nes-pascal

Array Implementation and Measurements (0.5.8)

Historical note: fixed-pool figures below record the 0.5.8 state. Milestone 0.5.11 now reserves 1 expression byte and 2 separate loop-cache bytes for this benchmark, recovering 13 ZP bytes.

English

Português (Brasil)

Milestone 0.5.8 adds fixed-size global arrays as a compiler language construct, not as a runtime subsystem. ArrayType retains element type, lower and upper bounds, element count, and total one-byte-element size. Resolved array reads and writes remain explicit through semantic analysis, memory layout, temporary analysis, and ca65 emission.

Lowering and memory model

• Arrays use array[$00..$NN] of byte|boolean; only lower bound $00 is supported.
• Arrays are allocated contiguously in declaration order in regular user RAM. They are deliberately excluded from automatic Zero Page promotion.
• Constant indexes become static symbol offsets. Variable indexes are evaluated into X and use absolute indexed addressing.
• An indexed assignment evaluates the index first, preserves it temporarily on the hardware stack, evaluates the value, and then stores through ,x. No fixed compiler symbol or array runtime state is introduced.
• An array element used by Boolean control flow feeds the 0.5.7 branch-oriented lowering directly. Stored Boolean elements remain canonical $00/$01.
• Compile-time-known indexes are folded and checked. Variable indexes have no runtime bounds check or metadata cost.

Array benchmark

The new arrays workload fills byte and Boolean arrays in loops, reads and writes constant and variable indexes, performs indexed arithmetic, and branches on Boolean elements.

Metric	Result
PRG code	382 B
PRG occupied	388 B
Instructions	182
Estimated static base cycles	569
Expression tree depth	2
Maximum live expression temporaries	1
Fixed temporary pool reservation	16 B
Temporaries/cache actually required	3 B
Array element storage	16 B regular RAM
Other regular runtime/user storage	4 B
Automatically promoted scalar storage	3 B ZP
ZP benchmark allocated/reserved	28 B
ZP allocator-visible free	125 B
Regular allocator-visible free	1,516 B
Runtime features	None

The three required temporary/cache bytes are one reusable expression byte for Sum + Values[Index] and two existing for_limit_* bytes. Array indexing itself adds no fixed or actual Zero Page temporary. Constant-only indexing requires no index temporary.

Estimated static base cycles uses the deterministic benchmark convention: each emitted instruction is counted once at its base Ricoh 2A03 cost, branches are not taken, and loop counts, page crossing, interrupts, and DMA are excluded.

Pre-existing corpus regression check

The complete 16-program 0.5.5/0.5.7 corpus was measured before and after array support. Every listed metric is identical; no-array programs emit no array runtime, descriptor, metadata, RAM, or Zero Page state.

Benchmark	PRG code/occupied B	Instructions	Est. cycles	Live temps	ZP temp required	Non-ZP allocated
minimal	239/245	108	367	0	0 B	7 B
arithmetic	251/257	115	383	0	0 B	7 B
boolean_expressions	382/388	170	525	0	0 B	12 B
conditionals	282/288	128	415	0	0 B	5 B
loops	317/323	146	460	0	0 B	4 B
counting	488/494	216	700	0	6 B	9 B
procedures	289/295	134	466	0	0 B	4 B
procedure_parameters	350/356	155	524	0	0 B	11 B
controller_input	704/710	318	945	0	0 B	265 B
sprite_support	583/589	273	911	0	0 B	326 B
metasprite_player	1,303/1,309	489	1,599	0	0 B	272 B
sprite_animation	1,875/1,881	614	2,035	0	0 B	276 B
palette_support	812/818	342	1,106	0	0 B	306 B
background_updates	2,166/2,172	522	1,773	1	0 B	995 B
frame_callbacks	272/278	124	438	0	0 B	6 B
gameplay_full_stack	3,350/3,356	815	2,712	1	0 B	1,260 B

Deliberately deferred

Runtime bounds checking, non-zero lower bounds, array parameters/returns, multidimensional and dynamic arrays, bit packing, arrays of records, pointer semantics, a generic runtime array system, and temporary allocator redesign remain outside this milestone.

This site is open source. Improve this page.