@fraczak/k 6.10.3 → 6.10.4

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -5,6 +5,49 @@ All notable changes to this project will be documented here.
5
5
  The format follows [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
6
6
  This project uses [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
7
7
 
8
+ ## [6.10.4] — 2026-10-06 — Karatsuba Multiplication, Multi-Backend Harness, and CLI Help
9
+
10
+ ### Arithmetic & Algorithms
11
+
12
+ - **Karatsuba Multiplication (`Examples/karatsuba-mult.k`)**:
13
+ - Implemented standalone, minimal Karatsuba integer multiplication with
14
+ asymptotic time complexity $O(n^{\log_2 3}) \approx O(n^{1.585})$.
15
+ - Utilizes unary counter (`$unat`) splitting with a 512-bit threshold
16
+ switching to grade-school multiplication (`times_bits`) for small
17
+ inputs.
18
+ - Implements direct bit-level arithmetic (`plus_bits`, `minus_bits`,
19
+ `times_bits`) and direct $2m$ bit shifting (`shift2_bits_u`).
20
+ - Added dedicated test suite in `tests/test-karatsuba.mjs` verifying
21
+ squaring cases and mixed-sign pair multiplications against BigInt
22
+ oracle.
23
+
24
+ ### Performance & Benchmarking
25
+
26
+ - **Multiplication Performance Harness**
27
+ (`scripts/perf-multiplication.mjs`):
28
+ - Added multi-backend performance runner comparing `times` and
29
+ `karatsuba` across LLVM, Wasm, kVM, and JS engines.
30
+ - Supports `--mode squaring` (doubling bit width per step) and
31
+ `--mode sizes` (random operands of fixed bit lengths).
32
+ - Includes per-step timeout protection and process RSS memory guards.
33
+ - Added `npm run perf:mult` script in `package.json`.
34
+
35
+ ### CLI Usability & Backend Enhancements
36
+
37
+ - **Unified `-h` and `--help` CLI Documentation**:
38
+ - Added comprehensive option and environment variable documentation
39
+ across all benchmark, performance, and test scripts in `scripts/`
40
+ and `backends/*/scripts/`.
41
+ - **LLVM Direct `.kvm` Bytecode Compilation**:
42
+ - Added `compileKVMToExecutable` in `backends/llvm/src/executable.mjs`,
43
+ enabling direct binary compilation from lowered `.kvm` IR.
44
+ - **Runtime Mark Compaction Fixes**:
45
+ - Updated LLVM runtime compaction tracking (`%k_rt_tail_mark`) to
46
+ accurately preserve base and last compaction arena marks in
47
+ tail loops.
48
+
49
+ ---
50
+
8
51
  ## [6.10.3] — 2026-10-04 — Unified File Formats Specification and Documentation Modernization
9
52
 
10
53
  ### Documentation & Core Architecture
@@ -14,7 +14,7 @@ type-checked artifacts and JIT/native execution.
14
14
  | **`.k`** | Human-readable source code | UTF-8 plain text | Optional (trailing expression) | Editor, `k-decompile` | `k.mjs`, `k-repl`, `k-compile`, backends |
15
15
  | **`.ko`** | Executable compiled object | Binary (`KOBJ\n` + JSON payload) | **Required** (`"__main__"`) | `k-compile`, REPL (`:ko`) | `k.mjs`, `k-wasm-compile`, `k-llvm-build`, inspect tools |
16
16
  | **`.klib`**| Reusable compiled library | UTF-8 JSON payload | **None** (`null`) | `k-compile --lib`, REPL (`:klib`) | REPL (`:load`), CLI linkers (`--lib`) |
17
- | **`.kvm`** | Polymorphic kVM bytecode template | JSON / CBOR (`format: "k-vm", layer: "KVM-P"`) | Per-relation functions | `k-compile --format kvm`, kvm.mjs | kVM runtime, JIT specializer, `k-wasm-compile` |
17
+ | **`.kvm`** | Polymorphic kVM bytecode template | JSON / CBOR (`format: "k-vm", layer: "KVM-P"`) | Per-relation functions | `k-compile --format kvm`, kvm.mjs | `kvm.mjs`, `k-wasm-compile`, `k-llvm-build` |
18
18
 
