go/src/runtime/alg.go

490 lines
13 KiB
Go
Raw Normal View History

// Copyright 2014 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package runtime
import (
"internal/abi"
"internal/byteorder"
"internal/cpu"
"internal/goarch"
"internal/runtime/sys"
"unsafe"
)
const (
c0 = uintptr((8-goarch.PtrSize)/4*2860486313 + (goarch.PtrSize-4)/4*33054211828000289)
c1 = uintptr((8-goarch.PtrSize)/4*3267000013 + (goarch.PtrSize-4)/4*23344194077549503)
)
func memhash0(p unsafe.Pointer, h uintptr) uintptr {
return h
}
func memhash8(p unsafe.Pointer, h uintptr) uintptr {
return memhash(p, h, 1)
}
func memhash16(p unsafe.Pointer, h uintptr) uintptr {
return memhash(p, h, 2)
}
func memhash128(p unsafe.Pointer, h uintptr) uintptr {
return memhash(p, h, 16)
}
//go:nosplit
func memhash_varlen(p unsafe.Pointer, h uintptr) uintptr {
ptr := sys.GetClosurePtr()
size := *(*uintptr)(unsafe.Pointer(ptr + unsafe.Sizeof(h)))
return memhash(p, h, size)
}
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
// runtime variable to check if the processor we're running on
// actually supports the instructions used by the AES-based
// hash implementation.
var useAeshash bool
// in asm_*.s
// memhash should be an internal detail,
// but widely used packages access it using linkname.
// Notable members of the hall of shame include:
// - github.com/aacfactory/fns
// - github.com/dgraph-io/ristretto
// - github.com/minio/simdjson-go
// - github.com/nbd-wtf/go-nostr
// - github.com/outcaste-io/ristretto
// - github.com/puzpuzpuz/xsync/v2
// - github.com/puzpuzpuz/xsync/v3
// - github.com/authzed/spicedb
// - github.com/pingcap/badger
//
// Do not remove or change the type signature.
// See go.dev/issue/67401.
//
//go:linkname memhash
func memhash(p unsafe.Pointer, h, s uintptr) uintptr
func memhash32(p unsafe.Pointer, h uintptr) uintptr
func memhash64(p unsafe.Pointer, h uintptr) uintptr
// strhash should be an internal detail,
// but widely used packages access it using linkname.
// Notable members of the hall of shame include:
// - github.com/aristanetworks/goarista
// - github.com/bytedance/sonic
// - github.com/bytedance/go-tagexpr/v2
// - github.com/cloudwego/dynamicgo
// - github.com/v2fly/v2ray-core/v5
//
// Do not remove or change the type signature.
// See go.dev/issue/67401.
//
//go:linkname strhash
func strhash(p unsafe.Pointer, h uintptr) uintptr
func strhashFallback(a unsafe.Pointer, h uintptr) uintptr {
x := (*stringStruct)(a)
return memhashFallback(x.str, h, uintptr(x.len))
}
// NOTE: Because NaN != NaN, a map can contain any
// number of (mostly useless) entries keyed with NaNs.
// To avoid long hash chains, we assign a random number
// as the hash value for a NaN.
func f32hash(p unsafe.Pointer, h uintptr) uintptr {
f := *(*float32)(p)
switch {
case f == 0:
return c1 * (c0 ^ h) // +0, -0
case f != f:
math/rand, math/rand/v2: use ChaCha8 for global rand Move ChaCha8 code into internal/chacha8rand and use it to implement runtime.rand, which is used for the unseeded global source for both math/rand and math/rand/v2. This also affects the calculation of the start point for iteration over very very large maps (when the 32-bit fastrand is not big enough). The benefit is that misuse of the global random number generators in math/rand and math/rand/v2 in contexts where non-predictable randomness is important for security reasons is no longer a security problem, removing a common mistake among programmers who are unaware of the different kinds of randomness. The cost is an extra 304 bytes per thread stored in the m struct plus 2-3ns more per random uint64 due to the more sophisticated algorithm. Using PCG looks like it would cost about the same, although I haven't benchmarked that. Before this, the math/rand and math/rand/v2 global generator was wyrand (https://github.com/wangyi-fudan/wyhash). For math/rand, using wyrand instead of the Mitchell/Reeds/Thompson ALFG was justifiable, since the latter was not any better. But for math/rand/v2, the global generator really should be at least as good as one of the well-studied, specific algorithms provided directly by the package, and it's not. (Wyrand is still reasonable for scheduling and cache decisions.) Good randomness does have a cost: about twice wyrand. Also rationalize the various runtime rand references. goos: linux goarch: amd64 pkg: math/rand/v2 cpu: AMD Ryzen 9 7950X 16-Core Processor │ bbb48afeb7.amd64 │ 5cf807d1ea.amd64 │ │ sec/op │ sec/op vs base │ ChaCha8-32 1.862n ± 2% 1.861n ± 2% ~ (p=0.825 n=20) PCG_DXSM-32 1.471n ± 1% 1.460n ± 2% ~ (p=0.153 n=20) SourceUint64-32 1.636n ± 2% 1.582n ± 1% -3.30% (p=0.000 n=20) GlobalInt64-32 2.087n ± 1% 3.663n ± 1% +75.54% (p=0.000 n=20) GlobalInt64Parallel-32 0.1042n ± 1% 0.2026n ± 1% +94.48% (p=0.000 n=20) GlobalUint64-32 2.263n ± 2% 3.724n ± 1% +64.57% (p=0.000 n=20) GlobalUint64Parallel-32 0.1019n ± 1% 0.1973n ± 1% +93.67% (p=0.000 n=20) Int64-32 1.771n ± 1% 1.774n ± 1% ~ (p=0.449 n=20) Uint64-32 1.863n ± 2% 1.866n ± 1% ~ (p=0.364 n=20) GlobalIntN1000-32 3.134n ± 3% 4.730n ± 2% +50.95% (p=0.000 n=20) IntN1000-32 2.489n ± 1% 2.489n ± 1% ~ (p=0.683 n=20) Int64N1000-32 2.521n ± 1% 2.516n ± 1% ~ (p=0.394 n=20) Int64N1e8-32 2.479n ± 1% 2.478n ± 2% ~ (p=0.743 n=20) Int64N1e9-32 2.530n ± 2% 2.514n ± 2% ~ (p=0.193 n=20) Int64N2e9-32 2.501n ± 1% 2.494n ± 1% ~ (p=0.616 n=20) Int64N1e18-32 3.227n ± 1% 3.205n ± 1% ~ (p=0.101 n=20) Int64N2e18-32 3.647n ± 1% 3.599n ± 1% ~ (p=0.019 n=20) Int64N4e18-32 5.135n ± 1% 5.069n ± 2% ~ (p=0.034 n=20) Int32N1000-32 2.657n ± 1% 2.637n ± 1% ~ (p=0.180 n=20) Int32N1e8-32 2.636n ± 1% 2.636n ± 1% ~ (p=0.763 n=20) Int32N1e9-32 2.660n ± 2% 2.638n ± 1% ~ (p=0.358 n=20) Int32N2e9-32 2.662n ± 2% 2.618n ± 2% ~ (p=0.064 n=20) Float32-32 2.272n ± 2% 2.239n ± 2% ~ (p=0.194 n=20) Float64-32 2.272n ± 1% 2.286n ± 2% ~ (p=0.763 n=20) ExpFloat64-32 3.762n ± 1% 3.744n ± 1% ~ (p=0.171 n=20) NormFloat64-32 3.706n ± 1% 3.655n ± 2% ~ (p=0.066 n=20) Perm3-32 32.93n ± 3% 34.62n ± 1% +5.13% (p=0.000 n=20) Perm30-32 202.9n ± 1% 204.0n ± 1% ~ (p=0.482 n=20) Perm30ViaShuffle-32 115.0n ± 1% 114.9n ± 1% ~ (p=0.358 n=20) ShuffleOverhead-32 112.8n ± 1% 112.7n ± 1% ~ (p=0.692 n=20) Concurrent-32 2.107n ± 0% 3.725n ± 1% +76.75% (p=0.000 n=20) goos: darwin goarch: arm64 pkg: math/rand/v2 │ bbb48afeb7.arm64 │ 5cf807d1ea.arm64 │ │ sec/op │ sec/op vs base │ ChaCha8-8 2.480n ± 0% 2.429n ± 0% -2.04% (p=0.000 n=20) PCG_DXSM-8 2.531n ± 0% 2.530n ± 0% ~ (p=0.877 