|
Name |
|
Date |
Size |
#Lines |
LOC |
| .. | | 03-May-2022 | - |
| doc/ | H | 07-May-2022 | - | 4,473 | 4,256 |
| examples/ | H | 02-Nov-2019 | - | 850 | 607 |
| orc/ | H | 03-May-2022 | - | 42,915 | 34,576 |
| orc-test/ | H | 02-Nov-2019 | - | 2,087 | 1,610 |
| testsuite/ | H | 02-Nov-2019 | - | 19,496 | 15,322 |
| tools/ | H | 02-Nov-2019 | - | 2,092 | 1,875 |
| COPYING | H A D | 02-Nov-2019 | 3.1 KiB | 63 | 49 |
| README | H A D | 02-Nov-2019 | 3.5 KiB | 97 | 63 |
| RELEASE | H A D | 02-Nov-2019 | 16.7 KiB | 504 | 367 |
| ROADMAP | H A D | 02-Nov-2019 | 414 | 25 | 12 |
| TODO | H A D | 02-Nov-2019 | 427 | 34 | 16 |
| meson.build | H A D | 02-Nov-2019 | 5.7 KiB | 190 | 160 |
| orc-test.pc.in | H A D | 02-Nov-2019 | 368 | 13 | 11 |
| orc.m4 | H A D | 02-Nov-2019 | 2.1 KiB | 71 | 63 |
| orc.pc.in | H A D | 02-Nov-2019 | 383 | 17 | 12 |
README
1
2ORC - The Oil Runtime Compiler
3==============================
4
5(and OIL stands for Optimized Inner Loops)
6
7Orc is the sucessor to Liboil - The Library of Optimized Inner Loops.
8
9Orc is a library and set of tools for compiling and executing
10very simple programs that operate on arrays of data. The "language"
11is a generic assembly language that represents many of the features
12available in SIMD architectures, including saturated addition and
13subtraction, and many arithmetic operations.
14
15At this point, developers interested in using Orc should look at the
16examples and try out a few Orc programs in an experimental branch
17of their own projects. And provide feedback on how it works. There
18will likely be some major changes in ease of use from a developer's
19perspective over the next few releases.
20
21The 0.4 series of Orc releases will be API and ABI compatible, and
22will be incompatible with the 0.5 series when it comes out.
23
24
25Features:
26
27 - Users can create, compile, and run simple programs that use the
28 vector extensions of the CPU, all directly from an application.
29
30 - Users can compile Orc programs to assembly source code to be
31 compiled and used without linking against the Orc library.
32
33 - The generic assembly language can be extended by an application
34 by adding new opcodes.
35
36 - An application can add rules for converting existing or new opcodes
37 to binary code for a specific target.
38
39 - Current targets: SSE, MMX, MIPS, Altivec, NEON, and TI C64x+.
40 (The c64x target only produces source code.)
41
42 - Programs can optionally be emulated, which is useful for testing, or
43 if no rules are available to convert Orc opcodes to executable code.
44
45
46More information:
47
48 Web : http://gstreamer.freedesktop.org/modules/orc.html
49 Download : http://gstreamer.freedesktop.org/data/src/orc/
50 Documentation : http://gstreamer.freedesktop.org/data/doc/orc/
51
52
53Questions and Answers:
54
55 - Q: Why not let gcc vectorize my code?
56
57 A: Two reasons: first, since Orc's assembly language is much more
58 restrictive than C, Orc can generate better code than gcc, and
59 second, Orc can generate code for functions you define at runtime.
60 Many algorithms require gluing together several stages of operations,
61 and if each stage has several options, the total amount of code to
62 cover all combinations could be inconveniently large.
63
64 - Q: Why not use compiler intrinsics for SIMD code?
65
66 A: Compiler intrinsics only work for one target, and need to be
67 hand written. Plus, some compilers are very picky about source
68 code that uses intrinsics, and will silently produce slow code.
69 And, of course, you can't compile intrinsics at runtime.
70
71 - Q: How big is the Orc library?
72
73 A: For embedded users, the orc-backend meson option can
74 be used to disable irrelvant targets. Compiled with only one target
75 (SSE), the library size is about 150 kB uncompressed, or 48 kB
76 compressed. The goal was to keep the uncompressed size under
77 about 100 kB (but that failed!). A typical build with all targets
78 and the full ABI is around 350 kB.
79
80
81Caveats (Known Bugs):
82
83 - ?
84
85
86Future directions (Possibly outdated):
87
88 - Addition of more complex loop control and array structures.
89
90 - Addition of an option to compile the Orc library with only the
91 runtime features for a single target, e.g., for embedded systems.
92
93 - Addition of rewrite rules, which convert an instruction that cannot
94 be converted to binary code into a series of instructions that can.
95 This is necessary since assembly instructions on most targets do
96 not cover all the features of the Orc assembly language.
97