[mips][builtin] Small optimization.

port dcf3b66 https://crrev.com/c/1825222
port 7675b95 https://crrev.com/c/1826724

[mips][wasm-simd] Implement f32x4.sqrt.

port 36f2ec1 https://crrev.com/c/1808045

Change-Id: Ib714d56f2b0062d5013364eeea30294cf0b92cd4
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1826588
Reviewed-by: Santiago Aboy Solanes <solanes@chromium.org>
Reviewed-by: Zhi An Ng <zhin@chromium.org>
Commit-Queue: Yu Yin <xwafish@gmail.com>
Cr-Commit-Position: refs/heads/master@{#64030}

src/builtins/mips/builtins-mips.cc

@@ -1085,18 +1085,16 @@ void Builtins::Generate_InterpreterEntryTrampoline(MacroAssembler* masm) {
   __ Push(kInterpreterBytecodeArrayRegister, t0);
 
   // Allocate the local and temporary register file on the stack.
+  Label stack_overflow;
   {
     // Load frame size from the BytecodeArray object.
     __ lw(t0, FieldMemOperand(kInterpreterBytecodeArrayRegister,
                               BytecodeArray::kFrameSizeOffset));
 
     // Do a stack check to ensure we don't go over the limit.
-    Label ok;
     __ Subu(t1, sp, Operand(t0));
     LoadRealStackLimit(masm, a2);
-    __ Branch(&ok, hs, t1, Operand(a2));
+    __ Branch(&stack_overflow, lo, t1, Operand(a2));
-    __ CallRuntime(Runtime::kThrowStackOverflow);
-    __ bind(&ok);
 
     // If ok, push undefined as the initial value for all register file entries.
     Label loop_header;
@@ -1169,6 +1167,11 @@ void Builtins::Generate_InterpreterEntryTrampoline(MacroAssembler* masm) {
   GenerateTailCallToReturnedCode(masm, Runtime::kCompileLazy);
   // Unreachable code.
   __ break_(0xCC);
+
+  __ bind(&stack_overflow);
+  __ CallRuntime(Runtime::kThrowStackOverflow);
+  // Unreachable code.
+  __ break_(0xCC);
 }
 
 static void Generate_InterpreterPushArgs(MacroAssembler* masm,
@@ -2131,7 +2134,7 @@ void Builtins::Generate_Call(MacroAssembler* masm, ConvertReceiverMode mode) {
   //  -- a1 : the target to call (can be any Object).
   // -----------------------------------
 
-  Label non_callable, non_function, non_smi;
+  Label non_callable, non_smi;
   __ JumpIfSmi(a1, &non_callable);
   __ bind(&non_smi);
   __ GetObjectType(a1, t1, t2);
@@ -2146,12 +2149,11 @@ void Builtins::Generate_Call(MacroAssembler* masm, ConvertReceiverMode mode) {
   __ Branch(&non_callable, eq, t1, Operand(zero_reg));
 
   // Check if target is a proxy and call CallProxy external builtin
-  __ Branch(&non_function, ne, t2, Operand(JS_PROXY_TYPE));
+  __ Jump(BUILTIN_CODE(masm->isolate(), CallProxy),
-  __ Jump(BUILTIN_CODE(masm->isolate(), CallProxy), RelocInfo::CODE_TARGET);
+          RelocInfo::CODE_TARGET, eq, t2, Operand(JS_PROXY_TYPE));
 
   // 2. Call to something else, which might have a [[Call]] internal method (if
   // not we raise an exception).
-  __ bind(&non_function);
   // Overwrite the original receiver with the (original) target.
   __ Lsa(kScratchReg, sp, a0, kPointerSizeLog2);
   __ sw(a1, MemOperand(kScratchReg));

src/builtins/mips64/builtins-mips64.cc

@@ -1103,18 +1103,16 @@ void Builtins::Generate_InterpreterEntryTrampoline(MacroAssembler* masm) {
   __ Push(kInterpreterBytecodeArrayRegister, a4);
 
   // Allocate the local and temporary register file on the stack.
+  Label stack_overflow;
   {
     // Load frame size (word) from the BytecodeArray object.
     __ Lw(a4, FieldMemOperand(kInterpreterBytecodeArrayRegister,
                               BytecodeArray::kFrameSizeOffset));
 
     // Do a stack check to ensure we don't go over the limit.
-    Label ok;
     __ Dsubu(a5, sp, Operand(a4));
     LoadRealStackLimit(masm, a2);
-    __ Branch(&ok, hs, a5, Operand(a2));
+    __ Branch(&stack_overflow, lo, a5, Operand(a2));
-    __ CallRuntime(Runtime::kThrowStackOverflow);
-    __ bind(&ok);
 
     // If ok, push undefined as the initial value for all register file entries.
     Label loop_header;
@@ -1188,6 +1186,11 @@ void Builtins::Generate_InterpreterEntryTrampoline(MacroAssembler* masm) {
   GenerateTailCallToReturnedCode(masm, Runtime::kCompileLazy);
   // Unreachable code.
   __ break_(0xCC);
+
+  __ bind(&stack_overflow);
+  __ CallRuntime(Runtime::kThrowStackOverflow);
+  // Unreachable code.
+  __ break_(0xCC);
 }
 
 static void Generate_InterpreterPushArgs(MacroAssembler* masm,
@@ -2170,7 +2173,7 @@ void Builtins::Generate_Call(MacroAssembler* masm, ConvertReceiverMode mode) {
   //  -- a1 : the target to call (can be any Object).
   // -----------------------------------
 
