc-apple-darwin7.2.0
  GCC host triplet: powerpc-apple-darwin7.2.0
GCC target triplet: powerpc-apple-darwin7.2.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-10-07 19:35 -------
Confirmed, this is caused by -ftree-ter.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |amacleod at redhat dot com
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
           Keywords|                            |wrong-code
   Last reconfirmed|0000-00-00 00:00:00         |2004-10-07 19:35:47
               date|                            |
            Summary|asm 'volatile' is not       |[4.0 Regression] asm
                   |honored as documented       |'volatile' is not honored as
                   |                            |documented
   Target Milestone|---                         |4.0.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[amacleod at redhat dot com]

------- Additional Comments From amacleod at redhat dot com  2004-10-07 20:02 -------
ok, so how is this suppose to work again?
Entering SSA->normal, I see:

  double FFF;
  double DDD;
  double fpenv;
  double sum;
 
  # BLOCK 0
  # PRED: ENTRY [100.0%]  (fallthru,exec)
  sum_3 = DDD_1 + FFF_2;
  __asm__ __volatile__("mtfsf 255,%0"::"f" fpenv_4);
  return sum_3;
  # SUCC: EXIT [100.0%]


is the volatile in ASM suppose to block all movement across it or something?
There is nothing in the asm instruction indicating it has anything to do with
sum_3....



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[fjahanian at apple dot com]

------- Additional Comments From fjahanian at apple dot com  2004-10-07 20:37 -------
From: http://gcc.gnu.org/onlinedocs/gcc-3.4.0/gcc/Extended-Asm.html#Extended%20Asm

The volatile keyword indicates that the instruction has important side-effects. GCC will not delete a 
volatile asm if it is reachable. (The instruction can still be deleted if GCC can prove that control-flow 
will never reach the location of the instruction.) In addition, GCC will not reschedule instructions across 
a volatile asm instruction. 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[fjahanian at apple dot com]

------- Additional Comments From fjahanian at apple dot com  2004-10-07 21:51 -------
Thanks a lot. I will test the patch on apple-ppc-darwin and let you know.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[fjahanian at apple dot com]

------- Additional Comments From fjahanian at apple dot com  2004-10-07 22:02 -------
My test cases look good with this patch. I will do the usual bootstrap, dejagnu testing on
apple-ppc-darwin. But I can't imagine this patch can break anything.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[amacleod at redhat dot com]

------- Additional Comments From amacleod at redhat dot com  2004-10-08 14:35 -------
So after a discussion, I dont think the patch will be applied.  

A follow on sentence in the documentation reads:

"Note that even a volatile asm instruction can be moved in ways
that appear insignificant to the compiler, such as across jump
instructions.  You can't expect a sequence of volatile asm
instructions to remain perfectly consecutive.  If you want consecutive
output, use a single asm.  Also, GCC will perform some
optimizations across a volatile asm instruction; GCC does not
``forget everything'' when it encounters a volatile asm
instruction the way some other compilers do."


Thats what it is doing in this case, we are moving unrelated instructions across
the volatile asm. 

RTH suggested that we remove the sentence "In addition, GCC will not reschedule
instructions across a volatile asm instruction." from the documention since it
is incorrect given the above note.

It you need instructions not to be moved across a volatile, they must be exposed
in the asm.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[fjahanian at apple dot com]

------- Additional Comments From fjahanian at apple dot com  2004-10-08 16:23 -------
But this is a regression from gcc-3.3. Also, without this patch, there is no other place which checks
for a volatility of an 'asm' statement. Then why not just say in the documentation that 'volatile'
has no effect on an 'asm'? BTW, thanks for preparing the patch for me.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[dalej at apple dot com]

------- Additional Comments From dalej at apple dot com  2004-10-08 18:27 -------
The doc also says "You can prevent an asm from being deleted, moved significantly, or combined,
by writing the keyword volatile after the asm."  Which seems to contradict the later statement
that they can be moved across jumps; if that's not significant motion, what is?  Both sentences
go back to at least 2.95.

So what are the semantics of 'asm volatile'?  If it does not disable code motion, it seems to me
Fariborz is right that it's a NOP and should be deprecated.  But there is a lot of code around here
that uses it, and worked as intended with 3.3.  IMO the incorrect statement in the doc is the
last one, not the first two.



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[amacleod at redhat dot com]

------- Additional Comments From amacleod at redhat dot com  2004-10-08 18:37 -------
The claim is the documentation has been wrong for a long time as well. I don't
beleive it has ever acted as a scheduling barrier.   

volatile does not behave as a NOP because it *does* guarantee that the stmt will
not be removed unless it is unreachable. And therefore anything which feeds it
will stick around and not be moved past it.  That gives a volatile asm the same
semantics as a volatile construct in C.  There aren't really other references to
it in the tree optimizers because I beleive we do the same thing for all asms...
ie, I dont beleive we delete ASM of any sort unless they are unreachable.