n=20) SourceUint64-8 2.534n ± 0% 2.533n ± 0% ~ (p=0.732 n=20) GlobalInt64-8 2.172n ± 1% 4.794n ± 0% +120.67% (p=0.000 n=20) GlobalInt64Parallel-8 0.4320n ± 0% 0.9605n ± 0% +122.32% (p=0.000 n=20) GlobalUint64-8 2.182n ± 0% 4.770n ± 0% +118.58% (p=0.000 n=20) GlobalUint64Parallel-8 0.4307n ± 0% 0.9583n ± 0% +122.51% (p=0.000 n=20) Int64-8 4.107n ± 0% 4.104n ± 0% ~ (p=0.416 n=20) Uint64-8 4.080n ± 0% 4.080n ± 0% ~ (p=0.052 n=20) GlobalIntN1000-8 2.814n ± 2% 5.643n ± 0% +100.50% (p=0.000 n=20) IntN1000-8 4.141n ± 0% 4.139n ± 0% ~ (p=0.140 n=20) Int64N1000-8 4.140n ± 0% 4.140n ± 0% ~ (p=0.313 n=20) Int64N1e8-8 4.140n ± 0% 4.139n ± 0% ~ (p=0.103 n=20) Int64N1e9-8 4.139n ± 0% 4.140n ± 0% ~ (p=0.761 n=20) Int64N2e9-8 4.140n ± 0% 4.140n ± 0% ~ (p=0.636 n=20) Int64N1e18-8 5.266n ± 0% 5.326n ± 1% +1.14% (p=0.001 n=20) Int64N2e18-8 6.052n ± 0% 6.167n ± 0% +1.90% (p=0.000 n=20) Int64N4e18-8 8.826n ± 0% 9.051n ± 0% +2.55% (p=0.000 n=20) Int32N1000-8 4.127n ± 0% 4.132n ± 0% +0.12% (p=0.000 n=20) Int32N1e8-8 4.126n ± 0% 4.131n ± 0% +0.12% (p=0.000 n=20) Int32N1e9-8 4.127n ± 0% 4.132n ± 0% +0.12% (p=0.000 n=20) Int32N2e9-8 4.132n ± 0% 4.131n ± 0% ~ (p=0.017 n=20) Float32-8 4.109n ± 0% 4.105n ± 0% ~ (p=0.379 n=20) Float64-8 4.107n ± 0% 4.106n ± 0% ~ (p=0.867 n=20) ExpFloat64-8 5.339n ± 0% 5.383n ± 0% +0.82% (p=0.000 n=20) NormFloat64-8 5.735n ± 0% 5.737n ± 1% ~ (p=0.856 n=20) Perm3-8 26.65n ± 0% 26.80n ± 1% +0.58% (p=0.000 n=20) Perm30-8 194.8n ± 1% 197.0n ± 0% +1.18% (p=0.000 n=20) Perm30ViaShuffle-8 156.6n ± 0% 157.6n ± 1% +0.61% (p=0.000 n=20) ShuffleOverhead-8 124.9n ± 0% 125.5n ± 0% +0.52% (p=0.000 n=20) Concurrent-8 2.434n ± 3% 5.066n ± 0% +108.09% (p=0.000 n=20) goos: linux goarch: 386 pkg: math/rand/v2 cpu: AMD Ryzen 9 7950X 16-Core Processor │ bbb48afeb7.386 │ 5cf807d1ea.386 │ │ sec/op │ sec/op vs base │ ChaCha8-32 11.295n ± 1% 4.748n ± 2% -57.96% (p=0.000 n=20) PCG_DXSM-32 7.693n ± 1% 7.738n ± 2% ~ (p=0.542 n=20) SourceUint64-32 7.658n ± 2% 7.622n ± 2% ~ (p=0.344 n=20) GlobalInt64-32 3.473n ± 2% 7.526n ± 2% +116.73% (p=0.000 n=20) GlobalInt64Parallel-32 0.3198n ± 0% 0.5444n ± 0% +70.22% (p=0.000 n=20) GlobalUint64-32 3.612n ± 0% 7.575n ± 1% +109.69% (p=0.000 n=20) GlobalUint64Parallel-32 0.3168n ± 0% 0.5403n ± 0% +70.51% (p=0.000 n=20) Int64-32 7.673n ± 2% 7.789n ± 1% ~ (p=0.122 n=20) Uint64-32 7.773n ± 1% 7.827n ± 2% ~ (p=0.920 n=20) GlobalIntN1000-32 6.268n ± 1% 9.581n ± 1% +52.87% (p=0.000 n=20) IntN1000-32 10.33n ± 2% 10.45n ± 1% ~ (p=0.233 n=20) Int64N1000-32 10.98n ± 2% 11.01n ± 1% ~ (p=0.401 n=20) Int64N1e8-32 11.19n ± 2% 10.97n ± 1% ~ (p=0.033 n=20) Int64N1e9-32 11.06n ± 1% 11.08n ± 1% ~ (p=0.498 n=20) Int64N2e9-32 11.10n ± 1% 11.01n ± 2% ~ (p=0.995 n=20) Int64N1e18-32 15.23n ± 2% 15.04n ± 1% ~ (p=0.973 n=20) Int64N2e18-32 15.89n ± 1% 15.85n ± 1% ~ (p=0.409 n=20) Int64N4e18-32 18.96n ± 2% 19.34n ± 2% ~ (p=0.048 n=20) Int32N1000-32 10.46n ± 2% 10.44n ± 2% ~ (p=0.480 n=20) Int32N1e8-32 10.46n ± 2% 10.49n ± 2% ~ (p=0.951 n=20) Int32N1e9-32 10.28n ± 2% 10.26n ± 1% ~ (p=0.431 n=20) Int32N2e9-32 10.50n ± 2% 10.44n ± 2% ~ (p=0.249 n=20) Float32-32 13.80n ± 2% 13.80n ± 2% ~ (p=0.751 n=20) Float64-32 23.55n ± 2% 23.87n ± 0% ~ (p=0.408 n=20) ExpFloat64-32 15.36n ± 1% 15.29n ± 2% ~ (p=0.316 n=20) NormFloat64-32 13.57n ± 1% 13.79n ± 1% +1.66% (p=0.005 n=20) Perm3-32 45.70n ± 2% 46.99n ± 2% +2.81% (p=0.001 n=20) Perm30-32 399.0n ± 1% 403.8n ± 1% +1.19% (p=0.006 n=20) Perm30ViaShuffle-32 349.0n ± 1% 350.4n ± 1% ~ (p=0.909 n=20) ShuffleOverhead-32 322.3n ± 1% 323.8n ± 1% ~ (p=0.410 n=20) Concurrent-32 3.331n ± 1% 7.312n ± 1% +119.50% (p=0.000 n=20) For #61716. Change-Id: Ibdddeed85c34d9ae397289dc899e04d4845f9ed2 Reviewed-on: https://go-review.googlesource.com/c/go/+/516860 Reviewed-by: Michael Pratt <mpratt@google.com> Reviewed-by: Filippo Valsorda <filippo@golang.org> LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
2023-08-06 13:26:28 +10:00
return c1 * (c0 ^ h ^ uintptr(rand())) // any kind of NaN
default:
return memhash(p, h, 4)
}
}
func f64hash(p unsafe.Pointer, h uintptr) uintptr {
f := *(*float64)(p)
switch {
case f == 0:
return c1 * (c0 ^ h) // +0, -0
case f != f:
math/rand, math/rand/v2: use ChaCha8 for global rand Move ChaCha8 code into internal/chacha8rand and use it to implement runtime.rand, which is used for the unseeded global source for both math/rand and math/rand/v2. This also affects the calculation of the start point for iteration over very very large maps (when the 32-bit fastrand is not big enough). The benefit is that misuse of the global random number generators in math/rand and math/rand/v2 in contexts where non-predictable randomness is important for security reasons is no longer a security problem, removing a common mistake among programmers who are unaware of the different kinds of randomness. The cost is an extra 304 bytes per thread stored in the m struct plus 2-3ns more per random uint64 due to the more sophisticated algorithm. Using PCG looks like it would cost about the same, although I haven't benchmarked that. Before this, the math/rand and math/rand/v2 global generator was wyrand (https://github.com/wangyi-fudan/wyhash). For math/rand, using wyrand instead of the Mitchell/Reeds/Thompson ALFG was justifiable, since the latter was not any better. But for math/rand/v2, the global generator really should be at least as good as one of the well-studied, specific algorithms provided directly by the package, and it's not. (Wyrand is still reasonable for scheduling and cache decisions.) Good randomness does have a cost: about twice wyrand. Also rationalize the various runtime rand references. goos: linux goarch: amd64 pkg: math/rand/v2 cpu: AMD Ryzen 9 7950X 16-Core Processor │ bbb48afeb7.amd64 │ 5cf807d1ea.amd64 │ │ sec/op │ sec/op vs base │ ChaCha8-32 1.862n ± 2% 1.861n ± 2% ~ (p=0.825 n=20) PCG_DXSM-32 1.471n ± 1% 1.460n ± 2% ~ (p=0.153 n=20) SourceUint64-32 1.636n ± 2% 1.582n ± 1% -3.30% (p=0.000 n=20) GlobalInt64-32 2.087n ± 1% 3.663n ± 1% +75.54% (p=0.000 n=20) GlobalInt64Parallel-32 0.1042n ± 1% 0.2026n ± 1% +94.48% (p=0.000 n=20) GlobalUint64-32 2.263n ± 2% 3.724n ± 1% +64.57% (p=0.000 n=20) GlobalUint64Parallel-32 0.1019n ± 1% 0.1973n ± 1% +93.67% (p=0.000 n=20) Int64-32 1.771n ± 1% 1.774n ± 1% ~ (p=0.449 n=20) Uint64-32 1.863n ± 2% 1.866n ± 1% ~ (p=0.364 n=20) GlobalIntN1000-32 3.134n ± 3% 4.730n ± 2% +50.95% (p=0.000 n=20) IntN1000-32 2.489n ± 1% 2.489n ± 1% ~ (p=0.683 n=20) Int64N1000-32 