🔥 Speed and memory optimizations (OpenMP !):

- OpenMP multiprocessing for dataset generation
- remove some temporaries
- add march_native flag

README.md

@@ -10,6 +10,11 @@ Implementation is based on the [spec revision 23 (2017-08-03)](https://github.co
 An unoptimized mining CPU backend is available through the compile-time flag ``-d:ethash_mining``.
 It requires compilation through the C++ backend.
 
+## Optimizations
+For maximum speed, compile Ethash with `-d:release -d:march_native -d:openmp`.
+This will compile Ethash in Nim release mode, with all supported CPU extensions (AVX2 especially) and with OpenMP multiprocessing
+On MacOS, OpenMP requires installing GCC-7. It can be done through homebrew.
+
 ## Original implementation
 Original Ethereum implementation is available [here](https://github.com/ethereum/ethash).
 

ethash.nimble

@@ -22,7 +22,9 @@ proc test(name: string, lang: string = "c") =
 task test, "Run Proof-of-Work tests (without mining)":
   test "all_tests"
 
-task test_mining, "Run Proof-of-Work and mining tests (test in release mode)":
+task test_mining, "Run Proof-of-Work and mining tests (test in release mode + OpenMP + march=native)":
   switch("define", "release")
+  switch("define", "openmp")
+  switch("define", "march_native")
   switch("define", "ethash_mining")
   test "all_tests"

nim.cfg

@@ -0,0 +1,51 @@
+# Nim eth compilation flag config
+
+####  From default nim.cfg, somehow it's not taken into account with a custom nim.cfg
+@if release or quick:
+  obj_checks:off
+  field_checks:off
+  range_checks:off
+  bound_checks:off
+  overflow_checks:off
+  assertions:off
+  stacktrace:off
+  linetrace:off
+  debugger:off
+  line_dir:off
+  dead_code_elim:on
+@end
+
+@if release:
+  opt:speed
+@end
+
+# Configuration for GCC compiler:
+gcc.options.speed = "-O3 -fno-strict-aliasing"
+gcc.options.size = "-Os"
+@if windows:
+  gcc.options.debug = "-g3 -O0 -gdwarf-3"
+@else:
+  gcc.options.debug = "-g3 -O0"
+@end
+gcc.cpp.options.speed = "-O3 -fno-strict-aliasing"
+gcc.cpp.options.size = "-Os"
+gcc.cpp.options.debug = "-g3 -O0"
+
+# Configuration for the LLVM Clang compiler:
+clang.options.debug = "-g"
+clang.options.always = "-w"
+clang.options.speed = "-O3"
+clang.options.size = "-Os"
+#######
+
+
+@if openmp:
+  # stackTrace:off # Stack traces are already removed selectively in the code
+                   # Otherwise heap alloc from string will crash the program
+  @if macosx: # Default compiler on Mac is clang without OpenMP and gcc is an alias to clang.
+              # Use Homebrew GCC instead for OpenMP support. GCC (v7), must be properly linked via `brew link gcc`
+    cc:"gcc"
+    gcc.exe:"/usr/local/bin/gcc-7"
+    gcc.linkerexe:"/usr/local/bin/gcc-7"
+  @end
+@end

src/ethash.nim

@@ -1,6 +1,13 @@
 # Copyright (c) 2018 Status Research & Development GmbH
 # Distributed under the Apache v2 License (license terms are at http://www.apache.org/licenses/LICENSE-2.0).
 
+when defined(openmp):
+  {.passC: "-fopenmp".}
+  {.passL: "-fopenmp".}
+
+when defined(march_native):
+  {.passC: "-march=native".}
+
 import ./proof_of_work
 export proof_of_work
 

src/mining.nim

@@ -120,7 +120,7 @@ proc mine*(full_size: Natural, dataset: seq[Hash[512]], header: Hash[256], diffi
 
   randomize()                       # Start with a completely random seed
   result = uint64 random(high(int)) # TODO: Nim random does not work on uint64 range.
-                                    #       Also random is deprecate and do not include the end of the range.
+                                    #       Also random is deprecated in devel and does not include the end of the range.
 
   while not result.isValid(difficulty, full_size, dataset, header):
-    inc(result) # we rely on uin overflow (mod 2^64) here.
+    inc(result) # we rely on uint overflow (mod 2^64) here.

src/proof_of_work.nim

@@ -126,18 +126,24 @@ proc calc_dataset_item*(cache: seq[Hash[512]], i: Natural): Hash[512] {.noSideEf
 
   result = keccak512 mix[]
 
-proc calc_dataset*(full_size: Natural, cache: seq[Hash[512]]): seq[Hash[512]] {.noSideEffect.} =
+when defined(openmp):
+  # Remove stacktraces when using OpenMP, heap alloc from strings will crash.
+  {.push stacktrace: off.}
+proc calc_dataset*(full_size: Natural, cache: seq[Hash[512]]): seq[Hash[512]] =
 
   result = newSeq[Hash[512]](full_size div HASH_BYTES)
+  for i in `||`(0, result.len - 1, "simd"):
+    # OpenMP loop
+    result[i] = calc_dataset_item(cache, i)
 
-  for i, hash in result.mpairs:
+when defined(openmp):
-    hash = calc_dataset_item(cache, i)
+  # Remove stacktraces when using OpenMP, heap alloc from strings will crash.
+  {.pop.}
 
 # ###############################################################################
 # Main loop
 
-type HashimotoHash = tuple[mix_digest: Hash[256], value: Hash[256]]
+type HashimotoHash = tuple[mix_digest, value: Hash[256]]
-type DatasetLookup = proc(i: Natural): Hash[512] {.noSideEffect.}
 
 template hashimoto(header: Hash[256],
               nonce: uint64,

tests/test_mining.nim

@@ -19,16 +19,16 @@ suite "Test mining":
       full_size = get_datasize(blck)
 
     echo "\nGenerating dataset"
-    var start = cpuTime()
+    var start = epochTime()
     let dag = calc_dataset(full_size, cache)
-    echo "    Done, time taken: ", $(cpuTime() - start), " seconds"
+    echo "    Done, time taken: ", $(epochTime() - start), " seconds"
 
     echo "\nStarting mining"
-    start = cpuTime()
+    start = epochTime()
     let mined_nonce = mine(full_size, dag, header, difficulty)
-    echo "    Done, time taken: ", $(cpuTime() - start), " seconds"
+    echo "    Done, time taken: ", $(epochTime() - start), " seconds"
 
-    echo "\nUnfortunately we can't really test Ethereu mining as multiple nonces are valid"
+    echo "\nUnfortunately we can't really test Ethereum mining as multiple nonces are valid"
     echo "for a set of parameters, so we only test that there is no exception or out of memory"
     echo "\nThe nonce mined was:"
     echo mined_nonce