Enhance address-mode folding of ISD::ADD to handle cases where the

operands can't both be fully folded at the same time. For example,
in the included testcase, a global variable is being added with
an add of two values. The global variable wants RIP-relative
addressing, so it can't share the address with another base
register, but it's still possible to fold the initial add.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@66865 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 38 insertions, 0 deletions

diff --git a/lib/Target/X86/X86ISelDAGToDAG.cpp b/lib/Target/X86/X86ISelDAGToDAG.cpp
index 855c890120..f81ab6f8a7 100644
--- a/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -907,6 +907,19 @@ bool X86DAGToDAGISel::MatchAddress(SDValue N, X86ISelAddressMode &AM,
         !MatchAddress(N.getNode()->getOperand(0), AM, false, Depth+1))
       return false;
     AM = Backup;
+
+    // If we couldn't fold both operands into the address at the same time,
+    // see if we can just put each operand into a register and fold at least
+    // the add.
+    if (AM.BaseType == X86ISelAddressMode::RegBase &&
+        !AM.Base.Reg.getNode() &&
+        !AM.IndexReg.getNode() &&
+        !AM.isRIPRel) {
+      AM.Base.Reg = N.getNode()->getOperand(0);
+      AM.IndexReg = N.getNode()->getOperand(1);
+      AM.Scale = 1;
+      return false;
+    }
     break;
   }
 
diff --git a/test/CodeGen/X86/fold-add.ll b/test/CodeGen/X86/fold-add.ll
new file mode 100644
index 0000000000..2828ad22ef
--- /dev/null
+++ b/test/CodeGen/X86/fold-add.ll
@@ -0,0 +1,25 @@
+; RUN: llvm-as < %s | llc -march=x86-64 | grep {cmpb	\$0, (%r.\*,%r.\*)}
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
+target triple = "x86_64-apple-darwin9.6"
+@prev_length = internal global i32 0		; <i32*> [#uses=1]
+@window = internal global [65536 x i8] zeroinitializer, align 32		; <[65536 x i8]*> [#uses=1]
+@llvm.used = appending global [1 x i8*] [i8* bitcast (i32 (i32)* @longest_match to i8*)]		; <[1 x i8*]*> [#uses=0]
+
+define fastcc i32 @longest_match(i32 %cur_match) nounwind {
+entry:
+	%0 = load i32* @prev_length, align 4		; <i32> [#uses=3]
+	%1 = zext i32 %cur_match to i64		; <i64> [#uses=1]
+	%2 = sext i32 %0 to i64		; <i64> [#uses=1]
+	%.sum3 = add i64 %1, %2		; <i64> [#uses=1]
+	%3 = getelementptr [65536 x i8]* @window, i64 0, i64 %.sum3		; <i8*> [#uses=1]
+	%4 = load i8* %3, align 1		; <i8> [#uses=1]
+	%5 = icmp eq i8 %4, 0		; <i1> [#uses=1]
+	br i1 %5, label %bb5, label %bb23
+
+bb5:		; preds = %entry
+	ret i32 %0
+
+bb23:		; preds = %entry
+	ret i32 %0
+}