[mips] Optimize stack pointer adjustments.

Instead of always using addu to adjust the stack pointer when the
size out is of the range of an addiu instruction, use subu so that
a smaller constant can be generated.

This can give savings of ~3 instructions whenever a function has a
a stack frame whose size is out of range of an addiu instruction.

This change may break some naive stack unwinders.

Partially resolves PR/26291.

Thanks to David Chisnall for reporting the issue.

Reviewers: dsanders, vkalintiris

Differential Review: http://reviews.llvm.org/D21321

llvm-svn: 272666

llvm/lib/Target/Mips/MCTargetDesc/MipsABIInfo.cpp

@@ -120,6 +120,10 @@ unsigned MipsABIInfo::GetPtrAddiuOp() const {
   return ArePtrs64bit() ? Mips::DADDiu : Mips::ADDiu;
 }
 
+unsigned MipsABIInfo::GetPtrSubuOp() const {
+  return ArePtrs64bit() ? Mips::DSUBu : Mips::SUBu;
+}
+
 unsigned MipsABIInfo::GetPtrAndOp() const {
   return ArePtrs64bit() ? Mips::AND64 : Mips::AND;
 }

llvm/lib/Target/Mips/MCTargetDesc/MipsABIInfo.h

@@ -69,6 +69,7 @@ public:
   unsigned GetZeroReg() const;
   unsigned GetPtrAdduOp() const;
   unsigned GetPtrAddiuOp() const;
+  unsigned GetPtrSubuOp() const;
   unsigned GetPtrAndOp() const;
   unsigned GetGPRMoveOp() const;
   inline bool ArePtrs64bit() const { return IsN64(); }

llvm/lib/Target/Mips/MipsSEInstrInfo.cpp

@@ -20,6 +20,7 @@
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/MathExtras.h"
 #include "llvm/Support/TargetRegistry.h"
 
 using namespace llvm;
@@ -440,17 +441,24 @@ void MipsSEInstrInfo::adjustStackPtr(unsigned SP, int64_t Amount,
                                      MachineBasicBlock::iterator I) const {
   MipsABIInfo ABI = Subtarget.getABI();
   DebugLoc DL;
-  unsigned ADDu = ABI.GetPtrAdduOp();
   unsigned ADDiu = ABI.GetPtrAddiuOp();
 
   if (Amount == 0)
     return;
 
-  if (isInt<16>(Amount))// addi sp, sp, amount
+  if (isInt<16>(Amount)) {
+    // addi sp, sp, amount
     BuildMI(MBB, I, DL, get(ADDiu), SP).addReg(SP).addImm(Amount);
-  else { // Expand immediate that doesn't fit in 16-bit.
+  } else {
+    // For numbers which are not 16bit integers we synthesize Amount inline
+    // then add or subtract it from sp.
+    unsigned Opc = ABI.GetPtrAdduOp();
+    if (Amount < 0) {
+      Opc = ABI.GetPtrSubuOp();
+      Amount = -Amount;
+    }
     unsigned Reg = loadImmediate(Amount, MBB, I, DL, nullptr);
-    BuildMI(MBB, I, DL, get(ADDu), SP).addReg(SP).addReg(Reg, RegState::Kill);
+    BuildMI(MBB, I, DL, get(Opc), SP).addReg(SP).addReg(Reg, RegState::Kill);
   }
 }
 

llvm/test/CodeGen/Mips/eh-dwarf-cfa.ll

@@ -13,6 +13,8 @@ entry:
   %0 = call i8* @llvm.eh.dwarf.cfa(i32 0)
   ret i8* %0
 
+; CHECK-LABEL: f1:
+
 ; CHECK:        addiu   $sp, $sp, -32
 ; CHECK:        addiu   $2, $sp, 32
 }
@@ -24,10 +26,12 @@ entry:
   %0 = call i8* @llvm.eh.dwarf.cfa(i32 0)
   ret i8* %0
 
+; CHECK-LABEL: f2:
+
 ; check stack size (65536 + 8)
-; CHECK:        lui     $[[R0:[a-z0-9]+]], 65535
-; CHECK:        addiu   $[[R0]], $[[R0]], -8
-; CHECK:        addu    $sp, $sp, $[[R0]]
+; CHECK:        lui     $[[R0:[a-z0-9]+]], 1
+; CHECK:        addiu   $[[R0]], $[[R0]], 8
+; CHECK:        subu    $sp, $sp, $[[R0]]
 
