 Eugene Kozlov
|
c93d95854d
Add helper getters for Matrix3, Matrix3x4, Matrix4. Minor style and typo fixes.
|
8 years ago |
 urho3d-travis-ci
|
ee054a1507
Travis CI: bump copyright to 2017.
|
9 years ago |
 Yao Wei Tjong 姚伟忠
|
c675421464
Normal comment to doxygen comment using regex + manual correction.
|
9 years ago |
|
urho3d-travis-ci
|
c4f6f315ff
Travis CI: bump copyright to 2016.
|
10 years ago |
 Jukka Jylänki
|
811712e141
Work around VS2010 compiler bug with SSE and copy construction/assignment misalignment. Closes #1044.
|
10 years ago |
|
Jukka Jylänki
|
04dbf71e71
Work around bug #1053 (https://connect.microsoft.com/VisualStudio/feedback/details/2053175) by replacing the SSE1 specific _mm_ucomige_ss() with SSE2 specific _mm_cvtsi128_si32(_mm_castps_si128(c)) == -1. Given that URHO3D_SSE is targeting SSE2, this workaround has no drawbacks. Closes #1053.
|
10 years ago |
|
Jukka Jylänki
|
e28a678914
Optimize away some uses of _MM_TRANSPOSE4_PS(), to remove two shuffles, two movs and replace shufps with the lighter movehl/lh and unpacks. Also do vector loads and stores manually in BoundingBox::Transformed() to avoid VS2015 from generating dirty asm code for the function. This has a few percent improvement on the heavy BoundingBox::Transformed() function call.
|
10 years ago |
 Lasse Öörni
|
af5a14c3d9
Mark potentially dangerous math class constructors explicit. Closes #947.
|
10 years ago |
|
Jukka Jylänki
|
4ffff7ab4c
SSE optimize Matrix3x4 construction from translation, rotation and scale.
|
10 years ago |
|
Jukka Jylänki
|
3ac2cb6e37
Avoid depending on 64-bit only function _mm_cvtsi128_si64() in Matrix operator == to allow targeting 32-bit as well.
|
10 years ago |
|
Jukka Jylänki
|
0107a0bf75
SSE optimize Urho3D/Math/Matrix3x4.h and Urho3D/Math/Matrix4.h. These yield a ~3% improvement in 06_SkeletalAnimation sample with 2000 Jacks in it.
|
10 years ago |
|
Jukka Jylänki
|
3ad9ba306d
SSE optimize matrix-matrix multiplications when URHO3D_SSE is enabled. These constitute 7.82% of 06_SkeletalAnimation sample with 2000 Jacks in it. After SSE optimizations applied, matrix multiplication takes only 2.13% of total time in AMD CodeXL profile. This change also seems to have the effect that in scalar version, VS2015 did not dare to inline Matrix3x4 multiplication in AnimatedModel::UpdateSkinning(), but after SSE, the instruction count is small enough that it happily does so.
|
10 years ago |
|
Yao Wei Tjong 姚伟忠
|
fa77a456f8
Reformat Urho3D source files to get rid of the indents on empty lines.
|
10 years ago |
|
Lasse Öörni
|
bb2ebf7568
Bump copyright to 2015.
|
11 years ago |
|
Yao Wei Tjong 姚伟忠
|
b51e419c88
Change impl. file to include the header file from corresponding subdir.
|
11 years ago |
|
Yao Wei Tjong 姚伟忠
|
bcf7c94ec8
Rename Source/Engine to Source/Urho3D.
|
11 years ago |
