wazero
wazero copied to clipboard
Published
20 hours ago •
tetratelabs
tetratelabs
ssa: empirically faster passRedundantPhiEliminationOpt
This changes the order of traversal of basic blocks in passRedundantPhiEliminationOpt
optimization pass in a way that the algorithm converges more quickly.
In order to do so, run the pass after the control flow analysis and traverse the blocks
in the reverse post order.
This improves the compilation performance like 10-30% especially for Go and TinyGo binaries without having any influence on the runtime performance (exactly the same executables as before).
Stdlibs
wasip1
goos: darwin
goarch: arm64
pkg: github.com/tetratelabs/wazero/internal/integration_test/stdlibs
│ old_wasip1.txt │ new_wasip1.txt │
│ sec/op │ sec/op vs base │
Wasip1/Compile/src_archive_tar.test-10 3.114 ± 2% 2.335 ± 7% -25.02% (p=0.001 n=7)
Wasip1/Run/src_archive_tar.test-10 399.7m ± 3% 399.9m ± 2% ~ (p=0.456 n=7)
Wasip1/Compile/src_bufio.test-10 1.874 ± 1% 1.574 ± 1% -16.00% (p=0.001 n=7)
Wasip1/Run/src_bufio.test-10 122.9m ± 13% 125.6m ± 4% ~ (p=0.620 n=7)
Wasip1/Compile/src_bytes.test-10 1.987 ± 1% 1.639 ± 1% -17.50% (p=0.001 n=7)
Wasip1/Run/src_bytes.test-10 471.4m ± 0% 470.4m ± 0% ~ (p=0.456 n=7)
Wasip1/Compile/src_context.test-10 2.086 ± 1% 1.752 ± 3% -16.02% (p=0.001 n=7)
Wasip1/Run/src_context.test-10 31.78m ± 9% 30.91m ± 2% -2.74% (p=0.038 n=7)
Wasip1/Compile/src_encoding_ascii85.test-10 1.727 ± 3% 1.407 ± 1% -18.54% (p=0.001 n=7)
Wasip1/Run/src_encoding_ascii85.test-10 7.536m ± 3% 7.530m ± 1% ~ (p=1.000 n=7)
Wasip1/Compile/src_encoding_asn1.test-10 1.948 ± 1% 1.637 ± 1% -15.99% (p=0.001 n=7)
Wasip1/Run/src_encoding_asn1.test-10 8.542m ± 6% 8.425m ± 2% ~ (p=0.128 n=7)
Wasip1/Compile/src_encoding_base32.test-10 1.790 ± 1% 1.462 ± 1% -18.32% (p=0.001 n=7)
Wasip1/Run/src_encoding_base32.test-10 28.82m ± 3% 27.81m ± 1% -3.53% (p=0.002 n=7)
Wasip1/Compile/src_encoding_base64.test-10 1.817 ± 1% 1.472 ± 0% -19.00% (p=0.001 n=7)
Wasip1/Run/src_encoding_base64.test-10 12.31m ± 2% 12.17m ± 2% ~ (p=0.209 n=7)
Wasip1/Compile/src_encoding_binary.test-10 1.843 ± 2% 1.651 ± 14% -10.38% (p=0.026 n=7)
Wasip1/Run/src_encoding_binary.test-10 9.444m ± 4% 9.583m ± 4% +1.46% (p=0.038 n=7)
Wasip1/Compile/src_encoding_csv.test-10 1.789 ± 1% 1.500 ± 1% -16.17% (p=0.001 n=7)
Wasip1/Run/src_encoding_csv.test-10 19.99m ± ∞ ¹ 20.85m ± 4% ~ (p=0.250 n=1+7)
Wasip1/Compile/src_encoding_gob.test-10 2.165 ± 1%
geomean 264.3m 314.9m -9.16%
TinyGo
goos: darwin
goarch: arm64
pkg: github.com/tetratelabs/wazero/internal/integration_test/stdlibs
│ old_tinygo.txt │ new_tinygo.txt │
│ sec/op │ sec/op vs base │
TinyGo/Compile/container_heap.test-10 668.8m ± 2% 445.6m ± 9% -33.38% (p=0.001 n=7)
TinyGo/Run/container_heap.test-10 14.42m ± 11% 14.53m ± 7% ~ (p=0.318 n=7)
TinyGo/Compile/container_list.test-10 663.8m ± 1% 433.9m ± 2% -34.64% (p=0.001 n=7)
TinyGo/Run/container_list.test-10 14.26m ± 10% 14.17m ± 7% ~ (p=0.318 n=7)
TinyGo/Compile/container_ring.test-10 665.9m ± 1% 435.6m ± 4% -34.58% (p=0.001 n=7)
TinyGo/Run/container_ring.test-10 14.24m ± 12% 14.49m ± 2% ~ (p=0.259 n=7)
