[PATCH v3] MIPSr6/math: Use builtin fma and fmaf

[PATCH v3] MIPSr6/math: Use builtin fma and fmaf

[YunQiang Su]

From: YunQiang Su <syq@gcc.gnu.org>

MIPSr6 has MADDF.s/MADDF.d instructions, which are fused.

In MIPS ISA, double support can be subsetted.  Only FMAF is enabled
for this case.

	* sysdeps/mips/fpu/math-use-builtins-fma.h

Signed-off-by: YunQiang Su <syq@gcc.gnu.org>
---
 sysdeps/mips/fpu/math-use-builtins-fma.h | 13 +++++++++++++
 1 file changed, 13 insertions(+)
 create mode 100644 sysdeps/mips/fpu/math-use-builtins-fma.h

diff --git a/sysdeps/mips/fpu/math-use-builtins-fma.h b/sysdeps/mips/fpu/math-use-builtins-fma.h
new file mode 100644
index 0000000000..af8c1f95ff
--- /dev/null
+++ b/sysdeps/mips/fpu/math-use-builtins-fma.h
@@ -0,0 +1,13 @@
+#if __mips_isa_rev >= 6
+# ifdef __mips_single_float
+#  define USE_FMA_BUILTIN 0
+# else
+#  define USE_FMA_BUILTIN 1
+# endif
+# define USE_FMAF_BUILTIN 1
+#else
+# define USE_FMA_BUILTIN 0
+# define USE_FMAF_BUILTIN 0
+#endif
+#define USE_FMAL_BUILTIN 0
+#define USE_FMAF128_BUILTIN 0
-- 
2.39.3 (Apple Git-146)

Re: [PATCH v3] MIPSr6/math: Use builtin fma and fmaf

[Adhemerval Zanella Netto]

On 03/06/24 22:31, YunQiang Su wrote:
> From: YunQiang Su <syq@gcc.gnu.org>
> 
> MIPSr6 has MADDF.s/MADDF.d instructions, which are fused.
> 
> In MIPS ISA, double support can be subsetted.  Only FMAF is enabled
> for this case.
> 
> 	* sysdeps/mips/fpu/math-use-builtins-fma.h
> 
> Signed-off-by: YunQiang Su <syq@gcc.gnu.org>

LGTM, thanks.

Reviewed-by: Adhemerval Zanella  <adhemerval.zanella@linaro.org>

> ---
>  sysdeps/mips/fpu/math-use-builtins-fma.h | 13 +++++++++++++
>  1 file changed, 13 insertions(+)
>  create mode 100644 sysdeps/mips/fpu/math-use-builtins-fma.h
> 
> diff --git a/sysdeps/mips/fpu/math-use-builtins-fma.h b/sysdeps/mips/fpu/math-use-builtins-fma.h
> new file mode 100644
> index 0000000000..af8c1f95ff
> --- /dev/null
> +++ b/sysdeps/mips/fpu/math-use-builtins-fma.h
> @@ -0,0 +1,13 @@
> +#if __mips_isa_rev >= 6
> +# ifdef __mips_single_float
> +#  define USE_FMA_BUILTIN 0
> +# else
> +#  define USE_FMA_BUILTIN 1
> +# endif
> +# define USE_FMAF_BUILTIN 1
> +#else
> +# define USE_FMA_BUILTIN 0
> +# define USE_FMAF_BUILTIN 0
> +#endif
> +#define USE_FMAL_BUILTIN 0
> +#define USE_FMAF128_BUILTIN 0

Re: [PATCH v3] MIPSr6/math: Use builtin fma and fmaf

[Maciej W. Rozycki]

On Tue, 4 Jun 2024, Adhemerval Zanella Netto wrote:

> > MIPSr6 has MADDF.s/MADDF.d instructions, which are fused.
> > 
> > In MIPS ISA, double support can be subsetted.  Only FMAF is enabled
> > for this case.
> > 
> > 	* sysdeps/mips/fpu/math-use-builtins-fma.h
> > 
> > Signed-off-by: YunQiang Su <syq@gcc.gnu.org>
> 
> LGTM, thanks.
> 
> Reviewed-by: Adhemerval Zanella  <adhemerval.zanella@linaro.org>

 Doesn't this need a copyright notice?