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[dnovillo at redhat dot com]

------- Additional Comments From dnovillo at redhat dot com  2004-10-08 18:45 -------
Subject: Re:  [4.0 Regression] asm 'volatile'
	is not honored as documented

On Fri, 2004-10-08 at 14:37, amacleod at redhat dot com wrote:

> ie, I dont beleive we delete ASM of any sort unless they are unreachable.
> 
Right.  DCE on trees always marks ASM expressions as inherently
necessary.


Diego.



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

Re: [Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[Diego Novillo]

On Fri, 2004-10-08 at 14:37, amacleod at redhat dot com wrote:

> ie, I dont beleive we delete ASM of any sort unless they are unreachable.
> 
Right.  DCE on trees always marks ASM expressions as inherently
necessary.


Diego.

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[fjahanian at apple dot com]

------- Additional Comments From fjahanian at apple dot com  2004-10-08 18:48 -------
If I read you correctly, first you say:
"volatile does not behave as a NOP because it *does* guarantee that the stmt will
not be removed unless it is unreachable".
And then:
" I dont beleive we delete ASM of any sort unless they are unreachable".

So, question remains. what does 'volatile' do to an asm in gcc-4.0?

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[amacleod at redhat dot com]

------- Additional Comments From amacleod at redhat dot com  2004-10-08 18:53 -------
It does whatever it use to do in the RTL part of the compiler. 4.0 hasnt changed
that, whatever it is.  I suspect the RTL optimizers are capable of removing
ASMs. So without volatile, the RTL backend might/could remove it.

IIn the future the tree optimizers might be smart enough to do something
differnt with ASMs, and then it could be relevent whether volatile is set or
not.  It just isn't right now.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[fjahanian at apple dot com]

------- Additional Comments From fjahanian at apple dot com  2004-10-08 19:25 -------
Following test case shows that 'volatile' is not needed to prevent removal of an 'asm'.

% cat t.c
int main()
{
        asm ("nop");
}

% mygccm5o -S -O4 t.c
% cat t.s
.text
        .align 2
        .globl _main
_main:
        nop
        blr
        .subsections_via_symbols


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-10-08 19:28 -------
asm with no operands always acts as volatile (this is documented too).

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[fjahanian at apple dot com]

------- Additional Comments From fjahanian at apple dot com  2004-10-08 19:41 -------
change nop->isync , or something else.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] [4.0 Regression] asm 'volatile' is not honored as documented

[dalej at apple dot com]

------- Additional Comments From dalej at apple dot com  2004-10-08 21:04 -------
The question was discussed extensively in 2001.  The phrase
"In addition, GCC will not reschedule
instructions across a volatile asm instruction." 
was added here, by Mark Mitchell based on an earlier comment by rth.
http://gcc.gnu.org/ml/gcc-patches/2001-04/msg01477.html
If you read the whole thread, rth is quite firm that 'asm volatile' is
a scheduling barrier.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] asm 'volatile' is not honored as documented

[rth at gcc dot gnu dot org]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1921 bytes --]


------- Additional Comments From rth at gcc dot gnu dot org  2004-10-12 17:39 -------
Ah, I see some amount of confusion here.

Yes, in the referenced message I do argue for "asm volatile" to be a scheduling
barrier.  You'll note that what I'm most concerned about there is that the asm
remain in the body of the function and not get scheduled into the epilogue.
Which does indeed sound like it might could cause problems.  Too bad I didn't
bother to write a test case demonstrating what I had in mind, but there ya go.

As for transformations that are not *scheduling*, the documentation is also quite
clear: 'GCC will perform some optimizations across a volatile asm instruction;
GCC does not “forget everything” when it encounters a volatile asm instruction'.

Which means that if you try to use volatile asms for fp control word
manipulation, you will be sorely disappointed, because *every* existing pass
of the compiler will thwart you.  If you thought this worked before gcc 4.0,
you were mistaken -- tweak the test case a bit and it'll fail on any arbitrary
gcc release.

I have no particular interest in changing this, because gcc doesn't understand
#pragma FENV_ACCESS, and nothing you do until that support is added is going to
help.  I'll state for the record that FENV_ACCESS will be a *huge* amount of work.

