diff options
Diffstat (limited to 'docs/user_guide/errata.dox')
| -rw-r--r-- | docs/user_guide/errata.dox | 136 |
1 files changed, 136 insertions, 0 deletions
diff --git a/docs/user_guide/errata.dox b/docs/user_guide/errata.dox new file mode 100644 index 0000000000..056e45a432 --- /dev/null +++ b/docs/user_guide/errata.dox @@ -0,0 +1,136 @@ +/// +/// Copyright (c) 2019-2023 Arm Limited. +/// +/// SPDX-License-Identifier: MIT +/// +/// Permission is hereby granted, free of charge, to any person obtaining a copy +/// of this software and associated documentation files (the "Software"), to +/// deal in the Software without restriction, including without limitation the +/// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +/// sell copies of the Software, and to permit persons to whom the Software is +/// furnished to do so, subject to the following conditions: +/// +/// The above copyright notice and this permission notice shall be included in all +/// copies or substantial portions of the Software. +/// +/// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +/// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +/// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +/// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +/// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +/// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +/// SOFTWARE. +/// +namespace arm_compute +{ +/** +@page errata Errata + +@tableofcontents + +@section S7_1_errata Errata + +- (COMPMID-6493) Crash when running Arm Compute Library compiled for SVE2 on a computer that support SVE only. + - Versions: >= v21.02 && <=v23.08 + - OSs: Linux, Android. + - Conditions: + - Compile the latest Arm Compute Library for SVE2 (arch=armv8.6-a-sve2). + - multi_isa = 0 + - Device with SVE but without SVE2 support. + - Result: + - Crash due to illegal instruction. + - To run SVE only, build with arch="armv8.2-a-sve", arch="armv8.6-a-sve", or with multi_isa=1. + +- (COMPMID-6404) Under certain conditions, CLTile may produce incorrect result. + - Versions: >= v19.02 && < v23.08 + - OSs: Linux, Android. + - Conditions: + - The size of the lowest dimension of the input tensor is greater than 16 bytes. + - The size of the lowest dimension of the input tensor is not a multiple of 16. + - Result: + - Incorrect result is produced. + +- (COMPMID-6271) Under certain conditions, CLArgMinMaxLayer validation tests may fail + - Versions Affected: >= v20.02 && < v23.08 + - OSs Affected: Linux + - Conditions: + - Backend: OpenCL + - Axis == 0 + - Result: + - Sporadic mismatches only on certain devices + +- (COMPMID-5324) Issue identified with direct and depthwise convolutions for certain Arm® Mali™ DDK versions. + - Versions Affected: < v22.08 + - Conditions: + - Arm® Mali™ DDK Versions : >= r23p0 && <= r38p0 + - Mali™ GPUs: Bifrost GPU family with the exception of G71 + - Backend: OpenCL + - Build options Include : "cl-fast-relaxed-math" + - Result: Reduced accuracy issue, while using direct and depthwise convolutions fused with LU_BOUNDED_RELU activation. + +- (COMPMID-5134) An issue has been identified when running the graph_deepspeech_v0_4_1 graph example. + - Versions Affected: >= v21.08 + - Conditions: + - Data type input: F32 + - Backend: OpenCL + - Result: The execution of the graph_deepspeech_v0_4_1 could fail on OpenCL backend for systems with a small RAM. The issue is due to the extra temporary memory required to reshape the network weights + +- (COMPMID-4013) Experimented performance regressions for some networks on OpenCL when using Arm® Mali™ DDK r8p0 + - Versions Affected: v21.05 + - OSs Affected: All + - Conditions: + - Arm® Mali™ DDK r8p0 + +- (COMPMID-5146) Under certain conditions, CLFullyConnectedLayer quantized tests may fail due to an issue in the test framework. + - Versions Affected: v21.02 + - OSs Affected: Linux + - Conditions: + - armv7a architecture + - release mode + - asserts enabled + +- (COMPMID-4367) Performance regression in Convolution Layer OpenCL backend on Mali™ G77 when QSYMM8_PER_CHANNEL is used as weights' data type. + - Versions Affected: >= v20.11 && < v21.08 + - OSs Affected: All + - Conditions: + - Mali™ G77 + - Convolution Layer in use + - OpenCL backend + - Convolution Layer uses QSYMM8_PER_CHANNEL as the data type of its weight + +- (COMPMID-4306) A wrong test configuration has been found in CLGEMMMatrixMultiplyReshapedOnlyRHS set of tests. + - Versions Affected: >= v20.11 && < v21.05 + - Conditions: + - Data type input: F32/F16 + - Fused bounded relu activation with coefficient 'a' being negative + +- (COMPMID-5135) Under certain conditions, the validation test case 'CL/DirectConvolutionLayer/Float/FP32/RunSmall9x9\@InputShape=32x37x3x4:StrideX=1:StrideY=1:PadX=0:PadY=0:KernelSize=9:NumKernels=1:DataType=F32:ActivationInfo=LU_BOUNDED_RELU:DataLayout=NHWC' may fail. + - Versions Affected: >= v20.08 + - Conditions: + - The validation suite has to run in nightly mode and execute 40k+ test cases before the test mentioned above + +- (COMPMID-5136) Under certain conditions, benchmark examples can hang when OpenCL profiling queues are enabled. + - Versions Affected: >= v19.11 + - OSs Affected: Linux + - Conditions: + - Arm® Mali™ DDK r1p0 - r8p0, and + - Linux kernel >= 4.4 + +- (COMPMID-5137) On Android with armv8a/armv8.2-a architecture, Arm® Neon™ validation tests can fail when compiled using Android Ndk + >= r18b in debug mode (https://github.com/android/ndk/issues/1135). + - Versions Affected: >= v19.11 + - OSs Affected: Android + - Conditions: + - armv8a/armv8.2-a architecture, and + - Compiled using Android NDK >= r18b in debug mode. + +- (COMPMID-4288) An issue has been identified with CLCast. + - Versions Affected: >= v18.11 && < v21.05 + - Conditions: + - Data type input: F32 + - Data type output: All integer types + - Conversion policy: SATURATE + - Result: OpenCL backend will always wrap around instead of saturating for out-of-range inputs + +*/ +} // namespace |