  Maciej

Re: [PATCH v3] MIPSr6/math: Use builtin fma and fmaf

[YunQiang Su]

Maciej W. Rozycki <macro@orcam.me.uk> 于2024年6月5日周三 23:16写道：
>
>  Doesn't this need a copyright notice?
>

There is no copyright notice in any math-use-builtins-*.h.
I guess that we treat them as so straight forward, thus no copyright
notice is needed.

>   Maciej

Re: [PATCH v3] MIPSr6/math: Use builtin fma and fmaf

[Philippe Mathieu-Daudé]

On 4/6/24 03:31, YunQiang Su wrote:
> From: YunQiang Su <syq@gcc.gnu.org>
> 
> MIPSr6 has MADDF.s/MADDF.d instructions, which are fused.
> 
> In MIPS ISA, double support can be subsetted.  Only FMAF is enabled
> for this case.
> 
> 	* sysdeps/mips/fpu/math-use-builtins-fma.h
> 
> Signed-off-by: YunQiang Su <syq@gcc.gnu.org>
> ---
>   sysdeps/mips/fpu/math-use-builtins-fma.h | 13 +++++++++++++
>   1 file changed, 13 insertions(+)
>   create mode 100644 sysdeps/mips/fpu/math-use-builtins-fma.h

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>

Re: [PATCH v3] MIPSr6/math: Use builtin fma and fmaf

[Maciej W. Rozycki]

On Wed, 5 Jun 2024, YunQiang Su wrote:

> >  Doesn't this need a copyright notice?
> >
> 
> There is no copyright notice in any math-use-builtins-*.h.

 Existing files do not matter, it's the current interpretation of rules 
that does.  Correcting preexisting issues incurs effort someone has to 
undertake, so they may well stay around for a while; there are many of 
them.  New submissions need to be correct in all cases though.

> I guess that we treat them as so straight forward, thus no copyright
> notice is needed.

 And who decides that a piece of code is "straightforward"?  We currently 
have these rules written down:

--------------------------------------------------------------------------
4. Creating files

   * Don't create empty files
   * Find new versions of the copyright headers to use as a template.
   * Make sure the top line is descriptive.
   * "Contributed by" statements are no longer used.
--------------------------------------------------------------------------

I'd argue that iff a file consists solely of a #include_next directive, it 
might possibly skip the copyright header (though I think it would best be 
consulted by a lawyer anyway).  Otherwise if you argue that a piece of 
code is "straightforward" (i.e. legally insignificant, I suppose), then we 
need to draw a line somewhere.  We currently have it stated that:

"Copyright assignment for commulative [sic!] changes by any one author of 
less than 15 lines do not require copyright assignment."

so by extension you could argue that your submission only has 13 lines and 
therefore *by itself* is not legally significant.  Then if someone changes 
the file in the future so that it has at least 2 lines more, then it will 
be their responsibility to add the copyright notice header at that time.  
But it could be easy to miss.  And then should we remove the header again 
if the file has shrunk with a later update?

 I would argue that adding the the copyright notice right away is the 
correct approach in this case.

  Maciej

Re: [PATCH v3] MIPSr6/math: Use builtin fma and fmaf

[YunQiang Su]

Maciej W. Rozycki <macro@orcam.me.uk> 于2024年6月8日周六 22:17写道：
>
> On Wed, 5 Jun 2024, YunQiang Su wrote:
>
> > >  Doesn't this need a copyright notice?
> > >
> >
> > There is no copyright notice in any math-use-builtins-*.h.
>
>  Existing files do not matter, it's the current interpretation of rules
> that does.  Correcting preexisting issues incurs effort someone has to
> undertake, so they may well stay around for a while; there are many of
> them.  New submissions need to be correct in all cases though.
>
> > I guess that we treat them as so straight forward, thus no copyright
> > notice is needed.
>
>  And who decides that a piece of code is "straightforward"?  We currently
> have these rules written down:
>
> --------------------------------------------------------------------------
> 4. Creating files
>
>    * Don't create empty files
>    * Find new versions of the copyright headers to use as a template.
>    * Make sure the top line is descriptive.
>    * "Contributed by" statements are no longer used.
> --------------------------------------------------------------------------
>
> I'd argue that iff a file consists solely of a #include_next directive, it
> might possibly skip the copyright header (though I think it would best be
> consulted by a lawyer anyway).  Otherwise if you argue that a piece of
> code is "straightforward" (i.e. legally insignificant, I suppose), then we
> need to draw a line somewhere.  We currently have it stated that:
>
> "Copyright assignment for commulative [sic!] changes by any one author of
> less than 15 lines do not require copyright assignment."
>

