Parser::State::Match Struct Reference

Classes
struct	Clot
struct	FieldMapping
struct	HdrLenIncStop
struct	Save
struct	Set

Public Types
enum	flags_t { OFFSET = 1 , ROTATE = 2 }

Public Member Functions
	Match (int lineno, gress_t, State *n)
	Match (int lineno, gress_t, State *s, match_t m, VECTOR(pair_t) &data)
std::set< Match * >	get_all_preds ()
std::set< Match * >	get_all_preds_impl (std::set< Match * > &visited)
void	pass1 (Parser *pa, State *state)
void	pass2 (Parser *pa, State *state)
void	unmark_reachable (Parser *, State *state, bitvec &unreach)
template<class REGS>
void	write_common_row_config (REGS &, Parser *, State *, int, Match *, json::map &)
template<class REGS>
void	write_config (REGS &, json::vector &)
template<class REGS>
void	write_config (REGS &, Parser *, State *, Match *, json::map &)
template<class EA_REGS>
void	write_counter_config (EA_REGS &) const
template<>
void	write_counter_config (Target::JBay::parser_regs::_memory::_ml_ea_row &ea_row) const
template<>
void	write_counter_config (Target::Tofino::parser_regs::_memory::_ml_ea_row &ea_row) const
template<class REGS>
int	write_load_config (REGS &, Parser *, State *, int) const
template<>
int	write_load_config (Target::JBay::parser_regs &regs, Parser *pa, State *state, int row) const
template<>
int	write_load_config (Target::Tofino::parser_regs &regs, Parser *pa, State *state, int row) const
template<class REGS>
void	write_lookup_config (REGS &, State *, int) const
template<>
void	write_lookup_config (Target::JBay::parser_regs &regs, State *state, int r) const
template<>
void	write_lookup_config (Target::Tofino::parser_regs &regs, State *state, int row) const
template<class REGS>
void	write_row_config (REGS &, Parser *, State *, int, Match *, json::map &)
template<>
void	write_row_config (Target::JBay::parser_regs &regs, Parser *pa, State *state, int row, Match *def, json::map &ctxt_json)
template<class REGS>
void	write_saves (REGS &regs, Match *def, void *output_map, int &max_off, unsigned &used, int csum_8b, int csum_16b)
template<class REGS>
void	write_sets (REGS &regs, Match *def, void *output_map, unsigned &used, int csum_8b, int csum_16b)

Public Attributes
int	buf_req = -1
std::vector< Clot * >	clots
std::vector< Checksum >	csum
int	ctr_imm_amt = 0
CounterInit *	ctr_instr = nullptr
int	ctr_ld_src = 0
int	ctr_load = 0
bool	ctr_stack_pop = false
bool	ctr_stack_push = false
bool	ctr_stack_upd_w_top = false
int	disable_partial_hdr_err = -1
std::vector< FieldMapping >	field_mapping
bool	has_narrow_to_wide_extract = false
struct Parser::State::Match::HdrLenIncStop	hdr_len_inc_stop
int	intr_md_bits = 0
int	lineno
MatchKey	load
match_t	match
std::vector< const Phv::Ref * >	narrow_to_wide_16b_8
std::vector< const Phv::Ref * >	narrow_to_wide_32b_16
std::vector< const Phv::Ref * >	narrow_to_wide_32b_8
Ref	next
int	offset_inc = 0
bool	offset_rst = false
int	partial_hdr_err_proc = -1
PriorityUpdate	priority
int	row = -1
std::vector< Save * >	save
std::vector< Set * >	set
int	shift = 0
State *	state = nullptr
int	value_set_handle = -1
std::string	value_set_name
int	value_set_size = 0

Member Function Documentation

get_all_preds()

std::set< Parser::State::Match * > Parser::State::Match::get_all_preds

(

)

Extractor config tracking and register config code Different tofino models have very different ways in which their parser extractors are managed, but all are common in that there are multiple extractions that can happen in parallel in a single parser match tcam row. We manage this by having a target-specific 'output_map' object passed via a void * to target-sepcific write_output_config methods along with an unsigned used mask that tracks which or how many extractors have been used, so as to issue errors for conflicting uses.

The setup_phv_output_map method creates the target specific output_map object that will be passed to subsequent write_output_config calls to deal with each individual extract. Finally, mark_unused_output_map is called to deal with any register setup needed for unused extractors. They're called 'outputs' as the are concerned with outputting PHV values from the parser.

PHV outputs are split into 'saves' and 'sets' which come from different syntax in the asm source. 'saves' copy data from the input buffer into PHVs, while 'sets' write constants into the PHVs. Different targets have different constraints on how flexible they are for saves vs sets, so some want to do saves first and other sets

• tofino1: do saves first (why? sets seem more constrained, but there's an issue with ganging smaller extractors to write larger PHVs)
• tofino2: do sets first as some extractors can only do saves

FIXME – should probably refactor this into a more C++ style base class pointer with derived classes for each target. Should move the 'used' mask into that object as well. Alternately, could move the entire setup to mark_unused process into a target specific method.

Member Data Documentation

has_narrow_to_wide_extract

bool Parser::State::Match::has_narrow_to_wide_extract = false

Data for narrow to wide extraction analysis, flag and vector of affected PHV locations