Support unaligned integer loads on ARM and MIPS.

runtime/vm/intermediate_language_arm.cc

 // Copyright (c) 2013, 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_ARM.
 #if defined(TARGET_ARCH_ARM)
 
 #include "vm/intermediate_language.h"
 
 #include "vm/compiler.h"
@@ skipping 1190 matching lines @@
     return true;
   }
 
   return false;
 }
 
 
 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());
   bool needs_base = false;
   if (CanBeImmediateIndex(index(), class_id(), IsExternal(),
                           true,  // Load.
                           &needs_base)) {
     // CanBeImmediateIndex must return false for unsafe smis.
     locs->set_in(1, Location::Constant(index()->definition()->AsConstant()));
   } else {
@@ skipping 14 matching lines @@
   } else if (representation() == kUnboxedUint32) {
     ASSERT(class_id() == kTypedDataUint32ArrayCid);
     locs->set_out(0, Location::RequiresRegister());
   } else if (representation() == kUnboxedInt32) {
     ASSERT(class_id() == kTypedDataInt32ArrayCid);
     locs->set_out(0, Location::RequiresRegister());
   } else {
     ASSERT(representation() == kTagged);
     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()
+  Address element_address(IP);

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

Can IP be kNoRegister instead, or maybe pass a bogus offset that the aligned()
case will overwrite? That way we can assert in the aligned cases below that
element_address wasn't used for something by mistake.

[rmacnak, 2016/10/31 22:57:26]

Using kNoRegister.

+  if (aligned()) {
+    element_address = index.IsRegister()
       ? __ ElementAddressForRegIndex(true,  // Load.
                                      IsExternal(), class_id(), index_scale(),
                                      array,
                                      index.reg())
       : __ ElementAddressForIntIndex(true,  // Load.
                                      IsExternal(), class_id(), index_scale(),
                                      array, Smi::Cast(index.constant()).Value(),
                                      IP);  // Temp register.
-  // Warning: element_address may use register IP as base.
+    // Warning: element_address may use register IP as base.
+  } else {
+    if (index.IsRegister()) {
+      __ LoadElementAddressForRegIndex(address,
+                                       true,  // Load.
+                                       IsExternal(), class_id(), index_scale(),
+                                       array,
+                                       index.reg());
+    } else {
+      __ LoadElementAddressForIntIndex(address,
+                                       true,  // Load.
+                                       IsExternal(), class_id(), index_scale(),
+                                       array,
+                                       Smi::Cast(index.constant()).Value());
+    }
+  }
 
   if ((representation() == kUnboxedDouble)    ||
       (representation() == kUnboxedFloat32x4) ||
       (representation() == kUnboxedInt32x4)   ||
       (representation() == kUnboxedFloat64x2)) {
     const QRegister result = locs()->out(0).fpu_reg();
     const DRegister dresult0 = EvenDRegisterOf(result);
     switch (class_id()) {
       case kTypedDataFloat32ArrayCid:
         // Load single precision float.
@@ skipping 15 matching lines @@
     }
     return;
   }
 
   if ((representation() == kUnboxedUint32) ||
       (representation() == kUnboxedInt32)) {
     Register result = locs()->out(0).reg();
     switch (class_id()) {
       case kTypedDataInt32ArrayCid:
         ASSERT(representation() == kUnboxedInt32);
-        __ ldr(result, element_address);
+        if (aligned()) {
+          __ ldr(result, element_address);

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

Likewise, maybe assert here that address == kNoRegister, or above before we
start doing the loads, something like:

ASSERT(aligned() ? (address == kNoRegister) : (element_address.something() ==
something));

+        } else {
+          __ LoadWordUnaligned(result, address, TMP);
+        }
       break;
       case kTypedDataUint32ArrayCid:
         ASSERT(representation() == kUnboxedUint32);
-        __ ldr(result, element_address);
+        if (aligned()) {
+          __ ldr(result, element_address);
+        } else {
+          __ LoadWordUnaligned(result, address, TMP);
+        }
       break;
       default:
         UNREACHABLE();
     }
     return;
   }
 
   ASSERT(representation() == kTagged);
 
   const Register result = locs()->out(0).reg();
   switch (class_id()) {
     case kTypedDataInt8ArrayCid:
       ASSERT(index_scale() == 1);
+      ASSERT(aligned());
       __ ldrsb(result, element_address);
       __ SmiTag(result);
       break;
     case kTypedDataUint8ArrayCid:
     case kTypedDataUint8ClampedArrayCid:
     case kExternalTypedDataUint8ArrayCid:
     case kExternalTypedDataUint8ClampedArrayCid:
     case kOneByteStringCid:
     case kExternalOneByteStringCid:
       ASSERT(index_scale() == 1);
+      ASSERT(aligned());
       __ ldrb(result, element_address);
       __ SmiTag(result);
       break;
     case kTypedDataInt16ArrayCid:
-      __ ldrsh(result, element_address);
+      if (aligned()) {
+        __ ldrsh(result, element_address);
+      } else {
+        __ LoadHalfWordUnaligned(result, address, TMP);
+      }
       __ SmiTag(result);
       break;
     case kTypedDataUint16ArrayCid:
     case kTwoByteStringCid:
     case kExternalTwoByteStringCid:
-      __ ldrh(result, element_address);
+      if (aligned()) {
+        __ ldrh(result, element_address);
+      } else {
+        __ LoadHalfWordUnsignedUnaligned(result, address, TMP);
+      }
       __ SmiTag(result);
       break;
     default:
       ASSERT((class_id() == kArrayCid) || (class_id() == kImmutableArrayCid));
       __ ldr(result, element_address);
       break;
   }
 }
 