2.521n ± 1% 2.516n ± 1% ~ (p=0.394 n=20) Int64N1e8-32 2.479n ± 1% 2.478n ± 2% ~ (p=0.743 n=20) Int64N1e9-32 2.530n ± 2% 2.514n ± 2% ~ (p=0.193 n=20) Int64N2e9-32 2.501n ± 1% 2.494n ± 1% ~ (p=0.616 n=20) Int64N1e18-32 3.227n ± 1% 3.205n ± 1% ~ (p=0.101 n=20) Int64N2e18-32 3.647n ± 1% 3.599n ± 1% ~ (p=0.019 n=20) Int64N4e18-32 5.135n ± 1% 5.069n ± 2% ~ (p=0.034 n=20) Int32N1000-32 2.657n ± 1% 2.637n ± 1% ~ (p=0.180 n=20) Int32N1e8-32 2.636n ± 1% 2.636n ± 1% ~ (p=0.763 n=20) Int32N1e9-32 2.660n ± 2% 2.638n ± 1% ~ (p=0.358 n=20) Int32N2e9-32 2.662n ± 2% 2.618n ± 2% ~ (p=0.064 n=20) Float32-32 2.272n ± 2% 2.239n ± 2% ~ (p=0.194 n=20) Float64-32 2.272n ± 1% 2.286n ± 2% ~ (p=0.763 n=20) ExpFloat64-32 3.762n ± 1% 3.744n ± 1% ~ (p=0.171 n=20) NormFloat64-32 3.706n ± 1% 3.655n ± 2% ~ (p=0.066 n=20) Perm3-32 32.93n ± 3% 34.62n ± 1% +5.13% (p=0.000 n=20) Perm30-32 202.9n ± 1% 204.0n ± 1% ~ (p=0.482 n=20) Perm30ViaShuffle-32 115.0n ± 1% 114.9n ± 1% ~ (p=0.358 n=20) ShuffleOverhead-32 112.8n ± 1% 112.7n ± 1% ~ (p=0.692 n=20) Concurrent-32 2.107n ± 0% 3.725n ± 1% +76.75% (p=0.000 n=20) goos: darwin goarch: arm64 pkg: math/rand/v2 │ bbb48afeb7.arm64 │ 5cf807d1ea.arm64 │ │ sec/op │ sec/op vs base │ ChaCha8-8 2.480n ± 0% 2.429n ± 0% -2.04% (p=0.000 n=20) PCG_DXSM-8 2.531n ± 0% 2.530n ± 0% ~ (p=0.877 n=20) SourceUint64-8 2.534n ± 0% 2.533n ± 0% ~ (p=0.732 n=20) GlobalInt64-8 2.172n ± 1% 4.794n ± 0% +120.67% (p=0.000 n=20) GlobalInt64Parallel-8 0.4320n ± 0% 0.9605n ± 0% +122.32% (p=0.000 n=20) GlobalUint64-8 2.182n ± 0% 4.770n ± 0% +118.58% (p=0.000 n=20) GlobalUint64Parallel-8 0.4307n ± 0% 0.9583n ± 0% +122.51% (p=0.000 n=20) Int64-8 4.107n ± 0% 4.104n ± 0% ~ (p=0.416 n=20) Uint64-8 4.080n ± 0% 4.080n ± 0% ~ (p=0.052 n=20) GlobalIntN1000-8 2.814n ± 2% 5.643n ± 0% +100.50% (p=0.000 n=20) IntN1000-8 4.141n ± 0% 4.139n ± 0% ~ (p=0.140 n=20) Int64N1000-8 4.140n ± 0% 4.140n ± 0% ~ (p=0.313 n=20) Int64N1e8-8 4.140n ± 0% 4.139n ± 0% ~ (p=0.103 n=20) Int64N1e9-8 4.139n ± 0% 4.140n ± 0% ~ (p=0.761 n=20) Int64N2e9-8 4.140n ± 0% 4.140n ± 0% ~ (p=0.636 n=20) Int64N1e18-8 5.266n ± 0% 5.326n ± 1% +1.14% (p=0.001 n=20) Int64N2e18-8 6.052n ± 0% 6.167n ± 0% +1.90% (p=0.000 n=20) Int64N4e18-8 8.826n ± 0% 9.051n ± 0% +2.55% (p=0.000 n=20) Int32N1000-8 4.127n ± 0% 4.132n ± 0% +0.12% (p=0.000 n=20) Int32N1e8-8 4.126n ± 0% 4.131n ± 0% +0.12% (p=0.000 n=20) Int32N1e9-8 4.127n ± 0% 4.132n ± 0% +0.12% (p=0.000 n=20) Int32N2e9-8 4.132n ± 0% 4.131n ± 0% ~ (p=0.017 n=20) Float32-8 4.109n ± 0% 4.105n ± 0% ~ (p=0.379 n=20) Float64-8 4.107n ± 0% 4.106n ± 0% ~ (p=0.867 n=20) ExpFloat64-8 5.339n ± 0% 5.383n ± 0% +0.82% (p=0.000 n=20) NormFloat64-8 5.735n ± 0% 5.737n ± 1% ~ (p=0.856 n=20) Perm3-8 26.65n ± 0% 26.80n ± 1% +0.58% (p=0.000 n=20) Perm30-8 194.8n ± 1% 197.0n ± 0% +1.18% (p=0.000 n=20) Perm30ViaShuffle-8 156.6n ± 0% 157.6n ± 1% +0.61% (p=0.000 n=20) ShuffleOverhead-8 124.9n ± 0% 125.5n ± 0% +0.52% (p=0.000 n=20) Concurrent-8 2.434n ± 3% 5.066n ± 0% +108.09% (p=0.000 n=20) goos: linux goarch: 386 pkg: math/rand/v2 cpu: AMD Ryzen 9 7950X 16-Core Processor │ bbb48afeb7.386 │ 5cf807d1ea.386 │ │ sec/op │ sec/op vs base │ ChaCha8-32 11.295n ± 1% 4.748n ± 2% -57.96% (p=0.000 n=20) PCG_DXSM-32 7.693n ± 1% 7.738n ± 2% ~ (p=0.542 n=20) SourceUint64-32 7.658n ± 2% 7.622n ± 2% ~ (p=0.344 n=20) GlobalInt64-32 3.473n ± 2% 7.526n ± 2% +116.73% (p=0.000 n=20) GlobalInt64Parallel-32 0.3198n ± 0% 0.5444n ± 0% +70.22% (p=0.000 n=20) GlobalUint64-32 3.612n ± 0% 7.575n ± 1% +109.69% (p=0.000 n=20) GlobalUint64Parallel-32 0.3168n ± 0% 0.5403n ± 0% +70.51% (p=0.000 n=20) Int64-32 7.673n ± 2% 7.789n ± 1% ~ (p=0.122 n=20) Uint64-32 7.773n ± 1% 7.827n ± 2% ~ (p=0.920 n=20) GlobalIntN1000-32 6.268n ± 1% 9.581n ± 1% +52.87% (p=0.000 n=20) IntN1000-32 10.33n ± 2% 10.45n ± 1% ~ (p=0.233 n=20) Int64N1000-32 10.98n ± 2% 11.01n ± 1% ~ (p=0.401 n=20) Int64N1e8-32 11.19n ± 2% 10.97n ± 1% ~ (p=0.033 n=20) Int64N1e9-32 11.06n ± 1% 11.08n ± 1% ~ (p=0.498 n=20) Int64N2e9-32 11.10n ± 1% 11.01n ± 2% ~ (p=0.995 n=20) Int64N1e18-32 15.23n ± 2% 15.04n ± 1% ~ (p=0.973 n=20) Int64N2e18-32 15.89n ± 1% 15.85n ± 1% ~ (p=0.409 n=20) Int64N4e18-32 18.96n ± 2% 19.34n ± 2% ~ (p=0.048 n=20) Int32N1000-32 10.46n ± 2% 10.44n ± 2% ~ (p=0.480 n=20) Int32N1e8-32 10.46n ± 2% 10.49n ± 2% ~ (p=0.951 n=20) Int32N1e9-32 10.28n ± 2% 10.26n ± 1% ~ (p=0.431 n=20) Int32N2e9-32 10.50n ± 2% 10.44n ± 2% ~ (p=0.249 n=20) Float32-32 13.80n ± 2% 13.80n ± 2% ~ (p=0.751 n=20) Float64-32 23.55n ± 2% 23.87n ± 0% ~ (p=0.408 n=20) ExpFloat64-32 15.36n ± 1% 15.29n ± 2% ~ (p=0.316 n=20) NormFloat64-32 13.57n ± 1% 13.79n ± 1% +1.66% (p=0.005 n=20) Perm3-32 45.70n ± 2% 46.99n ± 2% +2.81% (p=0.001 n=20) Perm30-32 399.0n ± 1% 403.8n ± 1% +1.19% (p=0.006 n=20) Perm30ViaShuffle-32 349.0n ± 1% 350.4n ± 1% ~ (p=0.909 n=20) ShuffleOverhead-32 322.3n ± 1% 323.8n ± 1% ~ (p=0.410 n=20) Concurrent-32 3.331n ± 1% 7.312n ± 1% +119.50% (p=0.000 n=20) For #61716. Change-Id: Ibdddeed85c34d9ae397289dc899e04d4845f9ed2 Reviewed-on: https://go-review.googlesource.com/c/go/+/516860 Reviewed-by: Michael Pratt <mpratt@google.com> Reviewed-by: Filippo Valsorda <filippo@golang.org> LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
2023-08-06 13:26:28 +10:00
return c1 * (c0 ^ h ^ uintptr(rand())) // any kind of NaN
default:
return memhash(p, h, 8)
}
}
func c64hash(p unsafe.Pointer, h uintptr) uintptr {
x := (*[2]float32)(p)
return f32hash(unsafe.Pointer(&x[1]), f32hash(unsafe.Pointer(&x[0]), h))
}
func c128hash(p unsafe.Pointer, h uintptr) uintptr {
x := (*[2]float64)(p)
return f64hash(unsafe.Pointer(&x[1]), f64hash(unsafe.Pointer(&x[0]), h))
}
func interhash(p unsafe.Pointer, h uintptr) uintptr {
a := (*iface)(p)
tab := a.tab
if tab == nil {
return h
}
t := tab.Type
if t.Equal == nil {
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
// Check hashability here. We could do this check inside
// typehash, but we want to report the topmost type in
// the error text (e.g. in a struct with a field of slice type
// we want to report the struct, not the slice).
panic(errorString("hash of unhashable type " + toRType(t).string()))
}
if isDirectIface(t) {
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return c1 * typehash(t, unsafe.Pointer(&a.data), h^c0)
} else {
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return c1 * typehash(t, a.data, h^c0)
}
}
// nilinterhash should be an internal detail,
// but widely used packages access it using linkname.
