is anyone using the alpha backend?

I'm removing old targets that no longer appear actively maintained,
to reduce the burden for target-independent codegen maintenance.

Does anyone object to the removal of the Alpha backend?

Dan

Hi Dan,

Do you mind if I ask you what is the target-independent
codegen maintenance?

I imagine that target-independent codegen interface have to
make with target-independent LLVM IR.

I'm also interested in target-independent properties on LLVM. :slight_smile:

Thanks,
Jin-Gu Kang

"Target-independent CodeGen" is jargon for code shared between the
various backends. Despite the name, this isn't directly related to
target-independent properties of LLVM IR.

Dan

Dan Gohman <gohman@apple.com> writes:

I'm removing old targets that no longer appear actively maintained,
to reduce the burden for target-independent codegen maintenance.

Does anyone object to the removal of the Alpha backend?

It would be a shame to lose it. Alpha is an excellent example of a good
RISC-style ISA. There's at least one simulator out there that uses it.

What is the burden you are trying to reduce here? I don't think we
should be throwing out targets for the sake of throwing them out.
Having multiple targets is a good way to ensure the target independent
stuff stays target independent.

                              -Dave

Dan Gohman <gohman@apple.com> writes:

I'm removing old targets that no longer appear actively maintained,
to reduce the burden for target-independent codegen maintenance.

Does anyone object to the removal of the Alpha backend?

It would be a shame to lose it. Alpha is an excellent example of a good
RISC-style ISA. There's at least one simulator out there that uses it.

In this case, I disagree. The reality is that the Alpha target is not in as good of shape as (e.g.) the Sparc backend, there really is almost no hardware, and there are much better simulators for mips, sparc, arm, etc.

What is the burden you are trying to reduce here? I don't think we
should be throwing out targets for the sake of throwing them out.
Having multiple targets is a good way to ensure the target independent
stuff stays target independent.

Having a target that isn't being used is just maintenance burden.

-Chris

Dan Gohman <gohman@apple.com> writes:

I'm removing old targets that no longer appear actively maintained,
to reduce the burden for target-independent codegen maintenance.

Does anyone object to the removal of the Alpha backend?

It would be a shame to lose it. Alpha is an excellent example of a good
RISC-style ISA. There's at least one simulator out there that uses it.

In this case, I disagree. The reality is that the Alpha target is not in as good of shape as (e.g.) the Sparc backend, there really is almost no hardware, and there are much better simulators for mips, sparc, arm, etc.

I only know a couple people using it, and those only with simulators.

The lack of reasonable speed real hardware is a major drawback to any
type of maintenance. (building takes almost 2 days on a 4 way smp).

Andrew

Sure that makes sense. Is there any reason to keep this target in mainline LLVM though, or is LLVM 3.0 and earlier good enough?

-Chris

Andrew Lenharth <andrewl <at> lenharth.org> writes:

I only know a couple people using it, and those only with simulators.

The lack of reasonable speed real hardware is a major drawback to any
type of maintenance. (building takes almost 2 days on a 4 way smp).

Would distcc not be of any help here?

Emil