[llvm] r301700 - [RDF] Correctly calculate lane masks for defs

[Krzysztof Parzyszek via llvm-commits]

Author: kparzysz
Date: Fri Apr 28 16:57:53 2017
New Revision: 301700

URL: http://llvm.org/viewvc/llvm-project?rev=301700&view=rev
Log:
[RDF] Correctly calculate lane masks for defs

Added:
    llvm/trunk/test/CodeGen/Hexagon/rdf-def-mask.ll
Modified:
    llvm/trunk/lib/Target/Hexagon/RDFLiveness.cpp

Modified: llvm/trunk/lib/Target/Hexagon/RDFLiveness.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/Hexagon/RDFLiveness.cpp?rev=301700&r1=301699&r2=301700&view=diff
==============================================================================
--- llvm/trunk/lib/Target/Hexagon/RDFLiveness.cpp (original)
+++ llvm/trunk/lib/Target/Hexagon/RDFLiveness.cpp Fri Apr 28 16:57:53 2017
@@ -759,8 +759,13 @@ void Liveness::computeLiveIns() {
             // all related shadows as a single use cluster.
             RegisterRef S(RS.first, P.second);
             NodeList Ds = getAllReachingDefs(S, PUA, true, false, NoRegs);
-            for (NodeAddr<DefNode*> D : Ds)
-              LOX[S.Reg].insert({D.Id, S.Mask});
+            for (NodeAddr<DefNode*> D : Ds) {
+              // Calculate the mask corresponding to the visited def.
+              RegisterAggr TA(PRI);
+              TA.insert(D.Addr->getRegRef(DFG)).intersect(S);
+              LaneBitmask TM = TA.makeRegRef().Mask;
+              LOX[S.Reg].insert({D.Id, TM});
+            }
           }
         }
 

Added: llvm/trunk/test/CodeGen/Hexagon/rdf-def-mask.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/Hexagon/rdf-def-mask.ll?rev=301700&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/Hexagon/rdf-def-mask.ll (added)
+++ llvm/trunk/test/CodeGen/Hexagon/rdf-def-mask.ll Fri Apr 28 16:57:53 2017
@@ -0,0 +1,52 @@
+; RUN: llc -march=hexagon -O3 -verify-machineinstrs < %s | FileCheck %s
+; REQUIRES: asserts
+
+; Check for sane output. This testcase used to crash.
+; CHECK: jumpr r31
+
+target triple = "hexagon"
+
+ at g0 = external hidden unnamed_addr constant [9 x i16], align 8
+
+; Function Attrs: nounwind readnone
+define i64 @fred(i32 %a0) local_unnamed_addr #0 {
+b1:
+  %v2 = icmp slt i32 %a0, 1
+  br i1 %v2, label %b26, label %b3
+
+b3:                                               ; preds = %b1
+  %v4 = tail call i32 @llvm.hexagon.S2.clb(i32 %a0)
+  %v5 = add nsw i32 %v4, -12
+  %v6 = add nsw i32 %v4, -28
+  %v7 = tail call i32 @llvm.hexagon.S2.asl.r.r(i32 %a0, i32 %v6)
+  %v8 = add nsw i32 %v7, -8
+  %v9 = tail call i32 @llvm.hexagon.S2.asl.r.r(i32 %a0, i32 %v5)
+  %v10 = getelementptr inbounds [9 x i16], [9 x i16]* @g0, i32 0, i32 %v8
+  %v11 = load i16, i16* %v10, align 2
+  %v12 = sext i16 %v11 to i32
+  %v13 = shl nsw i32 %v12, 16
+  %v14 = add nsw i32 %v7, -7
+  %v15 = getelementptr inbounds [9 x i16], [9 x i16]* @g0, i32 0, i32 %v14
+  %v16 = load i16, i16* %v15, align 2
+  %v17 = sub i16 %v11, %v16
+  %v18 = and i32 %v9, 65535
+  %v19 = zext i16 %v17 to i32
+  %v20 = tail call i32 @llvm.hexagon.M2.mpyu.nac.ll.s0(i32 %v13, i32 %v18, i32 %v19) #1
+  %v21 = add nsw i32 %v4, -32
+  %v22 = zext i32 %v21 to i64
+  %v23 = shl nuw i64 %v22, 32
+  %v24 = zext i32 %v20 to i64
+  %v25 = or i64 %v23, %v24
+  br label %b26
+
+b26:                                              ; preds = %b3, %b1
+  %v27 = phi i64 [ %v25, %b3 ], [ 2147483648, %b1 ]
+  ret i64 %v27
+}
+
+declare i32 @llvm.hexagon.S2.clb(i32) #1
+declare i32 @llvm.hexagon.S2.asl.r.r(i32, i32) #1
+declare i32 @llvm.hexagon.M2.mpyu.nac.ll.s0(i32, i32, i32) #1
+
+attributes #0 = { nounwind readnone "target-cpu"="hexagonv55" "target-features"="-hvx,-hvx-double,-long-calls" }
+attributes #1 = { nounwind readnone }