// Notable members of the hall of shame include:
// - github.com/anacrolix/stm
// - github.com/aristanetworks/goarista
//
// Do not remove or change the type signature.
// See go.dev/issue/67401.
//
//go:linkname nilinterhash
func nilinterhash(p unsafe.Pointer, h uintptr) uintptr {
a := (*eface)(p)
t := a._type
if t == nil {
return h
}
if t.Equal == nil {
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
// See comment in interhash above.
panic(errorString("hash of unhashable type " + toRType(t).string()))
}
if isDirectIface(t) {
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return c1 * typehash(t, unsafe.Pointer(&a.data), h^c0)
} else {
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return c1 * typehash(t, a.data, h^c0)
}
}
// typehash computes the hash of the object of type t at address p.
// h is the seed.
// This function is seldom used. Most maps use for hashing either
// fixed functions (e.g. f32hash) or compiler-generated functions
// (e.g. for a type like struct { x, y string }). This implementation
// is slower but more general and is used for hashing interface types
// (called from interhash or nilinterhash, above) or for hashing in
// maps generated by reflect.MapOf (reflect_typehash, below).
// Note: this function must match the compiler generated
// functions exactly. See issue 37716.
//
// typehash should be an internal detail,
// but widely used packages access it using linkname.
// Notable members of the hall of shame include:
// - github.com/puzpuzpuz/xsync/v2
// - github.com/puzpuzpuz/xsync/v3
//
// Do not remove or change the type signature.
// See go.dev/issue/67401.
//
//go:linkname typehash
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
func typehash(t *_type, p unsafe.Pointer, h uintptr) uintptr {
if t.TFlag&abi.TFlagRegularMemory != 0 {
// Handle ptr sizes specially, see issue 37086.
switch t.Size_ {
case 4:
return memhash32(p, h)
case 8:
return memhash64(p, h)
default:
return memhash(p, h, t.Size_)
}
}
switch t.Kind_ & abi.KindMask {
case abi.Float32:
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return f32hash(p, h)
case abi.Float64:
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return f64hash(p, h)
case abi.Complex64:
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return c64hash(p, h)
case abi.Complex128:
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return c128hash(p, h)
case abi.String:
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return strhash(p, h)
case abi.Interface:
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
i := (*interfacetype)(unsafe.Pointer(t))
if len(i.Methods) == 0 {
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return nilinterhash(p, h)
}
return interhash(p, h)
case abi.Array:
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
a := (*arraytype)(unsafe.Pointer(t))
for i := uintptr(0); i < a.Len; i++ {
h = typehash(a.Elem, add(p, i*a.Elem.Size_), h)
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
}
return h
case abi.Struct:
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
s := (*structtype)(unsafe.Pointer(t))
for _, f := range s.Fields {
if f.Name.IsBlank() {
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
continue
}
h = typehash(f.Typ, add(p, f.Offset), h)
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
}
return h
default:
// Should never happen, as typehash should only be called
// with comparable types.
panic(errorString("hash of unhashable type " + toRType(t).string()))
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
}
}
func mapKeyError(t *maptype, p unsafe.Pointer) error {
if !t.HashMightPanic() {
return nil
}
return mapKeyError2(t.Key, p)
}
func mapKeyError2(t *_type, p unsafe.Pointer) error {
if t.TFlag&abi.TFlagRegularMemory != 0 {
return nil
}
switch t.Kind_ & abi.KindMask {
case abi.Float32, abi.Float64, abi.Complex64, abi.Complex128, abi.String:
return nil
case abi.Interface:
i := (*interfacetype)(unsafe.Pointer(t))
var t *_type
var pdata *unsafe.Pointer
if len(i.Methods) == 0 {
a := (*eface)(p)
t = a._type
if t == nil {
return nil
}
pdata = &a.data
} else {
a := (*iface)(p)
if a.tab == nil {
return nil
}
t = a.tab.Type
pdata = &a.data
}
if t.Equal == nil {
return errorString("hash of unhashable type " + toRType(t).string())
}
if isDirectIface(t) {
return mapKeyError2(t, unsafe.Pointer(pdata))
} else {
return mapKeyError2(t, *pdata)
}
case abi.Array:
a := (*arraytype)(unsafe.Pointer(t))
for i := uintptr(0); i < a.Len; i++ {
if err := mapKeyError2(a.Elem, add(p, i*a.Elem.Size_)); err != nil {
return err
}
}
return nil
case abi.Struct:
s := (*structtype)(unsafe.Pointer(t))
for _, f := range s.Fields {
if f.Name.IsBlank() {
continue
}
if err := mapKeyError2(f.Typ, add(p, f.Offset)); err != nil {
return err
}
}
return nil
default:
// Should never happen, keep this case for robustness.
return errorString("hash of unhashable type " + toRType(t).string())
}
}
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
//go:linkname reflect_typehash reflect.typehash
func reflect_typehash(t *_type, p unsafe.Pointer, h uintptr) uintptr {
return typehash(t, p, h)
}
func memequal0(p, q unsafe.Pointer) bool {
return true
}
func memequal8(p, q unsafe.Pointer) bool {
return *(*int8)(p) == *(*int8)(q)
}
func memequal16(p, q unsafe.Pointer) bool {
return *(*int16)(p) == *(*int16)(q)
}
func memequal32(p, q unsafe.Pointer) bool {
return *(*int32)(p) == *(*int32)(q)
}
func memequal64(p, q unsafe.Pointer) bool {
return *(*int64)(p) == *(*int64)(q)
}
func memequal128(p, q unsafe.Pointer) bool {
return *(*[2]int64)(p) == *(*[2]int64)(q)
}
func f32equal(p, q unsafe.Pointer) bool {
return *(*float32)(p) == *(*float32)(q)
}
func f64equal(p, q unsafe.Pointer) bool {
return *(*float64)(p) == *(*float64)(q)
}
func c64equal(p, q unsafe.Pointer) bool {
return *(*complex64)(p) == *(*complex64)(q)
}
func c128equal(p, q unsafe.Pointer) bool {
return *(*complex128)(p) == *(*complex128)(q)
}
func strequal(p, q unsafe.Pointer) bool {
return *(*string)(p) == *(*string)(q)
}
func interequal(p, q unsafe.Pointer) bool {
x := *(*iface)(p)
y := *(*iface)(q)
return x.tab == y.tab && ifaceeq(x.tab, x.data, y.data)
}
func nilinterequal(p, q unsafe.Pointer) bool {
x := *(*eface)(p)
y := *(*eface)(q)
return x._type == y._type && efaceeq(x._type, x.data, y.data)
}
func efaceeq(t *_type, x, y unsafe.Pointer) bool {
if t == nil {
return true
}
eq := t.Equal
if eq == nil {
panic(errorString("comparing uncomparable type " + toRType(t).string()))
}
if isDirectIface(t) {
// Direct interface types are ptr, chan, map, func, and single-element structs/arrays thereof.
// Maps and funcs are not comparable, so they can't reach here.
// Ptrs, chans, and single-element items can be compared directly using ==.
return x == y
}
return eq(x, y)
}
func ifaceeq(tab *itab, x, y unsafe.Pointer) bool {
if tab == nil {
return true
}
t := tab.Type
eq := t.Equal
if eq == nil {
panic(errorString("comparing uncomparable type " + toRType(t).string()))
}
if isDirectIface(t) {
// See comment in efaceeq.
return x == y
}
return eq(x, y)
}
// Testing adapters for hash quality tests (see hash_test.go)
//
// stringHash should be an internal detail,
// but widely used packages access it using linkname.
// Notable members of the hall of shame include:
// - github.com/k14s/starlark-go
//
// Do not remove or change the type signature.
// See go.dev/issue/67401.
//
//go:linkname stringHash
func stringHash(s string, seed uintptr) uintptr {
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return strhash(noescape(unsafe.Pointer(&s)), seed)
}
func bytesHash(b []byte, seed uintptr) uintptr {
s := (*slice)(unsafe.Pointer(&b))
return memhash(s.array, seed, uintptr(s.len))
}
func int32Hash(i uint32, seed uintptr) uintptr {
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return memhash32(noescape(unsafe.Pointer(&i)), seed)
}
func int64Hash(i uint64, seed uintptr) uintptr {
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return memhash64(noescape(unsafe.Pointer(&i)), seed)
}
func efaceHash(i any, seed uintptr) uintptr {
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return nilinterhash(noescape(unsafe.Pointer(&i)), seed)
}
func ifaceHash(i interface {
F()
}, seed uintptr) uintptr {
cmd/compile,runtime: generate hash functions only for types which are map keys Right now we generate hash functions for all types, just in case they are used as map keys. That's a lot of wasted effort and binary size for types which will never be used as a map key. Instead, generate hash functions only for types that we know are map keys. Just doing that is a bit too simple, since maps with an interface type as a key might have to hash any concrete key type that implements that interface. So for that case, implement hashing of such types at runtime (instead of with generated code). It will be slower, but only for maps with interface types as keys, and maybe only a bit slower as the aeshash time probably dominates the dispatch time. Reorg where we keep the equals and hash functions. Move the hash function from the key type to the map type, saving a field in every non-map type. That leaves only one function in the alg structure, so get rid of that and just keep the equal function in the type descriptor itself. cmd/go now has 10 generated hash functions, instead of 504. Makes cmd/go 1.0% smaller. Update #6853. Speed on non-interface keys is unchanged. Speed on interface keys is ~20% slower: name old time/op new time/op delta MapInterfaceString-8 23.0ns ±21% 27.6ns ±14% +20.01% (p=0.002 n=10+10) MapInterfacePtr-8 19.4ns ±16% 23.7ns ± 7% +22.48% (p=0.000 n=10+8) Change-Id: I7c2e42292a46b5d4e288aaec4029bdbb01089263 Reviewed-on: https://go-review.googlesource.com/c/go/+/191198 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Martin Möhrmann <moehrmann@google.com>
2019-08-06 15:22:51 -07:00
return interhash(noescape(unsafe.Pointer(&i)), seed)
}
const hashRandomBytes = goarch.PtrSize / 4 * 64
// used in asm_{386,amd64,arm64}.s to seed the hash function
var aeskeysched [hashRandomBytes]byte
// used in hash{32,64}.go to seed the hash function
var hashkey [4]uintptr
func alginit() {
// Install AES hash algorithms if the instructions needed are present.
