[v3,1/4] infra: add support for MIPS NaN

MIPS supports two different NaN encondings, legacy and 2008. Information
about MIPS NaN encodings can be found here:

  https://sourceware.org/binutils/docs/as/MIPS-NaN-Encodings.html

NaN legacy is the only option available for R2 cores and older.
NaN 2008 is the only option available for R6 cores.
R5 cores can have either NaN legacy or NaN 2008, depending on the
implementation. So, if the user selects a generic R5 target architecture
variant, we show a choice menu with both options available. For well
known R5 cores we directly select the NaN enconding they use.

Signed-off-by: Vicente Olivert Riera <Vincent.Riera@imgtec.com>
---
Changes v2 -> v3:
 - Nothing
Changes v1 -> v2:
 - Define config symbol in arch/Config.in
 - Change string "NAN" to "NaN"
---
 arch/Config.in                                     |  3 ++
 arch/Config.in.mips                                | 32 ++++++++++++++++++++++
 package/gcc/gcc.mk                                 |  7 +++++
 .../toolchain-external/pkg-toolchain-external.mk   |  5 ++++
 toolchain/toolchain-wrapper.c                      |  3 ++
 5 files changed, 50 insertions(+)

Hi Vicente,

 Pfew, more complexity... Patch looks good in general, but I think it shows that
we need some refactoring.

 I also wonder about external toolchains. I guess they'll either be multiarch,
or they support just one specific NaN encoding, right? In case of multiarch, all
is well. But an external toolchain built for one specific NaN encoding, will
that still work when you select the other encoding in Buildroot? Or will the
external toolchain infra detect that and bail out?


On 23-06-17 19:28, Vicente Olivert Riera wrote:
> MIPS supports two different NaN encondings, legacy and 2008. Information
                                  ^^^^^^^^^^encodings

> about MIPS NaN encodings can be found here:
> 
>   https://sourceware.org/binutils/docs/as/MIPS-NaN-Encodings.html
> 
> NaN legacy is the only option available for R2 cores and older.
> NaN 2008 is the only option available for R6 cores.
> R5 cores can have either NaN legacy or NaN 2008, depending on the
> implementation. So, if the user selects a generic R5 target architecture
> variant, we show a choice menu with both options available. For well
> known R5 cores we directly select the NaN enconding they use.
> 
> Signed-off-by: Vicente Olivert Riera <Vincent.Riera@imgtec.com>
> ---
> Changes v2 -> v3:
>  - Nothing
> Changes v1 -> v2:
>  - Define config symbol in arch/Config.in
>  - Change string "NAN" to "NaN"
> ---
>  arch/Config.in                                     |  3 ++
>  arch/Config.in.mips                                | 32 ++++++++++++++++++++++
>  package/gcc/gcc.mk                                 |  7 +++++
>  .../toolchain-external/pkg-toolchain-external.mk   |  5 ++++
>  toolchain/toolchain-wrapper.c                      |  3 ++
>  5 files changed, 50 insertions(+)
> 
> diff --git a/arch/Config.in b/arch/Config.in
> index 50377a9af..e921879d0 100644
> --- a/arch/Config.in
> +++ b/arch/Config.in
> @@ -264,6 +264,9 @@ config BR2_GCC_TARGET_ARCH
>  config BR2_GCC_TARGET_ABI
>  	string
>  
> +config BR2_GCC_TARGET_NAN
> +	string
> +

 It's a pity that we need separate Kconfig options for all of these
arch-specific tuning options. That's where some refactoring would be useful, so
that all the arch-specific stuff can be collected in an arch-specific makefile,
and gcc.mk and toolchain-external.mk etc. just use one variable that contains
everything. If you then need to add another option like that, you just need to
modify the arch-specific files.

