From 2f4a7747025cd3288453fdabd520638e37e3633c Mon Sep 17 00:00:00 2001 From: Nathan Bossart Date: Mon, 18 Mar 2024 10:44:08 -0500 Subject: [PATCH v4 1/3] pg_lfind32(): Optimize processing remaining elements. Discussion: https://postgr.es/m/20231129171526.GA857928%40nathanxps13 --- src/include/port/pg_lfind.h | 42 +++++++++++++++++++++++++++++++++---- 1 file changed, 38 insertions(+), 4 deletions(-) diff --git a/src/include/port/pg_lfind.h b/src/include/port/pg_lfind.h index b8dfa66eef..bef0e2d5be 100644 --- a/src/include/port/pg_lfind.h +++ b/src/include/port/pg_lfind.h @@ -95,15 +95,16 @@ pg_lfind32(uint32 key, uint32 *base, uint32 nelem) /* * For better instruction-level parallelism, each loop iteration operates - * on a block of four registers. Testing for SSE2 has showed this is ~40% - * faster than using a block of two registers. + * on a block of registers. We first do as much processing as possible + * with a block of 4 registers, then we try to process what remains with a + * block of 2 registers. */ const Vector32 keys = vector32_broadcast(key); /* load copies of key */ const uint32 nelem_per_vector = sizeof(Vector32) / sizeof(uint32); - const uint32 nelem_per_iteration = 4 * nelem_per_vector; + uint32 nelem_per_iteration = 4 * nelem_per_vector; /* round down to multiple of elements per iteration */ - const uint32 tail_idx = nelem & ~(nelem_per_iteration - 1); + uint32 tail_idx = nelem & ~(nelem_per_iteration - 1); #if defined(USE_ASSERT_CHECKING) bool assert_result = false; @@ -157,6 +158,39 @@ pg_lfind32(uint32 key, uint32 *base, uint32 nelem) return true; } } + + /* + * Try processing the remaining elements using 2 registers instead of 4. + */ + nelem_per_iteration = 2 * nelem_per_vector; + tail_idx = nelem & ~(nelem_per_iteration - 1); + + for (; i < tail_idx; i += nelem_per_iteration) + { + Vector32 vals1, + vals2, + result1, + result2, + result; + + /* load the next block into 2 registers */ + vector32_load(&vals1, &base[i]); + vector32_load(&vals2, &base[i + nelem_per_vector]); + + /* compare each value to the key */ + result1 = vector32_eq(keys, vals1); + result2 = vector32_eq(keys, vals2); + + /* combine the results into a single variable */ + result = vector32_or(result1, result2); + + /* see if there was a match */ + if (vector32_is_highbit_set(result)) + { + Assert(assert_result); + return true; + } + } #endif /* ! USE_NO_SIMD */ /* Process the remaining elements one at a time. */ -- 2.25.1