 ; check return value ($sp + stack size)
 ; CHECK:        lui     $[[R1:[a-z0-9]+]], 1
@@ -46,6 +50,8 @@ entry:
   %add = add i32 %1, %3
   ret i32 %add
 
+; CHECK-LABEL: f3:
+
 ; CHECK:        addiu   $sp, $sp, -40
 
 ; check return value ($fp + stack size + $fp)
@@ -60,6 +66,8 @@ entry:
   %0 = call i8* @llvm.eh.dwarf.cfa(i32 0)
   ret i8* %0
 
+; CHECK-LABEL: f4:
+
 ; CHECK-MIPS64:        daddiu   $sp, $sp, -32
 ; CHECK-MIPS64:        daddiu   $2, $sp, 32
 }

llvm/test/CodeGen/Mips/largeimm1.ll

@@ -1,13 +1,19 @@
 ; RUN: llc -march=mipsel -relocation-model=pic < %s | FileCheck %s
 
-; CHECK: lui ${{[0-9]+}}, 49152
-; CHECK: lui ${{[0-9]+}}, 16384
 define void @f() nounwind {
 entry:
   %a1 = alloca [1073741824 x i8], align 1
   %arrayidx = getelementptr inbounds [1073741824 x i8], [1073741824 x i8]* %a1, i32 0, i32 1048676
   call void @f2(i8* %arrayidx) nounwind
   ret void
+; CHECK-LABEL: f:
+
+; CHECK: lui    $[[R0:[a-z0-9]+]], 16384
+; CHECK: addiu  $[[R1:[a-z0-9]+]], $[[R0]], 24
+; CHECK: subu   $sp, $sp, $[[R1]]
+
+; CHECK: lui    $[[R2:[a-z0-9]+]], 16384
+; CHECK: addu   ${{[0-9]+}}, $sp, $[[R2]]
 }
 
 declare void @f2(i8*)

llvm/test/CodeGen/Mips/largeimmprinting.ll

@@ -10,21 +10,19 @@
 
 define void @f() nounwind {
 entry:
-; 32:  lui $[[R0:[0-9]+]], 65535
-; 32:  addiu $[[R0]], $[[R0]], -24
-; 32:  addu $sp, $sp, $[[R0]]
-; 32:  lui $[[R1:[0-9]+]], 1
-; 32:  addu $[[R1]], $sp, $[[R1]]
-; 32:  sw $ra, 20($[[R1]])
-; 64:  daddiu  $[[R0:[0-9]+]], $zero, 1
-; 64:  dsll  $[[R0]], $[[R0]], 48
-; 64:  daddiu  $[[R0]], $[[R0]], -1
-; 64:  dsll  $[[R0]], $[[R0]], 16
-; 64:  daddiu  $[[R0]], $[[R0]], -32
-; 64:  daddu $sp, $sp, $[[R0]]
-; 64:  lui $[[R1:[0-9]+]], 1
-; 64:  daddu $[[R1]], $sp, $[[R1]]
-; 64:  sd  $ra, 24($[[R1]])
+; 32:  lui     $[[R0:[0-9]+]], 1
+; 32:  addiu   $[[R0]], $[[R0]], 24
+; 32:  subu    $sp, $sp, $[[R0]]
+; 32:  lui     $[[R1:[0-9]+]], 1
+; 32:  addu    $[[R1]], $sp, $[[R1]]
+; 32:  sw      $ra, 20($[[R1]])
+
+; 64:  lui     $[[R0:[0-9]+]], 1
+; 64:  daddiu  $[[R0]], $[[R0]], 32
+; 64:  dsubu   $sp, $sp, $[[R0]]
+; 64:  lui     $[[R1:[0-9]+]], 1
+; 64:  daddu   $[[R1]], $sp, $[[R1]]
+; 64:  sd      $ra, 24($[[R1]])
 
   %agg.tmp = alloca %struct.S1, align 1
   %tmp = getelementptr inbounds %struct.S1, %struct.S1* %agg.tmp, i32 0, i32 0, i32 0