From 6481443560a92d0a3a55a31807de0cd712cd4f88 Mon Sep 17 00:00:00 2001 From: Abseil Team Date: Mon, 6 Jun 2022 09:28:31 -0700 Subject: Optimize SwissMap for ARM by 3-8% for all operations https://pastebin.com/CmnzwUFN The key idea is to avoid using 16 byte NEON and use 8 byte NEON which has lower latency for BitMask::Match. Even though 16 byte NEON achieves higher throughput, in SwissMap it's very important to catch these Matches with low latency as probing on average happens at most once. I also introduced NonIterableMask as ARM has really great cbnz instructions and additional AND on scalar mask had 1 extra latency cycle PiperOrigin-RevId: 453216147 Change-Id: I842c50d323954f8383ae156491232ced55aacb78 --- absl/container/internal/raw_hash_set_test.cc | 24 ++++++++++++++---------- 1 file changed, 14 insertions(+), 10 deletions(-) (limited to 'absl/container/internal/raw_hash_set_test.cc') diff --git a/absl/container/internal/raw_hash_set_test.cc b/absl/container/internal/raw_hash_set_test.cc index c79f8641..dc6bc3d2 100644 --- a/absl/container/internal/raw_hash_set_test.cc +++ b/absl/container/internal/raw_hash_set_test.cc @@ -195,35 +195,39 @@ TEST(Group, Match) { } } -TEST(Group, MatchEmpty) { +TEST(Group, MaskEmpty) { if (Group::kWidth == 16) { ctrl_t group[] = {ctrl_t::kEmpty, CtrlT(1), ctrl_t::kDeleted, CtrlT(3), ctrl_t::kEmpty, CtrlT(5), ctrl_t::kSentinel, CtrlT(7), CtrlT(7), CtrlT(5), CtrlT(3), CtrlT(1), CtrlT(1), CtrlT(1), CtrlT(1), CtrlT(1)}; - EXPECT_THAT(Group{group}.MatchEmpty(), ElementsAre(0, 4)); + EXPECT_THAT(Group{group}.MaskEmpty().LowestBitSet(), 0); + EXPECT_THAT(Group{group}.MaskEmpty().HighestBitSet(), 4); } else if (Group::kWidth == 8) { ctrl_t group[] = {ctrl_t::kEmpty, CtrlT(1), CtrlT(2), ctrl_t::kDeleted, CtrlT(2), CtrlT(1), ctrl_t::kSentinel, CtrlT(1)}; - EXPECT_THAT(Group{group}.MatchEmpty(), ElementsAre(0)); + EXPECT_THAT(Group{group}.MaskEmpty().LowestBitSet(), 0); + EXPECT_THAT(Group{group}.MaskEmpty().HighestBitSet(), 0); } else { FAIL() << "No test coverage for Group::kWidth==" << Group::kWidth; } } -TEST(Group, MatchEmptyOrDeleted) { +TEST(Group, MaskEmptyOrDeleted) { if (Group::kWidth == 16) { - ctrl_t group[] = {ctrl_t::kEmpty, CtrlT(1), ctrl_t::kDeleted, CtrlT(3), - ctrl_t::kEmpty, CtrlT(5), ctrl_t::kSentinel, CtrlT(7), - CtrlT(7), CtrlT(5), CtrlT(3), CtrlT(1), - CtrlT(1), CtrlT(1), CtrlT(1), CtrlT(1)}; - EXPECT_THAT(Group{group}.MatchEmptyOrDeleted(), ElementsAre(0, 2, 4)); + ctrl_t group[] = {ctrl_t::kEmpty, CtrlT(1), ctrl_t::kEmpty, CtrlT(3), + ctrl_t::kDeleted, CtrlT(5), ctrl_t::kSentinel, CtrlT(7), + CtrlT(7), CtrlT(5), CtrlT(3), CtrlT(1), + CtrlT(1), CtrlT(1), CtrlT(1), CtrlT(1)}; + EXPECT_THAT(Group{group}.MaskEmptyOrDeleted().LowestBitSet(), 0); + EXPECT_THAT(Group{group}.MaskEmptyOrDeleted().HighestBitSet(), 4); } else if (Group::kWidth == 8) { ctrl_t group[] = {ctrl_t::kEmpty, CtrlT(1), CtrlT(2), ctrl_t::kDeleted, CtrlT(2), CtrlT(1), ctrl_t::kSentinel, CtrlT(1)}; - EXPECT_THAT(Group{group}.MatchEmptyOrDeleted(), ElementsAre(0, 3)); + EXPECT_THAT(Group{group}.MaskEmptyOrDeleted().LowestBitSet(), 0); + EXPECT_THAT(Group{group}.MaskEmptyOrDeleted().HighestBitSet(), 3); } else { FAIL() << "No test coverage for Group::kWidth==" << Group::kWidth; } -- cgit v1.2.3