Add helper functions to query properties of the lowered Target (#8192)

[steven-johnson]

No description provided.

[mcourteaux]

This one should perhaps get a big warning that this does not check runtime-features of the CPU or device, but uses the compile-time target. I know it says that in the doc above, but I feel like many people will get fooled by their wrong intuition. It's especially tricky, as it returns an Expr and not a bool, so it gives off a runtime-vibe.

[steven-johnson]

Done

[steven-johnson]

This could be expanded to more of the Target API, but I stopped here; some other possibilities would be

bool target_supports_device_api(DeviceAPI device)
bool target_supports_type(const Type &t, DeviceAPI device)

[abadams]

This would address my use-case

[mcourteaux]

@abadams Is there a way to get the device type of the enclosing loop for a given expression?

Context: I'm working on the fast_atan, and it seems that LLVM has a good version of atan already implemented for CUDA akin to what the fast-version would be. So, I'd like to just select the default atan instrinsic if the code will be run on a CUDA device. This would involve checking the device type of the enclosing loop. Turning that into a compile-time constant bool expression, would allow putting this into a select(), similar to this PR.

[steven-johnson]

Is there a way to get the device type of the enclosing loop for a given expression?

No, but it seems like it would be pretty simple to add, in the same manner of the helpers I just added. Feel free to open a PR to do so (it may be a little while before I can get to it).