TinyGo/Compile/crypto_des.test-10 692.5m ± 4% 452.5m ± 9% -34.66% (p=0.001 n=7)
TinyGo/Run/crypto_des.test-10 18.32m ± 4% 18.14m ± 3% ~ (p=0.128 n=7)
TinyGo/Compile/crypto_md5.test-10 681.3m ± 3% 450.5m ± 4% -33.88% (p=0.001 n=7)
TinyGo/Run/crypto_md5.test-10 20.90m ± 4% 20.80m ± 8% ~ (p=1.000 n=7)
TinyGo/Compile/crypto_rc4.test-10 673.2m ± 4% 429.0m ± 3% -36.28% (p=0.001 n=7)
TinyGo/Run/crypto_rc4.test-10 162.7m ± 1% 162.7m ± 1% ~ (p=0.902 n=7)
TinyGo/Compile/crypto_sha1.test-10 679.4m ± 3% 437.5m ± 3% -35.61% (p=0.001 n=7)
TinyGo/Run/crypto_sha1.test-10 16.02m ± 6% 16.33m ± 3% ~ (p=0.535 n=7)
TinyGo/Compile/crypto_sha256.test-10 697.0m ± 4% 445.0m ± 0% -36.15% (p=0.001 n=7)
TinyGo/Run/crypto_sha256.test-10 16.10m ± 7% 15.91m ± 1% -1.13% (p=0.017 n=7)
TinyGo/Compile/crypto_sha512.test-10 698.2m ± 3% 450.0m ± 2% -35.56% (p=0.001 n=7)
TinyGo/Run/crypto_sha512.test-10 20.26m ± 7% 18.76m ± 1% -7.41% (p=0.017 n=7)
TinyGo/Compile/encoding_ascii85.test-10 681.4m ± 4% 435.9m ± 1% -36.03% (p=0.001 n=7)
TinyGo/Run/encoding_ascii85.test-10 14.88m ± 1% 14.93m ± 12% ~ (p=0.318 n=7)
TinyGo/Compile/encoding_base32.test-10 1.645 ± 4% 1.165 ± 4% -29.19% (p=0.001 n=7)
TinyGo/Run/encoding_base32.test-10 86.46m ± 4% 88.18m ± 5% ~ (p=0.209 n=7)
TinyGo/Compile/encoding_csv.test-10 734.2m ± 1% 483.1m ± 5% -34.20% (p=0.001 n=7)
TinyGo/Run/encoding_csv.test-10 22.85m ± 6% 22.86m ± 0% ~ (p=1.000 n=7)
TinyGo/Compile/encoding_hex.test-10 699.8m ± 5% 455.9m ± 2% -34.86% (p=0.001 n=7)
TinyGo/Run/encoding_hex.test-10 10.23m ± 4% 10.23m ± 5% ~ (p=0.805 n=7)
TinyGo/Compile/go_scanner.test-10 752.3m ± 3% 485.0m ± 1% -35.53% (p=0.001 n=7)
TinyGo/Run/go_scanner.test-10 16.70m ± 5% 16.62m ± 8% ~ (p=0.620 n=7)
TinyGo/Compile/hash.test-10 948.3m ± 2% 690.5m ± 1% -27.19% (p=0.001 n=7)
TinyGo/Run/hash.test-10 16.19m ± 5% 16.14m ± 5% ~ (p=0.902 n=7)
TinyGo/Compile/hash_adler32.test-10 663.8m ± 3% 423.4m ± 3% -36.22% (p=0.001 n=7)
TinyGo/Run/hash_adler32.test-10 24.75m ± 8% 24.62m ± 1% ~ (p=0.535 n=7)
TinyGo/Compile/hash_crc64.test-10 887.8m ± 3% 653.2m ± 1% -26.43% (p=0.001 n=7)
TinyGo/Run/hash_crc64.test-10 17.27m ± 1% 17.53m ± 5% +1.48% (p=0.017 n=7)
TinyGo/Compile/hash_fnv.test-10 685.6m ± 7% 443.7m ± 4% -35.28% (p=0.001 n=7)
TinyGo/Run/hash_fnv.test-10 15.49m ± 4% 14.59m ± 4% -5.85% (p=0.002 n=7)
TinyGo/Compile/html.test-10 2.476 ± 5% 2.188 ± 1% -11.63% (p=0.001 n=7)
TinyGo/Compile/internal_itoa.test-10 417.6m ± 1%
geomean 127.6m 108.0m -18.34%
Zig
goos: darwin
goarch: arm64
pkg: github.com/tetratelabs/wazero/internal/integration_test/stdlibs
│ old_zig.txt │ new_zig.txt │
│ sec/op │ sec/op vs base │
Zig/Compile/test-opt.wasm-10 4.577 ± 1% 4.697 ± 4% +2.63% (p=0.038 n=7)
Zig/Run/test-opt.wasm-10 18.91 ± 4% 18.96 ± 2% ~ (p=0.902 n=7)
Zig/Compile/test.wasm-10 5.710 ± 0% 5.707 ± 0% ~ (p=0.456 n=7)
Zig/Run/test.wasm-10 19.39 ± 1% 19.08 ± 2% -1.60% (p=0.011 n=7)
geomean 9.894 9.924 +0.30%
Compiling wazero compiled as Wasm
goos: darwin
goarch: arm64
pkg: github.com/tetratelabs/wazero
│ old.txt │ new.txt │
│ sec/op │ sec/op vs base │
Compilation-10 3.304 ± 2% 2.533 ± 1% -23.33% (p=0.001 n=7)
