thezha
commented
3 years ago SpatialTransformer is supported internally, we will add a TIM-VX API for it, then Tengine can integrate the new API.
Closed sky-fun closed 3 years ago
thezha
commented
3 years ago SpatialTransformer is supported internally, we will add a TIM-VX API for it, then Tengine can integrate the new API.
thezha
commented
3 years ago 
sky-fun
commented
3 years ago @thezha i add the code to Tengine, and run example on x86_64 simulator platform, got error like
Kernel "com.vivantecorp.extension.vxcTransform_setupThres_F16toF16" does not exist Kernel "com.vivantecorp.extension.vxcTransform_Gemm_F16toF16" does not exist Kernel "com.vivantecorp.extension.vxcTransform_InterP_F16toF16" does not exist
Do i need to change something about the dependence ？
thezha
commented
3 years ago Did you specify the runtime compiler header file via "export VIVANTE_SDK_DIR"?
export VIVANTE_SDK_DIR=pwd/prebuilt-sdk/x86_64_linux
sky-fun
commented
3 years ago @thezha yes， i did add "export VIVANTE_SDK_DIR=pwd/prebuilt-sdk/x86_64_linux" before i run the example.
thezha
commented
3 years ago @sky-fun These messages are not errors, they are information. Does your case run OK? Please supply more logs if possible.
Kernel "com.vivantecorp.extension.vxcTransform_setupThres_F16toF16" does not exist Kernel "com.vivantecorp.extension.vxcTransform_Gemm_F16toF16" does not exist Kernel "com.vivantecorp.extension.vxcTransform_InterP_F16toF16" does not exist
sky-fun
commented
3 years ago @thezha i can run my demo sucess with an unexpect result, so i use "export VIV_VX_DEBUG_LEVEL=4" to see the those informations, then i thought it was error bucause the phrase： "does not exist", and it may not run SpatialTransformer sucess on tim-vx. downblew is all my logs:
#productname=VSI SIMULATOR, pid=0xc5
prev_ptrs = 0x55b69df8c080
prev_ptrs = 0x55b69df8cac0
Kernel "com.vivantecorp.extension.vxcTransform_setupThres_F16toF16" does not existKernel "com.vivantecorp.extension.vxcTransform_Gemm_F16toF16" does not existKernel "com.vivantecorp.extension.vxcTransform_InterP_F16toF16" does not exist---------------------------Begin VerifyTiling -------------------------
AXI-SRAM = 0 Bytes VIP-SRAM = 260096 Bytes SWTILING_PHASE_FEATURES[1, 1, 1]
0 TP [( 112 128 3 1, 43008, 0x0x55b69ded6710(0x0x55b69ded6710, 0x(nil)) -> 112 128 3 1, 86016, 0x0x55b69dfe5d10(0x0x55b69dfe5d10, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] C[ 11]
1 NN [( 112 128 3 1, 43008, 0x0x55b69ded6710(0x0x55b69ded6710, 0x(nil)) -> 56 64 3 1, 10752, 0x0x55b69dfce460(0x0x55b69dfce460, 0x(nil))) k(2 2 1, 256) pad(0 0) pool(2 2, 2 2)] C[ 2]
2 NN [( 56 64 3 1, 10752, 0x0x55b69dfce460(0x0x55b69dfce460, 0x(nil)) -> 27 31 24 1, 20088, 0x0x55b69dfd3de0(0x0x55b69dfd3de0, 0x(nil))) k(3 3 3, 1024) pad(0 0) pool(2 2, 2 2)] P[ 1] C[ 3]
3 NN [( 27 31 24 1, 20088, 0x0x55b69dfd3de0(0x0x55b69dfd3de0, 0x(nil)) -> 25 29 24 1, 17400, 0x0x55b69dfd8ae0(0x0x55b69dfd8ae0, 0x(nil))) k(3 3 24, 5888) pad(0 0) pool(0 0, 1 1)] P[ 2] C[ 4]
4 TP [( 25 29 24 1, 17400, 0x0x55b69dfd8ae0(0x0x55b69dfd8ae0, 0x(nil)) -> 12 14 24 1, 4032, 0x0x55b69dfd9c50(0x0x55b69dfd9c50, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(2 2, 2 2)] P[ 3] C[ 5]
5 NN [( 12 14 24 1, 4032, 0x0x55b69dfd9c50(0x0x55b69dfd9c50, 0x(nil)) -> 5 6 48 1, 1440, 0x0x55b69dfcd6e0(0x0x55b69dfcd6e0, 0x(nil))) k(3 3 24, 11520) pad(0 0) pool(2 2, 2 2)] P[ 4] C[ 6]
6 TP [(1440 1 1 1, 1440, 0x0x55b69dfcd6e0(0x0x55b69dfcd6e0, 0x(nil)) -> 32 1 1 1, 32, 0x0x55b69dfe2ed0(0x0x55b69dfe2ed0, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 5] C[ 7]
7 TP [( 32 1 1 1, 32, 0x0x55b69dfe2ed0(0x0x55b69dfe2ed0, 0x(nil)) -> 6 1 1 1, 6, 0x0x55b69dfe4040(0x0x55b69dfe4040, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 6] C[ 8]
8 TP [( 6 1 1 1, 6, 0x0x55b69dfe4040(0x0x55b69dfe4040, 0x(nil)) -> 6 1 1 1, 12, 0x0x55b69dfe6240(0x0x55b69dfe6240, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 7] C[ 9]
9 SH [( 6 1 1 1, 12, 0x0x55b69dff0750(0x0x55b69dff0750, 0x(nil)) -> 6 1 1 1, 12, 0x0x55b69dfe7f90(0x0x55b69dfe7f90, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 8] C[ 10]
10 SH [( 6 1 1 1, 12, 0x0x55b69dfe7f90(0x0x55b69dfe7f90, 0x(nil)) -> 224 128 1 1, 57344, 0x0x55b69dfe8da0(0x0x55b69dfe8da0, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 9] C[ 11]
