ARM: Make use of d16-d31 when available.

src/arm/code-stubs-arm.cc

Index: src/arm/code-stubs-arm.cc
diff --git a/src/arm/code-stubs-arm.cc b/src/arm/code-stubs-arm.cc
index 521231c02ab7efb5db9720fde2630983492204e3..eb697e4a20164e76f8da48d7816410141d78b28e 100644
--- a/src/arm/code-stubs-arm.cc
+++ b/src/arm/code-stubs-arm.cc
@@ -2066,17 +2066,25 @@ void StoreBufferOverflowStub::Generate(MacroAssembler* masm) {
   // store the registers in any particular way, but we do have to store and
   // restore them.
   __ stm(db_w, sp, kCallerSaved | lr.bit());
+
+  const Register scratch = r1;
+
   if (save_doubles_ == kSaveFPRegs) {
     CpuFeatures::Scope scope(VFP2);
+    // Check CPU flags for number of registers.
+    // 16 regs => Z==1, 32 regs => Z==0.
+    __ mov(scratch, Operand(ExternalReference::cpu_features()));
+    __ ldr(scratch, MemOperand(scratch));
+    __ and_(scratch, scratch, Operand(1u << VFP32DREGS), SetCC);
+
     __ sub(sp, sp, Operand(kDoubleSize * DwVfpRegister::kNumRegisters));
     for (int i = 0; i < DwVfpRegister::kNumRegisters; i++) {
       DwVfpRegister reg = DwVfpRegister::from_code(i);
-      __ vstr(reg, MemOperand(sp, i * kDoubleSize));
+      __ vstr(reg, MemOperand(sp, i * kDoubleSize), i < 16 ? al : ne);
     }
   }
   const int argument_count = 1;
   const int fp_argument_count = 0;
-  const Register scratch = r1;
 
   AllowExternalCallThatCantCauseGC scope(masm);
   __ PrepareCallCFunction(argument_count, fp_argument_count, scratch);
@@ -2086,9 +2094,16 @@ void StoreBufferOverflowStub::Generate(MacroAssembler* masm) {
       argument_count);
   if (save_doubles_ == kSaveFPRegs) {
     CpuFeatures::Scope scope(VFP2);
+
+    // Check CPU flags for number of registers.
+    // 16 regs => Z==1, 32 regs => Z==0.
+    __ mov(scratch, Operand(ExternalReference::cpu_features()));
+    __ ldr(scratch, MemOperand(scratch));
+    __ and_(scratch, scratch, Operand(1u << VFP32DREGS), SetCC);
+
     for (int i = 0; i < DwVfpRegister::kNumRegisters; i++) {
       DwVfpRegister reg = DwVfpRegister::from_code(i);
-      __ vldr(reg, MemOperand(sp, i * kDoubleSize));
+      __ vldr(reg, MemOperand(sp, i * kDoubleSize), i < 16 ? al : ne);
     }
     __ add(sp, sp, Operand(kDoubleSize * DwVfpRegister::kNumRegisters));
   }