learning...

This commit is contained in:
ed
2026-07-07 01:22:09 -04:00
parent 93bab89f76
commit 6eee0249ff
5 changed files with 127 additions and 74 deletions
+2 -1
View File
@@ -206,7 +206,8 @@
* writes at this granularity. * writes at this granularity.
* *
* ============================================================================*/ * ============================================================================*/
#define tape_regs(...) (__VA_ARGS__) #define atom_reads(...) (__VA_ARGS__)
#define atom_writes(...) (__VA_ARGS__)
/* ============================================================================ /* ============================================================================
* ATOM ANNOTATION MACROS * ATOM ANNOTATION MACROS
-18
View File
@@ -189,22 +189,6 @@ internal MipsAtom_(mips_flush_icache) {
, mac_yield() , mac_yield()
}; };
typedef Struct_(Binds_SyncPrimCursor) {
U4 PrimtiveArena_Used;
U4 PrimtiveBase;
};
internal MipsAtom_(sync_prim_cursor) {
/* Pop the C-struct address and base address from the tape */
// Note(Ed): Argument shuffle....
load_word(R_AT, R_TapePtr, O_(Binds_SyncPrimCursor,PrimtiveArena_Used)),
load_word(R_T0, R_TapePtr, O_(Binds_SyncPrimCursor,PrimtiveBase)),
add_ui_1( R_TapePtr, S_(Binds_SyncPrimCursor)),
/* Calculate byte offset and store directly back to RAM */
sub_u(R_T0, R_PrimCursor, R_T0), // R_T0 = PrimitiveArea_Used(R_AT) - R_PrimCur
store_word(R_T0, R_AT, 0), // PrimitiveBase(R_AT)[0] = R_T0
mac_yield()
};
internal MipsAtom_(set_gte_world) { internal MipsAtom_(set_gte_world) {
/* Pop matrix address from tape into R_T3 ($11) */ /* Pop matrix address from tape into R_T3 ($11) */
load_word(R_T3, R_TapePtr, 0), load_word(R_T3, R_TapePtr, 0),
@@ -221,7 +205,6 @@ internal MipsAtom_(set_gte_world) {
mac_yield() mac_yield()
}; };
// TODO(Ed): I'm not sure yet if the bindings are redundant with the floortri atom yet. // TODO(Ed): I'm not sure yet if the bindings are redundant with the floortri atom yet.
typedef Struct_(Binds_CubeTri) { typedef Struct_(Binds_CubeTri) {
@@ -278,7 +261,6 @@ internal MipsAtom_(rbind_cube_tri) {
* Inner branch (OTZ bounds): branch_equal(R_AT, R_0, 13) * Inner branch (OTZ bounds): branch_equal(R_AT, R_0, 13)
* → Skip 13 instructions from BD slot, land at add_ui(R_FaceCur,...) * → Skip 13 instructions from BD slot, land at add_ui(R_FaceCur,...)
* ============================================================================ */ * ============================================================================ */
atom_resource(cube_tri, "model_ship_cube")
atom_region (cube_tri, REGION_PRIM_ARENA) atom_region (cube_tri, REGION_PRIM_ARENA)
atom_group (cube_tri, GROUP_RENDER_PRIMS) atom_group (cube_tri, GROUP_RENDER_PRIMS)
atom_cadence (cube_tri, CADENCE_FRAME) atom_cadence (cube_tri, CADENCE_FRAME)
+3 -3
View File
@@ -259,7 +259,7 @@ void update(PrimitiveArena* pa, U4* ordering_buf)
tb_emit(& tb, code_cube_tri); tb_emit(& tb, code_cube_tri);
} }
tb_emit(& tb, code_sync_prim_cursor); tb_emit(& tb, code_sync_primitive_arena);
tb_data(& tb, u4_(& pa->used)); tb_data(& tb, u4_(& pa->used));
tb_data(& tb, prim_base); tb_data(& tb, prim_base);
} }
@@ -351,8 +351,8 @@ void update(PrimitiveArena* pa, U4* ordering_buf)
for (U4 i = 0; i < Floor_num_faces; i++) { for (U4 i = 0; i < Floor_num_faces; i++) {
tb_emit(& tb, code_floor_tri); tb_emit(& tb, code_floor_tri);
} }
// After code_floor_tri iterations complete, the primitive arena's used counter is wrong.
