1#3.0.1 2#3.1.1 3#3.2.0 43.2.4 5#5745:37f59e65eb6c 65891:d8652709345d # introduce AVX 7#5893:24b4dc92c6d3 # merge 85895:997c2ef9fc8b # introduce FMA 9#5904:e1eafd14eaa1 # complex and AVX 105908:f8ee3c721251 # improve packing with ptranspose 11#5921:ca808bb456b0 # merge 12#5927:8b1001f9e3ac 135937:5a4ca1ad8c53 # New gebp kernel handling up to 3 packets x 4 register-level blocks 14#5949:f3488f4e45b2 # merge 15#5969:e09031dccfd9 # Disable 3pX4 kernel on Altivec 16#5992:4a429f5e0483 # merge 17before-evaluators 18#6334:f6a45e5b8b7c # Implement evaluator for sparse outer products 19#6639:c9121c60b5c7 20#6655:06f163b5221f # Properly detect FMA support on ARM 21#6677:700e023044e7 # FMA has been wrongly disabled 22#6681:11d31dafb0e3 23#6699:5e6e8e10aad1 # merge default to tensors 24#6726:ff2d2388e7b9 # merge default to tensors 25#6742:0cbd6195e829 # merge default to tensors 26#6747:853d2bafeb8f # Generalized the gebp apis 276765:71584fd55762 # Made the blocking computation aware of the l3 cache; Also optimized the blocking parameters to take into account the number of threads used for a computation 28#6781:9cc5a931b2c6 # generalized gemv 29#6792:f6e1daab600a # ensured that contractions that can be reduced to a matrix vector product 30#6844:039efd86b75c # merge tensor 316845:7333ed40c6ef # change prefetching in gebp 32#6856:b5be5e10eb7f # merge index conversion 33#6893:c3a64aba7c70 # clean blocking size computation 34#6898:6fb31ebe6492 # rotating kernel for ARM 356899:877facace746 # rotating kernel for ARM only 36#6904:c250623ae9fa # result_of 376921:915f1b1fc158 # fix prefetching change for ARM 386923:9ff25f6dacc6 # prefetching 396933:52572e60b5d3 # blocking size strategy 406937:c8c042f286b2 # avoid redundant pack_rhs 416981:7e5d6f78da59 # dynamic loop swapping 426984:45f26866c091 # rm dynamic loop swapping, adjust lhs's micro panel height to fully exploit L1 cache 436986:a675d05b6f8f # blocking heuristic: block on the rhs in L1 if the lhs fit in L1. 447013:f875e75f07e5 # organize a little our default cache sizes, and use a saner default L1 outside of x86 (10% faster on Nexus 5) 457015:8aad8f35c955 # Refactor computeProductBlockingSizes to make room for the possibility of using lookup tables 467016:a58d253e8c91 # Polish lookup tables generation 477018:9b27294a8186 # actual_panel_rows computation should always be resilient to parameters not consistent with the known L1 cache size, see comment 487019:c758b1e2c073 # Provide a empirical lookup table for blocking sizes measured on a Nexus 5. Only for float, only for Android on ARM 32bit for now. 497085:627e039fba68 # Bug 986: add support for coefficient-based product with 0 depth. 507098:b6f1db9cf9ec # Bug 992: don't select a 3p GEMM path with non-vectorizable scalar types, this hits unsupported paths in symm/triangular products code 517591:09a8e2186610 # 3.3-alpha1 527650:b0f3c8f43025 # help clang inlining 53#8744:74b789ada92a # Improved the matrix multiplication blocking in the case where mr is not a power of 2 (e.g on Haswell CPUs) 548789:efcb912e4356 # Made the index type a template parameter to evaluateProductBlockingSizes. Use numext::mini and numext::maxi instead of std::min/std::max to compute blocking sizes 558972:81d53c711775 # Don't optimize the processing of the last rows of a matrix matrix product in cases that violate the assumptions made by the optimized code path 568985:d935df21a082 # Remove the rotating kernel. 578988:6c2dc56e73b3 # Bug 256: enable vectorization with unaligned loads/stores. 589148:b8b8c421e36c # Relax mixing-type constraints for binary coefficient-wise operators 599174:d228bc282ac9 # merge 609212:c90098affa7b # Fix performance regression introduced in changeset 8aad8f35c955 619213:9f1c14e4694b # Fix performance regression in dgemm introduced by changeset 81d53c711775 62