Compiler mixin for targeting x86-64. More...

#include <CompilerX64.hpp>

Inheritance diagram for tpde::x64::CompilerX64< Adaptor, Derived, BaseTy, Config >:

Classes
class	CallBuilder
	Helper class for building call sequences. More...

Public Types
enum class	Jump { jo = 0 , jno , jb , jae , je , jne , jbe , ja , js , jns , jp , jnp , jl , jge , jle , jg , jmp }
	Jump conditions. More...

Public Member Functions
void	prologue_begin (CCAssigner *cc_assigner) noexcept
	Begin prologue, prepare for assigning arguments.

std::optional< i32 >	prologue_assign_arg_part (ValuePart &&vp, CCAssignment cca) noexcept
	Assign argument part.

void	prologue_end (CCAssigner *cc_assigner) noexcept
	Finish prologue.

void	alloca_fixed (u64 size, u32 align, ValuePart &res) noexcept
	Dynamic alloca of a fixed-size region.

void	alloca_dynamic (u64 elem_size, ValuePart &&count, u32 align, ValuePart &res) noexcept
	Dynamic alloca of a dynamically-sized region (elem_size * count bytes).

void	materialize_constant (const u64 *data, RegBank bank, u32 size, AsmReg dst) noexcept
	Materialize constant into a register.

void	generate_raw_jump (Jump jmp, Label target) noexcept
	Generate jump instruction to target label.

void	generate_raw_set (Jump cc, AsmReg dst, bool zext=true) noexcept
	Set dst to 1 if cc is true, otherwise set it to zero.

void	generate_raw_mask (Jump cc, AsmReg dst) noexcept
	Set all bits of dst to 1 if cc is true, otherwise set it to zero.

void	generate_raw_cmov (Jump cc, AsmReg dst, AsmReg src, bool is_64) noexcept
	Move src into dst if cc is true, otherwise do nothing.

void	generate_raw_intext (AsmReg dst, AsmReg src, bool sign, u32 from, u32 to) noexcept
	Integer extension.

void	generate_raw_bfi (AsmReg dst, AsmReg src, u32 lsb, u32 width) noexcept
	Bitfield insert. Needs a temporary register, src is not modified.

void	generate_raw_bfiz (AsmReg dst, AsmReg src, u32 lsb, u32 width) noexcept
	Bitfield insert in zero.

void	generate_call (std::variant< SymRef, ValuePart > &&target, std::span< CallArg > arguments, typename Base::ValueRef *result, bool variable_args=false)
	Generate a function call.

ScratchReg	tls_get_addr (SymRef sym, TLSModel model) noexcept
	Generate code sequence to load address of sym into a register.

Public Member Functions inherited from tpde::CompilerBase< Adaptor, Derived, PlatformConfig >
	CompilerBase (Adaptor *adaptor)
	Initialize a CompilerBase, should be called by the derived classes.

Derived *	derived ()
	shortcut for casting to the Derived class so that overloading works

bool	compile ()
	Compile the functions returned by Adaptor::funcs.

void	reset ()
	Reset any leftover data from the previous compilation such that it will not affect the next compilation.

CCAssigner *	cur_cc_assigner () noexcept
	Get CCAssigner for current function.

void	release_assignment (ValLocalIdx local_idx, ValueAssignment *) noexcept
	Release an assignment when reference count drops to zero, either frees the assignment immediately or delays free to the end of the live range.

void	init_variable_ref (ValLocalIdx local_idx, u32 var_ref_data) noexcept
	Init a variable-ref assignment.

void	init_variable_ref (IRValueRef value, u32 var_ref_data) noexcept
	Init a variable-ref assignment.

void	prologue_assign_arg (CCAssigner *cc_assigner, u32 arg_idx, IRValueRef arg, u32 align=1, bool allow_split=false) noexcept
	Assign function argument in prologue.

