Namespaces
namespace	utils

Classes
class	AssignmentNode
	AST node for assignment statements. More...
class	ASTNode
	Abstract Syntax Tree (AST) node base class. More...
class	BinaryOpNode
	AST node for binary operations (arithmetic and relational). More...
class	CallNode
	AST node for function calls. More...
struct	ChainedBrightnessPattern
struct	ChainedInvertPattern
struct	ChainRotatePattern
class	CLIOptions
	Struct to hold command-line options for the compiler. More...
struct	CombineNestedCrops
	Canonicalization pattern for the CropOp that combines nested crop operations into a single crop operation. only applicable when both crop operations have constant integer coordinates and dimensions. More...
struct	CompileError
	Struct to hold compilation error information. More...
class	Compiler
	Compiler class that orchestrates the compilation process. More...
struct	CropToAffine
struct	ElementWiseBinaryOpToAffine
struct	ElementWiseUnaryOpToAffine
struct	FixedKernelFilterPattern
class	ForNode
	AST node for "for" statements. More...
class	FunctionNode
	AST node for function definitions. More...
struct	GenericTypeUpdatePattern
	A generic pattern that updates the types of operations based on a provided type converter. Used in KernelToMemref and OpsToFuncCalls passes to ensure that operations have the correct types after type conversion. More...
struct	IdentityConvolutionPattern
class	IfNode
	AST node for if conditional statements. More...
class	ImageAccessHelper
	Helper class to generate LLVM IR for accessing fields of the Image struct via an opaque pointer. Assumes the following C++ struct layout: struct Image { uint32_t _width; // Offset 0 uint32_t _height; // Offset 4 unsigned char *_data; // Offset 8 };. More...
class	IRPassManager
	Wrapper around MLIR PassManager to manage and run passes. More...
class	KernelData
class	KernelNode
	AST node for kernel definitions. More...
struct	KernelToMemref
class	Lexer
	Lexical analyzer for the picceler programming language. More...
struct	LoadImageToCall
class	Location
	Struct to hold location information. More...
class	MLIRGen
	MLIR code generator that converts AST nodes to MLIR constructs. This offers the initial IR generation from the AST. More...
class	ModuleNode
	AST node for the entire module. More...
struct	NeighbourhoodOpsToAffine
class	NumberNode
	AST node for numeric literals. More...
struct	ParameterizedFilterPattern
	Rewrites a filter op whose kernel depends on a single I64 parameter (sharpen's strength, box_blur/gaussian_blur's radius) into a convolution + kernel pair, using staticCalc when the parameter is a compile-time constant and dynamicBuilder otherwise. More...
class	Parser
	The Parser class that converts tokens into an AST. More...
struct	PassLogger
	Custom struct to log pass execution using spdlog. This struct implements the PassInstrumentation interface to hook into the pass execution lifecycle. More...
struct	PiccelerAddProfilingPass
	A pass that wraps every Picceler compute/IO op with piccelerTraceBegin/piccelerTraceEnd runtime calls. Excludes StringConstOp (instrumentation itself creates these) and ShowImageOp/ ReadNumberOp/ReadStringOp (block on a window/stdin, not compute – one call would swallow every real op's time into invisibility on the timeline), PrintOp (times stdout buffering, not compute), and KernelConstOp (sub-microsecond next to any real op, just adds a row). More...
struct	PiccelerFiltersToConvPass
	A pass that converts high-level image filter operations (like sharpen, blur, edge detect) into convolution operations with constant kernels. More...
struct	PiccelerKernelToMemrefPass
struct	PiccelerOpsToFuncCallsPass
	A pass that converts image operations (like load, show, save) into function calls. More...
struct	PiccelerToAffinePass
	A pass that converts Picceler operations to the Affine dialect. This is the crucial step where we lower from our high-level image processing operations to a more explicit representation that can be further lowered to LLVM IR. Each Picceler operation is matched and rewritten into one or more Affine loops that perform the equivalent computation. This pass also handles type conversions and ensures that necessary runtime functions (like piccelerCreateImage) are declared. More...
struct	PiccelerToLLVMIRPass
	A pass that helps convert last pieces of Picceler IR to LLVM IR. More...
struct	PrintToCalls
struct	ReadNumberToCall
struct	ReadStringToCall
struct	RemoveInvertPattern
class	ReturnNode
	AST node for "return" statements. More...
struct	RotateToAffine
struct	SameOperandPattern
	Canonicalization pattern for the BlendOp that replaces blend(img, img, weight) with img if both input images are the same. More...
struct	SaveImageToCall
	Pattern to lower SaveImageOp to a function call. More...
struct	ShowImageToCall
	Pattern to lower ShowImageOp to a function call. More...
struct	StringConstConverter
class	StringNode
	AST node for string literals. More...
struct	Token
	Represents a token produced by the lexer. More...
class	VariableNode
	AST node for variable references. More...

