From 9608ed1311ca15fc43ee09dabea97d12f32485a4 Mon Sep 17 00:00:00 2001
From: Richard Sandiford <rsandifo@linux.vnet.ibm.com>
Date: Tue, 20 Aug 2013 09:11:13 +0000
Subject: [PATCH] Fix overly pessimistic shortcut in post-RA MachineLICM

Post-RA LICM keeps three sets of registers: PhysRegDefs, PhysRegClobbers
and TermRegs.  When it sees a definition of R it adds all aliases of R
to the corresponding set, so that when it needs to test for membership
it only needs to test a single register, rather than worrying about
aliases there too.  E.g. the final candidate loop just has:

    unsigned Def = Candidates[i].Def;
    if (!PhysRegClobbers.test(Def) && ...) {

to test whether register Def is multiply defined.

However, there was also a shortcut in ProcessMI to make sure we didn't
add candidates if we already knew that they would fail the final test.
This shortcut was more pessimistic than the final one because it
checked whether _any alias_ of the defined register was multiply defined.
This is too conservative for targets that define register pairs.
E.g. on z, R0 and R1 are sometimes used as a pair, so there is a
128-bit register that aliases both R0 and R1.  If a loop used
R0 and R1 independently, and the definition of R0 came first,
we would be able to hoist the R0 assignment (because that used
the final test quoted above) but not the R1 assignment (because
that meant we had two definitions of the paired R0/R1 register
and would fail the shortcut in ProcessMI).

This patch just uses the same check for the ProcessMI shortcut as
we use in the final candidate loop.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188774 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/CodeGen/MachineLICM.cpp    |  8 ++++----
 test/CodeGen/SystemZ/asm-17.ll | 22 ++++++++++++++++++++++
 2 files changed, 26 insertions(+), 4 deletions(-)

diff --git a/lib/CodeGen/MachineLICM.cpp b/lib/CodeGen/MachineLICM.cpp
index 6ad4e392adf..596b7a3ada9 100644
--- a/lib/CodeGen/MachineLICM.cpp
+++ b/lib/CodeGen/MachineLICM.cpp
@@ -468,12 +468,12 @@ void MachineLICM::ProcessMI(MachineInstr *MI,
     for (MCRegAliasIterator AS(Reg, TRI, true); AS.isValid(); ++AS) {
       if (PhysRegDefs.test(*AS))
         PhysRegClobbers.set(*AS);
-      if (PhysRegClobbers.test(*AS))
-        // MI defined register is seen defined by another instruction in
-        // the loop, it cannot be a LICM candidate.
-        RuledOut = true;
       PhysRegDefs.set(*AS);
     }
+    if (PhysRegClobbers.test(Reg))
+      // MI defined register is seen defined by another instruction in
+      // the loop, it cannot be a LICM candidate.
+      RuledOut = true;
   }
 
   // Only consider reloads for now and remats which do not have register
diff --git a/test/CodeGen/SystemZ/asm-17.ll b/test/CodeGen/SystemZ/asm-17.ll
index 33234fcae1c..86b0bbff953 100644
--- a/test/CodeGen/SystemZ/asm-17.ll
+++ b/test/CodeGen/SystemZ/asm-17.ll
@@ -80,3 +80,25 @@ define float @f7(float %in) {
   call void asm sideeffect "blah", "~{f0},~{cc}"()
   ret float %in
 }
+
+; Test that both registers in a GR128 pair get hoisted.
+define void @f8(i32 %count) {
+; CHECK-LABE: f8
+; CHECK-DAG: lhi %r0, 0
+; CHECK-DAG: lhi %r1, 1
+; CHECK: %loop
+; CHECK-NOT: %r
+; CHECK: blah %r0, %r1
+entry:
+  br label %loop
+
+loop:
+  %this = phi i32 [ %count, %entry ], [ %next, %loop ]
+  call void asm sideeffect "blah $0, $1", "{r0},{r1}" (i32 0, i32 1)
+  %next = sub i32 %this, 1
+  %cmp = icmp ne i32 %next, 0
+  br i1 %cmp, label %loop, label %exit
+
+exit:
+  ret void
+}
-- 
2.34.1