1 Movit 1.6.0, January 24th, 2018
3 - Support for effects that work as compute shaders. Compute shaders are
4 generally slower than fragment shaders for the same algorithm,
5 but allow some forms of communication between shader invocations
6 and have more flexible output, which can enable more efficient algorithms.
7 See effect.h for more details. Note that the fastest rendering API on
8 EffectChain is now to a texture if possible, not to an FBO. This will
9 only matter if the last effect is a compute shader.
11 - Movit now includes a compute shader implementation of DeinterlaceEffect,
12 which is automatically used instead of the fragment shader implementation
13 if your GPU and OpenGL driver supports it (in practice, this means on
14 all platforms except on macOS). The compute shader version is typically
15 20–80% faster than the fragment shader version, depending on your GPU
18 A compute shader implementation of ResampleEffect was written but
19 ultimately failed to be faster, and so is not included.
21 - Support for microbenchmarks of effects through the Google microbenchmarking
22 framework (optional). Currently, DeinterlaceEffect and ResampleEffect has
23 benchmarks; enable them by running the unit test with --benchmark (also try
26 - Effects can now explicitly request _not_ to have mipmaps, which means they
27 can do so without needing to request bounce and fiddling with the sampler
28 state. Note that this is an API change for effects.
30 - Movit now requires C++11, both to build and to #include the header files.
31 Support for SDL1 has been dropped; unit tests and the demo program now need
34 - Various smaller bugfixes and optimizations.
37 Movit 1.5.3, August 10th, 2017
42 Movit 1.5.2, July 5th, 2017
44 - The texture generation for ResampleEffect (when the size is changed)
47 - Work around an NVIDIA driver threading bug by aggressively caching VAOs.
50 Movit 1.5.1, May 29th, 2017
52 - Allow YCbCrInput to change input format after finalize.
54 - Some minor YCbCrInput bugfixes.
57 Movit 1.5.0, March 21st, 2017
59 - Support interleaved Y'CbCr input (4:4:4 in a single texture).
61 - Support 10-bit and 12-bit Y'CbCr, both for input and output. For planar,
62 these are supported packed in 16-bit ints; for interleaved, 10:10:10:2 is
63 supported. (Efficient conversion to and from v210, ie. 10-bit 4:2:2,
64 is possible using compute shaders, but Movit does not include support
65 for them at the current point.) Note that this now means the num_levels
66 flag in YCbCrFormat actually matters, although 0 will be interpreted
67 as 256 (8-bit) for the benefit of older applications.
69 - Limited support for having multiple Y'CbCr outputs from a chain.
71 - Allow changing the Y'CbCr output coefficients runtime, ie., after finalize.
73 - Fix an issue where the last pass would have been rendered with the sRGB
74 flag set, which confused Qt applications running in certain NVIDIA
78 Movit 1.4.0, November 5th, 2016
80 - Allow setting the intermediate format for chains, instead of hardcoding
81 it at 16-bit RGBA; advanced users can use this to e.g. ask for 8-bit
82 sRGB intermediates, reducing the amount of memory bandwidth needed
83 at the cost of reduced precision. Whether this is a good tradeoff or not
84 depends on the exact chain and your requirements.
86 - Fix an issue where a (cached) shader program could be used from multiple
87 threads at a time, causing the uniforms to contain unpredictable values.
89 - Make the error printed on check_error() slightly friendlier: Include the
90 enum if possible, and print it to stderr instead of stdout.
93 Movit 1.3.2, February 23rd, 2016
95 - Fix an issue with initialization in certain locales. Patch from
96 Jean-Baptiste Mardelle.
99 Movit 1.3.1, February 15th, 2016
101 - Fix an issue where certain effect chains (particularly involving
102 out-of-tree effects that return only a constant color) could cause
103 texture coordinates not to be set properly. (The new code probably
104 also has slightly lower OpenGL driver overhead.) Reported by
105 Christophe Thommeret.
108 Movit 1.3.0, January 31st, 2016
110 - Movit now requires GLSL 1.30 (so a driver from 2008 or later);
111 before, it claimed to support 1.10, but actually used 1.30 features.
112 Note that some OpenGL drivers, in particular on OS X, only supports
113 GLSL 1.30 (actually, 1.50) if you have an OpenGL core context.
115 - Add a deinterlacer, based on YADIF.
117 - Allow parallel output to RGBA and Y'CbCr textures at the same time.
119 - Make FlatInput and YCbCrInput support taking in external OpenGL
120 textures. Also allow them to change width/height after instantiation.
