[llvm] [FastIsel] Get the right register type for call instruction (PR #164565)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-selectiondag

Author: Luo Yuanke (LuoYuanke)

<details>
<summary>Changes</summary>

When switch from fast isel to dag isel the input value is from llvm IR instruction.
If the instruction is call we should get the calling convention of the callee and
pass it to RegsForValue::getCopyFromRegs, so that it can deduce the right RegisterVT
of the returned value of the callee.


---
Full diff: https://github.com/llvm/llvm-project/pull/164565.diff


2 Files Affected:

- (modified) llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp (+6-1) 
- (added) llvm/test/CodeGen/X86/bf16-fast-isel.ll (+19) 


``````````diff
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index 20a0efd3afa1c..14f9fdfddf658 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -1977,8 +1977,13 @@ SDValue SelectionDAGBuilder::getValueImpl(const Value *V) {
   if (const Instruction *Inst = dyn_cast<Instruction>(V)) {
     Register InReg = FuncInfo.InitializeRegForValue(Inst);
 
+    std::optional<CallingConv::ID> CallConv = std::nullopt;
+    auto *Callee = dyn_cast<CallInst>(Inst);
+    if (Callee && !Callee->isInlineAsm())
+      CallConv = Callee->getCallingConv();
+
     RegsForValue RFV(*DAG.getContext(), TLI, DAG.getDataLayout(), InReg,
-                     Inst->getType(), std::nullopt);
+                     Inst->getType(), CallConv);
     SDValue Chain = DAG.getEntryNode();
     return RFV.getCopyFromRegs(DAG, FuncInfo, getCurSDLoc(), Chain, nullptr, V);
   }
diff --git a/llvm/test/CodeGen/X86/bf16-fast-isel.ll b/llvm/test/CodeGen/X86/bf16-fast-isel.ll
new file mode 100644
index 0000000000000..43622566d2974
--- /dev/null
+++ b/llvm/test/CodeGen/X86/bf16-fast-isel.ll
@@ -0,0 +1,19 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s
+
+define i8 @test(ptr %f) nounwind {
+; CHECK-LABEL: test:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    pushq %rax
+; CHECK-NEXT:    callq foo at PLT
+; CHECK-NEXT:    callq bar at PLT
+; CHECK-NEXT:    popq %rcx
+; CHECK-NEXT:    retq
+entry:
+  %call = call bfloat @foo(ptr %f)
+  %call2 = call noundef zeroext i8 @bar(bfloat %call)
+  ret i8 %call2
+}
+
+declare bfloat @foo(ptr %f)
+declare zeroext i8 @bar(bfloat)

``````````

</details>


https://github.com/llvm/llvm-project/pull/164565