-  Label non_callable, non_function, non_smi;
+  Label non_callable, non_smi;
   __ JumpIfSmi(a1, &non_callable);
   __ bind(&non_smi);
   __ GetObjectType(a1, t1, t2);
@@ -2184,12 +2187,11 @@ void Builtins::Generate_Call(MacroAssembler* masm, ConvertReceiverMode mode) {
   __ And(t1, t1, Operand(Map::IsCallableBit::kMask));
   __ Branch(&non_callable, eq, t1, Operand(zero_reg));
 
-  __ Branch(&non_function, ne, t2, Operand(JS_PROXY_TYPE));
+  __ Jump(BUILTIN_CODE(masm->isolate(), CallProxy),
-  __ Jump(BUILTIN_CODE(masm->isolate(), CallProxy), RelocInfo::CODE_TARGET);
+          RelocInfo::CODE_TARGET, eq, t2, Operand(JS_PROXY_TYPE));
 
   // 2. Call to something else, which might have a [[Call]] internal method (if
   // not we raise an exception).
-  __ bind(&non_function);
   // Overwrite the original receiver with the (original) target.
   __ Dlsa(kScratchReg, sp, a0, kPointerSizeLog2);
   __ Sd(a1, MemOperand(kScratchReg));

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

@@ -2053,6 +2053,11 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       __ bnegi_w(i.OutputSimd128Register(), i.InputSimd128Register(0), 31);
       break;
     }
+    case kMipsF32x4Sqrt: {
+      CpuFeatureScope msa_scope(tasm(), MIPS_SIMD);
+      __ fsqrt_w(i.OutputSimd128Register(), i.InputSimd128Register(0));
+      break;
+    }
     case kMipsF32x4RecipApprox: {
       CpuFeatureScope msa_scope(tasm(), MIPS_SIMD);
       __ frcp_w(i.OutputSimd128Register(), i.InputSimd128Register(0));

src/compiler/backend/mips/instruction-codes-mips.h

@@ -159,6 +159,7 @@ namespace compiler {
   V(MipsI32x4MinU)                 \
   V(MipsF32x4Abs)                  \
   V(MipsF32x4Neg)                  \
+  V(MipsF32x4Sqrt)                 \
   V(MipsF32x4RecipApprox)          \
   V(MipsF32x4RecipSqrtApprox)      \
   V(MipsF32x4Add)                  \

src/compiler/backend/mips/instruction-scheduler-mips.cc

@@ -54,6 +54,7 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kMipsF32x4Div:
     case kMipsF32x4Ne:
     case kMipsF32x4Neg:
+    case kMipsF32x4Sqrt:
     case kMipsF32x4RecipApprox:
     case kMipsF32x4RecipSqrtApprox:
     case kMipsF32x4ReplaceLane:

src/compiler/backend/mips/instruction-selector-mips.cc

@@ -2018,6 +2018,7 @@ void InstructionSelector::VisitInt64AbsWithOverflow(Node* node) {
   V(F32x4UConvertI32x4, kMipsF32x4UConvertI32x4)         \
   V(F32x4Abs, kMipsF32x4Abs)                             \
   V(F32x4Neg, kMipsF32x4Neg)                             \
+  V(F32x4Sqrt, kMipsF32x4Sqrt)                           \
   V(F32x4RecipApprox, kMipsF32x4RecipApprox)             \
   V(F32x4RecipSqrtApprox, kMipsF32x4RecipSqrtApprox)     \
   V(I32x4SConvertF32x4, kMipsI32x4SConvertF32x4)         \

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

@@ -2248,6 +2248,11 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       __ ftrunc_u_w(i.OutputSimd128Register(), i.InputSimd128Register(0));
       break;
     }
+    case kMips64F32x4Sqrt: {
+      CpuFeatureScope msa_scope(tasm(), MIPS_SIMD);
+      __ fsqrt_w(i.OutputSimd128Register(), i.InputSimd128Register(0));
+      break;
+    }
     case kMips64I32x4Neg: {
       CpuFeatureScope msa_scope(tasm(), MIPS_SIMD);
       __ xor_v(kSimd128RegZero, kSimd128RegZero, kSimd128RegZero);

src/compiler/backend/mips64/instruction-codes-mips64.h

@@ -189,6 +189,7 @@ namespace compiler {
   V(Mips64I32x4MinU)                         \
   V(Mips64F32x4Abs)                          \
   V(Mips64F32x4Neg)                          \
+  V(Mips64F32x4Sqrt)                          \
   V(Mips64F32x4RecipApprox)                  \
   V(Mips64F32x4RecipSqrtApprox)              \
   V(Mips64F32x4Add)                          \

src/compiler/backend/mips64/instruction-scheduler-mips64.cc

@@ -82,6 +82,7 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kMips64F32x4Div:
     case kMips64F32x4Ne:
     case kMips64F32x4Neg:
+    case kMips64F32x4Sqrt:
     case kMips64F32x4RecipApprox:
     case kMips64F32x4RecipSqrtApprox:
     case kMips64F32x4ReplaceLane:

src/compiler/backend/mips64/instruction-selector-mips64.cc

@@ -2681,6 +2681,7 @@ void InstructionSelector::VisitInt64AbsWithOverflow(Node* node) {
   V(F32x4UConvertI32x4, kMips64F32x4UConvertI32x4)         \
   V(F32x4Abs, kMips64F32x4Abs)                             \
   V(F32x4Neg, kMips64F32x4Neg)                             \
+  V(F32x4Sqrt, kMips64F32x4Sqrt)                           \
   V(F32x4RecipApprox, kMips64F32x4RecipApprox)             \
   V(F32x4RecipSqrtApprox, kMips64F32x4RecipSqrtApprox)     \
   V(I32x4SConvertF32x4, kMips64I32x4SConvertF32x4)         \