122 - Various compatibility and performance fixes. In particular,
123 fp16 conversion on older (non-Haswell) CPUs is much faster, due to
124 new conversion code by Fabian Giesen.
127 Movit 1.2.0, September 24th, 2015
129 - Movit now ships a version.h with a #define MOVIT_VERSION that
130 increases on every API change, including in git. (The number
131 itself carries no semantic meaning beyond this.) Movit 1.2.0
132 is not API or ABI compatible with 1.1.x, so there has been a
133 soname bump to libmovit.so.3.
135 - More flexible Y'CbCr input; in particular, a special input
136 for 4:2:2 interleaved data (UYVY), and support for semi-planar
137 input (Cb and Cr in the same texture, like in NV12). Note that
138 you should now also set the new “num_levels” member of
139 YCbYCbCrFormat to 256 (signifying 8-bit input), although
140 it is not used yet; this is a stepping stone towards supporting
141 10- and 12-bit Y'CbCr.
143 - Basic support for Y'CbCr _output_. Currently only 8-bit,
144 and only 4:4:4 (ie., you'll need to subsample yourself
145 afterwards). It is possible to split the output into multiple
146 textures, though, if you want a luma/chroma split (NV12-like)
149 - Support top-left origin for output. This is only really useful
150 if you are rendering directly into some memory area with top-left
151 origin; most users will get the expected behavior by using
152 bottom-left as before.
154 - Rework uniform handling for less OpenGL overhead. Note that
155 this means your effects now need to register uniforms in the
156 C++ code instead of declaring them in the .frag file.
158 - Make the PaddingEffect border subpixel-aware, and also support
159 an arbitrary (subpixel) border offset. This means that you can
160 compose PaddingEffect with an integral left/top offset
161 (use the new IntegralPaddingEffect for potentially more speed)
162 and ResampleEffect with a subpixel left/top offset (and zoom to
163 compensate) to get Lanczos interpolation for the pixels and a
164 simple bilinear interpolation for the border itself.
166 - Fix a bug that could cause very bad filter weights in
169 - Various performance improvements, in particular with regards to
170 CPU usage in ResampleEffect.
173 Movit 1.1.3, March 29th, 2015
175 - Fix accuracy issues in ResampleEffect, particularly when
176 zooming. Note that this might cost a few percent performance.
177 Reported by Christophe Thommeret.
179 - Make the number of BlurEffect taps configurable.
180 Based on patch by Christophe Thommeret.
182 - Multiple other bugfixes, some courtesy Dan Dennedy and
183 Christophe Thommeret. In particular, thread-safety fixes
184 related to locales and chain finalization.
187 Movit 1.1.2, August 12th, 2014
189 - Performance bugfix: Fix texture freelist behavior so that it's
190 LRU instead of the exact opposite. Patch by Christophe Thommeret.
192 - Performance bugfix: Correct the number of blur taps read in the
193 blur filter (it was reading about twice as many as it should).
194 Also found by Christophe Thommeret.
197 Movit 1.1.1, April 12th, 2014
199 - Fix an issue that could cause assertion failure in ResourcePool
200 when FBOs are reused, especially with NVIDIA's drivers.
203 Movit 1.1, April 10th, 2014
205 - NOTE: Movit now uses libepoxy rather than GLEW. This is needed because
206 of core context and GLES3 support (see below). Movit 1.1 should be
207 generally API-compatible with 1.0.3 (with the exception of the next
208 point), but is not ABI-compatible, so there has been a soname bump
211 - NOTE: The rules for using an EffectChain or ResourcePool in multiple
212 OpenGL contexts or threads have changed somewhat; see the comments in
215 - Movit will now work in OpenGL core contexts, and on GLES 3.0 or newer.
216 (GLES2 devices are not supported.) This also holds for all unit tests,
217 provided that you use SDL2 and not SDL1 (because SDL2 is needed to set
218 up such contexts). Note that the included demo application still requires
219 a classic OpenGL context.
221 - ResampleEffect can now do sub-pixel translate and/or zoom.
223 - LumaMixEffect now has an “invert” flag, as a convenience to e.g. change
224 a left-to-right wipe into a right-to-left one.
226 - Significant reduction in driver overhead, especially on NVIDIA drivers.
228 - Various smaller bugfixes and performance improvements.
231 Movit 1.0.3, March 16th, 2014
233 - Yet more build system tweaks mainly related to distribution packaging.
236 Movit 1.0.2, March 16th, 2014
238 - Make a few tweaks to “make install”, to make distributions' lives easier.
241 Movit 1.0.1, March 16th, 2014
243 - Fix so that shared libraries are built.
246 Movit 1.0, March 16th, 2014