I don't think that the line count is the most important: some code is
quite short
while its logic is quite complex.

For this patch, I don't think that it has logic at all. It is just a
statement about the
FMA support of MIPS ISA documents.
I mean I doubt whether it can be protected by copyright.

> so by extension you could argue that your submission only has 13 lines and
> therefore *by itself* is not legally significant.  Then if someone changes
> the file in the future so that it has at least 2 lines more, then it will
> be their responsibility to add the copyright notice header at that time.
> But it could be easy to miss.  And then should we remove the header again
> if the file has shrunk with a later update?
>

Personally, I don't think/wish a short and *non-logic* code can be protected
by copyright, as anybody who may do something for it will write the
almost same code.
I don't think that their work is a copy paste.

>  I would argue that adding the the copyright notice right away is the
> correct approach in this case.
>
>   Maciej

Re: [PATCH v3] MIPSr6/math: Use builtin fma and fmaf

[YunQiang Su]

Adhemerval Zanella Netto <adhemerval.zanella@linaro.org> 于2024年6月5日周三 02:00写道：
>
>
>
> On 03/06/24 22:31, YunQiang Su wrote:
> > From: YunQiang Su <syq@gcc.gnu.org>
> >
> > MIPSr6 has MADDF.s/MADDF.d instructions, which are fused.
> >
> > In MIPS ISA, double support can be subsetted.  Only FMAF is enabled
> > for this case.
> >
> >       * sysdeps/mips/fpu/math-use-builtins-fma.h
> >
> > Signed-off-by: YunQiang Su <syq@gcc.gnu.org>
>
> LGTM, thanks.
>
> Reviewed-by: Adhemerval Zanella  <adhemerval.zanella@linaro.org>
>

Can you help to commit it?

Re: [PATCH v3] MIPSr6/math: Use builtin fma and fmaf

[Adhemerval Zanella Netto]

On 09/06/24 03:43, YunQiang Su wrote:
> Maciej W. Rozycki <macro@orcam.me.uk> 于2024年6月8日周六 22:17写道：
>>
>> On Wed, 5 Jun 2024, YunQiang Su wrote:
>>
>>>>  Doesn't this need a copyright notice?
>>>>
>>>
>>> There is no copyright notice in any math-use-builtins-*.h.
>>
>>  Existing files do not matter, it's the current interpretation of rules
>> that does.  Correcting preexisting issues incurs effort someone has to
>> undertake, so they may well stay around for a while; there are many of
>> them.  New submissions need to be correct in all cases though.
>>
>>> I guess that we treat them as so straight forward, thus no copyright
>>> notice is needed.
>>
>>  And who decides that a piece of code is "straightforward"?  We currently
>> have these rules written down:
>>
>> --------------------------------------------------------------------------
>> 4. Creating files
>>
>>    * Don't create empty files
>>    * Find new versions of the copyright headers to use as a template.
>>    * Make sure the top line is descriptive.
>>    * "Contributed by" statements are no longer used.
>> --------------------------------------------------------------------------
>>
>> I'd argue that iff a file consists solely of a #include_next directive, it
>> might possibly skip the copyright header (though I think it would best be
>> consulted by a lawyer anyway).  Otherwise if you argue that a piece of
>> code is "straightforward" (i.e. legally insignificant, I suppose), then we
>> need to draw a line somewhere.  We currently have it stated that:
>>
>> "Copyright assignment for commulative [sic!] changes by any one author of
>> less than 15 lines do not require copyright assignment."
>>
> 
> I don't think that the line count is the most important: some code is
> quite short
> while its logic is quite complex.
> 
> For this patch, I don't think that it has logic at all. It is just a
> statement about the
> FMA support of MIPS ISA documents.
> I mean I doubt whether it can be protected by copyright.
> 
>> so by extension you could argue that your submission only has 13 lines and
>> therefore *by itself* is not legally significant.  Then if someone changes
>> the file in the future so that it has at least 2 lines more, then it will
>> be their responsibility to add the copyright notice header at that time.
>> But it could be easy to miss.  And then should we remove the header again
>> if the file has shrunk with a later update?
>>
> 
> Personally, I don't think/wish a short and *non-logic* code can be protected
> by copyright, as anybody who may do something for it will write the
> almost same code.
> I don't think that their work is a copy paste.

