Gitiles
Code Review Sign In
gerrit-public.fairphone.software / platform / external / XNNPACK / a49e41f0003a6a853607f271196b415a596216c5
  1. a49e41f QU8 4x16C4 NEON Dot Product GEMM/IGEMM microkernels for Cortex A55r1 by Frank Barchard · 3 years, 1 month ago
  2. 8589ecd QS8 IGEMM use x11 for params, x10 for a3 and x0 for cn_stride by Frank Barchard · 3 years, 1 month ago
  3. 4810905 Leverage v128.const WAsm SIMD instruction by Marat Dukhan · 3 years, 1 month ago
  4. 8dc106e QC8/QS8/QU8 GEMM/IGEMM WAsm SIMD microkernels using i32x4.dot_i16x8_s instruction by Marat Dukhan · 3 years, 1 month ago
  5. feee77f Leverage f32x4.nearest, f32x4.floor, f32x4.ceil, f32x4.trunc WAsm SIMD instructions by Marat Dukhan · 3 years, 1 month ago
  6. 5d27a7b Leverage f32x4.nearest, f32x4.floor, f32x4.ceil, f32x4.trunc WAsm SIMD instructions by Marat Dukhan · 3 years, 1 month ago
  7. 0a3093c QU8 vadd neon use x32 instead of x8 by Frank Barchard · 3 years, 1 month ago
  8. 7da8b02 Q8 dwconv switch from 8x25 to 16x25 by Frank Barchard · 3 years, 1 month ago
  9. e252f92 End-to-end benchmarks on QC8 MobileNet v1/v2 models by Marat Dukhan · 3 years, 1 month ago
  10. 0d06573 dwconv Q8 switch from 8x9 to 16x9 tile. by Frank Barchard · 3 years, 1 month ago
  11. aae722a Run template generators in parallel by Frank Barchard · 3 years, 1 month ago
  12. 1215c9a QS8 NEON GEMM microkernels use rewind instead of reload by Frank Barchard · 3 years, 1 month ago
  13. 6b30b73 Remainder branch move before label. by Frank Barchard · 3 years, 1 month ago
  14. fec7363 QU8 C4 4x8 rename registers to avoid 3 push/pops. by Frank Barchard · 3 years, 1 month ago
  15. 8b69802 Enable QU8 C4 NEON Dot Product GEMM/IGEMM microkernels for Cortex A55r1 by Frank Barchard · 3 years, 1 month ago
  16. ca4c68e QU8 C4 NEON Dot Product GEMM/IGEMM microkernels for Cortex A55r1 by Frank Barchard · 3 years, 1 month ago
  17. 2d33a9b Document WebAssembly SIMD as non-experimental platform by Marat Dukhan · 3 years, 1 month ago
  18. 56f157c Relabel branches for quantized assembly ARM microkernels by Frank Barchard · 3 years, 1 month ago
  19. 0c76422 QU8 NEON Assembly remove channel wise by Frank Barchard · 3 years, 1 month ago
  20. 408f153 Enable QU8 4x16 C4 NEON Assembly Dot Product GEMM/IGEMM microkernels by Frank Barchard · 3 years, 1 month ago
  21. 4066898 QU8 4x16 C4 NEON Assembly Dot Product GEMM/IGEMM microkernels by Frank Barchard · 3 years, 1 month ago
  22. 3b9b4bc Fix VCLAMP parameter initialization functions on pre-NEON ARM by Marat Dukhan · 3 years, 1 month ago
  23. a38bf33 QU8 4x8c4 rewind params with SUB by Frank Barchard · 3 years, 1 month ago
  24. b48f367 QU8 4x8 C4 NEON reload params during subtract by Frank Barchard · 3 years, 1 month ago
  25. 073185e QU8 4x8 C4 NEON Assembly Dot Product use partial sums on zero point by Frank Barchard · 3 years, 1 month ago
  26. 0049e89 QU8 C4 NEON Assembly Dot Product GEMM/IGEMM microkernels by Frank Barchard · 3 years, 1 month ago
  27. a29b57e QU8 e2e benchmark remove rndnu from benchmark names. by Frank Barchard · 3 years, 1 month ago
  28. 889ed10 QS8 gemm benchmarks switch from GEMMLOWP to RNDNU for AARCH64 assembly by Frank Barchard · 3 years, 1 month ago
  29. 52e4443 Rename linux_aarch64 to linux_arm64 by Marat Dukhan · 3 years, 1 month ago
  30. 6fe565e QU8 neondot use C2 partial sum for zero point accumulators. by Frank Barchard · 3 years, 1 month ago
  31. 3f2074f QU8 neondot use uint32x2 for zero point and accumulators by Frank Barchard · 3 years, 1 month ago
  32. 6507b17 Enable quantized inference by default on the Web platform by Marat Dukhan · 3 years, 1 month ago
