| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | /*
 | 
					
						
							| 
									
										
										
										
											2021-11-19 12:54:45 +01:00
										 |  |  |  * Copyright (c) Yann Collet, Facebook, Inc. | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |  * All rights reserved. | 
					
						
							|  |  |  |  * | 
					
						
							|  |  |  |  * This source code is licensed under both the BSD-style license (found in the | 
					
						
							|  |  |  |  * LICENSE file in the root directory of this source tree) and the GPLv2 (found | 
					
						
							|  |  |  |  * in the COPYING file in the root directory of this source tree). | 
					
						
							|  |  |  |  * You may select, at your option, one of the above-listed licenses. | 
					
						
							|  |  |  |  */ | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2019-11-09 22:31:00 -06:00
										 |  |  | #include "zstd_compress_internal.h"  /* ZSTD_hashPtr, ZSTD_count, ZSTD_storeSeq */
 | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | #include "zstd_fast.h"
 | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  | void ZSTD_fillHashTable(ZSTD_matchState_t* ms, | 
					
						
							| 
									
										
										
										
											2019-07-20 20:47:07 +02:00
										 |  |  |                         const void* const end, | 
					
						
							|  |  |  |                         ZSTD_dictTableLoadMethod_e dtlm) | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | { | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |     const ZSTD_compressionParameters* const cParams = &ms->cParams; | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |     U32* const hashTable = ms->hashTable; | 
					
						
							|  |  |  |     U32  const hBits = cParams->hashLog; | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |     U32  const mls = cParams->minMatch; | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |     const BYTE* const base = ms->window.base; | 
					
						
							|  |  |  |     const BYTE* ip = base + ms->nextToUpdate; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE; | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |     const U32 fastHashFillStep = 3; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |     /* Always insert every fastHashFillStep position into the hash table.
 | 
					
						
							|  |  |  |      * Insert the other positions if their hash entry is empty. | 
					
						
							|  |  |  |      */ | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |     for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) { | 
					
						
							| 
									
										
										
										
											2021-01-08 11:21:43 +01:00
										 |  |  |         U32 const curr = (U32)(ip - base); | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |         size_t const hash0 = ZSTD_hashPtr(ip, hBits, mls); | 
					
						
							| 
									
										
										
										
											2021-01-08 11:21:43 +01:00
										 |  |  |         hashTable[hash0] = curr; | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |         if (dtlm == ZSTD_dtlm_fast) continue; | 
					
						
							|  |  |  |         /* Only load extra positions for ZSTD_dtlm_full */ | 
					
						
							|  |  |  |         {   U32 p; | 
					
						
							|  |  |  |             for (p = 1; p < fastHashFillStep; ++p) { | 
					
						
							|  |  |  |                 size_t const hash = ZSTD_hashPtr(ip + p, hBits, mls); | 
					
						
							|  |  |  |                 if (hashTable[hash] == 0) {  /* not yet filled */ | 
					
						
							| 
									
										
										
										
											2021-01-08 11:21:43 +01:00
										 |  |  |                     hashTable[hash] = curr + p; | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |     }   }   }   } | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | } | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2019-07-20 20:47:07 +02:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  | /**
 | 
					
						
							|  |  |  |  * If you squint hard enough (and ignore repcodes), the search operation at any | 
					
						
							|  |  |  |  * given position is broken into 4 stages: | 
					
						
							|  |  |  |  * | 
					
						
							|  |  |  |  * 1. Hash   (map position to hash value via input read) | 
					
						
							|  |  |  |  * 2. Lookup (map hash val to index via hashtable read) | 
					
						
							|  |  |  |  * 3. Load   (map index to value at that position via input read) | 
					
						
							|  |  |  |  * 4. Compare | 
					
						
							|  |  |  |  * | 
					
						
							|  |  |  |  * Each of these steps involves a memory read at an address which is computed | 
					
						
							|  |  |  |  * from the previous step. This means these steps must be sequenced and their | 
					
						
							|  |  |  |  * latencies are cumulative. | 
					
						
							|  |  |  |  * | 
					
						
							|  |  |  |  * Rather than do 1->2->3->4 sequentially for a single position before moving | 
					
						
							|  |  |  |  * onto the next, this implementation interleaves these operations across the | 
					
						
							|  |  |  |  * next few positions: | 
					
						
							|  |  |  |  * | 
					
						
							|  |  |  |  * R = Repcode Read & Compare | 
					
						
							|  |  |  |  * H = Hash | 
					
						
							|  |  |  |  * T = Table Lookup | 
					
						
							|  |  |  |  * M = Match Read & Compare | 
					
						
							|  |  |  |  * | 
					
						
							|  |  |  |  * Pos | Time --> | 
					
						
							|  |  |  |  * ----+------------------- | 
					
						
							|  |  |  |  * N   | ... M | 
					
						
							|  |  |  |  * N+1 | ...   TM | 
					
						
							|  |  |  |  * N+2 |    R H   T M | 
					
						
							|  |  |  |  * N+3 |         H    TM | 
					
						
							|  |  |  |  * N+4 |           R H   T M | 
					
						
							|  |  |  |  * N+5 |                H   ... | 
					
						
							|  |  |  |  * N+6 |                  R ... | 
					
						
							|  |  |  |  * | 
					
						
							|  |  |  |  * This is very much analogous to the pipelining of execution in a CPU. And just | 
					
						
							|  |  |  |  * like a CPU, we have to dump the pipeline when we find a match (i.e., take a | 
					
						
							|  |  |  |  * branch). | 
					
						
							|  |  |  |  * | 
					
						
							|  |  |  |  * When this happens, we throw away our current state, and do the following prep | 
					
						
							|  |  |  |  * to re-enter the loop: | 
					
						
							|  |  |  |  * | 
					
						
							|  |  |  |  * Pos | Time --> | 
					
						
							|  |  |  |  * ----+------------------- | 
					
						
							|  |  |  |  * N   | H T | 
					
						
							|  |  |  |  * N+1 |  H | 
					
						
							|  |  |  |  * | 
					
						
							|  |  |  |  * This is also the work we do at the beginning to enter the loop initially. | 
					
						
							|  |  |  |  */ | 
					
						
							| 
									
										
										
