Support unaligned integer loads on ARM and MIPS.

runtime/vm/intermediate_language_arm64.cc

 // Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
 #include "vm/globals.h"  // Needed here to get TARGET_ARCH_ARM64.
 #if defined(TARGET_ARCH_ARM64)
 
 #include "vm/intermediate_language.h"
 
 #include "vm/compiler.h"
@@ skipping 1025 matching lines @@
   }
   return Address::CanHoldOffset(static_cast<int32_t>(offset),
                                 Address::Offset,
                                 Address::OperandSizeFor(cid));
 }
 
 
 LocationSummary* LoadIndexedInstr::MakeLocationSummary(Zone* zone,
                                                        bool opt) const {
   const intptr_t kNumInputs = 2;
-  const intptr_t kNumTemps = 0;
+  const intptr_t kNumTemps = aligned() ? 0 : 1;
   LocationSummary* locs = new(zone) LocationSummary(
       zone, kNumInputs, kNumTemps, LocationSummary::kNoCall);
   locs->set_in(0, Location::RequiresRegister());
   if (CanBeImmediateIndex(index(), class_id(), IsExternal())) {
     locs->set_in(1, Location::Constant(index()->definition()->AsConstant()));
   } else {
     locs->set_in(1, Location::RequiresRegister());
   }
   if ((representation() == kUnboxedDouble)    ||
       (representation() == kUnboxedFloat32x4) ||
       (representation() == kUnboxedInt32x4)   ||
       (representation() == kUnboxedFloat64x2)) {
     locs->set_out(0, Location::RequiresFpuRegister());
   } else {
     locs->set_out(0, Location::RequiresRegister());
   }
+  if (!aligned()) {
+    locs->set_temp(0, Location::RequiresRegister());
+  }
   return locs;
 }
 
 
 void LoadIndexedInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
   // The array register points to the backing store for external arrays.
   const Register array = locs()->in(0).reg();
   const Location index = locs()->in(1);
+  const Register address = aligned() ? kNoRegister : locs()->temp(0).reg();
 
-  Address element_address = index.IsRegister()
-      ? __ ElementAddressForRegIndex(true,  // Load.
-                                     IsExternal(), class_id(), index_scale(),
-                                     array, index.reg())
-      : __ ElementAddressForIntIndex(
-            IsExternal(), class_id(), index_scale(),
-            array, Smi::Cast(index.constant()).Value());
-  // Warning: element_address may use register TMP as base.
+  Address element_address(TMP);

[zra, 2016/10/26 06:59:41]

ditto

+  if (aligned()) {
+    element_address = index.IsRegister()
+        ? __ ElementAddressForRegIndex(true,  // Load.
+                                       IsExternal(), class_id(), index_scale(),
+                                       array, index.reg())
+        : __ ElementAddressForIntIndex(
+              IsExternal(), class_id(), index_scale(),
+              array, Smi::Cast(index.constant()).Value());
+    // Warning: element_address may use register TMP as base.
+  } else {
+    if (index.IsRegister()) {
+      __ LoadElementAddressForRegIndex(address,
+                                       true,  // Load.
+                                       IsExternal(), class_id(), index_scale(),
+                                       array, index.reg());
+    } else {
+      __ LoadElementAddressForIntIndex(address,
+                                       IsExternal(), class_id(), index_scale(),
+                                       array,
+                                       Smi::Cast(index.constant()).Value());
+    }
+  }
 
   if ((representation() == kUnboxedDouble)    ||
       (representation() == kUnboxedFloat32x4) ||
       (representation() == kUnboxedInt32x4)   ||
       (representation() == kUnboxedFloat64x2)) {
     const VRegister result = locs()->out(0).fpu_reg();
     switch (class_id()) {
+      ASSERT(aligned());
       case kTypedDataFloat32ArrayCid:
         // Load single precision float.
         __ fldrs(result, element_address);
         break;
       case kTypedDataFloat64ArrayCid:
         // Load double precision float.
         __ fldrd(result, element_address);
         break;
       case kTypedDataFloat64x2ArrayCid:
       case kTypedDataInt32x4ArrayCid:
       case kTypedDataFloat32x4ArrayCid:
         __ fldrq(result, element_address);
         break;
       default:
         UNREACHABLE();
     }
     return;
   }
 
   if ((representation() == kUnboxedInt32) ||
       (representation() == kUnboxedUint32)) {
     const Register result = locs()->out(0).reg();
     switch (class_id()) {
       case kTypedDataInt32ArrayCid:
         ASSERT(representation() == kUnboxedInt32);
-        __ ldr(result, element_address, kWord);
+        if (aligned()) {
+          __ ldr(result, element_address, kWord);
+        } else {
+          __ LoadUnaligned(result, address, TMP, kWord);
+        }
         break;
       case kTypedDataUint32ArrayCid:
         ASSERT(representation() == kUnboxedUint32);
-        __ ldr(result, element_address, kUnsignedWord);
+        if (aligned()) {
+          __ ldr(result, element_address, kUnsignedWord);
+        } else {
+          __ LoadUnaligned(result, address, TMP, kUnsignedWord);
+        }
         break;
       default:
         UNREACHABLE();
       }
     return;
   }
 
