Odin/core/rexcode/ppc/tablegen/gen.odin

// rexcode  ·  Brendan Punsky (dotbmp@github), original author

package rexcode_ppc_tablegen

// =============================================================================
// POWERPC TABLE GENERATOR  (Stage A)
// =============================================================================
//
// Reads the single-source-of-truth ENCODING_TABLE (encoding_table.odin, this
// package) and emits human-readable, type-checked Odin into ./generated/:
//
//   generated/encode_tables.odin   ENCODE_FORMS + ENCODE_RUNS + PREFIX_BITS_TABLE
//   generated/decode_tables.odin   DECODE_ENTRIES + DECODE_FORM_IDX +
//                                   DECODE_BUCKET_LIST + primary/sub index tables
//   generated/writer.odin          Stage B: serialize those globals to ../../tables/*.bin
//
// It also re-emits the library loader ../tables.odin. Run:
//   odin run ppc/tablegen            # Stage A
//   odin run ppc/tablegen/generated  # Stage B
//
// PowerPC dispatch is two-level (one DECODE_ENTRIES array, two index tables):
//   Primary (64 buckets):  key = bits[26:31] (6-bit primary opcode).
//   Secondary (16384):     key = primary*256 + bits[1:8] (low XO bits, skip Rc).
// Each bucket points to a contiguous run in DECODE_BUCKET_LIST of entry indices,
// sorted by mask popcount descending so the most-specific match wins first.

import "core:fmt"
import "core:os"
import "core:strings"
import "core:slice"
import "core:reflect"
import "core:math/bits"
import lib "../"

// Package-scope aliases so the moved SoT resolves Mnemonic/Encoding unqualified.
Encoding :: lib.Encoding
Mnemonic :: lib.Mnemonic

Blob :: struct { global, file, typ: string }
BLOBS := [?]Blob{
	{"ENCODE_FORMS",         "ppc.encode_forms.bin", "Encoding"},
	{"ENCODE_RUNS",          "ppc.encode_runs.bin",  "Encode_Run"},
	{"PREFIX_BITS_TABLE",    "ppc.prefix_bits.bin",  "u32"},
	{"DECODE_ENTRIES",       "ppc.entries.bin",      "Decode_Entry"},
	{"DECODE_FORM_IDX",      "ppc.form_idx.bin",     "u16"},
	{"DECODE_BUCKET_LIST",   "ppc.bucket_list.bin",  "u16"},
	{"DECODE_INDEX_PRIMARY", "ppc.idx_primary.bin",  "Decode_Index"},
	{"DECODE_INDEX_SUB",     "ppc.idx_sub.bin",      "Decode_Index"},
}

DIR_GEN     :: #directory + "/generated/"
PATH_LOADER :: #directory + "/../tables.odin"

PRIMARY_BUCKETS :: 64       // bits[26:31] of word
SUB_BITS        :: 8        // bits[1:8] of word (low XO bits, skipping Rc)
SUB_BUCKETS     :: PRIMARY_BUCKETS * (1 << SUB_BITS)   // 64 * 256 = 16384

Entry :: struct {
	mnemonic: lib.Mnemonic,
	ops:      [4]lib.Operand_Type,
	enc:      [4]lib.Operand_Encoding,
	bits:     u32,
	mask:     u32,
	feature:  lib.Feature,
	mode:     lib.Mode,
	flags:    lib.Encoding_Flags,
	form_idx: u16,
}

Range :: struct { start: u32, count: u16 }

Pair :: struct { bucket: u32, entry_idx: u16 }

main :: proc() {
	n := emit_encode_tables()
	ne := emit_decode_tables()
	emit_writer()
	emit_loader()
	fmt.printfln("ppc tablegen: %d encode forms, %d decode entries", n, ne)
}

// -----------------------------------------------------------------------------
// Encode side
// -----------------------------------------------------------------------------

emit_encode_tables :: proc() -> (total: int) {
	sb := strings.builder_make()
	strings.write_string(&sb, "package rexcode_ppc_generated\n\n")
	strings.write_string(&sb, "// GENERATED by ../gen.odin -- DO NOT EDIT.\n")
	strings.write_string(&sb, "// Flattened encode forms + per-mnemonic run index + prefix words (source: ENCODING_TABLE).\n\n")
	strings.write_string(&sb, "import lib \"../..\"\n\n")