										
											2019-11-09 22:31:00 -06:00
										 |  |  | FORCE_INLINE_TEMPLATE size_t | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  | ZSTD_compressBlock_fast_noDict_generic( | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], | 
					
						
							|  |  |  |         void const* src, size_t srcSize, | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         U32 const mls, U32 const hasStep) | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  | { | 
					
						
							|  |  |  |     const ZSTD_compressionParameters* const cParams = &ms->cParams; | 
					
						
							|  |  |  |     U32* const hashTable = ms->hashTable; | 
					
						
							|  |  |  |     U32 const hlog = cParams->hashLog; | 
					
						
							|  |  |  |     /* support stepSize of 0 */ | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |     size_t const stepSize = hasStep ? (cParams->targetLength + !(cParams->targetLength) + 1) : 2; | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |     const BYTE* const base = ms->window.base; | 
					
						
							|  |  |  |     const BYTE* const istart = (const BYTE*)src; | 
					
						
							| 
									
										
										
										
											2019-07-20 20:47:07 +02:00
										 |  |  |     const U32   endIndex = (U32)((size_t)(istart - base) + srcSize); | 
					
						
							| 
									
										
										
										
											2020-09-18 21:38:36 +02:00
										 |  |  |     const U32   prefixStartIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog); | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |     const BYTE* const prefixStart = base + prefixStartIndex; | 
					
						
							|  |  |  |     const BYTE* const iend = istart + srcSize; | 
					
						
							|  |  |  |     const BYTE* const ilimit = iend - HASH_READ_SIZE; | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  | 
 | 
					
						
							|  |  |  |     const BYTE* anchor = istart; | 
					
						
							|  |  |  |     const BYTE* ip0 = istart; | 
					
						
							|  |  |  |     const BYTE* ip1; | 
					
						
							|  |  |  |     const BYTE* ip2; | 
					
						
							|  |  |  |     const BYTE* ip3; | 
					
						
							|  |  |  |     U32 current0; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     U32 rep_offset1 = rep[0]; | 
					
						
							|  |  |  |     U32 rep_offset2 = rep[1]; | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |     U32 offsetSaved = 0; | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |     size_t hash0; /* hash for ip0 */ | 
					
						
							|  |  |  |     size_t hash1; /* hash for ip1 */ | 
					
						
							|  |  |  |     U32 idx; /* match idx for ip0 */ | 
					
						
							|  |  |  |     U32 mval; /* src value at match idx */ | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     U32 offcode; | 
					
						
							|  |  |  |     const BYTE* match0; | 
					
						
							|  |  |  |     size_t mLength; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     /* ip0 and ip1 are always adjacent. The targetLength skipping and
 | 
					
						
							|  |  |  |      * uncompressibility acceleration is applied to every other position, | 
					
						
							|  |  |  |      * matching the behavior of #1562. step therefore represents the gap | 
					
						
							|  |  |  |      * between pairs of positions, from ip0 to ip2 or ip1 to ip3. */ | 
					
						
							|  |  |  |     size_t step; | 
					
						
							|  |  |  |     const BYTE* nextStep; | 
					
						
							|  |  |  |     const size_t kStepIncr = (1 << (kSearchStrength - 1)); | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2019-11-01 21:36:06 -05:00
										 |  |  |     DEBUGLOG(5, "ZSTD_compressBlock_fast_generic"); | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |     ip0 += (ip0 == prefixStart); | 
					
						
							| 
									
										
										
										
											2021-01-08 11:21:43 +01:00
										 |  |  |     {   U32 const curr = (U32)(ip0 - base); | 
					
						
							|  |  |  |         U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog); | 
					
						
							|  |  |  |         U32 const maxRep = curr - windowLow; | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         if (rep_offset2 > maxRep) offsetSaved = rep_offset2, rep_offset2 = 0; | 
					
						
							|  |  |  |         if (rep_offset1 > maxRep) offsetSaved = rep_offset1, rep_offset1 = 0; | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |     } | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |     /* start each op */ | 
					
						
							|  |  |  | _start: /* Requires: ip0 */ | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     step = stepSize; | 
					
						
							|  |  |  |     nextStep = ip0 + kStepIncr; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     /* calculate positions, ip0 - anchor == 0, so we skip step calc */ | 
					
						
							|  |  |  |     ip1 = ip0 + 1; | 
					
						
							|  |  |  |     ip2 = ip0 + step; | 
					
						
							|  |  |  |     ip3 = ip2 + 1; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     if (ip3 >= ilimit) { | 
					
						
							|  |  |  |         goto _cleanup; | 
					
						
							|  |  |  |     } | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     hash0 = ZSTD_hashPtr(ip0, hlog, mls); | 
					
						
							|  |  |  |     hash1 = ZSTD_hashPtr(ip1, hlog, mls); | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     idx = hashTable[hash0]; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     do { | 
					