   ASSERT(representation() == kTagged);
   const Register result = locs()->out(0).reg();
   switch (class_id()) {
     case kTypedDataInt8ArrayCid:
       ASSERT(index_scale() == 1);
       __ ldr(result, element_address, kByte);
       __ SmiTag(result);
       break;
     case kTypedDataUint8ArrayCid:
     case kTypedDataUint8ClampedArrayCid:
     case kExternalTypedDataUint8ArrayCid:
     case kExternalTypedDataUint8ClampedArrayCid:
     case kOneByteStringCid:
     case kExternalOneByteStringCid:
       ASSERT(index_scale() == 1);
       __ ldr(result, element_address, kUnsignedByte);
       __ SmiTag(result);
       break;
     case kTypedDataInt16ArrayCid:
-      __ ldr(result, element_address, kHalfword);
+      if (aligned()) {
+        __ ldr(result, element_address, kHalfword);
+      } else {
+        __ LoadUnaligned(result, address, TMP, kHalfword);
+      }
       __ SmiTag(result);
       break;
     case kTypedDataUint16ArrayCid:
     case kTwoByteStringCid:
     case kExternalTwoByteStringCid:
-      __ ldr(result, element_address, kUnsignedHalfword);
+      if (aligned()) {
+        __ ldr(result, element_address, kUnsignedHalfword);
+      } else {
+        __ LoadUnaligned(result, address, TMP, kUnsignedHalfword);
+      }
       __ SmiTag(result);
       break;
     default:
       ASSERT((class_id() == kArrayCid) || (class_id() == kImmutableArrayCid));
+      ASSERT(aligned());
       __ ldr(result, element_address);
       break;
   }
 }
 
 
 LocationSummary* LoadCodeUnitsInstr::MakeLocationSummary(Zone* zone,
                                                          bool opt) const {
   const intptr_t kNumInputs = 2;
   const intptr_t kNumTemps = 0;
@@ skipping 76 matching lines @@
     default:
       UNREACHABLE();
       return kTagged;
   }
 }
 
 
 LocationSummary* StoreIndexedInstr::MakeLocationSummary(Zone* zone,
                                                         bool opt) const {
   const intptr_t kNumInputs = 3;
-  const intptr_t kNumTemps = 0;
+  const intptr_t kNumTemps = aligned() ? 0 : 2;
   LocationSummary* locs = new(zone) LocationSummary(
       zone, kNumInputs, kNumTemps, LocationSummary::kNoCall);
   locs->set_in(0, Location::RequiresRegister());
   if (CanBeImmediateIndex(index(), class_id(), IsExternal())) {
     locs->set_in(1, Location::Constant(index()->definition()->AsConstant()));
   } else {
     locs->set_in(1, Location::WritableRegister());
   }
   switch (class_id()) {
     case kArrayCid:
@@ skipping 19 matching lines @@
       break;
     case kTypedDataInt32x4ArrayCid:
     case kTypedDataFloat32x4ArrayCid:
     case kTypedDataFloat64x2ArrayCid:
       locs->set_in(2, Location::RequiresFpuRegister());
       break;
     default:
       UNREACHABLE();
       return NULL;
   }
+  if (!aligned()) {
+    locs->set_temp(0, Location::RequiresRegister());
+    locs->set_temp(1, Location::RequiresRegister());
+  }
   return locs;
 }
 
 
 void StoreIndexedInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
   // The array register points to the backing store for external arrays.
   const Register array = locs()->in(0).reg();
   const Location index = locs()->in(1);
+  const Register address = aligned() ? kNoRegister : locs()->temp(0).reg();
+  const Register scratch = aligned() ? kNoRegister : locs()->temp(1).reg();
 
-  Address element_address = index.IsRegister()
+  Address element_address(TMP);

