[llvm-dev] Testing "normal" cross-compilers versus GPU backends

[Krzysztof Parzyszek via llvm-dev]

On 9/3/2015 7:56 PM, Robinson, Paul via llvm-dev wrote:
> Right, intuitively it doesn't make sense.  Is it actually useful to build a
> GPU compiler that will crash unless you ask it to generate GPU code? Seems
> to me it should default to producing GPU code.

If the lack of the default triple was to become a feature, the compiler 
could simply request specific arch/cpu settings and exit gracefully. 
The default triple could be set to something that explicitly states the 
absence of a particular default target (e.g. triple=none, or something 
like that).  This may not be useful for popular targets (that are used 
as hosts), but it might be for embedded targets that come in a variety 
of different flavors.  The default choice may be a lot less obvious in 
those cases.  As an added bonus, if users have to explicitly specify 
target details, they would be immune to potential changes of the 
defaults if they were to upgrade the compiler.

-Krzysztof

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, 
hosted by The Linux Foundation