						
							|  |  |  |         /* load repcode match for ip[2]*/ | 
					
						
							|  |  |  |         const U32 rval = MEM_read32(ip2 - rep_offset1); | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         /* write back hash table entry */ | 
					
						
							|  |  |  |         current0 = (U32)(ip0 - base); | 
					
						
							|  |  |  |         hashTable[hash0] = current0; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         /* check repcode at ip[2] */ | 
					
						
							|  |  |  |         if ((MEM_read32(ip2) == rval) & (rep_offset1 > 0)) { | 
					
						
							|  |  |  |             ip0 = ip2; | 
					
						
							|  |  |  |             match0 = ip0 - rep_offset1; | 
					
						
							|  |  |  |             mLength = ip0[-1] == match0[-1]; | 
					
						
							|  |  |  |             ip0 -= mLength; | 
					
						
							|  |  |  |             match0 -= mLength; | 
					
						
							|  |  |  |             offcode = STORE_REPCODE_1; | 
					
						
							| 
									
										
										
										
											2020-09-18 21:38:36 +02:00
										 |  |  |             mLength += 4; | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |             goto _match; | 
					
						
							|  |  |  |         } | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  | 
 | 
					
						
							|  |  |  |         /* load match for ip[0] */ | 
					
						
							|  |  |  |         if (idx >= prefixStartIndex) { | 
					
						
							|  |  |  |             mval = MEM_read32(base + idx); | 
					
						
							|  |  |  |         } else { | 
					
						
							|  |  |  |             mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */ | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |         } | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  | 
 | 
					
						
							|  |  |  |         /* check match at ip[0] */ | 
					
						
							|  |  |  |         if (MEM_read32(ip0) == mval) { | 
					
						
							|  |  |  |             /* found a match! */ | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |             goto _offset; | 
					
						
							|  |  |  |         } | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  | 
 | 
					
						
							|  |  |  |         /* lookup ip[1] */ | 
					
						
							|  |  |  |         idx = hashTable[hash1]; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         /* hash ip[2] */ | 
					
						
							|  |  |  |         hash0 = hash1; | 
					
						
							|  |  |  |         hash1 = ZSTD_hashPtr(ip2, hlog, mls); | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         /* advance to next positions */ | 
					
						
							|  |  |  |         ip0 = ip1; | 
					
						
							|  |  |  |         ip1 = ip2; | 
					
						
							|  |  |  |         ip2 = ip3; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         /* write back hash table entry */ | 
					
						
							|  |  |  |         current0 = (U32)(ip0 - base); | 
					
						
							|  |  |  |         hashTable[hash0] = current0; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         /* load match for ip[0] */ | 
					
						
							|  |  |  |         if (idx >= prefixStartIndex) { | 
					
						
							|  |  |  |             mval = MEM_read32(base + idx); | 
					
						
							|  |  |  |         } else { | 
					
						
							|  |  |  |             mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */ | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |         } | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         /* check match at ip[0] */ | 
					
						
							|  |  |  |         if (MEM_read32(ip0) == mval) { | 
					
						
							|  |  |  |             /* found a match! */ | 
					
						
							|  |  |  |             goto _offset; | 
					
						
							|  |  |  |         } | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         /* lookup ip[1] */ | 
					
						
							|  |  |  |         idx = hashTable[hash1]; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         /* hash ip[2] */ | 
					
						
							|  |  |  |         hash0 = hash1; | 
					
						
							|  |  |  |         hash1 = ZSTD_hashPtr(ip2, hlog, mls); | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         /* advance to next positions */ | 
					
						
							|  |  |  |         ip0 = ip1; | 
					
						
							|  |  |  |         ip1 = ip2; | 
					
						
							|  |  |  |         ip2 = ip0 + step; | 
					
						
							|  |  |  |         ip3 = ip1 + step; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         /* calculate step */ | 
					
						
							|  |  |  |         if (ip2 >= nextStep) { | 
					
						
							|  |  |  |             step++; | 
					
						
							|  |  |  |             PREFETCH_L1(ip1 + 64); | 
					
						
							|  |  |  |             PREFETCH_L1(ip1 + 128); | 
					
						
							|  |  |  |             nextStep += kStepIncr; | 
					
						
							|  |  |  |         } | 
					
						
							|  |  |  |     } while (ip3 < ilimit); | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | _cleanup: | 
					
						
							|  |  |  |     /* Note that there are probably still a couple positions we could search.
 | 
					
						
							|  |  |  |      * However, it seems to be a meaningful performance hit to try to search | 
					
						
							|  |  |  |      * them. So let's not. */ | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  | 
 | 
					
						
							|  |  |  |     /* save reps for next block */ | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |     rep[0] = rep_offset1 ? rep_offset1 : offsetSaved; | 
					
						
							|  |  |  |     rep[1] = rep_offset2 ? rep_offset2 : offsetSaved; | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  | 
 | 
					
						
							|  |  |  |     /* Return the last literals size */ | 
					
						
							| 
									
										
										
