perf: bmt SIMD hasher

cmd/bee/cmd/cmd.go

@@ -88,6 +88,7 @@ const (
 	optionAutoTLSDomain                    = "autotls-domain"
 	optionAutoTLSRegistrationEndpoint      = "autotls-registration-endpoint"
 	optionAutoTLSCAEndpoint                = "autotls-ca-endpoint"
+	optionUseSIMD                          = "use-simd-hashing"
 
 	// blockchain-rpc
 	optionNameBlockchainRpcEndpoint    = "blockchain-rpc-endpoint"
@@ -337,6 +338,7 @@ func (c *command) setAllFlags(cmd *cobra.Command) {
 	cmd.Flags().String(optionAutoTLSDomain, p2pforge.DefaultForgeDomain, "autotls domain")
 	cmd.Flags().String(optionAutoTLSRegistrationEndpoint, p2pforge.DefaultForgeEndpoint, "autotls registration endpoint")
 	cmd.Flags().String(optionAutoTLSCAEndpoint, p2pforge.DefaultCAEndpoint, "autotls certificate authority endpoint")
+	cmd.Flags().Bool(optionUseSIMD, false, "use SIMD BMT hasher (available only on linux amd64 platforms)")
 }
 
 // preRun must be called from every command's PreRunE, after which c.logger is

cmd/bee/cmd/start.go

@@ -15,15 +15,19 @@ import (
 	"os"
 	"os/signal"
 	"path/filepath"
+	"runtime"
 	"strings"
 	"sync/atomic"
 	"syscall"
 	"time"
 
 	"github.com/ethersphere/bee/v2"
 	"github.com/ethersphere/bee/v2/pkg/accesscontrol"
+	"github.com/ethersphere/bee/v2/pkg/bmt"
+	"github.com/ethersphere/bee/v2/pkg/bmtpool"
 	chaincfg "github.com/ethersphere/bee/v2/pkg/config"
 	"github.com/ethersphere/bee/v2/pkg/crypto"
+	"github.com/ethersphere/bee/v2/pkg/keccak"
 	"github.com/ethersphere/bee/v2/pkg/keystore"
 	filekeystore "github.com/ethersphere/bee/v2/pkg/keystore/file"
 	memkeystore "github.com/ethersphere/bee/v2/pkg/keystore/mem"
@@ -266,6 +270,21 @@ func buildBeeNode(ctx context.Context, c *command, cmd *cobra.Command, logger lo
 		neighborhoodSuggester = c.config.GetString(optionNameNeighborhoodSuggester)
 	}
 
+	useSIMD := c.config.GetBool(optionUseSIMD)
+	if useSIMD {
+		if runtime.GOOS != "linux" || runtime.GOARCH != "amd64" {
+			return nil, fmt.Errorf("SIMD hashing requires linux/amd64 (this build is %s/%s)", runtime.GOOS, runtime.GOARCH)
+		}
+		if !keccak.HasSIMD() {
+			return nil, errors.New("SIMD hashing requires a CPU with AVX2 or AVX-512; this CPU has neither")
+		}
+		bmt.SetSIMDOptIn(true)
+		// Rebuild the global bmtpool instance so the new SIMDOptIn value
+		// is reflected in the pool created for hot-path BMT hashing.
+		bmtpool.Rebuild()
+		logger.Info("SIMD hashing enabled", "batch_width", keccak.BatchWidth(), "avx512", keccak.HasAVX512())
+	}
+
 	b, err := node.NewBee(ctx, c.config.GetString(optionNameP2PAddr), signerConfig.publicKey, signerConfig.signer, networkID, logger, signerConfig.libp2pPrivateKey, signerConfig.pssPrivateKey, signerConfig.session, &node.Options{
 		Addr:                          c.config.GetString(optionNameP2PAddr),
 		AllowPrivateCIDRs:             c.config.GetBool(optionNameAllowPrivateCIDRs),

go.mod

@@ -103,7 +103,7 @@ require (
 	github.com/ipfs/go-log/v2 v2.6.0 // indirect
 	github.com/jackpal/go-nat-pmp v1.0.2 // indirect
 	github.com/jbenet/go-temp-err-catcher v0.1.0 // indirect
-	github.com/klauspost/cpuid/v2 v2.2.10 // indirect
+	github.com/klauspost/cpuid/v2 v2.3.0
 	github.com/koron/go-ssdp v0.0.6 // indirect
 	github.com/leodido/go-urn v1.4.0 // indirect
 	github.com/libdns/libdns v0.2.2 // indirect

