- use better hash function.

This is djb2 which is supposedly one of the best string hashing functions around.
Avoids annoying dependencies and is a lot simpler and faster than the alternatives.
This commit is contained in:
Christoph Oelckers 2023-08-21 23:29:07 +02:00
parent 5433430767
commit 2a6fb6da84
1 changed files with 18 additions and 33 deletions

View File

@ -61,6 +61,16 @@ static void UpperCopy(char* to, const char* from)
to[i] = 0;
}
//djb2
static uint32_t MakeHash(const char* str, size_t length = SIZE_MAX)
{
uint32_t hash = 5381;
uint32_t c;
while (length-- > 0 && (c = *str++)) hash = hash * 33 + (c | 32);
return hash;
}
static void md5Hash(FileReader& reader, uint8_t* digest)
{
using namespace fs_private::md5;
@ -522,7 +532,7 @@ int FileSystem::CheckNumForName (const char *name, int space)
}
UpperCopy (uname, name);
i = FirstLumpIndex[LumpNameHash (uname) % NumEntries];
i = FirstLumpIndex[MakeHash(uname, 8) % NumEntries];
while (i != NULL_INDEX)
{
@ -560,7 +570,7 @@ int FileSystem::CheckNumForName (const char *name, int space, int rfnum, bool ex
}
UpperCopy (uname, name);
i = FirstLumpIndex[LumpNameHash (uname) % NumEntries];
i = FirstLumpIndex[MakeHash (uname, 8) % NumEntries];
// If exact is true if will only find lumps in the same WAD, otherwise
// also those in earlier WADs.
@ -619,7 +629,7 @@ int FileSystem::CheckNumForFullName (const char *name, bool trynormal, int names
uint32_t *nli = ignoreext ? NextLumpIndex_NoExt : NextLumpIndex_FullName;
auto len = strlen(name);
for (i = fli[MakeKey(name) % NumEntries]; i != NULL_INDEX; i = nli[i])
for (i = fli[MakeHash(name) % NumEntries]; i != NULL_INDEX; i = nli[i])
{
if (strnicmp(name, FileInfo[i].longName.c_str(), len)) continue;
if (FileInfo[i].longName[len] == 0) break; // this is a full match
@ -648,7 +658,7 @@ int FileSystem::CheckNumForFullName (const char *name, int rfnum)
return CheckNumForFullName (name);
}
i = FirstLumpIndex_FullName[MakeKey (name) % NumEntries];
i = FirstLumpIndex_FullName[MakeHash (name) % NumEntries];
while (i != NULL_INDEX &&
(stricmp(name, FileInfo[i].longName.c_str()) || FileInfo[i].rfnum != rfnum))
@ -700,7 +710,7 @@ int FileSystem::FindFileWithExtensions(const char* name, const char *const *exts
uint32_t* nli = NextLumpIndex_NoExt;
auto len = strlen(name);
for (i = fli[MakeKey(name) % NumEntries]; i != NULL_INDEX; i = nli[i])
for (i = fli[MakeHash(name) % NumEntries]; i != NULL_INDEX; i = nli[i])
{
if (strnicmp(name, FileInfo[i].longName.c_str(), len)) continue;
if (FileInfo[i].longName[len] != '.') continue; // we are looking for extensions but this file doesn't have one.
@ -820,31 +830,6 @@ int FileSystem::GetFileFlags (int lump)
return FileInfo[lump].lump->Flags ^ FileInfo[lump].flags;
}
//==========================================================================
//
// LumpNameHash
//
// NOTE: s should already be uppercase, in contrast to the BOOM version.
//
// Hash function used for lump names.
// Must be mod'ed with table size.
// Can be used for any 8-character names.
//
//==========================================================================
uint32_t FileSystem::LumpNameHash (const char *s)
{
const uint32_t *table = GetCRCTable ();
uint32_t hash = 0xffffffff;
int i;
for (i = 8; i > 0 && *s; --i, ++s)
{
hash = CRC1 (hash, *s, table);
}
return hash ^ 0xffffffff;
}
//==========================================================================
//
// InitHashChains
@ -875,14 +860,14 @@ void FileSystem::InitHashChains (void)
// Now set up the chains
for (i = 0; i < (unsigned)NumEntries; i++)
{
j = LumpNameHash (FileInfo[i].shortName.String) % NumEntries;
j = MakeHash (FileInfo[i].shortName.String, 8) % NumEntries;
NextLumpIndex[i] = FirstLumpIndex[j];
FirstLumpIndex[j] = i;
// Do the same for the full paths
if (!FileInfo[i].longName.empty())
{
j = MakeKey(FileInfo[i].longName.c_str()) % NumEntries;
j = MakeHash(FileInfo[i].longName.c_str()) % NumEntries;
NextLumpIndex_FullName[i] = FirstLumpIndex_FullName[j];
FirstLumpIndex_FullName[j] = i;
@ -891,7 +876,7 @@ void FileSystem::InitHashChains (void)
auto slash = nameNoExt.find_last_of('/');
if ((dot > slash || slash == std::string::npos) && dot != std::string::npos) nameNoExt.resize(dot);
j = MakeKey(nameNoExt.c_str()) % NumEntries;
j = MakeHash(nameNoExt.c_str()) % NumEntries;
NextLumpIndex_NoExt[i] = FirstLumpIndex_NoExt[j];
FirstLumpIndex_NoExt[j] = i;