Update Readme. Minor fix on demos.

Update README.md

Update README.md

Update README.md

Update README.md

Improve installation guide

Update README.md

Update README.md

README.md

@@ -1,20 +1,48 @@
-# CAUTION: Currently, this repo is under refactoring. Please checkout around Dec 15th. 
-
 # Collage
-System for automated integration of deep learning backends. Our implementation uses TVM as its code generator. 
+System for automated integration of deep learning backends. 
 
 # Installation
-1. Go to `tvm/` and install tvm. Make sure backend libaries of interest are built together. [TVM installation guide](https://tvm.apache.org/docs/install/index.html)
-2. Declare following environment variables
+Since our implementation uses TVM as the main code generator, install tvm under `tvm/`. [TVM installation guide](https://tvm.apache.org/docs/install/index.html)
+1. Install dependencies
+```
+sudo apt-get update
+sudo apt-get install -y python3 python3-dev python3-setuptools gcc libtinfo-dev zlib1g-dev build-essential cmake libedit-dev libxml2-dev
+```
+```
+pip3 install --user numpy decorator attrs tornado psutil xgboost cloudpickle pytest
+```
+
+2. Create build directory and go to build directory
+```
+mkdir tvm/build && cd tvm/build
+```
+3. Prepare `cmake` configuration file. Make sure backend libaries of interest are built together. We provide cmake config that we used for our GPU/CPU experiments (`config.cmake.gpu`, `config.cmake.cpu`) in `tvm/cmake/`. Users may copy it to their build directory and rename it to `config.cmake`
+ ```
+ cp ../cmake/config.cmake.gpu config.cmake
+ ```
+4. Run `cmake` and `make`
+```
+cmake .. && make -j$(nproc)
+```
+5. Declare following environment variables
 ```
 export COLLAGE_HOME=/path/to/collage/repo
 export COLLAGE_TVM_HOME=${COLLAGE_HOME}/tvm
 export PYTHONPATH=${COLLAGE_TVM_HOME}/python:${COLLAGE_HOME}/python:${PYTHONPATH}
 ```
 
 # Demo
-1. `cd demo/`
-2. `python3 demo.py`
+Install the following dependencies for deep learning models used for demo.
+```
+pip3 install --user torch torchvision tqdm onnx onnxruntime
+```
+
+We provide two demos (`demo_performance.py`, `demo_customization.py`) under `demo/`. 
+* `demo_performance.py` shows how collage optimizes given workloads with popular backends that Collage provides by default.
+* `demo_customization.py` shows how users can register new backend with their custom codegen, pattern, pattern rule.
+
+For the best result, it is highly recommend to create the tuning log by using `autotune_tvm_ops.py` before running those demos.
+
 
 # Note
 * As Collage uses TVM as its code generator, it cannot support backends that TVM is unable to build. Tested backends are

demo/demo_customization.py

@@ -47,16 +47,16 @@
 workload = {
     "optimizer": "op-level", 
     "backends": ["autotvm", "cudnn", "cublas", "tensorrt"], 
-    "network_name": "dcgan", #"resnext50_32x4d", 
+    "network_name": "dcgan", 
     "target": "cuda",
     "batch_size": 1,
 }
 
-# Default logging level
-#logging.basicConfig(level=logging.ERROR)
+# Default logging level. Skip messages during optimization
+logging.basicConfig(level=logging.ERROR)
 
 # Enable logging to monitor optimization progress e.g., operator matching, profiling...
-logging.basicConfig(level=logging.INFO)
+#logging.basicConfig(level=logging.INFO)
 
 def measure_perf(lib, workload):
     # Create workload
@@ -103,8 +103,8 @@ def check_dimension(config):
         return dim1 == 2 and dim2 == 2
 
     patterns = [
-        tuple([Pattern(is_op("nn.conv2d")(wildcard(), wildcard())), None]),
-        tuple([Pattern(is_op("nn.dense")(wildcard(), wildcard())), check_dimension])
+        tuple([Pattern(is_op("nn.conv2d")(wildcard(), wildcard())), check_dimension]),
+        tuple([Pattern(is_op("nn.dense")(wildcard(), wildcard())), None])
     ]
 
     collage_mod.register_new_backend(
@@ -141,11 +141,11 @@ def cg_VanillaTVM(net, target, params, **kwargs):
 
     # Run backend placement optimization with two custom backends
     workload["backends"] = ["VanillaTVM", "SimpleBackend"]
-    #lib = collage_mod.optimize_backend_placement(**workload)
-    #collage_mean_perf, collage_std_perf = measure_perf(lib, workload)
-    #print(f"# Network: {workload['network_name']}, Collage optimizer: {workload['optimizer']}")
-    #print(f"    - Provided backends: {workload['backends']}")
-    #print(f"    - Run with Collage  (mean, std) = ({collage_mean_perf:.4f}+-{collage_std_perf:.4f})")
+    lib = collage_mod.optimize_backend_placement(**workload)
+    collage_mean_perf, collage_std_perf = measure_perf(lib, workload)
+    print(f"# Network: {workload['network_name']}, Collage optimizer: {workload['optimizer']}")
+    print(f"    - Provided backends: {workload['backends']}")
+    print(f"    - Run with Collage  (mean, std) = ({collage_mean_perf:.4f}+-{collage_std_perf:.4f})")
 
 
     # 3. Register new backend with a pattern rule

demo/demo_performance.py

@@ -23,7 +23,7 @@
     "batch_size": 1,
 }
 
-# Default logging level
+# Default logging level. Skip messages during optimization
 logging.basicConfig(level=logging.ERROR)
 
 # Enable logging to monitor optimization progress e.g., operator matching, profiling...