Граф коммитов

447 Коммитов

Автор SHA1 Сообщение Дата
Tianqi Chen 2607a83619 [RUNTIME][PYTHON] More compatibility in ndarray (#463) 2017-09-18 23:11:19 -07:00
Tianqi Chen 0220abbafa [METAL] use 32bit indexing for metal until we have a bound adapted pass (#462)
* [METAL] use 32bit indexing for metal until we have a bound adapted pass

* fix lint
2017-09-18 13:14:31 -07:00
Tianqi Chen 324fe1653a [SUBMODULE] upgrade dmlc-core (#461) 2017-09-18 10:22:06 -07:00
Xingjian Shi 51f8327f73 Use ewise schedule for broadcasting (#460) 2017-09-17 23:32:05 -07:00
Tianqi Chen d6007a24c8 [RPC] Expose module handle (#459)
* [RPC] Expose module handle

* not include handle
2017-09-17 21:50:41 -07:00
Tianqi Chen fe564d9037 [RPC] Include rpc session info into context (#458)
* [RPC] Include rpc session info into context

* add type checker in return converison
2017-09-17 19:05:55 -07:00
Tianqi Chen 3a0d3a391f [PASS] Fix intrinsic lowering with fma and other intrin (#457)
* [PASS] Fix intrinsic lowering with fma and other intrin

* relax rtol for sqrt
2017-09-17 15:54:36 -07:00
Xingjian Shi 5ae1a07920 [TOPI] add binary broadacst (#456)
* add binary broadacst

* fix testing

* revise testing threshold
2017-09-17 09:34:34 -07:00
Aditya Atluri dd029c83fc added vim temporary files to gitignore (#453) 2017-09-13 20:29:54 -07:00
Aditya Atluri 891e226bae [BACKEND] initial llvm codegen for amdgpu (#402)
* added initial llvm codegen for amdgpu

* fixed whitespace

* fixed hsaco gen from ir

* fixed targetmachine for rocm and added GetSource for rocm

* fixed whitespace issues

* changed statement to use less than 100 lines

* added intrinsics for workgroup - rocm

* whitespace - newline error fix

* fixed error msg for workitem-workgroup intrinsics

* added llvm ir dump for rocm codegen

* [ROCM] changed codegen to emit proper amdgpu kernel header

* fixed whitespace error

* fixed whitespace error- 2

* fixed AddFunction to not to use extra arg

1. Changed AddFunctionInternal to not to take extra arg for target type
2. Use Target from CodeGenLLVM to check for AMDGPU target

* fixed whitespaces

* fixed whitespaces 2

* fixed codegen for AMDGPU - now generating valid IR

* fixed codegen depending on code review

* reviewed alignment for amd devices

* added code to dump code object to file

* fixed cpplint errors

* print out IR after pass manager

* added code to dump asm, obj to file and std string

* fixed whitespaces

* Update codegen_amdgpu.cc

* used registry for amdgpu llvm

* Fixed whitespaces

* added code for calling linker

* fixed formatting errors

* added rocm link python interface

* fixed pylint issues and added more body to the function

* added doc string

* added doc string for module

* fixed python code after review, fixed llvm object codegen

* fixed linker to generate code object

* removed dumping to output file and debugging log out

* fixed lint for python code

* added fault check after running linker

* removed print statement in rocm.py

* changed rocm lld linker to raise runtimeerror than emitting error log to stderr

* changed the way linker command line is pass to subprocess.popen

* removed redundant code and reuse tvm utils

* removed commented out code

* removed cloning of unused modules, and put IR into string
2017-09-13 12:29:20 -07:00
Tianqi Chen 5061a6da5e [RUNTIME] Add function to pack arguments (#452) 2017-09-12 22:08:52 -07:00
Leyuan Wang 769544ad7d conv2d schedule fall back warning fixed (#450) 2017-09-12 09:25:02 -07:00
Shuai Yuan 220fa04015 [RPC] clarify error message for unmatched context (#451)
Clarify confusing error message for unmatched context
2017-09-12 09:24:42 -07:00
wetliu ffff1e4932 [TOPI] Depth wise convolution backward methods for NHWC (#434)
* rename the nchw and pass the unit test; going to do it for nhwc depthwise

