[mips] Cleanup: Refactor AddPair and SubPair, implement MulPair

Refactor AddPair and SubPair Macro-assembler instructions to
prevent register overwriting, refactor all the places where
these instructions are used. Also, implement MulPair instruction.

Change-Id: I3f8f9d5fe6fa5bf25df3446614ac311cf886b6ac
Reviewed-on: https://chromium-review.googlesource.com/1013571Reviewed-by: Ivica Bogosavljevic <ivica.bogosavljevic@mips.com>
Commit-Queue: Sreten Kovacevic <sreten.kovacevic@mips.com>
Cr-Commit-Position: refs/heads/master@{#52642}

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

@@ -1174,19 +1174,18 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     } break;
     case kMipsAddPair:
       __ AddPair(i.OutputRegister(0), i.OutputRegister(1), i.InputRegister(0),
-                 i.InputRegister(1), i.InputRegister(2), i.InputRegister(3));
+                 i.InputRegister(1), i.InputRegister(2), i.InputRegister(3),
+                 kScratchReg, kScratchReg2);
       break;
     case kMipsSubPair:
       __ SubPair(i.OutputRegister(0), i.OutputRegister(1), i.InputRegister(0),
-                 i.InputRegister(1), i.InputRegister(2), i.InputRegister(3));
+                 i.InputRegister(1), i.InputRegister(2), i.InputRegister(3),
+                 kScratchReg, kScratchReg2);
       break;
     case kMipsMulPair: {
-      __ Mulu(i.OutputRegister(1), i.OutputRegister(0), i.InputRegister(0),
-              i.InputRegister(2));
-      __ Mul(kScratchReg, i.InputRegister(0), i.InputRegister(3));
-      __ Mul(kScratchReg2, i.InputRegister(1), i.InputRegister(2));
-      __ Addu(i.OutputRegister(1), i.OutputRegister(1), kScratchReg);
-      __ Addu(i.OutputRegister(1), i.OutputRegister(1), kScratchReg2);
+      __ MulPair(i.OutputRegister(0), i.OutputRegister(1), i.InputRegister(0),
+                 i.InputRegister(1), i.InputRegister(2), i.InputRegister(3),
+                 kScratchReg, kScratchReg2);
     } break;
     case kMipsAddD:
       // TODO(plind): add special case: combine mult & add.

src/mips/macro-assembler-mips.cc

@@ -1411,34 +1411,44 @@ void TurboAssembler::MultiPopFPU(RegList regs) {
   addiu(sp, sp, stack_offset);
 }
 
-
 void TurboAssembler::AddPair(Register dst_low, Register dst_high,
                              Register left_low, Register left_high,
-                             Register right_low, Register right_high) {
-  Register kScratchReg = s3;
-  if (left_low == right_low) {
-    // Special case for left = right and the sum potentially overwriting both
-    // left and right.
-    Slt(kScratchReg, left_low, zero_reg);
-    Addu(dst_low, left_low, right_low);
-  } else {
-    Addu(dst_low, left_low, right_low);
-    // If the sum overwrites right, left remains unchanged, otherwise right
-    // remains unchanged.
-    Sltu(kScratchReg, dst_low, (dst_low == right_low) ? left_low : right_low);
-  }
-  Addu(dst_high, left_high, right_high);
-  Addu(dst_high, dst_high, kScratchReg);
+                             Register right_low, Register right_high,
+                             Register scratch1, Register scratch2) {
+  BlockTrampolinePoolScope block_trampoline_pool(this);
+  Register scratch3 = t8;
+  Addu(scratch1, left_low, right_low);
+  Sltu(scratch3, scratch1, left_low);
+  Addu(scratch2, left_high, right_high);
+  Addu(dst_high, scratch2, scratch3);
+  Move(dst_low, scratch1);
 }
 
 void TurboAssembler::SubPair(Register dst_low, Register dst_high,
                              Register left_low, Register left_high,
-                             Register right_low, Register right_high) {
-  Register kScratchReg = s3;
-  Sltu(kScratchReg, left_low, right_low);
-  Subu(dst_low, left_low, right_low);
-  Subu(dst_high, left_high, right_high);
-  Subu(dst_high, dst_high, kScratchReg);
+                             Register right_low, Register right_high,
+                             Register scratch1, Register scratch2) {
+  BlockTrampolinePoolScope block_trampoline_pool(this);
+  Register scratch3 = t8;
+  Sltu(scratch3, left_low, right_low);
+  Subu(scratch1, left_low, right_low);
+  Subu(scratch2, left_high, right_high);
+  Subu(dst_high, scratch2, scratch3);
+  Move(dst_low, scratch1);
+}
+
+void TurboAssembler::MulPair(Register dst_low, Register dst_high,
+                             Register left_low, Register left_high,
+                             Register right_low, Register right_high,
+                             Register scratch1, Register scratch2) {
+  BlockTrampolinePoolScope block_trampoline_pool(this);
+  Register scratch3 = t8;
+  Mulu(scratch2, scratch1, left_low, right_low);
+  Mul(scratch3, left_low, right_high);
+  Addu(scratch2, scratch2, scratch3);
+  Mul(scratch3, left_high, right_low);
+  Addu(dst_high, scratch2, scratch3);
+  Move(dst_low, scratch1);
 }
 