tb_emit(& tb, code_sync_prim_cursor); tb_emit(& tb, code_sync_primitive_arena);
tb_data(& tb, u4_(& pa->used)); tb_data(& tb, u4_(& pa->used));
tb_data(& tb, prim_base); tb_data(& tb, prim_base);
} }
+22 -4
View File
@@ -27,8 +27,8 @@
atom_group( floor_tri, GROUP_RENDER_FLOOR) atom_group( floor_tri, GROUP_RENDER_FLOOR)
atom_cadence(floor_tri, CADENCE_FRAME) atom_cadence(floor_tri, CADENCE_FRAME)
atom_annot( floor_tri, phase_work, atom_annot( floor_tri, phase_work,
tape_regs(R_PrimCursor, R_FaceCursor, R_VertBase, R_OtBase), atom_reads( R_PrimCursor, R_FaceCursor, R_VertBase, R_OtBase),
tape_regs(R_PrimCursor, R_FaceCursor)) atom_writes(R_PrimCursor, R_FaceCursor))
internal internal
MipsAtom_(floor_tri) { MipsAtom_(floor_tri) {
// T0-T2 allocated // T0-T2 allocated
@@ -69,11 +69,29 @@ MipsAtom_(floor_tri) {
/* 8. Insert into Ordering Table Linked List */ /* 8. Insert into Ordering Table Linked List */
mac_insert_ot_tag(R_T1, 0x0400), mac_insert_ot_tag(R_T1, 0x0400),
add_ui(R_PrimCursor, R_PrimCursor, S_(Poly_F3)), /* Advance Prim Cursor (5 words) */ add_ui_1(R_PrimCursor, S_(Poly_F3)), /* Advance Prim Cursor (5 words) */
// Note(Ed): No bounds checking, should be checked before atom runs.
/* 9. Advance Input Cursor & Yield (Both branch targets land here) */ /* 9. Advance Input Cursor & Yield (Both branch targets land here) */
atom_label(floor_tri_exit) atom_label(floor_tri_exit)
add_ui(R_FaceCursor, R_FaceCursor, S_(S2) * 4), /* Advance Face Cursor (4 * S2 = 8 bytes) */ add_ui_1(R_FaceCursor, S_(S2) * 4), /* Advance Face Cursor (4 * S2 = 8 bytes) */
mac_yield()
};
typedef Struct_(Binds_SyncPrimitiveArena) { U4 PrimtiveArena_Used; U4 PrimtiveBase; };
atom_region( sync_primitive_arena, REGION_PRIM_ARENA)
atom_group( sync_primitive_arena, GROUP_RENDER_FLOOR)
atom_cadence(sync_primitive_arena, CADENCE_FRAME)
atom_annot( sync_primitive_arena, phase_work,
atom_reads( R_TapePtr, R_PrimCursor),
atom_writes(R_TapePtr))
internal MipsAtom_(sync_primitive_arena) {
load_word(R_AT, R_TapePtr, O_(Binds_SyncPrimitiveArena,PrimtiveArena_Used)),
load_word(R_T0, R_TapePtr, O_(Binds_SyncPrimitiveArena,PrimtiveBase)),
add_ui_1( R_TapePtr, S_(Binds_SyncPrimitiveArena)),
/* Calculate byte offset and store directly back to RAM */
sub_u( R_T0, R_PrimCursor, R_T0), // R_T0 = R_PrimCursor - prim-base
store_word(R_T0, R_AT, 0), // R_AT[0] = R_T0
mac_yield() mac_yield()
}; };
+100 -48
View File
@@ -265,7 +265,14 @@ local MACRO_EXPANSION = {
["CADENCE_ONDEMAND"] = "ondemand", ["CADENCE_ONDEMAND"] = "ondemand",
} }
local function valid_phase(p) return KNOWN_PHASES[p] end -- Phases + regions + cadences are defined further below (after TAPE_ATOM_MACROS).
-- We expose valid_phase via upvalue once KNOWN_PHASES is defined, but for now
-- it's a closure-resolved at call time.
-- Table refs initialised later; valid_phase is hooked up below KNOWN_PHASES.
local KNOWN_PHASES, KNOWN_REGIONS, KNOWN_CADENCES -- forward declarations
local function valid_phase(p) return KNOWN_PHASES and KNOWN_PHASES[p] or false end
local function is_wave_context_reg(name) return WAVE_CONTEXT_REGS[name] ~= nil end local function is_wave_context_reg(name) return WAVE_CONTEXT_REGS[name] ~= nil end
-- ============================================================ -- ============================================================
@@ -315,9 +322,10 @@ end
-- Extract identifier args from a parenthesized group. Returns a list -- Extract identifier args from a parenthesized group. Returns a list
-- of {kind, value} pairs where kind is one of: -- of {kind, value} pairs where kind is one of:
-- "ident" -- a bare identifier (e.g. TAPE_PHASE_WORK) -- "ident" -- a bare identifier (e.g. phase_work)
-- "regs" -- a TAPE_REGS(...) call whose args are extracted as a list -- "atom_reads" -- an atom_reads(...) call: value is the register list
-- "other" -- something we can't classify (preserved as text) -- "atom_writes"-- an atom_writes(...) call: value is the register list
-- "other" -- something we can't classify (preserved as text)
local function parse_atom_annot_args(inner) local function parse_atom_annot_args(inner)
-- Split at top-level commas, respecting nested parens. -- Split at top-level commas, respecting nested parens.
