Movatterモバイル変換

[0]ホーム
URL:

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

 Sign up
Appearance settings

Commitf8f4afe

Browse files
Optimize vector8_has_le() on AArch64.
Presently, the SIMD implementation of this function uses unsignedsaturating subtraction to find bytes less than or equal to thegiven value, which is a workaround for the lack of unsignedcomparison instructions on some architectures. However, Neonoffers vminvq_u8(), which returns the minimum (unsigned) value inthe vector. This commit adds a Neon-specific implementation thatuses vminvq_u8() to optimize vector8_has_le() on AArch64.In passing, adjust the SSE2 implementation to use vector8_min() andvector8_eq() to find values less than or equal to the given value.This was the only use of vector8_ssub(), so it has been removed.Reviewed-by: John Naylor <johncnaylorls@gmail.com>Discussion:https://postgr.es/m/aNHDNDSHleq0ogC_%40nathan
1 parent74b41f5 commitf8f4afe

File tree

1 file changed

+10
-27
lines changed

1 file changed

+10
-27
lines changed

‎src/include/port/simd.h‎

Lines changed: 10 additions & 27 deletions
Original file line numberDiff line numberDiff line change
@@ -86,7 +86,6 @@ static inline uint32 vector8_highbit_mask(const Vector8 v);
8686
staticinlineVector8vector8_or(constVector8v1,constVector8v2);
8787
#ifndefUSE_NO_SIMD
8888
staticinlineVector32vector32_or(constVector32v1,constVector32v2);
89-
staticinlineVector8vector8_ssub(constVector8v1,constVector8v2);
9089
#endif
9190

9291
/*
@@ -213,6 +212,10 @@ static inline bool
213212
vector8_has_le(constVector8v,constuint8c)
214213
{
215214
boolresult= false;
215+
#ifdefUSE_SSE2
216+
Vector8umin;
217+
Vector8cmpe;
218+
#endif
216219

217220
/* pre-compute the result for assert checking */
218221
#ifdefUSE_ASSERT_CHECKING
@@ -250,14 +253,12 @@ vector8_has_le(const Vector8 v, const uint8 c)
250253
}
251254
}
252255
}
253-
#else
254-
255-
/*
256-
* Use saturating subtraction to find bytes <= c, which will present as
257-
* NUL bytes. This approach is a workaround for the lack of unsigned
258-
* comparison instructions on some architectures.
259-
*/
260-
result=vector8_has_zero(vector8_ssub(v,vector8_broadcast(c)));
256+
#elif defined(USE_SSE2)
257+
umin=vector8_min(v,vector8_broadcast(c));
258+
cmpe=vector8_eq(umin,v);
259+
result=vector8_is_highbit_set(cmpe);
260+
#elif defined(USE_NEON)
261+
result=vminvq_u8(v) <=c;
261262
#endif
262263

263264
Assert(assert_result==result);
@@ -358,24 +359,6 @@ vector32_or(const Vector32 v1, const Vector32 v2)
358359
}
359360
#endif/* ! USE_NO_SIMD */
360361

361-
/*
362-
* Return the result of subtracting the respective elements of the input
363-
* vectors using saturation (i.e., if the operation would yield a value less
364-
* than zero, zero is returned instead). For more information on saturation
365-
* arithmetic, see https://en.wikipedia.org/wiki/Saturation_arithmetic
366-
*/
367-
#ifndefUSE_NO_SIMD
368-
staticinlineVector8
369-
vector8_ssub(constVector8v1,constVector8v2)
370-
{
371-
#ifdefUSE_SSE2
372-
return_mm_subs_epu8(v1,v2);
373-
#elif defined(USE_NEON)
374-
returnvqsubq_u8(v1,v2);
375-
#endif
376-
}
377-
#endif/* ! USE_NO_SIMD */
378-
379362
/*
380363
* Return a vector with all bits set in each lane where the corresponding
381364
* lanes in the inputs are equal.

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp