Gitiles
Code Review
Sign In
review.mlplatform.org
/
ml
/
ethos-u
/
ethos-u-vela
/
refs/tags/2.0.0
tag
ac9056b5a113ecc51481b0040849dbb8d8c95286
tagger
Tim Hall <tim.hall@arm.com>
Mon Nov 30 14:44:06 2020 +0000
object
b27d8957b1778aa99c6b1160032d8ce1f8af35e3
20.11 release.
« Previous
7ad862a
MLBEDSW-3032 Fixed issue in removal of reshapes
by Patrik Gustavsson
· 3 years, 8 months ago
f03bad3
MLBEDSW-2031: LUT support tanh/sigmoid
by Louis Verhaard
· 3 years, 8 months ago
eadb166
MLBEDSW-2885: Fix overflow from inf number
by Andreas Nevalainen
· 3 years, 9 months ago
fa2f92a
MLBEDSW-3035: Updated StridedSlice checks
by Louis Verhaard
· 3 years, 8 months ago
f3d737e
MLBEDSW-2811: Add rescaling to ReLus with different scaling
by Andreas Nevalainen
· 3 years, 8 months ago
0b9c9a3
MLBEDSW-2337: Intermediate feature maps in fast storage
by Louis Verhaard
· 3 years, 9 months ago
8854dc9
MLBEDSW-2788 Fix crash on non-constant weight tensors
by Andreas Nevalainen
· 3 years, 8 months ago
6c97e9a
MLBEDSW-3018 Change to check for avoiding NHCWB16
by Patrik Gustavsson
· 3 years, 8 months ago
9db529a
MLBEDSW-3070: Fix addressing of weights
by Louis Verhaard
· 3 years, 8 months ago
d8c032d
MLBEDSW-2813: Handle non-const weights and check shapes
by Andreas Nevalainen
· 3 years, 9 months ago
d9e38fe
Fix int8/uint8 softmax mul shape
by Fredrik Svedberg
· 3 years, 8 months ago
cb33704
MLBEDSW-1693 Convert batched FC to Conv
by Patrik Gustavsson
· 3 years, 9 months ago
504d6b6
MLBEDSW-2816: Fix assert in scheduler
by Diqing Zhong
· 3 years, 9 months ago
8daf6b7
MLBEDSW-2734: TensorFlow 2.3 compatibility
by Jacob Bohlin
· 3 years, 9 months ago
2b60ce9
[MLBEDSW-2638] Do not use 40bit ACC for softmax int16
by Fredrik Svedberg
· 3 years, 9 months ago
1a66697
MLBEDSW-2809: Redo the Tensor addressing
by Jacob Bohlin
· 3 years, 9 months ago
1356c2a
MLBEDSW-2377: Greedy allocator improvement
by Louis Verhaard
· 3 years, 9 months ago
c3862c2
MLBEDSW-3008: Replace Split ops with Identity
by Dwight Lidman
· 3 years, 9 months ago
2f6f379
[MLBEDSW-2845] Improve unit test coverage of fp_math
by Fredrik Svedberg
· 3 years, 9 months ago
55d9e33
MLBEDSW-2816: remove constant array from SRAM
by Diqing Zhong
· 3 years, 9 months ago
8dbf8cf
MLBEDSW-2745 Support relus with differing scales
by Michael McGeagh
· 3 years, 9 months ago
fa4cb29
MLBEDSW-2994 Remove undesired reshape OPs
by Patrik Gustavsson
· 3 years, 9 months ago
e1cc3de
MLBEDSW-2813: Handle None tensor when parsing operators
by Andreas Nevalainen
· 3 years, 9 months ago
2b7a162
vela: SHRAM alloc too big for some elementwise ops
by Tim Hall
· 3 years, 9 months ago
11b0bdb
vela: Refactor op sets in graph_optimiser
by Michael McGeagh
· 3 years, 9 months ago
34ad19b
optim: Fix issue with IFM streaming of LUT
by Michael McGeagh
· 3 years, 9 months ago
98a3499
MLBEDSW-2935: LUT fusing with preceding operator
by Louis Verhaard
· 3 years, 9 months ago
515c956
MLBEDSW-2874 Fix writing out empty tflite files
by Michael McGeagh
· 3 years, 9 months ago
835d8e1
[MLBEDSW-2928] Add batching to softmax
by Fredrik Svedberg
· 3 years, 9 months ago
e5cf95b
MLBEDSW-2819 Removed CLI-option permanent-storage
by Patrik Gustavsson
· 3 years, 9 months ago
271ddc3
MLBEDSW-2814 Add support for inferred size in SplitV
by Patrik Gustavsson
· 3 years, 9 months ago
0628a8c
MLBEDSW-2567: CLI option to specify allocation alignment
by Jacob Bohlin
· 3 years, 9 months ago