go.sum

@@ -541,8 +541,8 @@ github.com/klauspost/compress v1.18.0/go.mod h1:2Pp+KzxcywXVXMr50+X0Q/Lsb43OQHYW
 github.com/klauspost/cpuid v0.0.0-20170728055534-ae7887de9fa5/go.mod h1:Pj4uuM528wm8OyEC2QMXAi2YiTZ96dNQPGgoMS4s3ek=
 github.com/klauspost/cpuid/v2 v2.0.4/go.mod h1:FInQzS24/EEf25PyTYn52gqo7WaD8xa0213Md/qVLRg=
 github.com/klauspost/cpuid/v2 v2.0.6/go.mod h1:FInQzS24/EEf25PyTYn52gqo7WaD8xa0213Md/qVLRg=
-github.com/klauspost/cpuid/v2 v2.2.10 h1:tBs3QSyvjDyFTq3uoc/9xFpCuOsJQFNPiAhYdw2skhE=
-github.com/klauspost/cpuid/v2 v2.2.10/go.mod h1:hqwkgyIinND0mEev00jJYCxPNVRVXFQeu1XKlok6oO0=
+github.com/klauspost/cpuid/v2 v2.3.0 h1:S4CRMLnYUhGeDFDqkGriYKdfoFlDnMtqTiI/sFzhA9Y=
+github.com/klauspost/cpuid/v2 v2.3.0/go.mod h1:hqwkgyIinND0mEev00jJYCxPNVRVXFQeu1XKlok6oO0=
 github.com/klauspost/crc32 v0.0.0-20161016154125-cb6bfca970f6/go.mod h1:+ZoRqAPRLkC4NPOvfYeR5KNOrY6TD+/sAC3HXPZgDYg=
 github.com/klauspost/pgzip v1.0.2-0.20170402124221-0bf5dcad4ada/go.mod h1:Ch1tH69qFZu15pkjo5kYi6mth2Zzwzt50oCQKQE9RUs=
 github.com/klauspost/reedsolomon v1.11.8 h1:s8RpUW5TK4hjr+djiOpbZJB4ksx+TdYbRH7vHQpwPOY=

pkg/bmt/benchmark_test.go

@@ -26,7 +26,16 @@ func BenchmarkBMT(b *testing.B) {
 		b.Run(fmt.Sprintf("%v_size_%v", "REF", size), func(b *testing.B) {
 			benchmarkRefHasher(b, size)
 		})
-		b.Run(fmt.Sprintf("%v_size_%v", "BMT", size), func(b *testing.B) {
+		b.Run(fmt.Sprintf("%v_size_%v", "BMT_Goroutine", size), func(b *testing.B) {
+			prev := bmt.SIMDOptIn()
+			bmt.SetSIMDOptIn(false)
+			defer bmt.SetSIMDOptIn(prev)
+			benchmarkBMT(b, size)
+		})
+		b.Run(fmt.Sprintf("%v_size_%v", "BMT_SIMD", size), func(b *testing.B) {
+			prev := bmt.SIMDOptIn()
+			bmt.SetSIMDOptIn(true)
+			defer bmt.SetSIMDOptIn(prev)
 			benchmarkBMT(b, size)
 		})
 	}
@@ -81,13 +90,15 @@ func benchmarkBMTBaseline(b *testing.B, _ int) {
 	}
 }
 
-// benchmarks BMT Hasher
+// benchmarks BMT Hasher with whichever pool the current SIMDOptIn selects.
+// Tree is constructed inside the benchmark so the SIMDOptIn value at call time
+// determines which implementation gets exercised.
 func benchmarkBMT(b *testing.B, n int) {
 	b.Helper()
 
 	testData := testutil.RandBytesWithSeed(b, 4096, seed)
 
-	pool := bmt.NewPool(bmt.NewConf(swarm.NewHasher, testSegmentCount, testPoolSize))
+	pool := bmt.NewPool(bmt.NewConf(testSegmentCount, testPoolSize))
 	h := pool.Get()
 	defer pool.Put(h)
 
@@ -106,7 +117,7 @@ func benchmarkPool(b *testing.B, poolsize int) {
 
 	testData := testutil.RandBytesWithSeed(b, 4096, seed)
 
-	pool := bmt.NewPool(bmt.NewConf(swarm.NewHasher, testSegmentCount, poolsize))
+	pool := bmt.NewPool(bmt.NewConf(testSegmentCount, poolsize))
 	cycles := 100
 
 	b.ReportAllocs()