Question about ARM FASTFPE

Question about ARM FASTFPE

[Masahiro Yamada]

Hi.

I have a question about the following code
in arch/arm/Makefile:


# Do we have FASTFPE?
FASTFPE :=arch/arm/fastfpe
ifeq ($(FASTFPE),$(wildcard $(FASTFPE)))
FASTFPE_OBJ :=$(FASTFPE)/
endif


Since arch/arm/fastfpe does not exist in the upstream tree,
I guess this is a hook to compile downstream source code.

If a user puts arch/arm/fastfpe/ into their local source tree,
Kbuild is supposed to compile the files in it.

Is this correct?


If so, I am afraid this would not work for O= building.

$(wildcard ...) checks if this directory exists in the *objtree*,
while scripts/Makefile.build needs to include
arch/arm/fastfpe/Makefile from *srctree*.

I think the correct code should be like follows:

# Do we have FASTFPE?
FASTFPE :=arch/arm/fastfpe
ifneq ($(wildcard $(srctree)/$(FASTFPE)),)
FASTFPE_OBJ :=$(FASTFPE)/
endif


Having said that, I am not sure this code is worth fixing.

This code was added around v2.5.1.9,
and the actual source code for arch/arm/fastfpe/
was never upstreamed.


In general, we do not care much about the downstream code support.

What should we do about this?
Fix and keep maintaining? Delete?


-- 
Best Regards
Masahiro Yamada

Re: Question about ARM FASTFPE

[Russell King - ARM Linux admin]

On Wed, Jul 10, 2019 at 01:30:24PM +0900, Masahiro Yamada wrote:
> Hi.
> 
> I have a question about the following code
> in arch/arm/Makefile:
> 
> 
> # Do we have FASTFPE?
> FASTFPE :=arch/arm/fastfpe
> ifeq ($(FASTFPE),$(wildcard $(FASTFPE)))
> FASTFPE_OBJ :=$(FASTFPE)/
> endif
> 
> 
> Since arch/arm/fastfpe does not exist in the upstream tree,
> I guess this is a hook to compile downstream source code.
> 
> If a user puts arch/arm/fastfpe/ into their local source tree,
> Kbuild is supposed to compile the files in it.
> 
> Is this correct?
> 
> 
> If so, I am afraid this would not work for O= building.
> 
> $(wildcard ...) checks if this directory exists in the *objtree*,
> while scripts/Makefile.build needs to include
> arch/arm/fastfpe/Makefile from *srctree*.
> 
> I think the correct code should be like follows:
> 
> # Do we have FASTFPE?
> FASTFPE :=arch/arm/fastfpe
> ifneq ($(wildcard $(srctree)/$(FASTFPE)),)
> FASTFPE_OBJ :=$(FASTFPE)/
> endif
> 
> 
> Having said that, I am not sure this code is worth fixing.
> 
> This code was added around v2.5.1.9,

... as a _result_ of a discussion and deciding not to upstream it,
but to still allow its use.  Fastfpe is faster than nwfpe (so has
a definite advantage for FP intensive applications) but we decided
we didn't want two FP emulation codes in the kernel.  However, if
someone wants to use it, it has to be built into the kernel, it
can't be modular.

> and the actual source code for arch/arm/fastfpe/
> was never upstreamed.
> 
> 
> In general, we do not care much about the downstream code support.
> 
> What should we do about this?
> Fix and keep maintaining? Delete?
> 
> 
> -- 
> Best Regards
> Masahiro Yamada
> 

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up

Re: Question about ARM FASTFPE

[Masahiro Yamada]

