[0/9] LoongArch: Use builtins with GCC >= 13

Message ID 20221118112912.2501774-1-tangxiaolin@loongson.cn
Headers
Series LoongArch: Use builtins with GCC >= 13 |

Message

Xiaolin Tang Nov. 18, 2022, 11:29 a.m. UTC
  Patch 0001-0007:
GCC 13, LoongArch support builtins for {,l,ll}rint, logb,
use these built-ins instead of generic function implementation.

Patch 0008:
GCC 13, LoongArch support ___builtin_scalbn{,f} with -fno-math-errno,
but only "libm" can use -fno-math-errno in GLIBC, and scalbn is in libc
instead of libm because __printf_fp calls it. So inline assembler of
the scalbn functions is a better choice for LoongArch.

Patch 0009: Add LoongArch support for scalb[f].

Discuss with Xi Ruoyao<xry111@xry111.site>
Link:
https://sourceware.org/pipermail/libc-alpha/2022-November/143425.html

Xi Ruoyao (1):
  LoongArch: Use __builtin_rint{,f} with GCC >= 13

Xiaolin Tang (8):
  Use GCC builtins for lrint functions if desired.
  LoongArch: Use __builtin_lrint{,f} with GCC >= 13
  Use GCC builtins for llrint functions if desired.
  LoongArch: Use __builtin_llrint{,f} with GCC >= 13
  Use GCC builtins for logb functions if desired.
  LoongArch: Use __builtin_logb{,f} with GCC >= 13
  LoongArch: Add support for scalbn[f]
  LoongArch: Add support for scalb[f]

 sysdeps/generic/math-use-builtins-llrint.h    |  4 ++
 sysdeps/generic/math-use-builtins-logb.h      |  4 ++
 sysdeps/generic/math-use-builtins-lrint.h     |  4 ++
 sysdeps/generic/math-use-builtins.h           |  3 +
 sysdeps/ieee754/dbl-64/s_llrint.c             | 18 ++++--
 sysdeps/ieee754/dbl-64/s_logb.c               |  5 ++
 sysdeps/ieee754/dbl-64/s_lrint.c              | 18 ++++--
 sysdeps/ieee754/float128/float128_private.h   |  6 ++
 sysdeps/ieee754/flt-32/s_llrintf.c            | 18 ++++--
 sysdeps/ieee754/flt-32/s_logbf.c              |  5 ++
 sysdeps/ieee754/flt-32/s_lrintf.c             | 18 ++++--
 sysdeps/ieee754/ldbl-128/s_llrintl.c          | 17 ++++--
 sysdeps/ieee754/ldbl-128/s_logbl.c            |  6 ++
 sysdeps/ieee754/ldbl-128/s_lrintl.c           | 17 ++++--
 sysdeps/loongarch/fpu/e_scalb.c               | 60 +++++++++++++++++++
 sysdeps/loongarch/fpu/e_scalbf.c              | 60 +++++++++++++++++++
 .../loongarch/fpu/math-use-builtins-llrint.h  | 10 ++++
 .../loongarch/fpu/math-use-builtins-logb.h    | 10 ++++
 .../loongarch/fpu/math-use-builtins-lrint.h   | 10 ++++
 .../loongarch/fpu/math-use-builtins-rint.h    |  9 +++
 sysdeps/loongarch/fpu/s_scalbn.c              | 29 +++++++++
 sysdeps/loongarch/fpu/s_scalbnf.c             | 29 +++++++++
 22 files changed, 326 insertions(+), 34 deletions(-)
 create mode 100644 sysdeps/generic/math-use-builtins-llrint.h
 create mode 100644 sysdeps/generic/math-use-builtins-logb.h
 create mode 100644 sysdeps/generic/math-use-builtins-lrint.h
 create mode 100644 sysdeps/loongarch/fpu/e_scalb.c
 create mode 100644 sysdeps/loongarch/fpu/e_scalbf.c
 create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-llrint.h
 create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-logb.h
 create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-lrint.h
 create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-rint.h
 create mode 100644 sysdeps/loongarch/fpu/s_scalbn.c
 create mode 100644 sysdeps/loongarch/fpu/s_scalbnf.c
  

Comments

Xi Ruoyao Nov. 22, 2022, 4:20 p.m. UTC | #1
Hi Xiaolin,

I guess you need to resent the series because {1,8,9}/9 lacks the proper
In-Reply-To: header, causing them missing from the patchwork interface
(https://patchwork.sourceware.org/project/glibc/list/?series=14494). 
The reviewers will use the patchwork interface to track the patches.