19
19
  ### Format Transformation Workflow
20
20
 
@@ -63,6 +63,7 @@ flowchart TD
63
63
 
64
64
  %% Lowering from .kvm
65
65
  KVM -->|"k-wasm-compile"| WASM
66
+ KVM -->|"k-llvm-build"| NATIVE
66
67
  KVM -->|"kvm.mjs / specializeKVM"| INTERP
67
68
  ```
68
69
 
@@ -276,21 +277,38 @@ Export all currently loaded relations and types into a reusable library:
276
277
  `.kvm` is a polymorphic register-IR template artifact
277
278
  (`format: "k-vm"`, `layer: "KVM-P"`, `isPolymorphic: true`).
278
279
 
279
- #### How `.kvm` Differs from `.ko` and `.klib`:
280
- - **Abstraction Level:** `.ko` and `.klib` store high-level relational
281
- expressions (KIR) and structural type graphs. In contrast, `.kvm` is
282
- lowered into explicit register-based virtual machine instructions
283
- (`alloc`, `make_variant`, `project_variant`, `union`, `call`, and
284
- tail loops).
285
- - **Polymorphic Template vs. Fixed Object:**
286
- - `.ko` is a concrete executable bound to an entrypoint (`main`).
287
- - `.klib` is a library of high-level relation definitions for linking.
288
- - `.kvm` stores compiled bytecode functions paired with principal
289
- pattern graphs ($P_{in}, P_{out}$).
290
- - **Instant Specialization:** Instead of running the compiler's full
291
- constraint-derivation engine, the runtime specializer (`specializeKVM`)
292
- prunes dead union branches and folds redundant guards on-the-fly
293
- against concrete input envelopes $(v, P_v)$.
280
+ #### Why `.kvm` Exists (What You Gain):
281
+
282
+ 1. **Relational AST vs. Linear Register Machine:**
283
+ `.ko` and `.klib` store high-level relational combinators (compositions
284
+ `(f g)`, products `{a x, b y}`, unions `<f, g>`) and structural type
285
+ graphs. They contain no virtual registers, instruction sequences, or
286
+ control-flow branches. `.kvm` transforms this functional tree into a
287
+ linear 3-address virtual register machine IR (`alloc`, `project_field`,
288
+ `make_variant`, `project_variant`, `jump_if_fail`, `call`, and loops).
289
+
290
+ 2. **$1 \times N$ Backend Decoupling:**
291
+ Rather than having every code generator (LLVM, WebAssembly, ARM64,
292
+ and the JIT interpreter) independently lower the relational tree and
293
+ re-implement type-guided control flow, relational-to-imperative
294
+ lowering occurs once in `kvm.mjs`. All backends consume the exact same
295
+ ~10 linear kVM instructions (`kvm2llvm.mjs`, `kvm2wasm.mjs`,
296
+ `kvm2arm64.mjs`, `kvm.mjs`).
297
+
298
+ 3. **Microsecond Polymorphic Specialization (`layer: "KVM-P"`):**
299
+ In polymorphic programs, specializing `.ko` requires re-running the
300
+ compiler's full structural type derivation and constraint convergence
301
+ engine. A `.kvm` template stores pre-compiled register bytecode with
302
+ principal pattern graphs. The runtime specializer (`specializeKVM`)
303
+ prunes dead branches and folds redundant guards against concrete
304
+ runtime input envelopes $(v, P_v)$ in microseconds with zero AST
305
+ recompilation overhead.
306
+
307
+ 4. **Inspectable & Cacheable Register IR:**
308
+ `.kvm` produces a human- and machine-readable JSON or CBOR format.
309
+ You can inspect virtual register assignments, verify dead-branch
310
+ elimination, diff compiler output across runs, or cache the lowered IR
311
+ to bypass frontend compilation when building native or WASM targets.
294
312
 
295
313
  ### How to Generate
296
314
 
@@ -309,7 +327,24 @@ k-compile add5.ko add5.kvm
309
327
 
310
328
  ### How to Consume
311
329
 
330
+ - **Direct lowering from `.kvm` to Native Binaries (`k-llvm-build`):**
331
+ Compile a `.kvm` template directly into a native executable:
332
+ ```bash
333
+ k-llvm-build add5.kvm add5-bin
334
+ echo "10" | ./codecs/int.mjs --parse | \
335
+ ./add5-bin | \
336
+ ./codecs/int.mjs --print
337
+ # Output: 15
338
+ ```
339
+
340
+ - **Direct lowering from `.kvm` to LLVM IR (`k-llvm-compile`):**
341
+ Emit LLVM assembly directly from the register IR:
342
+ ```bash
343
+ k-llvm-compile add5.kvm add5.ll
344
+ ```
345
+
312
346
  - **Direct lowering from `.kvm` to WebAssembly (`k-wasm-compile`):**
347
+ Compile a `.kvm` template directly into a standalone `.wasm` module:
313
348
  ```bash
314
349
  k-wasm-compile add5.kvm add5.wasm
315
350
  echo "10" | ./codecs/int.mjs --parse | \
@@ -0,0 +1,215 @@
1
+ -- Karatsuba multiplication O(n^1.585)
2
+ -- Self-contained and minimal implementation
3
+ -- for arbitrary-precision integers
4
+
5
+ $ yes_no = < {} yes, {} no >;
6
+ yes = {} | yes $yes_no;
7
+ no = {} | no $yes_no;
8
+
9
+ $ bits = < {} _, bits 0, bits 1 >;
10
+ 0b = {} | _;
11
+ 0 = 0b |0;
12
+ 1 = 0b |1;
13
+
14
+ $ unat = < {} _, unat 1 >;
15
+ 0u = {} | _ $unat;
16
+
17
+ empty = $bits < /_ yes, no >;
18
+ empty? = {empty /yes if, () then} .then;
19
+
20
+ inv = {()x, 0b r} _inv;
21
+ _inv = ${bits x, bits r}
22
+ <
23
+ {.x empty? if, .r then} .then,
24
+ <
25
+ {.x /0 x, .r |0 r},
26
+ {.x /1 x, .r |1 r}
27
+ > _inv
28
+ >;
29
+
30
+ concat = ${bits x, bits y} {.x inv x, .y y} concat_;
31
+ concat_ = ${bits x, bits y}
32
+ <
33
+ {.x empty? if, .y then} .then,
34
+ <
35
+ {.x /0 x, .y |0 y},
36
+ {.x /1 x, .y |1 y}
37
+ > concat_
38
+ >;
39
+
40
+ remove_leading_zeros = < /0 remove_leading_zeros, () >;
41
+ nat = remove_leading_zeros <empty? 0, ()>;
42
+
43
+ $ int = < bits '+', bits '-' >;
44
+ int = nat | '+' $int;
45
+ abs = $int </'+', /'-'>;
46
+
47
+ $ sign = < {} '+', {} '-' >;
48
+ sign_plus = {} | '+' $sign;
49
+ sign_minus = {} | '-' $sign;
50
+ swap_sign = < /'+' | '-', /'-' | '+' >;
51
+
52
+ $ carry = < {} 0, {} 1 >;
53
+ c0 = {} | 0 $carry;
54
+ c1 = {} | 1 $carry;
55
+
56
+ _plus = {.x x, .y y, .c c, 0b r} _plus_;
57
+ _plus_ = ${bits x, bits y, carry c, bits r}
58
+ <
59
+ {.x empty? if, .y empty? and, .c/0 c, .r inv then} .then,
60
+ {.x empty? if, .y empty? and, {.r x, 1 r} _inv then} .then,
61
+ {.x empty? if, .c/0 and, {.r x, .y r} _inv then} .then,
62
+ <
63
+ {.x y, .y empty? x, .c c, .r r},
64
+ {<.c/0 0, 1> x, .y y, .x empty? c0 c, .r r},
65
+ {.x /0 x, .y /0 y, .c/0 c0 c, .r |0 r},
66
+ {.x /0 x, .y /0 y, .c/1 c0 c, .r |1 r},
67
+ {.x /0 x, .y /1 y, .c/0 c0 c, .r |1 r},
68
+ {.x /0 x, .y /1 y, .c/1 c1 c, .r |0 r},
69
+ {.x /1 x, .y /0 y, .c/0 c0 c, .r |1 r},
70
+ {.x /1 x, .y /0 y, .c/1 c1 c, .r |0 r},
71
+ {.x /1 x, .y /1 y, .c/0 c1 c, .r |0 r},
72
+ {.x /1 x, .y /1 y, .c/1 c1 c, .r |1 r}
73
+ > _plus_
74
+ >;
75
+
76
+ _minus = {.x x, .y y, .c c, 0b r} _minus_;
77
+ _minus_ = ${bits x, bits y, carry c, bits r}
78
+ <
79
+ {.y empty? if, .c/0 and, {.r x, .x r} _inv then} .then,
80
+ <
81
+ {.x/0 x, .y empty? y, .c/1 c1 c, .r |1 r},
82
+ {.x/1 x, .y empty? y, .c/1 c0 c, .r |0 r},
83
+ {.x/0 x, .y/0 y, .c/0 c0 c, .r |0 r},
84
+ {.x/0 x, .y/0 y, .c/1 c1 c, .r |1 r},
85
+ {.x/0 x, .y/1 y, .c/0 c1 c, .r |1 r},
86
+ {.x/0 x, .y/1 y, .c/1 c1 c, .r |0 r},
87
+ {.x/1 x, .y/0 y, .c/0 c0 c, .r |1 r},
88
+ {.x/1 x, .y/0 y, .c/1 c0 c, .r |0 r},
89
+ {.x/1 x, .y/1 y, .c/0 c0 c, .r |0 r},
90
+ {.x/1 x, .y/1 y, .c/1 c1 c, .r |1 r}
91
+ > _minus_
92
+ >;
93
+
94
+ plus_bits = ${bits x, bits y} {.x inv x, .y inv y, c0 c} _plus inv nat;
95
+ minus_bits = ${bits x, bits y} {.x inv x, .y inv y, c0 c} _minus inv nat;
96
+
97
+ _times = ${bits result, bits _x, bits y}
98
+ <
99
+ {.y empty? if, .result then} .then,
100
+ <