[zra, 2016/10/26 06:59:41]

ditto

+  if (aligned()) {
+    element_address = index.IsRegister()
       ? __ ElementAddressForRegIndex(false,  // Store.
                                      IsExternal(), class_id(), index_scale(),
                                      array, index.reg())
       : __ ElementAddressForIntIndex(
             IsExternal(), class_id(), index_scale(),
             array, Smi::Cast(index.constant()).Value());
+  } else {
+    if (index.IsRegister()) {
+      __ LoadElementAddressForRegIndex(address,
+                                       false,  // Store.
+                                       IsExternal(), class_id(), index_scale(),
+                                       array, index.reg());
+    } else {
+      __ LoadElementAddressForIntIndex(address,
+                                       IsExternal(), class_id(), index_scale(),
+                                       array,
+                                       Smi::Cast(index.constant()).Value());
+    }
+  }
 
   switch (class_id()) {
     case kArrayCid:
+      ASSERT(aligned());
       if (ShouldEmitStoreBarrier()) {
         const Register value = locs()->in(2).reg();
         __ StoreIntoObject(array, element_address, value);
       } else if (locs()->in(2).IsConstant()) {
         const Object& constant = locs()->in(2).constant();
         __ StoreIntoObjectNoBarrier(array, element_address, constant);
       } else {
         const Register value = locs()->in(2).reg();
         __ StoreIntoObjectNoBarrier(array, element_address, value);
       }
       break;
     case kTypedDataInt8ArrayCid:
     case kTypedDataUint8ArrayCid:
     case kExternalTypedDataUint8ArrayCid:
     case kOneByteStringCid: {
+      ASSERT(aligned());
       if (locs()->in(2).IsConstant()) {
         const Smi& constant = Smi::Cast(locs()->in(2).constant());
         __ LoadImmediate(TMP, static_cast<int8_t>(constant.Value()));
         __ str(TMP, element_address, kUnsignedByte);
       } else {
         const Register value = locs()->in(2).reg();
         __ SmiUntag(TMP, value);
         __ str(TMP, element_address, kUnsignedByte);
       }
       break;
     }
     case kTypedDataUint8ClampedArrayCid:
     case kExternalTypedDataUint8ClampedArrayCid: {
+      ASSERT(aligned());
       if (locs()->in(2).IsConstant()) {
         const Smi& constant = Smi::Cast(locs()->in(2).constant());
         intptr_t value = constant.Value();
         // Clamp to 0x0 or 0xFF respectively.
         if (value > 0xFF) {
           value = 0xFF;
         } else if (value < 0) {
           value = 0;
         }
         __ LoadImmediate(TMP, static_cast<int8_t>(value));
         __ str(TMP, element_address, kUnsignedByte);
       } else {
         const Register value = locs()->in(2).reg();
         __ CompareImmediate(value, 0x1FE);  // Smi value and smi 0xFF.
         // Clamp to 0x00 or 0xFF respectively.
         __ csetm(TMP, GT);  // TMP = value > 0x1FE ? -1 : 0.
         __ csel(TMP, value, TMP, LS);  // TMP = value in range ? value : TMP.
         __ SmiUntag(TMP);
         __ str(TMP, element_address, kUnsignedByte);
       }
       break;
     }
     case kTypedDataInt16ArrayCid:
     case kTypedDataUint16ArrayCid: {
       const Register value = locs()->in(2).reg();
       __ SmiUntag(TMP, value);
-      __ str(TMP, element_address, kUnsignedHalfword);
+      if (aligned()) {
+        __ str(TMP, element_address, kUnsignedHalfword);
+      } else {
+        __ StoreUnaligned(TMP, address, scratch, kUnsignedHalfword);
+      }
       break;
     }
     case kTypedDataInt32ArrayCid:
     case kTypedDataUint32ArrayCid: {
       const Register value = locs()->in(2).reg();
-      __ str(value, element_address, kUnsignedWord);
+      if (aligned()) {
+        __ str(value, element_address, kUnsignedWord);
+      } else {
+        __ StoreUnaligned(value, address, scratch, kUnsignedWord);
+      }
       break;
     }
     case kTypedDataFloat32ArrayCid: {
+      ASSERT(aligned());
       const VRegister value_reg = locs()->in(2).fpu_reg();
       __ fstrs(value_reg, element_address);
       break;
     }
     case kTypedDataFloat64ArrayCid: {
+      ASSERT(aligned());
       const VRegister value_reg = locs()->in(2).fpu_reg();
       __ fstrd(value_reg, element_address);
       break;
     }
     case kTypedDataFloat64x2ArrayCid:
     case kTypedDataInt32x4ArrayCid:
     case kTypedDataFloat32x4ArrayCid: {
+      ASSERT(aligned());
       const VRegister value_reg = locs()->in(2).fpu_reg();
       __ fstrq(value_reg, element_address);
       break;
     }
     default:
       UNREACHABLE();
   }
 }
 
 
@@ skipping 4420 matching lines @@
                                 1,
                                 locs());
   __ Drop(1);
   __ Pop(result);
 }
 
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_ARM64