Re: [perl #32996] Register coloring not dirtying registers for method calls properly

[Leopold Toetsch]

Dan Sugalski <[EMAIL PROTECTED]> wrote:

> The register coloring algorithm's not dirtying registers right, and
> looks to be assuming that things are in registers across method calls
> when they really aren't.

Fixed and thanks for the test case,
leo

Re: [perl #32996] Register coloring not dirtying registers for method calls properly

[Leopold Toetsch]

Leopold Toetsch <[EMAIL PROTECTED]> wrote:
> Dan Sugalski <[EMAIL PROTECTED]> wrote:

>> The register coloring algorithm's not dirtying registers right,

> There are still some compatibility hooks inside parrot that set P1 on
> the caller's end (or on both sides). I'll remove this remenants RSN.

Looking a bit deeper into this bug reveals that above sentence doesn't
describe fully what happens, and it additionally depends on the register
allocator (CVS or new).

1) current CVS allocator
* P28 .. P30 are used to allocate temps and not allocated else
* P31 is reserved for spilling
* the old allocator doesn't have the code WRT preserved registers, which
  was written for the new one (it could be backported)
* the register allocator needs 3 more registers and allocates P0, P1, P3
  (P2 is visible in "set P2, Px" that moves the object in place)

Now implicit register usage strikes back. The register allocator isn't
aware of the side effect that C clobbers P0, P1, and S0.

2) new allocator
* has a list of allocatables usable for temps and has a clear notion
  of preserved vs non-preserved registers

static const char assignable[4][5] = {
   /* 0  1  2  3  4  */
{ 0, 0, 0, 0, 0, }, /* I */
{ 0, 1, 1, 1, 1, }, /* S */
{ 0, 0, 0, 1, 1, }, /* P */
{ 1, 1, 1, 1, 1, }, /* N */
};

* code would execute correctly, one more used register would cause
  spilling

leo

Re: [perl #32996] Register coloring not dirtying registers for method calls properly

[Leopold Toetsch]

Dan Sugalski <[EMAIL PROTECTED]> wrote:

> The register coloring algorithm's not dirtying registers right,

There are still some compatibility hooks inside parrot that set P1 on
the caller's end (or on both sides). I'll remove this remenants RSN.

leo

[perl #32996] Register coloring not dirtying registers for method calls properly

[via RT]

# New Ticket Created by  Dan Sugalski 
# Please include the string:  [perl #32996]
# in the subject line of all future correspondence about this issue. 
# http://rt.perl.org:80/rt3/Ticket/Display.html?id=32996 >


The register coloring algorithm's not dirtying registers right, and 
looks to be assuming that things are in registers across method calls 
when they really aren't. The following test program shows the 
problem. While it should print "In method 1\nIn method 2\n" a few 
dozen times, instead it dies partway through. From a trace it looks 
like the coloring algorithm assumes that an object is in a register 
that gets clobbered (correctly) by the method calls.

.namespace ["Foo"]

.sub __biginit prototyped, @MAIN
newclass $P0, "Foo"
$I1 = find_type "Foo"
$P1 = new $I1
$P1.method1()
$P1.method2()

$P2 = new $I1
$P2.method1()
$P2.method2()

$P3 = new $I1
$P3.method1()
$P3.method2()

$P4 = new $I1
$P4.method1()
$P4.method2()

$P5 = new $I1
$P5.method1()
$P5.method2()

$P6 = new $I1
$P6.method1()
$P6.method2()

$P7 = new $I1
$P7.method1()
$P7.method2()

$P8 = new $I1
$P8.method1()
$P8.method2()

$P9 = new $I1
$P9.method1()
$P9.method2()

$P10 = new $I1
$P10.method1()
$P10.method2()

$P11 = new $I1
$P11.method1()
$P11.method2()

$P12 = new $I1
$P12.method1()
$P12.method2()

$P13 = new $I1
$P13.method1()
$P13.method2()

$P14 = new $I1
$P14.method1()
$P14.method2()

$P15 = new $I1
$P15.method1()
$P15.method2()

$P1.method1()
$P1.method2()
$P2.method1()
$P2.method2()
$P3.method1()
$P3.method2()
$P4.method1()
$P4.method2()
$P5.method1()
$P5.method2()
$P6.method1()
$P6.method2()
$P7.method1()
$P7.method2()
$P8.method1()
$P8.method2()
$P9.method1()
$P9.method2()
$P10.method1()
$P10.method2()
$P11.method1()
$P11.method2()
$P12.method1()
$P12.method2()
$P13.method1()
$P13.method2()
$P14.method1()
$P14.method2()
$P15.method1()
$P15.method2()

end
.end

.sub method1 prototyped, method
print "In method 1\n"
.pcc_begin_return
.pcc_end_return
.end

.sub method2 prototyped, method
print "In method 2\n"
.pcc_begin_return
.pcc_end_return
.end


-- 
Dan

--it's like this---
Dan Sugalski  even samurai
[EMAIL PROTECTED] have teddy bears and even
   teddy bears get drunk