DoubangoTelecom/compv issues and pull requests

#100 - Rename compv_convlt to compv_gblur and make ComVMathConvlt the default convolution impl

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#99 - BGR -> GRAY no math needed

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#98 - ymmMaskToExtractFirst64Bits and ymmMaskToExtractFirst128Bits not needed in asm

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#97 - ASM: loop unroll for Matrix::mulGA

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#96 - Add AVX version for MatrixMaxAbsOffDiagSymm

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#95 - mulABt_4x4

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#94 - mulAB_4x4

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#93 - mulABt_3x3

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#92 - mulAB_3x3

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#91 - Use AsyncTask11 everywhere

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#90 - ImageConv many x64 versions are missing

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#89 - add fast transpose9, mulGA9, mulAB9... (homography)

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#88 - Track "size_of_float_is4" and replace "float" with "compv_float32_t"

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#87 - Split intrinsic files per caps (_sse2.cxx, _ssse3.cxx, ....) to allow conditional compilation

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#86 - jacobiAngles() do not compute cos(theta) and sin(theta)

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#85 - Eigenvalue computation for 3x3

Issue - State: open - Opened by DoubangoTelecom over 8 years ago - 1 comment

#84 - Add support for fast_rand

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#83 - Add support for transposeS, transposeD, transposeA

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#82 - Add support for PRNG using Mersenne Twister

Issue - State: open - Opened by DoubangoTelecom over 8 years ago - 1 comment

#81 - Matrix inverse, add support for LU, Cholesky and QR decompositions

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#80 - Add support for QR decomposition

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#79 - Add support for LU decomposition

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#78 - Mem::cpy try with uint64_t to see

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#77 - Convlt1_verthz_float32 use dp

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#76 - Use CompVPtrXXXX everywhere

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#75 - Use reference (&) instead of pointer (*) in return args

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#74 - Replace sizeof(float) == 4 with compv_float32_t

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#73 - COMPV_ASSERT must break on release mode

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#72 - Some AVX2 functions should be AVX1

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#71 - Rectangular matrixes, optimize using MCOP

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#70 - Dimensional reduction for the covariance matrix (Eigen values/vectors)

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#69 - FAST detector, pushing interest points: optimization issue

Issue - State: open - Opened by DoubangoTelecom over 8 years ago - 1 comment

#68 - Samples

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#67 - When testing mode, memZero after malloc() to make sure to have same MD5 result

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#66 - Decreasing CPU usage / Speedup

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#65 - Add list of customers on doubango.org

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#64 - Disable x86 ASM when X64 version is available

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#63 - add support for fxp brief decription

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#62 - Add support for fxp convolution

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#61 - Promote float to compv_float32_t when sizeof(float) != 4

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#60 - convlt1_verthz: VS2013 intrin faster than our ASM (not the case for Intel C++)

Issue - State: open - Opened by DoubangoTelecom over 8 years ago - 1 comment

#59 - Add support for Intel® Integrated Performance Primitives (Intel® IPP)

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#58 - adds support for Short Vector Math Library (SVML) Functions

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#57 - Call vzeroupper after COMPV_YASM_RESTORE_YMM

Issue - State: closed - Opened by sarandogou over 8 years ago

#56 - Description, uses capacity() and size() to avoid allocating memory for each process

Issue - State: closed - Opened by sarandogou over 8 years ago

#55 - Remove ComVDescription and replace it with CompVArray

Issue - State: closed - Opened by sarandogou over 8 years ago

#54 - Win64 calling convention: stack must be aligned on 16bytes after params allocation -> SystemV redzone

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#53 - getThreadIdxForNextToCurrentCore not correct, must use getThreadIdxCurrent() instead of getCoreId()

Issue - State: open - Opened by sarandogou over 8 years ago

#52 - do not use -march=native on xcode

Issue - State: open - Opened by sarandogou over 8 years ago

#51 - Apple LLVM: fatal error: error in backend: Cannot select: intrinsic %llvm.x86.avx.vzeroupper

Issue - State: closed - Opened by sarandogou over 8 years ago

#50 - Add support for frame-based multi-threading

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago - 1 comment

#49 - RGBA->I420 with odd width doesn't work or YUV player is buggy

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#48 - ImageConv, add x64 versions

Issue - State: open - Opened by DoubangoTelecom over 8 years ago - 1 comment

#47 - Replace COMPV_YASM_SAVE_XMM with COMPV_YASM_SAVE_YMM in AVX code

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#46 - use vzeroall instead of vzeroupper and omit vpxor ymm, ymm

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#45 - Add FastStrenghts32 implementation using AVX-only code and VPHMINPOSUW

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#44 - Use "__stdcall" on all ASM extern functions to avoid surprises

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#43 - Add unit tests with resolution change

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#42 - Make memory checking thread-safe

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#41 - cache optimisations and many other stuffs (e.g cupid) must be tested on AMD CPUs

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#40 - Make box sorting multi-threaded

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#39 - Cache alignment issue on all SIMD code

Issue - State: open - Opened by DoubangoTelecom over 8 years ago - 1 comment

#38 - On windows, save XMM6-n registers for avx

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#37 - Compute function pointers onces and update when CPU flags is update

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#36 - replace all cmovs with setcc when possible. The later is supported on all i386

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#35 - False sharing issue on interest points vector for FAST

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#34 - CompVImage::wrap is very slooow

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#33 - RGBA->I420 C++ version doesn't work

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#32 - COMPV_INTERLEAVE_I8_XMM_SSE2, reverse line 2/3

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#31 - Use dr memory for memory checks (works on Windows unlike valgrind)

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#30 - Use valgrind for profiling

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#29 - Products

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#28 - COMPV_YASM_SAVE_RSP_TO_MM0, COMPV_YASM_RESTORE_RSP_FROM_MM0

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#27 - compv_popcnt16 is sloow because of the test.

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#26 - Use letsencrypt.org to secure compv.org

Issue - State: open - Opened by DoubangoTelecom over 8 years ago

#25 - Adds support for Intel C++ compiler

Issue - State: closed - Opened by DoubangoTelecom over 8 years ago

#24 - Do not use WaitForSingleObject for multithreading: performance issues

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#23 - add support video feature detection and description: motion estimation and compensation

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#22 - Always use 'lea edi, [edi +4]' instead of 'add edi, 4'

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#21 - COMPV_XXXXX_MIN_SAMPLES_PER_THREAD macros must depends on the CPU hz

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago - 1 comment

#20 - Use SpinWait in asynTask

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#19 - Review all x86/x64 assemply using Intel Optimization manual

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#18 - Add support for Apple's accelerate framework

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#17 - implement fastArctan2 for embedded devices

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#16 - Add support for PROSAC

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#15 - Create Doubango feature detector+descriptor

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago - 12 comments

#14 - Add support for DCCI interpolation (preserve edges)

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#13 - FAST9 and FAST12 using ragel state machine

Issue - State: closed - Opened by DoubangoTelecom almost 9 years ago

#12 - Convert SSSE3 functions to SSE2 when possible

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#11 - FAST detector: add support for both SAD and maxThreshold in NMS

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#10 - Adds support for FMA3 instruction set

Issue - State: closed - Opened by DoubangoTelecom almost 9 years ago

#9 - Align the data and stride on the cache line size if it's <128 and pow2(must be but who knows).

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#8 - Most of the asm code could be opti for x64 (remove, pushs, remove allocation, remove saving/restoring xmm/ymm registers...)

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#7 - replace _mm256_permute4x64_epi64 with _mm256_shuffle_epi8

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#6 - AVX-SSE Transition Penalties

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago - 1 comment

#5 - CompVThread::setAffinity not tested on Android

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#4 - CompVThread::setAffinity not implemented on OSX and iOS

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#3 - reallocAligned not optimized (free old then alloc new)

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago

#2 - reallocAligned doesn't copy old memory to the newly allocated one

Issue - State: closed - Opened by DoubangoTelecom almost 9 years ago

#1 - Find efficient way to malloc, realloc, and calloc aligned memory on all platforms

Issue - State: open - Opened by DoubangoTelecom almost 9 years ago - 1 comment

GitHub / DoubangoTelecom/compv issues and pull requests