[CGP] Permit tail call optimization on undefined return value

[antoniofrighetto]

We should be able to freely allow tail call optimization on undefined values as well.

Fixes: #82387.

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

@llvm/pr-subscribers-backend-x86

Author: Antonio Frighetto (antoniofrighetto)

Changes

We should be able to freely allow tail call optimization on undefined values as well.

Fixes: #82387.

---

Full diff: https://github.com/llvm/llvm-project/pull/82419.diff

2 Files Affected:

• (modified) llvm/lib/CodeGen/CodeGenPrepare.cpp (+3-2)
• (modified) llvm/test/CodeGen/X86/tailcall-cgp-dup.ll (+22)

diff --git a/llvm/lib/CodeGen/CodeGenPrepare.cpp b/llvm/lib/CodeGen/CodeGenPrepare.cpp
index 4036f18dbc6794..feefe87f406365 100644
--- a/llvm/lib/CodeGen/CodeGenPrepare.cpp
+++ b/llvm/lib/CodeGen/CodeGenPrepare.cpp
@@ -2686,8 +2686,9 @@ bool CodeGenPrepare::dupRetToEnableTailCallOpts(BasicBlock *BB,
             attributesPermitTailCall(F, CI, RetI, *TLI)) {
           // Either we return void or the return value must be the first
           // argument of a known intrinsic or library function.
-          if (!V || (isIntrinsicOrLFToBeTailCalled(TLInfo, CI) &&
-                     V == CI->getArgOperand(0))) {
+          if (!V || isa<UndefValue>(V) ||
+              (isIntrinsicOrLFToBeTailCalled(TLInfo, CI) &&
+               V == CI->getArgOperand(0))) {
             TailCallBBs.push_back(Pred);
           }
         }
diff --git a/llvm/test/CodeGen/X86/tailcall-cgp-dup.ll b/llvm/test/CodeGen/X86/tailcall-cgp-dup.ll
index 401ed9f7bc5a9e..92811c87f5623f 100644
--- a/llvm/test/CodeGen/X86/tailcall-cgp-dup.ll
+++ b/llvm/test/CodeGen/X86/tailcall-cgp-dup.ll
@@ -362,8 +362,30 @@ return:
   ret ptr %src
 }
 
+@i = global i32 0, align 4
+
+define i32 @undef_tailc() nounwind {
+; CHECK-LABEL: undef_tailc:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    cmpl $0, _i(%rip)
+; CHECK-NEXT:    jne _qux ## TAILCALL
+; CHECK-NEXT:  ## %bb.1:
+; CHECK-NEXT:    retq
+  %1 = load i32, ptr @i, align 4
+  %2 = icmp eq i32 %1, 0
+  br i1 %2, label %5, label %3
+
+3:
+  %4 = tail call i32 @qux()
+  br label %5
+
+5:
+  ret i32 undef
+}
+
 declare void @llvm.memcpy.p0.p0.i64(ptr noalias nocapture writeonly, ptr noalias nocapture readonly, i64, i1)
 declare void @llvm.memset.p0.i64(ptr nocapture writeonly, i8, i64, i1)
 declare noalias ptr @malloc(i64)
 declare ptr @strcpy(ptr noalias returned writeonly, ptr noalias nocapture readonly)
 declare ptr @baz(ptr, ptr)
+declare i32 @qux()

[topperc]

Could we just leave V as nullptr back at line 2602?

[nikic]

I think the explicit UndefValue check is fine. Especially if we also want to extend the phi case above to handle undef incoming values later.

[antoniofrighetto]

Shall we ever hit the phi case? Looking at https://clang.godbolt.org/z/eq8vTxvzM, it seems that in the first case it suffices one non-undef incoming value to tail call, whereas the second case is refined by CGP as follows:

define dso_local i32 @foo() local_unnamed_addr #0 {
  %1 = load i32, ptr @i, align 4
  switch i32 %1, label %6 [
    i32 2, label %2
    i32 5, label %4
  ]

2:                                                ; preds = %0
  %3 = tail call i32 @bar() #2
  br label %6

4:                                                ; preds = %0
  %5 = tail call i32 @qux() #2
  ret i32 %5

6:                                                ; preds = %2, %0
  ret i32 undef
}

Thus getting tail called as part of this change.

[widlarizer]

Confirmed on RISC-V to fix the reproducer for the issue mentioned above, thanks a lot for the fast PR!

[nikic]

LGTM

[nikic]

I think the explicit UndefValue check is fine. Especially if we also want to extend the phi case above to handle undef incoming values later.

[widlarizer]

• TODO: fix tests

From CI: Failed Tests (3):
LLVM :: CodeGen/AArch64/addsub.ll
LLVM :: CodeGen/AArch64/callbr-asm-obj-file.ll
LLVM :: CodeGen/RISCV/pr51206.ll