[llvm-branch-commits] [llvm] [AMDGPU] Intrinsic for launching whole wave functions (PR #145859)

[Matt Arsenault via llvm-branch-commits]

================
@@ -7997,6 +7997,43 @@ void SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I,
     HasTailCall = true;
     return;
   }
+  case Intrinsic::amdgcn_call_whole_wave: {
+    TargetLowering::ArgListTy Args;
+
+    // The first argument is the callee. Skip it when assembling the call args.
+    TargetLowering::ArgListEntry Arg;
+    for (unsigned Idx = 1; Idx < I.arg_size(); ++Idx) {
+      Arg.Node = getValue(I.getArgOperand(Idx));
+      Arg.Ty = I.getArgOperand(Idx)->getType();
+      Arg.setAttributes(&I, Idx);
+      Args.push_back(Arg);
+    }
+
+    SDValue ConvControlToken;
+    if (auto Bundle = I.getOperandBundle(LLVMContext::OB_convergencectrl)) {
+      auto *Token = Bundle->Inputs[0].get();
+      ConvControlToken = getValue(Token);
+    }
+
+    TargetLowering::CallLoweringInfo CLI(DAG);
+    CLI.setDebugLoc(getCurSDLoc())
+        .setChain(getRoot())
+        .setCallee(CallingConv::AMDGPU_Gfx_WholeWave, I.getType(),
+                   getValue(I.getArgOperand(0)), std::move(Args))
+        .setTailCall(false)
+        .setIsPreallocated(
+            I.countOperandBundlesOfType(LLVMContext::OB_preallocated) != 0)
+        .setConvergent(I.isConvergent())
+        .setConvergenceControlToken(ConvControlToken);
+    CLI.CB = &I;
+
+    std::pair<SDValue, SDValue> Result =
+        lowerInvokable(CLI, /*EHPadBB*/ nullptr);
----------------
arsenm wrote:

```suggestion
        lowerInvokable(CLI, /*EHPadBB=*/ nullptr);
```

https://github.com/llvm/llvm-project/pull/145859