Improved calling convention for aarch64

IDE/src/BuildContext.bf

@@ -317,12 +317,13 @@ namespace IDE
 
 		bool QueueProjectGNULink(Project project, String targetPath, Workspace.Options workspaceOptions, Project.Options options, String objectsArg)
 		{
+			if (options.mBuildOptions.mBuildKind == .Intermediate)
+				return true;
+
 			bool isDebug = gApp.mConfigName.IndexOf("Debug", true) != -1;
 
 			bool isMinGW = false;
 
-			
-
 #if BF_PLATFORM_WINDOWS
 			bool isWSL = mPlatformType == .Linux;
 			String llvmDir = scope String(IDEApp.sApp.mInstallDir);

IDE/src/Project.bf

@@ -879,6 +879,7 @@ namespace IDE
 			case Test;
 			case StaticLib;
 			case DynamicLib;
+			case Intermediate;
 			case NotSupported;
 		}
 

IDEHelper/Compiler/BfIRBuilder.cpp

@@ -350,6 +350,15 @@ int BfIRConstHolder::GetSize(BfTypeCode typeCode)
 	case BfTypeCode_Enum: return 0;
 	case BfTypeCode_TypeAlias: return 0;
 	case BfTypeCode_Extension: return 0;
+	case BfTypeCode_FloatX2: return 4*2;
+	case BfTypeCode_FloatX3: return 4*3;
+	case BfTypeCode_FloatX4: return 4*4;
+	case BfTypeCode_DoubleX2: return 8*2;
+	case BfTypeCode_DoubleX3: return 8*3;
+	case BfTypeCode_DoubleX4: return 8*4;
+	case BfTypeCode_Int64X2: return 8*2;
+	case BfTypeCode_Int64X3: return 8*3;
+	case BfTypeCode_Int64X4: return 8*4;
 	default: return 0;
 	}
 }

IDEHelper/Compiler/BfIRBuilder.h

@@ -109,7 +109,15 @@ enum BfTypeCode : uint8
 	BfTypeCode_Enum,
 	BfTypeCode_TypeAlias,
 	BfTypeCode_Extension,
-
+	BfTypeCode_FloatX2,
+	BfTypeCode_FloatX3,
+	BfTypeCode_FloatX4,
+	BfTypeCode_DoubleX2,
+	BfTypeCode_DoubleX3,
+	BfTypeCode_DoubleX4,
+	BfTypeCode_Int64X2,
+	BfTypeCode_Int64X3,
+	BfTypeCode_Int64X4,
 	BfTypeCode_Length
 };
 

IDEHelper/Compiler/BfIRCodeGen.cpp

@@ -601,6 +601,24 @@ llvm::Type* BfIRCodeGen::GetLLVMType(BfTypeCode typeCode, bool& isSigned)
 		return llvm::Type::getDoubleTy(*mLLVMContext);
 	case BfTypeCode_Float2:
 		return llvm::FixedVectorType::get(llvm::Type::getFloatTy(*mLLVMContext), 2);
+	case BfTypeCode_FloatX2:
+		return llvm::ArrayType::get(llvm::Type::getFloatTy(*mLLVMContext), 2);
+	case BfTypeCode_FloatX3:
+		return llvm::ArrayType::get(llvm::Type::getFloatTy(*mLLVMContext), 3);
+	case BfTypeCode_FloatX4:
+		return llvm::ArrayType::get(llvm::Type::getFloatTy(*mLLVMContext), 4);
+	case BfTypeCode_DoubleX2:
+		return llvm::ArrayType::get(llvm::Type::getDoubleTy(*mLLVMContext), 2);
+	case BfTypeCode_DoubleX3:
+		return llvm::ArrayType::get(llvm::Type::getDoubleTy(*mLLVMContext), 3);
+	case BfTypeCode_DoubleX4:
+		return llvm::ArrayType::get(llvm::Type::getDoubleTy(*mLLVMContext), 4);
+	case BfTypeCode_Int64X2:
+		return llvm::ArrayType::get(llvm::Type::getInt64Ty(*mLLVMContext), 2);
+	case BfTypeCode_Int64X3:
+		return llvm::ArrayType::get(llvm::Type::getInt64Ty(*mLLVMContext), 3);
+	case BfTypeCode_Int64X4:
+		return llvm::ArrayType::get(llvm::Type::getInt64Ty(*mLLVMContext), 4);
 	default: break;
 	}
 	return NULL;

IDEHelper/Compiler/BfResolvedTypeUtils.cpp

@@ -842,7 +842,8 @@ bool BfMethodInstance::WantsStructsAttribByVal(BfType* paramType)
 	if ((owner->mModule->mCompiler->mOptions.mPlatformType == BfPlatformType_Windows) &&
 		(owner->mModule->mCompiler->mOptions.mMachineType == BfMachineType_x64))
 		return false;
-
+	if (owner->mModule->mCompiler->mOptions.mMachineType == BfMachineType_AArch64)
+		return false;
 	auto typeInst = paramType->ToTypeInstance();
 	return (typeInst != NULL) && (typeInst->mIsCRepr);
 }
@@ -1202,7 +1203,7 @@ void BfMethodInstance::GetIRFunctionInfo(BfModule* module, BfIRType& returnType,
 
