perf: enable wasm simd (nodejs#735)

* perf: enable wasm simd

* bench: enable simd

* build: add wasm simd

* ci: use node 16 in benchmarks

* test: add simd test script

* ci: add simd bench

* enable simd by default in tests and benchmarks

* fix machine specs in README.md

Co-authored-by: Robert Nagy <ronagy@icloud.com>

* client: fallback to non-simd on all errors

* fixup: re-enable jest

* fixup

Co-authored-by: Daniele Belardi <dwon.dnl@gmail.com>

.taprc

@@ -5,3 +5,4 @@ coverage: true
 expose-gc: true
 timeout: 60
 check-coverage: false
+node-arg: --experimental-wasm-simd

README.md

@@ -17,32 +17,36 @@ npm i undici
 
 ## Benchmarks
 
-AMD EPYC 7502P 32 Core, Node 15
+Node 16
 
 The benchmark is a simple `hello world` [example](benchmarks/index.js) using a
 number of unix sockets (connections) with a pipelining depth of 10.
 
 ### Connections 1
 
-| Test                | Samples |        Result  | Tolerance | Difference with slowest |
-|---------------------|---------|----------------|-----------|-------------------------|
-| http - no keepalive |      99 |   812 reqs/sec |  ± 0.22 % |                         |
-| http - keepalive    |      99 |   819 reqs/sec |  ± 0.20 % | + 0.82 %                |
-| undici - pipeline   |      99 |  6632 reqs/sec |  ± 0.63 % | + 716.73 %              |
-| undici - request    |      99 |  6645 reqs/sec |  ± 1.34 % | + 718.34 %              |
-| undici - stream     |      99 |  7366 reqs/sec |  ± 0.59 % | + 807.11 %              |
-| undici - dispatch   |      99 |  7404 reqs/sec |  ± 0.37 % | + 811.76 %              |
+| Tests               | Samples |        Result | Tolerance | Difference with slowest |
+|---------------------|---------|---------------|-----------|-------------------------|
+| http - no keepalive |      15 |  4.63 req/sec |  ± 2.77 % |                       - |
+| http - keepalive    |      10 |  4.81 req/sec |  ± 2.16 % |                + 3.94 % |
+| undici - stream     |      25 | 62.22 req/sec |  ± 2.67 % |             + 1244.58 % |
+| undici - dispatch   |      15 | 64.33 req/sec |  ± 2.47 % |             + 1290.24 % |
+| undici - request    |      15 | 66.08 req/sec |  ± 2.48 % |             + 1327.88 % |
+| undici - pipeline   |      10 | 66.13 req/sec |  ± 1.39 % |             + 1329.08 % |
 
 ### Connections 50
 
-| Test                | Samples |        Result  | Tolerance | Difference with slowest |
-|---------------------|---------|----------------|-----------|-------------------------|
-| http - no keepalive |      99 | 12968 reqs/sec |  ± 1.86 % |                         |
-| http - keepalive    |      99 | 14745 reqs/sec |  ± 1.59 % | + 13.70 %               |
-| undici - pipeline   |      99 | 20051 reqs/sec |  ± 2.34 % | + 54.62 %               |
-| undici - stream     |     100 | 26456 reqs/sec |  ± 3.50 % | + 104.00 %              |
-| undici - request    |      99 | 29342 reqs/sec |  ± 1.26 % | + 126.26 %              |
-| undici - dispatch   |      99 | 35323 reqs/sec |  ± 0.77 % | + 172.38 %              |
+| Tests               | Samples |           Result | Tolerance | Difference with slowest |
+|---------------------|---------|------------------|-----------|-------------------------|
+| http - no keepalive |      50 |  3546.49 req/sec |  ± 2.90 % |                       - |
+| http - keepalive    |      15 |  5692.67 req/sec |  ± 2.48 % |               + 60.52 % |
+| undici - pipeline   |      25 |  8478.71 req/sec |  ± 2.62 % |              + 139.07 % |
+| undici - request    |      20 |  9766.66 req/sec |  ± 2.79 % |              + 175.39 % |
+| undici - stream     |      15 | 10109.74 req/sec |  ± 2.94 % |              + 185.06 % |
+| undici - dispatch   |      25 | 10949.73 req/sec |  ± 2.54 % |              + 208.75 % |
+
+#### Note
+
+The benchmarks have the [simd](https://github.com/WebAssembly/simd) feature enabled.
 
