1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397
|
//+build !noasm,!appengine,gc
// Copyright (c) 2020 MinIO Inc. All rights reserved.
// Use of this source code is governed by a license that can be
// found in the LICENSE file.
package md5simd
import (
"encoding/binary"
"fmt"
"runtime"
"sync"
"github.com/klauspost/cpuid/v2"
)
// MD5 initialization constants
const (
// Lanes is the number of concurrently calculated hashes.
Lanes = 16
init0 = 0x67452301
init1 = 0xefcdab89
init2 = 0x98badcfe
init3 = 0x10325476
// Use scalar routine when below this many lanes
useScalarBelow = 3
)
// md5ServerUID - Does not start at 0 but next multiple of 16 so as to be able to
// differentiate with default initialisation value of 0
const md5ServerUID = Lanes
const buffersPerLane = 3
// Message to send across input channel
type blockInput struct {
uid uint64
msg []byte
sumCh chan sumResult
reset bool
}
type sumResult struct {
digest [Size]byte
}
type lanesInfo [Lanes]blockInput
// md5Server - Type to implement parallel handling of MD5 invocations
type md5Server struct {
uidCounter uint64
cycle chan uint64 // client with uid has update.
newInput chan newClient // Add new client.
digests map[uint64][Size]byte // Map of uids to (interim) digest results
maskRounds16 [16]maskRounds // Pre-allocated static array for max 16 rounds
maskRounds8a [8]maskRounds // Pre-allocated static array for max 8 rounds (1st AVX2 core)
maskRounds8b [8]maskRounds // Pre-allocated static array for max 8 rounds (2nd AVX2 core)
allBufs []byte // Preallocated buffer.
buffers chan []byte // Preallocated buffers, sliced from allBufs.
i8 [2][8][]byte // avx2 temporary vars
d8a, d8b digest8
wg sync.WaitGroup
}
// NewServer - Create new object for parallel processing handling
func NewServer() Server {
if !cpuid.CPU.Supports(cpuid.AVX2) {
return &fallbackServer{}
}
md5srv := &md5Server{}
md5srv.digests = make(map[uint64][Size]byte)
md5srv.newInput = make(chan newClient, Lanes)
md5srv.cycle = make(chan uint64, Lanes*10)
md5srv.uidCounter = md5ServerUID - 1
md5srv.allBufs = make([]byte, 32+buffersPerLane*Lanes*internalBlockSize)
md5srv.buffers = make(chan []byte, buffersPerLane*Lanes)
// Fill buffers.
for i := 0; i < buffersPerLane*Lanes; i++ {
s := 32 + i*internalBlockSize
md5srv.buffers <- md5srv.allBufs[s : s+internalBlockSize : s+internalBlockSize]
}
// Start a single thread for reading from the input channel
go md5srv.process(md5srv.newInput)
return md5srv
}
type newClient struct {
uid uint64
input chan blockInput
}
// process - Sole handler for reading from the input channel.
func (s *md5Server) process(newClients chan newClient) {
// To fill up as many lanes as possible:
//
// 1. Wait for a cycle id.
// 2. If not already in a lane, add, otherwise leave on channel
// 3. Start timer
// 4. Check if lanes is full, if so, goto 10 (process).
// 5. If timeout, goto 10.
// 6. Wait for new id (goto 2) or timeout (goto 10).
// 10. Process.
// 11. Check all input if there is already input, if so add to lanes.
// 12. Goto 1
// lanes contains the lanes.
var lanes lanesInfo
// lanesFilled contains the number of filled lanes for current cycle.
var lanesFilled int
// clients contains active clients
var clients = make(map[uint64]chan blockInput, Lanes)
addToLane := func(uid uint64) {
cl, ok := clients[uid]
if !ok {
// Unknown client. Maybe it was already removed.
return
}
// Check if we already have it.
for _, lane := range lanes[:lanesFilled] {
if lane.uid == uid {
return
}
}
// Continue until we get a block or there is nothing on channel
for {
select {
case block, ok := <-cl:
if !ok {
// Client disconnected
delete(clients, block.uid)
return
}
if block.uid != uid {
panic(fmt.Errorf("uid mismatch, %d (block) != %d (client)", block.uid, uid))
}
// If reset message, reset and we're done
if block.reset {
delete(s.digests, uid)
continue
}
// If requesting sum, we will need to maintain state.
if block.sumCh != nil {
var dig digest
d, ok := s.digests[uid]
if ok {
dig.s[0] = binary.LittleEndian.Uint32(d[0:4])
dig.s[1] = binary.LittleEndian.Uint32(d[4:8])
dig.s[2] = binary.LittleEndian.Uint32(d[8:12])
dig.s[3] = binary.LittleEndian.Uint32(d[12:16])
} else {
dig.s[0], dig.s[1], dig.s[2], dig.s[3] = init0, init1, init2, init3
}
sum := sumResult{}
// Add end block to current digest.
blockScalar(&dig.s, block.msg)
binary.LittleEndian.PutUint32(sum.digest[0:], dig.s[0])
binary.LittleEndian.PutUint32(sum.digest[4:], dig.s[1])
binary.LittleEndian.PutUint32(sum.digest[8:], dig.s[2])
binary.LittleEndian.PutUint32(sum.digest[12:], dig.s[3])
block.sumCh <- sum
if block.msg != nil {
s.buffers <- block.msg
}
continue
}
if len(block.msg) == 0 {
continue
}
lanes[lanesFilled] = block
lanesFilled++
return
default:
return
}
}
}
addNewClient := func(cl newClient) {
if _, ok := clients[cl.uid]; ok {
panic("internal error: duplicate client registration")
}
clients[cl.uid] = cl.input
}
allLanesFilled := func() bool {
return lanesFilled == Lanes || lanesFilled >= len(clients)
}
for {
// Step 1.