101
+ {.result |0 result, ._x _x, .y /0 y},
102
+ {{._x x, .result |0 y, c0 c} _plus result, ._x _x, .y /1 y}
103
+ > _times
104
+ >;
105
+ times_bits = ${bits x, bits y} {0 result, .x inv _x, .y y} _times inv nat;
106
+
107
+ unat_to_zeros = $unat < /_ 0b, /1 unat_to_zeros |0 > $bits;
108
+ unat_to_double_zeros = $unat
109
+ < /_ 0b, /1 unat_to_double_zeros |0 |0 > $bits;
110
+
111
+ _peel2 = < </0,/1> </0,/1>, </0,/1>, () >;
112
+ _half_unat = ${bits x, bits y, unat m}
113
+ <
114
+ {.x empty? if, .y empty? and, .m then} .then,
115
+ {.x _peel2 x, .y _peel2 y, .m |1 m} _half_unat
116
+ >;
117
+ half_unat = {.x x, .y y, 0u m} _half_unat;
118
+
119
+ _split_rev_u = ${bits x_rev, unat m, bits low}
120
+ <
121
+ {.m /_ if, {.low nat low, .x_rev inv nat high} then} .then,
122
+ {.x_rev empty? if, {.low nat low, 0 high} then} .then,
123
+ <
124
+ {.x_rev /0 x_rev, .m /1 m, .low |0 low},
125
+ {.x_rev /1 x_rev, .m /1 m, .low |1 low}
126
+ > _split_rev_u
127
+ >;
128
+ split_at = {.x inv x_rev, .m m, 0b low} _split_rev_u;
129
+
130
+ at_least_256_u? = <
131
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
132
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
133
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
134
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
135
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
136
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
137
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
138
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
139
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
140
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
141
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
142
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
143
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
144
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
145
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
146
+ /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1 /1
147
+ yes, no
148
+ >;
149
+
150
+ shift_bits_u = ${bits val, unat m}
151
+ <
152
+ {.val empty? if, 0b then} .then,
153
+ {.val x, .m unat_to_zeros y} concat
154
+ >;
155
+
156
+ shift2_bits_u = ${bits val, unat m}
157
+ <
158
+ {.val empty? if, 0b then} .then,
159
+ {.val x, .m unat_to_double_zeros y} concat
160
+ >;
161
+
162
+ _karatsuba = ${bits x, bits y}
163
+ <
164
+ -- Base case: if m < 256 (< 512 bits), fall back to times_bits
165
+ {{.x x, .y y} half_unat m, .x x, .y y}
166
+ <
167
+ {
168
+ .m at_least_256_u? /no if,
169
+ {.x x, .y y} times_bits then
170
+ } .then,
171
+ -- Recursive Karatsuba:
172
+ {
173
+ {.x x, .m m} split_at sx,
174
+ {.y x, .m m} split_at sy,
175
+ .m m
176
+ }
177
+ {
178
+ {.sx.high x, .sy.high y} _karatsuba z2,
179
+ {.sx.low x, .sy.low y} _karatsuba z0,
180
+ {
181
+ {.sx.high x, .sx.low y} plus_bits x,
182
+ {.sy.high x, .sy.low y} plus_bits y
183
+ } _karatsuba z_sum,
184
+ .m m
185
+ }
186
+ {
187
+ .z2 z2,
188
+ .z0 z0,
189
+ {{.z_sum x, .z2 y} minus_bits x, .z0 y} minus_bits z1,
190
+ .m m
191
+ }
192
+ {
193
+ {.z2 val, .m m} shift2_bits_u z2_s,
194
+ {.z1 val, .m m} shift_bits_u z1_s,
195
+ .z0 z0
196
+ }
197
+ {{.z2_s x, .z1_s y} plus_bits x, .z0 y} plus_bits
198
+ >
199
+ >;
200
+
201
+ karatsuba = ${int x, int y}
202
+ {
203
+ <
204
+ {.x/'+' if, .y /'+' and} sign_plus,
205
+ {.x/'-' if, .y /'-' and} sign_plus,
206
+ sign_minus
207
+ > sign,
208
+ {.x abs x, .y abs y} _karatsuba result
209
+ }
210
+ <
211
+ {.sign/'+' if, .result int then} .then,
212
+ {.sign/'-' if, .result int swap_sign then} .then
213
+ >;
214
+
215
+ ()
@@ -6,6 +6,22 @@ import { fileURLToPath } from "node:url";
6
6
 
