[GISel] Add generic implementation for @llvm.expect.with.probability when optimizations are disabled

[antangelo]

Handle @llvm.expect.with.probability in GlobalISel in the same way @llvm.expect is handled, passing the value through as-is. This can be encountered if the intrinsic is used without optimizations, which would otherwise transform it out.

Fixes #115411 for GlobalISel

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

@llvm/pr-subscribers-llvm-globalisel

Author: None (antangelo)

Changes

Handle @llvm.expect.with.probability in GlobalISel in the same way @llvm.expect is handled, passing the value through as-is. This can be encountered if the intrinsic is used without optimizations, which would otherwise transform it out.

Fixes #115411 for GlobalISel

---

Full diff: https://github.com/llvm/llvm-project/pull/117835.diff

2 Files Affected:

• (modified) llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp (+1)
• (modified) llvm/test/CodeGen/Generic/builtin-expect-with-probability.ll (+1)

diff --git a/llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp b/llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp
index 675f55d8086bc3..f668e41094bbc8 100644
--- a/llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp
@@ -2431,6 +2431,7 @@ bool IRTranslator::translateKnownIntrinsic(const CallInst &CI, Intrinsic::ID ID,
   case Intrinsic::invariant_end:
     return true;
   case Intrinsic::expect:
+  case Intrinsic::expect_with_probability:
   case Intrinsic::annotation:
   case Intrinsic::ptr_annotation:
   case Intrinsic::launder_invariant_group:
diff --git a/llvm/test/CodeGen/Generic/builtin-expect-with-probability.ll b/llvm/test/CodeGen/Generic/builtin-expect-with-probability.ll
index aef134b636d5a7..76e9e5e81aae0f 100644
--- a/llvm/test/CodeGen/Generic/builtin-expect-with-probability.ll
+++ b/llvm/test/CodeGen/Generic/builtin-expect-with-probability.ll
@@ -1,4 +1,5 @@
 ; RUN: llc < %s
+; RUN: llc -global-isel < %s
 
 declare i32 @llvm.expect.with.probability(i32, i32, double)
 

[arsenm]

The existing test is woefully inadequate. Should fix this in a pre-commit. This needs to have checks, and should not be a "Generic" test. Those cannot exist

[antangelo]

For generic changes like this, is it sufficient to test against one backend or does each one need its own coverage?

[antangelo]

I have moved testing for GISel to the AArch64 IRTranslator no-op intrinsics suite, where similar tests lie, and removed the update to the generic one. I'll update the SelectionDAG tests to be specific and remove the generic one altogether separately.