Re: [PATCH] amdgcn: Add vectorized math routines

Hello

Sorry for the delay in replying. I've gone through the licenses for each 
file - apart from v64df_gamma.c ,v64sf_gamma.c, v64df_log10.c and 
v64sf_log10.c (which are all rather trivial), all the .c files already 
contain the copyright header from the file from which they were derived. 
A number of the original files simply had no license notice, so there 
was nothing to copy over.

No license:

v64_mathcnst.c, v64df_acos.c, v64df_asin.c, v64df_atan.c, v64df_atan2.c, 
v64df_cos.c, v64df_cosh.c, v64df_ispos.c, v64df_numtest.c, v64df_sin.c, 
v64df_sinh.c, v64sf_acos.c, v64sf_asin.c, v64sf_atan.c, v64sf_atan2.c, 
v64sf_cos.c, v64sf_cosh.c, v64sf_ispos.c, v64sf_numtest.c, v64sf_sin.c, 
v64sf_sinh.c

Sun Microsystems license:

v64df_acosh.c, v64df_asinh.c, v64df_atanh.c, v64df_copysign.c, 
v64df_erf.c, v64df_finite.c, v64df_fmod.c, v64df_gamma.c, v64df_hypot.c, 
v64df_ilogb.c, v64df_isnan.c, v64df_lgamma_r.c, v64df_lgamma.c, 
v64df_modf.c, v64df_pow.c, v64df_remainder.c, v64df_rint.c, 
v64df_scalb.c, v64df_scalbn.c, v64df_signif.c, v64df_tan.c, 
v64df_tgamma.c, v64sf_acosh.c, v64sf_asinh.c, v64sf_atanh.c, 
v64sf_copysign.c, v64sf_erf.c, v64sf_exp.c, v64sf_finite.c, 
v64sf_fmod.c, v64sf_gamma.c, v64sf_hypot.c, v64sf_ilogb.c, 
v64sf_isnan.c, v64sf_lgamma_r.c, v64sf_lgamma.c, v64sf_modf.c, 
v64sf_pow.c, v64sf_remainder.c, v64sf_rint.c, v64sf_scalb.c, 
v64sf_scalbn.c, v64sf_signif.c, v64sf_tan.c, v64sf_tgamma.c

Red Hat license:

v64df_exp2.c, v64sf_exp2.c

'Software Manual for the Elementary Functions' attribution:

v64df_asine.c, v64df_atangent.c, v64df_exp.c, v64df_log.c, 
v64df_log10.c, v64df_sine.c, v64df_sineh.c, v64df_sqrt.c, v64df_tanh.c, 
v64sf_asine.c, v64sf_atangent.c, v64sf_log.c, v64sf_log10.c, 
v64sf_sine.c, v64sf_sineh.c, v64sf_sqrt.c, v64sf_tanh.c

Original:

v64df_log2.c, v64sf_log2.c, v64_reent.c

The amdgcnmach.h file has portions from:

newlib/libm/common/fdlibm.h (Sun Microsystems license)
newlib/libm/mathfp/zmath.h (no license)
newlib/libm/common/math_errf.c (ARM license)
newlib/libm/common/math_config.h (ARM license)

so I have added the Sun and ARM copyright notices to the file. I have 
split off the vector macro library used by all the functions into a 
separate file amdgcn_veclib.h to make the authorship more clear, and 
have amdgcnmach.h include that.

I have added a copyright notice for Siemens (my parent company) at the 
top of each file, with terms similar to the usual MIT/BSD clauses.

Okay for inclusion in Newlib?

Thanks,

Kwok


> Hi Kwok,
> 
> You need to provide headers in all the source and include files.  Some
> files mention they are based on an existing
> newlib file.  Those files should copy the license header over and not just
> have a reference (imagine if
> someone were to delete the original file or rename it).  You are not
> allowed to change the licensing
> of a file you have based off of but you can add additional copyright info
> for your additions if you wish.
> 
> As mentioned you need to add a license header for the new amdgcnmatch.h
> header file that is used everywhere.
> 
> -- Jeff J.
> 
> On Fri, Dec 2, 2022 at 5:04 PM Kwok Cheung Yeung <[email protected]>
> wrote:
> 
>> Hello
>>
>> This patch implements a set of vectorized math routines for use by the
>> auto-vectorizer in GCC for the AMD GCN architecture. The implementation
>> is located in libm/machine/amdgcn, and is based on the current set of
>> scalar math routines in Newlib, from libm/common, libm/math and
>> libm/mathfp (where the level of accuracy was deemed sufficient for our
>> purposes). I have ensured that the copyright notices from the original
>> were retained.
>>
>> The vectorized routines make extensive use of the GCC vector extensions
>> for C, some GCN-specific builtins in GCC, and a SIMD abstraction library
>> located in libm/machine/amdgcn/amdgcnmach.h. As the builtins are a
>> recent addition to GCC, I have added #if statements to skip compilation
>> of the routines that use them to ensure that Newlib will still build for
>> older versions of GCC.
>>
>> Versions for 2-lane vectors up to 64-lane vectors (in powers of 2) are
>> provided for all public functions. Versions with <64 lanes (which is the
>> 'natural' vector width for AMD GCN) are shims that call the 64-lane
>> version with the unused lanes masked out.
>>
>> Okay for inclusion into Newlib?
>>
>> Thanks
>>
>> Kwok Yeung