AsmReg	gval_as_reg (GenericValuePart &gv) noexcept
	Get generic value part into a single register, evaluating expressions and materializing immediates as required.

AsmReg	gval_as_reg_reuse (GenericValuePart &gv, ScratchReg &dst) noexcept
	Like gval_as_reg; if the GenericValuePart owns a reusable register (either a ScratchReg, possibly due to materialization, or a reusable ValuePartRef), store it in dst.

void	label_place (Label label) noexcept
	Convenience function to place a label at the current position.

i32	allocate_stack_slot (u32 size) noexcept
	Allocate a static stack slot.

void	free_stack_slot (u32 slot, u32 size) noexcept
	Free a static stack slot.

ValueRef	val_ref (IRValueRef value) noexcept
	Get a using reference to a value.

std::pair< ValueRef, ValuePartRef >	val_ref_single (IRValueRef value) noexcept
	Get a using reference to a single-part value and provide direct access to the only part.

ValueRef	result_ref (IRValueRef value) noexcept
	Get a defining reference to a value.

std::pair< ValueRef, ValuePartRef >	result_ref_single (IRValueRef value) noexcept
	Get a defining reference to a single-part value and provide direct access to the only part.

ValueRef	result_ref_alias (IRValueRef dst, ValueRef &&src) noexcept
	Make dst an alias for src, which must be a non-constant value with an identical part configuration.

ValueRef	result_ref_stack_slot (IRValueRef value, AssignmentPartRef base, i32 off) noexcept
	Initialize value as a pointer into a stack variable (i.e., a value allocated from cur_static_allocas() or similar) with an offset.

Reg	select_reg (RegBank bank, u64 exclusion_mask) noexcept
	Select an available register, evicting loaded values if needed.

void	reload_to_reg (AsmReg dst, AssignmentPartRef ap) noexcept
	Reload a value part from memory or recompute variable address.

void	allocate_spill_slot (AssignmentPartRef ap) noexcept
	Allocate a stack slot for an assignment.

void	spill (AssignmentPartRef ap) noexcept
	Ensure the value is spilled in its stack slot (except variable refs).

void	evict (AssignmentPartRef ap) noexcept
	Evict the value from its register, spilling if needed, and free register.

void	evict_reg (Reg reg) noexcept
	Evict the value from the register, spilling if needed, and free register.

void	free_reg (Reg reg) noexcept
	Free the register. Requires that the contained value is already spilled.

void	generate_uncond_branch (IRBlockRef target) noexcept
	Generate an unconditional branch at the end of a basic block.

void	generate_cond_branch (Jump jmp, IRBlockRef true_target, IRBlockRef false_target) noexcept
	Generate an conditional branch at the end of a basic block.

void	generate_switch (ScratchReg &&cond, u32 width, IRBlockRef default_block, std::span< const std::pair< u64, IRBlockRef > > cases) noexcept
	Generate a switch at the end of a basic block.

RegisterFile::RegBitSet	spill_before_branch (bool force_spill=false) noexcept
	Spill values that need to be spilled for later blocks.

void	release_spilled_regs (typename RegisterFile::RegBitSet) noexcept
	Free registers marked by spill_before_branch().

void	release_regs_after_return () noexcept
	When reaching a point in the function where no other blocks will be reached anymore, use this function to release register assignments after the end of that block so the compiler does not accidentally use registers which don't contain any values.

void	begin_branch_region () noexcept
	Indicate beginning of region where value-state must not change.

void	end_branch_region () noexcept
	Indicate end of region where value-state must not change.

void	generate_branch_to_block (Jump jmp, IRBlockRef target, bool needs_split, bool last_inst) noexcept
	Generate a branch to a basic block; execution continues afterwards.

bool	branch_needs_split (IRBlockRef target) noexcept
	Whether branch to a block requires additional instructions and therefore a direct jump to the block is not possible.