 	BfTypeCode loweredReturnTypeCode = BfTypeCode_None;
 	BfTypeCode loweredReturnTypeCode2 = BfTypeCode_None;	
-	if ((!module->mIsComptimeModule) && (GetLoweredReturnType(&loweredReturnTypeCode, &loweredReturnTypeCode2, forceStatic)))
+	if ((!module->mIsComptimeModule) && (GetLoweredReturnType(&loweredReturnTypeCode, &loweredReturnTypeCode2, forceStatic)) && (loweredReturnTypeCode != BfTypeCode_None))
 	{
 		auto irReturnType = module->GetIRLoweredType(loweredReturnTypeCode, loweredReturnTypeCode2);
 		returnType = irReturnType;
@@ -1842,8 +1843,8 @@ bool BfTypeInstance::GetLoweredType(BfTypeUsage typeUsage, BfTypeCode* outTypeCo
 	{
 		// Odd Windows rule: composite returns for non-static methods are always sret
 		if (typeUsage == BfTypeUsage_Return_NonStatic)
-			return false;		
+			return false;
-	}
+	}	
 	else
 	{
 		// Non-Win64 systems allow lowered splitting of composites over multiple params
@@ -1860,12 +1861,15 @@ bool BfTypeInstance::GetLoweredType(BfTypeUsage typeUsage, BfTypeCode* outTypeCo
 		}
 	}
 
+	int maxInstSize = 16;
+	if (mModule->mCompiler->mOptions.mMachineType == BfMachineType_AArch64)
+		maxInstSize = 32;
+
 	if (deepCheck)
 	{
-
+		if ((mInstSize >= 4) && (mInstSize <= maxInstSize))
-		if ((mInstSize >= 4) && (mInstSize <= 16))
 		{
-			BfTypeCode types[4] = { BfTypeCode_None };
+			BfTypeCode types[8] = { BfTypeCode_None };			
 
 			std::function<void(BfType*, int)> _CheckType = [&](BfType* type, int offset)
 			{
@@ -1883,12 +1887,14 @@ bool BfTypeInstance::GetLoweredType(BfTypeUsage typeUsage, BfTypeCode* outTypeCo
 						}
 					}
 					else
-						types[offset / 4] = BfTypeCode_Object;
+					{
+						types[offset / 4] = BfTypeCode_Object;						
+					}
 				}
 				else if (type->IsPrimitiveType())
 				{
 					auto primType = (BfPrimitiveType*)type;
-					types[offset / 4] = primType->mTypeDef->mTypeCode;
+					types[offset / 4] = primType->mTypeDef->mTypeCode;					
 				}
 				else if (type->IsSizedArray())
 				{
@@ -1902,7 +1908,87 @@ bool BfTypeInstance::GetLoweredType(BfTypeUsage typeUsage, BfTypeCode* outTypeCo
 
 			bool handled = false;
 
-			if (mModule->mCompiler->mOptions.mPlatformType == BfPlatformType_Windows)
+			if (mModule->mCompiler->mOptions.mMachineType == BfMachineType_AArch64)
+			{
+				// For returns, we want to avoid sret but not actually lower
+				bool writeOutCode = (typeUsage != BfTypeUsage_Return_NonStatic) && (typeUsage != BfTypeUsage_Return_Static);
+
+				if ((types[0] == BfTypeCode_Float) &&
+					(types[1] == BfTypeCode_Float) &&
+					(types[2] == BfTypeCode_None))
+				{
+					if ((outTypeCode != NULL) && (writeOutCode))
+						*outTypeCode = BfTypeCode_FloatX2;
+					return true;
+				}
+
+				if ((types[0] == BfTypeCode_Float) &&
+					(types[1] == BfTypeCode_Float) &&
+					(types[2] == BfTypeCode_Float) &&
+					(types[3] == BfTypeCode_None))
+				{
+					if ((outTypeCode != NULL) && (writeOutCode))
+						*outTypeCode = BfTypeCode_FloatX3;
+					return true;
+				}
+
+				if ((types[0] == BfTypeCode_Float) &&
+					(types[1] == BfTypeCode_Float) &&
+					(types[2] == BfTypeCode_Float) &&
+					(types[3] == BfTypeCode_Float) &&
+					(types[4] == BfTypeCode_None))
+				{
+					if ((outTypeCode != NULL) && (writeOutCode))
+						*outTypeCode = BfTypeCode_FloatX4;
+					return true;
+				}
+
+				if ((types[0] == BfTypeCode_Double) &&
+					(types[2] == BfTypeCode_Double) &&
+					(types[4] == BfTypeCode_None))
+				{
+					if ((outTypeCode != NULL) && (writeOutCode))
+						*outTypeCode = BfTypeCode_DoubleX2;
+					return true;
+				}
+
+				if ((types[0] == BfTypeCode_Double) &&
+					(types[2] == BfTypeCode_Double) &&
+					(types[4] == BfTypeCode_Double) &&
+					(types[6] == BfTypeCode_None))
+				{
+					if ((outTypeCode != NULL) && (writeOutCode))
+						*outTypeCode = BfTypeCode_DoubleX3;
+					return true;
+				}
+
+				if ((types[0] == BfTypeCode_Double) &&
+					(types[2] == BfTypeCode_Double) &&
+					(types[4] == BfTypeCode_Double) &&
+					(types[6] == BfTypeCode_Double))
+				{
+					if ((outTypeCode != NULL) && (writeOutCode))
+						*outTypeCode = BfTypeCode_DoubleX4;
+					return true;
+				}
+
+				if (mInstSize <= 8)
+				{
+					if ((outTypeCode != NULL) && (writeOutCode))
+						*outTypeCode = BfTypeCode_Int64;
+					return true;
+				}
+
+				if (mInstSize <= 16)
+				{
+					if ((outTypeCode != NULL) && (writeOutCode))
+						*outTypeCode = BfTypeCode_Int64X2;
+					return true;
+				}
+
+				return false;
+			}
+			else if (mModule->mCompiler->mOptions.mPlatformType == BfPlatformType_Windows)
 			{
 				bool hasFloat = false;
 				for (int type = 0; type < 4; type++)