Here's a new version of 0002 with a modified SSE2 implementation, as
discussed elsewhere [0]. This allows us to remove vector8_ssub().
[0] https://postgr.es/m/aNWO7L43UevRErw_%40nathan
--
nathan
>From a644f049add68ea78326f88d8994898c92f23c20 Mon Sep 17 00:00:00 2001
From: Nathan Bossart <nat...@postgresql.org>
Date: Mon, 22 Sep 2025 16:17:09 -0500
Subject: [PATCH v2 1/1] Optimize vector8_has_le() on AArch64.
---
src/include/port/simd.h | 32 ++++++--------------------------
1 file changed, 6 insertions(+), 26 deletions(-)
diff --git a/src/include/port/simd.h b/src/include/port/simd.h
index 97c5f353022..e317e5fdfc0 100644
--- a/src/include/port/simd.h
+++ b/src/include/port/simd.h
@@ -86,7 +86,6 @@ static inline uint32 vector8_highbit_mask(const Vector8 v);
static inline Vector8 vector8_or(const Vector8 v1, const Vector8 v2);
#ifndef USE_NO_SIMD
static inline Vector32 vector32_or(const Vector32 v1, const Vector32 v2);
-static inline Vector8 vector8_ssub(const Vector8 v1, const Vector8 v2);
#endif
/*
@@ -250,14 +249,13 @@ vector8_has_le(const Vector8 v, const uint8 c)
}
}
}
-#else
+#elif defined(USE_SSE2)
+ Vector8 umin = _mm_min_epu8(v, vector8_broadcast(c));
+ Vector8 cmpe = _mm_cmpeq_epi8(umin, v);
- /*
- * Use saturating subtraction to find bytes <= c, which will present as
- * NUL bytes. This approach is a workaround for the lack of unsigned
- * comparison instructions on some architectures.
- */
- result = vector8_has_zero(vector8_ssub(v, vector8_broadcast(c)));
+ result = vector8_is_highbit_set(cmpe);
+#elif defined(USE_NEON)
+ result = vminvq_u8(v) <= c;
#endif
Assert(assert_result == result);
@@ -358,24 +356,6 @@ vector32_or(const Vector32 v1, const Vector32 v2)
}
#endif /* ! USE_NO_SIMD */
-/*
- * Return the result of subtracting the respective elements of the input
- * vectors using saturation (i.e., if the operation would yield a value less
- * than zero, zero is returned instead). For more information on saturation
- * arithmetic, see https://en.wikipedia.org/wiki/Saturation_arithmetic
- */
-#ifndef USE_NO_SIMD
-static inline Vector8
-vector8_ssub(const Vector8 v1, const Vector8 v2)
-{
-#ifdef USE_SSE2
- return _mm_subs_epu8(v1, v2);
-#elif defined(USE_NEON)
- return vqsubq_u8(v1, v2);
-#endif
-}
-#endif /* ! USE_NO_SIMD */
-
/*
* Return a vector with all bits set in each lane where the corresponding
* lanes in the inputs are equal.
--
2.39.5 (Apple Git-154)
