Commit Graph

103 Commits

Author SHA1 Message Date
Chao Liu
cd883e7581 experimenting
[ROCm/composable_kernel commit: 766b0a9eaf]
2019-03-24 12:09:57 -05:00
Chao Liu
6f1651f8a7 experimenting
[ROCm/composable_kernel commit: f35c64eb78]
2019-03-23 19:35:31 -05:00
Chao Liu
971442aab2 experiment with hip compiler
[ROCm/composable_kernel commit: 68ae0731f1]
2019-03-22 19:40:01 -05:00
Chao Liu
b24dc40c09 adding inline asm
[ROCm/composable_kernel commit: 52ae168b17]
2019-03-22 17:05:23 -05:00
Chao Liu
2c3bd06d25 Merge branch 'direct_fp16'
[ROCm/composable_kernel commit: fdaaaa500c]
2019-03-22 16:46:41 -05:00
Chao Liu
1f925812b2 hip build
[ROCm/composable_kernel commit: 8c923db423]
2019-03-22 14:22:58 -05:00
Chao Liu
9637b859d8 added int8x4
[ROCm/composable_kernel commit: e72eece8fc]
2019-03-21 09:59:40 -05:00
Chao Liu
99a6831f62 adding int8 direct that reads pre-vectorized data
[ROCm/composable_kernel commit: 02d72160dc]
2019-03-19 01:30:28 -05:00
Chao Liu
80f2ab2e09 adding int8 direct that reads pre-vectorized data
[ROCm/composable_kernel commit: 050a1a6890]
2019-03-19 00:05:41 -05:00
Chao Liu
732984e63b adding fp16 direct that reads pre-vectorized data
[ROCm/composable_kernel commit: 79d9b1084b]
2019-03-18 18:16:02 -05:00
Chao Liu
5ba0f64087 adding fp16 direct that reads pre-vectorized data
[ROCm/composable_kernel commit: 4f0fc72e91]
2019-03-18 15:03:17 -05:00
Chao Liu
da50d65ba0 refactoring block copy
[ROCm/composable_kernel commit: 03eef73c5b]
2019-03-17 15:36:38 -05:00
Chao Liu
6fd0910da8 refactoring ConstantTensorDescriptor
[ROCm/composable_kernel commit: a0584426ff]
2019-03-17 03:22:41 -05:00
Chao Liu
0485704cb3 refactor
[ROCm/composable_kernel commit: fd8de38417]
2019-03-16 10:50:46 -05:00
Chao Liu
7bfe330532 update hip build
[ROCm/composable_kernel commit: 2c9b8c2432]
2019-03-12 17:20:11 -05:00
Chao Liu
5c5f48e664 update
[ROCm/composable_kernel commit: 0c88a3d891]
2019-03-09 13:51:08 -06:00
Chao Liu
85e6b5db6d Merge branch 'master' into implicit_gemm_fp16
[ROCm/composable_kernel commit: ce0182ce05]
2019-03-09 13:46:47 -06:00
Chao Liu
973e3610cf refactor
[ROCm/composable_kernel commit: f54cad7d4f]
2019-03-09 13:39:24 -06:00
Chao Liu
8c2635a357 refactor
[ROCm/composable_kernel commit: 7a97087713]
2019-03-09 12:59:47 -06:00
Chao Liu
aec22f9cf9 refactor
[ROCm/composable_kernel commit: 43cd8529c2]
2019-03-09 12:52:16 -06:00
Chao Liu
005e6fe84d refactor
[ROCm/composable_kernel commit: 8edbc659b8]
2019-03-06 12:34:31 -06:00
Chao Liu
e3ab560c50 refactor
[ROCm/composable_kernel commit: 04c5527d07]
2019-03-04 17:09:20 -06:00
Chao Liu
77ec1e4016 clean up
[ROCm/composable_kernel commit: 5fd40ad768]
2019-03-02 17:27:37 -06:00
Chao Liu
7080421844 refactor
[ROCm/composable_kernel commit: 4543d17a71]
2019-02-19 22:07:15 -06:00
Chao Liu
23b7f3ef33 refactor
[ROCm/composable_kernel commit: b2b622e8b2]
2019-02-19 20:34:21 -06:00
Chao Liu
1c962a13ee device_implicit_gemm_convolution_1_chwn_csrk_khwn: use tensor copy (instead of pointwise) for writing output, 3x3 increased from 78% to 84%, 5x5 from 80% to 84%
