From 4e2b66100518432e1e4fa8c603601fd946c1f945 Mon Sep 17 00:00:00 2001
From: David Xu <David.Xu@arm.com>
Date: Tue, 2 Sep 2014 09:33:56 +0000
Subject: [PATCH] Merge Extend and Shift into a UBFX

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@216899 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/AArch64/AArch64ISelDAGToDAG.cpp | 28 ++++++++++++----------
 test/CodeGen/AArch64/andandshift.ll        | 28 ++++++++++++++++++++++
 2 files changed, 43 insertions(+), 13 deletions(-)
 create mode 100644 test/CodeGen/AArch64/andandshift.ll

diff --git a/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp b/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
index 5a624e381ca..c6bdf566c13 100644
--- a/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
+++ b/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
@@ -1381,20 +1381,21 @@ static bool isBitfieldExtractOpFromAnd(SelectionDAG *CurDAG, SDNode *N,
   return true;
 }
 
-static bool isOneBitExtractOpFromShr(SDNode *N, unsigned &Opc, SDValue &Opd0,
-                                     unsigned &LSB, unsigned &MSB) {
-  // We are looking for the following pattern which basically extracts a single
-  // bit from the source value and places it in the LSB of the destination
-  // value, all other bits of the destination value or set to zero:
+static bool isSeveralBitsExtractOpFromShr(SDNode *N, unsigned &Opc,
+                                          SDValue &Opd0, unsigned &LSB,
+                                          unsigned &MSB) {
+  // We are looking for the following pattern which basically extracts several
+  // continuous bits from the source value and places it from the LSB of the
+  // destination value, all other bits of the destination value or set to zero:
   //
   // Value2 = AND Value, MaskImm
   // SRL Value2, ShiftImm
   //
-  // with MaskImm >> ShiftImm == 1.
+  // with MaskImm >> ShiftImm to search for the bit width.
   //
   // This gets selected into a single UBFM:
   //
-  // UBFM Value, ShiftImm, ShiftImm
+  // UBFM Value, ShiftImm, BitWide + Srl_imm -1
   //
 
   if (N->getOpcode() != ISD::SRL)
@@ -1410,15 +1411,16 @@ static bool isOneBitExtractOpFromShr(SDNode *N, unsigned &Opc, SDValue &Opd0,
   if (!isIntImmediate(N->getOperand(1), Srl_imm))
     return false;
 
-  // Check whether we really have a one bit extract here.
-  if (And_mask >> Srl_imm == 0x1) {
+  // Check whether we really have several bits extract here.
+  unsigned BitWide = 64 - CountLeadingOnes_64(~(And_mask >> Srl_imm));
+  if (BitWide && isMask_64(And_mask >> Srl_imm)) {
     if (N->getValueType(0) == MVT::i32)
       Opc = AArch64::UBFMWri;
     else
       Opc = AArch64::UBFMXri;
 
-    LSB = MSB = Srl_imm;
-
+    LSB = Srl_imm;
+    MSB = BitWide + Srl_imm - 1;
     return true;
   }
 
@@ -1439,8 +1441,8 @@ static bool isBitfieldExtractOpFromShr(SDNode *N, unsigned &Opc, SDValue &Opd0,
   assert((VT == MVT::i32 || VT == MVT::i64) &&
          "Type checking must have been done before calling this function");
 
-  // Check for AND + SRL doing a one bit extract.
-  if (isOneBitExtractOpFromShr(N, Opc, Opd0, LSB, MSB))
+  // Check for AND + SRL doing several bits extract.
+  if (isSeveralBitsExtractOpFromShr(N, Opc, Opd0, LSB, MSB))
     return true;
 
   // we're looking for a shift of a shift
diff --git a/test/CodeGen/AArch64/andandshift.ll b/test/CodeGen/AArch64/andandshift.ll
new file mode 100644
index 00000000000..e2c7a098908
--- /dev/null
+++ b/test/CodeGen/AArch64/andandshift.ll
@@ -0,0 +1,28 @@
+; RUN: llc -O3 < %s | FileCheck %s 
+target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
+target triple = "arm64--linux-gnu"
+
+; Function Attrs: nounwind readnone
+define i32 @test1(i8 %a) {
+; CHECK-LABLE: @test1
+; CHECK: ubfx {{w[0-9]+}}, w0, #3, #5
+entry:
+  %conv = zext i8 %a to i32
+  %shr1 = lshr i32 %conv, 3
+  ret i32 %shr1
+}
+
+; Function Attrs: nounwind readnone
+define i32 @test2(i8 %a) {
+; CHECK-LABLE: @test2
+; CHECK: and {{w[0-9]+}}, w0, #0xff
+; CHECK: ubfx {{w[0-9]+}}, w0, #3, #5
+entry:
+  %conv = zext i8 %a to i32
+  %cmp = icmp ugt i8 %a, 47
+  %shr5 = lshr i32 %conv, 3
+  %retval.0 = select i1 %cmp, i32 %shr5, i32 %conv
+  ret i32 %retval.0
+}
+
+
-- 
2.34.1