From: Chris Duncan <chris@zoso.dev>
Date: Tue, 7 Jan 2025 13:54:18 +0000 (-0800)
Subject: Log some more benchmarks. Switch back to PowGl for blocks, with the plan to later... 
X-Git-Url: https://zoso.dev/?a=commitdiff_plain;h=a7914e49c045ae973b8ba1fac8ebb63ae614b578;p=libnemo.git

Log some more benchmarks. Switch back to PowGl for blocks, with the plan to later make PowGpu available behind a flag.
---

diff --git a/benchmarks.md b/benchmarks.md
index 3b0706f..389caa7 100644
--- a/benchmarks.md
+++ b/benchmarks.md
@@ -1,15 +1,14 @@
+PASS  Original PoW module: Time to calculate proof-of-work for a send block 16 times
 Total: 89756 ms
 Average: 5609.75 ms
 Harmonic: 2092.567565254879 ms
 Geometric: 3612.112662613675 ms
-PASS  Original PoW module: Time to calculate proof-of-work for a send block 16 times
-
 
+PASS  Customized PoW: Time to calculate proof-of-work for a send block 16 times
 Total: 33240 ms
 Average: 2077.5 ms
 Harmonic: 1328.5635414262717 ms
 Geometric: 1663.110986923899 ms
-PASS  Customized PoW: Time to calculate proof-of-work for a send block 16 times
 
 How much faster?
 Total:    56156 ms
@@ -17,6 +16,14 @@ Average:   3532 ms
 Harmonic:   764 ms
 Geometric: 1949 ms
 
+Another PowGl test:
+Total: 22831.300000041723 ms
+Average: 3805.2166666736207 ms
+Harmonic: 928.6432328540742 ms
+Geometric: 2500.810238375608 ms
+Minimum: 193 ms
+Maximum: 8361 ms
+
 The proof-of-work equation for Nano cryptocurrency is defined as `blake2b(nonce||blockhash)>=threshold` where blake2b is the hash function configured for an 8-byte output, nonce is a random 8-byte value, || is concatenation, blockhash is a 32-byte value, and threshold is 0xfffffff800000000.
 
 My code currently finds valid nonces on a gaming GPU without issue but only because the initial search space is so large due to using a workgroup size of 256 and a dispatch of (256, 256, 256), so the probability of finding a nonce in the first pass is extremely high. However, this does not perform well on less powerful hardware like smartphones. Please alter my code to perform the nonce search in `main()` in a loop until a valid nonce is found with the idea that a smaller workgroup size and/or smaller dispatch dimensions can allow weak hardware to nonetheless iterate through nonces to search for a valid one and return it.
diff --git a/perf/block.perf.js b/perf/block.perf.js
index fae621b..d1ee578 100644
--- a/perf/block.perf.js
+++ b/perf/block.perf.js
@@ -37,7 +37,7 @@ await suite('Block performance', async () => {
 		console.log(`Maximum: ${max} ms`)
 	})
 
-	await skip(`PowGl: Time to calculate proof-of-work for a block hash ${COUNT} times`, async () => {
+	await test(`PowGl: Time to calculate proof-of-work for a block hash ${COUNT} times`, async () => {
 		const times = []
 		const hashes = [
 			NANO_TEST_VECTORS.PRIVATE_0,
@@ -63,7 +63,7 @@ await suite('Block performance', async () => {
 		console.log(`Maximum: ${max} ms`)
 	})
 
-	await test(`PowGpu: Time to calculate proof-of-work for a send block ${COUNT} times`, async () => {
+	await skip(`PowGpu: Time to calculate proof-of-work for a send block ${COUNT} times`, async () => {
 		const times = []
 		const block = new SendBlock(
 			NANO_TEST_VECTORS.SEND_BLOCK.account,
diff --git a/src/lib/block.ts b/src/lib/block.ts
index 11b969f..1d36116 100644
--- a/src/lib/block.ts
+++ b/src/lib/block.ts
@@ -16,7 +16,7 @@ import { PowGl, PowGpu } from './workers.js'
 * of three derived classes: SendBlock, ReceiveBlock, ChangeBlock.
 */
 abstract class Block {
-	static #pool: Pool = new Pool(PowGpu)
+	static #pool: Pool = new Pool(PowGl)
 	account: Account
 	type: string = 'state'
 	abstract subtype: 'send' | 'receive' | 'change'