7
7
  const root = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..");
8
8
  const node = process.execPath;
9
+
10
+ function printHelp() {
11
+ console.log(`Usage: node backends/arm64/scripts/run-tests.mjs [options]
12
+
13
+ Runs ARM64 backend test suite and integration tests.
14
+
15
+ Options:
16
+ -h, --help
17
+ Show this help message and exit.`);
18
+ }
19
+
20
+ if (process.argv.includes("-h") || process.argv.includes("--help")) {
21
+ printHelp();
22
+ process.exit(0);
23
+ }
24
+
9
25
  const tests = [
10
26
  ["tests/test-arm64.mjs", node, ["tests/test-arm64.mjs"]],
11
27
  ["tests/integration.sh", path.join(root, "tests/integration.sh"), []]
@@ -1,21 +1,25 @@
1
1
  #!/usr/bin/env node
2
2
 
3
+ import fs from "node:fs";
3
4
  import { argv, exit, stdin } from "node:process";
4
5
  import {
5
6
  compileProgramInputToObject,
6
7
  parseCompileOptions,
7
8
  resolveProgramInput
8
9
  } from "../src/cli.mjs";
9
- import { compileObjectToExecutable } from "../src/executable.mjs";
10
+ import {
11
+ compileKVMToExecutable,
12
+ compileObjectToExecutable
13
+ } from "../src/executable.mjs";
10
14
 
