[loong64][mips64][fastcall] Enable float support on loong64 and mips64

Port commit 098f31f4

Bug: chromium:1052746

Change-Id: I4f9fd952c2ce8b51772eac89d4852d55363d1ed1
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3292352
Auto-Submit: Liu yu <liuyu@loongson.cn>
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Commit-Queue: Toon Verwaest <verwaest@chromium.org>
Cr-Commit-Position: refs/heads/main@{#78106}

src/codegen/loong64/macro-assembler-loong64.cc

@@ -3892,16 +3892,19 @@ void TurboAssembler::Float64MinOutOfLine(FPURegister dst, FPURegister src1,
 }
 
 static const int kRegisterPassedArguments = 8;
+static const int kFPRegisterPassedArguments = 8;
 
 int TurboAssembler::CalculateStackPassedWords(int num_reg_arguments,
                                               int num_double_arguments) {
   int stack_passed_words = 0;
-  num_reg_arguments += 2 * num_double_arguments;
 
   // Up to eight simple arguments are passed in registers a0..a7.
   if (num_reg_arguments > kRegisterPassedArguments) {
     stack_passed_words += num_reg_arguments - kRegisterPassedArguments;
   }
+  if (num_double_arguments > kFPRegisterPassedArguments) {
+    stack_passed_words += num_double_arguments - kFPRegisterPassedArguments;
+  }
   return stack_passed_words;
 }
 

src/codegen/mips64/macro-assembler-mips64.cc

@@ -5906,12 +5906,11 @@ static const int kRegisterPassedArguments = 8;
 int TurboAssembler::CalculateStackPassedWords(int num_reg_arguments,
                                               int num_double_arguments) {
   int stack_passed_words = 0;
-  num_reg_arguments += 2 * num_double_arguments;
+  int num_args = num_reg_arguments + num_double_arguments;
 
-  // O32: Up to four simple arguments are passed in registers a0..a3.
-  // N64: Up to eight simple arguments are passed in registers a0..a7.
-  if (num_reg_arguments > kRegisterPassedArguments) {
-    stack_passed_words += num_reg_arguments - kRegisterPassedArguments;
+  // Up to eight arguments are passed in FPURegisters and GPRegisters.
+  if (num_args > kRegisterPassedArguments) {
+    stack_passed_words = num_args - kRegisterPassedArguments;
   }
   stack_passed_words += kCArgSlotCount;
   return stack_passed_words;

src/compiler/backend/loong64/code-generator-loong64.cc

@@ -650,8 +650,9 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     case kArchPrepareCallCFunction: {
       UseScratchRegisterScope temps(tasm());
       Register scratch = temps.Acquire();
-      int const num_parameters = MiscField::decode(instr->opcode());
-      __ PrepareCallCFunction(num_parameters, scratch);
+      int const num_gp_parameters = ParamField::decode(instr->opcode());
+      int const num_fp_parameters = FPParamField::decode(instr->opcode());
+      __ PrepareCallCFunction(num_gp_parameters, num_fp_parameters, scratch);
       // Frame alignment requires using FP-relative frame addressing.
       frame_access_state()->SetFrameAccessToFP();
       break;
@@ -687,7 +688,8 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       AssemblePrepareTailCall();
       break;
     case kArchCallCFunction: {
-      int const num_parameters = MiscField::decode(instr->opcode());
+      int const num_gp_parameters = ParamField::decode(instr->opcode());
+      int const num_fp_parameters = FPParamField::decode(instr->opcode());
 #if V8_ENABLE_WEBASSEMBLY
       Label start_call;
       bool isWasmCapiFunction =
@@ -709,10 +711,10 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
 #endif  // V8_ENABLE_WEBASSEMBLY
       if (instr->InputAt(0)->IsImmediate()) {
         ExternalReference ref = i.InputExternalReference(0);
-        __ CallCFunction(ref, num_parameters);
+        __ CallCFunction(ref, num_gp_parameters, num_fp_parameters);
       } else {
         Register func = i.InputRegister(0);
-        __ CallCFunction(func, num_parameters);
+        __ CallCFunction(func, num_gp_parameters, num_fp_parameters);
       }
 #if V8_ENABLE_WEBASSEMBLY
       if (isWasmCapiFunction) {

src/compiler/backend/loong64/instruction-selector-loong64.cc

@@ -1671,8 +1671,10 @@ void InstructionSelector::EmitPrepareArguments(
 
   // Prepare for C function call.
   if (call_descriptor->IsCFunctionCall()) {
-    Emit(kArchPrepareCallCFunction | MiscField::encode(static_cast<int>(
-                                         call_descriptor->ParameterCount())),
+    int gp_param_count = static_cast<int>(call_descriptor->GPParameterCount());
+    int fp_param_count = static_cast<int>(call_descriptor->FPParameterCount());
+    Emit(kArchPrepareCallCFunction | ParamField::encode(gp_param_count) |
+             FPParamField::encode(fp_param_count),
          0, nullptr, 0, nullptr);
 
     // Poke any stack arguments.