local args = {} local args = {}
@@ -325,20 +333,37 @@ local function parse_atom_annot_args(inner)
for _, tok in ipairs(tokens) do for _, tok in ipairs(tokens) do
local s = trim(tok) local s = trim(tok)
if s ~= "" then if s ~= "" then
-- TAPE_REGS(...) → extract inner identifiers -- Detect register-list calls: atom_reads(...) / atom_writes(...) / tape_regs(...)
if s:sub(1, 10) == "TAPE_REGS(" and s:sub(-1) == ")" then local regs_kind = nil
local regs_inner = s:sub(11, -2) local regs_inner = nil
if s:sub(-1) == ")" then
if s:sub(1, 11) == "atom_reads(" then
regs_kind = "atom_reads"
regs_inner = s:sub(12, -2)
elseif s:sub(1, 12) == "atom_writes(" then
regs_kind = "atom_writes"
regs_inner = s:sub(13, -2)
end
end
if regs_kind then
local regs = {} local regs = {}
for r in regs_inner:gmatch("[^,]+") do for r in regs_inner:gmatch("[^,]+") do
local trimmed = trim(r) local trimmed = trim(r)
if trimmed ~= "" then table.insert(regs, trimmed) end if trimmed ~= "" then table.insert(regs, trimmed) end
end end
table.insert(args, {kind = "regs", value = regs}) -- Resolve any phase_* / R_* alias macros
for i, r in ipairs(regs) do
if MACRO_EXPANSION[r] then regs[i] = MACRO_EXPANSION[r] end
end
table.insert(args, {kind = regs_kind, value = regs})
else else
-- Bare identifier (e.g. TAPE_PHASE_WORK) -- Bare identifier (e.g. phase_work)
local id, _ = read_ident(s, 1) local id, _ = read_ident(s, 1)
if id and trim(s) == id then if id and trim(s) == id then
table.insert(args, {kind = "ident", value = id}) local v = id
if MACRO_EXPANSION[v] then v = MACRO_EXPANSION[v] end
table.insert(args, {kind = "ident", value = v})
else else
table.insert(args, {kind = "other", value = s}) table.insert(args, {kind = "other", value = s})
end end
@@ -358,20 +383,20 @@ local TAPE_ATOM_MACROS = {
} }
-- Phase token names (must match macros in tape_atom_dsl.h) -- Phase token names (must match macros in tape_atom_dsl.h)
local KNOWN_PHASES = { KNOWN_PHASES = {
["init"] = true, ["bind"] = true, ["setup"] = true, ["init"] = true, ["bind"] = true, ["setup"] = true,
["work"] = true, ["commit"] = true, ["terminate"] = true, ["work"] = true, ["commit"] = true, ["terminate"] = true,
} }
-- Region token names (must match REGION_* macros in tape_atom_dsl.h) -- Region token names (must match REGION_* macros in tape_atom_dsl.h)
local KNOWN_REGIONS = { KNOWN_REGIONS = {
["prim_arena"] = true, ["face_arena"] = true, ["vertex_arena"] = true, ["prim_arena"] = true, ["face_arena"] = true, ["vertex_arena"] = true,
["ot_arena"] = true, ["heap_3d_models"] = true, ["cdrom_stream"] = true, ["ot_arena"] = true, ["heap_3d_models"] = true, ["cdrom_stream"] = true,
["vram_heap"] = true, ["vram_heap"] = true,
} }
-- Cadence token names (must match CADENCE_* macros in tape_atom_dsl.h) -- Cadence token names (must match CADENCE_* macros in tape_atom_dsl.h)
local KNOWN_CADENCES = { KNOWN_CADENCES = {
["frame"] = true, ["once"] = true, ["ondemand"] = true, ["frame"] = true, ["once"] = true, ["ondemand"] = true,
} }
@@ -403,60 +428,81 @@ local function find_atom_annotations(source)
if #args < 1 then if #args < 1 then
table.insert(annots, { table.insert(annots, {
line = line_of(source, i), line = line_of(source, i),
macro = ident, macro = ident,
kind = macro_def.kind, kind = macro_def.kind,
error = "missing atom name (first arg)" error = "missing atom name (first arg)",
}) })
else else
local name = args[1].value local name = args[1].value
-- For TAPE_ATOM_BIND: arg layout is (name, Binds_Struct, writes) -- atom_bind: (name, Binds_Struct, writes)
-- For others: (name, phase, reads, writes) -- atom_annot: (name, phase, reads, writes)
-- INIT / TERMINATE: (name) -- atom_setup / atom_commit: (name, reads)
-- atom_init / atom_terminate: (name)
local entry = { local entry = {
line = line_of(source, i), line = line_of(source, i),
macro = ident, macro = ident,
name = name, name = name,
kind = macro_def.kind, kind = macro_def.kind,
binds = nil, binds = nil,
phase = nil, phase = nil,
reads = {}, reads = {},
writes = {}, writes = {},
errors = {},
} }
-- Is this arg a register-list call (any of the recognized forms)?