* bug with fusion

* nchw works fine; nhwc float32 problem remains

* still cannot bind them together

* fusion works

* syntax fix

* all bugs fixed; test cases pass

* minor fix on nn.h

* back wrt input

* backward wrt input nhwc; only test case in recipe

* test case for depthwise back wrt input

* test case for depthwise backward wrt weight

* tags

* minor fixes

* pylint test; add arch=3.7

* modify scheduler

* better backward depthwise w.r.t weight scheduler

* updated scheduler

* test_topi_depthwise_conv2d_back_input.py and test_topi_depthwise_conv2d_back_weight.py success

* all test cases wrt input pass

* update

* new test cases and scheduler

* not working 1 and 2

* good wrt weight, bad wrt input

* test cases added

* remove tf lines

* minor fix

* compute arch changed

* remove compile hook

* minor change

* pylint

* fix the float for python case

* fix cases for python3 case

* except for memoize

* fix most; memoize still wrong

* memoize added

* unexpected layout cases added for scheduler

* error message layout other than NHWC added

* improve padding

* fix as pr requests

* remove dilate in backward wrt weight
2017-09-11 19:54:27 -07:00
Tianqi Chen f2ab736b61 [RUNTIME] Enable extension type to PackedFunc. (#447)
* [RUNTIME] Enable extension type to PackedFunc.

* More comments
2017-09-11 17:29:18 -07:00
Shuai Yuan 3130f2d558 [DOCS] Add prerequisites about zlib1g-dev (#446)
* [DOCS] Add prerequisites about zlib1g-devin

Add prerequisites about zlib1g-dev. It occurs `/usr/bin/ld: cannot find -lz` without zlib1g-dev.

* Add prerequisites about python-setuptools 

Add prerequisites about python-setuptools. Otherwise, it will fail when executing `python setup install --user` command.

* [DOCS] Add prerequisites about python-dev

Add installation prerequisites about python-dev. Otherwise, it will fail with `SystemError: Cannot compile 'Python.h'. Perhaps you need to install python-dev|python-devel.` when executing `python setup install --user`.
2017-09-11 10:59:49 -07:00
Tianqi Chen 1077f8e814 [RUNTIME][RPC] Enable remote linking of device code. (#444)
* [RUNTIME][RPC] Enable remote linking of device code.

* fix build
2017-09-10 23:56:46 -07:00
Yizhi Liu 64402c14d3 disable fopen64 in dmlc-core (#443) 2017-09-10 11:32:47 -07:00
Yizhi Liu 28281c999f [APP] enhance android ui (#441) 2017-09-08 22:52:43 -07:00
Tianqi Chen a961b29c5c [TOPI] Fix softmax bug (#437) 2017-09-08 19:00:24 -07:00
Tianqi Chen 0c9adc5b1e [LLVM] Protect ll when emit pass (#436) 2017-09-08 18:46:46 -07:00
Leyuan Wang edca6830dc Conv2d updated (#435)
* improved conv2d for last group of workloads

* conv2d_nchw improved on 14_256_256 and 56_64_128
2017-09-08 16:15:11 -07:00
Tianqi Chen 400c1c483e [SCHEDULE] Enhance cache_write to enable layout change. (#432)
* [SCHEDULE] Enahance cache_write to enable layout change.

* more tests
2017-09-07 15:50:25 -07:00
Shuai Yuan 663d7c52f8 [DOCS] Fix markdown syntax error (#430)
Fix markdown syntax error (code shifts out of markdown-code box).
2017-09-07 09:03:43 -07:00
Tianqi Chen e4f588106a [METAL][RUNTIME] Fix bug of memcpy into metal buffer (#428) 2017-09-06 16:22:47 -07:00
Leyuan Wang ab17bf6569 [TOPI] Improve conv2d for resnet18 workload (#427)
* relu activation migrated to topi

* reviews addressed

* relu compute deleted

* conv2d_nchw updated

* resnet18 hand tuned schedule added

* pylint error fixed

* one more workload test for conv2d_nchw

* conv2d schedule subfunctions added for different patterns

* reviews addressed
2017-09-05 19:50:39 -07:00
Tianqi Chen 5ea4072c5a [PYTHON] Allow general types (#425) 2017-09-04 22:24:17 -07:00
Tianqi Chen df3c996b2a [TEST] Add memoize to save test data (#424)
* [TEST] Add memoize to save test data