Public Attributes
u32	scalar_arg_count = 0xFFFF'FFFF
	For vararg functions only: number of scalar and xmm registers used.

u32	max_callee_stack_arg_size
	For functions without dynamic allocas, the largest size used for arguments passed on the stack to callees.

bool	preserve_flags
	Whether flags must be preserved when materializing constants etc.

SymRef	sym_tls_get_addr
	Symbol for __tls_get_addr.

Public Attributes inherited from tpde::CompilerBase< Adaptor, Derived, PlatformConfig >
u32	frame_size
	The current size of the stack frame.

bool	has_dynamic_alloca
	Whether the stack frame might have dynamic alloca.

bool	is_leaf_function
	Whether the function is guaranteed to be a leaf function.

bool	generated_call
	Whether the function actually includes a call.

bool	frame_used
	Whether the stack frame is used.

util::SmallVector< i32, 16 >	fixed_free_lists [5]
	Free-Lists for 1/2/4/8/16 sized allocations.

std::unordered_map< u32, std::vector< i32 > >	dynamic_free_lists
	Free-Lists for all other sizes.

bool	generating_branch
	Whether we are currently in the middle of generating branch-related code and therefore must not change any value-related state.

Detailed Description

template<IRAdaptor Adaptor, typename Derived, template< typename, typename, typename > typename BaseTy = CompilerBase, typename Config = PlatformConfig>
struct tpde::x64::CompilerX64< Adaptor, Derived, BaseTy, Config >

Compiler mixin for targeting x86-64.

Definition at line 269 of file CompilerX64.hpp.

Member Enumeration Documentation

◆ Jump

template<IRAdaptor Adaptor, typename Derived, template< typename, typename, typename > typename BaseTy = CompilerBase, typename Config = PlatformConfig>

enum class tpde::x64::CompilerX64::Jump

strong

Jump conditions.

Enumerator
jo	Jump if overflow (OF=1).
jno	Jump if not overflow (OF=0).
jb	Jump if below/if carry (CF=1).
jae	Jump if above or equal/if not carry (CF=0).
je	Jump if equal/if zero (ZF=1).
jne	Jump if not equal/if not zero (ZF=0).
jbe	Jump if below or equal (CF=1 or ZF=1).
ja	Jump if above (CF=0 and ZF=0).
js	Jump if sign (SF=1).
jns	Jump if not sign (SF=0).
jp	Jump if parity even (PF=1).
jnp	Jump if parity odd (PF=0).
jl	Jump if less (SF!=OF).
jge	Jump if greater or equal (SF=OF).
jle	Jump if less or equal (ZF=1 or SF!=OF).
jg	Jump if greater (ZF=0 and SF=OF).
jmp	Unconditional jump.

Definition at line 454 of file CompilerX64.hpp.

Member Function Documentation

◆ alloca_dynamic()

template<IRAdaptor Adaptor, typename Derived, template< typename, typename, typename > typename BaseTy, typename Config>

void tpde::x64::CompilerX64< Adaptor, Derived, BaseTy, Config >::alloca_dynamic	(	u64	elem_size,
		ValuePart &&	count,
		u32	align,
		ValuePart &	res )

noexcept

Dynamic alloca of a dynamically-sized region (elem_size * count bytes).

count must have a size of 64 bit.

Definition at line 1156 of file CompilerX64.hpp.

◆ generate_call()

template<IRAdaptor Adaptor, typename Derived, template< typename, typename, typename > typename BaseTy, typename Config>

void tpde::x64::CompilerX64< Adaptor, Derived, BaseTy, Config >::generate_call	(	std::variant< SymRef, ValuePart > &&	target,
		std::span< CallArg >	arguments,
		typename Base::ValueRef *	result,
		bool	variable_args = false )

Generate a function call.

This will get the arguments into the correct registers according to the calling convention, clear non-callee-saved registers from the register file (make sure you do not have any fixed assignments left over) and fill the result registers (the u8 in the ScratchReg pair indicates the register bank)

Targets can be a symbol (call to PLT with relocation), or an indirect call to a ValuePart. Result is an optional reference.