	for m in Mnemonic { total += len(ENCODING_TABLE[m]) }

	strings.write_string(&sb, "@(rodata)\n")
	fmt.sbprintfln(&sb, "ENCODE_FORMS := [%d]lib.Encoding{{", total)
	for m in Mnemonic {
		forms := ENCODING_TABLE[m]
		if len(forms) == 0 { continue }
		fmt.sbprintfln(&sb, "\t// .%v", m)
		for f in forms {
			write_row(&sb, f.mnemonic, f.ops, f.enc, f.bits, f.mask, f.feature, f.mode, f.flags)
		}
	}
	strings.write_string(&sb, "}\n\n")

	run_w := 0
	for m in Mnemonic { run_w = max(run_w, len(reflect.enum_string(m))) }
	strings.write_string(&sb, "@(rodata)\n")
	strings.write_string(&sb, "ENCODE_RUNS := [lib.Mnemonic]lib.Encode_Run{\n")
	start := 0
	for m in Mnemonic {
		c := len(ENCODING_TABLE[m])
		name := reflect.enum_string(m)
		fmt.sbprintf(&sb, "\t.%s", name)
		for _ in 0..<run_w-len(name) { strings.write_byte(&sb, ' ') }
		fmt.sbprintfln(&sb, " = {{% 5d, % 3d}},", start, c)
		start += c
	}
	strings.write_string(&sb, "}\n\n")

	// PREFIX_BITS_TABLE: dense [len(Mnemonic)]u32, indexed by mnemonic ordinal.
	// (Non-prefixed mnemonics are 0; the library reads it as PREFIX_BITS_TABLE[u16(mn)].)
	nmn := len(Mnemonic)
	strings.write_string(&sb, "@(rodata)\n")
	fmt.sbprintfln(&sb, "PREFIX_BITS_TABLE := [%d]u32{{", nmn)
	for m in Mnemonic {
		v := PREFIX_BITS_TABLE[m]
		if v != 0 {
			fmt.sbprintfln(&sb, "\t%d = 0x%08X, // .%v", u16(m), v, m)
		}
	}
	strings.write_string(&sb, "}\n")
	emit_file(DIR_GEN + "encode_tables.odin", &sb)
	return
}

// -----------------------------------------------------------------------------
// Decode side  (ported from the old tools/gen_decode_tables.odin)
// -----------------------------------------------------------------------------

emit_decode_tables :: proc() -> (total: int) {
	all: [dynamic]Entry
	defer delete(all)
	for mn in Mnemonic {
		for f, fi in ENCODING_TABLE[mn] {
			append(&all, Entry{
				mnemonic = mn,
				ops      = f.ops,
				enc      = f.enc,
				bits     = f.bits,
				mask     = f.mask,
				feature  = f.feature,
				mode     = f.mode,
				flags    = f.flags,
				form_idx = u16(fi),
			})
		}
	}

	// Sort the global entries array: by primary opcode, then mask popcount
	// descending so within-bucket scan picks the most specific first.
	slice.sort_by(all[:], proc(x, y: Entry) -> bool {
		px := (x.bits >> 26) & 0x3F
		py := (y.bits >> 26) & 0x3F
		if px != py { return px < py }
		xc := bits.count_ones(x.mask)
		yc := bits.count_ones(y.mask)
		if xc != yc { return xc > yc }
		return u16(x.mnemonic) < u16(y.mnemonic)
	})

	// For each entry, enumerate the (primary, sub-key) pairs it can match.
	primary_pairs: [dynamic]Pair
	sub_pairs:     [dynamic]Pair
	defer delete(primary_pairs); defer delete(sub_pairs)

	keys: [dynamic]u32
	defer delete(keys)

	for e, i in all {
		// Primary key
		enumerate_keys(e.bits, e.mask, 26, 6, &keys)
		for k in keys {
			append(&primary_pairs, Pair{bucket = k, entry_idx = u16(i)})
		}

		// Sub key: primary << SUB_BITS | bits[1..SUB_BITS+1)
		prim_keys: [dynamic]u32
		sub_only:  [dynamic]u32
		defer delete(prim_keys); defer delete(sub_only)
		enumerate_keys(e.bits, e.mask, 26, 6, &prim_keys)
		enumerate_keys(e.bits, e.mask, 1,  SUB_BITS, &sub_only)
		for pk in prim_keys {
			for sk in sub_only {
				key := pk * (1 << SUB_BITS) + sk
				append(&sub_pairs, Pair{bucket = key, entry_idx = u16(i)})
			}
		}
	}