>  config BR2_GCC_TARGET_CPU
>  	string
>  
> diff --git a/arch/Config.in.mips b/arch/Config.in.mips
> index 4e9ad12ad..a9c27a0e8 100644
> --- a/arch/Config.in.mips
> +++ b/arch/Config.in.mips
> @@ -51,6 +51,7 @@ config BR2_mips_m5150
>  	bool "M5150"
>  	depends on !BR2_ARCH_IS_64
>  	select BR2_MIPS_CPU_MIPS32R5
> +	select BR2_MIPS_NAN_2008
>  config BR2_mips_m6250
>  	bool "M6250"
>  	depends on !BR2_ARCH_IS_64
> @@ -59,6 +60,7 @@ config BR2_mips_p5600
>  	bool "P5600"
>  	depends on !BR2_ARCH_IS_64
>  	select BR2_MIPS_CPU_MIPS32R5
> +	select BR2_MIPS_NAN_2008
>  config BR2_mips_xburst
>  	bool "XBurst"
>  	depends on !BR2_ARCH_IS_64
> @@ -126,6 +128,36 @@ config BR2_MIPS_SOFT_FLOAT
>  	  floating point functions, then everything will need to be
>  	  compiled with soft floating point support (-msoft-float).
>  
> +config BR2_MIPS_NAN_LEGACY
> +	bool
> +	default y		if BR2_MIPS_CPU_MIPS32 || BR2_MIPS_CPU_MIPS32R2 || BR2_MIPS_CPU_MIPS64 || BR2_MIPS_CPU_MIPS64R2

 Don't put spaces between y and if, and do only one per line.

 However, mixing these default y with select from BR2_mips_m5150 etc. is not
good. Either you have a list of default y here, or you do select from
BR2_MIPS_CPU_MIPS32 etc. I have a slight preference for the latter.

> +
> +config BR2_MIPS_NAN_2008
> +	bool
> +	default y		if BR2_MIPS_CPU_MIPS32R6 || BR2_MIPS_CPU_MIPS64R6
> +
> +choice
> +	prompt "Target NaN"
> +	depends on BR2_mips_32r5 || BR2_mips_64r5
> +	default BR2_MIPS_ENABLE_NAN_2008
> +
> +	help
> +	  NaN encoding to be used
> +
> +config BR2_MIPS_ENABLE_NAN_LEGACY
> +	bool "legacy"
> +	select BR2_MIPS_NAN_LEGACY
> +
> +config BR2_MIPS_ENABLE_NAN_2008
> +	bool "2008"
> +	depends on !BR2_MIPS_SOFT_FLOAT
> +	select BR2_MIPS_NAN_2008
> +endchoice
> +
> +config BR2_GCC_TARGET_NAN
> +	default "legacy"	if BR2_MIPS_NAN_LEGACY
> +	default "2008"		if BR2_MIPS_NAN_2008
> +
>  config BR2_ARCH
>  	default "mips"		if BR2_mips
>  	default "mipsel"	if BR2_mipsel
> diff --git a/package/gcc/gcc.mk b/package/gcc/gcc.mk
> index b52f9456b..c0249cd50 100644
> --- a/package/gcc/gcc.mk
> +++ b/package/gcc/gcc.mk
> @@ -204,6 +204,9 @@ endif
>  ifneq ($(call qstrip,$(BR2_GCC_TARGET_ABI)),)
>  HOST_GCC_COMMON_CONF_OPTS += --with-abi=$(BR2_GCC_TARGET_ABI)
>  endif
> +ifneq ($(call qstrip,$(BR2_GCC_TARGET_NAN)),)
> +HOST_GCC_COMMON_CONF_OPTS += --with-nan=$(BR2_GCC_TARGET_NAN)

 You're just doing the same thing as in the rest of the file, but this is not
good: it should be qstripped. And behold, 50 lines below it is already
qstripped, as HOST_GCC_COMMON_WRAPPER_TARGET_NAN :-). Refactoring opportunity!
But of course, that needs to be a separate patch (and you don't need to do that
now if you don't feel up to it).


