[Coroutines 2/2] Improve symmetric control transfer feature

Differential Revision: https://reviews.llvm.org/D76913

llvm/lib/Transforms/Coroutines/CoroSplit.cpp

@@ -934,7 +934,8 @@ static bool simplifyTerminatorLeadingToRet(Instruction *InitialInst) {
   BasicBlock *UnconditionalSucc = nullptr;
 
   Instruction *I = InitialInst;
-  while (I->isTerminator()) {
+  while (I->isTerminator() ||
+         (isa<CmpInst>(I) && I->getNextNode()->isTerminator())) {
     if (isa<ReturnInst>(I)) {
       if (I != InitialInst) {
         // If InitialInst is an unconditional branch,
@@ -958,6 +959,29 @@ static bool simplifyTerminatorLeadingToRet(Instruction *InitialInst) {
         I = BB->getFirstNonPHIOrDbgOrLifetime();
         continue;
       }
+    } else if (auto *CondCmp = dyn_cast<CmpInst>(I)) {
+      auto *BR = dyn_cast<BranchInst>(I->getNextNode());
+      if (BR && BR->isConditional() && CondCmp == BR->getCondition()) {
+        // If the case number of suspended switch instruction is reduced to
+        // 1, then it is simplified to CmpInst in llvm::ConstantFoldTerminator.
+        // And the comparsion looks like : %cond = icmp eq i8 %V, constant.
+        ConstantInt *CondConst = dyn_cast<ConstantInt>(CondCmp->getOperand(1));
+        if (CondConst && CondCmp->getPredicate() == CmpInst::ICMP_EQ) {
+          Value *V = CondCmp->getOperand(0);
+          auto it = ResolvedValues.find(V);
+          if (it != ResolvedValues.end())
+            V = it->second;
+
+          if (ConstantInt *Cond0 = dyn_cast<ConstantInt>(V)) {
+            BasicBlock *BB = Cond0->equalsInt(CondConst->getZExtValue())
+                                 ? BR->getSuccessor(0)
+                                 : BR->getSuccessor(1);
+            scanPHIsAndUpdateValueMap(I, BB, ResolvedValues);
+            I = BB->getFirstNonPHIOrDbgOrLifetime();
+            continue;
+          }
+        }
+      }
     } else if (auto *SI = dyn_cast<SwitchInst>(I)) {
       Value *V = SI->getCondition();
       auto it = ResolvedValues.find(V);