On Wed, Jul 10, 2019 at 5:23 PM Russell King - ARM Linux admin
<linux@armlinux.org.uk> wrote:
>
> On Wed, Jul 10, 2019 at 01:30:24PM +0900, Masahiro Yamada wrote:
> > Hi.
> >
> > I have a question about the following code
> > in arch/arm/Makefile:
> >
> >
> > # Do we have FASTFPE?
> > FASTFPE :=arch/arm/fastfpe
> > ifeq ($(FASTFPE),$(wildcard $(FASTFPE)))
> > FASTFPE_OBJ :=$(FASTFPE)/
> > endif
> >
> >
> > Since arch/arm/fastfpe does not exist in the upstream tree,
> > I guess this is a hook to compile downstream source code.
> >
> > If a user puts arch/arm/fastfpe/ into their local source tree,
> > Kbuild is supposed to compile the files in it.
> >
> > Is this correct?
> >
> >
> > If so, I am afraid this would not work for O= building.
> >
> > $(wildcard ...) checks if this directory exists in the *objtree*,
> > while scripts/Makefile.build needs to include
> > arch/arm/fastfpe/Makefile from *srctree*.
> >
> > I think the correct code should be like follows:
> >
> > # Do we have FASTFPE?
> > FASTFPE :=arch/arm/fastfpe
> > ifneq ($(wildcard $(srctree)/$(FASTFPE)),)
> > FASTFPE_OBJ :=$(FASTFPE)/
> > endif
> >
> >
> > Having said that, I am not sure this code is worth fixing.
> >
> > This code was added around v2.5.1.9,
>
> ... as a _result_ of a discussion and deciding not to upstream it,
> but to still allow its use.  Fastfpe is faster than nwfpe (so has
> a definite advantage for FP intensive applications) but we decided
> we didn't want two FP emulation codes in the kernel.  However, if
> someone wants to use it, it has to be built into the kernel, it
> can't be modular.


IMHO, the entry in Makefile and Kconfig should be removed
from upstream, then moved to a part of the fastfpe local patch.



-- 
Best Regards
Masahiro Yamada

Re: Question about ARM FASTFPE

[Russell King - ARM Linux admin]

On Wed, Jul 10, 2019 at 06:54:06PM +0900, Masahiro Yamada wrote:
> On Wed, Jul 10, 2019 at 5:23 PM Russell King - ARM Linux admin
> <linux@armlinux.org.uk> wrote:
> >
> > On Wed, Jul 10, 2019 at 01:30:24PM +0900, Masahiro Yamada wrote:
> > > Hi.
> > >
> > > I have a question about the following code
> > > in arch/arm/Makefile:
> > >
> > >
> > > # Do we have FASTFPE?
> > > FASTFPE :=arch/arm/fastfpe
> > > ifeq ($(FASTFPE),$(wildcard $(FASTFPE)))
> > > FASTFPE_OBJ :=$(FASTFPE)/
> > > endif
> > >
> > >
> > > Since arch/arm/fastfpe does not exist in the upstream tree,
> > > I guess this is a hook to compile downstream source code.
> > >
> > > If a user puts arch/arm/fastfpe/ into their local source tree,
> > > Kbuild is supposed to compile the files in it.
> > >
> > > Is this correct?
> > >
> > >
> > > If so, I am afraid this would not work for O= building.
> > >
> > > $(wildcard ...) checks if this directory exists in the *objtree*,
> > > while scripts/Makefile.build needs to include
> > > arch/arm/fastfpe/Makefile from *srctree*.
> > >
> > > I think the correct code should be like follows:
> > >
> > > # Do we have FASTFPE?
> > > FASTFPE :=arch/arm/fastfpe
> > > ifneq ($(wildcard $(srctree)/$(FASTFPE)),)
> > > FASTFPE_OBJ :=$(FASTFPE)/
> > > endif
> > >
> > >
> > > Having said that, I am not sure this code is worth fixing.
> > >
> > > This code was added around v2.5.1.9,
> >
> > ... as a _result_ of a discussion and deciding not to upstream it,
> > but to still allow its use.  Fastfpe is faster than nwfpe (so has
> > a definite advantage for FP intensive applications) but we decided
> > we didn't want two FP emulation codes in the kernel.  However, if
> > someone wants to use it, it has to be built into the kernel, it
> > can't be modular.
> 
> 
> IMHO, the entry in Makefile and Kconfig should be removed
> from upstream, then moved to a part of the fastfpe local patch.