11 SH [( 112 128 3 1, 86016, 0x0x55b69dfe5d10(0x0x55b69dfe5d10, 0x(nil)) -> 112 128 3 1, 86016, 0x0x55b69dfe6900(0x0x55b69dfe6900, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 0, 10] C[ 12]
12 TP [( 112 128 3 1, 86016, 0x0x55b69dfe6900(0x0x55b69dfe6900, 0x(nil)) -> 112 128 3 1, 43008, 0x0x55b69dfe5790(0x0x55b69dfe5790, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 11] C[ 13]
13 NN [( 112 128 3 1, 43008, 0x0x55b69dfe5790(0x0x55b69dfe5790, 0x(nil)) -> 112 128 24 1, 344064, 0x0x55b69e06fe60(0x0x55b69e06fe60, 0x(nil))) k(3 3 3, 1024) pad(1 1) pool(0 0, 1 1)] P[ 12] C[ 14, 17]
14 TP [( 112 128 24 1, 344064, 0x0x55b69e06fe60(0x0x55b69e06fe60, 0x(nil)) -> 57 65 96 1, 355680, 0x0x55b6a0375d70(0x0x55b6a0375d70, 0x(nil))) k(0 0 0, 0) pad(1 1) pool(0 0, 1 1)] P[ 13] C[ 15]
15 NN [( 57 65 96 1, 355680, 0x0x55b6a0375d70(0x0x55b6a0375d70, 0x(nil)) -> 56 64 24 1, 86016, 0x0x55b69e02bc30(0x0x55b69e02bc30, 0x(nil))) k(2 2 96, 10496) pad(0 0) pool(0 0, 1 1)] P[ 14] C[ 16]
16 NN [( 56 64 24 1, 86016, 0x0x55b69e02bc30(0x0x55b69e02bc30, 0x(nil)) -> 56 64 24 1, 172032, 0x0x55b69e08efc0(0x0x55b69e08efc0, 0x(nil))) k(3 3 24, 5888) pad(1 1) pool(0 0, 1 1)] P[ 15] C[ 18]
17 NN [( 112 128 24 1, 344064, 0x0x55b69e06fe60(0x0x55b69e06fe60, 0x(nil)) -> 56 64 24 1, 172032, 0x0x55b6a1ee6fc0(0x0x55b69e08efc0, 0x0x15000)) k(1 1 24, 768) pad(0 0) pool(2 2, 2 2)] P[ 13] C[ 18]
18 NN [( 56 64 48 1, 172032, 0x0x55b69e08efc0(0x0x55b69e08efc0, 0x(nil)) -> 56 64 24 1, 172032, 0x0x55b6a1eec5a0(0x0x55b69e0945a0, 0x0x15000)) k(1 1 48, 1536) pad(0 0) pool(0 0, 1 1)] P[ 16, 17] C[ 19, 22]
19 TP [( 56 64 24 1, 172032, 0x0x55b6a1eec5a0(0x0x55b69e0945a0, 0x0x15000) -> 56 64 24 1, 86016, 0x0x55b69e058fe0(0x0x55b69e058fe0, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 18] C[ 20]
20 NN [( 56 64 24 1, 86016, 0x0x55b69e058fe0(0x0x55b69e058fe0, 0x(nil)) -> 56 64 24 1, 86016, 0x0x55b69e05c740(0x0x55b69e05c740, 0x(nil))) k(3 3 24, 5888) pad(1 1) pool(0 0, 1 1)] P[ 19] C[ 21]
21 NN [( 56 64 24 1, 86016, 0x0x55b69e05c740(0x0x55b69e05c740, 0x(nil)) -> 56 64 24 1, 172032, 0x0x55b69e0945a0(0x0x55b69e0945a0, 0x(nil))) k(3 3 24, 5888) pad(1 1) pool(0 0, 1 1)] P[ 20] C[ 22]
22 NN [( 56 64 48 1, 172032, 0x0x55b69e0945a0(0x0x55b69e0945a0, 0x(nil)) -> 56 64 24 1, 86016, 0x0x55b69dffdde0(0x0x55b69dffdde0, 0x(nil))) k(1 1 48, 1536) pad(0 0) pool(0 0, 1 1)] P[ 18, 21] C[ 23, 26]
23 TP [( 56 64 24 1, 86016, 0x0x55b69dffdde0(0x0x55b69dffdde0, 0x(nil)) -> 29 33 96 1, 91872, 0x0x55b6a037aa70(0x0x55b6a037aa70, 0x(nil))) k(0 0 0, 0) pad(1 1) pool(0 0, 1 1)] P[ 22] C[ 24]
24 NN [( 29 33 96 1, 91872, 0x0x55b6a037aa70(0x0x55b6a037aa70, 0x(nil)) -> 28 32 48 1, 43008, 0x0x55b69e0018f0(0x0x55b69e0018f0, 0x(nil))) k(2 2 96, 20992) pad(0 0) pool(0 0, 1 1)] P[ 23] C[ 25]
25 NN [( 28 32 48 1, 43008, 0x0x55b69e0018f0(0x0x55b69e0018f0, 0x(nil)) -> 28 32 48 1, 86016, 0x0x55b69e0980f0(0x0x55b69e0980f0, 0x(nil))) k(3 3 48, 22784) pad(1 1) pool(0 0, 1 1)] P[ 24] C[ 27]
26 NN [( 56 64 24 1, 86016, 0x0x55b69dffdde0(0x0x55b69dffdde0, 0x(nil)) -> 28 32 48 1, 86016, 0x0x55b69ffc40f0(0x0x55b69e0980f0, 0x0xa800)) k(1 1 24, 1536) pad(0 0) pool(2 2, 2 2)] P[ 22] C[ 27]
27 NN [( 28 32 96 1, 86016, 0x0x55b69e0980f0(0x0x55b69e0980f0, 0x(nil)) -> 28 32 48 1, 86016, 0x0x55b69ffc7c40(0x0x55b69e09bc40, 0x0xa800)) k(1 1 96, 6144) pad(0 0) pool(0 0, 1 1)] P[ 25, 26] C[ 28, 31]
28 TP [( 28 32 48 1, 86016, 0x0x55b69ffc7c40(0x0x55b69e09bc40, 0x0xa800) -> 28 32 48 1, 43008, 0x0x55b69e009150(0x0x55b69e009150, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 27] C[ 29]
29 NN [( 28 32 48 1, 43008, 0x0x55b69e009150(0x0x55b69e009150, 0x(nil)) -> 28 32 48 1, 43008, 0x0x55b69e00cc60(0x0x55b69e00cc60, 0x(nil))) k(3 3 48, 23040) pad(1 1) pool(0 0, 1 1)] P[ 28] C[ 30]
30 NN [( 28 32 48 1, 43008, 0x0x55b69e00cc60(0x0x55b69e00cc60, 0x(nil)) -> 28 32 48 1, 86016, 0x0x55b69e09bc40(0x0x55b69e09bc40, 0x(nil))) k(3 3 48, 23040) pad(1 1) pool(0 0, 1 1)] P[ 29] C[ 31]
31 NN [( 28 32 96 1, 86016, 0x0x55b69e09bc40(0x0x55b69e09bc40, 0x(nil)) -> 28 32 48 1, 43008, 0x0x55b69e011970(0x0x55b69e011970, 0x(nil))) k(1 1 96, 6144) pad(0 0) pool(0 0, 1 1)] P[ 27, 30] C[ 32, 33]
32 NN [( 28 32 48 1, 43008, 0x0x55b69e011970(0x0x55b69e011970, 0x(nil)) -> 14 16 64 1, 28672, 0x0x55b69eb03790(0x0x55b69e09f790, 0x0x3800)) k(1 1 48, 3584) pad(0 0) pool(2 2, 2 2)] P[ 31] C[ 36]
