p13709645
/
InfiniTensor
forked from jiuyuan/InfiniTensor
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Commit Graph

  • 601bdd3093 Remove: NNet pure_engine Liyan Zheng 2023-06-26 14:57:19 +0800
  • 2d869dffe6 add test report markdown v0630 wanghailu 2023-06-26 14:32:35 +0800
  • 4fbeb2b233 fix reshape wanghailu 2023-06-26 11:33:51 +0800
  • 1c84c3ebb9 fix: 消除无效张量 YdrMaster 2023-06-26 09:36:56 +0800
  • 29071ddcac Rename: Expr construction helpers NNET_e2e Liyan Zheng 2023-06-25 20:31:08 +0800
  • c6c445991a Add: enable mutator search in python Liyan Zheng 2023-06-25 20:18:18 +0800
  • 2f64997142 feat: 导出到 python YdrMaster 2023-06-25 17:17:14 +0800
  • 89b3cf7c08 feat: 为 MLU 转换 Conv 的 weight YdrMaster 2023-06-25 15:25:22 +0800
  • d87777b4cd format fix whjthu 2023-06-25 12:00:34 +0800
  • 16af5df2bc update model surgeon for converting CNN models from NCHW layout to NHWC layout whjthu 2023-06-25 11:51:55 +0800
  • db9d5f4ffe add bang conv nhwc kernel whjthu 2023-06-25 08:24:02 +0800
  • fa62e35e41 add bang conv nhwc kernel whjthu 2023-06-25 08:22:54 +0800
  • aaeadf9211 add support for ConvNHWC operator and model converter for different layout whjthu 2023-06-25 07:19:20 +0800
  • 19d7dc871d
         update doc (#83) Hardy 2023-06-23 14:22:52 +0800
  • 8f89b68f3d code format update_doc wanghailu 2023-06-21 09:41:18 +0800
  • 4f7bb811b0 add code wanghailu 2023-06-19 16:26:17 +0800
  • b86656ecfc fix wanghailu 2023-06-19 10:20:48 +0800
  • 9cb0c49162 Merge branch 'update_doc' of github.com:InfiniTensor/InfiniTensor into update_doc wanghailu 2023-06-19 09:52:46 +0800
  • 47965ad06f fix bug wanghailu 2023-06-19 09:51:17 +0800
  • 1b8c69e7b8 fix wanghailu 2023-06-15 14:31:57 +0800
  • 9df61ceb0e add env.sh and update install guide wanghailu 2023-06-13 10:49:53 +0800
  • 6d925a961f add max pool2d root 2023-06-07 10:10:10 +0800
  • de2e0ac0ef add code root 2023-06-07 09:58:45 +0800
  • e1a6fd9c82 update doc wanghailu 2023-06-06 15:23:37 +0800
  • d3ac411ef2 update doc wanghailu 2023-06-06 14:29:20 +0800
  • f3a2d5b87d add code wanghailu 2023-06-06 11:12:55 +0800
  • 1b0dd36f30 add code wanghailu 2023-06-06 09:56:15 +0800
  • d1a05c03dd support run resnet18 on xpu root 2023-06-05 16:32:08 +0800
  • 924bd70d1f add code root 2023-06-02 14:30:19 +0800
  • a9a5a05acf add some operator root 2023-06-02 10:36:18 +0800
  • 221d64e1e1 add unary operator root 2023-06-01 14:45:25 +0800
  • d48dbf5008 add unary operator root 2023-05-31 14:43:58 +0800
  • 1317c6b4eb update doc wanghailu 2023-05-31 11:01:55 +0800
  • 0570e31202 update doc wanghailu 2023-05-31 10:58:00 +0800
  • 8de7cb7b46 add transpose root 2023-05-31 10:47:53 +0800
  • 943f5234a6 add matmul root 2023-05-31 10:06:13 +0800