* Update comment

* mark py version
2017-09-04 21:55:05 -07:00
Tianqi Chen 071b138f74 [BUILD] Improve build instruction with llvm. (#422) 2017-09-04 19:36:46 -07:00
Tianqi Chen c6c287bfa3 [SETUP] Always use relpath for setup (#421)
* [SETUP] Always use relpath for setup

* [CMAKE] Fix cmake llvm build
2017-09-04 17:53:18 -07:00
ziheng 3c9e262fbd [FIX] Fix build error: call to 'make_const' is ambiguous (#415) 2017-09-03 20:54:55 -07:00
ziheng e05f54bee0 [TOPI] Add topi.target; Schedule for raspberry pi (#406)
* CPU Schedule for raspberry pi

* Update

* Update

* Add topi.target

* Refactor

* Update

* Make python3 happy

* Improve

* Improve

* Improve

* Use get_const_int
2017-09-02 23:25:07 -07:00
Tianqi Chen f6bb7ababa [GPU][TOPI] Fix cross thread reduction schedule (#414) 2017-09-02 22:49:16 -07:00
Tianqi Chen adf39837aa [PASS] Improve double buffer (#413) 2017-09-02 22:49:08 -07:00
Tianqi Chen 5072efaeca [PASS] Improve vthread injection. (#411) 2017-09-02 21:16:57 -07:00
Tianqi Chen b0d9f2998a [PASS] Check memory info bound to guard failure (#409) 2017-09-01 16:43:56 -07:00
Tianqi Chen 0138997fbe [BUILD] Allow inject custom pass via phase (#408) 2017-09-01 16:00:04 -07:00
Tianqi Chen f73c461f50 [BACKEND] Explicitly allow specialization of FMA in llvm (#407) 2017-09-01 11:46:45 -07:00
Tianqi Chen a45d3b01f7 [PASS] InjectDoubleBuffer (#405) 2017-08-31 18:28:24 -07:00
Tianqi Chen b8c8aadfeb [BACKEND] Allow nvptx to pass ll ir to CUDAModule (#404) 2017-08-31 09:30:57 -07:00
ziheng 50c7a01b30 [submodule] update dlpack (#403) 2017-08-31 08:58:23 -07:00
Tianqi Chen 4578048ce1 [PASS] IRTransform to enable IR pass proptype in python (#401) 2017-08-30 16:01:35 -07:00
Tianqi Chen 8ef266063f [SCHEDULE][PASS] support storage_align of certain axis (#400)
* [SCHEDULE][PASS] support storage_align of certain axis

* fix lint
2017-08-30 13:11:18 -07:00
Tianqi Chen b03c324304 [CODEGEN] Multiple parallel in one launch (#399) 2017-08-30 09:59:43 -07:00
Tianqi Chen ad8733ea14 [DOCS][APP] Add Example for C++ deployment (#398)
* [DOCS][APP] Add Example for C++ deployment

* fix lint
2017-08-30 08:50:36 -07:00
Tianqi Chen 0eec95bf54 [DOC] Reorganize docs (#397) 2017-08-29 14:02:04 -07:00
Yizhi Liu 4e8002059d [APP] Android RPC README (#395) 2017-08-28 13:51:01 -07:00
Yuwei HU 7ad3c51e61 [TOPI] improve elemwise schedule (#393)
* [TOPI] improve elemwise schedule

* modify fuse
2017-08-27 22:05:22 -07:00
Tianqi Chen 0560e1569e [CODEGEN] NVPTX backend. (#392)
* [CODEGEN] NVPTX backend.

* Fix pylint

* use fix
2017-08-27 17:09:55 -07:00
Leyuan Wang efafa1a0dd [TOPI] Relu and schedule elemwise (#390)
* convolution compute typo fixed

* relu activation migrated to topi

* reviews addressed

* elemwise schedule added

* relu compute deleted
2017-08-27 16:47:22 -07:00