	// Re-sort pair lists: primary order, then mask popcount descending.
	rebuild :: proc(pairs: ^[dynamic]Pair, all: []Entry) {
		Sort_Pair :: struct { sort_key: u64, entry_idx: u16, bucket: u32 }
		sortable := make([dynamic]Sort_Pair, 0, len(pairs), context.temp_allocator)
		for pp in pairs^ {
			e := all[pp.entry_idx]
			pop := u64(bits.count_ones(e.mask))
			// (bucket << 40) | ((63 - pop) << 32) | mnemonic
			key := (u64(pp.bucket) << 40) | ((63 - pop) << 32) | u64(e.mnemonic)
			append(&sortable, Sort_Pair{sort_key = key, entry_idx = pp.entry_idx, bucket = pp.bucket})
		}
		slice.sort_by_key(sortable[:], proc(s: Sort_Pair) -> u64 { return s.sort_key })
		clear(pairs)
		for s in sortable { append(pairs, Pair{bucket = s.bucket, entry_idx = s.entry_idx}) }
	}
	rebuild(&primary_pairs, all[:])
	rebuild(&sub_pairs,     all[:])

	// Build flat u16 dispatch list. Each bucket points to a contiguous run.
	primary_idx: [PRIMARY_BUCKETS]Range
	sub_idx:     [SUB_BUCKETS]Range
	bucket_list: [dynamic]u16
	defer delete(bucket_list)

	emit_pairs :: proc(pairs: []Pair, idx: []Range, list: ^[dynamic]u16) {
		prev_bucket: i64 = -1
		for pp in pairs {
			cur := i64(pp.bucket)
			if cur != prev_bucket {
				idx[cur].start = u32(len(list))
				idx[cur].count = 0
				prev_bucket = cur
			}
			append(list, pp.entry_idx)
			idx[cur].count += 1
		}
	}
	emit_pairs(primary_pairs[:], primary_idx[:], &bucket_list)
	emit_pairs(sub_pairs[:],     sub_idx[:],     &bucket_list)

	sb := strings.builder_make()
	strings.write_string(&sb, "package rexcode_ppc_generated\n\n")
	strings.write_string(&sb, "// GENERATED by ../gen.odin -- DO NOT EDIT.\n")
	strings.write_string(&sb, "// Reverse decode tables (source: ENCODING_TABLE), two-level primary+sub dispatch.\n\n")
	strings.write_string(&sb, "import lib \"../..\"\n\n")

	strings.write_string(&sb, "@(rodata)\n")
	fmt.sbprintfln(&sb, "DECODE_ENTRIES := [%d]lib.Decode_Entry{{", len(all))
	for e in all {
		write_row(&sb, e.mnemonic, e.ops, e.enc, e.bits, e.mask, e.feature, e.mode, e.flags)
	}
	strings.write_string(&sb, "}\n\n")

	emit_form_idx(&sb, all[:])
	emit_bucket_list(&sb, bucket_list[:])
	emit_range_table(&sb, "DECODE_INDEX_PRIMARY", primary_idx[:])
	emit_range_table(&sb, "DECODE_INDEX_SUB",     sub_idx[:])
	emit_file(DIR_GEN + "decode_tables.odin", &sb)
	return len(all)
}

enumerate_keys :: proc(b, mask: u32, key_shift: u32, key_bits: u32, out: ^[dynamic]u32) {
	clear(out)
	key_mask := (u32(1) << key_bits) - 1
	fixed_key := ((b & mask) >> key_shift) & key_mask
	var_bits := (~mask >> key_shift) & key_mask
	sub: u32 = 0
	for {
		append(out, fixed_key | sub)
		if var_bits == 0 { break }
		if sub == var_bits { break }
		sub = (sub - var_bits) & var_bits
	}
}

emit_form_idx :: proc(sb: ^strings.Builder, entries: []Entry) {
	all_zero := true
	for e in entries {
		if e.form_idx != 0 { all_zero = false; break }
	}
	if all_zero {
		strings.write_string(sb, "@(rodata)\n")
		fmt.sbprintf(sb, "DECODE_FORM_IDX: [%d]u16\n\n", len(entries))
	} else {
		strings.write_string(sb, "@(rodata)\n")
		fmt.sbprintf(sb, "DECODE_FORM_IDX := [%d]u16{{", len(entries))
		for e, i in entries {
			if i % 64 == 0 { strings.write_string(sb, "\n\t") }
			fmt.sbprintf(sb, "%d, ", e.form_idx)
		}
		strings.write_string(sb, "\n}\n\n")
	}
}