33 TP [( 28 32 48 1, 43008, 0x0x55b69e011970(0x0x55b69e011970, 0x(nil)) -> 15 17 192 1, 48960, 0x0x55b6a037e410(0x0x55b6a037e410, 0x(nil))) k(0 0 0, 0) pad(1 1) pool(0 0, 1 1)] P[ 31] C[ 34]
34 NN [( 15 17 192 1, 48960, 0x0x55b6a037e410(0x0x55b6a037e410, 0x(nil)) -> 14 16 64 1, 14336, 0x0x55b69e017fd0(0x0x55b69e017fd0, 0x(nil))) k(2 2 192, 56320) pad(0 0) pool(0 0, 1 1)] P[ 33] C[ 35]
35 NN [( 14 16 64 1, 14336, 0x0x55b69e017fd0(0x0x55b69e017fd0, 0x(nil)) -> 14 16 64 1, 28672, 0x0x55b69e09f790(0x0x55b69e09f790, 0x(nil))) k(3 3 64, 40448) pad(1 1) pool(0 0, 1 1)] P[ 34] C[ 36]
36 NN [( 14 16 128 1, 28672, 0x0x55b69e09f790(0x0x55b69e09f790, 0x(nil)) -> 14 16 64 1, 14336, 0x0x55b69e01bc90(0x0x55b69e01bc90, 0x(nil))) k(1 1 128, 10496) pad(0 0) pool(0 0, 1 1)] P[ 32, 35] C[ 37, 40]
37 TP [( 14 16 64 1, 14336, 0x0x55b69e01bc90(0x0x55b69e01bc90, 0x(nil)) -> 14 16 64 1, 14336, 0x0x55b69e01cce0(0x0x55b69e01cce0, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 36] C[ 38]
38 NN [( 14 16 64 1, 14336, 0x0x55b69e01cce0(0x0x55b69e01cce0, 0x(nil)) -> 14 16 64 1, 14336, 0x0x55b69e0723c0(0x0x55b69e0723c0, 0x(nil))) k(3 3 64, 40704) pad(1 1) pool(0 0, 1 1)] P[ 37] C[ 39]
39 NN [( 14 16 64 1, 14336, 0x0x55b69e0723c0(0x0x55b69e0723c0, 0x(nil)) -> 14 16 64 1, 14336, 0x0x55b69e073530(0x0x55b69e073530, 0x(nil))) k(3 3 64, 40448) pad(1 1) pool(0 0, 1 1)] P[ 38] C[ 40]
40 SH [(14336 1 1 1, 14336, 0x0x55b69e073530(0x0x55b69e073530, 0x(nil)) -> 14336 1 1 1, 14336, 0x0x55b69e0770d0(0x0x55b69e0770d0, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 39, 36] C[ 41, 42]
41 NN [( 14 16 64 1, 14336, 0x0x55b69e0770d0(0x0x55b69e0770d0, 0x(nil)) -> 7 8 128 1, 7168, 0x0x55b69e078090(0x0x55b69e078090, 0x(nil))) k(1 1 64, 9472) pad(0 0) pool(2 2, 2 2)] P[ 40] C[ 45]
42 TP [( 14 16 64 1, 14336, 0x0x55b69e0770d0(0x0x55b69e0770d0, 0x(nil)) -> 8 9 256 1, 18432, 0x0x55b6a03831c0(0x0x55b6a03831c0, 0x(nil))) k(0 0 0, 0) pad(1 1) pool(0 0, 1 1)] P[ 40] C[ 43]
43 NN [( 8 9 256 1, 18432, 0x0x55b6a03831c0(0x0x55b6a03831c0, 0x(nil)) -> 7 8 128 1, 7168, 0x0x55b69e07d730(0x0x55b69e07d730, 0x(nil))) k(2 2 256, 149760) pad(0 0) pool(0 0, 1 1)] P[ 42] C[ 44]
44 NN [( 7 8 128 1, 7168, 0x0x55b69e07d730(0x0x55b69e07d730, 0x(nil)) -> 7 8 128 1, 7168, 0x0x55b69e07e8a0(0x0x55b69e07e8a0, 0x(nil))) k(3 3 128, 164864) pad(1 1) pool(0 0, 1 1)] P[ 43] C[ 45]
45 SH [(7168 1 1 1, 7168, 0x0x55b69e07e8a0(0x0x55b69e07e8a0, 0x(nil)) -> 7168 1 1 1, 7168, 0x0x55b69e0813f0(0x0x55b69e0813f0, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 44, 41] C[ 46, 49]
46 TP [( 7 8 128 1, 7168, 0x0x55b69e0813f0(0x0x55b69e0813f0, 0x(nil)) -> 7 8 128 1, 7168, 0x0x55b69e082440(0x0x55b69e082440, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 45] C[ 47]
47 NN [( 7 8 128 1, 7168, 0x0x55b69e082440(0x0x55b69e082440, 0x(nil)) -> 7 8 128 1, 7168, 0x0x55b69e085dc0(0x0x55b69e085dc0, 0x(nil))) k(3 3 128, 166656) pad(1 1) pool(0 0, 1 1)] P[ 46] C[ 48]
48 NN [( 7 8 128 1, 7168, 0x0x55b69e085dc0(0x0x55b69e085dc0, 0x(nil)) -> 7 8 128 1, 7168, 0x0x55b69e086f30(0x0x55b69e086f30, 0x(nil))) k(3 3 128, 165376) pad(1 1) pool(0 0, 1 1)] P[ 47] C[ 49]
49 SH [(7168 1 1 1, 7168, 0x0x55b69e086f30(0x0x55b69e086f30, 0x(nil)) -> 7168 1 1 1, 7168, 0x0x55b69e08aad0(0x0x55b69e08aad0, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 48, 45] C[ 50]
50 TP [( 7 8 128 1, 7168, 0x0x55b69e08aad0(0x0x55b69e08aad0, 0x(nil)) -> 3 4 128 1, 1536, 0x0x55b69dfcdd80(0x0x55b69dfcdd80, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(2 2, 2 2)] P[ 49] C[ 51]
51 SH [( 1 1536 1 1, 1536, 0x0x55b69dfcdd80(0x0x55b69dfcdd80, 0x(nil)) -> 1 1536 1 1, 1536, 0x0x55b69dee0ca0(0x0x55b69dee0ca0, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 50]
Detected Segments
AB_VS (0 - 8)
AB_VS (12 - 13)
TL_VS (14 - 15)
AB_VS (15 - 18)
AB_VS (19 - 36)
AB_VS (37 - 39)
AB_VS (42 - 44)
AB_VS (46 - 48)
======================== Block [0 - 8] ==============================
0 TP DD -> DD [( 43008, 86016), IC( 0), KC( 0), NNT( 0, 0)]
1 NN DD -> VS [( 43008, 10752), IC( 1792), KC( 256), NNT( 0, 0)]
2 NN VS -> VS [( 10752, 20224), IC( 0), KC( 1024), NNT( 0, 0)]
3 NN VS -> VS [( 20224, 17408), IC( 0), KC( 5888), NNT( 0, 0)]
4 TP VS -> VS [( 17408, 4096), IC( 0), KC( 0), NNT( 0, 0)]
