• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* -*- c++ -*- */
2 /*
3  * Copyright © 2020 Intel Corporation
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the "Software"),
7  * to deal in the Software without restriction, including without limitation
8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9  * and/or sell copies of the Software, and to permit persons to whom the
10  * Software is furnished to do so, subject to the following conditions:
11  *
12  * The above copyright notice and this permission notice (including the next
13  * paragraph) shall be included in all copies or substantial portions of the
14  * Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
22  * IN THE SOFTWARE.
23  */
24 
25 #ifndef BRW_IR_PERFORMANCE_H
26 #define BRW_IR_PERFORMANCE_H
27 
28 class fs_visitor;
29 
30 namespace brw {
31    /**
32     * Various estimates of the performance of a shader based on static
33     * analysis.
34     */
35    struct performance {
36       performance(const fs_visitor *v);
37       ~performance();
38 
39       analysis_dependency_class
dependency_classperformance40       dependency_class() const
41       {
42          return (DEPENDENCY_INSTRUCTIONS |
43                  DEPENDENCY_BLOCKS);
44       }
45 
46       bool
validateperformance47       validate(const backend_shader *) const
48       {
49          return true;
50       }
51 
52       /**
53        * Array containing estimates of the runtime of each basic block of the
54        * program in cycle units.
55        */
56       unsigned *block_latency;
57 
58       /**
59        * Estimate of the runtime of the whole program in cycle units assuming
60        * uncontended execution.
61        */
62       unsigned latency;
63 
64       /**
65        * Estimate of the throughput of the whole program in
66        * invocations-per-cycle units.
67        *
68        * Note that this might be lower than the ratio between the dispatch
69        * width of the program and its latency estimate in cases where
70        * performance doesn't scale without limits as a function of its thread
71        * parallelism, e.g. due to the existence of a bottleneck in a shared
72        * function.
73        */
74       float throughput;
75 
76    private:
77       performance(const performance &perf);
78       performance &
79       operator=(performance u);
80    };
81 }
82 
83 #endif
84