[AArch64][SME] Add remarks to flag lazy ZA saves, and SMSTART/SMSTOP transitions

[jroelofs]

No description provided.

[github-actions]

⚠️ C/C++ code formatter, clang-format found issues in your code. ⚠️

You can test this locally with the following command:

git-clang-format --diff 5082e827c1670f5c98d320d08979b8855253d0cc 2ed87db9fcfdb2c60bb1bf5ff95db3bf3f4943ad -- llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

View the diff from clang-format here.

diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 6c5a121d5c0f..db0103813411 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -7404,10 +7404,10 @@ AArch64TargetLowering::LowerCall(CallLoweringInfo &CLI,
         TPIDR2ObjAddr);
     OptimizationRemarkEmitter ORE(&MF.getFunction());
     ORE.emit([&]() {
-      auto R = CLI.CB ? OptimizationRemarkAnalysis("sme", "SMELazySaveZA",
-                                                   CLI.CB)
-                      : OptimizationRemarkAnalysis("sme", "SMELazySaveZA",
-                                                   &MF.getFunction());
+      auto R = CLI.CB
+                   ? OptimizationRemarkAnalysis("sme", "SMELazySaveZA", CLI.CB)
+                   : OptimizationRemarkAnalysis("sme", "SMELazySaveZA",
+                                                &MF.getFunction());
       DescribeCallsite(R) << " sets up a lazy save for ZA";
       if (CalleeAttrs.preservesZA())
         R << ", but callee preserves ZA, so we request 0 slices to be saved";
@@ -7426,10 +7426,10 @@ AArch64TargetLowering::LowerCall(CallLoweringInfo &CLI,
     PStateSM = getPStateSM(DAG, Chain, CallerAttrs, DL, MVT::i64);
     OptimizationRemarkEmitter ORE(&MF.getFunction());
     ORE.emit([&]() {
-      auto R = CLI.CB ? OptimizationRemarkAnalysis("sme", "SMETransition",
-                                                   CLI.CB)
-                      : OptimizationRemarkAnalysis("sme", "SMETransition",
-                                                   &MF.getFunction());
+      auto R = CLI.CB
+                   ? OptimizationRemarkAnalysis("sme", "SMETransition", CLI.CB)
+                   : OptimizationRemarkAnalysis("sme", "SMETransition",
+                                                &MF.getFunction());
       DescribeCallsite(R) << " requires a streaming mode transition";
       return R;
     });

[sdesmalen-arm]

Thanks for adding this, this is very useful.

[sdesmalen-arm]

nit: Would it be easier to have a function that takes just a StringRef, so that you can do EmitSMERemarkForCall("requires a streaming mode transition");? You can then hide the construction of the ORE and OptimizationRemarkAnalysis in this lambda.

[jroelofs]

I was thinking the current format would be a bit nicer for additions, i.e. if/when ZA liveness starts being tracked, the remark could be amended to mention the number of za slices that should be preserved.

[sdesmalen-arm]

I'm a bit weary of writing code in anticipation of future changes which may never happen. We can always rewrite the code in this form later if that makes more sense. That said, I don't feel too strongly about it since the current code isn't bad either.

[sdesmalen-arm]

I'm a bit weary of writing code in anticipation of future changes which may never happen. We can always rewrite the code in this form later if that makes more sense. That said, I don't feel too strongly about it since the current code isn't bad either.