microoptimization

README.md

@@ -47,10 +47,10 @@ Number of words|Time to check all anagrams no longer than that|Time to solve "ea
 ---------------|----------------------------------------------|-------------------------|-----------------------------------|-------------------------|---------------------------------------------
 3|0.04s||||4560
 4|0.45s|||0.08s|7,431,984
-5|12s|0.15s|0.06s|0.3s|1,347,437,484
+5|11.7s|0.15s|0.06s|0.3s|1,347,437,484
-6|6.5 minutes|1.2s|0.2s|3s|58,405,904,844
+6|6.5 minutes|1.2s|0.2s|2.9s|58,405,904,844
-7|2 hours|7s|0.9s|19s|1,070,307,744,114
+7|2 hours|6.7s|0.9s|19s|1,070,307,744,114
-8|21.5 hours|25s|2.6s|80s|10,893,594,396,594
+8|21.5 hours|25s|2.6s|79s|10,893,594,396,594
 9||2.5 minutes|13s|9.5 minutes|70,596,864,409,954
 10||5 minutes|21s|17.5 minutes|314,972,701,475,754
 

dotnet/WhiteRabbit.UnmanagedBridge/WhiteRabbit.UnmanagedBridge.cpp

@@ -27,7 +27,7 @@ void WhiteRabbitUnmanagedBridge::MD5Unmanaged::ComputeMD5(unsigned __int32 * inp
 #endif
 }
 
-void WhiteRabbitUnmanagedBridge::MD5Unmanaged::FillPhraseSet(__int64* bufferPointer, __int64* allWordsPointer, __int32* wordIndexes, unsigned __int64* permutationsPointer, int numberOfCharacters, int numberOfWords)
+void WhiteRabbitUnmanagedBridge::MD5Unmanaged::FillPhraseSet(unsigned __int64* initialBufferPointer, unsigned __int64* bufferPointer, unsigned __int64* allWordsPointer, __int32* wordIndexes, unsigned __int64* permutationsPointer, int numberOfWords)
 {
-    fillPhraseSet(bufferPointer, (unsigned __int64*)allWordsPointer, wordIndexes, permutationsPointer, numberOfCharacters, numberOfWords);
+    fillPhraseSet(initialBufferPointer, bufferPointer, allWordsPointer, wordIndexes, permutationsPointer, numberOfWords);
 }

dotnet/WhiteRabbit.UnmanagedBridge/WhiteRabbit.UnmanagedBridge.h

@@ -13,6 +13,6 @@ namespace WhiteRabbitUnmanagedBridge {
         public:
             literal int PhrasesPerSet = PHRASES_PER_SET;
             static void ComputeMD5(unsigned int* input);
-            static void FillPhraseSet(__int64* bufferPointer, __int64* allWordsPointer, __int32* wordIndexes, unsigned __int64* permutationsPointer, int numberOfCharacters, int numberOfWords);
+            static void FillPhraseSet(unsigned __int64* initialBufferPointer, unsigned __int64* bufferPointer, unsigned __int64* allWordsPointer, __int32* wordIndexes, unsigned __int64* permutationsPointer, int numberOfWords);
 	};
 }

dotnet/WhiteRabbit.UnmanagedBridge/phraseset.cpp

@@ -27,12 +27,12 @@
 #define INIT_WORD(phraseNumber) \
     auto permutation = permutationsPointer[phraseNumber]; \
     unsigned __int64 cumulativeWordOffset = 0; \
-    auto phrase = _mm256_set1_epi32(0);
+    auto phrase = avx2initialBuffer[phraseNumber];
 
 #define PROCESS_WORD(phraseNumber, wordNumber) \
     { \
         auto currentWord = allWordsPointer + wordIndexes[permutation % 16] * 128; \
-        phrase = _mm256_or_si256(phrase, *(__m256i*)(currentWord + cumulativeWordOffset)); \
+        phrase = _mm256_xor_si256(phrase, *(__m256i*)(currentWord + cumulativeWordOffset)); \
         permutation >>= 4; \
         cumulativeWordOffset += currentWord[127]; \
     }
@@ -118,8 +118,9 @@
 #define REPEAT_WORDS10(phraseNumber) REPEAT_WORDS(phraseNumber, REPEAT_WORDS_SIMPLE10)
 
 
-void fillPhraseSet(__int64* bufferPointer, unsigned __int64* allWordsPointer, __int32* wordIndexes, unsigned __int64* permutationsPointer, int numberOfCharacters, int numberOfWords)
+void fillPhraseSet(unsigned __int64* initialBufferPointer, unsigned __int64* bufferPointer, unsigned __int64* allWordsPointer, __int32* wordIndexes, unsigned __int64* permutationsPointer, int numberOfWords)
 {
+    auto avx2initialBuffer = (__m256i*)initialBufferPointer;
     auto avx2buffer = (__m256i*)bufferPointer;
 
     switch (numberOfWords)
@@ -155,15 +156,6 @@ void fillPhraseSet(__int64* bufferPointer, unsigned __int64* allWordsPointer, __
         REPEAT_PHRASES(REPEAT_WORDS10);
         break;
     }
-
-    auto length = numberOfCharacters + numberOfWords - 1;
-    auto lengthInBits = (unsigned __int32)(length << 3);
-
-#define FILL_PHRASE_LAST_BYTE(phraseNumber) ((unsigned char*)bufferPointer)[length + phraseNumber * 32] = 128;
-#define FILL_PHRASE_SET_LENGTH(phraseNumber) ((unsigned __int32*)bufferPointer)[7 + phraseNumber * 8] = lengthInBits;
-
-    REPEAT_PHRASES(FILL_PHRASE_LAST_BYTE);
-    REPEAT_PHRASES(FILL_PHRASE_SET_LENGTH);
 }
 