On Fri, 2022-11-18 at 19:29 +0800, Xiaolin Tang wrote:
> Patch 0001-0007:
> GCC 13, LoongArch support builtins for {,l,ll}rint, logb,
> use these built-ins instead of generic function implementation.
> 
> Patch 0008:
> GCC 13, LoongArch support ___builtin_scalbn{,f} with -fno-math-errno,
> but only "libm" can use -fno-math-errno in GLIBC, and scalbn is in
> libc
> instead of libm because __printf_fp calls it. So inline assembler of
> the scalbn functions is a better choice for LoongArch.
> 
> Patch 0009: Add LoongArch support for scalb[f].
> 
> Discuss with Xi Ruoyao<xry111@xry111.site>
> Link:
> https://sourceware.org/pipermail/libc-alpha/2022-November/143425.html
> 
> Xi Ruoyao (1):
>   LoongArch: Use __builtin_rint{,f} with GCC >= 13
> 
> Xiaolin Tang (8):
>   Use GCC builtins for lrint functions if desired.
>   LoongArch: Use __builtin_lrint{,f} with GCC >= 13
>   Use GCC builtins for llrint functions if desired.
>   LoongArch: Use __builtin_llrint{,f} with GCC >= 13
>   Use GCC builtins for logb functions if desired.
>   LoongArch: Use __builtin_logb{,f} with GCC >= 13
>   LoongArch: Add support for scalbn[f]
>   LoongArch: Add support for scalb[f]
> 
>  sysdeps/generic/math-use-builtins-llrint.h    |  4 ++
>  sysdeps/generic/math-use-builtins-logb.h      |  4 ++
>  sysdeps/generic/math-use-builtins-lrint.h     |  4 ++
>  sysdeps/generic/math-use-builtins.h           |  3 +
>  sysdeps/ieee754/dbl-64/s_llrint.c             | 18 ++++--
>  sysdeps/ieee754/dbl-64/s_logb.c               |  5 ++
>  sysdeps/ieee754/dbl-64/s_lrint.c              | 18 ++++--
>  sysdeps/ieee754/float128/float128_private.h   |  6 ++
>  sysdeps/ieee754/flt-32/s_llrintf.c            | 18 ++++--
>  sysdeps/ieee754/flt-32/s_logbf.c              |  5 ++
>  sysdeps/ieee754/flt-32/s_lrintf.c             | 18 ++++--
>  sysdeps/ieee754/ldbl-128/s_llrintl.c          | 17 ++++--
>  sysdeps/ieee754/ldbl-128/s_logbl.c            |  6 ++
>  sysdeps/ieee754/ldbl-128/s_lrintl.c           | 17 ++++--
>  sysdeps/loongarch/fpu/e_scalb.c               | 60
> +++++++++++++++++++
>  sysdeps/loongarch/fpu/e_scalbf.c              | 60
> +++++++++++++++++++
>  .../loongarch/fpu/math-use-builtins-llrint.h  | 10 ++++
>  .../loongarch/fpu/math-use-builtins-logb.h    | 10 ++++
>  .../loongarch/fpu/math-use-builtins-lrint.h   | 10 ++++
>  .../loongarch/fpu/math-use-builtins-rint.h    |  9 +++
>  sysdeps/loongarch/fpu/s_scalbn.c              | 29 +++++++++
>  sysdeps/loongarch/fpu/s_scalbnf.c             | 29 +++++++++
>  22 files changed, 326 insertions(+), 34 deletions(-)
>  create mode 100644 sysdeps/generic/math-use-builtins-llrint.h
>  create mode 100644 sysdeps/generic/math-use-builtins-logb.h
>  create mode 100644 sysdeps/generic/math-use-builtins-lrint.h
>  create mode 100644 sysdeps/loongarch/fpu/e_scalb.c
>  create mode 100644 sysdeps/loongarch/fpu/e_scalbf.c
>  create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-llrint.h
>  create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-logb.h
>  create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-lrint.h
>  create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-rint.h
>  create mode 100644 sysdeps/loongarch/fpu/s_scalbn.c
>  create mode 100644 sysdeps/loongarch/fpu/s_scalbnf.c
>
  
Xiaolin Tang Nov. 23, 2022, 2:14 a.m. UTC | #2
OK, thank you for pointing this out.  Please delete them, and I will 
resend this series.