> +endif
>  ifneq ($(call qstrip,$(BR2_GCC_TARGET_CPU)),)
>  ifneq ($(call qstrip,$(BR2_GCC_TARGET_CPU_REVISION)),)
>  HOST_GCC_COMMON_CONF_OPTS += --with-cpu=$(call qstrip,$(BR2_GCC_TARGET_CPU)-$(BR2_GCC_TARGET_CPU_REVISION))
> @@ -254,6 +257,7 @@ HOST_GCC_COMMON_WRAPPER_TARGET_CPU := $(call qstrip,$(BR2_GCC_TARGET_CPU)-$(BR2_
>  endif
>  HOST_GCC_COMMON_WRAPPER_TARGET_ARCH := $(call qstrip,$(BR2_GCC_TARGET_ARCH))
>  HOST_GCC_COMMON_WRAPPER_TARGET_ABI := $(call qstrip,$(BR2_GCC_TARGET_ABI))
> +HOST_GCC_COMMON_WRAPPER_TARGET_NAN := $(call qstrip,$(BR2_GCC_TARGET_NAN))

 Damn this file is crap :-) Should be = instead of :=. But of course keeping
consistency is more important.

>  HOST_GCC_COMMON_WRAPPER_TARGET_FPU := $(call qstrip,$(BR2_GCC_TARGET_FPU))
>  HOST_GCC_COMMON_WRAPPER_TARGET_FLOAT_ABI := $(call qstrip,$(BR2_GCC_TARGET_FLOAT_ABI))
>  HOST_GCC_COMMON_WRAPPER_TARGET_MODE := $(call qstrip,$(BR2_GCC_TARGET_MODE))
> @@ -267,6 +271,9 @@ endif
>  ifneq ($(HOST_GCC_COMMON_WRAPPER_TARGET_ABI),)
>  HOST_GCC_COMMON_TOOLCHAIN_WRAPPER_ARGS += -DBR_ABI='"$(HOST_GCC_COMMON_WRAPPER_TARGET_ABI)"'
>  endif
> +ifneq ($(HOST_GCC_COMMON_WRAPPER_TARGET_NAN),)
> +HOST_GCC_COMMON_TOOLCHAIN_WRAPPER_ARGS += -DBR_NAN='"$(HOST_GCC_COMMON_WRAPPER_TARGET_NAN)"'
> +endif
>  ifneq ($(HOST_GCC_COMMON_WRAPPER_TARGET_FPU),)
>  HOST_GCC_COMMON_TOOLCHAIN_WRAPPER_ARGS += -DBR_FPU='"$(HOST_GCC_COMMON_WRAPPER_TARGET_FPU)"'
>  endif
> diff --git a/toolchain/toolchain-external/pkg-toolchain-external.mk b/toolchain/toolchain-external/pkg-toolchain-external.mk
> index 826934505..29c0aade1 100644
> --- a/toolchain/toolchain-external/pkg-toolchain-external.mk
> +++ b/toolchain/toolchain-external/pkg-toolchain-external.mk
> @@ -156,6 +156,7 @@ CC_TARGET_CPU_ := $(call qstrip,$(BR2_GCC_TARGET_CPU)-$(BR2_GCC_TARGET_CPU_REVIS
>  endif
>  CC_TARGET_ARCH_ := $(call qstrip,$(BR2_GCC_TARGET_ARCH))
>  CC_TARGET_ABI_ := $(call qstrip,$(BR2_GCC_TARGET_ABI))
> +CC_TARGET_NAN_ := $(call qstrip,$(BR2_GCC_TARGET_NAN))

 Again, this doesn't make sense -- this is exactly the same as
HOST_GCC_COMMON_WRAPPER_TARGET_NAN...

 So it smells like one simple refactoring opportunity would be to move this
stuff to e.g. toolchain.mk, or toolchain-wrapper.mk.

 But OK, all that refactoring, pretty nice but not needed for acceptance of this
patch.

