S390: [wasm] Int64Lowering of Int64Mul on ia32 and arm.

src/compiler/s390/code-generator-s390.cc

Index: src/compiler/s390/code-generator-s390.cc
diff --git a/src/compiler/s390/code-generator-s390.cc b/src/compiler/s390/code-generator-s390.cc
index 2759c50dbdd649e2a85b132a4796385c9569a03a..f3314737b48afdd148ae9f94be9727d84980a05d 100644
--- a/src/compiler/s390/code-generator-s390.cc
+++ b/src/compiler/s390/code-generator-s390.cc
@@ -882,6 +882,36 @@ void CodeGenerator::AssembleArchInstruction(Instruction* instr) {
       __ SubLogicalWithBorrow32(i.OutputRegister(1), i.InputRegister(1),
                                 i.InputRegister(3));
       break;
+    case kS390_MulPair:

[JoranSiu, 2016/03/30 21:23:19]

Why don't we load the full 64-bit value in R0 and R1, and use a 64-bit MSGR
instead?   It'll be more efficient.  We are able to use 64-bit instructions on
32-bit.

[john.yan, 2016/03/31 03:50:36]

OK.

+      // r3 i.InputRegister(0) ... left low word.
+      // r2 i.InputRegister(1) ... left high word.
+      // r5 i.InputRegister(2) ... right low word.
+      // r4 i.InputRegister(3) ... right high word.
+      // The following sequence was inspired by gcc
+      // r1 = (I1 + (I0 >> 31)) * I2
+      // r0 = (I3 + (I2 >> 31)) * I0
+      // r1 += r0
+      // push r1
+      // r0:r1 = I0 * I2
+      // O0 = r1
+      // pop O1
+      // O1 += r0
+      __ lr(r1, i.InputRegister(0));
+      __ lr(r0, i.InputRegister(2));
+      __ srl(r1, Operand(31));
+      __ srl(r0, Operand(31));
+      __ ar(r1, i.InputRegister(1));
+      __ ar(r0, i.InputRegister(3));
+      __ msr(r1, i.InputRegister(2));
+      __ msr(r0, i.InputRegister(0));
+      __ ar(r1, r0);
+      __ push(r1);
+      __ lr(r1, i.InputRegister(0));
+      __ mr_z(r0, i.InputRegister(2));
+      __ lr(i.OutputRegister(0), r1);
+      __ pop(i.OutputRegister(1));
+      __ ar(i.OutputRegister(1), r0);
+      break;
     case kS390_ShiftLeftPair:
       if (instr->InputAt(2)->IsImmediate()) {
         __ ShiftLeftPair(i.OutputRegister(0), i.OutputRegister(1),