emit_bucket_list :: proc(sb: ^strings.Builder, items: []u16) {
	strings.write_string(sb, "@(rodata)\n")
	fmt.sbprintf(sb, "DECODE_BUCKET_LIST := [%d]u16{{", len(items))
	for v, i in items {
		if i % 64 == 0 { strings.write_string(sb, "\n\t") }
		fmt.sbprintf(sb, "% 4d, ", v)
	}
	strings.write_string(sb, "\n}\n\n")
}

emit_range_table :: proc(sb: ^strings.Builder, name: string, ranges: []Range) {
	strings.write_string(sb, "@(rodata)\n")
	fmt.sbprintf(sb, "%s := [%d]lib.Decode_Index{{", name, len(ranges))
	amount_set := 0
	for r, i in ranges {
		if r.count != 0 {
			if amount_set % 16 == 0 { strings.write_string(sb, "\n\t") }
			fmt.sbprintf(sb, "0x%04X = {{% 5d, % 4d, 0}}, ", i, r.start, r.count)
			amount_set += 1
		}
	}
	strings.write_string(sb, "\n}\n\n")
}

// -----------------------------------------------------------------------------
// Shared row + flags formatting
// -----------------------------------------------------------------------------

write_row :: proc(sb: ^strings.Builder, mn: lib.Mnemonic, ops: [4]lib.Operand_Type,
                  enc: [4]lib.Operand_Encoding, bits, mask: u32, feature: lib.Feature,
                  mode: lib.Mode, flags: lib.Encoding_Flags) {
	fmt.sbprintf(sb, "\t{{ .%v, {{.%v,.%v,.%v,.%v}}, {{.%v,.%v,.%v,.%v}}, 0x%08X, 0x%08X, .%v, .%v, {{%s}} }},\n",
		mn, ops[0], ops[1], ops[2], ops[3], enc[0], enc[1], enc[2], enc[3], bits, mask, feature, mode, flags_lit(flags))
}

flags_lit :: proc(f: lib.Encoding_Flags) -> string {
	parts: [dynamic]string
	defer delete(parts)
	if f.branch      { append(&parts, "branch=true")      }
	if f.cond_branch { append(&parts, "cond_branch=true") }
	if f.writes_lr   { append(&parts, "writes_lr=true")   }
	if f.sets_cr0    { append(&parts, "sets_cr0=true")    }
	if f.sets_cr1    { append(&parts, "sets_cr1=true")    }
	if f.abs_branch  { append(&parts, "abs_branch=true")  }
	if f.has_oe      { append(&parts, "has_oe=true")      }
	if f.prefixed    { append(&parts, "prefixed=true")    }
	if f.vle         { append(&parts, "vle=true")         }
	if f.vle_short   { append(&parts, "vle_short=true")   }
	return strings.join(parts[:], ", ", context.temp_allocator)
}

// -----------------------------------------------------------------------------
// Stage B writer + the library loader
// -----------------------------------------------------------------------------

emit_writer :: proc() {
	sb := strings.builder_make()
	strings.write_string(&sb, "package rexcode_ppc_generated\n\n")
	strings.write_string(&sb, "// GENERATED by ../gen.odin -- DO NOT EDIT.\n")
	strings.write_string(&sb, "// Stage B: serialize the typed tables above to raw blobs under ../../tables/.\n\n")
	strings.write_string(&sb, "import \"core:os\"\nimport \"core:fmt\"\n\n")
	strings.write_string(&sb, "TABLES :: #directory + \"/../../tables/\"\n\n")
	strings.write_string(&sb, "raw :: #force_inline proc \"contextless\" (p: rawptr, n: int) -> []u8 {\n\treturn (cast([^]u8)p)[:n]\n}\n\n")
	strings.write_string(&sb, "w :: proc(file: string, data: []u8) {\n")
	strings.write_string(&sb, "\tif err := os.write_entire_file(file, data); err != nil {\n")
	strings.write_string(&sb, "\t\tfmt.eprintfln(\"rexcode tablegen: failed to write %s: %v\", file, err)\n\t\tos.exit(1)\n\t}\n}\n\n")
	strings.write_string(&sb, "main :: proc() {\n")
	for b in BLOBS {
		fmt.sbprintfln(&sb, "\tw(TABLES + \"%s\", raw(&%s, size_of(%s)))", b.file, b.global, b.global)
	}
	strings.write_string(&sb, "}\n")
	emit_file(DIR_GEN + "writer.odin", &sb)
}

