[arm] Port native routines to use UseScratchRegisterScope

Make use of UseScratchRegisterScope instead of using the ip register directly in
code stubs, builtin and the deoptimizer. In a lot of cases, we can
simply use a different register rather than using the new scope.

Bug: v8:6553
Change-Id: Ibc8a9a78bb88f3850c6e8b45871cc3a5b3971b3b
Reviewed-on: https://chromium-review.googlesource.com/544837
Commit-Queue: Pierre Langlois <pierre.langlois@arm.com>
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#46429}

src/arm/codegen-arm.cc

@@ -143,7 +143,8 @@ MemCopyUint8Function CreateMemCopyUint8Function(Isolate* isolate,
     __ ldr(temp1, MemOperand(src, 4, PostIndex));
     __ str(temp1, MemOperand(dest, 4, PostIndex));
   } else {
-    Register temp2 = ip;
+    UseScratchRegisterScope temps(&masm);
+    Register temp2 = temps.Acquire();
     Label loop;
 
     __ bic(temp2, chars, Operand(0x3), SetCC);
@@ -219,8 +220,10 @@ MemCopyUint16Uint8Function CreateMemCopyUint16Uint8Function(
     __ vst1(Neon16, NeonListOperand(d0, 2), NeonMemOperand(dest));
     __ Ret();
   } else {
+    UseScratchRegisterScope temps(&masm);
+
     Register temp1 = r3;
-    Register temp2 = ip;
+    Register temp2 = temps.Acquire();
     Register temp3 = lr;
     Register temp4 = r4;
     Label loop;

src/arm/deoptimizer-arm.cc

@@ -111,9 +111,11 @@ void Deoptimizer::TableEntryGenerator::Generate() {
     // We use a run-time check for VFP32DREGS.
     CpuFeatureScope scope(masm(), VFP32DREGS,
                           CpuFeatureScope::kDontCheckSupported);
+    UseScratchRegisterScope temps(masm());
+    Register scratch = temps.Acquire();
 
     // Check CPU flags for number of registers, setting the Z condition flag.
-    __ CheckFor32DRegs(ip);
+    __ CheckFor32DRegs(scratch);
 
     // Push registers d0-d15, and possibly d16-d31, on the stack.
     // If d16-d31 are not pushed, decrease the stack pointer instead.
@@ -130,9 +132,13 @@ void Deoptimizer::TableEntryGenerator::Generate() {
   // handle this a bit differently.
   __ stm(db_w, sp, restored_regs  | sp.bit() | lr.bit() | pc.bit());
 
-  __ mov(ip, Operand(ExternalReference(IsolateAddressId::kCEntryFPAddress,
-                                       isolate())));
-  __ str(fp, MemOperand(ip));
+  {
+    UseScratchRegisterScope temps(masm());
+    Register scratch = temps.Acquire();
+    __ mov(scratch, Operand(ExternalReference(
+                        IsolateAddressId::kCEntryFPAddress, isolate())));
+    __ str(fp, MemOperand(scratch));
+  }
 
   const int kSavedRegistersAreaSize =
       (kNumberOfRegisters * kPointerSize) + kDoubleRegsSize + kFloatRegsSize;
@@ -294,15 +300,18 @@ void Deoptimizer::TableEntryGenerator::Generate() {
 
   // Restore the registers from the stack.
   __ ldm(ia_w, sp, restored_regs);  // all but pc registers.
-  __ pop(ip);  // remove sp
-  __ pop(ip);  // remove lr
 
   __ InitializeRootRegister();
 
-  __ pop(ip);  // remove pc
-  __ pop(ip);  // get continuation, leave pc on stack
-  __ pop(lr);
-  __ Jump(ip);
+  // Remove sp, lr and pc.
+  __ Drop(3);
+  {
+    UseScratchRegisterScope temps(masm());
+    Register scratch = temps.Acquire();
+    __ pop(scratch);  // get continuation, leave pc on stack
+    __ pop(lr);
+    __ Jump(scratch);
+  }
   __ stop("Unreachable.");
 }
 
@@ -315,13 +324,15 @@ void Deoptimizer::TableEntryGenerator::GeneratePrologue() {
   // ARMv7, we can use movw (with a maximum immediate of 0xffff). On ARMv6, we
   // need two instructions.
   STATIC_ASSERT((kMaxNumberOfEntries - 1) <= 0xffff);
+  UseScratchRegisterScope temps(masm());
+  Register scratch = temps.Acquire();
   if (CpuFeatures::IsSupported(ARMv7)) {
     CpuFeatureScope scope(masm(), ARMv7);
     Label done;
     for (int i = 0; i < count(); i++) {
       int start = masm()->pc_offset();
       USE(start);
-      __ movw(ip, i);
+      __ movw(scratch, i);
       __ b(&done);
       DCHECK_EQ(table_entry_size_, masm()->pc_offset() - start);
     }
@@ -337,14 +348,14 @@ void Deoptimizer::TableEntryGenerator::GeneratePrologue() {
     for (int i = 0; i < count(); i++) {
       int start = masm()->pc_offset();
       USE(start);
-      __ mov(ip, Operand(i & 0xff));  // Set the low byte.
+      __ mov(scratch, Operand(i & 0xff));  // Set the low byte.
       __ b(&high_fixes[i >> 8]);      // Jump to the secondary table.
       DCHECK_EQ(table_entry_size_, masm()->pc_offset() - start);
     }
     // Generate the secondary table, to set the high byte.
     for (int high = 1; high <= high_fix_max; high++) {
       __ bind(&high_fixes[high]);
-      __ orr(ip, ip, Operand(high << 8));
+      __ orr(scratch, scratch, Operand(high << 8));
       // If this isn't the last entry, emit a branch to the end of the table.
       // The last entry can just fall through.
       if (high < high_fix_max) __ b(&high_fixes[0]);
@@ -354,7 +365,7 @@ void Deoptimizer::TableEntryGenerator::GeneratePrologue() {
     // through with no additional branch.
     __ bind(&high_fixes[0]);
   }
-  __ push(ip);
+  __ push(scratch);
 }