Matt Hess
7730b6993f
- Create assembly/vm.ts with full RandomX VM implementation
- 256 instructions, 2048 iterations per hash
- Native u64/f64 operations in WebAssembly
- Full mode dataset lookups (2GB pre-computed)
- Update miner to use WASM VM for full mode
- mining-worker-asm.js uses pre-compiled WASM
- ~32 H/s per thread (4x faster than light mode)
- 8 threads achieves ~260 H/s
- Clean up redundant code
- Remove mining-worker-full.js (old JIT approach)
- Consolidate 'asm' mode into 'full' mode
61 lines
1.9 KiB
JavaScript
61 lines
1.9 KiB
JavaScript
/**
|
|
* Debug using @noble/hashes argon2 directly
|
|
*/
|
|
|
|
import { argon2d } from '@noble/hashes/argon2.js';
|
|
|
|
// Helper to convert bytes to hex
|
|
function bytesToHex(bytes) {
|
|
return Array.from(bytes).map(b => b.toString(16).padStart(2, '0')).join('');
|
|
}
|
|
|
|
// Helper to read little-endian uint64
|
|
function readLE64(bytes, offset) {
|
|
let val = 0n;
|
|
for (let i = 0; i < 8; i++) {
|
|
val |= BigInt(bytes[offset + i]) << BigInt(i * 8);
|
|
}
|
|
return val;
|
|
}
|
|
|
|
console.log('=== Test with @noble/hashes argon2d ===\n');
|
|
|
|
const key = new TextEncoder().encode("test key 000");
|
|
const salt = new Uint8Array([0x52, 0x61, 0x6e, 0x64, 0x6f, 0x6d, 0x58, 0x03]); // "RandomX\x03"
|
|
|
|
console.log('Key:', new TextDecoder().decode(key), `(${key.length} bytes)`);
|
|
console.log('Salt:', bytesToHex(salt), `(${salt.length} bytes)`);
|
|
console.log();
|
|
|
|
// RandomX uses Argon2d with:
|
|
// - Memory: 262144 KiB
|
|
// - Iterations (time cost): 3
|
|
// - Parallelism (lanes): 1
|
|
// - Output length: varies
|
|
|
|
// The noble argon2d function returns the hash output, not the memory
|
|
// For RandomX, we need the actual memory state after Argon2d
|
|
|
|
// Let's try to get the first 64 bytes of output
|
|
console.log('Testing noble argon2d with dkLen=64...');
|
|
|
|
try {
|
|
const result = argon2d(key, salt, {
|
|
t: 3, // iterations
|
|
m: 262144, // memory in KiB
|
|
p: 1, // parallelism
|
|
dkLen: 64 // output length
|
|
});
|
|
console.log('Result (64 bytes):', bytesToHex(result));
|
|
} catch (e) {
|
|
console.log('Error:', e.message);
|
|
}
|
|
|
|
// But for RandomX, the cache is the actual Argon2d memory, not the output
|
|
// The Argon2d output is derived from the final block, which is different from what RandomX uses
|
|
// RandomX uses the raw memory blocks as the cache
|
|
|
|
console.log('\nNote: RandomX uses the raw Argon2d memory as cache, not the output hash.');
|
|
console.log('The expected cacheMemory[0] = 0x191e0e1d23c02186 is the first qword of the first block.');
|
|
console.log('This is NOT the same as the Argon2d hash output.');
|