[Runtime] MISRA-C compliant TVM runtime by liangfu · Pull Request #3934 · apache/tvm

liangfu · 2019-09-11T11:14:19Z

This PR implements MISRA-C compliant TVM runtime proposed in #3159 .

See timing improvements in the following table:

	Create	SetInput	Run	GetOutput	Destroy
Existing Runtime	4.75 ms	0.50 ms	8.17 ms	0.01 ms	0.27 ms
Proposed Runtime	3.40 ms	0.25 ms	7.04 ms	0.00 ms	0.30 ms

@tqchen @ajtulloch @nhynes Please review.

tqchen · 2019-09-13T20:38:05Z

also cc @ajtulloch Now that we have quite a few runtimes, I wonder if it makes sense to consolidate some of them, for example, i do think it makes sense to make uTVM runtime MISRA-C compliant

liangfu · 2019-09-21T06:14:14Z

This PR also contains a rewritten version of JSON parser and NDArray reader, which is relative independent and concise. I would like to propose a replacement for the picojson that is currently used as a third-party dependency (introduced in #3567) . Does this make sense ?

ajtulloch · 2019-09-25T01:01:40Z

@liangfu absolutely, this would reduce code size substantially as well I believe. I suspect that it would make sense to remove the 'uTVM runtime' and replace it with the MISRA-C one - as long as you're comfortable with code size being an important consideration in the development of the MISRA-C one?

tqchen · 2019-10-14T03:55:33Z

@liangfu @ajtulloch would be great if we can followup on this topic. e.g. try to compare the standalone utvm and misra-c runtime, and review this part of the proposed code.

One related question is that is it still OK to use C++? I see most of the code are implemented in pure C, which is fine, but perhaps we could still benefit from namespace and simple class(no virtual methods).

liangfu · 2019-10-14T04:29:43Z

The major intention to implement this in pure C is to maximize portability. (Some vendors didn't offer C++ compiler for its vision processor, partly for the sake of misra-c compliance, e.g. Vision SDK from TI.) On the other hand, this reduces binary size at the cost of flexibility.

tqchen · 2019-10-14T05:02:28Z

OK, I think we could go with c API, just remember to keep naming style consistent with the current C API as in c_runtime_api.h

liangfu · 2019-10-14T05:27:50Z

Sure, I would update the naming accordingly, and add test cases as well.

tqchen · 2020-02-06T03:27:37Z

ping @liangfu please see if you are interested in continue pushing this thread

liangfu · 2020-02-06T15:11:05Z

@tqchen sure, I can continue to push this thread, although this has been suspended for a while.

liangfu · 2020-02-15T16:29:12Z

@tqchen For now, the implement reproduces apps/bundle_deploy with complete runtime implemented in pure C. Note that the build_mode.py script generates apps/bundle_deploy_c/build/bridge.c to load functions in the compiled object (model.o), so that runtime could load all the compiled functions. In addition, to keep the pure C code OOP, an additional underscore is used in member functions for structs.

Please take a review.

ajtulloch · 2020-02-18T23:05:01Z

This looks really neat, well done @liangfu.

liangfu · 2020-02-19T01:38:27Z

@ajtulloch Thanks for the comment. In addition, I was also trying to integrate this with your implement of the standalone uTVM runtime. Hopefully, we would have a successful removal of picojson as an external dependency.

Change-Id: I027ddff15c31fb4da0bd0e461427dce619de1f93

liangfu · 2020-03-05T06:54:36Z

Also is this MISRA-C compliant going to work on 32bit and 64bit systems?

I have tested this on both 32-bit ARM board, and 64-bit Linux PC, they produced exactly the same result.

One additional suggestion I had was to test that the demo builds and runs in CI;

Nice suggestion! I've made this runs in CI.

Please take anther look.

Change-Id: I5ad5bb8426468aac9fc8d074e56ddea358a7fd91

Change-Id: Ic2e82fb3051b6c254ef32a964f976b61e3e5fe4d

tmoreau89 · 2020-03-05T20:19:43Z

Excellent, thanks for adding this to CI so quickly! I was able to reproduce the demo by typing in make demo; it ran for the most part successfully, but I got an illegal instruction error in the end:

python3 build_model.py -o build
INFO:root:Model file not found. Downloading to /Users/moreau/.mxnet/models/mobilenet0.25-9f83e440.params.
Downloading /Users/moreau/.mxnet/models/mobilenet0.25-9f83e440.zip from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/mobilenet0.25-9f83e440.zip...
INFO:autotvm:Download pre-tuned parameters package from https://raw.githubusercontent.com/uwsampl/tvm-distro/master/tophub/llvm_v0.04.log
...100%, 0.02 MB, 121 KB/s, 0 seconds passed
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op sqrt
INFO:compile_engine:Use implementation injective.cpu for op divide
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op negative
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op sqrt
INFO:compile_engine:Use implementation injective.cpu for op divide
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op negative
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op sqrt
INFO:compile_engine:Use implementation injective.cpu for op divide
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op negative
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op sqrt
INFO:compile_engine:Use implementation injective.cpu for op divide
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op negative
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op sqrt
INFO:compile_engine:Use implementation injective.cpu for op divide
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op negative
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op sqrt
INFO:compile_engine:Use implementation injective.cpu for op divide
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op negative
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op squeeze
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op squeeze
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op squeeze
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op squeeze
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op squeeze
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op squeeze
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op multiply
INFO:compile_engine:Use implementation injective.cpu for op multiply
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 3, 224, 224), 'float32'), ('TENSOR', (8, 3, 3, 3), 'float32'), (2, 2), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('depthwise_conv2d_NCHWc.x86', ('TENSOR', (1, 8, 112, 112), 'float32'), ('TENSOR', (8, 1, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 8, 112, 112), 'float32'), ('TENSOR', (16, 8, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('depthwise_conv2d_NCHWc.x86', ('TENSOR', (1, 16, 112, 112), 'float32'), ('TENSOR', (16, 1, 3, 3), 'float32'), (2, 2), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 16, 56, 56), 'float32'), ('TENSOR', (32, 16, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('depthwise_conv2d_NCHWc.x86', ('TENSOR', (1, 32, 56, 56), 'float32'), ('TENSOR', (32, 1, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 32, 56, 56), 'float32'), ('TENSOR', (32, 32, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('depthwise_conv2d_NCHWc.x86', ('TENSOR', (1, 32, 56, 56), 'float32'), ('TENSOR', (32, 1, 3, 3), 'float32'), (2, 2), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 32, 28, 28), 'float32'), ('TENSOR', (64, 32, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('depthwise_conv2d_NCHWc.x86', ('TENSOR', (1, 64, 28, 28), 'float32'), ('TENSOR', (64, 1, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 64, 28, 28), 'float32'), ('TENSOR', (64, 64, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('depthwise_conv2d_NCHWc.x86', ('TENSOR', (1, 64, 28, 28), 'float32'), ('TENSOR', (64, 1, 3, 3), 'float32'), (2, 2), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 64, 14, 14), 'float32'), ('TENSOR', (128, 64, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('depthwise_conv2d_NCHWc.x86', ('TENSOR', (1, 128, 14, 14), 'float32'), ('TENSOR', (128, 1, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 128, 14, 14), 'float32'), ('TENSOR', (128, 128, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('depthwise_conv2d_NCHWc.x86', ('TENSOR', (1, 128, 14, 14), 'float32'), ('TENSOR', (128, 1, 3, 3), 'float32'), (2, 2), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 128, 7, 7), 'float32'), ('TENSOR', (256, 128, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('depthwise_conv2d_NCHWc.x86', ('TENSOR', (1, 256, 7, 7), 'float32'), ('TENSOR', (256, 1, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 7, 7), 'float32'), ('TENSOR', (256, 256, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op expand_dims
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation softmax.cpu for op nn.softmax
WARNING:autotvm:Cannot find config for target=llvm --system-lib, workload=('dense_nopack.x86', ('TENSOR', (1, 256), 'float32'), ('TENSOR', (1000, 256), 'float32'), None, 'float32'). A fallback configuration is used, which may bring great performance regression.
INFO:compile_engine:Use implementation dense_nopack.x86 for op nn.dense
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
INFO:compile_engine:Use implementation injective.cpu for op nn.batch_flatten
INFO:compile_engine:Use implementation injective.cpu for op nn.batch_flatten
INFO:compile_engine:Use implementation adaptive_pool.cpu for op nn.global_avg_pool2d