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

@@ -704,7 +704,8 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       AssemblePrepareTailCall();
       break;
     case kArchCallCFunction: {
-      int const num_parameters = MiscField::decode(instr->opcode());
+      int const num_gp_parameters = ParamField::decode(instr->opcode());
+      int const num_fp_parameters = FPParamField::decode(instr->opcode());
 #if V8_ENABLE_WEBASSEMBLY
       Label start_call;
       bool isWasmCapiFunction =
@@ -731,10 +732,10 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
 #endif  // V8_ENABLE_WEBASSEMBLY
       if (instr->InputAt(0)->IsImmediate()) {
         ExternalReference ref = i.InputExternalReference(0);
-        __ CallCFunction(ref, num_parameters);
+        __ CallCFunction(ref, num_gp_parameters, num_fp_parameters);
       } else {
         Register func = i.InputRegister(0);
-        __ CallCFunction(func, num_parameters);
+        __ CallCFunction(func, num_gp_parameters, num_fp_parameters);
       }
 #if V8_ENABLE_WEBASSEMBLY
       if (isWasmCapiFunction) {

src/compiler/c-linkage.cc

@@ -96,6 +96,9 @@ namespace {
 // == mips64 =================================================================
 // ===========================================================================
 #define PARAM_REGISTERS a0, a1, a2, a3, a4, a5, a6, a7
+#define FP_PARAM_REGISTERS f12, f13, f14, f15, f16, f17, f18, f19
+#define FP_RETURN_REGISTER f0
+
 #define CALLEE_SAVE_REGISTERS                                                  \
   s0.bit() | s1.bit() | s2.bit() | s3.bit() | s4.bit() | s5.bit() | s6.bit() | \
       s7.bit()
@@ -107,6 +110,8 @@ namespace {
 // == loong64 ================================================================
 // ===========================================================================
 #define PARAM_REGISTERS a0, a1, a2, a3, a4, a5, a6, a7
+#define FP_PARAM_REGISTERS f0, f1, f2, f3, f4, f5, f6, f7
+#define FP_RETURN_REGISTER f0
 #define CALLEE_SAVE_REGISTERS                                                  \
   s0.bit() | s1.bit() | s2.bit() | s3.bit() | s4.bit() | s5.bit() | s6.bit() | \
       s7.bit() | s8.bit() | fp.bit()
@@ -166,12 +171,15 @@ namespace {
 #endif
 }  // namespace
 
-#if defined(V8_TARGET_OS_WIN) && defined(V8_TARGET_ARCH_X64)
+#if (defined(V8_TARGET_OS_WIN) && defined(V8_TARGET_ARCH_X64)) || \
+    defined(V8_TARGET_ARCH_MIPS64)
 // As defined in
 // https://docs.microsoft.com/en-us/cpp/build/x64-calling-convention?view=vs-2019#parameter-passing,
 // Windows calling convention doesn't differentiate between GP and FP params
 // when counting how many of them should be placed in registers. That's why
 // we use the same counter {i} for both types here.
+// MIPS is the same, as defined in
+// https://techpubs.jurassic.nl/manuals/0630/developer/Mpro_n32_ABI/sgi_html/ch02.html#id52620.
 void BuildParameterLocations(const MachineSignature* msig,
                              size_t kFPParamRegisterCount,
                              size_t kParamRegisterCount,
@@ -203,7 +211,8 @@ void BuildParameterLocations(const MachineSignature* msig,
     }
   }
 }
-#else  // defined(V8_TARGET_OS_WIN) && defined(V8_TARGET_ARCH_X64)
+#else  // (defined(V8_TARGET_OS_WIN) && defined(V8_TARGET_ARCH_X64)) ||
+       // defined(V8_TARGET_ARCH_MIPS64)
 // As defined in https://www.agner.org/optimize/calling_conventions.pdf,
 // Section 7, Linux and Mac place parameters in consecutive registers,
 // differentiating between GP and FP params. That's why we maintain two
@@ -244,7 +253,8 @@ void BuildParameterLocations(const MachineSignature* msig,
     }
   }
 }
-#endif  // defined(V8_TARGET_OS_WIN) && defined(V8_TARGET_ARCH_X64)
+#endif  // (defined(V8_TARGET_OS_WIN) && defined(V8_TARGET_ARCH_X64)) ||
+        // defined(V8_TARGET_ARCH_MIPS64)
 