 #pragma managed

dotnet/WhiteRabbit.UnmanagedBridge/phraseset.h

@@ -1,3 +1,3 @@
 #pragma once
 
-void fillPhraseSet(__int64* bufferPointer, unsigned __int64* allWordsPointer, __int32* wordIndexes, unsigned __int64* permutationsPointer, int numberOfCharacters, int numberOfWords);
+void fillPhraseSet(unsigned __int64* initialBufferPointer, unsigned __int64* bufferPointer, unsigned __int64* allWordsPointer, __int32* wordIndexes, unsigned __int64* permutationsPointer, int numberOfWords);

dotnet/WhiteRabbit/PhraseSet.cs

@@ -1,6 +1,7 @@
 namespace WhiteRabbit
 {
     using System.Diagnostics;
+    using System.Linq;
     using System.Runtime.CompilerServices;
     using WhiteRabbitUnmanagedBridge;
 
@@ -17,11 +18,23 @@
             this.Buffer = new uint[8 * length];
         }
 
-        public unsafe void FillPhraseSet(Word[] allWords, int[] wordIndexes, ulong[] permutations, int permutationOffset, int numberOfCharacters)
+        public unsafe void InitPhraseSet(int numberOfCharacters, int numberOfWords)
         {
-            Debug.Assert(numberOfCharacters + wordIndexes.Length - 1 < 27);
-
             fixed (uint* bufferPointer = this.Buffer)
+            {
+                var length = (uint)(numberOfCharacters + numberOfWords - 1);
+                var lengthInBits = (uint)(length << 3);
+                for (var i = 0; i < this.Length; i++)
+                {
+                    bufferPointer[7 + i * 8] = lengthInBits;
+                    ((byte*)bufferPointer)[length + i * 32] = 128 ^ ' ';
+                }
+            }
+        }
+
+        public unsafe void FillPhraseSet(PhraseSet initial, Word[] allWords, int[] wordIndexes, ulong[] permutations, int permutationOffset)
+        {
+            fixed (uint* bufferPointer = this.Buffer, initialBufferPointer = initial.Buffer)
             {
                 fixed (ulong* permutationsPointer = permutations)
                 {
@@ -30,11 +43,11 @@
                         fixed (Word* allWordsPointer = allWords)
                         {
                             MD5Unmanaged.FillPhraseSet(
-                                (long*)bufferPointer,
+                                (ulong*)initialBufferPointer,
-                                (long*)allWordsPointer,
+                                (ulong*)bufferPointer,
+                                (ulong*)allWordsPointer,
                                 wordIndexesPointer,
                                 permutationsPointer + permutationOffset,
-                                numberOfCharacters,
                                 wordIndexes.Length);
                         }
                     }
@@ -58,7 +71,7 @@
 
         public unsafe byte[] GetBytes(int number)
         {
-            Debug.Assert(number < Constants.PhrasesPerSet);
+            Debug.Assert(number < this.Length);
 
             fixed (uint* bufferPointer = this.Buffer)
             {
@@ -82,5 +95,16 @@
                 return result;
             }
         }
+
+        public unsafe string DebugBytes(int number)
+        {
+            Debug.Assert(number < this.Length);
+
+            fixed (uint* bufferPointer = this.Buffer)
+            {
+                var bytes = (byte*)bufferPointer;
+                return string.Concat(Enumerable.Range(32 * number, 32).Select(i => bytes[i].ToString("X2")));
+            }
+        }
     }
 }

dotnet/WhiteRabbit/StringsProcessor.cs

@@ -120,6 +120,8 @@
 
         private void ProcessSum(int[] sum, Action<PhraseSet> action)
         {
+            var initialPhraseSet = new PhraseSet(Constants.PhrasesPerSet);
+            initialPhraseSet.InitPhraseSet(this.NumberOfCharacters, sum.Length);
             var phraseSet = new PhraseSet(Constants.PhrasesPerSet);
             var filter = ComputeFilter(sum);
             var wordsVariants = this.ConvertVectorsToWordIndexes(sum);
@@ -130,7 +132,7 @@
                 var permutations = PrecomputedPermutationsGenerator.HamiltonianPermutations(wordsArray.Length, filter);
                 for (var i = 0; i < permutations.Length; i += Constants.PhrasesPerSet)
                 {
-                    phraseSet.FillPhraseSet(this.AllWords, wordsArray, permutations, i, this.NumberOfCharacters);
+                    phraseSet.FillPhraseSet(initialPhraseSet, this.AllWords, wordsArray, permutations, i);
                     action(phraseSet);
                 }
             }