5 NN VS -> VS [( 4096, 1536), IC( 0), KC( 10752), NNT( 0, 0)]
6 TP VS -> VS [( 1536, 256), IC( 0), KC( 0), NNT( 0, 0)]
7 TP VS -> VS [( 256, 256), IC( 0), KC( 0), NNT( 0, 0)]
8 TP VS -> DD [( 256, 12), IC( 0), KC( 0), NNT( 0, 0)]
------------------------------------------------------------------
Segment AB (0 - 8)
AXISRAM: Peak used 0 0.000000% VIPSRAM: Peak used 43520 16.732283%
======================== Block [0 - 8] SUCCEED =========================
======================== Block [12 - 13] ==============================
12 TP DD -> VS [( 86016, 43008), IC( 0), KC( 0), NNT( 0, 0)]
13 NN VS -> DD [( 43008, 0), IC( 0), KC( 1024), NNT( 0, 0)]
------------------------------------------------------------------
Segment AB (12 - 13)
AXISRAM: Peak used 0 0.000000% VIPSRAM: Peak used 44032 16.929132%
======================== Block [12 - 13] SUCCEED =========================
======================== Block [14 - 18] ==============================
14 TP DD -> VS [( 0, 125952), IC( 0), KC( 0), NNT( 0, 0)]
15 NN VS -> VS [( 125952, 86016), IC( 0), KC( 7424), NNT( 0, 0)]
16 NN VS -> VS [( 86016, 172032), IC( 0), KC( 5120), NNT( 0, 0)]
17 NN DD -> VS [( 0, 172032), IC( 3072), KC( 1024), NNT( 0, 0)]
18 NN VS -> DD [( 172032, 0), IC( 0), KC( 768), NNT( 0, 0)]
------------------------------------------------------------------
Segment Tiling (14 - 15)
[DD 1( 0, 1)( 0) ->VS 1( 0, 1)( 23) P( 1) F(1)] [VS 23( 0, 23)( 23) ->VS 22( 0, 22)( 64) P( 0) F(0)]
[DD 44( 1, 45)( 0) ->VS 22( 1, 23)( 23) P( 0) F(1)] [VS 23( 22, 45)( 23) ->VS 22( 22, 44)( 64) P( 0) F(0)]
[DD 44( 45, 89)( 0) ->VS 22( 23, 45)( 23) P( 0) F(1)] [VS 21( 44, 65)( 23) ->VS 20( 44, 64)( 64) P( 0) F(1)]
[DD 39( 89, 128)( 0) ->VS 20( 45, 65)( 23) P( 0) F(1)]
0 [( 14, 0) F(1)]
1 [( 14, 1) F(1)]
2 [( 15, 0) F(0)] 3 [( 14, 2) F(1)]
4 [( 15, 1) F(0)] 5 [( 14, 3) F(1)]
6 [( 15, 2) F(1)]
AXISRAM: Estimate used 0 0.000000% VIPSRAM: Estimate used 136352 52.423721% M = 22
------------------------------------------------------------------
Segment AB (16 - 18)
AXISRAM: Peak used 0 0.000000% VIPSRAM: Peak used 260096 100.000000%
======================== Block [14 - 18] SUCCEED =========================
======================== Block [19 - 36] ==============================
19 TP DD -> VS [( 0, 86016), IC( 0), KC( 0), NNT( 0, 0)]
20 NN VS -> VS [( 86016, 86016), IC( 0), KC( 5888), NNT( 0, 0)]
21 NN VS -> DD [( 86016, 0), IC( 0), KC( 5888), NNT( 0, 0)]
22 NN DD -> VS [( 0, 86016), IC( 8192), KC( 768), NNT( 0, 0)]
23 TP VS -> VS [( 86016, 91904), IC( 0), KC( 0), NNT( 0, 0)]
24 NN VS -> VS [( 91904, 43008), IC( 0), KC( 14848), NNT( 0, 0)]
25 NN VS -> VS [( 43008, 86016), IC( 0), KC( 21248), NNT( 0, 0)]
26 NN VS -> VS [( 86016, 86016), IC( 0), KC( 1280), NNT( 0, 0)]
27 NN VS -> VS [( 86016, 86016), IC( 0), KC( 1024), NNT( 0, 0)]
28 TP VS -> VS [( 86016, 43008), IC( 0), KC( 0), NNT( 0, 0)]
29 NN VS -> VS [( 43008, 43008), IC( 0), KC( 20992), NNT( 0, 0)]
30 NN VS -> VS [( 43008, 86016), IC( 0), KC( 20224), NNT( 0, 0)]
31 NN VS -> VS [( 86016, 43008), IC( 0), KC( 1024), NNT( 0, 0)]
32 NN VS -> VS [( 43008, 28672), IC( 0), KC( 3584), NNT( 0, 0)]
33 TP VS -> VS [( 43008, 49152), IC( 0), KC( 0), NNT( 0, 0)]
34 NN VS -> VS [( 49152, 14336), IC( 0), KC( 35072), NNT( 0, 0)]
35 NN VS -> VS [( 14336, 28672), IC( 0), KC( 38912), NNT( 0, 0)]
36 NN VS -> DD [( 28672, 0), IC( 0), KC( 1280), NNT( 0, 0)]
------------------------------------------------------------------
Segment AB (19 - 36)
AXISRAM: Peak used 0 0.000000% VIPSRAM: Peak used 236288 90.846458%
======================== Block [19 - 36] SUCCEED =========================
======================== Block [37 - 39] ==============================
37 TP DD -> VS [( 0, 14336), IC( 0), KC( 0), NNT( 0, 0)]
38 NN VS -> VS [( 14336, 14336), IC( 0), KC( 36352), NNT( 0, 0)]
39 NN VS -> DD [( 14336, 0), IC( 0), KC( 38144), NNT( 0, 0)]
------------------------------------------------------------------
Segment AB (37 - 39)
AXISRAM: Peak used 0 0.000000% VIPSRAM: Peak used 65024 25.000000%
======================== Block [37 - 39] SUCCEED =========================
======================== Block [42 - 44] ==============================
42 TP DD -> VS [( 0, 18432), IC( 0), KC( 0), NNT( 0, 0)]
43 NN VS -> VS [( 18432, 7168), IC( 0), KC( 94720), NNT( 0, 0)]