										
											2019-07-20 20:47:07 +02:00
										 |  |  |     return (size_t)(iend - anchor); | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  | 
 | 
					
						
							|  |  |  | _offset: /* Requires: ip0, idx */ | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     /* Compute the offset code. */ | 
					
						
							|  |  |  |     match0 = base + idx; | 
					
						
							|  |  |  |     rep_offset2 = rep_offset1; | 
					
						
							|  |  |  |     rep_offset1 = (U32)(ip0-match0); | 
					
						
							|  |  |  |     offcode = STORE_OFFSET(rep_offset1); | 
					
						
							|  |  |  |     mLength = 4; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     /* Count the backwards match length. */ | 
					
						
							|  |  |  |     while (((ip0>anchor) & (match0>prefixStart)) && (ip0[-1] == match0[-1])) { | 
					
						
							|  |  |  |         ip0--; | 
					
						
							|  |  |  |         match0--; | 
					
						
							|  |  |  |         mLength++; | 
					
						
							|  |  |  |     } | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | _match: /* Requires: ip0, match0, offcode */ | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     /* Count the forward length. */ | 
					
						
							|  |  |  |     mLength += ZSTD_count(ip0 + mLength, match0 + mLength, iend); | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength); | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     ip0 += mLength; | 
					
						
							|  |  |  |     anchor = ip0; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     /* write next hash table entry */ | 
					
						
							|  |  |  |     if (ip1 < ip0) { | 
					
						
							|  |  |  |         hashTable[hash1] = (U32)(ip1 - base); | 
					
						
							|  |  |  |     } | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     /* Fill table and check for immediate repcode. */ | 
					
						
							|  |  |  |     if (ip0 <= ilimit) { | 
					
						
							|  |  |  |         /* Fill Table */ | 
					
						
							|  |  |  |         assert(base+current0+2 > istart);  /* check base overflow */ | 
					
						
							|  |  |  |         hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */ | 
					
						
							|  |  |  |         hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base); | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         if (rep_offset2 > 0) { /* rep_offset2==0 means rep_offset2 is invalidated */ | 
					
						
							|  |  |  |             while ( (ip0 <= ilimit) && (MEM_read32(ip0) == MEM_read32(ip0 - rep_offset2)) ) { | 
					
						
							|  |  |  |                 /* store sequence */ | 
					
						
							|  |  |  |                 size_t const rLength = ZSTD_count(ip0+4, ip0+4-rep_offset2, iend) + 4; | 
					
						
							|  |  |  |                 { U32 const tmpOff = rep_offset2; rep_offset2 = rep_offset1; rep_offset1 = tmpOff; } /* swap rep_offset2 <=> rep_offset1 */ | 
					
						
							|  |  |  |                 hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base); | 
					
						
							|  |  |  |                 ip0 += rLength; | 
					
						
							|  |  |  |                 ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, STORE_REPCODE_1, rLength); | 
					
						
							|  |  |  |                 anchor = ip0; | 
					
						
							|  |  |  |                 continue;   /* faster when present (confirmed on gcc-8) ... (?) */ | 
					
						
							|  |  |  |     }   }   } | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     goto _start; | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  | } | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  | #define ZSTD_GEN_FAST_FN(dictMode, mls, step)                                                            \
 | 
					
						
							|  |  |  |     static size_t ZSTD_compressBlock_fast_##dictMode##_##mls##_##step(                                      \ | 
					
						
							|  |  |  |             ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],                    \ | 
					
						
							|  |  |  |             void const* src, size_t srcSize)                                                       \ | 
					
						
							|  |  |  |     {                                                                                              \ | 
					
						
							|  |  |  |         return ZSTD_compressBlock_fast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls, step); \ | 
					
						
							|  |  |  |     } | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(noDict, 4, 1) | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(noDict, 5, 1) | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(noDict, 6, 1) | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(noDict, 7, 1) | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(noDict, 4, 0) | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(noDict, 5, 0) | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(noDict, 6, 0) | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(noDict, 7, 0) | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  | 
 | 
					
						
							|  |  |  | size_t ZSTD_compressBlock_fast( | 
					
						
							|  |  |  |         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], | 
					
						
							|  |  |  |         void const* src, size_t srcSize) | 
					
						
							|  |  |  | { | 
					
						
							| 
									
										
										
										
											2019-11-09 22:31:00 -06:00
										 |  |  |     U32 const mls = ms->cParams.minMatch; | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |     assert(ms->dictMatchState == NULL); | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |     if (ms->cParams.targetLength > 1) { | 
					
						
							|  |  |  |         switch(mls) | 
					
						
							|  |  |  |         { | 
					
						
							|  |  |  |         default: /* includes case 3 */ | 
					
						
							|  |  |  |         case 4 : | 
					
						
							|  |  |  |             return ZSTD_compressBlock_fast_noDict_4_1(ms, seqStore, rep, src, srcSize); | 
					
						
							|  |  |  |         case 5 : | 
					
						
							|  |  |  |             return ZSTD_compressBlock_fast_noDict_5_1(ms, seqStore, rep, src, srcSize); | 
					
						
							|  |  |  |         case 6 : | 
					
						
							|  |  |  |             return ZSTD_compressBlock_fast_noDict_6_1(ms, seqStore, rep, src, srcSize); | 
					
						
							|  |  |  |         case 7 : | 
					
						
							|  |  |  |             return ZSTD_compressBlock_fast_noDict_7_1(ms, seqStore, rep, src, srcSize); | 
					
						
							|  |  |  |         } | 
					
						
							|  |  |  |     } else { | 
					
						
							|  |  |  |         switch(mls) | 
					
						
							|  |  |  |         { | 
					
						
							|  |  |  |         default: /* includes case 3 */ | 
					
						
							|  |  |  |         case 4 : | 
					
						
							|  |  |  |             return ZSTD_compressBlock_fast_noDict_4_0(ms, seqStore, rep, src, srcSize); | 
					
						
							|  |  |  |         case 5 : | 
					
						
							|  |  |  |             return ZSTD_compressBlock_fast_noDict_5_0(ms, seqStore, rep, src, srcSize); | 
					
