From 22b557d94277947d88029ed20a89381842ec22b9 Mon Sep 17 00:00:00 2001
From: Juergen Ributzka <juergen@apple.com>
Date: Thu, 18 Sep 2014 05:40:47 +0000
Subject: [PATCH] [FastISel][AArch64] Try to fold the offset into the add
 instruction when simplifying a memory address.

Small optimization in 'simplifyAddress'. When the offset cannot be encoded in
the load/store instruction, then we need to materialize the address manually.
The add instruction can encode a wider range of immediates than the load/store
instructions. This change tries to fold the offset into the add instruction
first before materializing the offset in a register.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218031 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/AArch64/AArch64FastISel.cpp        | 14 +++++--
 .../AArch64/fast-isel-addressing-modes.ll     | 38 +++++++------------
 2 files changed, 23 insertions(+), 29 deletions(-)

diff --git a/lib/Target/AArch64/AArch64FastISel.cpp b/lib/Target/AArch64/AArch64FastISel.cpp
index 0e9ef990962..ff56209d2ff 100644
--- a/lib/Target/AArch64/AArch64FastISel.cpp
+++ b/lib/Target/AArch64/AArch64FastISel.cpp
@@ -918,10 +918,16 @@ bool AArch64FastISel::simplifyAddress(Address &Addr, MVT VT) {
   // reg+offset into a register.
   if (ImmediateOffsetNeedsLowering) {
     unsigned ResultReg = 0;
-    if (Addr.getReg())
-      ResultReg = fastEmit_ri_(MVT::i64, ISD::ADD, Addr.getReg(),
-                               /*IsKill=*/false, Offset, MVT::i64);
-    else
+    if (Addr.getReg()) {
+      // Try to fold the immediate into the add instruction.
+      ResultReg = emitAddSub_ri(/*UseAdd=*/true, MVT::i64, Addr.getReg(),
+                                /*IsKill=*/false, Offset);
+      if (!ResultReg) {
+        unsigned ImmReg = fastEmit_i(MVT::i64, MVT::i64, ISD::Constant, Offset);
+        ResultReg = emitAddSub_rr(/*UseAdd=*/true, MVT::i64, Addr.getReg(),
+                                  /*IsKill=*/false, ImmReg, /*IsKill=*/true);
+      }
+    } else
       ResultReg = fastEmit_i(MVT::i64, MVT::i64, ISD::Constant, Offset);
 
     if (!ResultReg)
diff --git a/test/CodeGen/AArch64/fast-isel-addressing-modes.ll b/test/CodeGen/AArch64/fast-isel-addressing-modes.ll
index b33c06ee278..88a845821f2 100644
--- a/test/CodeGen/AArch64/fast-isel-addressing-modes.ll
+++ b/test/CodeGen/AArch64/fast-isel-addressing-modes.ll
@@ -154,12 +154,9 @@ define i32 @load_breg_immoff_3(i64 %a) {
 
 ; Min un-supported unscaled offset
 define i32 @load_breg_immoff_4(i64 %a) {
-; SDAG-LABEL: load_breg_immoff_4
-; SDAG:       add [[REG:x[0-9]+]], x0, #257
-; SDAG-NEXT:  ldr {{w[0-9]+}}, {{\[}}[[REG]]{{\]}}
-; FAST-LABEL: load_breg_immoff_4
-; FAST:       add [[REG:x[0-9]+]], x0, {{x[0-9]+}}
-; FAST-NEXT:  ldr {{w[0-9]+}}, {{\[}}[[REG]]{{\]}}
+; CHECK-LABEL: load_breg_immoff_4
+; CHECK:       add [[REG:x[0-9]+]], x0, #257
+; CHECK-NEXT:  ldr {{w[0-9]+}}, {{\[}}[[REG]]{{\]}}
   %1 = add i64 %a, 257
   %2 = inttoptr i64 %1 to i32*
   %3 = load i32* %2
@@ -178,12 +175,9 @@ define i32 @load_breg_immoff_5(i64 %a) {
 
