Fix some bugs.

See merge request !1127

1. Add transform for Shape and InferConv2dShape ops.
2. Remove matmul gpu kernel.
3. Fix the length of abbreviated commit id in MACE version.

Fix host link issue

See merge request !1125

Fix some bugs.

See merge request !1110

1. OpenCL Runtime destructor bug.
2. Net's outputs are used in net again.
3. Average pooling precision bug.

* Integrate APU

* fix code style issue

fix compile error on darwin

See merge request !1091

N/A
Signed-off-by: NLuxuhui <luxuhui@xiaomi.com>

Support accuracy validation using python script as a plugin.

See merge request !1074

add script to build and update images in registry server

See merge request !1084

Fix scalar output shape bug in operator.jinja2

See merge request !1083

Fix some bugs.

See merge request !1082

1. Bug in operator source template.
2. Output shapes of Op with data format may be not 4D.

Fix wrong checking in crop OpenCL kernel.

See merge request !1081

Add arguments checking for mace_run tool.

See merge request !1080

fix lstm_nonlinear name

See merge request !1075

Fp16

See merge request !1073

Add input size validation for Eltwise Op.

See merge request !1060

Update quantization doc

See merge request !1079

fix sched_setaffinity'bug and opt the performance of threadpool

See merge request !1078

Add data format explanation document.

See merge request !1072

N/A
Signed-off-by: NLuxuhui <luxuhui@xiaomi.com>

Add more log for locating out of range error of OpenCL.

See merge request !1077

Summary: Reformat fp16 gemv, add FP16_MATMUL_WEIGHT

Differential Revision: https://phabricator.d.xiaomi.net/D144134

Refactor data format

See merge request !1069

opt the compile option  and related code

See merge request !1068