root
/
MNN
mirror of https://github.com/alibaba/MNN.git
1
0
Fork 0
Code Issues Actions 2 Packages Projects Releases Wiki Activity
787 Commits 20 Branches 109 Tags 396 MiB
Commit Graph

365 Commits

Author SHA1 Message Date
gdh1995 1eaa7a632b fix Reduce may crash if it follows Reshape 
When converting a caffe model, `ReductionTransform` thinks
there're always 4 dims in the input of a `Reduction` layer.

However, when forwarding `OpCommonUtils::computeReduceDims`
had an assumption that dim numbers wouldn't exceed
`inputs[0]->dimensions()`.

So, MNN would write unexpected addresses and then crash,
if a Reduction op follows a Reshape / InnerProduct op,
since both ops may output a tensor of 2 dimensions.

_Comment_: I know `TransformInnerProduct` often converts caffe's
InnerProduct into 1x1 Convolution, but Reshape can also cause a
tensor has only 2 dimensions, so this still deserves a fix.
 2020-12-19 02:42:07 +08:00
Hui Shu ab711d484c Synchronize internal master to Github 2020-12-15 14:12:35 +08:00
DaydreamCoding d4316ada8f
fix threadpool may not destroyed 2020-12-03 15:25:45 +08:00
xiaying 29a39f1c10 [MNN:Sync] Sync internal gitlab 2020-11-25 19:03:07 +08:00
tianbu.xsw e784b3981d [PATCH 77/78] delete unuseful code 2020-11-25 18:57:56 +08:00
tianbu.xsw a87431f67e [PATCH 76/78] [MNN OpenCL] sync cpu2gpu data copy temporarily 2020-11-25 18:57:56 +08:00
雁行 a7acf0e90e [PATCH 71/78] [MNN:Bugfix] fix bug of fuseRegion 2020-11-25 18:57:56 +08:00
雁行 08d6ae5ee7 [PATCH 70/78] [MNN:Bugfix] fix bug of ArgMax whith topk and value 2020-11-25 18:57:56 +08:00
xiaying aecb905947 [PATCH 67/78] [MNN:Bugfix] Fix bug for ShapeConcat and OnnxTopK 2020-11-25 18:57:55 +08:00
tianbu.xsw a8ec4ca4fc [PATCH 63/78] [MNN Bugfix] fix call backup when create CL wrong 2020-11-25 18:57:55 +08:00
xiaying 2b9e9e75ca [PATCH 62/78] [MNN:Bugfix] Add depthwise convint8 in imageop 2020-11-25 18:57:55 +08:00
xiaying 703697d720 [PATCH 61/78] [MNN:Refractor] move unuseful code to backupcode 2020-11-25 18:57:55 +08:00
xiaying e53d24db47 [PATCH 60/78] [MNN:Bugfix] Compability for old model's nearest don't support half pixels 2020-11-25 18:57:55 +08:00
xiaying ed794cd29d [PATCH 59/78] [Metal:Refractor] Avoid threadgroup < 64 error 2020-11-25 18:57:55 +08:00
xiaying 2f198233be [PATCH 58/78] [Metal:Speed] Optimize MetalConvolution1x1 by replace z unrool to x unrool 2020-11-25 18:57:55 +08:00
xiaying 7874581431 [PATCH 57/78] [Metal:Speed] Optimize thread group set 2020-11-25 18:57:55 +08:00
xiaying 79837b7e1d [PATCH 56/78] [Metal:Speed] MetalConvolutionGEMM small opt 2020-11-25 18:57:55 +08:00
tianbu.xsw 7967511f39 [PATCH 54/78] delete when create fail 2020-11-25 18:57:55 +08:00
tianbu.xsw 88e8c9e47d [PATCH 53/78] [OpenCL bugfix] back to backupBackend when create fail 2020-11-25 18:57:54 +08:00
xiaying cf7aa5f8ef [PATCH 52/78] [OpenCL:Bugfix] Fix bug for clGetPlatform not check error 2020-11-25 18:57:54 +08:00
xiaying d19adf3ed2 [PATCH 51/78] [MNN:Refractor] Solve compile warning in Session 2020-11-25 18:57:54 +08:00
xiaying 816a455c23 [PATCH 46/78] [Metal:Bugfix] Fix metal crash bug on iphone6p 2020-11-25 18:57:54 +08:00
