Tencent
Unsupported data format for device acc: Transpose input or reshape output
1. 环境（environment）
- Build OS and Version: Ubuntu 20.04
- RunTime OS Version: Android
- RunTime DEVICE: OPENCL
2. Github版本
- branch：master
- commit(optional):
3. 编译方式(compile method) CMake完整编译参数(full cmake arguments) ABI="arm64-v8a" STL="c++_static" PROFILING="OFF" CLEAN="" DEVICE_TYPE="" INPUT_DATA_TYPE=0 MODEL_TYPE=TNN USE_NCNN_MODEL=0 KERNEL_TUNE="-et" THREAD_NUM=1 ADB=adb BENCHMARK_TYPE="APP" BENCHMARK_APP_DIR=benchmark_app/benchmark/
WORK_DIR=pwd
BENCHMARK_MODEL_DIR=$WORK_DIR/../benchmark-model
BUILD_DIR=build
BUILD_APP_DIR=build_app
ANDROID_DIR=/data/local/tmp/tnn-benchmark
ANDROID_DATA_DIR=$ANDROID_DIR/benchmark-model
OUTPUT_LOG_FILE=benchmark_models_result.txt
LOOP_COUNT=16
WARM_UP_COUNT=5
INTERVAL=5
benchmark_model_list=( mtinet.opt.tnnproto ) 4. 编译日志(build log) E/tnn: tnn::Status tnn::OpenCLLayerAcc::CheckBlob(const std::vector<Blob *> &, const std::vector<Blob *> &) [File /home/mzz/TNN/source/tnn/device/opencl/acc/opencl_layer_acc.cc][Line 451] Resolve Layer(Transpose_962)-Input Blob(2475) Data Format(0) failed E/tnn: virtual tnn::Status tnn::OpenCLPermuteLayerAcc::Reshape(const std::vector<Blob *> &, const std::vector<Blob *> &) [File /home/mzz/TNN/source/tnn/device/opencl/acc/opencl_permute_layer_acc.cc][Line 103] code: 0x9004 msg: unsupported data format for device acc E/tnn: bool tnn::test::CheckResult(std::string, tnn::Status) [File /home/mzz/TNN/test/test.cc][Line 338] create instance failed: code: 0x9004 msg: unsupported data format for device acc /data/local/tmp/tnn-benchmark/benchmark_models_result.txt: 1 file pulled, 0 skipped. 21.7 MB/s (111508 bytes in 0.005s) ZTE A2020N2 Pro
benchmark device: OPENCL
I/tnn: tnn::Status tnn::OpenCLRuntime::Init() [File /home/mzz/TNN/source/tnn/device/opencl/opencl_runtime.cc][Line 120] OpenCL version: CL_TARGET_OPENCL_VERSION 200 CL_HPP_TARGET_OPENCL_VERSION 110 CL_HPP_MINIMUM_OPENCL_VERSION 110 I/tnn: tnn::Status tnn::OpenCLRuntime::Init() [File /home/mzz/TNN/source/tnn/device/opencl/opencl_runtime.cc][Line 155] Create common opencl context
Fri Jul 22 14:08:00 CST 2022
5. 详细描述bug 情况 (Describe the bug)
Resolve Layer(Transpose_962)-Input Blob(2475) Data Format(0) failed
6. bug所在模型代码
7. tnnproto
模型压缩在附件里了，感谢！
------------------ 原始邮件 ------------------ 发件人: "Tencent/TNN" @.>; 发送时间: 2022年7月23日(星期六) 下午3:05 @.>; @.@.>; 主题: Re: [Tencent/TNN] Unsupported data format for device acc: Transpose input or reshape output (Issue #1737)
方便提供一下 tnn 模型吗，我们复现一下这个问题看看
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
从QQ邮箱发来的超大附件
mtinet.zip (60.74M, 2022年08月22日 15:31 到期)进入下载页面：http://mail.qq.com/cgi-bin/ftnExs_download?t=exs_ftn_download&k=0e386661184600c511a965ef10660616530003530352030c1c085252034b0000545c4b00545701140109505007540609540b5705366c34544551080442484e5041385b&code=18fa6f49
我这边使用附件里的模型没有复现这个问题，看了一下模型结构，和 issue 里提到的有点不太一样，辛苦确认一下模型 附件里的模型，softmax 的 axis 大于了 3，目前 tnn 暂时不支持，需要把维度压缩一下