AMDGPU: Create a dummy call sequence when emitting call error

[arsenm]

At least one special case call lowering tries to parse the call
sequence and asserts when it can't find a callseq_end.

[arsenm]

• DAG: Check if deoptimize call is available before emitting it #170940
• AMDGPU: Avoid crashing on statepoint-like pseudoinstructions #170657
• AMDGPU: Create a dummy call sequence when emitting call error #170656 👈 (View in Graphite)
• AMDGPU: Use correct chain when emitting error on a call #170645
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Matt Arsenault (arsenm)

Changes

At least one special case call lowering tries to parse the call
sequence and asserts when it can't find a callseq_end.

---

Full diff: https://github.com/llvm/llvm-project/pull/170656.diff

1 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp (+6-1)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index ba80c0695115c..128125d05a3c8 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -1412,7 +1412,12 @@ SDValue AMDGPUTargetLowering::lowerUnhandledCall(CallLoweringInfo &CLI,
       InVals.push_back(DAG.getPOISON(Arg.VT));
   }
 
-  return CLI.Chain;
+  // FIXME: Hack because R600 doesn't handle callseq pseudos yet.
+  if (getTargetMachine().getTargetTriple().getArch() == Triple::r600)
+    return CLI.Chain;
+
+  SDValue Chain = DAG.getCALLSEQ_START(CLI.Chain, 0, 0, CLI.DL);
+  return DAG.getCALLSEQ_END(Chain, 0, 0, /*InGlue=*/SDValue(), CLI.DL);
 }
 
 SDValue AMDGPUTargetLowering::LowerCall(CallLoweringInfo &CLI,

[ro-i]

would the call being a tail call matter? (probably not since this is "unhandled call" anyway)

[arsenm]

Tail calls would be opt-in by the lowering, and we don't try to handle musttail today