Typedefs
template<typename T>
using	Result = std::expected<T, CompileError>
	A type alias for the result of a compilation operation.

Enumerations
enum class	Channel : uint8_t { R = 0 , G = 1 , B = 2 , A = 3 , Count = 4 }
enum class	BinaryOperatorType : uint8_t { Add , Sub , Mul , Div , Equal , NotEqual , LessThan , LessEqual , GreaterThan , GreaterEqual }

Functions
template<typename T>
auto	getRawPointers (const std::vector< std::unique_ptr< T > > &vec)
std::ostream &	operator<< (std::ostream &os, const Token &token)
	Outputs a token to the given output stream.
mlir::Value	createFloatConstant (mlir::OpBuilder &builder, mlir::Location loc, double value)
	Common utility function to create a constant float value in MLIR.
mlir::Value	createIntConstant (mlir::OpBuilder &builder, mlir::Location loc, int64_t value)
	Common utility function to create a constant int value in MLIR.
mlir::affine::AffineParallelOp	createAffineParallel (mlir::ConversionPatternRewriter &rewriter, mlir::Location loc, mlir::ValueRange upperBounds, mlir::TypeRange resultTypes={}, llvm::ArrayRef< mlir::arith::AtomicRMWKind > reductions={})
	Builds an affine.parallel band with one induction variable per entry in upperBounds, lower bound 0 and step 1. Upper bounds may be dynamic SSA values, not just compile-time constants. Pass resultTypes/reductions to get a reduction band whose body must end with a matching affine.yield; left empty, the trivial affine.yield terminator is inserted automatically.
BinaryOperatorType	getBinaryOpType (const std::string &opStr)
mlir::Value	coerceValueToInt64 (mlir::OpBuilder &builder, mlir::Location loc, mlir::Value value, llvm::StringRef opName, llvm::StringRef argName)
	Coerces a given MLIR value to a 64-bit integer if possible, with special handling for constants.
Result< std::pair< mlir::Value, mlir::Value > >	getKernelNeighborhoodSize (mlir::OpBuilder &builder, mlir::Location loc, mlir::Value kernelOperand)
Result< KernelData >	calculateSharpenKernel (SharpenOp op, SharpenOpAdaptor adaptor)
Result< KernelData >	calculateBoxBlurKernel (BoxBlurOp op, BoxBlurOpAdaptor adaptor)
Result< KernelData >	calculateGaussianKernel (GaussianBlurOp op, GaussianBlurOpAdaptor adaptor)
Result< KernelData >	calculateEdgeDetectKernel (EdgeDetectOp op, EdgeDetectOpAdaptor adaptor)
Result< KernelData >	calculateEmbossKernel (EmbossOp op, EmbossOpAdaptor adaptor)
void	emitRuntimeRadiusBoundsCheck (mlir::ConversionPatternRewriter &rewriter, mlir::Location loc, mlir::Value radius)
Result< mlir::Value >	buildBoxBlurKernelDynamic (BoxBlurOp op, BoxBlurOpAdaptor adaptor, mlir::ConversionPatternRewriter &rewriter, mlir::Location loc)
Result< mlir::Value >	buildGaussianKernelDynamic (GaussianBlurOp op, GaussianBlurOpAdaptor adaptor, mlir::ConversionPatternRewriter &rewriter, mlir::Location loc)
Result< mlir::Value >	buildSharpenKernelDynamic (SharpenOp op, SharpenOpAdaptor adaptor, mlir::ConversionPatternRewriter &rewriter, mlir::Location loc)
mlir::func::FuncOp	ensureRuntimeFunc (mlir::ModuleOp module, mlir::StringRef name, llvm::ArrayRef< mlir::Type > inputs, llvm::ArrayRef< mlir::Type > results, mlir::PatternRewriter &rewriter, mlir::Location loc)
	Ensures that a runtime function is declared in the module. If the function does not exist, it is created.
mlir::Value	buildImageMemref (mlir::PatternRewriter &rewriter, mlir::Location loc, mlir::Value dataPtr, mlir::Value heightI64, mlir::Value widthI64)
	Builds a memref<?x?x4xi8> LLVM descriptor directly from a raw data pointer and dynamic height/width (both i64-typed, matching LLVMTypeConverter's default index type), emitting a single UnrealizedConversionCastOp from the LLVM struct to the memref type. This cast is a structural identity (LLVMTypeConverter converts memref<?x?x4xi8> to exactly this struct type), so it folds away cleanly under ReconcileUnrealizedCastsPass, unlike a cast straight from !llvm.ptr to memref.
mlir::FailureOr< mlir::Value >	hoistArithChainToFunctionTop (mlir::ConversionPatternRewriter &rewriter, mlir::func::FuncOp funcOp, mlir::Operation *insertBefore, mlir::Value value)
	Relocates value's defining chain to funcOp's top level, since MLIR requires an affine.parallel/affine.apply dim or symbol to be defined there. Only safe for a pure arith chain (a compile-time-constant kernel size); fails otherwise.
The following functions create instances of the various passes used in the compilation process.
Each function corresponds to a specific pass that transforms the IR in a particular way.
std::unique_ptr< mlir::Pass >	createPiccelerKernelToMemrefPass ()
std::unique_ptr< mlir::Pass >	createPiccelerOpsToFuncCallsPass ()
std::unique_ptr< mlir::Pass >	createPiccelerToAffinePass ()
std::unique_ptr< mlir::Pass >	createPiccelerToLLVMIRPass ()
std::unique_ptr< mlir::Pass >	createPiccelerFiltersToConvPass ()
std::unique_ptr< mlir::Pass >	createPiccelerAddProfilingPass ()

