Age | Commit message (Collapse) | Author |
|
NEGEMMLowpOffsetContributionOutputStageKernel
This change reduces the core's library size by 191Kb.
Change-Id: Ifb8eb0d7f8bc7713f2368803a62a4c9277cc5c87
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3439
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Change-Id: I315520ca825e2a420bb28308f020b95ff2969042
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3341
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
|
|
Added support on NEArithmeticAddition and NEArithmeticSubtraction
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Change-Id: Ifa805f8455ef6eff1ee627752dc1c7fe9740ec47
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3451
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
|
|
* Add positive/negative API validation tests for both CL/NEON
* Add unit-tests for both CL/NEON
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: Ie33df7a9d30eaccad69dd6c3b48a305bcc023bfc
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3441
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Change-Id: I2102bfe95c2c2335bb587842f9d860cf939a9026
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3315
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Change-Id: I631c566fdf38ff0388b38e71549a1291d351f8b6
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3426
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Change-Id: I208287b44ece051e95f891d43a691cb0ac6e56c5
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3419
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Change-Id: Id13d4f9860f25eeb8c4188ac9248653c4f7c1dc2
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3349
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
COMPMID-3321: Add cl_image support for GEMMReshaped NT_T
- Added support for cl_image in CLGEMMMatrixMultiplyReshapedKernel (both
NT and T kernels)
- Extended the tests for the validating rhs_info.export_to_cl_image =
true
- Added utility macros in OpenCL to load data from a OpenCL image object
- Updated doxygen documentation in CLGEMMMatrixMultiplyReshapedKernel.h
- Updated doxygen documentation in CLGEMMReshapeRHSMatrixKernel.h
Change-Id: I953b10e4ef205d1b76dcbc366e5a91fd5a8e1d5c
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3329
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
|
|
NEArithmeticAdditionKernel"
This reverts commit 4a61653202afb018f4f259d3c144a735d73f0a20.
Reason for revert: We will allow in-place computations by providing the same input1 (or input2) as output, thus avoiding changes in the interface.
Change-Id: I7c8669e207e15731dc26dc366150bf960508a879
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3035
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Change-Id: I6352a520ce38230cdfbad346b176cb659ab242a7
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3327
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
different tensors.
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I98782bb73e9dc0899ffb1796aca6f99714adea94
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3343
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
|
|
Add sub/exp/splitv support in graph api
Signed-off-by: Sheri Zhang <sheri.zhang@arm.com>
Change-Id: I4e08cc19a46655717068b12c93d67e619a595d9a
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3309
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Change-Id: I8168cea5056ff48a0253ebb8c88ea549a3ea69a2
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3335
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
|
|
Change-Id: I5677c87bba97dd395a3e13dbce34a3dd2c437033
Signed-off-by: morgolock <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3289
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Change-Id: I0089657dd95d7c7b8592984def8e8de1d7e6d085
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3308
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
- Make Scale and ScaleKernel (on CL and GLES) use ScaleKernelInfo
- Deprecate previous configure/validate functions on NEON, CL and GLES
- Make adjustments required by deprecation
Change-Id: I7e81f4ee9ae919392137b92f91e9bc002b7ae277
Signed-off-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3317
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
|
|
* Move helper function dim_index_2_num_dims from header to inl file
Signed-off-by: SiCong Li <sicong.li@arm.com>
Change-Id: I204756997316eae8cc2c9d4ba6a91d5974c75965
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3325
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Change-Id: I5568ca1e9383f3960886f211d8320fc4525a8804
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3326
Reviewed-by: James Conroy <james.conroy@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
|
|
and CLSoftmaxLayer
* [Interface change] "axis" argument is renamed to "reduce_end_axis"
* Unify the meaning of "axis"(now "reduce_end_axis") to be the last axis
of the first n dimensions (inclusive)to reduce.
This way the meaning of reduce_end_axis stays the same for both
positive and negative values: it selects a dimension before which all
dimensions (including the selected dimension) are reduced.