if (GOARCH == "386" || GOARCH == "amd64") &&
cpu.X86.HasAES && // AESENC
cpu.X86.HasSSSE3 && // PSHUFB
cpu.X86.HasSSE41 { // PINSR{D,Q}
initAlgAES()
return
}
if GOARCH == "arm64" && cpu.ARM64.HasAES {
initAlgAES()
return
}
math/rand, math/rand/v2: use ChaCha8 for global rand Move ChaCha8 code into internal/chacha8rand and use it to implement runtime.rand, which is used for the unseeded global source for both math/rand and math/rand/v2. This also affects the calculation of the start point for iteration over very very large maps (when the 32-bit fastrand is not big enough). The benefit is that misuse of the global random number generators in math/rand and math/rand/v2 in contexts where non-predictable randomness is important for security reasons is no longer a security problem, removing a common mistake among programmers who are unaware of the different kinds of randomness. The cost is an extra 304 bytes per thread stored in the m struct plus 2-3ns more per random uint64 due to the more sophisticated algorithm. Using PCG looks like it would cost about the same, although I haven't benchmarked that. Before this, the math/rand and math/rand/v2 global generator was wyrand (https://github.com/wangyi-fudan/wyhash). For math/rand, using wyrand instead of the Mitchell/Reeds/Thompson ALFG was justifiable, since the latter was not any better. But for math/rand/v2, the global generator really should be at least as good as one of the well-studied, specific algorithms provided directly by the package, and it's not. (Wyrand is still reasonable for scheduling and cache decisions.) Good randomness does have a cost: about twice wyrand. Also rationalize the various runtime rand references. goos: linux goarch: amd64 pkg: math/rand/v2 cpu: AMD Ryzen 9 7950X 16-Core Processor │ bbb48afeb7.amd64 │ 5cf807d1ea.amd64 │ │ sec/op │ sec/op vs base │ ChaCha8-32 1.862n ± 2% 1.861n ± 2% ~ (p=0.825 n=20) PCG_DXSM-32 1.471n ± 1% 1.460n ± 2% ~ (p=0.153 n=20) SourceUint64-32 1.636n ± 2% 1.582n ± 1% -3.30% (p=0.000 n=20) GlobalInt64-32 2.087n ± 1% 3.663n ± 1% +75.54% (p=0.000 n=20) GlobalInt64Parallel-32 0.1042n ± 1% 0.2026n ± 1% +94.48% (p=0.000 n=20) GlobalUint64-32 2.263n ± 2% 3.724n ± 1% +64.57% (p=0.000 n=20) GlobalUint64Parallel-32 0.1019n ± 1% 0.1973n ± 1% +93.67% (p=0.000 n=20) Int64-32 1.771n ± 1% 1.774n ± 1% ~ (p=0.449 n=20) Uint64-32 1.863n ± 2% 1.866n ± 1% ~ (p=0.364 n=20) GlobalIntN1000-32 3.134n ± 3% 4.730n ± 2% +50.95% (p=0.000 n=20) IntN1000-32 2.489n ± 1% 2.489n ± 1% ~ (p=0.683 n=20) Int64N1000-32 2.521n ± 1% 2.516n ± 1% ~ (p=0.394 n=20) Int64N1e8-32 2.479n ± 1% 2.478n ± 2% ~ (p=0.743 n=20) Int64N1e9-32 2.530n ± 2% 2.514n ± 2% ~ (p=0.193 n=20) Int64N2e9-32 2.501n ± 1% 2.494n ± 1% ~ (p=0.616 n=20) Int64N1e18-32 3.227n ± 1% 3.205n ± 1% ~ (p=0.101 n=20) Int64N2e18-32 3.647n ± 1% 3.599n ± 1% ~ (p=0.019 n=20) Int64N4e18-32 5.135n ± 1% 5.069n ± 2% ~ (p=0.034 n=20) Int32N1000-32 2.657n ± 1% 2.637n ± 1% ~ (p=0.180 n=20) Int32N1e8-32 2.636n ± 1% 2.636n ± 1% ~ (p=0.763 n=20) Int32N1e9-32 2.660n ± 2% 2.638n ± 1% ~ (p=0.358 n=20) Int32N2e9-32 2.662n ± 2% 2.618n ± 2% ~ (p=0.064 n=20) Float32-32 2.272n ± 2% 2.239n ± 2% ~ (p=0.194 n=20) Float64-32 2.272n ± 1% 2.286n ± 2% ~ (p=0.763 n=20) ExpFloat64-32 3.762n ± 1% 3.744n ± 1% ~ (p=0.171 n=20) NormFloat64-32 3.706n ± 1% 3.655n ± 2% ~ (p=0.066 n=20) Perm3-32 32.93n ± 3% 34.62n ± 1% +5.13% (p=0.000 n=20) Perm30-32 202.9n ± 1% 204.0n ± 1% ~ (p=0.482 n=20) Perm30ViaShuffle-32 115.0n ± 1% 114.9n ± 1% ~ (p=0.358 n=20) ShuffleOverhead-32 112.8n ± 1% 112.7n ± 1% ~ (p=0.692 n=20) Concurrent-32 2.107n ± 0% 3.725n ± 1% +76.75% (p=0.000 n=20) goos: darwin goarch: arm64 pkg: math/rand/v2 │ bbb48afeb7.arm64 │ 5cf807d1ea.arm64 │ │ sec/op │ sec/op vs base │ ChaCha8-8 2.480n ± 0% 2.429n ± 0% -2.04% (p=0.000 n=20) PCG_DXSM-8 2.531n ± 0% 2.530n ± 0% ~ (p=0.877 n=20) SourceUint64-8 2.534n ± 0% 2.533n ± 0% ~ (p=0.732 n=20) GlobalInt64-8 2.172n ± 1% 4.794n ± 0% +120.67% (p=0.000 n=20) GlobalInt64Parallel-8 0.4320n ± 0% 0.9605n ± 0% +122.32% (p=0.000 n=20) GlobalUint64-8 2.182n ± 0% 4.770n ± 0% +118.58% (p=0.000 n=20) GlobalUint64Parallel-8 0.4307n ± 0% 0.9583n ± 0% +122.51% (p=0.000 n=20) Int64-8 4.107n ± 0% 4.104n ± 0% ~ (p=0.416 n=20) Uint64-8 4.080n ± 0% 4.080n ± 0% ~ (p=0.052 n=20) GlobalIntN1000-8 2.814n ± 2% 5.643n ± 0% +100.50% (p=0.000 n=20) IntN1000-8 4.141n ± 0% 4.139n ± 0% ~ (p=0.140 n=20) Int64N1000-8 4.140n ± 0% 4.140n ± 0% ~ (p=0.313 n=20) Int64N1e8-8 4.140n ± 0% 4.139n ± 0% ~ (p=0.103 n=20) Int64N1e9-8 4.139n ± 0% 4.140n ± 0% ~ (p=0.761 n=20) Int64N2e9-8 4.140n ± 0% 4.140n ± 0% ~ (p=0.636 n=20) Int64N1e18-8 5.266n ± 0% 5.326n ± 1% +1.14% (p=0.001 n=20) Int64N2e18-8 6.052n ± 0% 6.167n ± 0% +1.90% (p=0.000 n=20) Int64N4e18-8 8.826n ± 0% 9.051n ± 0% +2.55% (p=0.000 n=20) Int32N1000-8 4.127n ± 0% 4.132n ± 0% +0.12% (p=0.000 n=20) Int32N1e8-8 4.126n ± 0% 4.131n ± 0% +0.12% (p=0.000 n=20) Int32N1e9-8 4.127n ± 0% 4.132n ± 0% +0.12% (p=0.000 n=20) Int32N2e9-8 4.132n ± 0% 4.131n ± 0% ~ (p=0.017 n=20) Float32-8 4.109n ± 0% 4.105n ± 0% ~ (p=0.379 n=20) Float64-8 4.107n ± 0% 4.106n ± 0% ~ (p=0.867 n=20) ExpFloat64-8 5.339n ± 0% 5.383n ± 0% +0.82% (p=0.000 n=20) NormFloat64-8 5.735n ± 0% 5.737n ± 1% ~ (p=0.856 n=20) Perm3-8 26.65n ± 0% 26.80n ± 1% +0.58% (p=0.000 n=20) Perm30-8 194.8n ± 1% 197.0n ± 0% +1.18% (p=0.000 n=20) Perm30ViaShuffle-8 156.6n ± 0% 157.6n ± 1% +0.61% (p=0.000 n=20) ShuffleOverhead-8 124.9n ± 0% 125.5n ± 0% +0.52% (p=0.000 n=20) Concurrent-8 2.434n ± 3% 5.066n ± 0% +108.09% (p=0.000 n=20) goos: linux goarch: 386 pkg: math/rand/v2 cpu: AMD Ryzen 9 7950X 16-Core Processor │ bbb48afeb7.386 │ 5cf807d1ea.386 │ │ sec/op │ sec/op vs base │ ChaCha8-32 11.295n ± 1% 4.748n ± 2% -57.96% (p=0.000 n=20) PCG_DXSM-32 7.693n ± 1% 7.738n ± 2% ~ (p=0.542 n=20) SourceUint64-32 7.658n ± 2% 7.622n ± 2% ~ (p=0.344 n=20) GlobalInt64-32 3.473n ± 2% 7.526n ± 