  • c4ee6ecdf9 add code root 2023-05-30 15:00:54 +0800
  • 750d345060 update doc wanghailu 2023-05-30 13:58:39 +0800
  • 84c1365dc4 update doc wanghailu 2023-05-30 11:27:21 +0800
  • df17080b63 add xpu code root 2023-05-29 16:40:43 +0800
  • dec1aa9e34 add code root 2023-05-29 13:34:07 +0800
  • 20b44edb08 add sub, mul, div, pow, maximum, minimum root 2023-05-26 11:21:39 +0800
  • 13d6a6144f support kunlun xpu and add an operator named Add root 2023-05-25 16:50:00 +0800
  • d25b606e12 Add: TF32 supports and accurate timing for conv Liyan Zheng 2023-05-07 13:22:39 +0800
  • cd9d2a3ffa fix unsqueeze parse model_test wangw 2023-04-23 18:01:57 +0800
  • abcfa76fb5 Add: efficient CUDA transpose for last two dims Liyan Zheng 2023-05-05 15:16:07 +0800
  • c983cd083d Hack for constructing an empty graph; Handle FlattenOp in Cuda Graph. TC_revision whjthu 2023-05-02 19:58:09 +0800
  • 0dd5f5003f Fix cuda kernel bugs; Fix search engine bugs. whjthu 2023-05-02 18:04:21 +0800
  • 5f0f1ce3c7 Add PET optimization rules whjthu 2023-05-02 15:50:21 +0800
  • 98dce3e4f1 Fix: using a reshape operator to handle the situation when all the operators are eliminated. whjthu 2023-05-02 05:21:26 +0800
  • b5211ec75e Fix axes attr in ReduceMean op. whjthu 2023-05-02 05:07:59 +0800
  • e8bb1c0bd8 Fix: Sort input order for broadcast element wize operators. Now the first input always has the same shape with the output. whjthu 2023-05-02 04:50:38 +0800
  • 33cb5e224a Replace sycl with OpenMP; Use gcc instead of dpcpp in IntelCPU runtime; Fix some bugs. whjthu 2023-05-01 17:34:26 +0800
  • 6a70555892 Add: TensorRT backend Liyan Zheng 2023-04-30 23:44:10 +0800
  • f47a411095 Add: export with random weight Liyan Zheng 2023-04-30 22:25:07 +0800
  • df2534d209 Fix: fuse Relu to GEMM/Conv Liyan Zheng 2023-04-30 16:16:16 +0800
  • a1f02593d3 Add: export Flatten operator to ONNX Liyan Zheng 2023-04-30 16:09:59 +0800
  • 65b4b42fa0 Merge remote-tracking branch 'origin/NNET_OpSearch' into NNET_e2e Liyan Zheng 2023-04-30 15:34:47 +0800
  • b068442bfb Add: print time in op evalution Liyan Zheng 2023-04-30 00:47:57 +0800
  • c6e7748786 Add: rule of Any+Relu->Any Liyan Zheng 2023-04-28 21:46:54 +0800
  • d0ae48d21d Add: CUDA Matmul selection Liyan Zheng 2023-04-28 19:13:19 +0800
  • c875f3cbb8 Add: Matmul Transpose plans Liyan Zheng 2023-04-28 19:13:01 +0800
  • 95a8b90fa7 Fix: add virutal method sync in Runtime Liyan Zheng 2023-04-28 00:44:54 +0800
  • c58b67f743 Chore: suppress output Liyan Zheng 2023-04-26 14:11:53 +0800
  • 75c9226164 Merge branch 'NNET_e2e' into NNET_op_test Liyan Zheng 2023-04-25 04:32:18 +0800
  • f877eca517 Add: IT_ASSERT in checkCudaError NNET_gcn Liyan Zheng 2023-04-25 04:25:10 +0800