Typedef Documentation

◆ Result

template<typename T>

using picceler::Result = std::expected<T, CompileError>

A type alias for the result of a compilation operation.

Template Parameters

T	The type of the successful result.

Enumeration Type Documentation

◆ BinaryOperatorType

enum class picceler::BinaryOperatorType : uint8_t

strong

Enumerator
Add
Sub
Mul
Div
Equal
NotEqual
LessThan
LessEqual
GreaterThan
GreaterEqual

◆ Channel

enum class picceler::Channel : uint8_t

strong

Enumerator
R
G
B
A
Count

Function Documentation

◆ buildBoxBlurKernelDynamic()

Result< mlir::Value > picceler::buildBoxBlurKernelDynamic	(	BoxBlurOp	op,
		BoxBlurOpAdaptor	adaptor,
		mlir::ConversionPatternRewriter &	rewriter,
		mlir::Location	loc )

◆ buildGaussianKernelDynamic()

Result< mlir::Value > picceler::buildGaussianKernelDynamic	(	GaussianBlurOp	op,
		GaussianBlurOpAdaptor	adaptor,
		mlir::ConversionPatternRewriter &	rewriter,
		mlir::Location	loc )

◆ buildImageMemref()

mlir::Value picceler::buildImageMemref	(	mlir::PatternRewriter &	rewriter,
		mlir::Location	loc,
		mlir::Value	dataPtr,
		mlir::Value	heightI64,
		mlir::Value	widthI64 )

Builds a memref<?x?x4xi8> LLVM descriptor directly from a raw data pointer and dynamic height/width (both i64-typed, matching LLVMTypeConverter's default index type), emitting a single UnrealizedConversionCastOp from the LLVM struct to the memref type. This cast is a structural identity (LLVMTypeConverter converts memref<?x?x4xi8> to exactly this struct type), so it folds away cleanly under ReconcileUnrealizedCastsPass, unlike a cast straight from !llvm.ptr to memref.

