This is the mail archive of the
mailing list for the GCC project.
Re: [wide-int] Add fast path for hosts with HWI widening multiplication
- From: Richard Sandiford <rdsandiford at googlemail dot com>
- To: Kenneth Zadeck <zadeck at naturalbridge dot com>
- Cc: "Joseph S. Myers" <joseph at codesourcery dot com>, ramrad01 at arm dot com, gcc-patches <gcc-patches at gcc dot gnu dot org>, Mike Stump <mikestump at comcast dot net>, Richard Henderson <rth at redhat dot com>
- Date: Thu, 08 May 2014 20:18:00 +0100
- Subject: Re: [wide-int] Add fast path for hosts with HWI widening multiplication
- Authentication-results: sourceware.org; auth=none
- References: <8761r8kgph dot fsf at talisman dot default> <871u1sdbb6 dot fsf at talisman dot default> <CAJA7tRa8sMQbg==u8srmfRGKpbjA2+cF5c466xAHvpjZB+8_4w at mail dot gmail dot com> <Pine dot LNX dot 4 dot 64 dot 1405081636260 dot 4990 at digraph dot polyomino dot org dot uk> <87vbtg2kcq dot fsf at talisman dot default> <536BCECB dot 6070001 at naturalbridge dot com>
Kenneth Zadeck <email@example.com> writes:
> everyone who has a private port will hate you forever. note that i
> have 2 of them.
Got any other ideas though? I suppose if we're prepared to break
compatibility with whatever the upstream of longlong.h is, we could
make more use of intN_t and uintN_t.
Having a whitelist of hosts seems like the best fix though.
I'm not sure the default umul_ppmm is going to be any better
than not defining it.
> On 05/08/2014 02:31 PM, Richard Sandiford wrote:
>> "Joseph S. Myers" <firstname.lastname@example.org> writes:
>>> On Thu, 8 May 2014, Ramana Radhakrishnan wrote:
>>>> <DATE> Ramana Radhakrishnan <email@example.com>
>>>> * wide-int.cc (UTItype): Define.
>>>> (UDWtype): Define for appropriate W_TYPE_SIZE.
>>> This breaks builds for 32-bit hosts, where TImode isn't supported. You
>>> can only use TImode on the host if it's 64-bit.
>>> wide-int.cc:37:56: error: unable to emulate 'TI'
>> The longlong.h interface seems to be designed to be as difficult to use
>> as possible :-( So maybe we really do need to limit it to hosts that are
>> known to work and benefit from it.
>> How about the following? I tested that it produces identical
>> wide-int.o .text for x86_64.
>> I think additions to or removals from the list should be treated as
>> * wide-int.cc: Only include longlong.h for certain targets.
>> Index: gcc/wide-int.cc
>> --- gcc/wide-int.cc 2014-05-08 19:13:15.782158808 +0100
>> +++ gcc/wide-int.cc 2014-05-08 19:28:52.880742385 +0100
>> @@ -27,19 +27,20 @@ along with GCC; see the file COPYING3.
>> #include "tree.h"
>> #include "dumpfile.h"
>> -#if GCC_VERSION >= 3000
>> +#if (GCC_VERSION >= 3000 \
>> + && (defined __aarch64 \
>> + || defined __alpha \
>> + || defined __ia64 \
>> + || defined __powerpc64__ \
>> + || defined __sparcv9 \
>> + || defined __x86_64__))
>> #define W_TYPE_SIZE HOST_BITS_PER_WIDE_INT
>> -typedef unsigned HOST_HALF_WIDE_INT UHWtype;
>> -typedef unsigned HOST_WIDE_INT UWtype;
>> typedef unsigned int UQItype __attribute__ ((mode (QI)));
>> typedef unsigned int USItype __attribute__ ((mode (SI)));
>> typedef unsigned int UDItype __attribute__ ((mode (DI)));
>> -typedef unsigned int UTItype __attribute__ ((mode (TI)));
>> -#if W_TYPE_SIZE == 32
>> -# define UDWtype UDItype
>> -#elif W_TYPE_SIZE == 64
>> -# define UDWtype UTItype
>> +typedef unsigned HOST_HALF_WIDE_INT UHWtype;
>> +typedef unsigned HOST_WIDE_INT UWtype;
>> +typedef unsigned int UDWtype __attribute__ ((mode (TI)));
>> #include "longlong.h"