						
							|  |  |  |         case 6 : | 
					
						
							|  |  |  |             return ZSTD_compressBlock_fast_noDict_6_0(ms, seqStore, rep, src, srcSize); | 
					
						
							|  |  |  |         case 7 : | 
					
						
							|  |  |  |             return ZSTD_compressBlock_fast_noDict_7_0(ms, seqStore, rep, src, srcSize); | 
					
						
							|  |  |  |         } | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |     } | 
					
						
							|  |  |  | } | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | FORCE_INLINE_TEMPLATE | 
					
						
							|  |  |  | size_t ZSTD_compressBlock_fast_dictMatchState_generic( | 
					
						
							|  |  |  |         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         void const* src, size_t srcSize, U32 const mls, U32 const hasStep) | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | { | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |     const ZSTD_compressionParameters* const cParams = &ms->cParams; | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |     U32* const hashTable = ms->hashTable; | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |     U32 const hlog = cParams->hashLog; | 
					
						
							|  |  |  |     /* support stepSize of 0 */ | 
					
						
							|  |  |  |     U32 const stepSize = cParams->targetLength + !(cParams->targetLength); | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |     const BYTE* const base = ms->window.base; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     const BYTE* const istart = (const BYTE*)src; | 
					
						
							|  |  |  |     const BYTE* ip = istart; | 
					
						
							|  |  |  |     const BYTE* anchor = istart; | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |     const U32   prefixStartIndex = ms->window.dictLimit; | 
					
						
							|  |  |  |     const BYTE* const prefixStart = base + prefixStartIndex; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     const BYTE* const iend = istart + srcSize; | 
					
						
							|  |  |  |     const BYTE* const ilimit = iend - HASH_READ_SIZE; | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |     U32 offset_1=rep[0], offset_2=rep[1]; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     U32 offsetSaved = 0; | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |     const ZSTD_matchState_t* const dms = ms->dictMatchState; | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |     const ZSTD_compressionParameters* const dictCParams = &dms->cParams ; | 
					
						
							|  |  |  |     const U32* const dictHashTable = dms->hashTable; | 
					
						
							|  |  |  |     const U32 dictStartIndex       = dms->window.dictLimit; | 
					
						
							|  |  |  |     const BYTE* const dictBase     = dms->window.base; | 
					
						
							|  |  |  |     const BYTE* const dictStart    = dictBase + dictStartIndex; | 
					
						
							|  |  |  |     const BYTE* const dictEnd      = dms->window.nextSrc; | 
					
						
							|  |  |  |     const U32 dictIndexDelta       = prefixStartIndex - (U32)(dictEnd - dictBase); | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |     const U32 dictAndPrefixLength  = (U32)(ip - prefixStart + dictEnd - dictStart); | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |     const U32 dictHLog             = dictCParams->hashLog; | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2019-07-20 20:47:07 +02:00
										 |  |  |     /* if a dictionary is still attached, it necessarily means that
 | 
					
						
							|  |  |  |      * it is within window size. So we just check it. */ | 
					
						
							|  |  |  |     const U32 maxDistance = 1U << cParams->windowLog; | 
					
						
							|  |  |  |     const U32 endIndex = (U32)((size_t)(ip - base) + srcSize); | 
					
						
							|  |  |  |     assert(endIndex - prefixStartIndex <= maxDistance); | 
					
						
							|  |  |  |     (void)maxDistance; (void)endIndex;   /* these variables are not used when assert() is disabled */ | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |     (void)hasStep; /* not currently specialized on whether it's accelerated */ | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2021-11-19 12:54:45 +01:00
										 |  |  |     /* ensure there will be no underflow
 | 
					
						
							| 
									
										
										
										
											2019-07-20 20:47:07 +02:00
										 |  |  |      * when translating a dict index into a local index */ | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |     assert(prefixStartIndex >= (U32)(dictEnd - dictBase)); | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     /* init */ | 
					
						
							| 
									
										
										
										
											2019-11-01 21:36:06 -05:00
										 |  |  |     DEBUGLOG(5, "ZSTD_compressBlock_fast_dictMatchState_generic"); | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |     ip += (dictAndPrefixLength == 0); | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |     /* dictMatchState repCode checks don't currently handle repCode == 0
 | 
					
						
							|  |  |  |      * disabling. */ | 
					
						
							|  |  |  |     assert(offset_1 <= dictAndPrefixLength); | 
					
						
							|  |  |  |     assert(offset_2 <= dictAndPrefixLength); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | 
 | 
					
						
							|  |  |  |     /* Main Search Loop */ | 
					
						
							|  |  |  |     while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */ | 
					
						
							|  |  |  |         size_t mLength; | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |         size_t const h = ZSTD_hashPtr(ip, hlog, mls); | 
					
						
							| 
									
										
										
										
											2021-01-08 11:21:43 +01:00
										 |  |  |         U32 const curr = (U32)(ip-base); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |         U32 const matchIndex = hashTable[h]; | 
					
						
							|  |  |  |         const BYTE* match = base + matchIndex; | 
					
						
							| 
									
										
										
										
											2021-01-08 11:21:43 +01:00
										 |  |  |         const U32 repIndex = curr + 1 - offset_1; | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |         const BYTE* repMatch = (repIndex < prefixStartIndex) ? | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |                                dictBase + (repIndex - dictIndexDelta) : | 
					
						
							|  |  |  |                                base + repIndex; | 
					
						
							| 
									
										
										
