[llvm-commits] [llvm] r166068 - in /llvm/trunk: lib/CodeGen/LiveRangeEdit.cpp test/CodeGen/X86/pr14098.ll

[Jakob Stoklund Olesen]

Author: stoklund
Date: Tue Oct 16 17:51:58 2012
New Revision: 166068

URL: http://llvm.org/viewvc/llvm-project?rev=166068&view=rev
Log:
Avoid rematerializing a redef immediately after the old def.

PR14098 contains an example where we would rematerialize a MOV8ri
immediately after the original instruction:

  %vreg7:sub_8bit<def> = MOV8ri 9; GR32_ABCD:%vreg7
  %vreg22:sub_8bit<def> = MOV8ri 9; GR32_ABCD:%vreg7

Besides being pointless, it is also wrong since the original instruction
only redefines part of the register, and the value read by the new
instruction is wrong.

The problem was the LiveRangeEdit::allUsesAvailableAt() didn't
special-case OrigIdx == UseIdx and found the wrong SSA value.

Added:
    llvm/trunk/test/CodeGen/X86/pr14098.ll
Modified:
    llvm/trunk/lib/CodeGen/LiveRangeEdit.cpp

Modified: llvm/trunk/lib/CodeGen/LiveRangeEdit.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/LiveRangeEdit.cpp?rev=166068&r1=166067&r2=166068&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/LiveRangeEdit.cpp (original)
+++ llvm/trunk/lib/CodeGen/LiveRangeEdit.cpp Tue Oct 16 17:51:58 2012
@@ -96,6 +96,13 @@
     const VNInfo *OVNI = li.getVNInfoAt(OrigIdx);
     if (!OVNI)
       continue;
+
+    // Don't allow rematerialization immediately after the original def.
+    // It would be incorrect if OrigMI redefines the register.
+    // See PR14098.
+    if (SlotIndex::isSameInstr(OrigIdx, UseIdx))
+      return false;
+
     if (OVNI != li.getVNInfoAt(UseIdx))
       return false;
   }

Added: llvm/trunk/test/CodeGen/X86/pr14098.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/pr14098.ll?rev=166068&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/pr14098.ll (added)
+++ llvm/trunk/test/CodeGen/X86/pr14098.ll Tue Oct 16 17:51:58 2012
@@ -0,0 +1,23 @@
+; RUN: llc -mtriple i386-unknown-linux-gnu -relocation-model=pic -verify-machineinstrs < %s
+; We used to crash on this.
+
+declare void @foo()
+declare void @foo3(i1 %x)
+define void @bar(i1 %a1, i16 %a2) nounwind align 2 {
+bb0:
+  %a3 = trunc i16 %a2 to i8
+  %a4 = lshr i16 %a2, 8
+  %a5 = trunc i16 %a4 to i8
+  br i1 %a1, label %bb1, label %bb2
+bb1:
+  br label %bb2
+bb2:
+  %a6 = phi i8 [ 3, %bb0 ], [ %a5, %bb1 ]
+  %a7 = phi i8 [ 9, %bb0 ], [ %a3, %bb1 ]
+  %a8 = icmp eq i8 %a6, 1
+  call void @foo()
+  %a9 = icmp eq i8 %a7, 0
+  call void @foo3(i1 %a9)
+  call void @foo3(i1 %a8)
+  ret void
+}