From 86eb22976f590b37ea16170b2d10b2425fc39661 Mon Sep 17 00:00:00 2001
From: Evan Cheng <evan.cheng@apple.com>
Date: Fri, 7 May 2010 02:04:02 +0000
Subject: [PATCH] Use VLD2q32 / VST2q32 to reload / spill QQ (pair of Q)
 registers when stack slot is sufficiently aligned. Use VLDMD / VSTMD
 otherwise.

llvm-svn: 103235
---
 llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp | 44 +++++++++++++++++++++---
 llvm/lib/Target/ARM/ARMInstrNEON.td      | 10 ------
 2 files changed, 39 insertions(+), 15 deletions(-)

diff --git a/llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp b/llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp
index eff2008e6139..a37e4c67d552 100644
--- a/llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp
+++ b/llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp
@@ -788,14 +788,31 @@ storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
                        .addImm(ARM_AM::getAM5Opc(ARM_AM::ia, 4)))
         .addMemOperand(MMO);
       MIB = AddDReg(MIB, SrcReg, ARM::DSUBREG_0, getKillRegState(isKill), TRI);
-      AddDReg(MIB, SrcReg, ARM::DSUBREG_1, 0, TRI);
-        
+      AddDReg(MIB, SrcReg, ARM::DSUBREG_1, 0, TRI);        
     }
   } else {
     assert((RC == ARM::QQPRRegisterClass ||
             RC == ARM::QQPR_VFP2RegisterClass ||
             RC == ARM::QQPR_8RegisterClass) && "Unknown regclass!");
-    llvm_unreachable("Not yet implemented!");
+    if (Align >= 16 && getRegisterInfo().canRealignStack(MF)) {
+      MachineInstrBuilder MIB = BuildMI(MBB, I, DL, get(ARM::VST2q32))
+        .addFrameIndex(FI).addImm(128);
+      MIB = AddDReg(MIB, SrcReg, ARM::DSUBREG_0, getKillRegState(isKill), TRI);
+      MIB = AddDReg(MIB, SrcReg, ARM::DSUBREG_1, 0, TRI);
+      MIB = AddDReg(MIB, SrcReg, ARM::DSUBREG_2, 0, TRI);
+      MIB = AddDReg(MIB, SrcReg, ARM::DSUBREG_3, 0, TRI);
+      AddDefaultPred(MIB.addMemOperand(MMO));
+    } else {
+      MachineInstrBuilder MIB =
+        AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTMD))
+                       .addFrameIndex(FI)
+                       .addImm(ARM_AM::getAM5Opc(ARM_AM::ia, 4)))
+        .addMemOperand(MMO);
+      MIB = AddDReg(MIB, SrcReg, ARM::DSUBREG_0, getKillRegState(isKill), TRI);
+      MIB = AddDReg(MIB, SrcReg, ARM::DSUBREG_1, 0, TRI);
+      MIB = AddDReg(MIB, SrcReg, ARM::DSUBREG_2, 0, TRI);
+            AddDReg(MIB, SrcReg, ARM::DSUBREG_3, 0, TRI);
+    }
   }
 }
 
@@ -847,13 +864,30 @@ loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
                        .addImm(ARM_AM::getAM5Opc(ARM_AM::ia, 4)))
         .addMemOperand(MMO);
       MIB = AddDReg(MIB, DestReg, ARM::DSUBREG_0, RegState::Define, TRI);
-      AddDReg(MIB, DestReg, ARM::DSUBREG_1, RegState::Define, TRI);
+            AddDReg(MIB, DestReg, ARM::DSUBREG_1, RegState::Define, TRI);
     }
   } else {
     assert((RC == ARM::QQPRRegisterClass ||
             RC == ARM::QQPR_VFP2RegisterClass ||
             RC == ARM::QQPR_8RegisterClass) && "Unknown regclass!");
-    llvm_unreachable("Not yet implemented!");
+    if (Align >= 16 && getRegisterInfo().canRealignStack(MF)) {
+      MachineInstrBuilder MIB = BuildMI(MBB, I, DL, get(ARM::VLD2q32));
+      MIB = AddDReg(MIB, DestReg, ARM::DSUBREG_0, RegState::Define, TRI);
+      MIB = AddDReg(MIB, DestReg, ARM::DSUBREG_1, RegState::Define, TRI);
+      MIB = AddDReg(MIB, DestReg, ARM::DSUBREG_2, RegState::Define, TRI);
+      MIB = AddDReg(MIB, DestReg, ARM::DSUBREG_3, RegState::Define, TRI);
+      AddDefaultPred(MIB.addFrameIndex(FI).addImm(128).addMemOperand(MMO));
+    } else {
+      MachineInstrBuilder MIB =
+        AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDMD))
+                       .addFrameIndex(FI)
+                       .addImm(ARM_AM::getAM5Opc(ARM_AM::ia, 4)))
+        .addMemOperand(MMO);
+      MIB = AddDReg(MIB, DestReg, ARM::DSUBREG_0, RegState::Define, TRI);
+      MIB = AddDReg(MIB, DestReg, ARM::DSUBREG_1, RegState::Define, TRI);
+      MIB = AddDReg(MIB, DestReg, ARM::DSUBREG_2, RegState::Define, TRI);
+            AddDReg(MIB, DestReg, ARM::DSUBREG_3, RegState::Define, TRI);
+    }
   }
 }
 
diff --git a/llvm/lib/Target/ARM/ARMInstrNEON.td b/llvm/lib/Target/ARM/ARMInstrNEON.td
index af401ffdef46..df3c53a4df15 100644
--- a/llvm/lib/Target/ARM/ARMInstrNEON.td
+++ b/llvm/lib/Target/ARM/ARMInstrNEON.td
@@ -123,11 +123,6 @@ def VLDMQ
   : AXDI5<(outs QPR:$dst), (ins addrmode5:$addr, pred:$p),
           IndexModeNone, IIC_fpLoadm,
           "vldm${addr:submode}${p}\t${addr:base}, ${dst:dregpair}", "", []>;
-def VLDMQ_UPD
-  : AXDI5<(outs QPR:$dst, GPR:$wb), (ins addrmode5:$addr, pred:$p),
-          IndexModeUpd, IIC_fpLoadm,
-          "vldm${addr:submode}${p}\t${addr:base}!, ${dst:dregpair}",
-          "$addr.base = $wb", []>;
 } // mayLoad = 1
 
 let mayStore = 1 in {
@@ -138,11 +133,6 @@ def VSTMQ
   : AXDI5<(outs), (ins QPR:$src, addrmode5:$addr, pred:$p),
           IndexModeNone, IIC_fpStorem,
           "vstm${addr:submode}${p}\t${addr:base}, ${src:dregpair}", "", []>;
-def VSTMQ_UPD
-  : AXDI5<(outs GPR:$wb), (ins QPR:$src, addrmode5:$addr, pred:$p),
-          IndexModeUpd, IIC_fpStorem,
-          "vstm${addr:submode}${p}\t${addr:base}!, ${src:dregpair}",
-          "$addr.base = $wb", []>;
 } // mayStore = 1
 
 let mayLoad = 1, hasExtraDefRegAllocReq = 1 in {
-- 
GitLab