• Remove VIACopy changes and fix formating

LGTM

• Rename CSR_SRV4 to CSR_PPC64

Posted the NFC patch suggested as D77235

• Add newline to test

Are we forgetting the case of commons? They also have visibility as part of the assembler directive on AIX.

On AIX and PPC LE Linux after this change we are seeing invalid accesses when the backend asserts/fatal_errors. Looks like the driver and CC1 now share some global TimerGroup state that points to Timers which got created by CC1 but didn't get cleanup after the assert. Reported as: https://bugs.llvm.org/show_bug.cgi?id=45164

Rebase

• Update comment

• Update tests with labels that shouldn't be in the objectfile

LGTM

• Use external to determine if we should emit the symbol table entry
• Revert to not emitting linkage on the descriptor if it's a static
• Update test to refer to csect

Minor nit, but otherwise LGTM

LGTM

LGTM

• Add missing newline

In D74911#1885002, @sfertile wrote:

Would it make sense to move the logic that defines BEGIN_TWO_BYTE_PACK and END_TWO_BYTE_PACK into cmake now that its done in more then one place?

• Fix formating

• Update comment

• Update lang doc to indicate XCOFF doesn't support COMDAT

In D74631#1876995, @sfertile wrote:

1. We should probably update the COMDAT section of the lang ref to mention XCOFF doesn't support COMDATS.

• Add missing newline

LGTM

• Don't present the LLVM_BUILD_32_BITS option on AIX

This change regresses a whole bunch of lit tests when running on AIX, which change just the march and then stumble into the assert(MAI->needsFunctionDescriptors() && "AIX ABI is descriptor based.");, for example test/CodeGen/XCore/section-name.ll. Perhaps this should live in the PPC target printer instead?

Please add an option to control this output change.

• Don't set -q64 in the CMakeLists, we should use OBJECT_MODE instead on AIX if we are planning to build 64-bit so CMake knows what we are up to.
• Restore comment

LGTM (with note about later follow on)

LGTM with minor nit

• Update comment
• Change assertion into report fatal error
• Address comments

• Check for -Wl, form of passing -G

• Use IntNumBytes to get the length of fill

• Leave -G on for modules and brtl for executables that load them
• Use REGEX replace to filter flags
• Guard flags and make sure to add brtl if needed

Other than minor nit, LGTM

• Remove erroneously emiting fill length

• Update test to add byte splat
• Rename MAI query to doesZeroDirectiveSupportNonZeroValue()

• Remove include addition