INFO:compile_engine:Use implementation conv2d_NCHWc.x86 for op nn.contrib_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation depthwise_conv2d_NCHWc.x86 for op nn.contrib_depthwise_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation conv2d_NCHWc.x86 for op nn.contrib_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation depthwise_conv2d_NCHWc.x86 for op nn.contrib_depthwise_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation conv2d_NCHWc.x86 for op nn.contrib_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation depthwise_conv2d_NCHWc.x86 for op nn.contrib_depthwise_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation conv2d_NCHWc.x86 for op nn.contrib_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation depthwise_conv2d_NCHWc.x86 for op nn.contrib_depthwise_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation conv2d_NCHWc.x86 for op nn.contrib_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation depthwise_conv2d_NCHWc.x86 for op nn.contrib_depthwise_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation conv2d_NCHWc.x86 for op nn.contrib_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation depthwise_conv2d_NCHWc.x86 for op nn.contrib_depthwise_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation conv2d_NCHWc.x86 for op nn.contrib_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation depthwise_conv2d_NCHWc.x86 for op nn.contrib_depthwise_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation conv2d_NCHWc.x86 for op nn.contrib_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation depthwise_conv2d_NCHWc.x86 for op nn.contrib_depthwise_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation conv2d_NCHWc.x86 for op nn.contrib_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation depthwise_conv2d_NCHWc.x86 for op nn.contrib_depthwise_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation conv2d_NCHWc.x86 for op nn.contrib_conv2d_NCHWc
INFO:compile_engine:Use implementation injective.cpu for op add
INFO:compile_engine:Use implementation injective.cpu for op nn.relu
INFO:compile_engine:Use implementation injective.cpu for op layout_transform
Downloading from url https://homes.cs.washington.edu/~moreau/media/vta/cat.jpg to /Users/moreau/Documents/Projects/tvm-misra/apps/bundle_deploy/build/cat.png
...100%, 0.12 MB, 2633 KB/s, 0 seconds passed
x (1, 3, 224, 224)
xxd -i build/graph.json  > build/graph.json.c
xxd -i build/params.bin  > build/params.bin.c
g++ -std=c++14 -O2 -fPIC -I/Users/moreau/Documents/Projects/tvm-misra/include -I/Users/moreau/Documents/Projects/tvm-misra/3rdparty/dmlc-core/include -I/Users/moreau/Documents/Projects/tvm-misra/3rdparty/dlpack/include -o build/demo  demo.cc -ldl
g++ -shared -std=c++14 -O2 -fPIC -I/Users/moreau/Documents/Projects/tvm-misra/include -I/Users/moreau/Documents/Projects/tvm-misra/3rdparty/dmlc-core/include -I/Users/moreau/Documents/Projects/tvm-misra/3rdparty/dlpack/include -fvisibility=hidden -o build/bundle.so  bundle.cc runtime.cc build/model.o -pthread
gcc -shared -std=c99 -O2 -fPIC -I/Users/moreau/Documents/Projects/tvm-misra/include -I/Users/moreau/Documents/Projects/tvm-misra/3rdparty/dmlc-core/include -I/Users/moreau/Documents/Projects/tvm-misra/3rdparty/dlpack/include -fvisibility=hidden -o build/bundle_c.so  bundle.c runtime.c build/model.o -pthread
In file included from runtime.c:47:
In file included from ./../../src/runtime/crt/crt_runtime_api.c:28:
In file included from ./../../src/runtime/crt/graph_runtime.h:31:
./../../src/runtime/crt/packed_func.h:105:3: warning: redefinition of typedef 'TVMPackedFunc' is a C11 feature [-Wtypedef-redefinition]
} TVMPackedFunc;
  ^
./../../src/runtime/crt/module.h:31:30: note: previous definition is here
typedef struct TVMPackedFunc TVMPackedFunc;
                             ^
In file included from runtime.c:47:
./../../src/runtime/crt/crt_runtime_api.c:82:12: warning: expression result unused [-Wunused-value]
    status -1;
    ~~~~~~ ^~
In file included from runtime.c:48:
./../../src/runtime/crt/crt_backend_api.c:55:82: warning: format string is not a string literal (potentially insecure) [-Wformat-security]
  snprintf(g_fexecs[g_fexecs_count].name, sizeof(g_fexecs[g_fexecs_count].name), name);
                                                                                 ^~~~
/Library/Developer/CommandLineTools/SDKs/MacOSX.sdk/usr/include/secure/_stdio.h:57:62: note: expanded from macro 'snprintf'
  __builtin___snprintf_chk (str, len, 0, __darwin_obsz(str), __VA_ARGS__)
                                                             ^~~~~~~~~~~
./../../src/runtime/crt/crt_backend_api.c:55:82: note: treat the string as an argument to avoid this
  snprintf(g_fexecs[g_fexecs_count].name, sizeof(g_fexecs[g_fexecs_count].name), name);
                                                                                 ^
                                                                                 "%s",
/Library/Developer/CommandLineTools/SDKs/MacOSX.sdk/usr/include/secure/_stdio.h:57:62: note: expanded from macro 'snprintf'
  __builtin___snprintf_chk (str, len, 0, __darwin_obsz(str), __VA_ARGS__)
                                                             ^
In file included from runtime.c:49:
./../../src/runtime/crt/graph_runtime.c:542:84: warning: format specifies type 'int' but the argument has type 'size_t' (aka 'unsigned long') [-Wformat]
      fprintf(stderr, "fail to create for node with idx=%d, storage_id=%d\n", idx, storage_id);
                                                                       ~~          ^~~~~~~~~~
                                                                       %zu
In file included from runtime.c:51:
./../../src/runtime/crt/ndarray.c:95:13: warning: format specifies type 'long' but the argument has type 'int64_t' (aka 'long long') [-Wformat]
            data_byte_size, (num_elems * elem_bytes));
            ^~~~~~~~~~~~~~
./../../src/runtime/crt/ndarray.c:95:29: warning: format specifies type 'long' but the argument has type 'long long' [-Wformat]
            data_byte_size, (num_elems * elem_bytes));
                            ^~~~~~~~~~~~~~~~~~~~~~~~
6 warnings generated.
build/demo build/bundle.so build/cat.bin
The maximum position in output vector is: 278, with max-value 0.613490.
timing: 5.07 ms (create), 0.74 ms (set_input), 3.60 ms (run), 0.01 ms (get_output), 0.10 ms (destroy)
build/demo build/bundle_c.so build/cat.bin
make: *** [demo] Illegal instruction: 4```