 // General code uses the above configuration data.
 CallDescriptor* Linkage::GetSimplifiedCDescriptor(Zone* zone,

src/compiler/globals.h

@@ -93,7 +93,9 @@ const int kMaxFastLiteralProperties = JSObject::kMaxInObjectProperties;
 // (using FP stack). As support is added to more platforms, please make sure
 // to list them here in order to enable tests of this functionality.
 // Make sure to sync the following with src/d8/d8-test.cc.
-#if defined(V8_TARGET_ARCH_X64) || defined(V8_TARGET_ARCH_ARM64)
+#if defined(V8_TARGET_ARCH_X64) || defined(V8_TARGET_ARCH_ARM64) || \
+    (defined(V8_TARGET_ARCH_MIPS64) && !defined(USE_SIMULATOR)) ||  \
+    (defined(V8_TARGET_ARCH_LOONG64) && !defined(USE_SIMULATOR))
 #define V8_ENABLE_FP_PARAMS_IN_C_LINKAGE
 #endif
 

src/d8/d8-test.cc

@@ -17,7 +17,9 @@
 // and resetting these counters.
 
 // Make sure to sync the following with src/compiler/globals.h.
-#if defined(V8_TARGET_ARCH_X64) || defined(V8_TARGET_ARCH_ARM64)
+#if defined(V8_TARGET_ARCH_X64) || defined(V8_TARGET_ARCH_ARM64) || \
+    (defined(V8_TARGET_ARCH_MIPS64) && !defined(USE_SIMULATOR)) ||  \
+    (defined(V8_TARGET_ARCH_LOONG64) && !defined(USE_SIMULATOR))
 #define V8_ENABLE_FP_PARAMS_IN_C_LINKAGE
 #endif
 

src/execution/loong64/simulator-loong64.cc

@@ -2228,7 +2228,18 @@ void Simulator::SoftwareInterrupt() {
           reinterpret_cast<SimulatorRuntimeProfilingGetterCall>(external);
       target(arg0, arg1, Redirection::ReverseRedirection(arg2));
     } else {
-      DCHECK(redirection->type() == ExternalReference::BUILTIN_CALL ||
+      // FAST_C_CALL is temporarily handled here as well, because we lack
+      // proper support for direct C calls with FP params in the simulator.
+      // The generic BUILTIN_CALL path assumes all parameters are passed in
+      // the GP registers, thus supporting calling the slow callback without
+      // crashing. The reason for that is that in the mjsunit tests we check
+      // the `fast_c_api.supports_fp_params` (which is false on simulator
+      // builds for loong64), thus we expect that the slow path will be
+      // called. And since the slow path passes the arguments as a `const
+      // FunctionCallbackInfo<Value>&` (which is a GP argument), the call is
+      // made correctly.
+      DCHECK(redirection->type() == ExternalReference::FAST_C_CALL ||
+             redirection->type() == ExternalReference::BUILTIN_CALL ||
              redirection->type() == ExternalReference::BUILTIN_CALL_PAIR);
       SimulatorRuntimeCall target =
           reinterpret_cast<SimulatorRuntimeCall>(external);

src/execution/mips64/simulator-mips64.cc

@@ -2423,7 +2423,18 @@ void Simulator::SoftwareInterrupt() {
           reinterpret_cast<SimulatorRuntimeProfilingGetterCall>(external);
       target(arg0, arg1, Redirection::ReverseRedirection(arg2));
     } else {
-      DCHECK(redirection->type() == ExternalReference::BUILTIN_CALL ||
+      // FAST_C_CALL is temporarily handled here as well, because we lack
+      // proper support for direct C calls with FP params in the simulator.
+      // The generic BUILTIN_CALL path assumes all parameters are passed in
+      // the GP registers, thus supporting calling the slow callback without
+      // crashing. The reason for that is that in the mjsunit tests we check
+      // the `fast_c_api.supports_fp_params` (which is false on simulator
+      // builds for mips64), thus we expect that the slow path will be
+      // called. And since the slow path passes the arguments as a `const
+      // FunctionCallbackInfo<Value>&` (which is a GP argument), the call is
+      // made correctly.
+      DCHECK(redirection->type() == ExternalReference::FAST_C_CALL ||
+             redirection->type() == ExternalReference::BUILTIN_CALL ||
              redirection->type() == ExternalReference::BUILTIN_CALL_PAIR);
       SimulatorRuntimeCall target =
           reinterpret_cast<SimulatorRuntimeCall>(external);