Mike Stump <mikest...@comcast.net> writes:
> On Jan 14, 2014, at 7:25 AM, Richard Sandiford
> <rsand...@linux.vnet.ibm.com> wrote:
>> Mike Stump <mikest...@comcast.net> writes:
>>> diff --git a/gcc/expmed.c b/gcc/expmed.c
>>> index ce063eb..720d8c1 100644
>>> --- a/gcc/expmed.c
>>> +++ b/gcc/expmed.c
>>> @@ -4963,6 +4963,7 @@ make_tree (tree type, rtx x)
>>> return t;
>>>
>>> case CONST_DOUBLE:
>>> + gcc_assert (HOST_BITS_PER_WIDE_INT * 2 <= MAX_BITSIZE_MODE_ANY_INT);
>>> if (TARGET_SUPPORTS_WIDE_INT == 0 && GET_MODE (x) == VOIDmode)
>>> t = wide_int_to_tree (type,
>>> wide_int::from_array (&CONST_DOUBLE_LOW (x), 2,
>>
>> I think this would be better as a STATIC_ASSERT.
>
> The cost should be the same in generated code. The C folks can read and
> understand the gcc_assert easier, and the static property of the check
> we don't make any use of. I could change it, but there are 0 of these
> in *.c presently, and I think we want to tread lightly on those folks
> that know and like C, and don't want gcc to be `different' for no good
> reason. That said, if people think it would be good to use it, I'd be
> happy to change it.
Static asserts are better than run-time asserts because they're caught
earlier, at GCC compile time. Yes, wide-int is the first thing to make
much use of it, but something has to be first.
>>> @@ -1440,10 +1442,10 @@ real_to_integer (const REAL_VALUE_TYPE *r, bool
>>> *fail, int precision)
>>> }
>>> #endif
>>> w = SIGSZ * HOST_BITS_PER_LONG + words * HOST_BITS_PER_WIDE_INT;
>>> - result = wide_int::from_array
>>> + tmp = real_int::from_array
>>> (val, (w + HOST_BITS_PER_WIDE_INT - 1) / HOST_BITS_PER_WIDE_INT, w);
>>> - result = wi::lrshift (result, (words * HOST_BITS_PER_WIDE_INT) -
>>> exp);
>>> - result = wide_int::from (result, precision, UNSIGNED);
>>> + tmp = wi::lrshift<real_int> (tmp, (words * HOST_BITS_PER_WIDE_INT) -
>>> exp);
>>> + result = wide_int::from (tmp, precision, UNSIGNED);
>>
>> Why did you need the <real_int>? It was supposed to work without.
>
> The code in question needs something that is max int + max significand
> real in size, we made the max int smaller (smaller than this quantity on
> x86) so, this code needs a special wide int that is bigger. The type is
> free as vrp uses the same type. As for why Kenny choose this method,
> I'd defer to him.
To be clear, I was only talking about the <real_int> in
"wi::lrshift<real_int>". Just "wi::lrshift" should be fine.
Tested on x86_64-linux-gnu. OK to install?
Thanks,
Richard
Index: gcc/expmed.c
===================================================================
--- gcc/expmed.c 2014-01-15 16:39:39.881276576 +0000
+++ gcc/expmed.c 2014-01-15 16:39:40.375274550 +0000
@@ -4963,7 +4963,7 @@ make_tree (tree type, rtx x)
return t;
case CONST_DOUBLE:
- gcc_assert (HOST_BITS_PER_WIDE_INT * 2 <= MAX_BITSIZE_MODE_ANY_INT);
+ STATIC_ASSERT (HOST_BITS_PER_WIDE_INT * 2 <= MAX_BITSIZE_MODE_ANY_INT);
if (TARGET_SUPPORTS_WIDE_INT == 0 && GET_MODE (x) == VOIDmode)
t = wide_int_to_tree (type,
wide_int::from_array (&CONST_DOUBLE_LOW (x), 2,
Index: gcc/real.c
===================================================================
--- gcc/real.c 2014-01-15 16:39:39.883276568 +0000
+++ gcc/real.c 2014-01-15 16:39:40.376274546 +0000
@@ -1444,7 +1444,7 @@ real_to_integer (const REAL_VALUE_TYPE *
w = SIGSZ * HOST_BITS_PER_LONG + words * HOST_BITS_PER_WIDE_INT;
tmp = real_int::from_array
(val, (w + HOST_BITS_PER_WIDE_INT - 1) / HOST_BITS_PER_WIDE_INT, w);
- tmp = wi::lrshift<real_int> (tmp, (words * HOST_BITS_PER_WIDE_INT) -
exp);
+ tmp = wi::lrshift (tmp, (words * HOST_BITS_PER_WIDE_INT) - exp);
result = wide_int::from (tmp, precision, UNSIGNED);
if (r->sign)