2% +116.73% (p=0.000 n=20) GlobalInt64Parallel-32 0.3198n ± 0% 0.5444n ± 0% +70.22% (p=0.000 n=20) GlobalUint64-32 3.612n ± 0% 7.575n ± 1% +109.69% (p=0.000 n=20) GlobalUint64Parallel-32 0.3168n ± 0% 0.5403n ± 0% +70.51% (p=0.000 n=20) Int64-32 7.673n ± 2% 7.789n ± 1% ~ (p=0.122 n=20) Uint64-32 7.773n ± 1% 7.827n ± 2% ~ (p=0.920 n=20) GlobalIntN1000-32 6.268n ± 1% 9.581n ± 1% +52.87% (p=0.000 n=20) IntN1000-32 10.33n ± 2% 10.45n ± 1% ~ (p=0.233 n=20) Int64N1000-32 10.98n ± 2% 11.01n ± 1% ~ (p=0.401 n=20) Int64N1e8-32 11.19n ± 2% 10.97n ± 1% ~ (p=0.033 n=20) Int64N1e9-32 11.06n ± 1% 11.08n ± 1% ~ (p=0.498 n=20) Int64N2e9-32 11.10n ± 1% 11.01n ± 2% ~ (p=0.995 n=20) Int64N1e18-32 15.23n ± 2% 15.04n ± 1% ~ (p=0.973 n=20) Int64N2e18-32 15.89n ± 1% 15.85n ± 1% ~ (p=0.409 n=20) Int64N4e18-32 18.96n ± 2% 19.34n ± 2% ~ (p=0.048 n=20) Int32N1000-32 10.46n ± 2% 10.44n ± 2% ~ (p=0.480 n=20) Int32N1e8-32 10.46n ± 2% 10.49n ± 2% ~ (p=0.951 n=20) Int32N1e9-32 10.28n ± 2% 10.26n ± 1% ~ (p=0.431 n=20) Int32N2e9-32 10.50n ± 2% 10.44n ± 2% ~ (p=0.249 n=20) Float32-32 13.80n ± 2% 13.80n ± 2% ~ (p=0.751 n=20) Float64-32 23.55n ± 2% 23.87n ± 0% ~ (p=0.408 n=20) ExpFloat64-32 15.36n ± 1% 15.29n ± 2% ~ (p=0.316 n=20) NormFloat64-32 13.57n ± 1% 13.79n ± 1% +1.66% (p=0.005 n=20) Perm3-32 45.70n ± 2% 46.99n ± 2% +2.81% (p=0.001 n=20) Perm30-32 399.0n ± 1% 403.8n ± 1% +1.19% (p=0.006 n=20) Perm30ViaShuffle-32 349.0n ± 1% 350.4n ± 1% ~ (p=0.909 n=20) ShuffleOverhead-32 322.3n ± 1% 323.8n ± 1% ~ (p=0.410 n=20) Concurrent-32 3.331n ± 1% 7.312n ± 1% +119.50% (p=0.000 n=20) For #61716. Change-Id: Ibdddeed85c34d9ae397289dc899e04d4845f9ed2 Reviewed-on: https://go-review.googlesource.com/c/go/+/516860 Reviewed-by: Michael Pratt <mpratt@google.com> Reviewed-by: Filippo Valsorda <filippo@golang.org> LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
2023-08-06 13:26:28 +10:00
for i := range hashkey {
hashkey[i] = uintptr(bootstrapRand())
math/rand, math/rand/v2: use ChaCha8 for global rand Move ChaCha8 code into internal/chacha8rand and use it to implement runtime.rand, which is used for the unseeded global source for both math/rand and math/rand/v2. This also affects the calculation of the start point for iteration over very very large maps (when the 32-bit fastrand is not big enough). The benefit is that misuse of the global random number generators in math/rand and math/rand/v2 in contexts where non-predictable randomness is important for security reasons is no longer a security problem, removing a common mistake among programmers who are unaware of the different kinds of randomness. The cost is an extra 304 bytes per thread stored in the m struct plus 2-3ns more per random uint64 due to the more sophisticated algorithm. Using PCG looks like it would cost about the same, although I haven't benchmarked that. Before this, the math/rand and math/rand/v2 global generator was wyrand (https://github.com/wangyi-fudan/wyhash). For math/rand, using wyrand instead of the Mitchell/Reeds/Thompson ALFG was justifiable, since the latter was not any better. But for math/rand/v2, the global generator really should be at least as good as one of the well-studied, specific algorithms provided directly by the package, and it's not. (Wyrand is still reasonable for scheduling and cache decisions.) Good randomness does have a cost: about twice wyrand. Also rationalize the various runtime rand references. goos: linux goarch: amd64 pkg: math/rand/v2 cpu: AMD Ryzen 9 7950X 16-Core Processor │ bbb48afeb7.amd64 │ 5cf807d1ea.amd64 │ │ sec/op │ sec/op vs base │ ChaCha8-32 1.862n ± 2% 1.861n ± 2% ~ (p=0.825 n=20) PCG_DXSM-32 1.471n ± 1% 1.460n ± 2% ~ (p=0.153 n=20) SourceUint64-32 1.636n ± 2% 1.582n ± 1% -3.30% (p=0.000 n=20) GlobalInt64-32 2.087n ± 1% 3.663n ± 1% +75.54% (p=0.000 n=20) GlobalInt64Parallel-32 0.1042n ± 1% 0.2026n ± 1% +94.48% (p=0.000 n=20) GlobalUint64-32 2.263n ± 2% 3.724n ± 1% +64.57% (p=0.000 n=20) GlobalUint64Parallel-32 0.1019n ± 1% 0.1973n ± 1% +93.67% (p=0.000 n=20) Int64-32 1.771n ± 1% 1.774n ± 1% ~ (p=0.449 n=20) Uint64-32 1.863n ± 2% 1.866n ± 1% ~ (p=0.364 n=20) GlobalIntN1000-32 3.134n ± 3% 4.730n ± 2% +50.95% (p=0.000 n=20) IntN1000-32 2.489n ± 1% 2.489n ± 1% ~ (p=0.683 n=20) Int64N1000-32 2.521n ± 1% 2.516n ± 1% ~ (p=0.394 n=20) Int64N1e8-32 2.479n ± 1% 2.478n ± 2% ~ (p=0.743 n=20) Int64N1e9-32 2.530n ± 2% 2.514n ± 2% ~ (p=0.193 n=20) Int64N2e9-32 2.501n ± 1% 2.494n ± 1% ~ (p=0.616 n=20) Int64N1e18-32 3.227n ± 1% 3.205n ± 1% ~ (p=0.101 n=20) Int64N2e18-32 3.647n ± 1% 3.599n ± 1% ~ (p=0.019 n=20) Int64N4e18-32 5.135n ± 1% 5.069n ± 2% ~ (p=0.034 n=20) Int32N1000-32 2.657n ± 1% 2.637n ± 1% ~ (p=0.180 n=20) Int32N1e8-32 2.636n ± 1% 2.636n ± 1% ~ (p=0.763 n=20) Int32N1e9-32 2.660n ± 2% 2.638n ± 1% ~ (p=0.358 n=20) Int32N2e9-32 2.662n ± 2% 2.618n ± 2% ~ (p=0.064 n=20) Float32-32 2.272n ± 2% 2.239n ± 2% ~ (p=0.194 n=20) Float64-32 2.272n ± 1% 2.286n ± 2% ~ (p=0.763 n=20) ExpFloat64-32 3.762n ± 1% 3.744n ± 1% ~ (p=0.171 n=20) NormFloat64-32 3.706n ± 1% 3.655n ± 2% ~ (p=0.066 n=20) Perm3-32 32.93n ± 3% 34.62n ± 1% +5.13% (p=0.000 n=20) Perm30-32 202.9n ± 1% 204.0n ± 1% ~ (p=0.482 n=20) Perm30ViaShuffle-32 115.0n ± 1% 114.9n ± 1% ~ (p=0.358 n=20) ShuffleOverhead-32 112.8n ± 1% 112.7n ± 1% ~ (p=0.692 n=20) Concurrent-32 2.107n ± 0% 3.725n ± 1% +76.75% (p=0.000 n=20) goos: darwin goarch: arm64 pkg: math/rand/v2 │ bbb48afeb7.arm64 │ 5cf807d1ea.arm64 │ │ sec/op │ sec/op vs base │ ChaCha8-8 2.480n ± 0% 2.429n ± 0% -2.04% (p=0.000 n=20) PCG_DXSM-8 2.531n ± 0% 2.530n ± 0% ~ (p=0.877 n=20) SourceUint64-8 2.534n ± 0% 2.533n ± 0% ~ (p=0.732 n=20) GlobalInt64-8 2.172n ± 1% 4.794n ± 0% +120.67% (p=0.000 n=20) GlobalInt64Parallel-8 0.4320n ± 0% 0.9605n ± 0% +122.32% (p=0.000 n=20) GlobalUint64-8 2.182n ± 0% 4.770n ± 0% +118.58% (p=0.000 n=20) GlobalUint64Parallel-8 0.4307n ± 0% 0.9583n ± 0% +122.51% (p=0.000 