[ROCm/composable_kernel commit: a65ef90308]
2019-02-19 11:47:46 -06:00
Chao Liu
f6f3f52faa gridwise_implicit_gemm_convolution_1_chwn_csrk_khwn use khwn for thread C data now
[ROCm/composable_kernel commit: 50b96745c6]
2019-02-17 02:28:20 -06:00
Chao Liu
13388cef47 add anther verision of batch gemm
[ROCm/composable_kernel commit: 1cb9885058]
2019-02-17 01:50:57 -06:00
Chao Liu
8bde1393d4 2-type implicit gemm using chwn
[ROCm/composable_kernel commit: 9f2e8f8bb4]
2019-02-15 22:51:51 -06:00
Chao Liu
64620dfcd4 delete useless code
[ROCm/composable_kernel commit: d7c84daf66]
2019-02-15 22:24:18 -06:00
Chao Liu
c0baa18a3f change file extension to hip.hpp and hip.cpp
[ROCm/composable_kernel commit: b2888adfbe]
2019-02-15 02:13:21 -06:00
Chao Liu
153629655f update build
[ROCm/composable_kernel commit: a414e3fdf8]
2019-02-15 02:06:34 -06:00
Chao Liu
e7f6b820cd hip build
[ROCm/composable_kernel commit: 67c6f73ffe]
2019-02-15 00:54:30 -06:00
Chao Liu
662ba16dc6 refactor build, clean up
[ROCm/composable_kernel commit: e80fbbdd71]
2019-02-14 15:10:16 -06:00
Chao Liu
485d762f55 make LDS double buffer works, 1x1 conv now hits 80% of peak
[ROCm/composable_kernel commit: 28354a0fa3]
2019-02-12 00:57:08 -06:00
Chao Liu
f4ca4c6710 tune for 1x1
[ROCm/composable_kernel commit: 61ac08661d]
2019-02-11 22:36:17 -06:00
Chao Liu
4e28dc5d98 refactor
[ROCm/composable_kernel commit: abf75ac039]
2019-02-11 17:45:22 -06:00
Chao Liu
12ac44917f update with new copy op
[ROCm/composable_kernel commit: 120ab94aa1]
2019-02-07 01:31:00 -06:00
Chao Liu
590a9ff0d3 add lds double buffer for cnhw implicit gemm
[ROCm/composable_kernel commit: 07f16673c9]
2019-02-07 00:56:53 -06:00
Chao Liu
f1bb74f29a unroll some loop, register double buffer gemm
[ROCm/composable_kernel commit: c866773642]
2019-02-06 23:44:21 -06:00
Chao Liu
0d9cae1613 add another blockwise gemm
[ROCm/composable_kernel commit: 1b323316a8]
2019-02-06 23:10:08 -06:00
Chao Liu
7967bb4f94 fixed LDS alignment bug
[ROCm/composable_kernel commit: 5e77650415]
2019-02-06 01:54:13 -06:00
Chao Liu
663703f1e9 bug fixes
[ROCm/composable_kernel commit: 079d63a788]
2019-02-05 23:19:57 -06:00
Chao Liu
008027fef6 refactor
[ROCm/composable_kernel commit: 42f4c7fd56]
2019-02-05 18:04:23 -06:00
Chao Liu
ebf3d27947 add another version of blockwise 2d copy, refactor
[ROCm/composable_kernel commit: 6614729a68]
2019-02-05 17:06:53 -06:00
Chao Liu
4627565374 refactor
[ROCm/composable_kernel commit: 4b616aad52]
2019-02-05 00:51:37 -06:00
Chao Liu
38cb63f129 working on reducing index calculation...
[ROCm/composable_kernel commit: 0741a8ab88]
2019-02-04 17:16:28 -06:00
Chao Liu
2e0ff7d413 refactor
[ROCm/composable_kernel commit: 9bbe9073ab]
2019-02-04 15:40:34 -06:00
Chao Liu
6a96ec4a50 padding works (sort of), but code looks ugly. Tuned some resnet configs
[ROCm/composable_kernel commit: 3439e4b5b7]
2019-01-25 02:50:28 -06:00
Chao Liu
7544bd482b improve implicit gemm NCHW, SRCK, NKHW, and tuned
[ROCm/composable_kernel commit: 8bd6ea1a97]
2019-01-24 22:24:30 -06:00