Change-Id: I4ab03bd8360b1cd8cac4998df0b1571064a9d4ed
Signed-off-by: SiCong Li <sicong.li@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3278
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I9be23e6ef1f552eb159e39fda16c82fa20124094
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3307
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
- Removes padding from NEThresholdKernel
- Alters configuration interface to use a descriptor
Change-Id: I394d5e1375454813856d9d206e61dc9a87c2cadc
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3300
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
|
|
- Both NEON and CL's Scale Kernel now supports aligned corners
with nearest neighbor interpolation
- Unsupported combination (center sampling policy with aligned
corners) now fails on validation
- Test suites for CL and NEON are pruned by removing unsupported
combination
Change-Id: Ieea4f145a131593b89b471dcec2b09619136b17c
Signed-off-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3297
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Change-Id: Ib96b7c8e86311acc1859c44afa3731e56373aa89
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3299
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
NEScaleKernel and NEScale are modified to use the
new kernel descriptor.
Change-Id: I32e3e2a5209a38f5bf32c43f64b2aa4430825988
Signed-off-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3239
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Added padding requirement in CLGEMMReshapeRHSMatrixKernel in order to
create 2d image from a cl_buffer.
Test extended in order to validate the padding requirement
Change-Id: I36bcaf3e9299ee186602b4e3456851cc8cda6ce6
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3292
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
|
|
NEROIAlignLayerKernel
Change-Id: I19106e4bb0f6886a23dccc5c71e864a8364f40f9
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3288
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
|
|
Change-Id: I5164db1fa4cef003875647ed01879c5dbdfb9b35
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3286
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Change-Id: Ia356b3545e01ccc98056951f3a20ed2712240dac
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3283
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
- Added flag to export to cl_image in GEMMRHSMatrixInfo
- Returned an error in the GEMM/Lowp kernels without this support
Change-Id: I4a523d93c0984626bbf23e2efeb114f9c7c20a24
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3274
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
|
|
ndrange.hpp file is needed by arm_gemm_compute_iface.hpp when building
armnn.
In a cross compilation environment like Yocto Openembedded, the installed
include files comes from either /arm_compute, /include or /support
directories.
So to avoid armnn build failure, ndrange.hpp need to be moved in
arm_compute/core/NEON/kernels/arm_gemm directory.
Other files that need ndrange.hpp to build are updated accordingly.
Signed-off-by: Vincent ABRIOU <vincent.abriou@st.com>
Change-Id: Icc8e18b23dd06c56e0c49a84d4a51bfd85fe2290
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3271
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
|
|
Change-Id: Ic4d20995d6c6bb76d07113e86247bad2722e4e83
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3244
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Removed most of the templates and refactored the code.
Performance is the same but the libary size dropped by
52Kb.
Change-Id: I41ff0c0853c923d925cdaeb05f4a58c9086fff94
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3190
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
COMPMID-3487: Refactor NEArithmeticSubtractionKernel
Refactored code in order to remove paddings. This resulted in
a big increase in libary size so after some rework the total size
dropped by 4Kb.
Change-Id: I4e3014c2ae49c29c6090b195ea16620afcf6c09f
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3206
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
|
|
Copy of missing parameters is added to the build function.
Change-Id: I06d665729a61f2f01dcfab21c6e6d55359fc81cb
Signed-off-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3226
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: James Conroy <james.conroy@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
|
|
Change-Id: Ia6c2f115849889baceafaf716477456e41f96037
Signed-off-by: morgolock <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3186
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Sang-Hoon Park <sang-hoon.park@arm.com>
|
|
Change-Id: Ied11a4a3e9d04615a1a1f0bfa552f3dd8293a170
Signed-off-by: morgolock <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3178
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Removed is_per_channel template arguments since it wasn't used
anywhere and also made has_biases a runtime parameter. The total
size reduction from this change is 28.6kb.
Change-Id: I292ac27ae3ea2885b8438f613390486323982664
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3189
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
|
|
This patch introduces CL kernel cache feature for
NN runtimes such as ARMNN or other in-house versions.
Originally, this feature came from utils/Utils.cpp and utils/Utils.h
files but was used just for testing.
Caching CL kernel binaries is mandatory and required for real product
because inference performance at first run is one of very important
factors. This patch exposes two API - save_program_cache_to_file
and restore_program_cache_from_file so that NN runtimes can use
these API to cache the compiled CL kernel binaries.
Changelog v1.
- Reordered including headers suggested by Manuel Bottini.
- Updated description.
Changelog v2.
- Drop function descriptions.
Changelog v3.
- Fix build error with opencl = 0. As for this, it moves two functions,
save_program_cache_to_file and restore_program_cache_from_file,
to src/runtime/CL/Utils.cpp and arm_compute/runtime/CL/Utils.h,
and set ARM_COMPUTE_CL flag only when opencl = 1.