在 2022/11/23 上午12:20, Xi Ruoyao 写道:
> Hi Xiaolin,
>
> I guess you need to resent the series because {1,8,9}/9 lacks the proper
> In-Reply-To: header, causing them missing from the patchwork interface
> (https://patchwork.sourceware.org/project/glibc/list/?series=14494).
> The reviewers will use the patchwork interface to track the patches.
>
> On Fri, 2022-11-18 at 19:29 +0800, Xiaolin Tang wrote:
>> Patch 0001-0007:
>> GCC 13, LoongArch support builtins for {,l,ll}rint, logb,
>> use these built-ins instead of generic function implementation.
>>
>> Patch 0008:
>> GCC 13, LoongArch support ___builtin_scalbn{,f} with -fno-math-errno,
>> but only "libm" can use -fno-math-errno in GLIBC, and scalbn is in
>> libc
>> instead of libm because __printf_fp calls it. So inline assembler of
>> the scalbn functions is a better choice for LoongArch.
>>
>> Patch 0009: Add LoongArch support for scalb[f].
>>
>> Discuss with Xi Ruoyao<xry111@xry111.site>
>> Link:
>> https://sourceware.org/pipermail/libc-alpha/2022-November/143425.html
>>
>> Xi Ruoyao (1):
>>    LoongArch: Use __builtin_rint{,f} with GCC >= 13
>>
>> Xiaolin Tang (8):
>>    Use GCC builtins for lrint functions if desired.
>>    LoongArch: Use __builtin_lrint{,f} with GCC >= 13
>>    Use GCC builtins for llrint functions if desired.
>>    LoongArch: Use __builtin_llrint{,f} with GCC >= 13
>>    Use GCC builtins for logb functions if desired.
>>    LoongArch: Use __builtin_logb{,f} with GCC >= 13
>>    LoongArch: Add support for scalbn[f]
>>    LoongArch: Add support for scalb[f]
>>
>>   sysdeps/generic/math-use-builtins-llrint.h    |  4 ++
>>   sysdeps/generic/math-use-builtins-logb.h      |  4 ++
>>   sysdeps/generic/math-use-builtins-lrint.h     |  4 ++
>>   sysdeps/generic/math-use-builtins.h           |  3 +
>>   sysdeps/ieee754/dbl-64/s_llrint.c             | 18 ++++--
>>   sysdeps/ieee754/dbl-64/s_logb.c               |  5 ++
>>   sysdeps/ieee754/dbl-64/s_lrint.c              | 18 ++++--
>>   sysdeps/ieee754/float128/float128_private.h   |  6 ++
>>   sysdeps/ieee754/flt-32/s_llrintf.c            | 18 ++++--
>>   sysdeps/ieee754/flt-32/s_logbf.c              |  5 ++
>>   sysdeps/ieee754/flt-32/s_lrintf.c             | 18 ++++--
>>   sysdeps/ieee754/ldbl-128/s_llrintl.c          | 17 ++++--
>>   sysdeps/ieee754/ldbl-128/s_logbl.c            |  6 ++
>>   sysdeps/ieee754/ldbl-128/s_lrintl.c           | 17 ++++--
>>   sysdeps/loongarch/fpu/e_scalb.c               | 60
>> +++++++++++++++++++
>>   sysdeps/loongarch/fpu/e_scalbf.c              | 60
>> +++++++++++++++++++
>>   .../loongarch/fpu/math-use-builtins-llrint.h  | 10 ++++
>>   .../loongarch/fpu/math-use-builtins-logb.h    | 10 ++++
>>   .../loongarch/fpu/math-use-builtins-lrint.h   | 10 ++++
>>   .../loongarch/fpu/math-use-builtins-rint.h    |  9 +++
>>   sysdeps/loongarch/fpu/s_scalbn.c              | 29 +++++++++
>>   sysdeps/loongarch/fpu/s_scalbnf.c             | 29 +++++++++
>>   22 files changed, 326 insertions(+), 34 deletions(-)
>>   create mode 100644 sysdeps/generic/math-use-builtins-llrint.h
>>   create mode 100644 sysdeps/generic/math-use-builtins-logb.h
>>   create mode 100644 sysdeps/generic/math-use-builtins-lrint.h
>>   create mode 100644 sysdeps/loongarch/fpu/e_scalb.c
>>   create mode 100644 sysdeps/loongarch/fpu/e_scalbf.c
>>   create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-llrint.h
>>   create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-logb.h
>>   create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-lrint.h
>>   create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-rint.h
>>   create mode 100644 sysdeps/loongarch/fpu/s_scalbn.c
>>   create mode 100644 sysdeps/loongarch/fpu/s_scalbnf.c
>>
  
Xiaolin Tang Nov. 23, 2022, 2:55 a.m. UTC | #3
Oops, I just resend {1,8,9}/9 with cover-patch, and still haven't solved 
this problem.

I will resend this series again with v2 version, which is added ilogb 
function.