 ; Min un-supported scaled offset
 define i32 @load_breg_immoff_6(i64 %a) {
-; SDAG-LABEL: load_breg_immoff_6
-; SDAG:       add [[REG:x[0-9]+]], x0, #4, lsl #12
-; SDAG-NEXT:  ldr {{w[0-9]+}}, {{\[}}[[REG]]{{\]}}
-; FAST-LABEL: load_breg_immoff_6
-; FAST:       add [[REG:x[0-9]+]], x0, {{x[0-9]+}}
-; FAST-NEXT:  ldr {{w[0-9]+}}, {{\[}}[[REG]]{{\]}}
+; CHECK-LABEL: load_breg_immoff_6
+; CHECK:       add [[REG:x[0-9]+]], x0, #4, lsl #12
+; CHECK-NEXT:  ldr {{w[0-9]+}}, {{\[}}[[REG]]{{\]}}
   %1 = add i64 %a, 16384
   %2 = inttoptr i64 %1 to i32*
   %3 = load i32* %2
@@ -226,12 +220,9 @@ define void @store_breg_immoff_3(i64 %a) {
 
 ; Min un-supported unscaled offset
 define void @store_breg_immoff_4(i64 %a) {
-; SDAG-LABEL: store_breg_immoff_4
-; SDAG:       add [[REG:x[0-9]+]], x0, #257
-; SDAG-NEXT:  str wzr, {{\[}}[[REG]]{{\]}}
-; FAST-LABEL: store_breg_immoff_4
-; FAST:       add [[REG:x[0-9]+]], x0, {{x[0-9]+}}
-; FAST-NEXT:  str wzr, {{\[}}[[REG]]{{\]}}
+; CHECK-LABEL: store_breg_immoff_4
+; CHECK:       add [[REG:x[0-9]+]], x0, #257
+; CHECK-NEXT:  str wzr, {{\[}}[[REG]]{{\]}}
   %1 = add i64 %a, 257
   %2 = inttoptr i64 %1 to i32*
   store i32 0, i32* %2
@@ -250,12 +241,9 @@ define void @store_breg_immoff_5(i64 %a) {
 
 ; Min un-supported scaled offset
 define void @store_breg_immoff_6(i64 %a) {
-; SDAG-LABEL: store_breg_immoff_6
-; SDAG:       add [[REG:x[0-9]+]], x0, #4, lsl #12
-; SDAG-NEXT:  str wzr, {{\[}}[[REG]]{{\]}}
-; FAST-LABEL: store_breg_immoff_6
-; FAST:       add [[REG:x[0-9]+]], x0, {{x[0-9]+}}
-; FAST-NEXT:  str wzr, {{\[}}[[REG]]{{\]}}
+; CHECK-LABEL: store_breg_immoff_6
+; CHECK:       add [[REG:x[0-9]+]], x0, #4, lsl #12
+; CHECK-NEXT:  str wzr, {{\[}}[[REG]]{{\]}}
   %1 = add i64 %a, 16384
   %2 = inttoptr i64 %1 to i32*
   store i32 0, i32* %2
@@ -319,7 +307,7 @@ define i64 @load_breg_offreg_immoff_2(i64 %a, i64 %b) {
 ; SDAG-NEXT:  add [[REG2:x[0-9]+]], [[REG1]], #15, lsl #12
 ; SDAG-NEXT:  ldr x0, {{\[}}[[REG2]]{{\]}}
 ; FAST-LABEL: load_breg_offreg_immoff_2
-; FAST:       add [[REG:x[0-9]+]], x0, {{x[0-9]+}}
+; FAST:       add [[REG:x[0-9]+]], x0, #15, lsl #12
 ; FAST-NEXT:  ldr x0, {{\[}}[[REG]], x1{{\]}}
   %1 = add i64 %a, %b
   %2 = add i64 %1, 61440