From f8070cc44a1dd2482b04d81994062bd65ea8b307 Mon Sep 17 00:00:00 2001 From: Thomas Schatzl Date: Fri, 1 Apr 2016 11:57:58 +0200 Subject: [PATCH] 8148099: Improve memory access to FromCardCache during GC Transpose the FromCardCache data array so the access during GC is linear instead of element-by-element with stride. Reviewed-by: mgerdin, brutisso --- hotspot/src/share/vm/gc/g1/g1FromCardCache.cpp | 6 +++--- hotspot/src/share/vm/gc/g1/g1FromCardCache.hpp | 11 +++++++---- 2 files changed, 10 insertions(+), 7 deletions(-) diff --git a/hotspot/src/share/vm/gc/g1/g1FromCardCache.cpp b/hotspot/src/share/vm/gc/g1/g1FromCardCache.cpp index 41b129c2111..75bf9a85861 100644 --- a/hotspot/src/share/vm/gc/g1/g1FromCardCache.cpp +++ b/hotspot/src/share/vm/gc/g1/g1FromCardCache.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2001, 2015, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2001, 2016, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -37,8 +37,8 @@ void G1FromCardCache::initialize(uint num_par_rem_sets, uint max_num_regions) { guarantee(_cache == NULL, "Should not call this multiple times"); _max_regions = max_num_regions; - _cache = Padded2DArray::create_unfreeable(num_par_rem_sets, - _max_regions, + _cache = Padded2DArray::create_unfreeable(_max_regions, + num_par_rem_sets, &_static_mem_size); invalidate(0, _max_regions); diff --git a/hotspot/src/share/vm/gc/g1/g1FromCardCache.hpp b/hotspot/src/share/vm/gc/g1/g1FromCardCache.hpp index 67c8ec65a52..8c0864609ae 100644 --- a/hotspot/src/share/vm/gc/g1/g1FromCardCache.hpp +++ b/hotspot/src/share/vm/gc/g1/g1FromCardCache.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2013, 2015, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2013, 2016, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -32,8 +32,11 @@ // a per-region and per-thread basis. class G1FromCardCache : public AllStatic { private: - // Array of card indices. Indexed by thread X and heap region to minimize + // Array of card indices. Indexed by heap region (rows) and thread (columns) to minimize // thread contention. + // This order minimizes the time to clear all entries for a given region during region + // freeing. I.e. a single clear of a single memory area instead of multiple separate + // accesses with a large stride per region. static int** _cache; static uint _max_regions; static size_t _static_mem_size; @@ -58,11 +61,11 @@ class G1FromCardCache : public AllStatic { } static int at(uint worker_id, uint region_idx) { - return _cache[worker_id][region_idx]; + return _cache[region_idx][worker_id]; } static void set(uint worker_id, uint region_idx, int val) { - _cache[worker_id][region_idx] = val; + _cache[region_idx][worker_id] = val; } static void initialize(uint num_par_rem_sets, uint max_num_regions);