[mips] Fix {AssembleArchLookupSwitch} implementation

Previous implementation used delay slot the wrong way.
Also, trampoline pools were not generated as they should.
MIPS64 changed to be the same as MIPS since there is probably
same problem that is masked.

test=wasm-spec-tests/tests/br_table

Change-Id: I94786233714a4a2f5eb86e74e02b7e7a7328bf2b
Reviewed-on: https://chromium-review.googlesource.com/901883Reviewed-by: Ivica Bogosavljevic <ivica.bogosavljevic@mips.com>
Commit-Queue: Ivica Bogosavljevic <ivica.bogosavljevic@mips.com>
Cr-Commit-Position: refs/heads/master@{#51100}

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

@@ -3254,9 +3254,8 @@ void CodeGenerator::AssembleArchLookupSwitch(Instruction* instr) {
   Register input = i.InputRegister(0);
   for (size_t index = 2; index < instr->InputCount(); index += 2) {
     __ li(at, Operand(i.InputInt32(index + 0)));
-    __ beq(input, at, GetLabel(i.InputRpo(index + 1)));
+    __ Branch(GetLabel(i.InputRpo(index + 1)), eq, input, Operand(at));
   }
-  __ nop();  // Branch delay slot of the last beq.
   AssembleArchJump(i.InputRpo(1));
 }
 

src/compiler/mips64/code-generator-mips64.cc

@@ -3523,9 +3523,8 @@ void CodeGenerator::AssembleArchLookupSwitch(Instruction* instr) {
   Register input = i.InputRegister(0);
   for (size_t index = 2; index < instr->InputCount(); index += 2) {
     __ li(at, Operand(i.InputInt32(index + 0)));
-    __ beq(input, at, GetLabel(i.InputRpo(index + 1)));
+    __ Branch(GetLabel(i.InputRpo(index + 1)), eq, input, Operand(at));
   }
-  __ nop();  // Branch delay slot of the last beq.
   AssembleArchJump(i.InputRpo(1));
 }