在 2022/11/23 上午12:20, Xi Ruoyao 写道:
> Hi Xiaolin,
>
> I guess you need to resent the series because {1,8,9}/9 lacks the proper
> In-Reply-To: header, causing them missing from the patchwork interface
> (https://patchwork.sourceware.org/project/glibc/list/?series=14494).
> The reviewers will use the patchwork interface to track the patches.
>
> On Fri, 2022-11-18 at 19:29 +0800, Xiaolin Tang wrote:
>> Patch 0001-0007:
>> GCC 13, LoongArch support builtins for {,l,ll}rint, logb,
>> use these built-ins instead of generic function implementation.
>>
>> Patch 0008:
>> GCC 13, LoongArch support ___builtin_scalbn{,f} with -fno-math-errno,
>> but only "libm" can use -fno-math-errno in GLIBC, and scalbn is in
>> libc
>> instead of libm because __printf_fp calls it. So inline assembler of
>> the scalbn functions is a better choice for LoongArch.
>>
>> Patch 0009: Add LoongArch support for scalb[f].
>>
>> Discuss with Xi Ruoyao<xry111@xry111.site>
>> Link:
>> https://sourceware.org/pipermail/libc-alpha/2022-November/143425.html
>>
>> Xi Ruoyao (1):
>>    LoongArch: Use __builtin_rint{,f} with GCC >= 13
>>
>> Xiaolin Tang (8):
>>    Use GCC builtins for lrint functions if desired.
>>    LoongArch: Use __builtin_lrint{,f} with GCC >= 13
>>    Use GCC builtins for llrint functions if desired.
>>    LoongArch: Use __builtin_llrint{,f} with GCC >= 13
>>    Use GCC builtins for logb functions if desired.
>>    LoongArch: Use __builtin_logb{,f} with GCC >= 13
>>    LoongArch: Add support for scalbn[f]
>>    LoongArch: Add support for scalb[f]
>>
>>   sysdeps/generic/math-use-builtins-llrint.h    |  4 ++
>>   sysdeps/generic/math-use-builtins-logb.h      |  4 ++
>>   sysdeps/generic/math-use-builtins-lrint.h     |  4 ++
>>   sysdeps/generic/math-use-builtins.h           |  3 +
>>   sysdeps/ieee754/dbl-64/s_llrint.c             | 18 ++++--
>>   sysdeps/ieee754/dbl-64/s_logb.c               |  5 ++
>>   sysdeps/ieee754/dbl-64/s_lrint.c              | 18 ++++--
>>   sysdeps/ieee754/float128/float128_private.h   |  6 ++
>>   sysdeps/ieee754/flt-32/s_llrintf.c            | 18 ++++--
>>   sysdeps/ieee754/flt-32/s_logbf.c              |  5 ++
>>   sysdeps/ieee754/flt-32/s_lrintf.c             | 18 ++++--
>>   sysdeps/ieee754/ldbl-128/s_llrintl.c          | 17 ++++--
>>   sysdeps/ieee754/ldbl-128/s_logbl.c            |  6 ++
>>   sysdeps/ieee754/ldbl-128/s_lrintl.c           | 17 ++++--
>>   sysdeps/loongarch/fpu/e_scalb.c               | 60
>> +++++++++++++++++++
>>   sysdeps/loongarch/fpu/e_scalbf.c              | 60
>> +++++++++++++++++++
>>   .../loongarch/fpu/math-use-builtins-llrint.h  | 10 ++++
>>   .../loongarch/fpu/math-use-builtins-logb.h    | 10 ++++
>>   .../loongarch/fpu/math-use-builtins-lrint.h   | 10 ++++
>>   .../loongarch/fpu/math-use-builtins-rint.h    |  9 +++
>>   sysdeps/loongarch/fpu/s_scalbn.c              | 29 +++++++++
>>   sysdeps/loongarch/fpu/s_scalbnf.c             | 29 +++++++++
>>   22 files changed, 326 insertions(+), 34 deletions(-)
>>   create mode 100644 sysdeps/generic/math-use-builtins-llrint.h
>>   create mode 100644 sysdeps/generic/math-use-builtins-logb.h
>>   create mode 100644 sysdeps/generic/math-use-builtins-lrint.h
>>   create mode 100644 sysdeps/loongarch/fpu/e_scalb.c
>>   create mode 100644 sysdeps/loongarch/fpu/e_scalbf.c
>>   create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-llrint.h
>>   create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-logb.h
>>   create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-lrint.h
>>   create mode 100644 sysdeps/loongarch/fpu/math-use-builtins-rint.h
>>   create mode 100644 sysdeps/loongarch/fpu/s_scalbn.c
>>   create mode 100644 sysdeps/loongarch/fpu/s_scalbnf.c
>>
  
Xi Ruoyao Nov. 23, 2022, 3:18 a.m. UTC | #4
On Wed, 2022-11-23 at 10:55 +0800, tangxiaolin wrote:
> Oops, I just resend {1,8,9}/9 with cover-patch, and still haven't solved 
> this problem.
> 
> I will resend this series again with v2 version, which is added ilogb 
> function.

Try "git send-email".  I had used to send long series manually with a
email client, but at last I found "git send-email" really makes our life
easier :).