Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Appearance settings

Commit 03cd792

Browse filesBrowse files
nodejs-github-bottargos
nodejs-github-bot
authored and
targos
committed
deps: update simdjson to 3.12.2
PR-URL: #57084 Reviewed-By: Luigi Pinca <luigipinca@gmail.com> Reviewed-By: Rafael Gonzaga <rafael.nunu@hotmail.com>
1 parent 9e1fce9 commit 03cd792
Copy full SHA for 03cd792

File tree

Expand file treeCollapse file tree

2 files changed

+28
-12
lines changed
Open diff view settings
Filter options
Expand file treeCollapse file tree

2 files changed

+28
-12
lines changed
Open diff view settings
Collapse file

‎deps/simdjson/simdjson.cpp‎

Copy file name to clipboardExpand all lines: deps/simdjson/simdjson.cpp
+13-5Lines changed: 13 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
/* auto-generated on 2025-01-27 20:34:35 -0500. Do not edit! */
1+
/* auto-generated on 2025-02-14 16:11:36 -0500. Do not edit! */
22
/* including simdjson.cpp: */
33
/* begin file simdjson.cpp */
44
#define SIMDJSON_SRC_SIMDJSON_CPP
@@ -20813,14 +20813,18 @@ namespace simd {
2081320813

2081420814
// Copies to 'output" all bytes corresponding to a 0 in the mask (interpreted as a bitset).
2081520815
// Passing a 0 value for mask would be equivalent to writing out every byte to output.
20816-
// Only the first 32 - count_ones(mask) bytes of the result are significant but 32 bytes
20816+
// Only the first 64 - count_ones(mask) bytes of the result are significant but 64 bytes
2081720817
// get written.
2081820818
// Design consideration: it seems like a function with the
2081920819
// signature simd8<L> compress(uint32_t mask) would be
2082020820
// sensible, but the AVX ISA makes this kind of approach difficult.
2082120821
template<typename L>
2082220822
simdjson_inline void compress(uint64_t mask, L * output) const {
20823-
_mm512_mask_compressstoreu_epi8 (output,~mask,*this);
20823+
// we deliberately avoid _mm512_mask_compressstoreu_epi8 for portability
20824+
// (AMD Zen4 has terrible performance with it, it is effectively broken)
20825+
// _mm512_mask_compressstoreu_epi8 (output,~mask,*this);
20826+
__m512i compressed = _mm512_maskz_compress_epi8(~mask, *this);
20827+
_mm512_storeu_si512(output, compressed); // could use a mask
2082420828
}
2082520829

2082620830
template<typename L>
@@ -23443,14 +23447,18 @@ namespace simd {
2344323447

2344423448
// Copies to 'output" all bytes corresponding to a 0 in the mask (interpreted as a bitset).
2344523449
// Passing a 0 value for mask would be equivalent to writing out every byte to output.
23446-
// Only the first 32 - count_ones(mask) bytes of the result are significant but 32 bytes
23450+
// Only the first 64 - count_ones(mask) bytes of the result are significant but 64 bytes
2344723451
// get written.
2344823452
// Design consideration: it seems like a function with the
2344923453
// signature simd8<L> compress(uint32_t mask) would be
2345023454
// sensible, but the AVX ISA makes this kind of approach difficult.
2345123455
template<typename L>
2345223456
simdjson_inline void compress(uint64_t mask, L * output) const {
23453-
_mm512_mask_compressstoreu_epi8 (output,~mask,*this);
23457+
// we deliberately avoid _mm512_mask_compressstoreu_epi8 for portability
23458+
// (AMD Zen4 has terrible performance with it, it is effectively broken)
23459+
// _mm512_mask_compressstoreu_epi8 (output,~mask,*this);
23460+
__m512i compressed = _mm512_maskz_compress_epi8(~mask, *this);
23461+
_mm512_storeu_si512(output, compressed); // could use a mask
2345423462
}
2345523463

