Dup of D59610 ?

LGTM

Added docs and tests. Addressed review feedback.

This still requires some more tests to be added (which I am working on), and docs updated. I just wanted to get this out here for some early code reviews and testing on hw.

Address review feedback.

Backend change will be posted later today. I am currently working on adding backend tests.

LGTM

Address review feedback.

LGTM

LGTM

Can SIProgramInfo::DebuggerWavefrontPrivateSegmentOffsetSGPR and SIProgramInfo::DebuggerPrivateSegmentBufferSGPR be completely deleted?

https://github.com/llvm-mirror/llvm/blob/master/lib/Target/AMDGPU/SIProgramInfo.h#L55
https://github.com/llvm-mirror/llvm/blob/master/lib/Target/AMDGPU/SIProgramInfo.h#L61

I am in favor of removing these. Clang will also need to be updated around here:
https://github.com/llvm-mirror/clang/blob/master/lib/Driver/ToolChains/AMDGPU.cpp#L41

In D57811#1395643, @ruiu wrote:

Just want to know a bit more about background of this change. Are you making an ABI-breaking change, so you are defining a new ABI version number for the AMDGPU target. Is this understanding correct? Is the AMDGPU target migrating to ABI version 2, or will version 1 and 2 co-exist in the future?

ping

ping

Address review feedback.

LGTM. Please make sure to update your commit message as your patch processes all m options now.

Is it still code object v2 by default?

Can you add some tests? Preferably with -mxnack/-mno-xnack, -msram-ecc/-mno-sram-ecc, -mcode-object-v3/-mno-code-object-v3

In D58026#1393837, @ruiu wrote:

The interpretation of ABI version of an ELF object file depends on the target machine type. For example, we can think of a target that allows a mix of version X and version Y objects let a linker to create a shim between cross-version function call. At least most ABIs don't define anything about the ABI version, so it is perhaps most natural to ignore the field at all, unless you are linking for a specific target in which the ABI version has a special meaning. Therefore, I'd think code that works only for AMDGPU target is better.

In D58026#1393777, @ruiu wrote:

I quickly hacked up this patch: https://gist.github.com/rui314/f3cd508b6cbbde763fc13bc6190d7505 Does this look good?

Make sure ABI versions across all object files are the same.

https://github.com/llvm-mirror/clang/blob/master/lib/Driver/ToolChains/AMDGPU.cpp#L52

Can you handle all features as follows instead of checking a particular option?

LGTM

LGTM

LGTM

LGTM

LGTM

This review fell through the cracks and never got published.

LGTM

Will fix during committing. Thanks.

LGTM

LGTM

LGTM

LGTM

LGTM

In D54545#1299055, @arsenm wrote:

Is this the one that zeros the high bits vs. leaves them intact?

LGTM

LGTM