 
@@ skipping 36 matching lines @@
 
 LocationSummary* StoreIndexedInstr::MakeLocationSummary(Zone* zone,
                                                         bool opt) const {
   const intptr_t kNumInputs = 3;
   LocationSummary* locs;
 
   bool needs_base = false;
   if (CanBeImmediateIndex(index(), class_id(), IsExternal(),
                           false,  // Store.
                           &needs_base)) {
-    const intptr_t kNumTemps = needs_base ? 1 : 0;
+    intptr_t kNumTemps = needs_base ? 1 : 0;

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

Maybe:

const intptr_t kNumTemps = aligned() ? (needs_base ? 1 : 0) : 2;

so that we can follow the pattern and leave kNumTemps const.

[Cutch, 2016/10/26 09:00:29]

+1

[rmacnak, 2016/10/31 22:57:26]

Done.

+    if (!aligned()) {
+      kNumTemps = 2;
+    }
     locs = new(zone) LocationSummary(
         zone, kNumInputs, kNumTemps, LocationSummary::kNoCall);
 
     // CanBeImmediateIndex must return false for unsafe smis.
     locs->set_in(1, Location::Constant(index()->definition()->AsConstant()));
     if (needs_base) {
       locs->set_temp(0, Location::RequiresRegister());
     }
+    if (!aligned()) {
+      locs->set_temp(0, Location::RequiresRegister());
+      locs->set_temp(1, Location::RequiresRegister());
+    }
   } else {
-    const intptr_t kNumTemps = 0;
+    const intptr_t kNumTemps = aligned() ? 0 : 2;
     locs = new(zone) LocationSummary(
         zone, kNumInputs, kNumTemps, LocationSummary::kNoCall);
 
     locs->set_in(1, Location::WritableRegister());
+    if (!aligned()) {
+      locs->set_temp(0, Location::RequiresRegister());
+      locs->set_temp(1, Location::RequiresRegister());
+    }
   }
   locs->set_in(0, Location::RequiresRegister());
 
   switch (class_id()) {
     case kArrayCid:
       locs->set_in(2, ShouldEmitStoreBarrier()
                         ? Location::WritableRegister()
                         : Location::RegisterOrConstant(value()));
       break;
     case kExternalTypedDataUint8ArrayCid:
@@ skipping 25 matching lines @@
   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 temp =
       (locs()->temp_count() > 0) ? locs()->temp(0).reg() : kNoRegister;
+  const Register temp2 =
+      (locs()->temp_count() > 1) ? locs()->temp(1).reg() : kNoRegister;
 
-  Address element_address = index.IsRegister()
+  Address element_address(IP);

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

Same comments here about asserts as above.

+  if (aligned()) {
+    element_address = index.IsRegister()
       ? __ ElementAddressForRegIndex(false,  // Store.
                                      IsExternal(), class_id(), index_scale(),
                                      array,
                                      index.reg())
       : __ ElementAddressForIntIndex(false,  // Store.
                                      IsExternal(), class_id(), index_scale(),
                                      array, Smi::Cast(index.constant()).Value(),
                                      temp);
+  } else {
+    if (index.IsRegister()) {
+      __ LoadElementAddressForRegIndex(temp,
+                                       false,  // Store.
+                                       IsExternal(), class_id(), index_scale(),
+                                       array,
+                                       index.reg());
+    } else {
+      __ LoadElementAddressForIntIndex(temp,
+                                       false,  // Store.
+                                       IsExternal(), class_id(), index_scale(),
+                                       array,
+                                       Smi::Cast(index.constant()).Value());
+    }
+  }
 
   switch (class_id()) {
     case kArrayCid:
       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 {
@@ skipping 38 matching lines @@
         __ mov(IP, Operand(value), LS);  // IP = value in range ? value : IP.
         __ SmiUntag(IP);
         __ strb(IP, element_address);
       }
       break;
     }
     case kTypedDataInt16ArrayCid:
     case kTypedDataUint16ArrayCid: {
       const Register value = locs()->in(2).reg();
       __ SmiUntag(IP, value);
-      __ strh(IP, element_address);
+      if (aligned()) {
+        __ strh(IP, element_address);
+      } else {
+        __ StoreHalfWordUnaligned(IP, temp, temp2);
+      }
       break;
     }
     case kTypedDataInt32ArrayCid:
     case kTypedDataUint32ArrayCid: {
       const Register value = locs()->in(2).reg();
-      __ str(value, element_address);
+      if (aligned()) {
+        __ str(value, element_address);
+      } else {
+        __ StoreWordUnaligned(value, temp, temp2);
+      }
       break;
     }
     case kTypedDataFloat32ArrayCid: {
       const SRegister value_reg =
           EvenSRegisterOf(EvenDRegisterOf(locs()->in(2).fpu_reg()));
       __ vstrs(value_reg, element_address);
       break;
     }
     case kTypedDataFloat64ArrayCid: {
       const DRegister value_reg = EvenDRegisterOf(locs()->in(2).fpu_reg());
@@ skipping 5520 matching lines @@
                                 1,
                                 locs());
   __ Drop(1);
   __ Pop(result);
 }
 
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_ARM