My working unpac space for OCaml projects in development
Merge pull request #4509 from arpadpanyik-arm/revert_branch_opt
AArch64: Revert previous branch optimization
authored by
Yann Collet
and committed by
GitHub
7ef9b303
7dd8bb53
+5
-22
+5
-22
vendor/git/zstd-c/lib/decompress/zstd_decompress_block.c
+5
-22
vendor/git/zstd-c/lib/decompress/zstd_decompress_block.c
···
1261
1261
const ZSTD_seqSymbol* const mlDInfo = seqState->stateML.table + seqState->stateML.state;
1262
1262
const ZSTD_seqSymbol* const ofDInfo = seqState->stateOffb.table + seqState->stateOffb.state;
1263
1263
# endif
1264
+
(void)longOffsets;
1264
1265
seq.matchLength = mlDInfo->baseValue;
1265
1266
seq.litLength = llDInfo->baseValue;
1266
1267
{ U32 const ofBase = ofDInfo->baseValue;
···
1290
1291
ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
1291
1292
ZSTD_STATIC_ASSERT(STREAM_ACCUMULATOR_MIN_32 > LONG_OFFSETS_MAX_EXTRA_BITS_32);
1292
1293
ZSTD_STATIC_ASSERT(STREAM_ACCUMULATOR_MIN_32 - LONG_OFFSETS_MAX_EXTRA_BITS_32 >= MaxMLBits);
1293
-
if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) {
1294
-
/* Always read extra bits, this keeps the logic simple,
1295
-
* avoids branches, and avoids accidentally reading 0 bits.
1296
-
*/
1297
-
U32 const extraBits = LONG_OFFSETS_MAX_EXTRA_BITS_32;
1298
-
offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
1299
-
BIT_reloadDStream(&seqState->DStream);
1300
-
offset += BIT_readBitsFast(&seqState->DStream, extraBits);
1301
-
} else {
1302
-
offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/); /* <= (ZSTD_WINDOWLOG_MAX-1) bits */
1303
-
if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
1304
-
}
1294
+
offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/); /* <= (ZSTD_WINDOWLOG_MAX-1) bits */
1305
1295
prevOffset2 = prevOffset1;
1306
1296
prevOffset1 = prevOffset0;
1307
1297
prevOffset0 = offset;
···
1332
1322
seq.offset = offset;
1333
1323
}
1334
1324
1335
-
if (mlBits > 0) {
1325
+
if (mlBits > 0)
1336
1326
seq.matchLength += BIT_readBitsFast(&seqState->DStream, mlBits/*>0*/);
1337
1327
1338
-
if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32))
1339
-
BIT_reloadDStream(&seqState->DStream);
1340
-
if (MEM_64bits() && (totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog)))
1341
-
BIT_reloadDStream(&seqState->DStream);
1342
-
}
1328
+
if (UNLIKELY(totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog)))
1329
+
BIT_reloadDStream(&seqState->DStream);
1343
1330
1344
1331
/* Ensure there are enough bits to read the rest of data in 64-bit mode. */
1345
1332
ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64);
···
1347
1334
if (llBits > 0)
1348
1335
seq.litLength += BIT_readBitsFast(&seqState->DStream, llBits/*>0*/);
1349
1336
1350
-
if (MEM_32bits())
1351
-
BIT_reloadDStream(&seqState->DStream);
1352
-
1353
1337
DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u",
1354
1338
(U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
1355
1339
···
1357
1341
/* Don't update FSE state for last sequence. */
1358
1342
ZSTD_updateFseStateWithDInfo(&seqState->stateLL, &seqState->DStream, llNext, llnbBits); /* <= 9 bits */
1359
1343
ZSTD_updateFseStateWithDInfo(&seqState->stateML, &seqState->DStream, mlNext, mlnbBits); /* <= 9 bits */
1360
-
if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream); /* <= 18 bits */
1361
1344
ZSTD_updateFseStateWithDInfo(&seqState->stateOffb, &seqState->DStream, ofNext, ofnbBits); /* <= 8 bits */
1362
1345
BIT_reloadDStream(&seqState->DStream);
1363
1346
}