FYI, if I add -mattr=+sve2 to your test arguments, I get:

An observation: nounwind only appears to affect output for 5 tests, in case there is an appetite to remove it elsewhere: insert_v2i64_nxv2i64_idx2 / insert_v4i32_nxv4i32_idx4 / insert_v16i8_nxv16i8_idx16 / insert_nxv8f16_nxv2f16 / insert_nxv4bf16_v4bf16.

For anyone wandering the internet in search of an answer to all their woes (probably future me, Hi future me!), this is the relevant error.

FAILED: obj/llvm/lib/Support/Support.Process.o
clang++ -MMD -MF obj/llvm/lib/Support/Support.Process.o.d -o obj/llvm/lib/Support/Support.Process.o -c ../../llvm/lib/Support/Process.cpp  -I../../llvm/lib/Support/Unix -I../../llvm/lib/Support/Windows -I../../llvm/include -Igen/llvm/include -O3 -fdiagnostics-color -Wall -Wextra -Wno-unused-parameter -Wdelete-non-virtual-dtor -Wstring-conversion -no-canonical-prefixes -Werror=date-time -fPIC -Wcovered-switch-default -std=c++14 -fvisibility-inlines-hidden -fno-exceptions -fno-rtti
In file included from ../../llvm/lib/Support/Process.cpp:107:
../../llvm/lib/Support/Unix/Process.inc:93:20: error: variable has incomplete type 'struct mallinfo2'
  struct mallinfo2 mi;
                   ^
../../llvm/lib/Support/Unix/Process.inc:93:10: note: forward declaration of 'mallinfo2'
  struct mallinfo2 mi;
         ^
../../llvm/lib/Support/Unix/Process.inc:94:10: error: no member named 'mallinfo2' in the global namespace
  mi = ::mallinfo2();
       ~~^
2 errors generated.

Thanks for the note, I appreciate it.

Forgot to press the abandon button, and changed my mind, I'll leave it in changes planned for a bit since there are no reviewers tagged yet.

I'm going to abandon this for now, though I may take another swipe at this later. I've identified while testing other types that this patch is wrong because it's not returning VectorRC when it should.

• Remove target-features test attribute per review and rebase.

On the effect of this patch: there are 10 translation units out of ~2,000 in the LNT benchmarks which have differing codegen as a consequence. I'll check for any performance effect next week before proposing this.

A fair point on the vector/fp overlap, will consider.

• Address Paul's suggestion: make getRegUsageForType return unsigned and simplify.

Looking pretty good. A couple of test cases to consider:

Looks good to me with minor nits.

LGTM with a naming nit, please leave others time to chime in before submitting as usual.

• Drop unneeded braces.
• Drop unneeded default argument.

In the previous patch I missed that the choice of CPU can override the default
value, and by chance this continued to work. Add a test and fix this.

• Make the flag absolute, per review comment (-aarch64-insert-extract-cost-adjustment => -aarch64-insert-extract-base-cost)

• Add tests:
  • Switch to utils/update_analyze_test_checks.py
  • Increasing
  • Decreasing
  • Underflow
  • FP

Accept with some nits in the tests which need fixing.

LGTM, looks like the formatter might have hit some lines you didn't intend to change -- if it's not too invasive perhaps worth a seperate NFC patch to format the file?

Seems reasonable to me.

In D121792#3388585, @dmgreen wrote:

Why do we have llvm.aarch64.sve.sel if we are always going to replace it with a select? Why not remove llvm.aarch64.sve.sel entirely?

LGTM, modulo that the require lines look like they need fixing.

Looking reasonable to me, one comment.

@aeubanks FYI this patch breaks test autogeneration. The reason is that changes the print header such that the regular expressions in update_analyze_test_checks.py no longer match.

LGTM.

Some comments.

Functionally I think it's looking reasonable to me. A few more stylistic nits.

LGTM with one more consistency tweak.

LGTM. I see the one additional instruction in the NOFP16 case which is an artefact of how the BSP node is defined, which is not changed in this patch. We've discussed it and attempted to remove it, but failed for now.

Getting there, thanks for the improvements. I've picked up on a few more things, but I think this is close to ready.

Thanks for the patch, I think this is looking reasonable to my knowledge. I've spotted the cause of the issues you've mentioned.

Looks like the build is broken due to PPC_LINUX_DEFAULT_IEEELONGDOUBLE, but reverting 08e535a50797, this differential fixes it.

Heads up, this did break my dev build and switching out libc is a nontrivial prospect for me for now. A configurable would be nice to have.

LGTM. Thanks for taking the time to fix this.

Looks reasonable to me. A suggestion inline.

LGTM with some editorial suggestions.

A couple of initial comments. It also looks like the code needs formatting.

LGTM with a nit.