Google Git Google Git
Sign in
chromium/external/github.com/google/XNNPACK/HEAD
  1. e3de491 Fix qd8 HVX GEMM/IGEMM build on Hexagon V69 by Frank Barchard · 27 minutes ago upstream/master
  2. a936a0e Add ARM SME2 and SME ACLE GEMM microkernels for `pf32-qc8w` and `pf32-qc4w` along with SME `x32` LHS packing microkernels, unit tests, and benchmarks. by Jeff Lee · 16 hours ago
  3. b94680b Add qp8_f32_qc2w_gemm benchmark for SME2 and add k_large unit tests by Frank Barchard · 17 hours ago
  4. 163fab0 When `A` has unit extent along `m` and `C` has `m > 1`, compute `dot(A, B)` by Marie White · 2 days ago
  5. 52fd02e Account for subnormal flush-to-zero in fully connected test tolerance. by Frank Barchard · 3 days ago
  6. 020575c Add ARM SME and SME2 ACLE x8-packq (f32qp8) microkernels and build rules by Jeff Lee · 3 days ago
  7. d4269f5 Fix infinity range and bf16 domains in test random generators by Frank Barchard · 3 days ago
  8. 46b6943 Look up deconvolution weights in the cache before packing them. by Quentin Khan · 3 days ago
  9. 6fb94d5 Add build configuration flags (disabled by default via by Frank Barchard · 3 days ago
  10. 95a0001 Disable ARM64 SME features when kernel SME is disabled. by Frank Barchard · 3 days ago
  11. 5c7bdb4 Simplify AMX IGEMM microkernels for m < 16 and remove 7x16c4 kernels. by Frank Barchard · 3 days ago
  12. 38762be Detect RISC-V uarch via hwprobe to enable RVV FP16 on SiFive X280/X390. by Frank Barchard · 3 days ago
  13. af32a5e Fix return types and signedness of HVX s8 and u8 SIMD reductions by Frank Barchard · 4 days ago
  14. b57d7c6 Add multiply_add elementwise kernels for (int8, int8, int32), (bf16, bf16, fp32), and (fp16, fp16, fp32). by Dillon Sharlet · 4 days ago
  15. 3552b03 Update XNNPACK dependencies by Dillon Sharlet · 4 days ago
  16. aa4235f In the 8-bit view of ZA, horizontal slices of 4 consecutive rows each, by Samuel Fuller · 4 days ago
  17. 8d7a8ef Support transposed inputs in pack_b and absorb transposes. by Dillon Sharlet · 4 days ago
  18. c9024d1 Add Qualcomm Hexagon HMX c4 qs8-qc4w and qs8-qc8w GEMM microkernels. by Frank Barchard · 5 days ago
  19. 82be863 Fix Hexagon HVX FP16 subgraph fallback and optimize f16<->f32 vcvt by Frank Barchard · 6 days ago chromium/8145 chromium/8146 chromium/8147 chromium/8148 chromium/8149 chromium/8150 chromium/8151 chromium/8152 chromium/8153
  20. d6d1497 Remove unit input dimensions in reduction normalization. by Frank Barchard · 6 days ago
  21. 81e7d0d Enforce 4D input shape in deconvolution-2d reshape by Reilly Grant · 6 days ago
  22. 7528581 Avoid lowering tensor rank in LUT fusion when broadcasting static operands by Reilly Grant · 6 days ago
  23. d1d59d4 Replace the internal static_reshape, split_dim, fuse_dim, split_dims, and fuse_dims nodes with a single reshape node. A reshape is described by groups of input dimensions that are fused and split into groups of output dimensions. The public API is unchanged. by Dillon Sharlet · 7 days ago
  24. f745c33 Fix 1x1 dynamic quantization check in convolution-2d subgraph test. by Frank Barchard · 7 days ago
  25. 40f3bc4 Fix slow random shapes and increase shard count in YNNPACK dot test by Frank Barchard · 7 days ago
  26. 46c2f29 Check unary datatypes when the node is defined. by Quentin Khan · 7 days ago
  27. 80bc909 Add fp64 support to dot_bench by Dillon Sharlet · 7 days ago
  28. a7b933c Add Qualcomm Hexagon HMX c2 f16-gemm microkernel. by Frank Barchard · 7 days ago
  29. 9b706e7 Use AVX-512 masked loads for AMX GEMM and IGEMM k remainders. by Frank Barchard · 7 days ago
  30. 9075d64 Extend Hexagon HVX `c4-hvx.c.in` GEMM and IGEMM templates to support by Frank Barchard · 7 days ago
  31. 51b179d Disable ARM SME on Windows ARM64 and guard naked SME stubs for GCC. by Frank Barchard · 8 days ago
  32. 08cc38b Rollback Skip dequant BMM rewrite when M > mr. by Changming Sun · 8 days ago
  33. 1043ad9 Choose order of m or n loops based on the size of A or B by Dillon Sharlet · 10 days ago
  34. fc1a262 Disable KleidiAI assembly and Arm64 SVE/SME architecture flags when building with MemorySanitizer, since MSan does not instrument inline assembly in KleidiAI or SVE/SME ACLE intrinsics. by XNNPACK Team · 10 days ago
