GX_TF_RGBA8 texture decoder optimized with SSE2 producing a ~68% speed increase over reference C implementation.
TABified the entire document per NeoBrainX. :)
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@6706 8ced0084-cf51-0410-be5f-012b33b47a6e
~25% speed improvement in decoding GX_TF_RGB5A3 textures which aren't used very much. I thought it would help for movie playback but I misled myself. Video playback has nothing to do with this texture format.
Next I'll see if I can knock out some of these other texture decoders. Byte swizzling I'm sure can somehow be accomplished using _mm_unpacklo_epi8 trickery, so that'd be another big win I hope.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@6698 8ced0084-cf51-0410-be5f-012b33b47a6e
TextureDecoder.cpp: merged two functionally identical decode5A3RGBA and decode5A3rgba methods.
OpcodeDecoding.cpp and DLCache.cpp: optimization for GX_LOAD_XF_REG. The PSUHFB solution sounds better for SSSE3, but this is a small win for the default case.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@6692 8ced0084-cf51-0410-be5f-012b33b47a6e
use plain vertex arrays instead of VBOs to render in Opengl plugin as the nature of the data make VBOs slower. This must bring, depending on the implementation, a good speedup in opengl.
in my system now opengl and d3d9 have a difference of 1 to 5 fps depending of the game.
some cleanup and a little work pointing to future improvements in the way of rendering.
please test and check for any errors.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@6139 8ced0084-cf51-0410-be5f-012b33b47a6e
The GCC model for extended instructions like these is that you compile
with -msse3 etc. These affect code generation for whole compilation units,
so the idea is that you have a separate .c file for each instruction set
class and then indirect to the desired one at runtime.
Without e.g. -msse4.1, the GCC built-ins used by <foointrin.h> are not
available. However, in our specific case of compiling with -msse2 and
wanting to use SSE3.1 code, enough built-ins are available that we only
need to provide a little hack for pshufb.
Upgrading this to also use SSE4.1 instructions doesn't appear feasible
without a lot of undesirable duplication of GCC built-in functions and
headers, so we'd probably have to move to the GCC model of separate
source files for that.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@6014 8ced0084-cf51-0410-be5f-012b33b47a6e
Fix the decoder codepath when OpenCL is enabled and the DX11 plugin is used.
Added the DX11 plugin to the Dolphin project dependencies.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@5764 8ced0084-cf51-0410-be5f-012b33b47a6e
These codes don't work unless "_M_SSE=0x301", for SSSE3, or "_M_SSE=0x401", for SSE4.1, is defined as a preprocessor definition.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@5300 8ced0084-cf51-0410-be5f-012b33b47a6e
fixed fps display in the top bar, now it shows the real fps of the game.
some code clean up and some corrections to make everything work right in the reference renderer.
multiples xfb now is broken even in single core as is was not an error caused by dual core, i really dono where the error is,everything looks correct but if you test a game with multiples xfb or the ipl you will see the error.
ector if you can take a look at he code and throw me some ideas i'll thank you.
please test.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@5272 8ced0084-cf51-0410-be5f-012b33b47a6e
fixed a misbehavior in the clear code that causes depth clear problems in reference hardware (Intel as example).
add 6 parameters to optimize Safe Texture Cache:
SafeTextureCacheColorSamples, SafeTextureCacheIndexedSamples, SafeTextureCacheTlutSamples:
this 3 parameters gives the number of samples taken to calculate the final hash value, less samples = more speed, more samples = more accuracy
if 0 is specified the hash is calculated using all the data in the texture.