Nope.  It means that rather than it being merely a drop-in, it has
to be maintained against changes to both these files.  Sorry, that's
more work.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up

Re: Question about ARM FASTFPE

[Russell King - ARM Linux admin]

On Wed, Jul 10, 2019 at 11:02:06AM +0100, Russell King - ARM Linux admin wrote:
> On Wed, Jul 10, 2019 at 06:54:06PM +0900, Masahiro Yamada wrote:
> > On Wed, Jul 10, 2019 at 5:23 PM Russell King - ARM Linux admin
> > <linux@armlinux.org.uk> wrote:
> > > On Wed, Jul 10, 2019 at 01:30:24PM +0900, Masahiro Yamada wrote:
> > > > $(wildcard ...) checks if this directory exists in the *objtree*,
> > > > while scripts/Makefile.build needs to include
> > > > arch/arm/fastfpe/Makefile from *srctree*.
> > > >
> > > > I think the correct code should be like follows:
> > > >
> > > > # Do we have FASTFPE?
> > > > FASTFPE :=arch/arm/fastfpe
> > > > ifneq ($(wildcard $(srctree)/$(FASTFPE)),)
> > > > FASTFPE_OBJ :=$(FASTFPE)/
> > > > endif

That does indeed work for split object builds.
 
-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up

Re: Question about ARM FASTFPE

[Masahiro Yamada]

On Wed, Jul 10, 2019 at 7:02 PM Russell King - ARM Linux admin
<linux@armlinux.org.uk> wrote:
>
> On Wed, Jul 10, 2019 at 06:54:06PM +0900, Masahiro Yamada wrote:
> > On Wed, Jul 10, 2019 at 5:23 PM Russell King - ARM Linux admin
> > <linux@armlinux.org.uk> wrote:
> > >
> > > On Wed, Jul 10, 2019 at 01:30:24PM +0900, Masahiro Yamada wrote:
> > > > Hi.
> > > >
> > > > I have a question about the following code
> > > > in arch/arm/Makefile:
> > > >
> > > >
> > > > # Do we have FASTFPE?
> > > > FASTFPE :=arch/arm/fastfpe
> > > > ifeq ($(FASTFPE),$(wildcard $(FASTFPE)))
> > > > FASTFPE_OBJ :=$(FASTFPE)/
> > > > endif
> > > >
> > > >
> > > > Since arch/arm/fastfpe does not exist in the upstream tree,
> > > > I guess this is a hook to compile downstream source code.
> > > >
> > > > If a user puts arch/arm/fastfpe/ into their local source tree,
> > > > Kbuild is supposed to compile the files in it.
> > > >
> > > > Is this correct?
> > > >
> > > >
> > > > If so, I am afraid this would not work for O= building.
> > > >
> > > > $(wildcard ...) checks if this directory exists in the *objtree*,
> > > > while scripts/Makefile.build needs to include
> > > > arch/arm/fastfpe/Makefile from *srctree*.
> > > >
> > > > I think the correct code should be like follows:
> > > >
> > > > # Do we have FASTFPE?
> > > > FASTFPE :=arch/arm/fastfpe
> > > > ifneq ($(wildcard $(srctree)/$(FASTFPE)),)
> > > > FASTFPE_OBJ :=$(FASTFPE)/
> > > > endif
> > > >
> > > >
> > > > Having said that, I am not sure this code is worth fixing.
> > > >
> > > > This code was added around v2.5.1.9,
> > >
> > > ... as a _result_ of a discussion and deciding not to upstream it,
> > > but to still allow its use.  Fastfpe is faster than nwfpe (so has
> > > a definite advantage for FP intensive applications) but we decided
> > > we didn't want two FP emulation codes in the kernel.  However, if
> > > someone wants to use it, it has to be built into the kernel, it
> > > can't be modular.
> >
> >
> > IMHO, the entry in Makefile and Kconfig should be removed
> > from upstream, then moved to a part of the fastfpe local patch.
>
> Nope.  It means that rather than it being merely a drop-in, it has
> to be maintained against changes to both these files.  Sorry, that's
> more work.