11
15
  function usage(stream = console.error) {
12
16
  const prog = argv[1] || "k-llvm-build.mjs";
13
17
  stream(`Usage: node ${prog} [options] [source-snippet | input-file [output-exe]]`);
14
- stream("Compile k source, .ko, or .klib input into a native executable.");
18
+ stream("Compile k source, .ko, .klib, or .kvm input into a native executable.");
15
19
  stream("");
16
20
  stream("Arguments:");
17
21
  stream(" source-snippet Inline k source, in the same style as k.mjs.");
18
- stream(" input-file Source .k, .ko, or .klib file. Reads UTF-8 source from stdin when omitted.");
22
+ stream(" input-file Source .k, .ko, .klib, or .kvm file. Reads UTF-8 source from stdin when omitted.");
19
23
  stream(" output-exe Output executable path.");
20
24
  stream("");
21
25
  stream("The executable reads a binary k pattern+value envelope from stdin and");
@@ -24,11 +28,21 @@ function usage(stream = console.error) {
24
28
  stream("stdout is encoded with the compiled output pattern.");
25
29
  stream("");
26
30
  stream("Options:");
27
- stream(" -o, --output path Specify output file path explicitly.");
28
- stream(" --lib file Load one .klib dependency before compiling.");
29
- stream(" --export spec Export a library alias into source scope. May be repeated.");
30
- stream(" spec is 'name' or 'libname:localname'.");
31
- stream(" -h, --help Show this help.");
31
+ stream(" -o, --output path Specify output file path explicitly.");
32
+ stream(" --input-pattern pattern Input pattern JSON or file path for specialization.");
33
+ stream(" --lib file Load one .klib dependency before compiling.");
34
+ stream(" --export spec Export a library alias into source scope. May be repeated.");
35
+ stream(" spec is 'name' or 'libname:localname'.");
36
+ stream(" -h, --help Show this help.");
37
+ }
38
+
39
+ function readMaybeFile(textOrPath) {
40
+ return fs.existsSync(textOrPath) ? fs.readFileSync(textOrPath, "utf8") : textOrPath;
41
+ }
42
+
43
+ function readPattern(inputPattern) {
44
+ if (inputPattern == null) return null;
45
+ return JSON.parse(readMaybeFile(inputPattern));
32
46
  }
33
47
 
34
48
  try {
@@ -39,6 +53,7 @@ try {
39
53
  }
40
54
 
41
55
  let explicitOutput = null;
56
+ let rawInputPattern = null;
42
57
  const remainingArgs = [];
43
58
  while (args.length > 0) {
44
59
  const arg = args[0];
@@ -46,11 +61,17 @@ try {
46
61
  args.shift();
47
62
  explicitOutput = args.shift();
48
63
  if (!explicitOutput) throw new Error("-o/--output requires a file argument");
64
+ } else if (arg === "--input-pattern") {
65
+ args.shift();
66
+ rawInputPattern = args.shift();
67
+ if (!rawInputPattern) throw new Error("--input-pattern requires a JSON or file argument");
49
68
  } else {
50
69
  remainingArgs.push(args.shift());
51
70
  }
52
71
  }
53
72
 
73
+ const inputPattern = readPattern(rawInputPattern);
74
+
54
75
  const { libraries, exportSpecs } = parseCompileOptions(remainingArgs);
55
76
  const input = resolveProgramInput(remainingArgs, { allowStdinSource: true });
56
77
  const positionalOutput = remainingArgs.shift();
@@ -59,13 +80,18 @@ try {
59
80
  const outputPath = explicitOutput || positionalOutput || null;
60
81
  if (outputPath == null) throw new Error("Output executable path is required (pass output-exe or -o/--output)");
61
82
 
62
- const object = await compileProgramInputToObject(input, {
63
- libraries,
64
- exportSpecs,
65
- stdin
66
- });
83
+ if (input.kind === "kvm") {
84
+ const kvm = JSON.parse(fs.readFileSync(input.path, "utf8"));
85
+ compileKVMToExecutable(kvm, outputPath, { inputPattern });
86
+ } else {
87
+ const object = await compileProgramInputToObject(input, {
88
+ libraries,
89
+ exportSpecs,
90
+ stdin
91
+ });
67
92
 
68
- compileObjectToExecutable(object, outputPath);
93
+ compileObjectToExecutable(object, outputPath, { inputPattern });
94
+ }
69
95
  } catch (error) {
70
96
  console.error(error.stack || error.message || String(error));
71
97
  usage();
@@ -7,17 +7,20 @@ import {
7
7
  parseCompileOptions,
8
8
  resolveProgramInput
9
9
  } from "../src/cli.mjs";
10
- import { compileObjectToLLVM } from "../src/llvm.mjs";
10
+ import {
11
+ compileObjectToLLVM,
12
+ compileLLVMArtifactFromKVM
13
+ } from "../src/llvm.mjs";
11
14
  import { inputPatternForObjectRelation } from "../src/executable.mjs";
12
15
 