tmoreau89 · 2020-03-05T20:20:45Z

Also @u99127 and @weberlo may be interested in this PR

Change-Id: Ide466a5cf21cf8bb990dcd4a1189ba17594e3c51

tqchen · 2020-03-07T01:31:06Z

@liangfu The e2e code cannot be run as part of the unittest because the CPU module do not have mxnet installed. Let us create a simple function(e.g. elemwise add) that makes use of the runtime as the test function.

tmoreau89 · 2020-03-07T01:35:32Z

ah, could we update the dockerfile to install mxnet? or will that be too wasteful in terms of CPU cycles

tqchen · 2020-03-07T02:42:34Z

In terms of testing the runtime out, using simple add one example would be sufficient.

tqchen · 2020-03-08T03:39:19Z

@liangfu would be great if you can try to update(use simple testing as in other all in one deployment that does not depend on mxnet) :)

liangfu · 2020-03-08T03:48:42Z

sure

Change-Id: Ie0dfd0ade6be4665b4384db7d260a6c69b35010f

Change-Id: Ie7fbc16b4b0b9509918d986a841f443900813bef

liangfu · 2020-03-09T10:15:44Z

@tmoreau89 @tqchen A simple test case has been added, and the testing results are successful. Please take another look.

tmoreau89 · 2020-03-09T16:49:20Z

Thank you @liangfu, this PR is pending on @tqchen 's approval

tqchen · 2020-03-09T18:21:45Z

Thanks @liangfu @tmoreau89 @ajtulloch , this is now merged!

mehrdadh · 2020-03-24T16:26:40Z

Great work. @liangfu have you considered using "system lib" approach since dlopen is banned in some environments?

mikeseven · 2020-03-29T01:40:23Z

@liangfu awesome work, very useful to our use-cases.

* implement of MISRA-C compliant TVM runtime; * working on bundle_deploy_c demo * move header files into include dir * fix compatibility issues * fix compatibility issues * resolve most of the warnings and errros * implement c_backend_api * introduce bridge * working well * move to header files and bundle.c into src/runtime/crt * clean up * satisfy linter * clean up * test with the cat image * remove synset * refactoring * refactoring * refactoring * initial crt_runtime_api.c * improved compatibility with g++ * using exposed API in c_runtime_api.h * call from c_runtime_api.h * clean up * lint * merge into apps/bundle_deploy directory Change-Id: I51904db81b8589e65d107d8ca77b47452e3812b5 * make the demo runs in ci Change-Id: I2c24f8b592508833d3555311c2b24d1931f19385 * address review comments Change-Id: I027ddff15c31fb4da0bd0e461427dce619de1f93 * release Change-Id: I5ad5bb8426468aac9fc8d074e56ddea358a7fd91 * fix ci testing Change-Id: Ic2e82fb3051b6c254ef32a964f976b61e3e5fe4d * add test case for misra c runtime Change-Id: Ie0dfd0ade6be4665b4384db7d260a6c69b35010f * fread files in testing to avoid calling xxd Change-Id: Ie7fbc16b4b0b9509918d986a841f443900813bef

ritabeczi · 2023-08-14T13:15:00Z

May I ask some more details like which MISRA rules(MISRA-C 2004, MISRA-C++-2008, MISRA-C-2012) are fixed other than what is mentioned on the #3159 . Thanks!

