dolysis
/
snappy
mirror of https://github.com/google/snappy.git
3
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Optimize check for uncommon decompression for ARM, saving two instructions and three cycles. 

PiperOrigin-RevId: 517141646
This commit is contained in:
Snappy Team 2023-03-16 09:20:43 -07:00 committed by Victor Costan
parent dc05e02648
commit 9c42b71b19
1 changed files with 10 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
snappy.cc
View File

@ -1234,16 +1234,21 @@ std::pair<const uint8_t*, ptrdiff_t> DecompressBranchless(
assert(tag == ip[-1]); assert(tag == ip[-1]);
// For literals tag_type = 0, hence we will always obtain 0 from // For literals tag_type = 0, hence we will always obtain 0 from
// ExtractLowBytes. For literals offset will thus be kLiteralOffset. // ExtractLowBytes. For literals offset will thus be kLiteralOffset.
ptrdiff_t len_min_offset = kLengthMinusOffset[tag]; ptrdiff_t len_minus_offset = kLengthMinusOffset[tag];
uint32_t next;
#if defined(__aarch64__) #if defined(__aarch64__)
size_t tag_type = AdvanceToNextTagARMOptimized(&ip, &tag); size_t tag_type = AdvanceToNextTagARMOptimized(&ip, &tag);
// We never need more than 16 bits. Doing a Load16 allows the compiler
// to elide the masking operation in ExtractOffset.
next = LittleEndian::Load16(old_ip);
#else #else
size_t tag_type = AdvanceToNextTagX86Optimized(&ip, &tag); size_t tag_type = AdvanceToNextTagX86Optimized(&ip, &tag);
next = LittleEndian::Load32(old_ip);
#endif #endif
uint32_t next = LittleEndian::Load32(old_ip); size_t len = len_minus_offset & 0xFF;
size_t len = len_min_offset & 0xFF; ptrdiff_t extracted = ExtractOffset(next, tag_type);
len_min_offset -= ExtractOffset(next, tag_type); ptrdiff_t len_min_offset = len_minus_offset - extracted;
if (SNAPPY_PREDICT_FALSE(len_min_offset > 0)) { if (SNAPPY_PREDICT_FALSE(len_minus_offset > extracted)) {
if (SNAPPY_PREDICT_FALSE(len & 0x80)) { if (SNAPPY_PREDICT_FALSE(len & 0x80)) {
// Exceptional case (long literal or copy 4). // Exceptional case (long literal or copy 4).
// Actually doing the copy here is negatively impacting the main // Actually doing the copy here is negatively impacting the main