I agree that for this specific case, no math-use-builtins-*.h file have the
Copyright headers and require it for this patch does not seem productive.  Maybe
a better approach would to send a subsequent patch to add the missing headers.

Re: [PATCH v3] MIPSr6/math: Use builtin fma and fmaf

[Maciej W. Rozycki]

On Sun, 9 Jun 2024, YunQiang Su wrote:

> > I'd argue that iff a file consists solely of a #include_next directive, it
> > might possibly skip the copyright header (though I think it would best be
> > consulted by a lawyer anyway).  Otherwise if you argue that a piece of
> > code is "straightforward" (i.e. legally insignificant, I suppose), then we
> > need to draw a line somewhere.  We currently have it stated that:
> >
> > "Copyright assignment for commulative [sic!] changes by any one author of
> > less than 15 lines do not require copyright assignment."
> >
> 
> I don't think that the line count is the most important: some code is
> quite short
> while its logic is quite complex.

 It's the copyright law that matters, not your judgement, I'm afraid.

> For this patch, I don't think that it has logic at all. It is just a
> statement about the
> FMA support of MIPS ISA documents.
> I mean I doubt whether it can be protected by copyright.

 Maybe, maybe not, ultimately it would have to be taken to a court of law 
to know for sure.  But in any case transforming documentation into code is 
creative work and therefore in principle covered by copyright law.  And 
this piece could be written in different ways.

 Hence in my opinion it is not clear that it is trivial and in any case 
adding a copyright header makes no harm, but makes it clear to the reader 
what the intent of the copyright holder (in this case FSF) has been when 
it comes to the terms of use of this piece.

> > so by extension you could argue that your submission only has 13 lines and
> > therefore *by itself* is not legally significant.  Then if someone changes
> > the file in the future so that it has at least 2 lines more, then it will
> > be their responsibility to add the copyright notice header at that time.
> > But it could be easy to miss.  And then should we remove the header again
> > if the file has shrunk with a later update?
> >
> 
> Personally, I don't think/wish a short and *non-logic* code can be protected
> by copyright, as anybody who may do something for it will write the
> almost same code.
> I don't think that their work is a copy paste.

 If you assert that this piece of work is trivial, then they can well 
trivially recreate it from documentation rather than copying from our 
sources, so what's the problem with having a copyright notice?

  Maciej

[COMMITTED] MIPSr6/math: Use builtin fma and fmaf

[Andreas K. Hüttel]

From: YunQiang Su <syq@gcc.gnu.org>

MIPSr6 has MADDF.s/MADDF.d instructions, which are fused.

In MIPS ISA, double support can be subsetted.  Only FMAF is enabled
for this case.

	* sysdeps/mips/fpu/math-use-builtins-fma.h

Signed-off-by: YunQiang Su <syq@gcc.gnu.org>
Reviewed-by: Adhemerval Zanella  <adhemerval.zanella@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
---
 sysdeps/mips/fpu/math-use-builtins-fma.h | 13 +++++++++++++
 1 file changed, 13 insertions(+)
 create mode 100644 sysdeps/mips/fpu/math-use-builtins-fma.h

diff --git a/sysdeps/mips/fpu/math-use-builtins-fma.h b/sysdeps/mips/fpu/math-use-builtins-fma.h
new file mode 100644
index 0000000000..af8c1f95ff
--- /dev/null
+++ b/sysdeps/mips/fpu/math-use-builtins-fma.h
@@ -0,0 +1,13 @@
+#if __mips_isa_rev >= 6
+# ifdef __mips_single_float
+#  define USE_FMA_BUILTIN 0
+# else
+#  define USE_FMA_BUILTIN 1
+# endif
+# define USE_FMAF_BUILTIN 1
+#else
+# define USE_FMA_BUILTIN 0
+# define USE_FMAF_BUILTIN 0
+#endif
+#define USE_FMAL_BUILTIN 0
+#define USE_FMAF128_BUILTIN 0
-- 
2.44.2