										
											2021-01-08 11:21:43 +01:00
										 |  |  |         hashTable[h] = curr;   /* update hash table */ | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |         if ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */ | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) { | 
					
						
							|  |  |  |             const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend; | 
					
						
							|  |  |  |             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4; | 
					
						
							|  |  |  |             ip++; | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |             ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_REPCODE_1, mLength); | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |         } else if ( (matchIndex <= prefixStartIndex) ) { | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |             size_t const dictHash = ZSTD_hashPtr(ip, dictHLog, mls); | 
					
						
							|  |  |  |             U32 const dictMatchIndex = dictHashTable[dictHash]; | 
					
						
							|  |  |  |             const BYTE* dictMatch = dictBase + dictMatchIndex; | 
					
						
							|  |  |  |             if (dictMatchIndex <= dictStartIndex || | 
					
						
							|  |  |  |                 MEM_read32(dictMatch) != MEM_read32(ip)) { | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |                 assert(stepSize >= 1); | 
					
						
							|  |  |  |                 ip += ((ip-anchor) >> kSearchStrength) + stepSize; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |                 continue; | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |             } else { | 
					
						
							|  |  |  |                 /* found a dict match */ | 
					
						
							| 
									
										
										
										
											2021-01-08 11:21:43 +01:00
										 |  |  |                 U32 const offset = (U32)(curr-dictMatchIndex-dictIndexDelta); | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |                 mLength = ZSTD_count_2segments(ip+4, dictMatch+4, iend, dictEnd, prefixStart) + 4; | 
					
						
							|  |  |  |                 while (((ip>anchor) & (dictMatch>dictStart)) | 
					
						
							|  |  |  |                      && (ip[-1] == dictMatch[-1])) { | 
					
						
							|  |  |  |                     ip--; dictMatch--; mLength++; | 
					
						
							|  |  |  |                 } /* catch up */ | 
					
						
							|  |  |  |                 offset_2 = offset_1; | 
					
						
							|  |  |  |                 offset_1 = offset; | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |                 ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |             } | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |         } else if (MEM_read32(match) != MEM_read32(ip)) { | 
					
						
							|  |  |  |             /* it's not a match, and we're not going to check the dictionary */ | 
					
						
							|  |  |  |             assert(stepSize >= 1); | 
					
						
							|  |  |  |             ip += ((ip-anchor) >> kSearchStrength) + stepSize; | 
					
						
							|  |  |  |             continue; | 
					
						
							|  |  |  |         } else { | 
					
						
							|  |  |  |             /* found a regular match */ | 
					
						
							|  |  |  |             U32 const offset = (U32)(ip-match); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |             mLength = ZSTD_count(ip+4, match+4, iend) + 4; | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |             while (((ip>anchor) & (match>prefixStart)) | 
					
						
							|  |  |  |                  && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */ | 
					
						
							|  |  |  |             offset_2 = offset_1; | 
					
						
							|  |  |  |             offset_1 = offset; | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |             ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength); | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |         } | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | 
 | 
					
						
							|  |  |  |         /* match found */ | 
					
						
							|  |  |  |         ip += mLength; | 
					
						
							|  |  |  |         anchor = ip; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         if (ip <= ilimit) { | 
					
						
							|  |  |  |             /* Fill Table */ | 
					
						
							| 
									
										
										
										
											2021-01-08 11:21:43 +01:00
										 |  |  |             assert(base+curr+2 > istart);  /* check base overflow */ | 
					
						
							|  |  |  |             hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;  /* here because curr+2 could be > iend-8 */ | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |             hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base); | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |             /* check immediate repcode */ | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |             while (ip <= ilimit) { | 
					
						
							|  |  |  |                 U32 const current2 = (U32)(ip-base); | 
					
						
							|  |  |  |                 U32 const repIndex2 = current2 - offset_2; | 
					
						
							|  |  |  |                 const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? | 
					
						
							|  |  |  |                         dictBase - dictIndexDelta + repIndex2 : | 
					
						
							|  |  |  |                         base + repIndex2; | 
					
						
							|  |  |  |                 if ( ((U32)((prefixStartIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */) | 
					
						
							|  |  |  |                    && (MEM_read32(repMatch2) == MEM_read32(ip)) ) { | 
					
						
							|  |  |  |                     const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend; | 
					
						
							|  |  |  |                     size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4; | 
					
						
							|  |  |  |                     U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */ | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |                     ZSTD_storeSeq(seqStore, 0, anchor, iend, STORE_REPCODE_1, repLength2); | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |                     hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2; | 
					
						
							|  |  |  |                     ip += repLength2; | 
					
						
							|  |  |  |                     anchor = ip; | 
					
						
							|  |  |  |                     continue; | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |                 } | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |                 break; | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |             } | 
					
						
							| 
									
										
										
										
											2019-04-18 11:53:29 +02:00
										 |  |  |         } | 
					
						
							|  |  |  |     } | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | 
 | 
					
						
							|  |  |  |     /* save reps for next block */ | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |     rep[0] = offset_1 ? offset_1 : offsetSaved; | 
					
						
							|  |  |  |     rep[1] = offset_2 ? offset_2 : offsetSaved; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | 
 | 
					
						
							|  |  |  |     /* Return the last literals size */ | 
					
						
							| 
									
										
										
										
											2019-07-20 20:47:07 +02:00
										 |  |  |     return (size_t)(iend - anchor); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | } | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  | 
 | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(dictMatchState, 4, 0) | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(dictMatchState, 5, 0) | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(dictMatchState, 6, 0) | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(dictMatchState, 7, 0) | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  | size_t ZSTD_compressBlock_fast_dictMatchState( | 
					
						
							|  |  |  |         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], | 
					
						
							|  |  |  |         void const* src, size_t srcSize) | 
					
						
							|  |  |  | { | 
					
						
							| 
									
										
										