44 NN VS -> DD [( 7168, 0), IC( 0), KC( 127488), NNT( 0, 0)]
------------------------------------------------------------------
Segment AB (42 - 44)
AXISRAM: Peak used 0 0.000000% VIPSRAM: Peak used 134656 51.771652%
======================== Block [42 - 44] SUCCEED =========================
======================== Block [46 - 48] ==============================
46 TP DD -> VS [( 0, 7168), IC( 0), KC( 0), NNT( 0, 0)]
47 NN VS -> VS [( 7168, 7168), IC( 0), KC( 116480), NNT( 0, 0)]
48 NN VS -> DD [( 7168, 0), IC( 0), KC( 123904), NNT( 0, 0)]
------------------------------------------------------------------
Segment AB (46 - 48)
AXISRAM: Peak used 0 0.000000% VIPSRAM: Peak used 131072 50.393700%
======================== Block [46 - 48] SUCCEED =========================
id IN [ x y w h ] OUT [ x y w h ] (tx, ty, kpc) (ic, kc, kc/ks, ks/eks, kernel_type)
0 TP DD 0x(nil) [ 0 0 112 128] -> DD 0x(nil) [ 0 0 112 128] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
1 NN DD 0x(nil) [ 0 0 112 128] -> VS 0x0xff000800 [ 0 0 56 64] ( 64, 8, 1) ( 1792, 256, 100.000000%, 100.000000%, DD)
2 NN VS 0x0xff000800 [ 0 0 56 64] -> VS 0x0xff006300 [ 0 0 27 31] ( 28, 16, 3) ( 0, 1024, 100.000000%, 100.000000%, DD)
3 NN VS 0x0xff006300 [ 0 0 27 31] -> VS 0x0xff000800 [ 0 0 25 29] ( 25, 12, 6) ( 0, 5888, 100.000000%, 100.000000%, DD)
4 TP VS 0x0xff000800 [ 0 0 25 29] -> VS 0x0xff00a200 [ 0 0 12 14] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
5 NN VS 0x0xff00a200 [ 0 0 12 14] -> VS 0x0xff000800 [ 0 0 5 6] ( 10, 12, 12) ( 0, 10752, 100.000000%, 93.333334%, DD)
6 TP VS 0x0xff000800 [ 0 0 1440 1] -> VS 0x0xff00b100 [ 0 0 32 1] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
7 TP VS 0x0xff00b100 [ 0 0 32 1] -> VS 0x0xff000800 [ 0 0 6 1] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
8 TP VS 0x0xff000800 [ 0 0 6 1] -> DD 0x(nil) [ 0 0 6 1] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
9 SH DD 0x(nil) [ 0 0 0 0] -> DD 0x(nil) [ 0 0 0 0] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
10 SH DD 0x(nil) [ 0 0 0 0] -> DD 0x(nil) [ 0 0 0 0] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
11 SH DD 0x(nil) [ 0 0 0 0] -> DD 0x(nil) [ 0 0 0 0] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
12 TP DD 0x(nil) [ 0 0 112 128] -> VS 0x0xff000800 [ 0 0 112 128] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
13 NN VS 0x0xff000800 [ 0 0 112 128] -> DD 0x(nil) [ 0 0 112 128] ( 64, 4, 6) ( 0, 1024, 100.000000%, 100.000000%, DD)
14 TP DD 0x(nil) [ 0 0 112 1] -> VS 0x0xff000800 [ 0 0 57 1] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
14 TP DD 0x0x70 [ 0 1 112 44] -> VS 0x0xff000839 [ 0 1 57 22] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
15 NN VS 0x0xff000800 [ 0 0 57 23] -> VS 0x0xff02b000 [ 0 0 56 22] ( 56, 8, 3) ( 0, 7424, 100.000000%, 70.731705%, DD)
14 TP DD 0x0x13b0 [ 0 45 112 44] -> VS 0x0xff000d1f [ 0 23 57 22] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
15 NN VS 0x0xff000ce6 [ 0 22 57 23] -> VS 0x0xff02b4d0 [ 0 22 56 22] ( 56, 8, 3) ( 0, 7424, 100.000000%, 70.731705%, DD)
14 TP DD 0x0x26f0 [ 0 89 112 39] -> VS 0x0xff001205 [ 0 45 57 20] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
15 NN VS 0x0xff0011cc [ 0 44 57 21] -> VS 0x0xff02b9a0 [ 0 44 56 20] ( 56, 8, 3) ( 0, 7424, 100.000000%, 70.731705%, DD)
16 NN VS 0x0xff02b000 [ 0 0 56 64] -> VS 0x0xff000800 [ 0 0 56 64] ( 28, 9, 6) ( 0, 2048, 40.000001%, 86.956519%, DD)
17 NN DD 0x(nil) [ 0 0 112 128] -> VS 0x0xff015800 [ 0 0 56 64] ( 64, 2, 6) ( 3072, 1024, 100.000000%, 133.333337%, DD)
18 NN VS 0x0xff000800 [ 0 0 56 64] -> DD 0x(nil) [ 0 0 56 64] ( 56, 8, 3) ( 0, 768, 100.000000%, 50.000000%, DD)
19 TP DD 0x(nil) [ 0 0 56 64] -> VS 0x0xff000800 [ 0 0 56 64] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
20 NN VS 0x0xff000800 [ 0 0 56 64] -> VS 0x0xff025300 [ 0 0 56 64] ( 28, 9, 6) ( 0, 5888, 100.000000%, 100.000000%, DD)
21 NN VS 0x0xff025300 [ 0 0 56 64] -> DD 0x(nil) [ 0 0 56 64] ( 56, 6, 3) ( 0, 5888, 100.000000%, 100.000000%, DD)
22 NN DD 0x(nil) [ 0 0 56 64] -> VS 0x0xff000800 [ 0 0 56 64] ( 56, 3, 6) ( 8192, 768, 100.000000%, 50.000000%, DD)
23 TP VS 0x0xff000800 [ 0 0 56 64] -> VS 0x0xff023c00 [ 0 0 29 33] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