 Regards,
 Arnout

>  CC_TARGET_FPU_ := $(call qstrip,$(BR2_GCC_TARGET_FPU))
>  CC_TARGET_FLOAT_ABI_ := $(call qstrip,$(BR2_GCC_TARGET_FLOAT_ABI))
>  CC_TARGET_MODE_ := $(call qstrip,$(BR2_GCC_TARGET_MODE))
> @@ -178,6 +179,10 @@ ifneq ($(CC_TARGET_ABI_),)
>  TOOLCHAIN_EXTERNAL_CFLAGS += -mabi=$(CC_TARGET_ABI_)
>  TOOLCHAIN_EXTERNAL_TOOLCHAIN_WRAPPER_ARGS += -DBR_ABI='"$(CC_TARGET_ABI_)"'
>  endif
> +ifneq ($(CC_TARGET_NAN_),)
> +TOOLCHAIN_EXTERNAL_CFLAGS += -mnan=$(CC_TARGET_NAN_)
> +TOOLCHAIN_EXTERNAL_TOOLCHAIN_WRAPPER_ARGS += -DBR_NAN='"$(CC_TARGET_NAN_)"'
> +endif
>  ifneq ($(CC_TARGET_FPU_),)
>  TOOLCHAIN_EXTERNAL_CFLAGS += -mfpu=$(CC_TARGET_FPU_)
>  TOOLCHAIN_EXTERNAL_TOOLCHAIN_WRAPPER_ARGS += -DBR_FPU='"$(CC_TARGET_FPU_)"'
> diff --git a/toolchain/toolchain-wrapper.c b/toolchain/toolchain-wrapper.c
> index 100aa181c..28066e425 100644
> --- a/toolchain/toolchain-wrapper.c
> +++ b/toolchain/toolchain-wrapper.c
> @@ -51,6 +51,9 @@ static char *predef_args[] = {
>  #ifdef BR_ABI
>  	"-mabi=" BR_ABI,
>  #endif
> +#ifdef BR_NAN
> +	"-mnan=" BR_NAN,
> +#endif
>  #ifdef BR_FPU
>  	"-mfpu=" BR_FPU,
>  #endif
>

On 23-06-17 23:56, Arnout Vandecappelle wrote:
>> +config BR2_MIPS_NAN_LEGACY
>> +	bool
>> +	default y		if BR2_MIPS_CPU_MIPS32 || BR2_MIPS_CPU_MIPS32R2 || BR2_MIPS_CPU_MIPS64 || BR2_MIPS_CPU_MIPS64R2
>  Don't put spaces between y and if, and do only one per line.
> 
>  However, mixing these default y with select from BR2_mips_m5150 etc. is not
> good. Either you have a list of default y here, or you do select from
> BR2_MIPS_CPU_MIPS32 etc. I have a slight preference for the latter.

 Just to be clear: this is the only thing that needs to be changed in this
patch, then it's OK for me.

 Oh, and also what happens with external toolchains configured for the wrong NaN.

 Regards,
 Arnout

Hello Arnout,

first of all, thanks for the review.

On 24/06/17 09:51, Arnout Vandecappelle wrote:
> 
> 
> On 23-06-17 23:56, Arnout Vandecappelle wrote:
>>> +config BR2_MIPS_NAN_LEGACY
>>> +	bool
>>> +	default y		if BR2_MIPS_CPU_MIPS32 || BR2_MIPS_CPU_MIPS32R2 || BR2_MIPS_CPU_MIPS64 || BR2_MIPS_CPU_MIPS64R2
>>  Don't put spaces between y and if, and do only one per line.
>>
>>  However, mixing these default y with select from BR2_mips_m5150 etc. is not
>> good. Either you have a list of default y here, or you do select from
>> BR2_MIPS_CPU_MIPS32 etc. I have a slight preference for the latter.
> 
>  Just to be clear: this is the only thing that needs to be changed in this
> patch, then it's OK for me.

I'll change that. Thanks.

There is also one more important change to do for fixing uClibc
compilation. I'll include it as well.

> 
>  Oh, and also what happens with external toolchains configured for the wrong NaN.

Some external toolchains can build code for both NaN. For instance, the
ones that are included in Buildroot...

  toolchain-external-codescape-img-mips
  toolchain-external-codescape-mti-mips
  toolchain-external-codesourcery-mips

...they do support both NaN. In that case, nothing has to be done.

But, if your external toolchain only supports one NaN, then you need to
make that external toolchain depend on either BR2_MIPS_NAN_LEGACY or
BR2_MIPS_NAN_2008.

Regards,

Vincent

> 
>  Regards,
>  Arnout
>