  33. 7a8dd87 Work around generating v128.storeXX_lane for quantized WAsm SIMD microkernels by Marat Dukhan · 3 years, 1 month ago
  34. fb53a8c Enable quantized inference by default on the Web platform by XNNPACK Team · 3 years, 1 month ago
  35. 0f66135 Enable quantized inference by default on the Web platform by Marat Dukhan · 3 years, 1 month ago
  36. de9c64a Enable 4x16 QU8 dot production microkernels by Frank Barchard · 3 years, 1 month ago
  37. 9cedb59 Accumulate in 16 bits once in WAsm SIMD MUL16 QS8/QC8 DWCONV before extending to 32 bits by Marat Dukhan · 3 years, 1 month ago
  38. a74310a Remove UDOT by zero point along the N axis by Frank Barchard · 3 years, 1 month ago
  39. 41f8f7c Merge pull request #1826 from peterjc123:clang_windows_fixes by XNNPACK Team · 3 years, 1 month ago
  40. 65692c7 Fix build for Clang on Windows by peter · 3 years, 1 month ago
  41. 0d00baa Support quantized Clamp and Max Pooling operators in Subgraph API by Marat Dukhan · 3 years, 1 month ago
  42. 61c0c9e Clamp NC operator for S8 data type by Marat Dukhan · 3 years, 1 month ago
  43. 9491279 Refactor parameter initialization for VCLAMP microkernels by Marat Dukhan · 3 years, 1 month ago
  44. 4c3e5a9 GEMM benchmark assembly microkernels before intrinsics. by Frank Barchard · 3 years, 1 month ago
  45. e79acb7 S8 VCLAMP microkernels by Marat Dukhan · 3 years, 1 month ago
  46. 1f5b108 Refactor U8 CLAMP microkernels by Marat Dukhan · 3 years, 1 month ago
  47. 2ea50a0 Refactor U8 MAXPOOL microkernels similarly to S8 MAXPOOL by Marat Dukhan · 3 years, 1 month ago
  48. dc5c148 S8 Max Pooling operator by Marat Dukhan · 3 years, 1 month ago
  49. 2314753 S8 MAXPOOL microkernels for all architectures by Marat Dukhan · 3 years, 1 month ago
  50. f158942 WAsm SIMD implementation of U8 MAXPOOL microkernel by Marat Dukhan · 3 years, 1 month ago
  51. 91ae165 Refactor initialization of MAXPOOL microkernel parameters by Marat Dukhan · 3 years, 1 month ago
  52. ee69093 Enable shell and node environments for WAsm binaries by Marat Dukhan · 3 years, 2 months ago
  53. 9098aba E2E for QU8 GEMM microkernels by Frank Barchard · 3 years, 2 months ago
  54. e033126 Generate more tile sizes for QU8 gemm/igemm by Frank Barchard · 3 years, 2 months ago
  55. b1cd381 Enable dot production microkernels for QU8 on Cortex A55 by Frank Barchard · 3 years, 2 months ago
  56. 2025515 Enable dot production microkernels for QU8 on ARM by Frank Barchard · 3 years, 2 months ago
  57. 88e839c QU8 C4 NEON Dot Product GEMM/IGEMM microkernels by Frank Barchard · 3 years, 2 months ago
  58. cf557d4 Increase timeout for Multiply ND operator test by Marat Dukhan · 3 years, 2 months ago
  59. c67779e Fix flakiness in QS8/QC8 NHWC Deconvolution tests by Marat Dukhan · 3 years, 2 months ago
  60. e7991e7 Minor refactoring of RNG in Fully Connected tester by Marat Dukhan · 3 years, 2 months ago
  61. 57c7827 Fix flakiness in QS8/QC8 NHWC Convolution tests by Marat Dukhan · 3 years, 2 months ago
  62. b3faed3 Fix flakiness in QC8/QS8 GEMM/IGEMM/DWCONV microkernels by Marat Dukhan · 3 years, 2 months ago
  63. 0c2a31e Improve unpacking in SSE4+ QC8/QS8/QU8 GEMM/IGEMM microkernels by Marat Dukhan · 3 years, 2 months ago
  64. d960231 Remove tests for WAsm SIMD QS8 DWCONV microkernels with GEMMLOWP requantization by Marat Dukhan · 3 years, 2 months ago
  65. 36fe5aa Remove WAsm SIMD QS8 DWCONV microkernels with GEMMLOWP requantization by Marat Dukhan · 3 years, 2 months ago
  66. 88c2da6 Run template code generators by Frank Barchard · 3 years, 2 months ago
  67. b43c5ef Fix indent on C4 Neon Dot Product GEMM/IGEMM microkernels by Frank Barchard · 3 years, 2 months ago