24 NN VS 0x0xff023c00 [ 0 0 29 33] -> VS 0x0xff015800 [ 0 0 28 32] ( 28, 16, 4) ( 0, 14848, 100.000000%, 70.731705%, DD)
25 NN VS 0x0xff015800 [ 0 0 28 32] -> VS 0x0xff025300 [ 0 0 28 32] ( 28, 16, 4) ( 0, 21248, 100.000000%, 93.258429%, DD)
26 NN VS 0x0xff000800 [ 0 0 56 64] -> VS 0x0xff02fb00 [ 0 0 28 32] ( 56, 2, 12) ( 0, 1280, 100.000000%, 83.333331%, DD)
27 NN VS 0x0xff025300 [ 0 0 28 32] -> VS 0x0xff00b000 [ 0 0 28 32] ( 28, 4, 12) ( 0, 1024, 100.000000%, 16.666667%, DD)
28 TP VS 0x0xff00b000 [ 0 0 28 32] -> VS 0x0xff02fb00 [ 0 0 28 32] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
29 NN VS 0x0xff02fb00 [ 0 0 28 32] -> VS 0x0xff015800 [ 0 0 28 32] ( 28, 16, 4) ( 0, 20992, 100.000000%, 91.111112%, DD)
30 NN VS 0x0xff015800 [ 0 0 28 32] -> VS 0x0xff000800 [ 0 0 28 32] ( 28, 16, 4) ( 0, 20224, 100.000000%, 87.777776%, DD)
31 NN VS 0x0xff000800 [ 0 0 28 32] -> VS 0x0xff02fb00 [ 0 0 28 32] ( 28, 4, 12) ( 0, 1024, 100.000000%, 16.666667%, DD)
32 NN VS 0x0xff02fb00 [ 0 0 28 32] -> VS 0x0xff004000 [ 0 0 14 16] ( 28, 4, 16) ( 0, 3584, 100.000000%, 100.000000%, DD)
33 TP VS 0x0xff02fb00 [ 0 0 28 32] -> VS 0x0xff007800 [ 0 0 15 17] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
34 NN VS 0x0xff007800 [ 0 0 15 17] -> VS 0x0xff036b00 [ 0 0 14 16] ( 14, 13, 8) ( 0, 35072, 100.000000%, 62.272727%, DD)
35 NN VS 0x0xff036b00 [ 0 0 14 16] -> VS 0x0xff000800 [ 0 0 14 16] ( 14, 12, 8) ( 0, 38912, 100.000000%, 96.202534%, DD)
36 NN VS 0x0xff000800 [ 0 0 14 16] -> DD 0x(nil) [ 0 0 14 16] ( 14, 16, 8) ( 0, 1280, 100.000000%, 12.195122%, DD)
37 TP DD 0x(nil) [ 0 0 14 16] -> VS 0x0xff000800 [ 0 0 14 16] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
38 NN VS 0x0xff000800 [ 0 0 14 16] -> VS 0x0xff00ce00 [ 0 0 14 16] ( 14, 12, 8) ( 0, 36352, 100.000000%, 89.308178%, DD)
39 NN VS 0x0xff00ce00 [ 0 0 14 16] -> DD 0x(nil) [ 0 0 14 16] ( 14, 16, 8) ( 0, 38144, 100.000000%, 94.303799%, DD)
40 SH DD 0x(nil) [ 0 0 0 0] -> DD 0x(nil) [ 0 0 0 0] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
41 NN DD 0x(nil) [ 0 0 14 16] -> DD 0x(nil) [ 0 0 7 8] ( 14, 16, 8) ( 14848, 9728, 100.000000%, 102.702701%, DD)
42 TP DD 0x(nil) [ 0 0 14 16] -> VS 0x0xff000800 [ 0 0 8 9] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
43 NN VS 0x0xff000800 [ 0 0 8 9] -> VS 0x0xff01fa00 [ 0 0 7 8] ( 7, 8, 16) ( 0, 94720, 100.000000%, 63.247865%, DD)
44 NN VS 0x0xff01fa00 [ 0 0 7 8] -> DD 0x(nil) [ 0 0 7 8] ( 7, 8, 16) ( 0, 127488, 100.000000%, 77.329195%, DD)
45 SH DD 0x(nil) [ 0 0 0 0] -> DD 0x(nil) [ 0 0 0 0] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
46 TP DD 0x(nil) [ 0 0 7 8] -> VS 0x0xff000800 [ 0 0 7 8] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
47 NN VS 0x0xff000800 [ 0 0 7 8] -> VS 0x0xff01ec00 [ 0 0 7 8] ( 7, 8, 16) ( 0, 116480, 100.000000%, 69.892472%, DD)
48 NN VS 0x0xff01ec00 [ 0 0 7 8] -> DD 0x(nil) [ 0 0 7 8] ( 7, 8, 16) ( 0, 123904, 100.000000%, 74.922603%, DD)
49 SH DD 0x(nil) [ 0 0 0 0] -> DD 0x(nil) [ 0 0 0 0] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
50 TP DD 0x(nil) [ 0 0 7 8] -> DD 0x(nil) [ 0 0 3 4] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
51 SH DD 0x(nil) [ 0 0 0 0] -> DD 0x(nil) [ 0 0 0 0] ( 0, 0, 0) ( 0, 0, 0.000000%, 0.000000%, NONE)
PreLoadWeightBiases = 0 -nan%
---------------------------End VerifyTiling -------------------------
vxo_insertHandle(out): old physical address: 00296300, mapIndex: 2, node_id: 15
success prerun_graph
##################VXEngineRun
VXEngineRun 0 4
prev_ptrs = 0x55b69df8cac0
Warning: swapHandel, CMD changed
NN/TP: pre_physical:0x00296940, new_physical:0x00296940
NN/TP: pre_physical:0x00296978, new_physical:0x00296978
Warning: swapHandel, CMD changed
NN/TP: pre_physical:0x00296940, new_physical:0x00296940
Warning: swapHandel, CMD changed
SH: pre_physical:0x00296300, new_physical:0x00296300, mapIndex: 2, node_id: 15
prev_ptrs = 0x55b69df8c080Is the following layer SpecialTransformer?
10 SH [( 6 1 1 1, 12, 0x0x55b69dfe7f90(0x0x55b69dfe7f90, 0x(nil)) -> 224 128 1 1, 57344, 0x0x55b69dfe8da0(0x0x55b69dfe8da0, 0x(nil))) k(0 0 0, 0) pad(0 0) pool(0 0, 1 1)] P[ 9] C[ 11]
Input shape(1, 1, 6), output shape(1, 128, 224)?
sky-fun
commented
3 years ago yes, but in the origin network target shape of SpecialTransformer should be (3,128, 112)
nightingalei
commented