This change makes CL kernel cache feature to depend on OpenCL support.
Changelog v4.
- Fix coding_standard_checks error.
Changelog v5.
- Add a new cpp file to Android.bp to fix coding_standard_checks error.
Changelog v6.
- rebased on top of master, and dropped _new_built_programs_map object
so it adds compiled CL kerne binaries per model to a given file.
Changelog v7.
- Correct descriptions.
- Correct arm_compute/runtime/CL/Utils.h file description.
- Correct src/runtime/CL/Utils.cpp file description.
- Drop ARM_COMPUTE_CL ifdef ~ endif from examples/cl_cache.cpp
- Do not add 'ARM_COMPUTE_CL' flag at SConscript.
Change-Id: I8c640fb1809ca79191daf5b1ff72a50dd492c36d
Signed-off-by: Inki Dae <inki.dae@samsung.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2914
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Change-Id: I20680dc74a3d709297539e2132417308a7aecc9d
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3159
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Removed most of the templates used in this kernel. This
resulted in a reduction of 35Kb.
Change-Id: I0d50aa769b361790d47a8017d795f2c92c6d8a6f
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3158
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Sang-Hoon Park <sang-hoon.park@arm.com>
|
|
Normalization with the maximum value was causing results to wrap-around
As a work-around we use saturating intrinsics to perform the operation
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I719b7ac7ad274dc2ae339bc4a055f9200134ed97
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3184
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
The followings are essential to make it work
- QSYMM16 is added as supported data type in CLGEMMLowpOutputStage
- Internal TensorCopyKernel is added similar to NEQLSTMLayer
The followings are fix for related things.
- Projection is modified to remove copy of projection_bias from
NEQLSTMLayer.
- Fix wrong argument for validate_mm()
- validate_mm() now returns on error.
Change-Id: Icbd04e9fdb8821eb41dd3e0a6a0980965b779714
Signed-off-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3177
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
|
|
Signed-off-by: Sheri Zhang <sheri.zhang@arm.com>
Change-Id: I285cc3b74ac0a45f0ad5830baed5237cea568f15
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3147
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Change-Id: I895b697c89c9a7509d48a54ac1effb7fbd8cca19
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3174
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Sang-Hoon Park <sang-hoon.park@arm.com>
|
|
* Retain data layout for SpaceToBatchLayerKernel and BatchToSpaceLayerKernel
Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Change-Id: I5bd0c3ca90a4964eda9f838c97c784aced0f4579
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3166
Reviewed-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
- Using NEDirectConvolution for big shapes since the memory
required explodes for 9x9 kernel
- Adding test cases
- Fix enables only the NEON Deconvolution for NHWC
Change-Id: I8a541346428e5686818f8ecb7f69e2a9106cbceb
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3135
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Sang-Hoon Park <sang-hoon.park@arm.com>
|
|
- Peephole and projection has been fixed to be working
- Small internal kernel copying data between tensors to
cover the case where num_units and output_size is different
is added.
Below is strictly outside of this patch's scope but are changes
helping this patch working (directly or indirectly) or making
NEQLSTM more complete.
- Consideration for layer normalization is added to InfoHelpers
- QSYMM8 data type is added to helper function to
print out tensors.
- NE/CLLSTMLayer::validate() logic has been modified to use correct
value for shape validation.
Change-Id: I40b4e71dfdbe8432caa2fe4a9af60a725362cc33
Signed-off-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3157
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
Currently 1D ranges of work are specified by the scheduler
via two integers, start and end. This limit opportunities
for advance parallelism and scheduling
This patch expands the interfaces to allow for ND parallism.
`GemmCommon::get_window_size` now returns an `NDRange` specifying the work
in N-dimensions rather than with the single integer it used prior (1D)
Execute now takes an `NDCoordinate` which specifies an `NDRange` with a start
position for that work along with an `NDCoordinate` to specify the thread location
In addition to expanding the interface to enable this functionality,
we have added the capability to SGEMM when the number of threads is high
this has the effective of allowing a much greater degree of parallelism
where te problem dimension would previously have limited the number of threads.
Change-Id: I3e1a8b7276216627bec4ff6f24ac2147552ea9fb
Signed-off-by: Joseph Dobson <joseph.dobson@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2962
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|
|
* Added support for pooling indices in NHWC Poolsize 2x2
Change-Id: Ib2a3468e794f58bbf2c03aba9f6b184b9d76b183
Signed-off-by: morgolock <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2997
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
|