  35. f11425b Update XNNPACK dependencies by Dillon Sharlet · 10 days ago
  36. 74fc960 Make ARM SME ABI stubs naked and uninstrumented to preserve caller registers by Samuel Fuller · 10 days ago
  37. 2691822 Return `unsupported_parameter` instead of `invalid_parameter` for unimplemented operator cases. by Quentin Khan · 10 days ago
  38. 9167c95 Define quantized internal values in `SubgraphTester`. by Quentin Khan · 10 days ago
  39. e877337 YNNPACK: Fix RISC-V Zvfh Clang version check in half.h by Frank Barchard · 10 days ago
  40. 7662831 f16-vlog test fix by Frank Barchard · 10 days ago
  41. 4a85b29 Add `+sme` target feature to `sme2_microkernels` in `BUILD.gn`. by Samuel Fuller · 10 days ago
  42. 264a9a8 Return `unsupported_parameter` for unsupported datatypes in `xnn_define_softmax`. by Quentin Khan · 10 days ago
  43. 09d1f6b Return `unsupported_parameter` for unimplemented `xnn_define_static_transpose` cases. by Quentin Khan · 10 days ago
  44. 113dc4a Don't read sub-byte static scalars as floats in the subgraph optimizer. by Quentin Khan · 10 days ago
  45. 9c1ad7d Reject per-tensor definitions of channelwise and blockwise datatypes. by Quentin Khan · 10 days ago
  46. 31e694a Return `unsupported_parameter` for sub-byte datatypes in copy-like nodes. by Quentin Khan · 10 days ago
  47. 1499bfa Return `unsupported_parameter` for unimplemented non-zero zero points. by Quentin Khan · 10 days ago
  48. 6304ce4 Return `unsupported_parameter` for unimplemented `xnn_define_batch_matrix_multiply` cases. by Quentin Khan · 10 days ago
  49. bc2bbfb Add ARM SME2 ACLE GEMM for pf16 and pf32 and LHS packing microkernels, along with unit tests and benchmarks. by Jeff Lee · 11 days ago
  50. c58baee Fix RISC-V Vector build errors, crashes, and test bugs across Coral by Frank Barchard · 11 days ago
  51. 5896bd0 Use uint64_t for Shape dimensions and ops calculation in YNNPACK dot benchmark. The previous usage of `int` would overflow in the calculation if the shape dimensions were too large. (i.e. `m * n * k >= 2^32`). This would make the benchmark erroneously report "0" for ops calculations. by Samuel Fuller · 11 days ago
  52. a30d821 Fix Unpooling2D.ReshapeOverflowOutputSize test attempting a huge allocation on 32-bit targets. by Dillon Sharlet · 11 days ago
  53. f7b0697 [jax-buildcop] [YNNPACK] Use unpacked_kernel in define_dot when pack_b is false by XNNPACK Team · 11 days ago
  54. 1025db7 Return `unsupported_parameter` for unsupported datatypes in `xnn_define_rope`. by Quentin Khan · 11 days ago
  55. e216faa Return `unsupported_parameter` for unsupported datatypes in `xnn_define_unpooling_2d`. by Quentin Khan · 11 days ago
  56. 890fe8a Return `unsupported_parameter` for unsupported datatypes in `xnn_define_average_pooling_2d`. by Quentin Khan · 11 days ago
  57. 98188a5 Map XNNPACK statuses to absl status codes. by Quentin Khan · 11 days ago
  58. 5ebc3f1 Add AVX512SKX c2 x16-packw microkernels for AMX-FP16 GEMM. by Frank Barchard · 11 days ago
  59. cde8417 [YNNPACK] Fix src2 operand in GCC Intel-syntax AMX tile dot product by Frank Barchard · 11 days ago
  60. cb7afb3 Use dynamic pre-shift in qs8/qu8 vcvt to increase fixed-point scale precision from 8 bits to up to 15 bits. by XNNPACK Team · 11 days ago
  61. d1002e1 Check the constant pad datatype before defining the node. by Quentin Khan · 11 days ago
  62. d2b705e Return `unsupported_parameter` for unsupported datatypes in `xnn_define_max_pooling_2d`. by Quentin Khan · 11 days ago
  63. f1e70ce Return `unsupported_parameter` for unsupported datatypes in `xnn_define_argmax_pooling_2d`. by Quentin Khan · 11 days ago
  64. a18e5fe Return `unsupported_parameter` for unsupported datatypes in `xnn_define_static_resize_bilinear_2d`. by Quentin Khan · 11 days ago
  65. 12e6692 Make all x86 and ARM tile_k = 1 dot kernels unaligned_b. by Dillon Sharlet · 12 days ago
  66. 4d9ce95 Use tail- and mask-undisturbed RVV intrinsics in reductions and vlrelu. by Frank Barchard · 12 days ago
  67. c361311 Skip unused tiles in AVX512-AMX INT8 GEMM and IGEMM microkernels. by Frank Barchard · 12 days ago