3 years ago @sky-fun What's the datatype and parameters to the SpacialTransformer?
nightingalei
commented
3 years ago @sky-fun Could please also provide the log with the env "export VSI_NN_LOG_LEVEL=5"? Thanks.
sky-fun
commented
3 years ago datatype is uint8 and the parameters is (output_h, output_w, false, false, false, false, false, false, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0)
sky-fun
commented
3 years ago @nightingalei
D [vsi_nn_SetupGraph:617]Sort graph nodes.
no Unprocessed node
D [setup_node:441]Setup node id[0] uid[1] op[POOL]
D [validate_op_io_types:165]Validate [POOL]
D [print_tensor:146]in(0) : id[ 1] vtl[0] const[0] shape[ 112, 128, 3, 1 ] fmt[u8 ] qnt[ASM zp=127, scale=0.007782]
D [print_tensor:146]out(0): id[ 4] vtl[1] const[0] shape[ 56, 64, 3, 1 ] fmt[u8 ] qnt[ASM zp=127, scale=0.007805]
D [setup_node:441]Setup node id[1] uid[2] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 4] vtl[1] const[0] shape[ 56, 64, 3, 1 ] fmt[u8 ] qnt[ASM zp=127, scale=0.007805]
D [print_tensor:146]in(1) : id[ 3] vtl[0] const[1] shape[ 3, 3, 3, 24 ] fmt[u8 ] qnt[ASM zp=144, scale=0.008708]
D [print_tensor:146]in(2) : id[ 2] vtl[0] const[1] shape[ 24 ] fmt[i32] qnt[ASM zp= 0, scale=0.000068]
D [print_tensor:146]out(0): id[ 105] vtl[1] const[0] shape[ 54, 62, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.014651]
D [setup_node:441]Setup node id[2] uid[3] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 105] vtl[1] const[0] shape[ 54, 62, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.014651]
D [print_tensor:146]out(0): id[ 5] vtl[1] const[0] shape[ 54, 62, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.014651]
D [setup_node:441]Setup node id[3] uid[4] op[POOL]
D [validate_op_io_types:165]Validate [POOL]
D [print_tensor:146]in(0) : id[ 5] vtl[1] const[0] shape[ 54, 62, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.014651]
D [print_tensor:146]out(0): id[ 8] vtl[1] const[0] shape[ 27, 31, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.014651]
D [setup_node:441]Setup node id[4] uid[5] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 8] vtl[1] const[0] shape[ 27, 31, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.014651]
D [print_tensor:146]in(1) : id[ 7] vtl[0] const[1] shape[ 3, 3, 24, 24 ] fmt[u8 ] qnt[ASM zp=123, scale=0.003316]
D [print_tensor:146]in(2) : id[ 6] vtl[0] const[1] shape[ 24 ] fmt[i32] qnt[ASM zp= 0, scale=0.000049]
D [print_tensor:146]out(0): id[ 106] vtl[1] const[0] shape[ 25, 29, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.006747]
D [setup_node:441]Setup node id[5] uid[6] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 106] vtl[1] const[0] shape[ 25, 29, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.006747]
D [print_tensor:146]out(0): id[ 9] vtl[1] const[0] shape[ 25, 29, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.006747]
D [setup_node:441]Setup node id[6] uid[7] op[POOL]
D [validate_op_io_types:165]Validate [POOL]
D [print_tensor:146]in(0) : id[ 9] vtl[1] const[0] shape[ 25, 29, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.006747]
D [print_tensor:146]out(0): id[ 12] vtl[1] const[0] shape[ 12, 14, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.006747]
D [setup_node:441]Setup node id[7] uid[8] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 12] vtl[1] const[0] shape[ 12, 14, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.006747]
D [print_tensor:146]in(1) : id[ 11] vtl[0] const[1] shape[ 3, 3, 24, 48 ] fmt[u8 ] qnt[ASM zp=132, scale=0.003830]
D [print_tensor:146]in(2) : id[ 10] vtl[0] const[1] shape[ 48 ] fmt[i32] qnt[ASM zp= 0, scale=0.000026]
D [print_tensor:146]out(0): id[ 107] vtl[1] const[0] shape[ 10, 12, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.004216]
D [setup_node:441]Setup node id[8] uid[9] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 107] vtl[1] const[0] shape[ 10, 12, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.004216]
D [print_tensor:146]out(0): id[ 13] vtl[1] const[0] shape[ 10, 12, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.004216]
D [setup_node:441]Setup node id[9] uid[10] op[POOL]
D [validate_op_io_types:165]Validate [POOL]
D [print_tensor:146]in(0) : id[ 13] vtl[1] const[0] shape[ 10, 12, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.004216]
