[llvm-commits] [llvm] r143006 - /llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
Jakob Stoklund Olesen
stoklund at 2pi.dk
Tue Oct 25 18:47:48 PDT 2011
Author: stoklund
Date: Tue Oct 25 20:47:48 2011
New Revision: 143006
URL: http://llvm.org/viewvc/llvm-project?rev=143006&view=rev
Log:
Don't use floating point to do an integer's job.
This code makes different decisions when compiled into x87 instructions
because of different rounding behavior. That caused phase 2/3
miscompares on 32-bit Linux when the phase 1 compiler was built with gcc
(using x87), and the phase 2 compiler was built with clang (using SSE).
This fixes PR11200.
Modified:
llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
Modified: llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp?rev=143006&r1=143005&r2=143006&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp Tue Oct 25 20:47:48 2011
@@ -2034,14 +2034,17 @@
return false;
APInt Range = ComputeRange(First, Last);
- double Density = TSize.roundToDouble() / Range.roundToDouble();
- if (Density < 0.4)
+ // The density is TSize / Range. Require at least 40%.
+ // It should not be possible for IntTSize to saturate for sane code, but make
+ // sure we handle Range saturation correctly.
+ uint64_t IntRange = Range.getLimitedValue(UINT64_MAX/10);
+ uint64_t IntTSize = TSize.getLimitedValue(UINT64_MAX/10);
+ if (IntTSize * 10 < IntRange * 4)
return false;
DEBUG(dbgs() << "Lowering jump table\n"
<< "First entry: " << First << ". Last entry: " << Last << '\n'
- << "Range: " << Range
- << ". Size: " << TSize << ". Density: " << Density << "\n\n");
+ << "Range: " << Range << ". Size: " << TSize << ".\n\n");
// Get the MachineFunction which holds the current MBB. This is used when
// inserting any additional MBBs necessary to represent the switch.
More information about the llvm-commits
mailing list