13
16
  function usage(stream = console.error) {
14
17
  const prog = argv[1] || "k-llvm-compile.mjs";
15
18
  stream(`Usage: node ${prog} [options] [source-snippet | input-file [output.ll]]`);
16
- stream("Compile a k .ko/.klib object into prototype LLVM IR.");
19
+ stream("Compile a k .ko/.klib object, source, or .kvm input into prototype LLVM IR.");
17
20
  stream("");
18
21
  stream("Arguments:");
19
22
  stream(" source-snippet Inline k source, in the same style as k.mjs.");
20
- stream(" input-file Source .k, .ko, or .klib file. Reads UTF-8 source from stdin when omitted.");
23
+ stream(" input-file Source .k, .ko, .klib, or .kvm file. Reads UTF-8 source from stdin when omitted.");
21
24
  stream(" output.ll Output LLVM IR path. Writes to stdout when omitted.");
22
25
  stream("");
23
26
  stream("Options:");
@@ -55,16 +58,30 @@ try {
55
58
 
56
59
  const outputPath = explicitOutput || positionalOutput || null;
57
60
 
58
- const object = await compileProgramInputToObject(input, {
59
- libraries,
60
- exportSpecs,
61
- stdin
62
- });
61
+ let llvm;
62
+ if (input.kind === "kvm") {
63
+ const kvm = JSON.parse(fs.readFileSync(input.path, "utf8"));
64
+ const entry = kvm.entry || kvm.relation || "__main__";
65
+ const mainFunc = kvm.functions?.[entry];
66
+ const artifact = compileLLVMArtifactFromKVM(kvm, {
67
+ relation: entry,
68
+ inputPattern: mainFunc?.inputPattern || kvm.inputPattern,
69
+ outputPattern: mainFunc?.outputPattern || kvm.outputPattern
70
+ });
71
+ llvm = typeof artifact === "string" ? artifact : artifact.llvm;
72
+ } else {
73
+ const object = await compileProgramInputToObject(input, {
74
+ libraries,
75
+ exportSpecs,
76
+ stdin
77
+ });
63
78
 
64
- const { llvm } = compileObjectToLLVM(object, {
65
- relation: object.main,
66
- inputPattern: inputPatternForObjectRelation(object, object.main)
67
- });
79
+ const res = compileObjectToLLVM(object, {
80
+ relation: object.main,
81
+ inputPattern: inputPatternForObjectRelation(object, object.main)
82
+ });
83
+ llvm = res.llvm;
84
+ }
68
85
 
69
86
  if (outputPath == null) {
70
87
  stdout.write(llvm);
@@ -553,17 +553,17 @@ static void rt_rewind_and_free(k_rt *rt, k_rt_checkpoint mark) {
553
553
  rt->has_reusable_blocks = 0;
554
554
  }
555
555
 
556
- k_value *k_rt_compact(k_rt *rt, k_value *root, k_rt_checkpoint *mark_ptr) {
557
- if (rt == NULL || root == NULL || mark_ptr == NULL) return root;
556
+ k_value *k_rt_compact(k_rt *rt, k_value *root, k_rt_tail_mark *tm) {
557
+ if (rt == NULL || root == NULL || tm == NULL) return root;
558
558
 
559
- size_t bytes = bytes_since_mark(rt, *mark_ptr);
559
+ size_t bytes = bytes_since_mark(rt, tm->last_mark);
560
560
  if (bytes < 4194304) {
561
561
  return root;
562
562
  }
563
563
 
564
- if (!is_after_mark(rt, root, *mark_ptr)) {
565
- rt_rewind_and_free(rt, *mark_ptr);
566
- *mark_ptr = k_rt_mark(rt);
564
+ if (!is_after_mark(rt, root, tm->base_mark)) {
565
+ rt_rewind_and_free(rt, tm->base_mark);
566
+ tm->last_mark = tm->base_mark;
567
567
  return root;
568
568
  }
569
569
 