d2e3355
[MLBEDSW-2905] Fix int16 multiplier saturation
by Fredrik Svedberg
· 3 years, 9 months ago
90033f3
MLBEDSW-2906: Only add padding fields to Npu ops
by Jacob Bohlin
· 3 years, 9 months ago
438e563
MLBEDSW-2903 Split mapping to tensor
by Patrik Gustavsson
· 3 years, 9 months ago
1.2.0
1.2.0.rc3
c3c08d8
MLBEDSW-2804: Removed invalid assert for shift range
by Jacob Bohlin
· 3 years, 9 months ago
258ebba
MLBEDSW-2915: Added None check for bias before checking restrictions
by Jacob Bohlin
· 3 years, 9 months ago
749bfd5
MLBEDSW-2919: bringup: int16 avgpool diff on FPGA
by Tim Hall
· 3 years, 9 months ago
7152517
MLBEDSW-2804: Bias data type check wrong
by Tim Hall
· 3 years, 9 months ago
d775e37
MLBEDSW-2900: softmax output diff on FPGA
by Tim Hall
· 3 years, 9 months ago
237d72d
[MLBEDSW-2900] zero point fix for int32 ifm
by Fredrik Svedberg
· 3 years, 9 months ago
d7911c4
MLBEDSW-2688: LUT calculation with different in/out scale
by Louis Verhaard
· 3 years, 9 months ago
7201932
MLBEDSW-2411: Update README with example networks
by Tim Hall
· 3 years, 10 months ago
5e990c8
MLBEDSW-2889: NHCWB16 format issue at end of subgraph
by Tim Hall
· 3 years, 9 months ago
49d9212
MLBEDSW-2804: Added bias data type check
by Jacob Bohlin
· 3 years, 10 months ago
99fcb89
MLBEDSW-2839 Update release notes for 1.2.0
by Michael McGeagh
· 3 years, 9 months ago
bb1a92a
[MLBEDSW-2857] Fix serialisation of non byte tensors
by Fredrik Svedberg
· 3 years, 9 months ago
880e735
[MLBEDSW-2846] Do not use NHCWB16 for reduce_sum int32
by Fredrik Svedberg
· 3 years, 9 months ago
a41cd4d
Small fix for Softmax regression
by Jacob Bohlin
· 3 years, 9 months ago
2abd3dd
MLBEDSW-2786: Fix IFM order in binary operation
by Diqing Zhong
· 3 years, 9 months ago
f0c59bf
MLBEDSW-2719: Update kernel size check
by Andreas Nevalainen
· 3 years, 9 months ago
ba69518
MLBEDSW-2686: Use NPU tensor format for noop reshapes.
by Tim Hall
· 3 years, 9 months ago
1.2.0.rc2
e55e274
Update to HI 1.0.6
by Stefan Nannesson
· 3 years, 10 months ago
58520b9
MLBEDSW-2688: use LeakyRelu for int16
by Louis Verhaard
· 3 years, 9 months ago
90831bc
MLBED-2822 Added CLI-opt for weight size est.
by Patrik Gustavsson
· 3 years, 9 months ago
d47cc27
MLBEDSW-2847: Fix for TransposeConv crash and u8 output diff
by Jacob Bohlin
· 3 years, 9 months ago
814cfbb
MLBEDSW-2688: LUT DMA may require kernel wait
by Louis Verhaard
· 3 years, 10 months ago
7579c75
MLBEDSW-2867: Split operators get placed on CPU
by Dwight Lidman
· 3 years, 9 months ago
87c1350
MLBEDSW-2654: Convert Resizebilinear to a number of 2x2 pools
by Charles Xu
· 3 years, 10 months ago
b9fc33c
MLBEDSW-2688: LeakyRelu rewrite to LUT or MUL/MAX
by Louis Verhaard
· 3 years, 10 months ago
e9194df
MLBEDSW-2646: Refactor unknown operator serialisation
by Tim Hall
· 3 years, 10 months ago
1.2.0.rc1
e3786ac
MLBEDSW-2679: Tensor quant comparison is incorrect
by Tim Hall
· 3 years, 10 months ago
67e0d8f
MLBEDSW-2663: Handle optional tensors
by Jacob Bohlin
· 3 years, 10 months ago
1575b94
[MLBEDSW-2730] Implement LUT generation for softmax uint8/int8
by Fredrik Svedberg
· 3 years, 10 months ago
1cdc467
Added a lower bound for the valid range of shift
by Jacob Bohlin
· 3 years, 10 months ago
6c87807
MLBEDSW-2664 Clarify help for CLI-opt permanent-storage
by Patrik Gustavsson
· 3 years, 10 months ago
9baa4c3
MLBEDSW-2822 Account for NHCWB16 in scheduler est.
