[MBP] Add flags to disable the BadCFGConflict check in MachineBlockPlacement.

Some benchmarks have shown that this could lead to a potential
performance benefit, and so adding some flags to try to help measure the
difference.

A possible explanation. In diamond-shaped CFGs (A followed by either
B or C both followed by D), putting B and C both in between A and
D leads to the code being less dense than it could be. Always either
B or C have to be skipped increasing the chance of cache misses etc.
Moving either B or C to after D might be beneficial on average.

In the long run, but we should probably do a better job of analyzing the
basic block and branch probabilities to move the correct one of B or
C to after D. But even if we don't use this in the long run, it is
a good baseline for benchmarking.

Original patch authored by Daniel Jasper with test tweaks and a second
flag added by me.

Differential Revision: http://reviews.llvm.org/D6969

llvm-svn: 226034

1 files changed, 112 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/X86/code_placement_bad_cfg_check.ll b/llvm/test/CodeGen/X86/code_placement_bad_cfg_check.ll
new file mode 100644
index 00000000000..a5f9e8aab1b
--- /dev/null
+++ b/llvm/test/CodeGen/X86/code_placement_bad_cfg_check.ll
@@ -0,0 +1,112 @@
+; RUN: llc -mcpu=corei7 -mtriple=x86_64-linux < %s | FileCheck %s -check-prefix=CHECK-BAD-CFG
+; RUN: llc -mcpu=corei7 -mtriple=x86_64-linux -no-bad-cfg-conflict-check < %s | FileCheck %s -check-prefix=CHECK-NO-BAD-CFG
+; RUN: llc -mcpu=corei7 -mtriple=x86_64-linux -only-hot-bad-cfg-conflict-check < %s | FileCheck %s -check-prefix=CHECK-HOT-BAD-CFG
+
+define void @foo(i32 %t) {
+; Test that we lift the call to 'c' up to immediately follow the call to 'b'
+; when we disable the cfg conflict check.
+;
+; CHECK-BAD-CFG-LABEL: foo:
+; CHECK-BAD-CFG: callq b
+; CHECK-BAD-CFG: callq a
+; CHECK-BAD-CFG: callq c
+;
+; CHECK-NO-BAD-CFG-LABEL: foo:
+; CHECK-NO-BAD-CFG: callq b
+; CHECK-NO-BAD-CFG: callq c
+; CHECK-NO-BAD-CFG: callq a
+;
+; CHECK-HOT-BAD-CFG-LABEL: foo:
+; CHECK-HOT-BAD-CFG: callq b
+; CHECK-HOT-BAD-CFG: callq c
+; CHECK-HOT-BAD-CFG: callq a
+
+entry:
+  %cmp = icmp eq i32 %t, 0
+  br i1 %cmp, label %if.then, label %if.else
+
+if.then:
+  call void @a()
+  br label %if.end
+
+if.else:
+  call void @b()
+  br label %if.end
+
+if.end:
+  call void @c()
+  ret void
+}
+
+define void @bar(i32 %t1, i32 %t2, i32 %t3) {
+; Test that we lift the call to 'c' up to immediately follow the call to 'b'
+; when we disable the cfg conflict check.
+;
+; CHECK-BAD-CFG-LABEL: bar:
+; CHECK-BAD-CFG: callq a
+; CHECK-BAD-CFG: callq c
+; CHECK-BAD-CFG: callq d
+; CHECK-BAD-CFG: callq f
+; CHECK-BAD-CFG: callq b
+; CHECK-BAD-CFG: callq e
+; CHECK-BAD-CFG: callq g
+;
+; CHECK-NO-BAD-CFG-LABEL: bar:
+; CHECK-NO-BAD-CFG: callq a
+; CHECK-NO-BAD-CFG: callq c
+; CHECK-NO-BAD-CFG: callq g
+; CHECK-NO-BAD-CFG: callq d
+; CHECK-NO-BAD-CFG: callq f
+; CHECK-NO-BAD-CFG: callq b
+; CHECK-NO-BAD-CFG: callq e
+;
+; CHECK-HOT-BAD-CFG-LABEL: bar:
+; CHECK-HOT-BAD-CFG: callq a
+; CHECK-HOT-BAD-CFG: callq c
+; CHECK-HOT-BAD-CFG: callq d
+; CHECK-HOT-BAD-CFG: callq f
+; CHECK-HOT-BAD-CFG: callq g
+; CHECK-HOT-BAD-CFG: callq b
+; CHECK-HOT-BAD-CFG: callq e
+
+entry:
+  br i1 undef, label %if1.then, label %if1.else
+
+if1.then:
+  call void @a()
+  %cmp2 = icmp eq i32 %t2, 0
+  br i1 %cmp2, label %if2.then, label %if2.else
+
+if2.then:
+  call void @b()
+  br label %if.end
+
+if2.else:
+  call void @c()
+  br label %if.end
+
+if1.else:
+  call void @d()
+  %cmp3 = icmp eq i32 %t3, 0
+  br i1 %cmp3, label %if3.then, label %if3.else
+
+if3.then:
+  call void @e()
+  br label %if.end
+
+if3.else:
+  call void @f()
+  br label %if.end
+
+if.end:
+  call void @g()
+  ret void
+}
+
+declare void @a()
+declare void @b()
+declare void @c()
+declare void @d()
+declare void @e()
+declare void @f()
+declare void @g()