From ccf53013cdc56b2c4ca9dd24688f71d29b007630 Mon Sep 17 00:00:00 2001
From: Juergen Ributzka <juergen@apple.com>
Date: Wed, 27 Aug 2014 21:38:33 +0000
Subject: [PATCH] [FastISel][AArch64] Fix simplify address when the address
 comes from a shift.

When the address comes directly from a shift instruction then the address
computation cannot be folded into the memory instruction, because the zero
register is not available as a base register. Simplify addess needs to emit the
shift instruction and use the result as base register.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@216621 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/AArch64/AArch64FastISel.cpp        |  4 ++++
 .../AArch64/fast-isel-addressing-modes.ll     | 21 +++++++++++++++++++
 2 files changed, 25 insertions(+)

diff --git a/lib/Target/AArch64/AArch64FastISel.cpp b/lib/Target/AArch64/AArch64FastISel.cpp
index a6e7daa0d9b..2faa1398064 100644
--- a/lib/Target/AArch64/AArch64FastISel.cpp
+++ b/lib/Target/AArch64/AArch64FastISel.cpp
@@ -708,6 +708,10 @@ bool AArch64FastISel::SimplifyAddress(Address &Addr, MVT VT) {
       Addr.getOffsetReg())
     RegisterOffsetNeedsLowering = true;
 
+  // Cannot encode zero register as base.
+  if (Addr.isRegBase() && Addr.getOffsetReg() && !Addr.getReg())
+    RegisterOffsetNeedsLowering = true;
+
   // If this is a stack pointer and the offset needs to be simplified then put
   // the alloca address into a register, set the base type back to register and
   // continue. This should almost never happen.
diff --git a/test/CodeGen/AArch64/fast-isel-addressing-modes.ll b/test/CodeGen/AArch64/fast-isel-addressing-modes.ll
index 86ba400cff2..750e081d423 100644
--- a/test/CodeGen/AArch64/fast-isel-addressing-modes.ll
+++ b/test/CodeGen/AArch64/fast-isel-addressing-modes.ll
@@ -107,6 +107,16 @@ define void @store_breg_f64(double* %a) {
   ret void
 }
 
+; Load Immediate
+define i32 @load_immoff_1() {
+; CHECK-LABEL: load_immoff_1
+; CHECK:       orr {{w|x}}[[REG:[0-9]+]], {{wzr|xzr}}, #0x80
+; CHECK:       ldr {{w[0-9]+}}, {{\[}}x[[REG]]{{\]}}
+  %1 = inttoptr i64 128 to i32*
+  %2 = load i32* %1
+  ret i32 %2
+}
+
 ; Load / Store Base Register + Immediate Offset
 ; Max supported negative offset
 define i32 @load_breg_immoff_1(i64 %a) {
@@ -318,6 +328,17 @@ define i64 @load_breg_offreg_immoff_2(i64 %a, i64 %b) {
   ret i64 %4
 }
 
+; Load Scaled Register Offset
+define i32 @load_shift_offreg_1(i64 %a) {
+; CHECK-LABEL: load_shift_offreg_1
+; CHECK:       lsl [[REG:x[0-9]+]], x0, #2
+; CHECK:       ldr {{w[0-9]+}}, {{\[}}[[REG]]{{\]}}
+  %1 = shl i64 %a, 2
+  %2 = inttoptr i64 %1 to i32*
+  %3 = load i32* %2
+  ret i32 %3
+}
+
 ; Load Base Register + Scaled Register Offset
 define i32 @load_breg_shift_offreg_1(i64 %a, i64 %b) {
 ; CHECK-LABEL: load_breg_shift_offreg_1