forked from LLNL/zfp
-
Notifications
You must be signed in to change notification settings - Fork 0
/
VERSIONS
161 lines (103 loc) · 5.78 KB
/
VERSIONS
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
zfp 0.5.0, February 29, 2016
- Modified CODEC to more efficiently encode blocks whose values are all
zero or are smaller in magnitude than the absolute error tolerance.
This allows representing "empty" blocks using only one bit each. This
version is not backwards compatible with prior zfp versions.
- Changed behavior of zfp_compress and zfp_decompress to not automatically
rewind the bit stream. This makes it easier to concatenate multiple
compressed bit streams, e.g. when compressing vector fields or multiple
scalars together.
- Added functions for compactly encoding the compression parameters
and field meta data, e.g. for producing self-contained compressed
streams. Also added functions for reading and writing a header
containing these parameters.
- Changed the zfp example program interface to allow reading and writing
compressed streams, optionally with a header. The zfp tool can now be
used to compress and decompress files as a stand alone utility.
zfp 0.4.1, December 28, 2015
- Fixed bug that caused segmentation fault when compressing 3D arrays
whose dimensions are not multiples of four. Specifically, arrays of
dimensions nx * ny * nz, with ny not a multiple of four, were not
handled correctly.
- Modified examples/fields.h to ensure standard compliance. Previously,
C99 support was needed to handle the hex float constants, which are
not supported in C++98.
- Added simple.c as a minimal example of how to call the compressor.
- Changed compilation of diffusion example to output two executables:
one with and one without compression.
zfp 0.4.0, December 5, 2015
- Substantial changes to the compression algorithm that improve PSNR
by about 6 dB and speed by a factor of 2-3. These changes are not
backward compatible with previous versions of zfp.
- Added support for 31-bit and 63-bit integer data, as well as shorter
integer types.
- Rewrote compression codec entirely in C to make linking and calling
easier from other programming languages, and to expose the low-level
interface through C instead of C++. This necessitated significant
changes to the API as well.
- Minor changes to the C++ compressed array API, as well as major
implementation changes to support the C library. The namespace and
public types are now all in lower case.
- Deprecated support for general fixed-point decorrelating transforms
and slimmed down implementation.
- Added new examples for evaluating the throughput of the (de)compressor
and for compressing grayscale images in the pgm format.
- Added FAQ.
zfp 0.3.2, December 3, 2015
- Fixed bug in Array::get() that caused the wrong cached block to be
looked up, thus occasionally copying incorrect values back to parts
of the array.
zfp 0.3.1, May 6, 2015
- Fixed rare bug caused by exponent underflow in blocks with no normal
and some denormal numbers.
zfp 0.3.0, March 3, 2015
- Modified the default decorrelating transform to one that uses only
additions and bit shifts. This new transform, in addition to being
faster, also has some theoretical optimality properties and tends to
improve rate distortion.
- Added compile-time support for parameterized transforms, e.g. to
support other popular transforms like DCT, HCT, and Walsh-Hadamard.
- Made forward transform range preserving: (-1, 1) is mapped to (-1, 1).
Consequently Q1.62 fixed point can be used throughout.
- Changed the order in which bits are emitted within each bit plane
to be more intelligent. Group tests are now deferred until they
are needed, i.e. just before the value bits for the group being
tested. This improves the quality of fixed-rate encodings, but
has no impact on compressed size.
- Made several optimizations to improve performance.
- Added floating-point traits to reduce the number of template
parameters. It is now possible to declare a 3D array as
Array3<float>, for example.
- Added functions for setting the array scalar type and dimensions.
- Consolidated several header files.
- Added testzfp for regression testing.
zfp 0.2.1, December 12, 2014
- Added Win64 support via Microsoft Visual Studio compiler.
- Fixed broken support for IBM's xlc compiler.
- Made several minor changes to suppress compiler warnings.
- Documented expected output for the diffusion example.
zfp 0.2.0, December 2, 2014
- The compression interface from zfpcompress was relocated to a
separate library, called libzfp, and modified to be callable from C.
This API now uses a parameter object (zfp_params) to specify array
type and dimensions as well as compression parameters.
- Several utility functions were added to simplify libzfp usage:
o Functions for setting the rate, precision, and accuracy.
Corresponding functions were also added to the Codec class.
o A function for estimating the buffer size needed for compression.
- The Array class functionality was expanded:
o Support for accessing the compressed bit stream stored with an
array, e.g. for offline compressed storage and for initializing
an already compressed array.
o Functions for dynamically specifying the cache size.
o The default cache is now direct-mapped instead of two-way
associative.
- Minor bug fixes:
o Corrected the value of the lowest possible bit plane to account for
both the smallest exponent and the number of bits in the significand.
o Corrected inconsistent use of rate and precision. The rate refers
to the number of compressed bits per floating-point value, while
the precision refers to the number of uncompressed bits. The Array
API was changed accordingly.
zfp 0.1.0, November 12, 2014
- Initial beta release.