◆ buildSharpenKernelDynamic()

Result< mlir::Value > picceler::buildSharpenKernelDynamic	(	SharpenOp	op,
		SharpenOpAdaptor	adaptor,
		mlir::ConversionPatternRewriter &	rewriter,
		mlir::Location	loc )

◆ calculateBoxBlurKernel()

Result< KernelData > picceler::calculateBoxBlurKernel	(	BoxBlurOp	op,
		BoxBlurOpAdaptor	adaptor )

◆ calculateEdgeDetectKernel()

Result< KernelData > picceler::calculateEdgeDetectKernel	(	EdgeDetectOp	op,
		EdgeDetectOpAdaptor	adaptor )

◆ calculateEmbossKernel()

Result< KernelData > picceler::calculateEmbossKernel	(	EmbossOp	op,
		EmbossOpAdaptor	adaptor )

◆ calculateGaussianKernel()

Result< KernelData > picceler::calculateGaussianKernel	(	GaussianBlurOp	op,
		GaussianBlurOpAdaptor	adaptor )

◆ calculateSharpenKernel()

Result< KernelData > picceler::calculateSharpenKernel	(	SharpenOp	op,
		SharpenOpAdaptor	adaptor )

◆ coerceValueToInt64()

mlir::Value picceler::coerceValueToInt64	(	mlir::OpBuilder &	builder,
		mlir::Location	loc,
		mlir::Value	value,
		llvm::StringRef	opName,
		llvm::StringRef	argName )

Coerces a given MLIR value to a 64-bit integer if possible, with special handling for constants.

Parameters

builder	The MLIR OpBuilder to use for creating new operations if coercion is needed.
loc	The location to use for any new operations created during coercion.
value	The MLIR value to coerce.
opName	The name of the operation for error reporting purposes.
argName	The name of the argument for error reporting purposes.

Returns: The coerced MLIR value as a 64-bit integer.

Exceptions

std::runtime_error if the value cannot be coerced to a 64-bit integer, or if a floating-point constant is not a whole number or is out of range.

◆ createAffineParallel()

mlir::affine::AffineParallelOp picceler::createAffineParallel	(	mlir::ConversionPatternRewriter &	rewriter,
		mlir::Location	loc,
		mlir::ValueRange	upperBounds,
		mlir::TypeRange	resultTypes = {},
		llvm::ArrayRef< mlir::arith::AtomicRMWKind >	reductions = {} )

Builds an affine.parallel band with one induction variable per entry in upperBounds, lower bound 0 and step 1. Upper bounds may be dynamic SSA values, not just compile-time constants. Pass resultTypes/reductions to get a reduction band whose body must end with a matching affine.yield; left empty, the trivial affine.yield terminator is inserted automatically.

Parameters

rewriter	The rewriter to use for creating the affine.parallel op and its bounds.
loc	The location to associate with the created ops.
upperBounds	One SSA value per loop dimension, used as the (exclusive) upper bound.
resultTypes	Result types for a reduction band, or empty for a plain loop.
reductions	One reduction kind per result, matching resultTypes.

Returns: The created affine.parallel op.

◆ createFloatConstant()

mlir::Value picceler::createFloatConstant	(	mlir::OpBuilder &	builder,
		mlir::Location	loc,
		double	value )

Common utility function to create a constant float value in MLIR.

Parameters

builder	The MLIR OpBuilder to use for creating the operation.
loc	The MLIR Location to associate with the operation.
value	The double value to create as a constant.

Returns: An MLIR Value representing the constant float.

◆ createIntConstant()

mlir::Value picceler::createIntConstant	(	mlir::OpBuilder &	builder,
		mlir::Location	loc,
		int64_t	value )

Common utility function to create a constant int value in MLIR.

Parameters

builder	The MLIR OpBuilder to use for creating the operation.
loc	The MLIR Location to associate with the operation.
value	The int64_t value to create as a constant.

Returns: An MLIR Value representing the constant int.