llvm/test/Transforms/Coroutines/coro-split-musttail3.ll

@@ -0,0 +1,101 @@
+; Tests that coro-split will convert coro.resume followed by a suspend to a
+; musttail call.
+; RUN: opt < %s -coro-split -S | FileCheck %s
+; RUN: opt < %s -passes=coro-split -S | FileCheck %s
+
+define void @f() #0 {
+entry:
+  %id = call token @llvm.coro.id(i32 0, i8* null, i8* null, i8* null)
+  %alloc = call i8* @malloc(i64 16) #3
+  %vFrame = call noalias nonnull i8* @llvm.coro.begin(token %id, i8* %alloc)
+
+  %save = call token @llvm.coro.save(i8* null)
+  %addr1 = call i8* @llvm.coro.subfn.addr(i8* null, i8 0)
+  %pv1 = bitcast i8* %addr1 to void (i8*)*
+  call fastcc void %pv1(i8* null)
+
+  %suspend = call i8 @llvm.coro.suspend(token %save, i1 false)
+  %cmp = icmp eq i8 %suspend, 0
+  br i1 %cmp, label %await.suspend, label %exit
+await.suspend:
+  %save2 = call token @llvm.coro.save(i8* null)
+  %br0 = call i8 @switch_result()
+  switch i8 %br0, label %unreach [
+    i8 0, label %await.resume3
+    i8 1, label %await.resume1
+    i8 2, label %await.resume2
+  ]
+await.resume1:
+  %hdl = call i8* @g()
+  %addr2 = call i8* @llvm.coro.subfn.addr(i8* %hdl, i8 0)
+  %pv2 = bitcast i8* %addr2 to void (i8*)*
+  call fastcc void %pv2(i8* %hdl)
+  br label %final.suspend
+await.resume2:
+  %hdl2 = call i8* @h()
+  %addr3 = call i8* @llvm.coro.subfn.addr(i8* %hdl2, i8 0)
+  %pv3 = bitcast i8* %addr3 to void (i8*)*
+  call fastcc void %pv3(i8* %hdl2)
+  br label %final.suspend
+await.resume3:
+  %addr4 = call i8* @llvm.coro.subfn.addr(i8* null, i8 0)
+  %pv4 = bitcast i8* %addr4 to void (i8*)*
+  call fastcc void %pv4(i8* null)
+  br label %final.suspend
+final.suspend:
+  %suspend2 = call i8 @llvm.coro.suspend(token %save2, i1 false)
+  %cmp2 = icmp eq i8 %suspend2, 0
+  br i1 %cmp2, label %pre.exit, label %exit
+pre.exit:
+  br label %exit
+exit:
+  call i1 @llvm.coro.end(i8* null, i1 false)
+  ret void
+unreach:
+  unreachable
+}
+
+; Verify that in the initial function resume is not marked with musttail.
+; CHECK-LABEL: @f(
+; CHECK: %[[addr1:.+]] = call i8* @llvm.coro.subfn.addr(i8* null, i8 0)
+; CHECK-NEXT: %[[pv1:.+]] = bitcast i8* %[[addr1]] to void (i8*)*
+; CHECK-NOT: musttail call fastcc void %[[pv1]](i8* null)
+
+; Verify that in the resume part resume call is marked with musttail.
+; CHECK-LABEL: @f.resume(
+; CHECK: %[[hdl:.+]] = call i8* @g()
+; CHECK-NEXT: %[[addr2:.+]] = call i8* @llvm.coro.subfn.addr(i8* %[[hdl]], i8 0)
+; CHECK-NEXT: %[[pv2:.+]] = bitcast i8* %[[addr2]] to void (i8*)*
+; CHECK-NEXT: musttail call fastcc void %[[pv2]](i8* %[[hdl]])
+; CHECK-NEXT: ret void
+; CHECK: %[[hdl2:.+]] = call i8* @h()
+; CHECK-NEXT: %[[addr3:.+]] = call i8* @llvm.coro.subfn.addr(i8* %[[hdl2]], i8 0)
+; CHECK-NEXT: %[[pv3:.+]] = bitcast i8* %[[addr3]] to void (i8*)*
+; CHECK-NEXT: musttail call fastcc void %[[pv3]](i8* %[[hdl2]])
+; CHECK-NEXT: ret void
+; CHECK: %[[addr4:.+]] = call i8* @llvm.coro.subfn.addr(i8* null, i8 0)
+; CHECK-NEXT: %[[pv4:.+]] = bitcast i8* %[[addr4]] to void (i8*)*
+; CHECK-NEXT: musttail call fastcc void %[[pv4]](i8* null)
+; CHECK-NEXT: ret void
+
+
+
+declare token @llvm.coro.id(i32, i8* readnone, i8* nocapture readonly, i8*) #1
+declare i1 @llvm.coro.alloc(token) #2
+declare i64 @llvm.coro.size.i64() #3
+declare i8* @llvm.coro.begin(token, i8* writeonly) #2
+declare token @llvm.coro.save(i8*) #2
+declare i8* @llvm.coro.frame() #3
+declare i8 @llvm.coro.suspend(token, i1) #2
+declare i8* @llvm.coro.free(token, i8* nocapture readonly) #1
+declare i1 @llvm.coro.end(i8*, i1) #2
+declare i8* @llvm.coro.subfn.addr(i8* nocapture readonly, i8) #1
+declare i8* @malloc(i64)
+declare i8 @switch_result()
+declare i8* @g()
+declare i8* @h()
+
+attributes #0 = { "coroutine.presplit"="1" }
+attributes #1 = { argmemonly nounwind readonly }
+attributes #2 = { nounwind }
+attributes #3 = { nounwind readnone }