  68. 8140a97 Check binary datatypes when the node is defined. by Quentin Khan · 12 days ago
  69. d302556 Return unsupported_parameter for unimplemented constant pad requests. by Quentin Khan · 12 days ago
  70. 9b0b1bf Fix AMX IGEMM row-stride crash by using direct _tile_loadd for mr == 1 by Frank Barchard · 12 days ago
  71. f967326 Document when XNNPACK returns invalid_parameter vs unsupported_parameter. by Quentin Khan · 12 days ago
  72. 1380e84 Fix SIGSEGV in pf32 SME/SME2 GEMM microkernel benchmarks. by Samuel Fuller · 12 days ago
  73. 9dd61a6 Lower ML Drift quantizations to real float to int conversions. by Quentin Khan · 12 days ago
  74. e8039b8 Improve performance on large square shapes. by Marie White · 12 days ago
  75. bd35a70 Lower ML Drift dequantizations of 8-bit integers to int to float conversions. by Quentin Khan · 12 days ago
  76. 4206de5 Only use tail loads when out of bounds in unaligned_b dot kernels. by Dillon Sharlet · 13 days ago
  77. 80bfb76 Fix premature overflow in SIMD exp and expm1 by Dillon Sharlet · 13 days ago
  78. f71e54d Don't lose a fused producer's parallelism to a consumer's single-iteration loop by Dillon Sharlet · 13 days ago
  79. 2ebdb70 Fix unused-variable warnings by Anshul Nigham · 13 days ago
  80. 48e3be0 Add AVX512-AMX-FP16 GEMM and IGEMM microkernels for Granite Rapids. by Frank Barchard · 13 days ago
  81. 0b1f641 Zero the column padding of packed B in the transpose packing path by Dillon Sharlet · 13 days ago
  82. a404f8f Use 5x64 avx512skx microkernel for f16_f32acc GEMM and IGEMM by Frank Barchard · 14 days ago
  83. 818a954 [YNNPACK] Fix Hexagon HVX dot bitfield overflow, NaN math, and tests by Frank Barchard · 14 days ago
  84. 059a89b [gn] Separate prod and non-prod microkernels by Richard Townsend · 14 days ago
  85. 51645a0 Adds new ARM SME2 ACLE based microkernels for QP8 F32 QC4W/QC8W GEMM and packing. by Jeff Lee · 14 days ago
  86. 401d450 Unpoison external output buffers in NnpackRunner under MSAN. by XNNPACK Team · 2 weeks ago
  87. 24ab388 Optimize AMX GEMM and IGEMM quantization, 4-bit unpacking, and loads by Frank Barchard · 2 weeks ago
  88. 9e22e5c Allow concurrent constant folding across YNNPACK subgraphs. by Dillon Sharlet · 2 weeks ago
  89. c2aa288 Support std::optional<T> in ScopeGuard::Bind. by Misha Gutman · 2 weeks ago
  90. f2e3ba2 [HVX] Replace vsplat + vector ops with immediate and unary instructions by Frank Barchard · 2 weeks ago
  91. 23dd507 Avoid static initializers for the constant folding cache. by XNNPACK Team · 2 weeks ago
  92. 26186f8 Support std::optional<T> in CheckTensors. by Misha Gutman · 2 weeks ago
  93. fa0fa36 Schedule non-aliasing `static_transpose` over outer batch dimensions (`d >= pin_end`) while omitting loop splits on the inner transposed dimensions (`0..pin_end-1`). by Misha Gutman · 3 weeks ago
  94. 0e7357e Support strided and broadcast inputs in matrix packing by Dillon Sharlet · 3 weeks ago
  95. 78d255d Fix genxnn race where update-microkernels.py ran before codegen finished. Default genxnn is a full regeneration like the old -f; -f adds the gemm_compiler update (rarely changes); -n regenerates changed files only; -u skips update-microkernels.py. by Frank Barchard · 3 weeks ago
  96. 115e015 Add a shared numerical test for per-channel dequantizations. by Quentin Khan · 3 weeks ago
  97. 533bf3f Enforce consistent arithmetic using fixed k-loop sizes. Reduction dimensions are now only dependent on input shape and data type, and not CPU-dependent parameters like kernel block sizes. by Marie White · 3 weeks ago
  98. 05576dc Add Hexagon HVX `x32_packw_gemm_goi` microkernels (`xnn_x32_packw_gemm_goi_ukernel_x32__hvx_u32` and `xnn_x32_packw_gemm_goi_ukernel_x64__hvx_u32`) using a 5-stage 32x32 HVX `Q6_W_vshuff_VVR` butterfly transpose and wire them into `f32_gemm_config`, `f32_igemm_config`, and `f32_gemm_nr2_config`. by Frank Barchard · 3 weeks ago
  99. ef04478 Define XNN_ENABLE_RISCV_VECTOR in Bazel xnnpack_configurable_defines(). by Frank Barchard · 3 weeks ago
  100. 43dd28e Add copyright year and clean working tree guidelines to AGENTS.md and CONTRIBUTING.md. by Frank Barchard · 3 weeks ago