n=20) Int64-8 4.107n ± 0% 4.104n ± 0% ~ (p=0.416 n=20) Uint64-8 4.080n ± 0% 4.080n ± 0% ~ (p=0.052 n=20) GlobalIntN1000-8 2.814n ± 2% 5.643n ± 0% +100.50% (p=0.000 n=20) IntN1000-8 4.141n ± 0% 4.139n ± 0% ~ (p=0.140 n=20) Int64N1000-8 4.140n ± 0% 4.140n ± 0% ~ (p=0.313 n=20) Int64N1e8-8 4.140n ± 0% 4.139n ± 0% ~ (p=0.103 n=20) Int64N1e9-8 4.139n ± 0% 4.140n ± 0% ~ (p=0.761 n=20) Int64N2e9-8 4.140n ± 0% 4.140n ± 0% ~ (p=0.636 n=20) Int64N1e18-8 5.266n ± 0% 5.326n ± 1% +1.14% (p=0.001 n=20) Int64N2e18-8 6.052n ± 0% 6.167n ± 0% +1.90% (p=0.000 n=20) Int64N4e18-8 8.826n ± 0% 9.051n ± 0% +2.55% (p=0.000 n=20) Int32N1000-8 4.127n ± 0% 4.132n ± 0% +0.12% (p=0.000 n=20) Int32N1e8-8 4.126n ± 0% 4.131n ± 0% +0.12% (p=0.000 n=20) Int32N1e9-8 4.127n ± 0% 4.132n ± 0% +0.12% (p=0.000 n=20) Int32N2e9-8 4.132n ± 0% 4.131n ± 0% ~ (p=0.017 n=20) Float32-8 4.109n ± 0% 4.105n ± 0% ~ (p=0.379 n=20) Float64-8 4.107n ± 0% 4.106n ± 0% ~ (p=0.867 n=20) ExpFloat64-8 5.339n ± 0% 5.383n ± 0% +0.82% (p=0.000 n=20) NormFloat64-8 5.735n ± 0% 5.737n ± 1% ~ (p=0.856 n=20) Perm3-8 26.65n ± 0% 26.80n ± 1% +0.58% (p=0.000 n=20) Perm30-8 194.8n ± 1% 197.0n ± 0% +1.18% (p=0.000 n=20) Perm30ViaShuffle-8 156.6n ± 0% 157.6n ± 1% +0.61% (p=0.000 n=20) ShuffleOverhead-8 124.9n ± 0% 125.5n ± 0% +0.52% (p=0.000 n=20) Concurrent-8 2.434n ± 3% 5.066n ± 0% +108.09% (p=0.000 n=20) goos: linux goarch: 386 pkg: math/rand/v2 cpu: AMD Ryzen 9 7950X 16-Core Processor │ bbb48afeb7.386 │ 5cf807d1ea.386 │ │ sec/op │ sec/op vs base │ ChaCha8-32 11.295n ± 1% 4.748n ± 2% -57.96% (p=0.000 n=20) PCG_DXSM-32 7.693n ± 1% 7.738n ± 2% ~ (p=0.542 n=20) SourceUint64-32 7.658n ± 2% 7.622n ± 2% ~ (p=0.344 n=20) GlobalInt64-32 3.473n ± 2% 7.526n ± 2% +116.73% (p=0.000 n=20) GlobalInt64Parallel-32 0.3198n ± 0% 0.5444n ± 0% +70.22% (p=0.000 n=20) GlobalUint64-32 3.612n ± 0% 7.575n ± 1% +109.69% (p=0.000 n=20) GlobalUint64Parallel-32 0.3168n ± 0% 0.5403n ± 0% +70.51% (p=0.000 n=20) Int64-32 7.673n ± 2% 7.789n ± 1% ~ (p=0.122 n=20) Uint64-32 7.773n ± 1% 7.827n ± 2% ~ (p=0.920 n=20) GlobalIntN1000-32 6.268n ± 1% 9.581n ± 1% +52.87% (p=0.000 n=20) IntN1000-32 10.33n ± 2% 10.45n ± 1% ~ (p=0.233 n=20) Int64N1000-32 10.98n ± 2% 11.01n ± 1% ~ (p=0.401 n=20) Int64N1e8-32 11.19n ± 2% 10.97n ± 1% ~ (p=0.033 n=20) Int64N1e9-32 11.06n ± 1% 11.08n ± 1% ~ (p=0.498 n=20) Int64N2e9-32 11.10n ± 1% 11.01n ± 2% ~ (p=0.995 n=20) Int64N1e18-32 15.23n ± 2% 15.04n ± 1% ~ (p=0.973 n=20) Int64N2e18-32 15.89n ± 1% 15.85n ± 1% ~ (p=0.409 n=20) Int64N4e18-32 18.96n ± 2% 19.34n ± 2% ~ (p=0.048 n=20) Int32N1000-32 10.46n ± 2% 10.44n ± 2% ~ (p=0.480 n=20) Int32N1e8-32 10.46n ± 2% 10.49n ± 2% ~ (p=0.951 n=20) Int32N1e9-32 10.28n ± 2% 10.26n ± 1% ~ (p=0.431 n=20) Int32N2e9-32 10.50n ± 2% 10.44n ± 2% ~ (p=0.249 n=20) Float32-32 13.80n ± 2% 13.80n ± 2% ~ (p=0.751 n=20) Float64-32 23.55n ± 2% 23.87n ± 0% ~ (p=0.408 n=20) ExpFloat64-32 15.36n ± 1% 15.29n ± 2% ~ (p=0.316 n=20) NormFloat64-32 13.57n ± 1% 13.79n ± 1% +1.66% (p=0.005 n=20) Perm3-32 45.70n ± 2% 46.99n ± 2% +2.81% (p=0.001 n=20) Perm30-32 399.0n ± 1% 403.8n ± 1% +1.19% (p=0.006 n=20) Perm30ViaShuffle-32 349.0n ± 1% 350.4n ± 1% ~ (p=0.909 n=20) ShuffleOverhead-32 322.3n ± 1% 323.8n ± 1% ~ (p=0.410 n=20) Concurrent-32 3.331n ± 1% 7.312n ± 1% +119.50% (p=0.000 n=20) For #61716. Change-Id: Ibdddeed85c34d9ae397289dc899e04d4845f9ed2 Reviewed-on: https://go-review.googlesource.com/c/go/+/516860 Reviewed-by: Michael Pratt <mpratt@google.com> Reviewed-by: Filippo Valsorda <filippo@golang.org> LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
2023-08-06 13:26:28 +10:00
}
}
func initAlgAES() {
useAeshash = true
// Initialize with random data so hash collisions will be hard to engineer.
math/rand, math/rand/v2: use ChaCha8 for global rand Move ChaCha8 code into internal/chacha8rand and use it to implement runtime.rand, which is used for the unseeded global source for both math/rand and math/rand/v2. This also affects the calculation of the start point for iteration over very very large maps (when the 32-bit fastrand is not big enough). The benefit is that misuse of the global random number generators in math/rand and math/rand/v2 in contexts where non-predictable randomness is important for security reasons is no longer a security problem, removing a common mistake among programmers who are unaware of the different kinds of randomness. The cost is an extra 304 bytes per thread stored in the m struct plus 2-3ns more per random uint64 due to the more sophisticated algorithm. Using PCG looks like it would cost about the same, although I haven't benchmarked that. Before this, the math/rand and math/rand/v2 global generator was wyrand (https://github.com/wangyi-fudan/wyhash). For math/rand, using wyrand instead of the Mitchell/Reeds/Thompson ALFG was justifiable, since the latter was not any better. But for math/rand/v2, the global generator really should be at least as good as one of the well-studied, specific algorithms provided directly by the package, and it's not. (Wyrand is still reasonable for scheduling and cache decisions.) Good randomness does have a cost: about twice wyrand. Also rationalize the various runtime rand references. goos: linux goarch: amd64 pkg: math/rand/v2 cpu: AMD Ryzen 9 7950X 16-Core Processor │ bbb48afeb7.amd64 │ 5cf807d1ea.amd64 │ │ sec/op │ sec/op vs base │ ChaCha8-32 1.862n ± 2% 1.861n ± 2% ~ (p=0.825 n=20) PCG_DXSM-32 1.471n ± 1% 1.460n ± 2% ~ (p=0.153 n=20) SourceUint64-32 1.636n ± 2% 1.582n ± 1% -3.30% (p=0.000 n=20) GlobalInt64-32 2.087n ± 1% 3.663n ± 1% +75.54% (p=0.000 n=20) GlobalInt64Parallel-32 0.1042n ± 1% 0.2026n ± 1% +94.48% (p=0.000 n=20) GlobalUint64-32 2.263n ± 2% 3.724n ± 1% +64.57% (p=0.000 n=20) GlobalUint64Parallel-32 0.1019n ± 1% 0.1973n ± 1% +93.67% (p=0.000 n=20) Int64-32 1.771n ± 1% 1.774n ± 1% ~ (p=0.449 n=20) Uint64-32 1.863n ± 2% 1.866n ± 1% ~ (p=0.364 n=20) GlobalIntN1000-32 3.134n ± 3% 4.730n ± 2% +50.95% (p=0.000 n=20) IntN1000-32 2.489n ± 1% 2.489n ± 1% ~ (p=0.683 n=20) Int64N1000-32 2.521n ± 1% 2.516n ± 1% ~ (p=0.394 n=20) Int64N1e8-32 2.479n ± 1% 2.478n ± 2% ~ (p=0.743 n=20) Int64N1e9-32 2.530n ± 2% 2.514n ± 2% ~ (p=0.193 n=20) Int64N2e9-32 2.501n ± 1% 2.494n ± 1% ~ (p=0.616 n=20) Int64N1e18-32 3.227n ± 1% 3.205n ± 1% ~ (p=0.101 n=20) Int64N2e18-32 3.647n ± 1% 3.599n ± 1% ~ (p=0.019 n=20) Int64N4e18-32 5.135n ± 1% 5.069n ± 2% ~ (p=0.034 n=20) Int32N1000-32 2.657n ± 1% 2.637n ± 1% ~ (p=0.180 n=20) Int32N1e8-32 2.636n ± 1% 2.636n ± 1% ~ (p=0.763 n=20) Int32N1e9-32 2.660n ± 2% 2.638n ± 1% ~ (p=0.358 n=20) Int32N2e9-32 2.662n ± 2% 2.618n ± 2% ~ (p=0.064 n=20) Float32-32 2.272n ± 2% 2.239n ± 2% ~ (p=0.194 n=20) Float64-32 2.272n ± 1% 2.286n ± 2% ~ (p=0.763 n=20) ExpFloat64-32 3.762n ± 1% 3.744n ± 1% ~ (p=0.171 n=20) NormFloat64-32 3.706n ± 1% 3.655n ± 2% ~ (p=0.066 n=20) Perm3-32 32.93n ± 3% 34.62n ± 1% +5.13% (p=0.000 n=20) Perm30-32 202.9n ± 1% 204.0n ± 1% ~ (p=0.482 n=20) Perm30ViaShuffle-32 115.0n ± 1% 114.9n ± 1% ~ (p=0.358 n=20) ShuffleOverhead-32 112.8n ± 1% 112.7n ± 1% ~ (p=0.692 n=20) Concurrent-32 2.107n ± 0% 3.725n ± 1% +76.75% (p=0.000 n=20) goos: darwin goarch: arm64 pkg: math/rand/v2 │ bbb48afeb7.arm64 │ 5cf807d1ea.arm64 │ │ sec/op │ sec/op vs base │ ChaCha8-8 2.480n ± 0% 2.429n ± 0% -2.04% (p=0.000 n=20) PCG_DXSM-8 2.531n ± 0% 2.530n ± 0% ~ (p=0.877 n=20) SourceUint64-8 2.534n ± 0% 2.533n ± 0% ~ (p=0.732 n=20) GlobalInt64-8 2.172n ± 1% 4.794n ± 0% +120.67% (p=0.000 n=20) GlobalInt64Parallel-8 0.4320n ± 0% 0.9605n ± 0% +122.32% (p=0.000 n=20) GlobalUint64-8 2.182n ± 0% 4.770n ± 0% +118.58% (p=0.000 n=20) GlobalUint64Parallel-8 0.4307n ± 0% 0.9583n ± 0% +122.51% (p=0.000 n=20) Int64-8 4.107n ± 0% 4.104n ± 0% ~ (p=0.416 n=20) Uint64-8 4.080n ± 0% 4.080n ± 0% ~ (p=0.052 n=20) GlobalIntN1000-8 2.814n ± 2% 5.643n ± 0% +100.50% (p=0.000 n=20) IntN1000-8 4.141n ± 0% 4.139n ± 0% ~ (p=0.140 n=20) Int64N1000-8 4.140n ± 0% 4.140n ± 0% ~ (p=0.313 n=20) Int64N1e8-8 4.140n ± 0% 4.139n ± 0% ~ (p=0.103 n=20) Int64N1e9-8 4.139n ± 0% 4.140n ± 0% ~ (p=0.761 n=20) Int64N2e9-8 4.140n ± 0% 4.140n ± 0% ~ (p=0.636 n=20) Int64N1e18-8 5.266n ± 0% 5.326n ± 1% +1.14% (p=0.001 n=20) Int64N2e18-8 6.052n ± 0% 6.167n ± 0% +1.90% (p=0.000 n=20) Int64N4e18-8 8.826n ± 0% 9.051n ± 0% +2.55% (p=0.000 n=20) Int32N1000-8 4.127n ± 0% 4.132n ± 0% +0.12% (p=0.000 n=20) Int32N1e8-8 4.126n ± 0% 4.131n ± 0% +0.12% (p=0.000 n=20) Int32N1e9-8 4.127n ± 0% 4.132n ± 0% +0.12% (p=0.000 n=20) Int32N2e9-8 4.132n ± 0% 4.131n ± 0% ~ (p=0.017 n=20) Float32-8 4.109n ± 0% 4.105n ± 0% ~ (p=0.379 n=20) Float64-8 4.107n ± 0% 4.106n ± 0% ~ (p=0.867 n=20) ExpFloat64-8 5.339n ± 0% 5.383n ± 0% +0.82% (p=0.000 n=20) NormFloat64-8 5.735n ± 0% 5.737n ± 1% ~ (p=0.856 n=20) Perm3-8 26.65n ± 0% 26.80n ± 1% +0.58% (p=0.000 n=20) Perm30-8 194.8n ± 1% 197.0n ± 0% +1.18% (p=0.000 n=20) Perm30ViaShuffle-8 156.6n ± 0% 157.6n ± 1% +0.61% (p=0.000 n=20) ShuffleOverhead-8 124.9n ± 0% 125.5n ± 0% +0.52% (p=0.000 n=20) Concurrent-8 2.434n ± 3% 5.066n ± 0% +108.09% (p=0.000 n=20) goos: linux goarch: 386 pkg: math/rand/v2 cpu: AMD Ryzen 9 7950X 16-Core Processor │ bbb48afeb7.386 │ 5cf807d1ea.386 │ │ sec/op │ sec/op vs base │ ChaCha8-32 11.295n ± 1% 4.748n ± 2% -57.96% (p=0.000 n=20) PCG_DXSM-32 7.693n ± 1% 7.738n ± 2% ~ (p=0.542 n=20) SourceUint64-32 7.658n ± 2% 7.622n ± 2% ~ (p=0.344 n=20) GlobalInt64-32 3.473n ± 2% 7.526n ± 2% +116.73% (p=0.000 n=20) GlobalInt64Parallel-32 0.3198n ± 0% 0.5444n ± 0% +70.22% (p=0.000 n=20) GlobalUint64-32 3.612n ± 0% 7.575n ± 1% +109.69% (p=0.000 n=20) GlobalUint64Parallel-32 0.3168n ± 0% 0.5403n ± 0% +70.51% (p=0.000 n=20) Int64-32 7.673n ± 2% 7.789n ± 1% ~ (p=0.122 n=20) Uint64-32 7.773n ± 1% 7.827n ± 2% ~ (p=0.920 n=20) GlobalIntN1000-32 6.268n ± 1% 9.581n ± 1% +52.87% (p=0.000 n=20) IntN1000-32 10.33n ± 2% 10.45n ± 1% ~ (p=0.233 n=20) Int64N1000-32 10.98n ± 2% 11.01n ± 1% ~ (p=0.401 n=20) Int64N1e8-32 11.19n ± 2% 10.97n ± 1% ~ (p=0.033 n=20) Int64N1e9-32 11.06n ± 1% 11.08n ± 1% ~ (p=0.498 n=20) Int64N2e9-32 11.10n ± 1% 11.01n ± 2% ~ (p=0.995 n=20) Int64N1e18-32 15.23n ± 2% 15.04n ± 1% ~ (p=0.973 n=20) Int64N2e18-32 15.89n ± 1% 15.85n ± 1% ~ (p=0.409 n=20) Int64N4e18-32 18.96n ± 2% 19.34n ± 2% ~ (p=0.048 n=20) Int32N1000-32 10.46n ± 2% 10.44n ± 2% ~ (p=0.480 n=20) Int32N1e8-32 10.46n ± 2% 10.49n ± 2% ~ (p=0.951 n=20) Int32N1e9-32 10.28n ± 2% 10.26n ± 1% ~ (p=0.431 n=20) Int32N2e9-32 10.50n ± 2% 10.44n ± 2% ~ (p=0.249 n=20) Float32-32 13.80n ± 2% 13.80n ± 2% ~ (p=0.751 n=20) Float64-32 23.55n ± 2% 23.87n ± 0% ~ (p=0.408 n=20) ExpFloat64-32 15.36n ± 1% 15.29n ± 2% ~ (p=0.316 n=20) NormFloat64-32 13.57n ± 1% 13.79n ± 1% +1.66% (p=0.005 n=20) Perm3-32 45.70n ± 2% 46.99n ± 2% +2.81% (p=0.001 n=20) Perm30-32 399.0n ± 1% 403.8n ± 1% +1.19% (p=0.006 n=20) Perm30ViaShuffle-32 349.0n ± 1% 350.4n ± 1% ~ (p=0.909 n=20) ShuffleOverhead-32 322.3n ± 1% 323.8n ± 1% ~ (p=0.410 n=20) Concurrent-32 3.331n ± 1% 7.312n ± 1% +119.50% (p=0.000 n=20) For #61716. Change-Id: Ibdddeed85c34d9ae397289dc899e04d4845f9ed2 Reviewed-on: https://go-review.googlesource.com/c/go/+/516860 Reviewed-by: Michael Pratt <mpratt@google.com> Reviewed-by: Filippo Valsorda <filippo@golang.org> LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
2023-08-06 13:26:28 +10:00
key := (*[hashRandomBytes / 8]uint64)(unsafe.Pointer(&aeskeysched))
for i := range key {
key[i] = bootstrapRand()
}
}
// Note: These routines perform the read with a native endianness.
func readUnaligned32(p unsafe.Pointer) uint32 {
q := (*[4]byte)(p)
if goarch.BigEndian {
return byteorder.BEUint32(q[:])
}
return byteorder.LEUint32(q[:])
}
func readUnaligned64(p unsafe.Pointer) uint64 {
q := (*[8]byte)(p)
if goarch.BigEndian {
return byteorder.BEUint64(q[:])
}
return byteorder.LEUint64(q[:])
}