build: add --v8-target-arch and cross compiling options

[xbolshe]

Checklist

• commit message follows commit guidelines

Affected core subsystem(s)

build

Description of change

• Add --v8-target-arch option to apply a correct v8_target_arch like x87 in config.gypi while --dest-cpu=ia32.
• Add --host-cc, --host-cxx, --host-ar options to set environment variables required for a cross compiling process.

Signed-off-by: Eugene Bolshakov pub@relvarsoft.com

[bnoordhuis]

I don't think --v8-target-arch would work. Your x87 example results in a binary where V8 emits x87 code but everything else still uses SSE. Is there a plausible scenario for target_arch != v8_target_arch?

[xbolshe]

Let me provide some experience about --v8-target-arch:

1. My experience to build my own Linux images with NodeJS for Intel Galileo Gen1&Gen2 boards:

There is no SSE support, my NodeJS works on Quark CPU:

2. I found this presentation too late (after digging into V8 code and finding a solution):

http://ark.intel.com/products/79084/Intel-Quark-SoC-X1000-16K-Cache-400-MHz

http://events.linuxfoundation.org/sites/events/files/slides/V8%20engine%20of%20Node.js%20on%20IA%20JavaScript-JITTED%20x86%20machine%20code%20mapping%20profiling%20support%20and%20X87%20Quark%20processor%20enabling.pdf

x87 support in V8 belongs to Intel.
On a slide below you may find that target_arch != v8_target_arch.

3. Users' experience when x87 is not selected and "Illegal instruction" is present:

https://communities.intel.com/thread/52059

https://forums.gentoo.org/viewtopic-p-7804792.html?sid=dee8f41663e159818a8f913e37b757bf#7804792

https://communities.intel.com/thread/80904

4. About "everything else still uses SSE":
   I can't agree with you because GCC and all code is done without SSE support using -march=i586 option:

https://gcc.gnu.org/onlinedocs/gcc/x86-Options.html

‘i586’
‘pentium’
Intel Pentium CPU with no MMX support.

/сс @cdai2

BR,
xbolshe

[bnoordhuis]

4. About "everything else still uses SSE": I can't agree with you because GCC and all code is done without SSE support using -march=i586 option:

I know that V8 is compiled with -march=i586 when v8_target_arch=x87, but node is not - that is what I mean with "everything else".

If you get a working binary, that probably means your toolchain is configured for i586, i.e., it defaults to -march=i586. But someone with, say, an i686 toolchain would end up with a malfunctioning binary.

[xbolshe]

Sure. And a malfunctioning binary may be done with another wrong compiler options (like big endian instead of little endian).

[jasnell]

Is this still needed?

[bnoordhuis]

I had another look and I still think this isn't the right approach. For one, this PR lets you build a node binary that reports process.arch is ia32 while it's really x87 (assuming the binary actually works, see my #7787 (comment) here.)

The --host-* switches might be acceptable, although I wouldn't want to vouch they are sufficient in all cases.

[xbolshe]

@jasnell Yep, it is needed.
@bnoordhuis It is not just an arch like i586. x87 option enables a part of V8 code like https://github.com/v8/v8/tree/master/src/x87

[fhinkel]

There hasn't been any activity here. I'm closing this. Feel free to reopen (or ping a collaborator) if I closed this in error.