 ## Quick Start
 

build/wasm.js

@@ -42,3 +42,25 @@ execSync(`${WASI_ROOT}/bin/clang \
  ${join(WASM_SRC, 'src')}/*.c \
  -I${join(WASM_SRC, 'include')} \
  -o ${join(WASM_OUT, 'llhttp.wasm')}`, { stdio: 'inherit' })
+
+// Build wasm simd binary
+execSync(`${WASI_ROOT}/bin/clang \
+ --sysroot=${WASI_ROOT}/share/wasi-sysroot \
+ -target wasm32-unknown-wasi \
+ -msimd128 \
+ -Ofast \
+ -fno-exceptions \
+ -fvisibility=hidden \
+ -mexec-model=reactor \
+ -Wl,-error-limit=0 \
+ -Wl,-O3 \
+ -Wl,--lto-O3 \
+ -Wl,--strip-all \
+ -Wl,--allow-undefined \
+ -Wl,--export-dynamic \
+ -Wl,--export-table \
+ -Wl,--export=malloc \
+ -Wl,--export=free \
+ ${join(WASM_SRC, 'src')}/*.c \
+ -I${join(WASM_SRC, 'include')} \
+ -o ${join(WASM_OUT, 'llhttp_simd.wasm')}`, { stdio: 'inherit' })

lib/client.js

@@ -23,12 +23,6 @@ const {
   BodyTimeoutError
 } = require('./core/errors')
 
-const { resolve } = require('path')
-const { readFileSync } = require('fs')
-const constants = require('./llhttp/constants')
-const EMPTY_BUF = Buffer.alloc(0)
-const mod = new WebAssembly.Module(readFileSync(resolve(__dirname, './llhttp/llhttp.wasm')))
-
 const {
   kUrl,
   kReset,
@@ -387,6 +381,26 @@ class HTTPParserError extends Error {
   }
 }
 
+let mod, build
+const { resolve } = require('path')
+const { readFileSync } = require('fs')
+const constants = require('./llhttp/constants')
+const EMPTY_BUF = Buffer.alloc(0)
+
+try {
+  build = resolve(__dirname, './llhttp/llhttp_simd.wasm')
+  const bin = readFileSync(build)
+  mod = new WebAssembly.Module(bin)
+} catch (e) {
+  // We could check if the error was caused by the simd option not
+  // being enabled, but the occurring of this other error
+  // * https://github.com/emscripten-core/emscripten/issues/11495
+  // got me to remove that check to avoid breaking Node 12.
+  build = resolve(__dirname, './llhttp/llhttp.wasm')
+  const bin = readFileSync(build)
+  mod = new WebAssembly.Module(bin)
+}
+
 const llhttp = new WebAssembly.Instance(mod, {
   env: {
     /* eslint-disable camelcase */

package.json

@@ -39,9 +39,10 @@
     "coverage": "standard | snazzy && tap test/*.js",
     "coverage:ci": "npm run coverage -- --coverage-report=lcovonly",
     "bench": "concurrently -k -s first npm:bench:server npm:bench:run",
+    "bench:simd": "concurrently -k -s first npm:bench:server npm:bench:run:simd",
     "bench:server": "node benchmarks/server.js",
     "prebench:run": "node benchmarks/wait.js",
-    "bench:run": "CONNECTIONS=1 node benchmarks/benchmark.js && CONNECTIONS=50 node benchmarks/benchmark.js",
+    "bench:run": "CONNECTIONS=1 node --experimental-wasm-simd benchmarks/benchmark.js && CONNECTIONS=50 node --experimental-wasm-simd benchmarks/benchmark.js",
     "serve:website": "docsify serve .",
     "prepare": "husky install"
   },