• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2020 Collabora, Ltd.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21  * SOFTWARE.
22  *
23  * Authors (Collabora):
24  *   Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
25  */
26 
27 /* Index buffer min/max cache. We need to calculate the min/max for arbitrary
28  * slices (start, start + count) of the index buffer at drawtime. As this can
29  * be quite expensive, we cache. Conceptually, we just use a hash table mapping
30  * the key (start, count) to the value (min, max). In practice, mesa's hash
31  * table implementation is higher overhead than we would like and makes
32  * handling memory usage a little complicated. So we use this data structure
33  * instead. Searching is O(n) to the size, but the size is capped at the
34  * PANFROST_MINMAX_SIZE constant (so this is a tradeoff between cache hit/miss
35  * ratio and cache search speed). Note that keys are adjacent so we get cache
36  * line alignment benefits. Insertion is O(1) and in-order until the cache
37  * fills up, after that it evicts the oldest cached value in a ring facilitated
38  * by index.
39  */
40 
41 #include "pan_minmax_cache.h"
42 
43 bool
panfrost_minmax_cache_get(struct panfrost_minmax_cache * cache,unsigned start,unsigned count,unsigned * min_index,unsigned * max_index)44 panfrost_minmax_cache_get(struct panfrost_minmax_cache *cache, unsigned start,
45                           unsigned count, unsigned *min_index,
46                           unsigned *max_index)
47 {
48    uint64_t ht_key = (((uint64_t)count) << 32) | start;
49    bool found = false;
50 
51    if (!cache)
52       return false;
53 
54    for (unsigned i = 0; i < cache->size; ++i) {
55       if (cache->keys[i] == ht_key) {
56          uint64_t hit = cache->values[i];
57 
58          *min_index = hit & 0xffffffff;
59          *max_index = hit >> 32;
60          found = true;
61          break;
62       }
63    }
64 
65    return found;
66 }
67 
68 void
panfrost_minmax_cache_add(struct panfrost_minmax_cache * cache,unsigned start,unsigned count,unsigned min_index,unsigned max_index)69 panfrost_minmax_cache_add(struct panfrost_minmax_cache *cache, unsigned start,
70                           unsigned count, unsigned min_index,
71                           unsigned max_index)
72 {
73    uint64_t ht_key = (((uint64_t)count) << 32) | start;
74    uint64_t value = min_index | (((uint64_t)max_index) << 32);
75    unsigned index = 0;
76 
77    if (!cache)
78       return;
79 
80    if (cache->size == PANFROST_MINMAX_SIZE) {
81       index = cache->index++;
82       cache->index = cache->index % PANFROST_MINMAX_SIZE;
83    } else {
84       index = cache->size++;
85    }
86 
87    cache->keys[index] = ht_key;
88    cache->values[index] = value;
89 }
90 
91 /* If we've been caching min/max indices and we update the index
92  * buffer, that may invalidate the min/max. Check what's been cached vs
93  * what we've written, and throw out invalid entries. */
94 
95 void
panfrost_minmax_cache_invalidate(struct panfrost_minmax_cache * cache,struct pipe_transfer * transfer)96 panfrost_minmax_cache_invalidate(struct panfrost_minmax_cache *cache,
97                                  struct pipe_transfer *transfer)
98 {
99    /* Ensure there is a cache to invalidate and a write */
100    if (!cache)
101       return;
102 
103    if (!(transfer->usage & PIPE_MAP_WRITE))
104       return;
105 
106    unsigned valid_count = 0;
107 
108    for (unsigned i = 0; i < cache->size; ++i) {
109       uint64_t key = cache->keys[i];
110 
111       uint32_t start = key & 0xffffffff;
112       uint32_t count = key >> 32;
113 
114       /* 1D range intersection */
115       bool invalid = MAX2(transfer->box.x, start) <
116                      MIN2(transfer->box.x + transfer->box.width, start + count);
117       if (!invalid) {
118          cache->keys[valid_count] = key;
119          cache->values[valid_count] = cache->values[i];
120          valid_count++;
121       }
122    }
123 
124    cache->size = valid_count;
125    cache->index = 0;
126 }
127