										
											2019-11-09 22:31:00 -06:00
										 |  |  |     U32 const mls = ms->cParams.minMatch; | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |     assert(ms->dictMatchState != NULL); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     switch(mls) | 
					
						
							|  |  |  |     { | 
					
						
							|  |  |  |     default: /* includes case 3 */ | 
					
						
							|  |  |  |     case 4 : | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         return ZSTD_compressBlock_fast_dictMatchState_4_0(ms, seqStore, rep, src, srcSize); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     case 5 : | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         return ZSTD_compressBlock_fast_dictMatchState_5_0(ms, seqStore, rep, src, srcSize); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     case 6 : | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         return ZSTD_compressBlock_fast_dictMatchState_6_0(ms, seqStore, rep, src, srcSize); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     case 7 : | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         return ZSTD_compressBlock_fast_dictMatchState_7_0(ms, seqStore, rep, src, srcSize); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     } | 
					
						
							|  |  |  | } | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  | static size_t ZSTD_compressBlock_fast_extDict_generic( | 
					
						
							|  |  |  |         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         void const* src, size_t srcSize, U32 const mls, U32 const hasStep) | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | { | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |     const ZSTD_compressionParameters* const cParams = &ms->cParams; | 
					
						
							|  |  |  |     U32* const hashTable = ms->hashTable; | 
					
						
							|  |  |  |     U32 const hlog = cParams->hashLog; | 
					
						
							|  |  |  |     /* support stepSize of 0 */ | 
					
						
							|  |  |  |     U32 const stepSize = cParams->targetLength + !(cParams->targetLength); | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |     const BYTE* const base = ms->window.base; | 
					
						
							|  |  |  |     const BYTE* const dictBase = ms->window.dictBase; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     const BYTE* const istart = (const BYTE*)src; | 
					
						
							|  |  |  |     const BYTE* ip = istart; | 
					
						
							|  |  |  |     const BYTE* anchor = istart; | 
					
						
							| 
									
										
										
										
											2019-07-20 20:47:07 +02:00
										 |  |  |     const U32   endIndex = (U32)((size_t)(istart - base) + srcSize); | 
					
						
							| 
									
										
										
										
											2019-11-01 21:36:06 -05:00
										 |  |  |     const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog); | 
					
						
							| 
									
										
										
										
											2019-07-20 20:47:07 +02:00
										 |  |  |     const U32   dictStartIndex = lowLimit; | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |     const BYTE* const dictStart = dictBase + dictStartIndex; | 
					
						
							| 
									
										
										
										
											2019-07-20 20:47:07 +02:00
										 |  |  |     const U32   dictLimit = ms->window.dictLimit; | 
					
						
							|  |  |  |     const U32   prefixStartIndex = dictLimit < lowLimit ? lowLimit : dictLimit; | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |     const BYTE* const prefixStart = base + prefixStartIndex; | 
					
						
							|  |  |  |     const BYTE* const dictEnd = dictBase + prefixStartIndex; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     const BYTE* const iend = istart + srcSize; | 
					
						
							|  |  |  |     const BYTE* const ilimit = iend - 8; | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |     U32 offset_1=rep[0], offset_2=rep[1]; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |     (void)hasStep; /* not currently specialized on whether it's accelerated */ | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2020-09-18 21:38:36 +02:00
										 |  |  |     DEBUGLOG(5, "ZSTD_compressBlock_fast_extDict_generic (offset_1=%u)", offset_1); | 
					
						
							| 
									
										
										
										
											2019-11-01 21:36:06 -05:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2019-07-20 20:47:07 +02:00
										 |  |  |     /* switch to "regular" variant if extDict is invalidated due to maxDistance */ | 
					
						
							|  |  |  |     if (prefixStartIndex == dictStartIndex) | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         return ZSTD_compressBlock_fast(ms, seqStore, rep, src, srcSize); | 
					
						
							| 
									
										
										
										
											2019-07-20 20:47:07 +02:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     /* Search Loop */ | 
					
						
							|  |  |  |     while (ip < ilimit) {  /* < instead of <=, because (ip+1) */ | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |         const size_t h = ZSTD_hashPtr(ip, hlog, mls); | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |         const U32    matchIndex = hashTable[h]; | 
					
						
							|  |  |  |         const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base; | 
					
						
							|  |  |  |         const BYTE*  match = matchBase + matchIndex; | 
					
						
							| 
									
										
										
										
											2021-01-08 11:21:43 +01:00
										 |  |  |         const U32    curr = (U32)(ip-base); | 
					
						
							|  |  |  |         const U32    repIndex = curr + 1 - offset_1; | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |         const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base; | 
					
						
							|  |  |  |         const BYTE* const repMatch = repBase + repIndex; | 
					
						
							| 
									
										
										
										
											2021-01-08 11:21:43 +01:00
										 |  |  |         hashTable[h] = curr;   /* update hash table */ | 
					
						
							|  |  |  |         DEBUGLOG(7, "offset_1 = %u , curr = %u", offset_1, curr); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2021-11-19 12:54:45 +01:00
										 |  |  |         if ( ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow */ | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |              & (offset_1 <= curr+1 - dictStartIndex) ) /* note: we are searching at curr+1 */ | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |            && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) { | 
					
						
							| 
									
										
										