xiaying aba789b391 [PATCH 45/78] [Metal:Bugfix] Fix crash bug for metal reduce not support type and multi convolution input 2020-11-25 18:57:54 +08:00
xiaying ec08d24f0d [PATCH 44/78] [Metal:Refractor] Reduce memory alloc for MetalBackend.mm 2020-11-25 18:57:54 +08:00
xiaying b0c881c698 [PATCH 43/78] [Metal:Feature] Support count memory for metal 2020-11-25 18:57:54 +08:00
xiaying b4e546abdf [PATCH 42/78] [Metal:Refractor] Temply remove metal profile for op 2020-11-25 18:57:54 +08:00
xiaying bce4b3a3bc [PATCH 41/78] [Metal:Speed] Reduce metal encoder and buffer alloc, delete unuseful code 2020-11-25 18:57:53 +08:00
xiaying 8483942c17 [PATCH 40/78] [Metal:Speed] Optimize MetalRaster's onCopyBuffer 2020-11-25 18:57:53 +08:00
xiaying 56e3557254 [PATCH 39/78] [Metal:Bugfix, Tools:Bugfix] Fix bug for origin quan compute error 2020-11-25 18:57:53 +08:00
xiaying 9c42dd49ff [PATCH 38/78] [Metal:Bugfix] Fix memory leak for MNNMetalContext's commit 2020-11-25 18:57:53 +08:00
xiaying bc86f31fe1 [PATCH 37/78] [Metal:Speed] Optimize deconvolution of metal 2020-11-25 18:57:53 +08:00
xiaying 3560e2b970 [PATCH 36/78] [Metal:Speed] Optimize Metal Reduce 2020-11-25 18:57:53 +08:00
xiaying b2a0c63917 [PATCH 35/78] [Metal:Speed] Optimize metal scale's execute 2020-11-25 18:57:53 +08:00
xiaying 5e770d5576 [PATCH 34/78] [Metal:Bugfix] Fix bug for MetalEltwise outof range 2020-11-25 18:57:53 +08:00
xiaying c8162015d8 [PATCH 33/78] [Metal:Speed] move metal op's allock into onResize 2020-11-25 18:57:53 +08:00
xiaying ab69f80723 [PATCH 32/78] [Metal:Speed] Optimize raster by move compute size into onResize, metal 2020-11-25 18:57:53 +08:00
xiaying 10fa9fc483 [PATCH 31/78] [Metal:Speed] Optimized MetalPooling by move operate into onResize 2020-11-25 18:57:53 +08:00
xiaying 80963bd19a [PATCH 30/78] [Metal:Speed] Support Metal C4 opt for raster 2020-11-25 18:57:53 +08:00
xiaying 14510593f8 [PATCH 26/78] [MNN:Speed] Add avx2 expc8 2020-11-25 18:57:52 +08:00
xiaying 1bd8d27131 [PATCH 25/78] [MNN:Speed] Add asm for avx int8 2020-11-25 18:57:52 +08:00
tianbu.xsw 466ed6d52c [PATCH 19/78] [MNN METAL] fix unary precision 2020-11-25 18:57:52 +08:00
riddick 7cc4ddc585 feat(backend): add M1 chip support 
- add cpu backend support for M1 chip
- add arm82 backend support for M1 chip
 2020-11-24 22:48:48 +08:00
jxt1234 5be85d7177
Merge pull request #1230 from zjd1988/master 
fix log error for cuda
 2020-11-24 10:47:14 +08:00
tzongw 440f570b4b fix newSrc out of range error 2020-11-23 16:58:55 +08:00
xiaying 81df3a48d7 Fix compile bug for old gcc version 2020-11-22 10:26:29 +08:00
zjd1988 357795a949 vector index should not equal to vector size 2020-11-18 10:17:09 +08:00
xiaying d47849cb5e Fix bug for MetalInterp compute error for pytorch half pixels 2020-11-13 15:26:31 +08:00
xiaying 9198251290 Optimize ShapeExpandDims 2020-11-13 15:26:20 +08:00
zjd1988 b7c8390068 fix log error 2020-11-13 09:01:15 +08:00
xiaying 6e0e9e95cb Fix memory leak for serveral model, the children don't release parent 2020-11-09 21:52:39 +08:00