@@ -574,14 +574,14 @@ k_value *k_rt_compact(k_rt *rt, k_value *root, k_rt_checkpoint *mark_ptr) {
574
574
 
575
575
  k_fixup_list fixups = {NULL, 0, 0};
576
576
 
577
- k_value *staged_root = copy_value(rt, root, &staging, &staging_used, &staging_cap, &fixups, *mark_ptr);
577
+ k_value *staged_root = copy_value(rt, root, &staging, &staging_used, &staging_cap, &fixups, tm->base_mark);
578
578
 
579
- rt_rewind_and_free(rt, *mark_ptr);
579
+ rt_rewind_and_free(rt, tm->base_mark);
580
580
 
581
581
  if (staging_used == 0) {
582
582
  free(staging);
583
583
  if (fixups.offsets != NULL) free(fixups.offsets);
584
- *mark_ptr = k_rt_mark(rt);
584
+ tm->last_mark = tm->base_mark;
585
585
  return staged_root;
586
586
  }
587
587
 
@@ -605,7 +605,7 @@ k_value *k_rt_compact(k_rt *rt, k_value *root, k_rt_checkpoint *mark_ptr) {
605
605
  free(staging);
606
606
  if (fixups.offsets != NULL) free(fixups.offsets);
607
607
 
608
- *mark_ptr = k_rt_mark(rt);
608
+ tm->last_mark = k_rt_mark(rt);
609
609
 
610
610
  return new_root;
611
611
  }
@@ -13,6 +13,11 @@ typedef struct {
13
13
  size_t used;
14
14
  } k_rt_checkpoint;
15
15
 
16
+ typedef struct {
17
+ k_rt_checkpoint base_mark;
18
+ k_rt_checkpoint last_mark;
19
+ } k_rt_tail_mark;
20
+
16
21
  typedef struct {
17
22
  const char *label;
18
23
  size_t label_length;
@@ -66,7 +71,7 @@ k_rt *k_rt_new(void);
66
71
  void k_rt_reset(k_rt *rt);
67
72
  k_rt_checkpoint k_rt_mark(k_rt *rt);
68
73
  void k_rt_rewind(k_rt *rt, k_rt_checkpoint mark);
69
- k_value *k_rt_compact(k_rt *rt, k_value *root, k_rt_checkpoint *mark_ptr);
74
+ k_value *k_rt_compact(k_rt *rt, k_value *root, k_rt_tail_mark *mark_ptr);
70
75
  void *k_rt_alloc(k_rt *rt, size_t size);
71
76
  void k_rt_free(k_rt *rt);
72
77
 
@@ -6,6 +6,22 @@ import { fileURLToPath } from "node:url";
6
6
 
7
7
  const root = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..");
8
8
  const node = process.execPath;
9
+
10
+ function printHelp() {
11
+ console.log(`Usage: node backends/llvm/scripts/run-tests.mjs [options]
12
+
13
+ Runs LLVM backend test suite, integration tests, and conformance suite.
14
+
15
+ Options:
16
+ -h, --help
17
+ Show this help message and exit.`);
18
+ }
19
+
20
+ if (process.argv.includes("-h") || process.argv.includes("--help")) {
21
+ printHelp();
22
+ process.exit(0);
23
+ }
24
+
9
25
  const tests = [
10
26
  ["tests/test-llvm.mjs", node, ["tests/test-llvm.mjs"]],
11
27
  ["tests/integration.sh", path.join(root, "tests/integration.sh"), []],
@@ -135,7 +135,7 @@ export async function compileProgramInputToObject(
135
135
  } else if (input.kind === "ko" || input.kind === "klib") {
136
136
  object = decodeObject(fs.readFileSync(input.path));
137
137
  } else if (input.kind === "kvm") {
138
- throw new Error("Compiling directly from .kvm to LLVM is not yet supported; use .k or .ko input.");
138
+ throw new Error("Input is already lowered to .kvm bytecode; use compileLLVMArtifactFromKVM or compileKVMToExecutable.");
139
139
  } else {
140
140
  throw new Error(`Unsupported input kind: ${input.kind}`);
141
141
  }