local function is_regs(a)
return a and (a.kind == "atom_reads" or a.kind == "atom_writes" or a.kind == "regs")
end
if macro_def.binds then if macro_def.binds then
-- (name, Binds_Struct, writes) -- atom_bind(name, Binds_Struct, writes)
if #args >= 2 and args[2].kind == "ident" then if #args >= 2 and args[2].kind == "ident" then
entry.binds = args[2].value entry.binds = args[2].value
end end
if #args >= 3 and args[3].kind == "regs" then if #args >= 3 and is_regs(args[3]) then
-- A bind writes the wave-context, so atom_writes(...) is the
-- canonical form, but legacy regs(...) is also accepted.
entry.writes = args[3].value entry.writes = args[3].value
end end
elseif ident == "TAPE_ATOM_INIT" or ident == "TAPE_ATOM_TERMINATE" then elseif ident == "atom_init" or ident == "atom_terminate" then
-- (name) — no phase, no reads/writes to extract -- (name) only, no reads/writes to extract
elseif ident == "TAPE_ATOM_SETUP" then elseif ident == "atom_setup" then
-- (name, reads) -- atom_setup(name, reads)
if #args >= 2 and args[2].kind == "regs" then if #args >= 2 and is_regs(args[2]) then
entry.reads = args[2].value entry.reads = args[2].value
end end
elseif ident == "TAPE_ATOM_COMMIT" then elseif ident == "atom_commit" then
-- (name, reads) -- atom_commit(name, reads)
if #args >= 2 and args[2].kind == "regs" then if #args >= 2 and is_regs(args[2]) then
entry.reads = args[2].value entry.reads = args[2].value
end end
elseif ident == "TAPE_ATOM_ANNOT" then elseif ident == "atom_annot" then
-- (name, phase, reads, writes) -- atom_annot(name, phase, reads, writes)
if #args >= 2 and args[2].kind == "ident" then if #args >= 2 and args[2].kind == "ident" then
-- Expand TAPE_PHASE_* macro references entry.phase = MACRO_EXPANSION[args[2].value] or args[2].value
local phase_id = args[2].value
entry.phase = MACRO_EXPANSION[phase_id] or phase_id
end end
if #args >= 3 and args[3].kind == "regs" then -- reads slot: atom_reads(...) is the canonical form;
-- atom_writes(...) in the reads slot is a likely bug.
if #args >= 3 and is_regs(args[3]) then
if args[3].kind == "atom_writes" then
table.insert(entry.errors,
"reads slot has atom_writes — swap order?")
end
entry.reads = args[3].value entry.reads = args[3].value
end end
if #args >= 4 and args[4].kind == "regs" then -- writes slot: atom_writes(...) is canonical;
-- atom_reads(...) here is a likely bug.
if #args >= 4 and is_regs(args[4]) then
if args[4].kind == "atom_reads" then
table.insert(entry.errors,
"writes slot has atom_reads — swap order?")
end
entry.writes = args[4].value entry.writes = args[4].value
end end
end end
table.insert(annots, entry) table.insert(annots, entry)
end end
i = after_paren i = after_paren
@@ -865,11 +911,11 @@ end
local function validate(source_path, word_counts) local function validate(source_path, word_counts)
local source = read_file(source_path) local source = read_file(source_path)
local annots = find_atom_annotations(source) local annots = find_atom_annotations(source)
local macros = find_macro_word_annotations(source) local macros = find_macro_word_annotations(source)
local pragmas = find_atom_pragmas(source) local pragmas = find_atom_pragmas(source)
local binds = find_binds_structs(source) local binds = find_binds_structs(source)
local atoms = find_atom_names(source) local atoms = find_atom_names(source)
-- Index for O(1) lookup -- Index for O(1) lookup
local atom_index = {} local atom_index = {}
@@ -892,6 +938,12 @@ local function validate(source_path, word_counts)
msg = string.format("annotation for '%s' has no matching MipsAtom_(%s) { ... }", a.name, a.name) msg = string.format("annotation for '%s' has no matching MipsAtom_(%s) { ... }", a.name, a.name)
}) })
end end
-- Per-entry parser errors (e.g. reads/writes slot mix-ups)
if a.errors then
for _, msg in ipairs(a.errors) do
table.insert(errors, {line = a.line, msg = string.format("'%s': %s", a.name, msg)})
end
end
end end
-- 2. Every atom must have exactly one annotation (no orphans, no duplicates). -- 2. Every atom must have exactly one annotation (no orphans, no duplicates).