open source compiler match with cmodel master version?

[andyzheng-snps]

Does somebody try to use open source compiler to compile loadable and run vp? I tried several configuration including int8 and fp16, vp can not run successfully.
1 compiler setting
./nvdla_compiler --prototxt ResNet-50-deploy.prototxt --caffemodel ResNet-50-model.caffemodel --configtarget nv_small --profile basic --cprecision fp16

./nvdla_compiler --prototxt ResNet-50-deploy.prototxt --caffemodel ResNet-50-model.caffemodel --configtarget nv_small --profile basic --cprecision int8 --calibtable /remote/ailab2/zhaoqing/nvdla/sw-new/umd/utils/calibdata/resnet50.json

2 add #define DLA_2_CONFIG into below file to match cmodel for base address of registers.
kmd/firmware/include/opendla.h

2 c model master version with nv_small configuration.

[prasshantg]

fp16 is not supported with nv_small

I have some instructions documented at https://github.com/prasshantg/personal check if it helps

[andyzheng-snps]

thanks prasshantg for sharing.
I used below setting same with you, there are some errors. How about you?

---

% ./nvdla_compiler --prototxt ResNet-50-deploy.prototxt --caffemodel ResNet-50-model.caffemodel --configtarget nv_small --profile fast-math --cprecision int8 --calibtable nvdla/sw-new/umd/utils/calibdata/resnet50.json --quantizationMode per-filter --batch 1 --informat nhwc
creating new wisdom context...
opening wisdom context...
parsing caffe network...
libnvdla<3> mark prob
Marking total 1 outputs
parsing calibration table...
attaching parsed network to the wisdom...
compiling profile "fast-math"... config "nv_small"...
(DLA) Error 0x00000002: Don't support 0 (in engine-ast/ConvCoreNode.cpp, function tryToMergeWithBatchNormOp(), line 1240)
(DLA) Error 0x00000002: Don't support 0 (in engine-ast/ConvCoreNode.cpp, function tryToMergeWithBatchNormOp(), line 1240)
(DLA) Error 0x00000002: Don't support 0 (in engine-ast/ConvCoreNode.cpp, function tryToMergeWithBatchNormOp(), line 1240)
(DLA) Error 0x00000002: Don't support 0 (in engine-ast/ConvCoreNode.cpp, function tryToMergeWithBatchNormOp(), line 1240)
libnvdla<2> Scale value 1 for 1.41276e-06 is too small (threshold: 2)
libnvdla<2> Scale value 1 for 1.01866e-06 is too small (threshold: 2)
libnvdla<2> Scale value 1 for 1.48463e-06 is too small (threshold: 2)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
libnvdla<2> Scale value 1 for 1.62639e-06 is too small (threshold: 2)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
libnvdla<2> Scale value 1 for 6.84193e-07 is too small (threshold: 2)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
libnvdla<2> Scale value 1 for 5.3635e-07 is too small (threshold: 2)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
libnvdla<2> Scale value 1 for 2.58676e-07 is too small (threshold: 2)
closing wisdom context...

[ICfirefire]

thanks prasshantg for sharing.
I used below setting same with you, there are some errors. How about you?

% ./nvdla_compiler --prototxt ResNet-50-deploy.prototxt --caffemodel ResNet-50-model.caffemodel --configtarget nv_small --profile fast-math --cprecision int8 --calibtable nvdla/sw-new/umd/utils/calibdata/resnet50.json --quantizationMode per-filter --batch 1 --informat nhwc
creating new wisdom context...
opening wisdom context...
parsing caffe network...
libnvdla<3> mark prob
Marking total 1 outputs
parsing calibration table...
attaching parsed network to the wisdom...
compiling profile "fast-math"... config "nv_small"...
(DLA) Error 0x00000002: Don't support 0 (in engine-ast/ConvCoreNode.cpp, function tryToMergeWithBatchNormOp(), line 1240)
(DLA) Error 0x00000002: Don't support 0 (in engine-ast/ConvCoreNode.cpp, function tryToMergeWithBatchNormOp(), line 1240)
(DLA) Error 0x00000002: Don't support 0 (in engine-ast/ConvCoreNode.cpp, function tryToMergeWithBatchNormOp(), line 1240)
(DLA) Error 0x00000002: Don't support 0 (in engine-ast/ConvCoreNode.cpp, function tryToMergeWithBatchNormOp(), line 1240)
libnvdla<2> Scale value 1 for 1.41276e-06 is too small (threshold: 2)
libnvdla<2> Scale value 1 for 1.01866e-06 is too small (threshold: 2)
libnvdla<2> Scale value 1 for 1.48463e-06 is too small (threshold: 2)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
libnvdla<2> Scale value 1 for 1.62639e-06 is too small (threshold: 2)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
libnvdla<2> Scale value 1 for 6.84193e-07 is too small (threshold: 2)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
libnvdla<2> Scale value 1 for 5.3635e-07 is too small (threshold: 2)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
(DLA) Error 0x0000000b: Scale value is 0 for 0.000000
(in ./include/priv/LowPrecision.h, function scaleAndShiftFromScalarImpl2(), line 281)
libnvdla<2> Scale value 1 for 2.58676e-07 is too small (threshold: 2)
closing wisdom context...

hello, did you fix this problem?

[H2OD]

hello, I have the same problem, did you fix it?

hello, I have the same problem, did you fix it?