Re: [COMMITTED] MIPSr6/math: Use builtin fma and fmaf

[Maciej W. Rozycki]

On Mon, 24 Jun 2024, Andreas K. Hüttel wrote:

> From: YunQiang Su <syq@gcc.gnu.org>
> 
> MIPSr6 has MADDF.s/MADDF.d instructions, which are fused.
> 
> In MIPS ISA, double support can be subsetted.  Only FMAF is enabled
> for this case.
> 
> 	* sysdeps/mips/fpu/math-use-builtins-fma.h
> 
> Signed-off-by: YunQiang Su <syq@gcc.gnu.org>
> Reviewed-by: Adhemerval Zanella  <adhemerval.zanella@linaro.org>
> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>

 It seems odd to me to commit an obsolete change reported to cause actual 
compilation errors as well as raising copyright notice concerns where 3 
subsequent revisions have been already posted that addressed these issues.

  Maciej

Re: [COMMITTED] MIPSr6/math: Use builtin fma and fmaf

[Andreas K. Huettel]

[-- Attachment #1: Type: text/plain, Size: 980 bytes --]

Oops. I'm sorry, I must have picked the wrong version by accident. On it...

On June 25, 2024 12:55:41 AM GMT+02:00, "Maciej W. Rozycki" <macro@orcam.me.uk> wrote:
>On Mon, 24 Jun 2024, Andreas K. Hüttel wrote:
>
>> From: YunQiang Su <syq@gcc.gnu.org>
>> 
>> MIPSr6 has MADDF.s/MADDF.d instructions, which are fused.
>> 
>> In MIPS ISA, double support can be subsetted.  Only FMAF is enabled
>> for this case.
>> 
>> 	* sysdeps/mips/fpu/math-use-builtins-fma.h
>> 
>> Signed-off-by: YunQiang Su <syq@gcc.gnu.org>
>> Reviewed-by: Adhemerval Zanella  <adhemerval.zanella@linaro.org>
>> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
>
> It seems odd to me to commit an obsolete change reported to cause actual 
>compilation errors as well as raising copyright notice concerns where 3 
>subsequent revisions have been already posted that addressed these issues.
>
>  Maciej

-- 
Sent from my Android device with K-9 Mail. Please excuse my brevity.

[REVERT] Revert "MIPSr6/math: Use builtin fma and fmaf"

[Andreas K. Hüttel]

Apologies, I mistakenly interpreted this to be already accepted.
Reverting until v6 or later is reviewed and approved.

This reverts commit 9e06e4a43b58519991acbed1d7f33abc40249226.
---
 sysdeps/mips/fpu/math-use-builtins-fma.h | 13 -------------
 1 file changed, 13 deletions(-)
 delete mode 100644 sysdeps/mips/fpu/math-use-builtins-fma.h

diff --git a/sysdeps/mips/fpu/math-use-builtins-fma.h b/sysdeps/mips/fpu/math-use-builtins-fma.h
deleted file mode 100644
index af8c1f95ff..0000000000
--- a/sysdeps/mips/fpu/math-use-builtins-fma.h
+++ /dev/null
@@ -1,13 +0,0 @@
-#if __mips_isa_rev >= 6
-# ifdef __mips_single_float
-#  define USE_FMA_BUILTIN 0
-# else
-#  define USE_FMA_BUILTIN 1
-# endif
-# define USE_FMAF_BUILTIN 1
-#else
-# define USE_FMA_BUILTIN 0
-# define USE_FMAF_BUILTIN 0
-#endif
-#define USE_FMAL_BUILTIN 0
-#define USE_FMAF128_BUILTIN 0
-- 
2.44.2