SafeTextureCacheColorMaxSize, SafeTextureCacheIndexedMaxSize, SafeTextureCacheTlutMaxSize:
this parameters limits the amount of data used for the hash calculation, it could appear as redundant but in some games is better to make a full hash of the first bytes instead of some samples of all the texture.
color, indexed, tlut : define the texture type, full color data, indexed, and the tlut memory.
the parameters are available in the config , no GUI at this time, if the test are OK will add it to the GUI.
if someone needs it will give more examples on how to configure the values for specific games.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@5116 8ced0084-cf51-0410-be5f-012b33b47a6e
sorry to tell but this will bring a speed drop, so let you decide if this change stay or not.( used the fastest open source hash algorithm i know)
do not apply full hashing to other format because it kills the performance.
for popular request added 9x SSAA believe me will kill your graphic card even if is the best but the image quality is exceptional.
as always please test and let me know the results.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@5034 8ced0084-cf51-0410-be5f-012b33b47a6e
chage the global format of the screenshots to bmp, is correctly supported by both plugins and is faster.
reverted the changes in safe texture cache, will try to make them more stable then commit them, this should fix compilation in linux and macand error introduced in MP games
corrected all the issues commented by ector, thanks for the comments alway is good that the code is revised by others to find missed spots.
please report any remaining issue to solve them.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@5001 8ced0084-cf51-0410-be5f-012b33b47a6e
first fixed scaling when updating backbuffer to make it friendly with encoders, now frame dumping must work without errors in any codec.
clean screenshot and frame dumping code now is more correct, faster and stable.
improve safe texture cache, improving the distribution of the hash algorithm, including tlut hash in the final hash of the texture, and making use of a 64 bit hash to make it more accurate.
clean a lot of code and corrected some missused vertex formats when drawing full screen quads.
and biggest change last:
implemented pseudo antialiasing: a image post-process algorithm that mimics antialiazing and is fare more easier to implement in this scenario.
you can change the intensity of the effect changing the values of the antialiasing combo. the right value depends on the game.
for example mkwii looks awesome with 8x.
please try all the changes and let me know the results.
if something is broken, please let me know and will fix it asap.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@5000 8ced0084-cf51-0410-be5f-012b33b47a6e
Advantages:
* Less confusion for users
* No need to build twice to make sure you didn't break something
* Easier to switch between the cores for testing
Disadvantages:
* None, as far as I can tell :) Maybe some extra code complexity, but not much.
Also break some include chains that caused <windows.h> to get included into everything, slowing down the build on Windows. There's more to do here though, there's still a lot of files that get it included that don't need it at all.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@4891 8ced0084-cf51-0410-be5f-012b33b47a6e
in videocommon little fix for the alpha test values, return to the original values as they are more accurate.
in D3D:
huge change in state management, now all the state management is centralized and redundant state changes are eliminated.
Fixed the overlapped viewport error in non ati cards:
the error was caused by this: when a viewport is defined larger than the current rendertarget, an error is thrown and the last valid viewport is used, this is the reference behavior, in ati cards if a larger viewport is defined, no eror is returned, the rendering is valid and is rendered using the projection defined by the viewport but limited to the rendertarget are, exactly like opengl or the GC hardware.
to solve this in reference drivers defined a large rendertarget (2x the size of the original) and proceed to render in a centered quad insithe the larger rendertarget, in this way larger viewports always falls inside a valid rendertarget size, the drawback of this is the waste of resources. it can be dynamized, depending or games or changed at runtime when a oversized viewport is detected, but i live that to future commits.
please test this and let me know the results.
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@4841 8ced0084-cf51-0410-be5f-012b33b47a6e
in D3D and Opengl:
fixed one nasty bug in texture loading where if a dynamic texture keeps his format but the tlut format is changed, the try or reloading the texture in the same texture could cause a hang if the size of the resulting texture is different than the original (size in bytes)
Applied a ugly temporal hack to the texture conversor to solve efb to ram misalignments and effect distortions.
in D3D:
Pseudo implementation of logic ops using basic blending: the first 8 operations are "good approximations", the remaining 8 are bullshit :) if someone have a better approximation to emulate this logic please let me know.
please test if i don't break anything in the process and test Mario kart wee you will get a nice surprise.:)
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@4656 8ced0084-cf51-0410-be5f-012b33b47a6e
james.jdunne