 void TurboAssembler::ShlPair(Register dst_low, Register dst_high,

src/mips/macro-assembler-mips.h

@@ -579,10 +579,16 @@ class TurboAssembler : public Assembler {
 
   // Int64Lowering instructions
   void AddPair(Register dst_low, Register dst_high, Register left_low,
-               Register left_high, Register right_low, Register right_high);
+               Register left_high, Register right_low, Register right_high,
+               Register scratch1, Register scratch2);
 
   void SubPair(Register dst_low, Register dst_high, Register left_low,
-               Register left_high, Register right_low, Register right_high);
+               Register left_high, Register right_low, Register right_high,
+               Register scratch1, Register scratch2);
+
+  void MulPair(Register dst_low, Register dst_high, Register left_low,
+               Register left_high, Register right_low, Register right_high,
+               Register scratch1, Register scratch2);
 
   void ShlPair(Register dst_low, Register dst_high, Register src_low,
                Register src_high, Register shift, Register scratch1,

src/wasm/baseline/mips/liftoff-assembler-mips.h

@@ -584,60 +584,23 @@ I32_SHIFTOP(shr, srlv)
 
 void LiftoffAssembler::emit_i64_mul(LiftoffRegister dst, LiftoffRegister lhs,
                                     LiftoffRegister rhs) {
-  LiftoffRegister org_dst = dst;
-  // Allocate registers for dst, to prevent overwriting source.
-  if (dst.overlaps(lhs) || dst.overlaps(rhs)) {
-    dst = GetUnusedRegister(kGpRegPair, LiftoffRegList::ForRegs(lhs, rhs));
-  }
-
-  // Multiply.
-  TurboAssembler::Mulu(dst.high_gp(), dst.low_gp(), lhs.low_gp(), rhs.low_gp());
-  LiftoffRegister tmp =
-      GetUnusedRegister(kGpReg, LiftoffRegList::ForRegs(dst, lhs, rhs));
-  TurboAssembler::Mul(tmp.gp(), lhs.low_gp(), rhs.high_gp());
-  TurboAssembler::Addu(dst.high_gp(), dst.high_gp(), tmp.gp());
-
-  TurboAssembler::Mul(tmp.gp(), lhs.high_gp(), rhs.low_gp());
-  TurboAssembler::Addu(dst.high_gp(), dst.high_gp(), tmp.gp());
-
-  // Move result to original dst if needed.
-  if (dst != org_dst) {
-    Move(org_dst, dst, kWasmI64);
-  }
+  TurboAssembler::MulPair(dst.low_gp(), dst.high_gp(), lhs.low_gp(),
+                          lhs.high_gp(), rhs.low_gp(), rhs.high_gp(),
+                          kScratchReg, kScratchReg2);
 }
 
 void LiftoffAssembler::emit_i64_add(LiftoffRegister dst, LiftoffRegister lhs,
                                     LiftoffRegister rhs) {
-  LiftoffRegister org_dst = dst;
-  // Allocate registers for dst, to prevent overwriting source.
-  if (dst.overlaps(lhs) || dst.overlaps(rhs)) {
-    dst = GetUnusedRegister(kGpRegPair, LiftoffRegList::ForRegs(lhs, rhs));
-  }
-
   TurboAssembler::AddPair(dst.low_gp(), dst.high_gp(), lhs.low_gp(),
-                          lhs.high_gp(), rhs.low_gp(), rhs.high_gp());
-
-  // Move result to original dst if needed.
-  if (dst != org_dst) {
-    Move(org_dst, dst, kWasmI64);
-  }
+                          lhs.high_gp(), rhs.low_gp(), rhs.high_gp(),
+                          kScratchReg, kScratchReg2);
 }
 
 void LiftoffAssembler::emit_i64_sub(LiftoffRegister dst, LiftoffRegister lhs,
                                     LiftoffRegister rhs) {
-  LiftoffRegister org_dst = dst;
-  // Allocate registers for dst, to prevent overwriting source.
-  if (dst.overlaps(lhs) || dst.overlaps(rhs)) {
-    dst = GetUnusedRegister(kGpRegPair, LiftoffRegList::ForRegs(lhs, rhs));
-  }
-
   TurboAssembler::SubPair(dst.low_gp(), dst.high_gp(), lhs.low_gp(),
-                          lhs.high_gp(), rhs.low_gp(), rhs.high_gp());
-
-  // Move result to original dst if needed.
-  if (dst != org_dst) {
-    Move(org_dst, dst, kWasmI64);
-  }
+                          lhs.high_gp(), rhs.low_gp(), rhs.high_gp(),
+                          kScratchReg, kScratchReg2);
 }
 
 namespace liftoff {