We can clarify the documentation, if you can be specific about what you thought
was confusing, but that's the most I'm willing to accept for this PR.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|wrong-code                  |documentation
            Summary|[4.0 Regression] asm        |asm 'volatile' is not
                   |'volatile' is not honored as|honored as documented
                   |documented                  |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] asm 'volatile' is not honored as documented

[dalej at apple dot com]

------- Additional Comments From dalej at apple dot com  2004-10-12 18:30 -------
OK, thanks.  From this it appears that the only effect of 'asm volatile' that users
can safely rely on is that such an instruction will not be deleted.  If this is
agreeable to everybody, I will revise the documentation to say that.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] asm 'volatile' is not honored as documented

[das at FreeBSD dot ORG]

------- Additional Comments From das at FreeBSD dot ORG  2005-01-14 21:34 -------
Here's another test case that shows that gcc 3.4.2 on i386 will reschedule other
instructions (in particular, other asms that are not volatile) across a volatile
asm:

#include <sys/types.h>
#define __fldenv(env)           __asm __volatile("fldenv %0" : : "m" (env))
#define __fnstsw(__sw)          __asm("fnstsw %0" : "=am" (*(__sw)))
#define __ldmxcsr(csr)          __asm __volatile("ldmxcsr %0" : : "m" (csr))
#define __stmxcsr(csr)          __asm("stmxcsr %0" : "=m" (*(csr)))

typedef struct {
        struct {
                uint32_t        control;
                uint32_t        status;
                uint32_t        tag;
                char            other[16];
        } x87;
        uint32_t                mxcsr;
} fenv_t;

int feraiseexcept(int __excepts);

int
feupdateenv(const fenv_t *envp)
{
        int mxcsr, status;

        __fnstsw(&status);
        __stmxcsr(&mxcsr);
        __fldenv(envp->x87);            /* volatile */
        __ldmxcsr(envp->mxcsr);         /* volatile */
        feraiseexcept((mxcsr | status) & 0x3f);
        return (0);
}

The code generated at -O2 is as follows:

feupdateenv:
        pushl   %ebp
        movl    %esp, %ebp
        subl    $20, %esp
        movl    8(%ebp), %eax
#APP
        fnstsw -4(%ebp)
        stmxcsr -8(%ebp)
        fldenv (%eax)
        ldmxcsr 28(%eax)
        fnstsw %eax              <--- should be movl -4(%ebp), %eax
#NO_APP
        orl     -8(%ebp), %eax
        andl    $63, %eax
        pushl   %eax
        call    feraiseexcept
        xorl    %eax, %eax
        leave
        ret

The fnstsw is moved (duplicated, actually) across the fldenv and ldmxcsr, which
are volatile.  At -O1, gcc gets this "right".

My vote is to fix the code rather than the documentation; if volatile asms don't
act as sequence points, then in code such as the above, it becomes necessary to
apply the sledgehammer of marking *all* asms volatile.  This is because there's
no way to express to gcc that an asm reads from a hard register such as the
FPSR.  If the instructions that write the FPSR (already marked volatile in the
example) acted as sequence points, we would be guaranteed that instructions that
read the FPSR could not be moved past them, without having to make the latter
volatile.

Note that even if you don't go with my idea, the following proposal isn't quite
accurate:

> From this it appears that the only effect of 'asm volatile' that users
> can safely rely on is that such an instruction will not be deleted.

It is also true (hopefully) that volatile asms cannot be reordered with respect
to each other, w.r.t. function calls, etc.  This is not true of ordinary asms,
as seen above.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] asm 'volatile' is not honored as documented

[rth at gcc dot gnu dot org]

------- Additional Comments From rth at gcc dot gnu dot org  2005-01-20 01:33 -------
In reply to comment #20:

Again, this is not scheduling, per se.  This is register
rematerialization.  We have a value at some point, and we
decide that it's cheaper to move the computation rather
than store and reload it.

This is really no different than if we decided to CSE the
computation as in

    __fnstsw(&s1);
    __fldenv(envp->x87);            /* volatile */
    __fnstsw(&s2);
->
    __fnstsw(&s1);
    __fldenv(envp->x87);            /* volatile */
    s2 = s1;

I must repeat myself that the original source code is 
buggy.  You've got asms that affect, or are affected by,
architectural state that is not visible to the compiler.
As such you REALLY REALLY MUST mark the asm as volatile.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] asm 'volatile' is not honored as documented

[rth at gcc dot gnu dot org]

------- Additional Comments From rth at gcc dot gnu dot org  2005-02-02 07:18 -------
No one suggested anything for clarification, but I don't see a bug here.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] asm 'volatile' is not honored as documented

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-10-01 19:48 -------
*** Bug 24165 has been marked as a duplicate of this bug. ***

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bjoern dot m dot haase at
                   |                            |web dot de


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884

[Bug tree-optimization/17884] asm 'volatile' is not honored as documented

[pinskia at gcc dot gnu dot org]

------- Comment #26 from pinskia at gcc dot gnu dot org  2008-09-23 23:50 -------
*** Bug 37631 has been marked as a duplicate of this bug. ***


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pardo at google dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17884