										
											2019-11-01 21:36:06 -05:00
										 |  |  |             const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend; | 
					
						
							| 
									
										
										
										
											2019-11-09 22:31:00 -06:00
										 |  |  |             size_t const rLength = ZSTD_count_2segments(ip+1 +4, repMatch +4, iend, repMatchEnd, prefixStart) + 4; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |             ip++; | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |             ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_REPCODE_1, rLength); | 
					
						
							| 
									
										
										
										
											2019-11-09 22:31:00 -06:00
										 |  |  |             ip += rLength; | 
					
						
							|  |  |  |             anchor = ip; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |         } else { | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |             if ( (matchIndex < dictStartIndex) || | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |                  (MEM_read32(match) != MEM_read32(ip)) ) { | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |                 assert(stepSize >= 1); | 
					
						
							|  |  |  |                 ip += ((ip-anchor) >> kSearchStrength) + stepSize; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |                 continue; | 
					
						
							|  |  |  |             } | 
					
						
							| 
									
										
										
										
											2019-11-01 21:36:06 -05:00
										 |  |  |             {   const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend; | 
					
						
							|  |  |  |                 const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart; | 
					
						
							| 
									
										
										
										
											2021-01-08 11:21:43 +01:00
										 |  |  |                 U32 const offset = curr - matchIndex; | 
					
						
							| 
									
										
										
										
											2019-11-09 22:31:00 -06:00
										 |  |  |                 size_t mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |                 while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */ | 
					
						
							| 
									
										
										
										
											2019-11-09 22:31:00 -06:00
										 |  |  |                 offset_2 = offset_1; offset_1 = offset;  /* update offset history */ | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |                 ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength); | 
					
						
							| 
									
										
										
										
											2019-11-09 22:31:00 -06:00
										 |  |  |                 ip += mLength; | 
					
						
							|  |  |  |                 anchor = ip; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |         }   } | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         if (ip <= ilimit) { | 
					
						
							|  |  |  |             /* Fill Table */ | 
					
						
							| 
									
										
										
										
											2021-01-08 11:21:43 +01:00
										 |  |  |             hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2; | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |             hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |             /* check immediate repcode */ | 
					
						
							|  |  |  |             while (ip <= ilimit) { | 
					
						
							|  |  |  |                 U32 const current2 = (U32)(ip-base); | 
					
						
							|  |  |  |                 U32 const repIndex2 = current2 - offset_2; | 
					
						
							| 
									
										
										
										
											2019-11-09 22:31:00 -06:00
										 |  |  |                 const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2; | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |                 if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (offset_2 <= curr - dictStartIndex))  /* intentional overflow */ | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |                    && (MEM_read32(repMatch2) == MEM_read32(ip)) ) { | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |                     const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend; | 
					
						
							|  |  |  |                     size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4; | 
					
						
							| 
									
										
										
										
											2019-11-09 22:31:00 -06:00
										 |  |  |                     { U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; }  /* swap offset_2 <=> offset_1 */ | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |                     ZSTD_storeSeq(seqStore, 0 /*litlen*/, anchor, iend, STORE_REPCODE_1, repLength2); | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |                     hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |                     ip += repLength2; | 
					
						
							|  |  |  |                     anchor = ip; | 
					
						
							|  |  |  |                     continue; | 
					
						
							|  |  |  |                 } | 
					
						
							|  |  |  |                 break; | 
					
						
							|  |  |  |     }   }   } | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |     /* save reps for next block */ | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  |     rep[0] = offset_1; | 
					
						
							|  |  |  |     rep[1] = offset_2; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | 
 | 
					
						
							|  |  |  |     /* Return the last literals size */ | 
					
						
							| 
									
										
										
										
											2019-07-20 20:47:07 +02:00
										 |  |  |     return (size_t)(iend - anchor); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | } | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  | ZSTD_GEN_FAST_FN(extDict, 4, 0) | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(extDict, 5, 0) | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(extDict, 6, 0) | 
					
						
							|  |  |  | ZSTD_GEN_FAST_FN(extDict, 7, 0) | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | 
 | 
					
						
							| 
									
										
										
										
											2018-05-16 02:45:22 +09:00
										 |  |  | size_t ZSTD_compressBlock_fast_extDict( | 
					
						
							|  |  |  |         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], | 
					
						
							| 
									
										
										
										
											2019-01-03 22:30:03 -02:00
										 |  |  |         void const* src, size_t srcSize) | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  | { | 
					
						
							| 
									
										
										
										
											2019-11-09 22:31:00 -06:00
										 |  |  |     U32 const mls = ms->cParams.minMatch; | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     switch(mls) | 
					
						
							|  |  |  |     { | 
					
						
							|  |  |  |     default: /* includes case 3 */ | 
					
						
							|  |  |  |     case 4 : | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         return ZSTD_compressBlock_fast_extDict_4_0(ms, seqStore, rep, src, srcSize); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     case 5 : | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         return ZSTD_compressBlock_fast_extDict_5_0(ms, seqStore, rep, src, srcSize); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     case 6 : | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         return ZSTD_compressBlock_fast_extDict_6_0(ms, seqStore, rep, src, srcSize); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     case 7 : | 
					
						
							| 
									
										
										
										
											2022-01-24 11:04:45 +01:00
										 |  |  |         return ZSTD_compressBlock_fast_extDict_7_0(ms, seqStore, rep, src, srcSize); | 
					
						
							| 
									
										
										
										
											2017-10-26 16:41:47 -04:00
										 |  |  |     } | 
					
						
							|  |  |  | } |