This is the motivation of upstreaming for everybody.

We never know the code that does not exist in upstream.
Downstream code must pay maintenance cost for ever.


-- 
Best Regards
Masahiro Yamada

Re: Question about ARM FASTFPE

[Russell King - ARM Linux admin]

On Wed, Jul 10, 2019 at 07:43:19PM +0900, Masahiro Yamada wrote:
> On Wed, Jul 10, 2019 at 7:02 PM Russell King - ARM Linux admin
> <linux@armlinux.org.uk> wrote:
> >
> > On Wed, Jul 10, 2019 at 06:54:06PM +0900, Masahiro Yamada wrote:
> > > On Wed, Jul 10, 2019 at 5:23 PM Russell King - ARM Linux admin
> > > <linux@armlinux.org.uk> wrote:
> > > >
> > > > On Wed, Jul 10, 2019 at 01:30:24PM +0900, Masahiro Yamada wrote:
> > > > > Hi.
> > > > >
> > > > > I have a question about the following code
> > > > > in arch/arm/Makefile:
> > > > >
> > > > >
> > > > > # Do we have FASTFPE?
> > > > > FASTFPE :=arch/arm/fastfpe
> > > > > ifeq ($(FASTFPE),$(wildcard $(FASTFPE)))
> > > > > FASTFPE_OBJ :=$(FASTFPE)/
> > > > > endif
> > > > >
> > > > >
> > > > > Since arch/arm/fastfpe does not exist in the upstream tree,
> > > > > I guess this is a hook to compile downstream source code.
> > > > >
> > > > > If a user puts arch/arm/fastfpe/ into their local source tree,
> > > > > Kbuild is supposed to compile the files in it.
> > > > >
> > > > > Is this correct?
> > > > >
> > > > >
> > > > > If so, I am afraid this would not work for O= building.
> > > > >
> > > > > $(wildcard ...) checks if this directory exists in the *objtree*,
> > > > > while scripts/Makefile.build needs to include
> > > > > arch/arm/fastfpe/Makefile from *srctree*.
> > > > >
> > > > > I think the correct code should be like follows:
> > > > >
> > > > > # Do we have FASTFPE?
> > > > > FASTFPE :=arch/arm/fastfpe
> > > > > ifneq ($(wildcard $(srctree)/$(FASTFPE)),)
> > > > > FASTFPE_OBJ :=$(FASTFPE)/
> > > > > endif
> > > > >
> > > > >
> > > > > Having said that, I am not sure this code is worth fixing.
> > > > >
> > > > > This code was added around v2.5.1.9,
> > > >
> > > > ... as a _result_ of a discussion and deciding not to upstream it,
> > > > but to still allow its use.  Fastfpe is faster than nwfpe (so has
> > > > a definite advantage for FP intensive applications) but we decided
> > > > we didn't want two FP emulation codes in the kernel.  However, if
> > > > someone wants to use it, it has to be built into the kernel, it
> > > > can't be modular.
> > >
> > >
> > > IMHO, the entry in Makefile and Kconfig should be removed
> > > from upstream, then moved to a part of the fastfpe local patch.
> >
> > Nope.  It means that rather than it being merely a drop-in, it has
> > to be maintained against changes to both these files.  Sorry, that's
> > more work.
> 
> 
> This is the motivation of upstreaming for everybody.
> 
> We never know the code that does not exist in upstream.
> Downstream code must pay maintenance cost for ever.

I'm the maintainer of the files in question.  I also care-take
fastfpe.  It's my choice to make.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up