[loongarch][DAG][FREEZE] Fix crash when FREEZE a half(f16) type on loongarch

[yxd-ym]

For zig with LLVM 19.1.0rc4, we are seeing the following error when bootstrapping a loongarch64-linux-musl target.

ziglang/zig-bootstrap#164 (comment)

It seems that this issue is caused by PromoteFloatResult is not handling FREEZE OP on loongarch.

Here is the reproduction of the error: https://godbolt.org/z/PPfvWjjG5

This patch adds the FREEZE OP handling with PromoteFloatRes_UnaryOp and adds a test case.

This patch changes loongarch's way of floating point promotion to soft promotion to avoid this problem.

See: loongarch's handling of half:

• loongarch64 half (f16) failure to select with +f and +d #93894
• [LoongArch] Allow f16 codegen with expansion to libcalls #94456

Also see: other float promotion FREEZE handling

• 0019c2f

[github-actions]

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

[llvmbot]

@llvm/pr-subscribers-llvm-selectiondag

@llvm/pr-subscribers-backend-loongarch

Author: YANG Xudong (yxd-ym)

Changes

For zig with LLVM 19.1.0rc4, we are seeing the following error when bootstrapping a loongarch64-linux-musl target.

ziglang/zig-bootstrap#164 (comment)

It seems that this issue is caused by PromoteFloatResult is not handling FREEZE OP on loongarch.

Here is the reproduction of the error: https://godbolt.org/z/PPfvWjjG5

This patch adds the FREEZE OP handling with PromoteFloatRes_UnaryOp and adds a test case.

See: loongarch's handling of half:

• loongarch64 half (f16) failure to select with +f and +d #93894
• [LoongArch] Allow f16 codegen with expansion to libcalls #94456

Also see: other float promotion FREEZE handling

• 0019c2f

---

Full diff: https://github.com/llvm/llvm-project/pull/107791.diff

2 Files Affected:

• (modified) llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp (+1)
• (modified) llvm/test/CodeGen/LoongArch/fp16-promote.ll (+15)

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
index b5c80005a0ecc1..51357dad5b30ec 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
@@ -2655,6 +2655,7 @@ void DAGTypeLegalizer::PromoteFloatResult(SDNode *N, unsigned ResNo) {
     case ISD::FLOG10:
     case ISD::FNEARBYINT:
     case ISD::FNEG:
+    case ISD::FREEZE:
     case ISD::FRINT:
     case ISD::FROUND:
     case ISD::FROUNDEVEN:
diff --git a/llvm/test/CodeGen/LoongArch/fp16-promote.ll b/llvm/test/CodeGen/LoongArch/fp16-promote.ll
index 75f920b43a06ce..a139428f2d61e9 100644
--- a/llvm/test/CodeGen/LoongArch/fp16-promote.ll
+++ b/llvm/test/CodeGen/LoongArch/fp16-promote.ll
@@ -324,3 +324,18 @@ define void @test_fmul_mem(ptr %p, ptr %q) nounwind {
   store half %r, ptr %p
   ret void
 }
+
+define half @freeze_half() nounwind {
+; LA32-LABEL: freeze_half:
+; LA32:       # %bb.0:
+; LA32-NEXT:    fadd.s	$fa0, $fa0, $fa0
+; LA32-NEXT:    ret
+;
+; LA64-LABEL: freeze_half:
+; LA64:       # %bb.0:
+; LA64-NEXT:    fadd.s	$fa0, $fa0, $fa0
+; LA64-NEXT:    ret
+  %y1 = freeze half undef
+  %t1 = fadd half %y1, %y1
+  ret half %t1
+}

[wangleiat]

Thanks.
Perhaps we should override this interface.

bool softPromoteHalfType() const override { return true; }

We currently do not have any parameter passing specifications for f16 types，this may also cause the generated code to not meet expectations.

[yxd-ym]

Thanks. Perhaps we should override this interface.

bool softPromoteHalfType() const override { return true; }

We currently do not have any parameter passing specifications for f16 types，this may also cause the generated code to not meet expectations.

@wangleiat Changed the fix according to your suggestion and updated the tests. Would you please check it?

[wangleiat]

LGTM

[yxd-ym]

cc @heiher @SixWeining

Also, is it possible to add this fix to release 19? Thanks!

[heiher]

cc @heiher @SixWeining

Also, is it possible to add this fix to release 19? Thanks!

Yes, this patch is also applicable to release 19.x. Thanks.

[SixWeining]

Should the commit message be updated?

[yxd-ym]

Should the commit message be updated?

Updated.

[github-actions]

@yxd-ym Congratulations on having your first Pull Request (PR) merged into the LLVM Project!

Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR.

Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues.

How to do this, and the rest of the post-merge process, is covered in detail here.

If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again.

If you don't get any reports, no action is required from you. Your changes are working as expected, well done!

[yxd-ym]

/cherry-pick 13280d9

[llvmbot]

/cherry-pick 13280d9

Error: Command failed due to missing milestone.

[nikic]

/cherry-pick 13280d9

[nikic]

Does this also fix #97975 on loongarch?

[llvmbot]

/pull-request #109093

[heiher]

Does this also fix #97975 on loongarch?

Yes. it also fix #97981