LOADER_TYPES :: `// -----------------------------------------------------------------------------
// Subsidiary table types (generated scaffolding)
// -----------------------------------------------------------------------------

// Companion run index: ENCODE_RUNS[mnemonic] -> contiguous run in ENCODE_FORMS.
Encode_Run :: struct {
	start: u32,
	count: u32,
}

Decode_Entry :: struct #packed {
	mnemonic: Mnemonic,             // 2
	ops:      [4]Operand_Type,      // 4
	enc:      [4]Operand_Encoding,  // 4
	bits:     u32,                  // 4
	mask:     u32,                  // 4
	feature:  Feature,              // 1
	mode:     Mode,                 // 1
	flags:    Encoding_Flags,       // 2
}
#assert(size_of(Decode_Entry) == 22)

Decode_Index :: struct #packed {
	start: u32,
	count: u16,
	_:     u16,
}
#assert(size_of(Decode_Index) == 8)

DECODE_SUB_BUCKETS :: 256 // per primary
`

LOADER_ACCESSORS :: `// -----------------------------------------------------------------------------
// Accessors
// -----------------------------------------------------------------------------

// Per-mnemonic encode forms: the run of ENCODE_FORMS belonging to ` + "`m`" + `.
// Replaces the old ENCODING_TABLE[m] slice; the returned view is into rodata.
@(private, require_results)
encoding_forms :: #force_inline proc "contextless" (m: Mnemonic) -> []Encoding {
	r := ENCODE_RUNS[u16(m)]
	return ENCODE_FORMS[r.start:][:r.count]
}
`

emit_loader :: proc() {
	sb := strings.builder_make()
	strings.write_string(&sb, "package rexcode_ppc\n\n")
	strings.write_string(&sb, "// =============================================================================\n")
	strings.write_string(&sb, "// GENERATED FILE - DO NOT EDIT\n")
	strings.write_string(&sb, "// =============================================================================\n")
	strings.write_string(&sb, "//\n")
	strings.write_string(&sb, "// Loads the flat binary encode/decode tables into @(rodata). Produced by tablegen:\n")
	strings.write_string(&sb, "//\n")
	strings.write_string(&sb, "//   odin run tablegen            # Stage A: ENCODING_TABLE -> generated/ + this file\n")
	strings.write_string(&sb, "//   odin run tablegen/generated  # Stage B: typed Odin literals -> tables/*.bin\n")
	strings.write_string(&sb, "//\n")
	strings.write_string(&sb, "// The .bin blobs are raw, host-endian, packed struct images.\n\n")
	strings.write_string(&sb, LOADER_TYPES)
	strings.write_string(&sb, "\n// -----------------------------------------------------------------------------\n")
	strings.write_string(&sb, "// Loaded tables (rodata, embedded from tables/*.bin at compile time)\n")
	strings.write_string(&sb, "// -----------------------------------------------------------------------------\n\n")

	gmax, fmax := 0, 0
	for b in BLOBS { gmax = max(gmax, len(b.global)); fmax = max(fmax, len(b.file)) }
	for b in BLOBS {
		fmt.sbprintf(&sb, "@(rodata) %s", b.global)
		for _ in 0..<gmax-len(b.global) { strings.write_byte(&sb, ' ') }
		path := fmt.tprintf("\"tables/%s\",", b.file)
		fmt.sbprintf(&sb, " := #load(%s", path)
		for _ in 0..<fmax-len(b.file) { strings.write_byte(&sb, ' ') }
		fmt.sbprintfln(&sb, " []%s)", b.typ)
	}
	strings.write_string(&sb, "\n")
	strings.write_string(&sb, LOADER_ACCESSORS)
	emit_file(PATH_LOADER, &sb)
}

GEN_ATTRIB :: "// rexcode  ·  Brendan Punsky (dotbmp@github), original author\n\n"

emit_file :: proc(path: string, sb: ^strings.Builder) {
	if err := os.write_entire_file(path, transmute([]u8)strings.concatenate({GEN_ATTRIB, strings.to_string(sb^)})); err != nil {
		fmt.eprintfln("rexcode tablegen: failed to write %s: %v", path, err)
		os.exit(1)
	}
}