implement of MISRA-C compliant TVM runtime;

e63516d

tqchen added the status: need review label Sep 13, 2019

liangfu changed the title ~~[Runtime] MISRA-C compliant TVM runtime~~ [WIP][Runtime] MISRA-C compliant TVM runtime Sep 27, 2019

liangfu added 13 commits February 10, 2020 12:34

Merge branch 'master' into tvmcrt

dd2df9a

working on bundle_deploy_c demo

63ea62f

move header files into include dir

dfffa76

fix compatibility issues

75ef2c0

fix compatibility issues

e2f9475

resolve most of the warnings and errros

3b5c012

implement c_backend_api

ad3aa53

introduce bridge

2d60116

working well

200842d

move to header files and bundle.c into src/runtime/crt

8d5a409

clean up

cc49b17

satisfy linter

8b34d2e

clean up

70c269b

liangfu requested a review from tqchen February 15, 2020 16:30

liangfu changed the title ~~[WIP][Runtime] MISRA-C compliant TVM runtime~~ [Runtime] MISRA-C compliant TVM runtime Feb 20, 2020

address review comments

b65b62e

Change-Id: I027ddff15c31fb4da0bd0e461427dce619de1f93

liangfu added 2 commits March 5, 2020 15:57

release

682a74b

Change-Id: I5ad5bb8426468aac9fc8d074e56ddea358a7fd91

fix ci testing

0ee43ca

Change-Id: Ic2e82fb3051b6c254ef32a964f976b61e3e5fe4d

Merge branch 'master' into tvmcrt

f548fcf

Change-Id: Ide466a5cf21cf8bb990dcd4a1189ba17594e3c51

liangfu added 2 commits March 9, 2020 11:47

add test case for misra c runtime

521f7a0

Change-Id: Ie0dfd0ade6be4665b4384db7d260a6c69b35010f

fread files in testing to avoid calling xxd

1a1b660

Change-Id: Ie7fbc16b4b0b9509918d986a841f443900813bef

tqchen approved these changes Mar 9, 2020

View reviewed changes

tqchen merged commit 450f716 into apache:master Mar 9, 2020

tqchen added status: accepted and removed status: need review labels Mar 9, 2020

tqchen mentioned this pull request Mar 11, 2020

[RFC] MISRA-C/C++ Compliant Runtime #3159

Closed

This was referenced Mar 13, 2020

[uTVM][Runtime] Deprecate uTVM Standalone Runtime #5060

Closed

[Runtime] Parameterize constants in MISRA-C runtime #5062

Closed

ZihengJiang mentioned this pull request Sep 17, 2020

TVM v0.7 Release Note Candidate #6486

Closed

Conversation

liangfu commented Sep 11, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tqchen commented Sep 13, 2019

Uh oh!

liangfu commented Sep 21, 2019

Uh oh!

ajtulloch commented Sep 25, 2019

Uh oh!

tqchen commented Oct 14, 2019

Uh oh!

liangfu commented Oct 14, 2019

Uh oh!

tqchen commented Oct 14, 2019

Uh oh!

liangfu commented Oct 14, 2019

Uh oh!

tqchen commented Feb 6, 2020

Uh oh!

liangfu commented Feb 6, 2020

Uh oh!

liangfu commented Feb 15, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ajtulloch commented Feb 18, 2020

Uh oh!

liangfu commented Feb 19, 2020

Uh oh!

liangfu commented Mar 5, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tmoreau89 commented Mar 5, 2020

Uh oh!

tmoreau89 commented Mar 5, 2020

Uh oh!

tqchen commented Mar 7, 2020

Uh oh!

tmoreau89 commented Mar 7, 2020

Uh oh!

tqchen commented Mar 7, 2020

Uh oh!

tqchen commented Mar 8, 2020

Uh oh!

liangfu commented Mar 8, 2020

Uh oh!

liangfu commented Mar 9, 2020

Uh oh!

tmoreau89 commented Mar 9, 2020

Uh oh!

tqchen commented Mar 9, 2020

Uh oh!

mehrdadh commented Mar 24, 2020

Uh oh!

mikeseven commented Mar 29, 2020

Uh oh!

ritabeczi commented Aug 14, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

liangfu commented Sep 11, 2019 •

edited

Loading

liangfu commented Feb 15, 2020 •

edited

Loading

liangfu commented Mar 5, 2020 •

edited

Loading

ritabeczi commented Aug 14, 2023 •

edited

Loading