[MiSched] Fix a logic error in tryPressure()

Fixes a logic error in the MachineScheduler found by Steve Montgomery (and
confirmed by Andy). This has gone unfixed for months because the fix has been
found to introduce some small performance regressions. However, Andy has
recommended that, at this point, we fix this to avoid further dependence on the
incorrect behavior (and then follow-up separately on any regressions), and I
agree.

Fixes PR18883.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219512 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/MachineScheduler.cpp b/lib/CodeGen/MachineScheduler.cpp
index 7e3c33b..b140fc6 100644 (file)
--- a/lib/CodeGen/MachineScheduler.cpp
+++ b/lib/CodeGen/MachineScheduler.cpp
@@ -2495,8 +2495,8 @@ static bool tryPressure(const PressureChange &TryP,
   }
   // If one candidate decreases and the other increases, go with it.
   // Invalid candidates have UnitInc==0.
-  if (tryLess(TryP.getUnitInc() < 0, CandP.getUnitInc() < 0, TryCand, Cand,
-              Reason)) {
+  if (tryGreater(TryP.getUnitInc() < 0, CandP.getUnitInc() < 0, TryCand, Cand,
+                 Reason)) {
     return true;
   }
   // If the candidates are decreasing pressure, reverse priority.

diff --git a/test/CodeGen/X86/misched-matmul.ll b/test/CodeGen/X86/misched-matmul.ll
index 3ea6512..5454b7c 100644 (file)
--- a/test/CodeGen/X86/misched-matmul.ll
+++ b/test/CodeGen/X86/misched-matmul.ll
@@ -10,7 +10,7 @@
 ; more complex cases.
 ;
 ; CHECK: @wrap_mul4
-; CHECK: 22 regalloc - Number of spills inserted
+; CHECK: 23 regalloc - Number of spills inserted
 
 define void @wrap_mul4(double* nocapture %Out, [4 x double]* nocapture %A, [4 x double]* nocapture %B) #0 {
 entry:

diff --git a/test/CodeGen/X86/ragreedy-last-chance-recoloring.ll b/test/CodeGen/X86/ragreedy-last-chance-recoloring.ll
index d8e4572..49d58f4 100644 (file)
--- a/test/CodeGen/X86/ragreedy-last-chance-recoloring.ll
+++ b/test/CodeGen/X86/ragreedy-last-chance-recoloring.ll
@@ -2,10 +2,12 @@
 ; Without the last chance recoloring, this test fails with:
 ; "ran out of registers".
 
-; RUN: not llc -regalloc=greedy -relocation-model=pic -lcr-max-depth=0  < %s 2>&1 | FileCheck %s --check-prefix=CHECK-DEPTH
+; NOTE: With the fix to PR18883, we don't actually run out of registers here
+; any more, and so those checks are disabled. This test remains only for general coverage.
+; XXX: not llc -regalloc=greedy -relocation-model=pic -lcr-max-depth=0  < %s 2>&1 | FileCheck %s --check-prefix=CHECK-DEPTH
 ; Test whether failure due to cutoff for depth is reported
 
-; RUN: not llc -regalloc=greedy -relocation-model=pic -lcr-max-interf=1  < %s 2>&1 | FileCheck %s --check-prefix=CHECK-INTERF
+; XXX: not llc -regalloc=greedy -relocation-model=pic -lcr-max-interf=1  < %s 2>&1 | FileCheck %s --check-prefix=CHECK-INTERF
 ; Test whether failure due to cutoff for interference is reported
 
 ; RUN: llc -regalloc=greedy -relocation-model=pic -lcr-max-interf=1 -lcr-max-depth=0 -exhaustive-register-search < %s > %t 2>&1