  • b13b799fbe All mdoels E2E Liyan Zheng 2023-04-25 04:24:43 +0800
  • 350fc01d39 Add: Search Depth Liyan Zheng 2023-04-25 01:07:21 +0800
  • 71f4f6e9d9 add op test for einnet NNET_op_test whjthu 2023-04-24 21:48:29 +0800
  • 1408d308cc Add: FCRS log NNET_OpSearch Liyan Zheng 2023-04-24 21:20:29 +0800
  • 11229a2baa Add: Figure 17 Liyan Zheng 2023-04-24 21:07:30 +0800
  • 0fb59c4dd6 merge gcn NNET_gcn_fuse xxcclong 2023-04-24 16:23:36 +0800
  • bdeaeae3b1 tmp xxcclong 2023-04-24 16:22:22 +0800
  • 2b85ac41ef Fix: CUDA Relu for 2D/1D tensor Liyan Zheng 2023-04-24 16:14:26 +0800
  • d248e227bb fuse any + relu xxcclong 2023-04-24 15:49:38 +0800
  • 1e46750159 Add conv2bgemm and fix mutator::runtime Liyan Zheng 2023-04-24 13:06:24 +0800
  • 079985bc8c Add: efficient transpose Liyan Zheng 2023-04-24 12:42:07 +0800
  • 8fe34c9e9b hack fsrcnn-conv-bias-act-fuse xxcclong 2023-04-24 03:55:23 +0800
  • 1eeb2d771e merge xxcclong 2023-04-24 03:45:10 +0800
  • 7270561728 Merge branch 'NNET_gcn' into fsrcnn-conv-bias-act-fuse xxcclong 2023-04-24 03:44:03 +0800
  • c1275cddb6 Fix: conv2dreduce_kernel_ offset Liyan Zheng 2023-04-24 02:30:41 +0800
  • 51cc042f56 Add: nchw to nhwc conversion Liyan Zheng 2023-04-24 02:29:53 +0800
  • 18d6ba4022 Merge branch 'NNET_e2e' into NNET_gcn Liyan Zheng 2023-04-23 23:07:42 +0800
  • 4211fd1f32 Fix: matmul transpose in convNHWC2gemm rule NNET_e2e_for_merge Liyan Zheng 2023-04-23 22:54:50 +0800
  • 8409c1f9d4 tested fsrcnn xxcclong 2023-04-23 22:19:51 +0800
  • 830b28913c better transposed convreduce xxcclong 2023-04-23 21:36:25 +0800
  • 1ba78d7f89 Add: reduce in Any Liyan Zheng 2023-04-23 21:36:12 +0800
  • 777aebafc9 fsrcnn xxcclong 2023-04-23 20:56:19 +0800
  • 131a679340 gcn optimization whjthu 2023-04-23 03:44:44 +0800
  • 5df2524ff9 Merge branch 'NNET_eliminateOP' into NNET_e2e Liyan Zheng 2023-04-23 13:35:29 +0800
  • f204866d93 Fix: reduce workspace size NNET_eliminateOP Liyan Zheng 2023-04-23 13:34:07 +0800
  • b9819e65c1 Fix: allow eliminate and fusion failure in search Liyan Zheng 2023-04-23 13:15:34 +0800
  • 7277356744 Add: Reshape/Transpose elimination Liyan Zheng 2023-04-23 02:10:05 +0800
  • 24f6eb273e
         Chore: remove redundant semicolon NNET_anyOp zhengly123 2023-04-23 00:25:33 +0800
  • f820117acd fix unused code whjthu 2023-04-23 00:18:26 +0800
  • 1ab2118716 add AnyOp and cuda kernel whjthu 2023-04-23 00:16:03 +0800
  • ff97c879fb add ConvNHWC and FSRCNN graph huangshuhong 2023-04-23 00:02:22 +0800
  • acc64fd32c Merge branch 'NNET_transpose' into NNET_e2e Liyan Zheng 2023-04-22 21:32:31 +0800
  • 33ab5dcd3e Fix: gbmm kernel Liyan Zheng 2023-04-22 21:14:52 +0800
  • e2f18272c9 Add: no malloc for reshape outputs Liyan Zheng 2023-04-22 21:13:57 +0800