2345623464
template<typename L>
Collapse file

‎deps/simdjson/simdjson.h‎

Copy file name to clipboardExpand all lines: deps/simdjson/simdjson.h
+15-7Lines changed: 15 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
/* auto-generated on 2025-01-27 20:34:35 -0500. Do not edit! */
1+
/* auto-generated on 2025-02-14 16:11:36 -0500. Do not edit! */
22
/* including simdjson.h: */
33
/* begin file simdjson.h */
44
#ifndef SIMDJSON_H
@@ -2437,7 +2437,7 @@ namespace std {
24372437
#define SIMDJSON_SIMDJSON_VERSION_H
24382438

24392439
/** The version of simdjson being used (major.minor.revision) */
2440-
#define SIMDJSON_VERSION "3.12.0"
2440+
#define SIMDJSON_VERSION "3.12.2"
24412441

24422442
namespace simdjson {
24432443
enum {
@@ -2452,7 +2452,7 @@ enum {
24522452
/**
24532453
* The revision (major.minor.REVISION) of simdjson being used.
24542454
*/
2455-
SIMDJSON_VERSION_REVISION = 0
2455+
SIMDJSON_VERSION_REVISION = 2
24562456
};
24572457
} // namespace simdjson
24582458

@@ -17948,14 +17948,18 @@ namespace simd {
1794817948

1794917949
// Copies to 'output" all bytes corresponding to a 0 in the mask (interpreted as a bitset).
1795017950
// Passing a 0 value for mask would be equivalent to writing out every byte to output.
17951-
// Only the first 32 - count_ones(mask) bytes of the result are significant but 32 bytes
17951+
// Only the first 64 - count_ones(mask) bytes of the result are significant but 64 bytes
1795217952
// get written.
1795317953
// Design consideration: it seems like a function with the
1795417954
// signature simd8<L> compress(uint32_t mask) would be
1795517955
// sensible, but the AVX ISA makes this kind of approach difficult.
1795617956
template<typename L>
1795717957
simdjson_inline void compress(uint64_t mask, L * output) const {
17958-
_mm512_mask_compressstoreu_epi8 (output,~mask,*this);
17958+
// we deliberately avoid _mm512_mask_compressstoreu_epi8 for portability
17959+
// (AMD Zen4 has terrible performance with it, it is effectively broken)
17960+
// _mm512_mask_compressstoreu_epi8 (output,~mask,*this);
17961+
__m512i compressed = _mm512_maskz_compress_epi8(~mask, *this);
17962+
_mm512_storeu_si512(output, compressed); // could use a mask
1795917963
}
1796017964

1796117965
template<typename L>
@@ -65401,14 +65405,18 @@ namespace simd {
6540165405

6540265406
// Copies to 'output" all bytes corresponding to a 0 in the mask (interpreted as a bitset).
6540365407
// Passing a 0 value for mask would be equivalent to writing out every byte to output.
65404-
// Only the first 32 - count_ones(mask) bytes of the result are significant but 32 bytes
65408+
// Only the first 64 - count_ones(mask) bytes of the result are significant but 64 bytes
6540565409
// get written.
6540665410
// Design consideration: it seems like a function with the
6540765411
// signature simd8<L> compress(uint32_t mask) would be
6540865412
// sensible, but the AVX ISA makes this kind of approach difficult.
6540965413
template<typename L>
6541065414
simdjson_inline void compress(uint64_t mask, L * output) const {
65411-
_mm512_mask_compressstoreu_epi8 (output,~mask,*this);
65415+
// we deliberately avoid _mm512_mask_compressstoreu_epi8 for portability
65416+
// (AMD Zen4 has terrible performance with it, it is effectively broken)
65417+
// _mm512_mask_compressstoreu_epi8 (output,~mask,*this);
65418+
__m512i compressed = _mm512_maskz_compress_epi8(~mask, *this);
65419+
_mm512_storeu_si512(output, compressed); // could use a mask
6541265420
}
6541365421

6541465422
template<typename L>

0 commit comments

Comments
0 (0)
Morty Proxy This is a proxified and sanitized view of the page, visit original site.