for lanesFilled == 0 {
select {
case cl, ok := <-newClients:
if !ok {
return
}
addNewClient(cl)
// Check if it already sent a payload.
addToLane(cl.uid)
continue
case uid := <-s.cycle:
addToLane(uid)
}
}
fillLanes:
for !allLanesFilled() {
select {
case cl, ok := <-newClients:
if !ok {
return
}
addNewClient(cl)
case uid := <-s.cycle:
addToLane(uid)
default:
// Nothing more queued...
break fillLanes
}
}
// If we did not fill all lanes, check if there is more waiting
if !allLanesFilled() {
runtime.Gosched()
for uid := range clients {
addToLane(uid)
if allLanesFilled() {
break
}
}
}
if false {
if !allLanesFilled() {
fmt.Println("Not all lanes filled", lanesFilled, "of", len(clients))
//pprof.Lookup("goroutine").WriteTo(os.Stdout, 1)
} else if true {
fmt.Println("all lanes filled")
}
}
// Process the lanes we could collect
s.blocks(lanes[:lanesFilled])
// Clear lanes...
lanesFilled = 0
// Add all current queued
for uid := range clients {
addToLane(uid)
if allLanesFilled() {
break
}
}
}
}
func (s *md5Server) Close() {
if s.newInput != nil {
close(s.newInput)
s.newInput = nil
}
}
// Invoke assembly and send results back
func (s *md5Server) blocks(lanes []blockInput) {
if len(lanes) < useScalarBelow {
// Use scalar routine when below this many lanes
switch len(lanes) {
case 0:
case 1:
lane := lanes[0]
var d digest
a, ok := s.digests[lane.uid]
if ok {
d.s[0] = binary.LittleEndian.Uint32(a[0:4])
d.s[1] = binary.LittleEndian.Uint32(a[4:8])
d.s[2] = binary.LittleEndian.Uint32(a[8:12])
d.s[3] = binary.LittleEndian.Uint32(a[12:16])
} else {
d.s[0] = init0
d.s[1] = init1
d.s[2] = init2
d.s[3] = init3
}
if len(lane.msg) > 0 {
// Update...
blockScalar(&d.s, lane.msg)
}
dig := [Size]byte{}
binary.LittleEndian.PutUint32(dig[0:], d.s[0])
binary.LittleEndian.PutUint32(dig[4:], d.s[1])
binary.LittleEndian.PutUint32(dig[8:], d.s[2])
binary.LittleEndian.PutUint32(dig[12:], d.s[3])
s.digests[lane.uid] = dig
if lane.msg != nil {
s.buffers <- lane.msg
}
lanes[0] = blockInput{}
default:
s.wg.Add(len(lanes))
var results [useScalarBelow]digest
for i := range lanes {
lane := lanes[i]
go func(i int) {
var d digest
defer s.wg.Done()
a, ok := s.digests[lane.uid]
if ok {
d.s[0] = binary.LittleEndian.Uint32(a[0:4])
d.s[1] = binary.LittleEndian.Uint32(a[4:8])
d.s[2] = binary.LittleEndian.Uint32(a[8:12])
d.s[3] = binary.LittleEndian.Uint32(a[12:16])
} else {
d.s[0] = init0
d.s[1] = init1
d.s[2] = init2
d.s[3] = init3
}
if len(lane.msg) == 0 {
results[i] = d
return
}
// Update...
blockScalar(&d.s, lane.msg)
results[i] = d
}(i)
}
s.wg.Wait()
for i, lane := range lanes {
dig := [Size]byte{}
binary.LittleEndian.PutUint32(dig[0:], results[i].s[0])
binary.LittleEndian.PutUint32(dig[4:], results[i].s[1])
binary.LittleEndian.PutUint32(dig[8:], results[i].s[2])
binary.LittleEndian.PutUint32(dig[12:], results[i].s[3])
s.digests[lane.uid] = dig
if lane.msg != nil {
s.buffers <- lane.msg
}
lanes[i] = blockInput{}
}
}
return
}
inputs := [16][]byte{}
for i := range lanes {
inputs[i] = lanes[i].msg
}
// Collect active digests...
state := s.getDigests(lanes)
// Process all lanes...
s.blockMd5_x16(&state, inputs, len(lanes) <= 8)
for i, lane := range lanes {
uid := lane.uid
dig := [Size]byte{}
binary.LittleEndian.PutUint32(dig[0:], state.v0[i])
binary.LittleEndian.PutUint32(dig[4:], state.v1[i])
binary.LittleEndian.PutUint32(dig[8:], state.v2[i])
binary.LittleEndian.PutUint32(dig[12:], state.v3[i])
s.digests[uid] = dig
if lane.msg != nil {
s.buffers <- lane.msg
}
lanes[i] = blockInput{}
}
}
func (s *md5Server) getDigests(lanes []blockInput) (d digest16) {
for i, lane := range lanes {
a, ok := s.digests[lane.uid]
if ok {
d.v0[i] = binary.LittleEndian.Uint32(a[0:4])
d.v1[i] = binary.LittleEndian.Uint32(a[4:8])
d.v2[i] = binary.LittleEndian.Uint32(a[8:12])
d.v3[i] = binary.LittleEndian.Uint32(a[12:16])
} else {
d.v0[i] = init0
d.v1[i] = init1
d.v2[i] = init2
d.v3[i] = init3
}
}
return
}
|