Add a late optimization to BranchFolding that hoist common instruction sequences

at the start of basic blocks to their common predecessor. It's actually quite
common (e.g. about 50 times in JM/lencod) and has shown to be a nice code size
benefit. e.g.

        pushq   %rax
        testl   %edi, %edi
        jne     LBB0_2
## BB#1:
        xorb    %al, %al
        popq    %rdx
        ret
LBB0_2:
        xorb    %al, %al
        callq   _foo
        popq    %rdx
        ret

=>

        pushq   %rax
        xorb    %al, %al
        testl   %edi, %edi
        je      LBB0_2
## BB#1:
        callq   _foo
LBB0_2:
        popq    %rdx
        ret

rdar://9145558

llvm-svn: 131172

1 files changed, 28 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/X86/hoist-common.ll b/llvm/test/CodeGen/X86/hoist-common.ll
new file mode 100644
index 00000000000..2b1550a4ffa
--- /dev/null
+++ b/llvm/test/CodeGen/X86/hoist-common.ll
@@ -0,0 +1,28 @@
+; RUN: llc < %s -march=x86-64  | FileCheck %s
+
+; Common "xorb al, al" instruction in the two successor blocks should be
+; moved to the entry block above the test + je.
+
+; rdar://9145558
+
+define zeroext i1 @t(i32 %c) nounwind ssp {
+entry:
+; CHECK: t:
+; CHECK: xorb %al, %al
+; CHECK: test
+; CHECK: je
+  %tobool = icmp eq i32 %c, 0
+  br i1 %tobool, label %return, label %if.then
+
+if.then:
+; CHECK: callq
+  %call = tail call zeroext i1 (...)* @foo() nounwind
+  br label %return
+
+return:
+; CHECK: ret
+  %retval.0 = phi i1 [ %call, %if.then ], [ false, %entry ]
+  ret i1 %retval.0
+}
+
+declare zeroext i1 @foo(...)