  68. 8c96521 Fix Static Constant Padding + Convolution 2D fusion with quantization by Marat Dukhan · 3 years, 2 months ago
  69. e0a20d6 Expose quantized Static Constant Pad operator in Subgraph API by Marat Dukhan · 3 years, 2 months ago
  70. 139e961 X8 version of Constand Pad ND operator by Marat Dukhan · 3 years, 2 months ago
  71. 07706f6 Replace generic shuffle with narrow instructions in WAsm SIMD QS8/QU8/QC8 microkernels by Marat Dukhan · 3 years, 2 months ago
  72. dfc2db0 Add prefix to QC8/QS8/QU8 WAsm SIMD GEMM/IGEMM microkernels by Marat Dukhan · 3 years, 2 months ago
  73. 637a038 Fix implicit pointer cast warning in FP16ARITH microkernels by Marat Dukhan · 3 years, 2 months ago
  74. 0461f2d Generalize PAD microkernels to all 8-/16-/32-bit data types by Marat Dukhan · 3 years, 2 months ago
  75. 3e9dc22 Remove WAsm SIMD GEMM/IGEMM microkernels with GEMMLOWP requantization by Marat Dukhan · 3 years, 2 months ago
  76. 933051b Generalize FILL microkernels to all 8-/16-/32-bit data types by Marat Dukhan · 3 years, 2 months ago
  77. 7c74aff Add F32 VLRELU benchmarks by Marat Dukhan · 3 years, 2 months ago
  78. 4486f87 Prune NEON-DOT QS8 GEMM/IGEMM microkernels with FP32 & GEMMLOWP requantization by Marat Dukhan · 3 years, 2 months ago
  79. 400e7cb Prune WAsm SIMD QS8 GEMM/IGEMM microkernels with GEMMLOWP requantization by Marat Dukhan · 3 years, 2 months ago
  80. e16bf7d Prune AVX2/AVX512 QS8 GEMM/IGEMM microkernels with GEMMLOWP requantization by Marat Dukhan · 3 years, 2 months ago
  81. dc020ff Add 4x16c4 rndnu e2e benchmark for qs8. by Frank Barchard · 3 years, 2 months ago
  82. 8634f7e Refactor F32 VHSWISH benchmarks by Marat Dukhan · 3 years, 2 months ago
  83. 12e426c Refactor F32 VELU benchmarks by Marat Dukhan · 3 years, 2 months ago
  84. 9f8ea9b Refactor F32 VSIGMOID benchmarks by Marat Dukhan · 3 years, 2 months ago
  85. 5aeb32b Refactor F32 VSQRT benchmarks by Marat Dukhan · 3 years, 2 months ago
  86. 3b6c36e Refactor F32 VRELU benchmarks by Marat Dukhan · 3 years, 2 months ago
  87. 66a3ca1 Initialize QS8 microkernel pointers on pre-NEON ARM architecture by Marat Dukhan · 3 years, 2 months ago
  88. 8674629 Use QS8 GEMM WAsm SIMD microkernels with FP32 requantization in the benchmark by Marat Dukhan · 3 years, 2 months ago
  89. 0ff7989 Use FP32 requantization for extended-weights QS8 GEMM microkernels on x86 by Marat Dukhan · 3 years, 2 months ago
  90. 529d2c1 Remove x86 QS8 GEMM microkernels with GEMMLOWP requantization from benchmarks by Marat Dukhan · 3 years, 2 months ago
  91. ec47958 Prune redundant NEON GEMM/IGEMM microkernels with GEMMLOWP requantization by Marat Dukhan · 3 years, 2 months ago
  92. 348ed6d Add ISA checks in QS8/QU8 requantization tests by Marat Dukhan · 3 years, 2 months ago
  93. f879d9e Add qs8-requantization-test to CMake build by Marat Dukhan · 3 years, 2 months ago
  94. 3c5e662 Initialize QU8 VMUL[C] microkernels for pre-NEON ARM by Marat Dukhan · 3 years, 2 months ago
  95. 066a0cb Evaluate convertsion-based WAsm SIMD implementations in the rounding benchmark by Marat Dukhan · 3 years, 2 months ago
  96. 2dac7bb Unify on wasm_f64x2_spalt(0.0) to materialize zero SIMD vector in WAsm by Marat Dukhan · 3 years, 2 months ago
  97. d4db6af Replace wasm_i32x4_lt(vzero, vXX) with wasm_i32x4_shr(vXX, 31) by Marat Dukhan · 3 years, 2 months ago
  98. ebb6207 QU8 4x16 IGEMM remove push for X21 register by Frank Barchard · 3 years, 2 months ago
  99. 8a211a3 Check parameter initialization functions for non-NULL before calling by Marat Dukhan · 3 years, 2 months ago
  100. 085883b Remove references to Google-specific headers in BUILD.bazel by Marat Dukhan · 3 years, 2 months ago