by Patrik Gustavsson
· 3 years, 10 months ago
f899231
MLBEDSW-2611: Update global scale for 16 bit to tanh and sigmoid
by Charles Xu
· 3 years, 10 months ago
bb1b09e
MLBEDSW-2783 Vela crashed on empty tflite file
by Michael McGeagh
· 3 years, 10 months ago
a3aad93
MLBEDSW-2824: Add mapping for ROUND operator
by Dwight Lidman
· 3 years, 10 months ago
b16dcce
vela: Minor refactor of operation class
by Tim Hall
· 3 years, 10 months ago
e5b0163
MLBEDSW-2683: Neural Network MACs is wrong
by Tim Hall
· 3 years, 10 months ago
fb85873
MLBEDSW-2728: Only insert primary op for NPU ops
by Jacob Bohlin
· 3 years, 10 months ago
d2a972d
MLBEDSW-2731: Allow all TensorFlow Lite operators to pass through
by Dwight Lidman
· 3 years, 10 months ago
2573311
MLBEDSW-2729: Add restrictions for shapeless tensors
by Dwight Lidman
· 3 years, 10 months ago
fa34c6f
MLBEDSW-2636 Prevent DMA of weight to Sram in some cases
by Patrik Gustavsson
· 3 years, 10 months ago
eebb1c2
MLBEDSW-2779 Avoid NHCWB16 in some SplitSliceRead cases
by Patrik Gustavsson
· 3 years, 10 months ago
597fd3f
[MLBEDSW-2657] Softmax uint8/int8
by Fredrik Svedberg
· 3 years, 10 months ago
30cb47a
MLBEDSW-2718: Fix a bug that would reshape TransposeConv input
by Jacob Bohlin
· 3 years, 10 months ago
940fdee
MLBEDSW-2589: Skip weight compression for CPU ops
by Dwight Lidman
· 3 years, 10 months ago
29d568e
MLBEDSW-2779 Consider num dimensions, in check for NHCWB16
by Patrik Gustavsson
· 3 years, 10 months ago
289a41d
Vela: Rework NPU/DMA dependency insertion (for MLBEDSW-2620)
by Tim Hall
· 3 years, 10 months ago
f767b93
MLBEDSW-2732: Added complex64 to datatypes
by Jacob Bohlin
· 3 years, 10 months ago
0b8268a
MLBEDSW-2688: Improved LUT support
by Louis Verhaard
· 3 years, 10 months ago
458a208
MLBEDSW-2570 Avoid usage of NHCWB16 for some cases
by Patrik Gustavsson
· 3 years, 10 months ago
be733cf
MLBEDSW-2639: Remove reverse_op_order attribute
by Jacob Bohlin
· 3 years, 10 months ago
bf61268
MLBEDSW-2755: Added check that ifm2_tensor is set
by Jacob Bohlin
· 3 years, 10 months ago
8b24f2b
[MLBEDSW-2749] removed the decorator for typecheck
by Manupa Karunaratne
· 3 years, 10 months ago
1a9d20e
MLBEDSW-2726: Vela crashes when marking tensor with TensorPurpose.Unknown
by Dwight Lidman
· 3 years, 10 months ago
7ad408b
MLBEDSW-2586: Null check before accessing tensor resampling mode
by Dwight Lidman
· 3 years, 10 months ago
c0bb899
MLBEDSW-2696 Fix Sram exceeded for Sram spilling
by Patrik Gustavsson
· 3 years, 10 months ago
89a6bbf
MLBEDSW-2681: Ceiling the upscale for OFM/IFM
by Charles Xu
· 3 years, 10 months ago
de2a57f
MLBEDSW-2684: Fix weight compression scale calculations for FC
by Jacob Bohlin
· 3 years, 10 months ago
8d3216f
vela: Remove redundant import, reuse existing func
by Michael McGeagh
· 3 years, 10 months ago
c5b549b
MLBEDSW-2637 Utilise new tensor and operator funcs
by Michael McGeagh
· 3 years, 10 months ago
5778ffd
MLBEDSW-2637 Refactor util funcs out of softmax.py
by Michael McGeagh
· 3 years, 10 months ago
22f74e1
MLBEDSW-2383 Preserve previous metadata
by Michael McGeagh
· 3 years, 10 months ago
e99b893
MLBEDSW-2639: Moved the IFM/IFM2 order switch to register cmd stream generator
by Jacob Bohlin
· 3 years, 11 months ago
ecd2052
MLBEDSW-2549 Crash with incorrect strided slice op
by Michael McGeagh
· 3 years, 10 months ago
36ffaf3
Skip the NOP resizebilinear op
by Charles Xu
· 3 years, 10 months ago
Next »