[wasm] Int64Lowering of Int64Mul on ia32 and arm.

src/compiler/arm/code-generator-arm.cc

 // Copyright 2014 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "src/compiler/code-generator.h"
 
 #include "src/arm/macro-assembler-arm.h"
 #include "src/ast/scopes.h"
 #include "src/compiler/code-generator-impl.h"
 #include "src/compiler/gap-resolver.h"
@@ skipping 791 matching lines @@
       // i.InputRegister(0) ... left low word.
       // i.InputRegister(1) ... left high word.
       // i.InputRegister(2) ... right low word.
       // i.InputRegister(3) ... right high word.
       __ sub(i.OutputRegister(0), i.InputRegister(0), i.InputRegister(2),
              SBit::SetCC);
       __ sbc(i.OutputRegister(1), i.InputRegister(1),
              Operand(i.InputRegister(3)));
       DCHECK_EQ(LeaveCC, i.OutputSBit());
       break;
+    case kArmMulPair:
+      // i.InputRegister(0) ... left low word.
+      // i.InputRegister(1) ... left high word.
+      // i.InputRegister(2) ... right low word.
+      // i.InputRegister(3) ... right high word.
+      __ mul(i.TempRegister(0), i.InputRegister(0), i.InputRegister(3));
+      __ mul(i.TempRegister(1), i.InputRegister(2), i.InputRegister(1));
+      __ add(i.TempRegister(0), i.TempRegister(0), i.TempRegister(1));
+      __ umull(i.OutputRegister(0), i.OutputRegister(1), i.InputRegister(0),
+               i.InputRegister(2));
+      __ add(i.OutputRegister(1), i.OutputRegister(1), i.TempRegister(0));

[martyn.capewell, 2016/03/17 18:20:29]

I think if you use multiply-accumulates here, you won't need temp registers:
  __ mul(i.OutputRegister(1), i.InputRegister(0), i.InputRegister(3));
  __ mla(i.OutputRegister(1), i.InputRegister(2), i.InputRegister(1),
i.OutputRegister(1));
  __ mov(i.OutputRegister(0), Operand(0));
  __ umlal(i.OutputRegister(0), i.OutputRegister(1), i.InputRegister(0),
i.InputRegister(2));

[martyn.capewell, 2016/03/18 11:42:49]

A bit more thought gives:
  __ umull(i.OutputRegister(0), i.OutputRegister(1), i.InputRegister(0),
i.InputRegister(2));
  __ mla(i.OutputRegister(1), i.InputRegister(0), i.InputRegister(3),
i.OutputRegister(1));
  __ mla(i.OutputRegister(1), i.InputRegister(2), i.InputRegister(1),
i.OutputRegister(1));

[ahaas, 2016/03/29 11:11:50]

Done. It was necessary for this change to use a unique register for input 3.

+      break;
     case kArmLslPair:
       if (instr->InputAt(2)->IsImmediate()) {
         __ LslPair(i.OutputRegister(0), i.OutputRegister(1), i.InputRegister(0),
                    i.InputRegister(1), i.InputInt32(2));
       } else {
         __ LslPair(i.OutputRegister(0), i.OutputRegister(1), i.InputRegister(0),
                    i.InputRegister(1), kScratchReg, i.InputRegister(2));
       }
       break;
     case kArmLsrPair:
@@ skipping 745 matching lines @@
       padding_size -= v8::internal::Assembler::kInstrSize;
     }
   }
 }
 
 #undef __
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8