Definition at line 1796 of file CompilerX64.hpp.

◆ generate_raw_bfiz()

template<IRAdaptor Adaptor, typename Derived, template< typename, typename, typename > class BaseTy, typename Config>

void tpde::x64::CompilerX64< Adaptor, Derived, BaseTy, Config >::generate_raw_bfiz	(	AsmReg	dst,
		AsmReg	src,
		u32	lsb,
		u32	width )

noexcept

Bitfield insert in zero.

src is not modified, but src and dst must be different.

Definition at line 1625 of file CompilerX64.hpp.

◆ generate_raw_intext()

template<IRAdaptor Adaptor, typename Derived, template< typename, typename, typename > class BaseTy, typename Config>

void tpde::x64::CompilerX64< Adaptor, Derived, BaseTy, Config >::generate_raw_intext	(	AsmReg	dst,
		AsmReg	src,
		bool	sign,
		u32	from,
		u32	to )

noexcept

Integer extension.

Might need a temporary register, src is not modified, might clobber flags.

Definition at line 1524 of file CompilerX64.hpp.

◆ generate_raw_set()

template<IRAdaptor Adaptor, typename Derived, template< typename, typename, typename > class BaseTy, typename Config>

void tpde::x64::CompilerX64< Adaptor, Derived, BaseTy, Config >::generate_raw_set	(	Jump	cc,
		AsmReg	dst,
		bool	zext = true )

noexcept

Set dst to 1 if cc is true, otherwise set it to zero.

If zext is false, only the lowest 8 bit are set. Flags are not clobbered.

Definition at line 1489 of file CompilerX64.hpp.

◆ prologue_assign_arg_part()

template<IRAdaptor Adaptor, typename Derived, template< typename, typename, typename > typename BaseTy, typename Config>

std::optional< i32 > tpde::x64::CompilerX64< Adaptor, Derived, BaseTy, Config >::prologue_assign_arg_part	(	ValuePart &&	vp,
		CCAssignment	cca )

noexcept

Assign argument part.

Returns the stack offset if the value should be initialized as stack variable.

Definition at line 638 of file CompilerX64.hpp.

◆ tls_get_addr()

template<IRAdaptor Adaptor, typename Derived, template< typename, typename, typename > typename BaseTy, typename Config>

CompilerX64< Adaptor, Derived, BaseTy, Config >::ScratchReg tpde::x64::CompilerX64< Adaptor, Derived, BaseTy, Config >::tls_get_addr	(	SymRef	sym,
		TLSModel	model )

noexcept

Generate code sequence to load address of sym into a register.

This will generate a function call for dynamic TLS access models.

Definition at line 1930 of file CompilerX64.hpp.

Member Data Documentation

◆ max_callee_stack_arg_size

template<IRAdaptor Adaptor, typename Derived, template< typename, typename, typename > typename BaseTy = CompilerBase, typename Config = PlatformConfig>

u32 tpde::x64::CompilerX64< Adaptor, Derived, BaseTy, Config >::max_callee_stack_arg_size

For functions without dynamic allocas, the largest size used for arguments passed on the stack to callees.

This size is added to the stack pointer subtraction/addition in prologue/epilogue to avoid stack pointer adjustments at call sites.

Definition at line 345 of file CompilerX64.hpp.

The documentation for this struct was generated from the following file:

tpde/include/tpde/x64/CompilerX64.hpp

Classes

Public Types

Public Member Functions

Public Attributes

Detailed Description

Member Enumeration Documentation

◆ Jump

Member Function Documentation

◆ alloca_dynamic()

◆ generate_call()

◆ generate_raw_bfiz()

◆ generate_raw_intext()

◆ generate_raw_set()

◆ prologue_assign_arg_part()

◆ tls_get_addr()

Member Data Documentation

◆ max_callee_stack_arg_size