[llvm] r236897 - [Fast-ISel] Clear kill flags on registers replaced by updateValueMap.

[Pete Cooper]

Author: pete
Date: Fri May  8 15:46:54 2015
New Revision: 236897

URL: http://llvm.org/viewvc/llvm-project?rev=236897&view=rev
Log:
[Fast-ISel] Clear kill flags on registers replaced by updateValueMap.

When selecting an extract instruction, we don't actually generate code but instead work out which register we are reading, and rewrite uses of the extract def to the source register.  This is done via updateValueMap,.

However, its possible that the source register we are rewriting *to* to also have uses.  If those uses are after a kill of the value we are rewriting *from* then we have uses after a kill and the verifier fails.

This code checks for the case where the to register is also used, and if so it clears all kill on the from register.  This is conservative, but better that always clearing kills on the from register.

Added:
    llvm/trunk/test/CodeGen/ARM/fast-isel-update-valuemap-for-extract.ll
Modified:
    llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp

Modified: llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp?rev=236897&r1=236896&r2=236897&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp Fri May  8 15:46:54 2015
@@ -578,6 +578,13 @@ bool SelectionDAGISel::runOnMachineFunct
         TargetRegisterInfo::isVirtualRegister(To))
       MRI.constrainRegClass(To, MRI.getRegClass(From));
     // Replace it.
+
+
+    // Replacing one register with another won't touch the kill flags.
+    // We need to conservatively clear the kill flags as a kill on the old
+    // register might dominate existing uses of the new register.
+    if (!MRI.use_empty(To))
+      MRI.clearKillFlags(From);
     MRI.replaceRegWith(From, To);
   }
 

Added: llvm/trunk/test/CodeGen/ARM/fast-isel-update-valuemap-for-extract.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/fast-isel-update-valuemap-for-extract.ll?rev=236897&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/ARM/fast-isel-update-valuemap-for-extract.ll (added)
+++ llvm/trunk/test/CodeGen/ARM/fast-isel-update-valuemap-for-extract.ll Fri May  8 15:46:54 2015
@@ -0,0 +1,24 @@
+; RUN: llc %s -o - -fast-isel=true -O0 -verify-machineinstrs | FileCheck %s
+
+target datalayout = "e-m:o-p:32:32-f64:32:64-v64:32:64-v128:32:128-a:0:32-n32-S32"
+target triple = "thumbv7-apple-ios8.0.0"
+
+; This test ensures that when fast-isel rewrites uses of the vreg for %tmp29, it also
+; updates kill flags on the shift instruction generated as part of the gep.
+; This was failing instruction verification.
+
+; CHECK-LABEL: @test
+
+%struct.node = type opaque
+
+declare void @foo([4 x i32], %struct.node*)
+
+define void @test([4 x i32] %xpic.coerce, %struct.node* %t) {
+bb:
+  %tmp29 = extractvalue [4 x i32] %xpic.coerce, 0
+  %tmp40 = bitcast %struct.node* %t to [8 x %struct.node*]*
+  %tmp41 = getelementptr inbounds [8 x %struct.node*], [8 x %struct.node*]* %tmp40, i32 0, i32 %tmp29
+  %tmp42 = load %struct.node*, %struct.node** %tmp41, align 4
+  call void @foo([4 x i32] %xpic.coerce, %struct.node* %tmp42)
+  ret void
+}