Mercurial > hg > audiostuff

comparison intercom/gsm/changes @ 2:13be24d74cd2

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
import intercom-0.4.1
author Peter Meerwald <pmeerw@cosy.sbg.ac.at>
date Fri, 25 Jun 2010 09:57:52 +0200
parents
children
comparison
equal deleted inserted replaced
1:9cadc470e3da 2:13be24d74cd2
1 File "changes"
2 ~~~~~~~~~~~~~~
3
4 LOG of the changes introduced by Simao and summary of results.
5
6 1. first of all, file names had to fit in DOS limitations. Some names had
7 to be truncated, and the mapping is as follows:
8 gsm_create.c -> gsm_crea.c
9 gsm_decode.c -> gsm_deco.c
10 gsm_destroy.c -> gsm_dest.c
11 gsm_encode.c -> gsm_enco.c
12 gsm_explode.c -> gsm_expl.c
13 gsm_implode.c -> gsm_impl.c
14 gsm_option.c -> gsm_opti.c
15 gsm_print.c -> gsm_prin.c
16 long_term.c -> long_ter.c
17 preprocess.c -> preproce.c
18 short_term.c -> short_te.c
19
20 2. gsm.h had added prototypes for all the existing functions
21
22 3. add.c had to be changed to make add_test.c work
23
24 4. private.h had to be changed to use either true-function or the safe
25 pseudo-functions for the GSM_... macros. This was the main reason for
26 the errors that were happening.
27
28 5. undefs were preceeded by a check of their definition to avoid some
29 complaints from VAX-C
30
31 6. rpe.c had to be changed in a strange hacked "switch" by its "normal"
32 implementation for VMS (compilation error!) - the code is as it was
33 in comments.
34
35 7. To keep compatibility, the change in patch 3 to rpe.c
36 shall not be
37 #define STEP( i, H ) (e[ k + i ] * (long)H)
38 but
39 #define STEP( i, H ) (e[ k + i ] * (longword)H)
40
41 8. long_term.c (now long_ter.c) had to have added some castings, such
42 as:
43 temp = gsm_norm( (longword)dmax << 16 );
44 ^^^^^^^^^^
45 because gsm_norm uses longs and dmax is short: (short)<<16 = 0 always!
46 Also, if MSDOS/__MSDOS__ is defined, then USE_TABLE_MUL is
47 #undef'd (see tests below).
48
49 9. other small changes that I don't recall from my notes. These may include
50 type castings, and other changes done in the desperate search for the bugs.
51
52
53 Simplifications:
54 ~~~~~~~~~~~~~~~~
55
56 Since I don't want a compress-like utility, I haven't put my hands in toast.
57 To ease my work, I've put all src, inc and add-test in one single directory
58 and simplified the makefile. I also added labels "all", "rpedemo" and "clean",
59 "anyway".
60
61
62 Extensions:
63 ~~~~~~~~~~~
64
65 I added a DCL for compiling on VMS using either gcc or vax-c, depending on
66 the comments. I have also added a borland bcc make file for the pc, as well
67 as a gcc makefile for the pc.
68
69 Added a demo program and 2 driving routines (currently embedded in the demo
70 program). With them, interfacing with the test sequence file is
71 straight-forward.
72
73
74 Test results:
75 ~~~~~~~~~~~~~
76
77 the bcc-compiled version processed correctly all the 5 test vectors for
78 the following conditions:
79
80 (SASR always defined)
81
82 USE_FLOAT_MUL undefined
83 USE_FLOAT_MUL defined and FAST undefined
84 USE_FLOAT_MUL and FAST defined
85
86 bcc failed compilation when USE_FLOAT_MUL undefined and USE_TABLE_MUL defined
87 because the table matrix is greater than 64kbytes! (No way out!)
88
89 When compiled by bcc with both USE_FLOAT_MUL and FAST undefined, bcc
90 failed passing the test sequences.
91
92 Besides compilation worked for USE_FLOAT_MUL and FAST defined, in fact, all
93 the runs with the test sequences were made in the "compatible mode", i.e.,
94 with the state variable structure fuield "fast" always set as 0. Therefore,
95 it was not tested. Although, it is known from the original authors that
96 the fast mode does not work in Unix (32bit), hence it is very likely not
97 to work with DOS either!
98
99 NeedFunctionPrototypes was always defined for bcc, gcc, acc, and vax-c,
100 and undefined for unix cc.
101
102 Since gcc is the same implementation ported across platforms, it should
103 work as for SunOS, but it was not tested.
104
105 The program worked properly for the following environments (see summary below):
106 SunOS 4.3: cc, acc, gcc
107 MS/DOS: bcc, gcc (djcpp)
108 VAX/VMS: cc (Vax-c), gcc
109
110 Summary of the tests:
111 -------------------------------------------------------------------------
112 Environment Compiler Symbols defined (Note: SASR always def'd)
113 USE_FLOAT_MUL FAST USE_TABLE_MUL Worked?
114 -------------------------------------------------------------------------
115 SunOS 4.3: cc yes no - yes
116 yes yes - yes*
117 no - no yes
118 no - yes yes
119 acc same as for cc in SunOS 4.3 yes
120 gcc same as for cc in SunOS 4.3 yes
121 MS/DOS: bcc yes no - yes
122 yes yes - yes
123 no - no no
124 no - yes no
125 gcc Not tested other than yes/no/- yes
126 VAX/VMS: Vax-c same as for cc in SunOS 4.3 yes
127 gcc Not tested other than yes/no/- yes
128 -------------------------------------------------------------------------
129 * Note: all the runs were made in the compatible mode, therefore the
130 fast computation of the data was not performed!
131 -------------------------------------------------------------------------
132
133 Therefore, the safest approach is to compile always with USE_FLOAT_MUL
134 defined and with and FAST defined if appropriate for speed!
135
136 In Vax-C there were warnings relating to function names longer than 31
137 characters. No way to circumvent this, but it is not a problem because the
138 31 first letters generate unique names anyway. Because of this, link will
139 also complain!
140
141 The test sequence files had the byte order swaped in the Sun workstation
142 because of the Big Endian data orientation.
143
144 -- <simao@cpqd.ansp.br> -- 8/Apr/94
145

Repositories maintained by Peter Meerwald, pmeerw@pmeerw.net.