◆ createPiccelerAddProfilingPass()

std::unique_ptr< mlir::Pass > picceler::createPiccelerAddProfilingPass ( )

◆ createPiccelerFiltersToConvPass()

std::unique_ptr< mlir::Pass > picceler::createPiccelerFiltersToConvPass ( )

◆ createPiccelerKernelToMemrefPass()

std::unique_ptr< mlir::Pass > picceler::createPiccelerKernelToMemrefPass ( )

◆ createPiccelerOpsToFuncCallsPass()

std::unique_ptr< mlir::Pass > picceler::createPiccelerOpsToFuncCallsPass ( )

◆ createPiccelerToAffinePass()

std::unique_ptr< mlir::Pass > picceler::createPiccelerToAffinePass ( )

◆ createPiccelerToLLVMIRPass()

std::unique_ptr< mlir::Pass > picceler::createPiccelerToLLVMIRPass ( )

◆ emitRuntimeRadiusBoundsCheck()

void picceler::emitRuntimeRadiusBoundsCheck	(	mlir::ConversionPatternRewriter &	rewriter,
		mlir::Location	loc,
		mlir::Value	radius )

◆ ensureRuntimeFunc()

mlir::func::FuncOp picceler::ensureRuntimeFunc	(	mlir::ModuleOp	module,
		mlir::StringRef	name,
		llvm::ArrayRef< mlir::Type >	inputs,
		llvm::ArrayRef< mlir::Type >	results,
		mlir::PatternRewriter &	rewriter,
		mlir::Location	loc )

Ensures that a runtime function is declared in the module. If the function does not exist, it is created.

Parameters

module	The MLIR module to check.
name	The name of the function.
inputs	The input types of the function.
results	The result types of the function.
rewriter	The pattern rewriter to use for creating the function.
loc	The location to use for the function.

Returns: The function operation.

◆ getBinaryOpType()

BinaryOperatorType picceler::getBinaryOpType ( const std::string & opStr )

inline

◆ getKernelNeighborhoodSize()

Result< std::pair< mlir::Value, mlir::Value > > picceler::getKernelNeighborhoodSize	(	mlir::OpBuilder &	builder,
		mlir::Location	loc,
		mlir::Value	kernelOperand )

◆ getRawPointers()

template<typename T>

auto picceler::getRawPointers ( const std::vector< std::unique_ptr< T > > & vec )

◆ hoistArithChainToFunctionTop()

mlir::FailureOr< mlir::Value > picceler::hoistArithChainToFunctionTop	(	mlir::ConversionPatternRewriter &	rewriter,
		mlir::func::FuncOp	funcOp,
		mlir::Operation *	insertBefore,
		mlir::Value	value )

Relocates value's defining chain to funcOp's top level, since MLIR requires an affine.parallel/affine.apply dim or symbol to be defined there. Only safe for a pure arith chain (a compile-time-constant kernel size); fails otherwise.

◆ operator<<()

std::ostream & picceler::operator<<	(	std::ostream &	os,
		const Token &	token )

Outputs a token to the given output stream.

Parameters

os	The output stream.
token	The token to output.

Returns: The output stream.

Namespaces

Classes

Typedefs

Enumerations

Functions

Typedef Documentation

◆ Result

Enumeration Type Documentation

◆ BinaryOperatorType

◆ Channel

Function Documentation

◆ buildBoxBlurKernelDynamic()

◆ buildGaussianKernelDynamic()

◆ buildImageMemref()

◆ buildSharpenKernelDynamic()

◆ calculateBoxBlurKernel()

◆ calculateEdgeDetectKernel()

◆ calculateEmbossKernel()

◆ calculateGaussianKernel()

◆ calculateSharpenKernel()

◆ coerceValueToInt64()

◆ createAffineParallel()

◆ createFloatConstant()

◆ createIntConstant()

◆ createPiccelerAddProfilingPass()

◆ createPiccelerFiltersToConvPass()

◆ createPiccelerKernelToMemrefPass()

◆ createPiccelerOpsToFuncCallsPass()

◆ createPiccelerToAffinePass()

◆ createPiccelerToLLVMIRPass()

◆ emitRuntimeRadiusBoundsCheck()

◆ ensureRuntimeFunc()

◆ getBinaryOpType()

◆ getKernelNeighborhoodSize()

◆ getRawPointers()

◆ hoistArithChainToFunctionTop()

◆ operator<<()