[RDF] Improve handling of inline-asm

- Keep implicit defs from inline-asm instructions.
- Treat register references from inline-asm as fixed.

llvm-svn: 267936

lib/Target/Hexagon/RDFGraph.cpp

@@ -686,10 +686,10 @@ bool TargetOperandInfo::isClobbering(const MachineInstr &In, unsigned OpNum)
   return false;
 }
 
-// Check if the given instruction specifically requires 
+// Check if the given instruction specifically requires
 bool TargetOperandInfo::isFixedReg(const MachineInstr &In, unsigned OpNum)
       const {
-  if (In.isCall() || In.isReturn())
+  if (In.isCall() || In.isReturn() || In.isInlineAsm())
     return true;
   const MCInstrDesc &D = In.getDesc();
   if (!D.getImplicitDefs() && !D.getImplicitUses())
@@ -1180,6 +1180,7 @@ void DataFlowGraph::buildStmt(NodeAddr<BlockNode*> BA, MachineInstr &In) {
       ImpUses.insert({R, 0});
 
   bool IsCall = In.isCall(), IsReturn = In.isReturn();
+  bool IsInlineAsm = In.isInlineAsm();
   bool IsPredicated = TII.isPredicated(In);
   unsigned NumOps = In.getNumOperands();
 
@@ -1213,7 +1214,7 @@ void DataFlowGraph::buildStmt(NodeAddr<BlockNode*> BA, MachineInstr &In) {
     if (!Op.isReg() || !Op.isDef() || !Op.isImplicit())
       continue;
     RegisterRef RR = { Op.getReg(), Op.getSubReg() };
-    if (!IsCall && !ImpDefs.count(RR))
+    if (!IsCall && !IsInlineAsm && !ImpDefs.count(RR))
       continue;
     if (DoneDefs.count(RR))
       continue;
@@ -1238,7 +1239,7 @@ void DataFlowGraph::buildStmt(NodeAddr<BlockNode*> BA, MachineInstr &In) {
     // instructions regardless of whether or not they appear in the instruction
     // descriptor's list.
     bool Implicit = Op.isImplicit();
-    bool TakeImplicit = IsReturn || IsCall || IsPredicated;
+    bool TakeImplicit = IsReturn || IsCall || IsInlineAsm || IsPredicated;
     if (Implicit && !TakeImplicit && !ImpUses.count(RR))
       continue;
     uint16_t Flags = NodeAttrs::None;

test/CodeGen/Hexagon/rdf-inline-asm-fixed.ll

@@ -0,0 +1,37 @@
+; RUN: llc -march=hexagon < %s | FileCheck %s
+; CHECK: r0 = #24
+; CHECK-NEXT: r1 =
+; // R2 should be assigned a value from R3+.
+; CHECK-NEXT: r2 = r{{[3-9]}}
+; CHECK-NEXT: trap0
+
+target datalayout = "e-m:e-p:32:32:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f64:64:64-f32:32:32-v64:64:64-v32:32:32-a:0-n16:32"
+target triple = "hexagon"
+
+; Function Attrs: nounwind
+define i32 @foo(i32 %status) #0 {
+entry:
+  %arg1 = alloca i32, align 4
+  %0 = bitcast i32* %arg1 to i8*
+  call void @llvm.lifetime.start(i64 4, i8* %0) #2
+  store i32 %status, i32* %arg1, align 4, !tbaa !1
+  %1 = call i32 asm sideeffect "r0 = #$1\0Ar1 = $2\0Ar2 = $4\0Atrap0 (#0)\0A$0 = r0", "=r,i,r,*m,r,~{r0},~{r1},~{r2}"(i32 24, i32* nonnull %arg1, i32* nonnull %arg1, i32 %status) #2, !srcloc !5
+  call void @llvm.lifetime.end(i64 4, i8* %0) #2
+  ret i32 %1
+}
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start(i64, i8* nocapture) #1
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end(i64, i8* nocapture) #1
+
+attributes #0 = { nounwind "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="hexagonv5" "target-features"="-hvx,-hvx-double" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind }
+attributes #2 = { nounwind }
+
+!1 = !{!2, !2, i64 0}
+!2 = !{!"int", !3, i64 0}
+!3 = !{!"omnipotent char", !4, i64 0}
+!4 = !{!"Simple C/C++ TBAA"}
+!5 = !{i32 110, i32 129, i32 146, i32 163, i32 183}

test/CodeGen/Hexagon/rdf-inline-asm.ll

@@ -0,0 +1,36 @@
+; RUN: llc -march=hexagon < %s
+; REQUIRES: asserts
+
+target datalayout = "e-m:e-p:32:32:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f64:64:64-f32:32:32-v64:64:64-v32:32:32-a:0-n16:32"
+target triple = "hexagon"
+
+@x = common global i32* null, align 4
+
+; Function Attrs: nounwind
+define i32 @inotify_init() #0 {
+entry:
+  %0 = tail call i32 asm sideeffect "trap0(#1);\0A", "={r0},{r6},~{memory}"(i32 1043) #1, !srcloc !1
+  %cmp = icmp sgt i32 %0, -4096
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %sub = sub nsw i32 0, %0
+  %1 = load i32*, i32** @x, align 4, !tbaa !2
+  store i32 %sub, i32* %1, align 4, !tbaa !6
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  %retval1.0 = phi i32 [ -1, %if.then ], [ %0, %entry ]
+  ret i32 %retval1.0
+}
+
+attributes #0 = { nounwind "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="hexagonv60" "target-features"="+hvx,-hvx-double" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { nounwind }
+
+!1 = !{i32 155}
+!2 = !{!3, !3, i64 0}
+!3 = !{!"any pointer", !4, i64 0}
+!4 = !{!"omnipotent char", !5, i64 0}
+!5 = !{!"Simple C/C++ TBAA"}
+!6 = !{!7, !7, i64 0}
+!7 = !{!"long", !4, i64 0}