D [print_tensor:146]out(0): id[ 14] vtl[1] const[0] shape[ 5, 6, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.004216]
D [setup_node:441]Setup node id[10] uid[11] op[RESHAPE]
D [print_tensor:146]in(0) : id[ 14] vtl[1] const[0] shape[ 5, 6, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.004216]
D [print_tensor:146]out(0): id[ 15] vtl[1] const[0] shape[ 1, 1, 1440, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.004216]
D [setup_node:441]Setup node id[11] uid[12] op[FCL2]
D [validate_op_io_types:165]Validate [FCL_RELU]
D [print_tensor:146]in(0) : id[ 15] vtl[1] const[0] shape[ 1, 1, 1440, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.004216]
D [print_tensor:146]in(1) : id[ 17] vtl[0] const[1] shape[ 1440, 32 ] fmt[u8 ] qnt[ASM zp=167, scale=0.005804]
D [print_tensor:146]in(2) : id[ 16] vtl[0] const[1] shape[ 32 ] fmt[i32] qnt[ASM zp= 0, scale=0.000024]
D [print_tensor:146]out(0): id[ 18] vtl[1] const[0] shape[ 32, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.005455]
D [setup_node:441]Setup node id[12] uid[13] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 18] vtl[1] const[0] shape[ 32, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.005455]
D [print_tensor:146]out(0): id[ 19] vtl[1] const[0] shape[ 32, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.005455]
D [setup_node:441]Setup node id[13] uid[14] op[FCL2]
D [validate_op_io_types:165]Validate [FCL_RELU]
D [print_tensor:146]in(0) : id[ 19] vtl[1] const[0] shape[ 32, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.005455]
D [print_tensor:146]in(1) : id[ 21] vtl[0] const[1] shape[ 32, 6 ] fmt[u8 ] qnt[ASM zp=121, scale=0.007150]
D [print_tensor:146]in(2) : id[ 20] vtl[0] const[1] shape[ 6 ] fmt[i32] qnt[ASM zp= 0, scale=0.000039]
D [print_tensor:146]out(0): id[ 22] vtl[1] const[0] shape[ 6, 1 ] fmt[u8 ] qnt[ASM zp=114, scale=0.013708]
D [setup_node:441]Setup node id[14] uid[15] op[SPATIAL_TRANSFORMER]
D [print_tensor:146]in(0) : id[ 1] vtl[0] const[0] shape[ 112, 128, 3, 1 ] fmt[u8 ] qnt[ASM zp=127, scale=0.007782]
D [print_tensor:146]in(1) : id[ 22] vtl[1] const[0] shape[ 6, 1 ] fmt[u8 ] qnt[ASM zp=114, scale=0.013708]
D [print_tensor:146]out(0): id[ 25] vtl[1] const[0] shape[ 112, 128, 3, 1 ] fmt[u8 ] qnt[ASM zp=127, scale=0.007805]
D [setup_node:441]Setup node id[15] uid[16] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 25] vtl[1] const[0] shape[ 112, 128, 3, 1 ] fmt[u8 ] qnt[ASM zp=127, scale=0.007805]
D [print_tensor:146]in(1) : id[ 24] vtl[0] const[1] shape[ 3, 3, 3, 24 ] fmt[u8 ] qnt[ASM zp= 98, scale=0.020664]
D [print_tensor:146]in(2) : id[ 23] vtl[0] const[1] shape[ 24 ] fmt[i32] qnt[ASM zp= 0, scale=0.000161]
D [print_tensor:146]out(0): id[ 108] vtl[1] const[0] shape[ 112, 128, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.003564]
D [setup_node:441]Setup node id[16] uid[17] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 108] vtl[1] const[0] shape[ 112, 128, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.003564]
D [print_tensor:146]out(0): id[ 28] vtl[1] const[0] shape[ 112, 128, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.003564]
D [setup_node:441]Setup node id[17] uid[18] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 28] vtl[1] const[0] shape[ 112, 128, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.003564]
D [print_tensor:146]in(1) : id[ 27] vtl[0] const[1] shape[ 3, 3, 24, 24 ] fmt[u8 ] qnt[ASM zp=161, scale=0.007057]
D [print_tensor:146]in(2) : id[ 26] vtl[0] const[1] shape[ 24 ] fmt[i32] qnt[ASM zp= 0, scale=0.000025]
D [print_tensor:146]out(0): id[ 109] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.003384]
D [setup_node:441]Setup node id[18] uid[19] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 109] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.003384]
D [print_tensor:146]out(0): id[ 31] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.003384]
D [setup_node:441]Setup node id[19] uid[20] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 28] vtl[1] const[0] shape[ 112, 128, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.003564]
D [print_tensor:146]in(1) : id[ 30] vtl[0] const[1] shape[ 1, 1, 24, 24 ] fmt[u8 ] qnt[ASM zp=138, scale=0.003668]
D [print_tensor:146]in(2) : id[ 29] vtl[0] const[1] shape[ 24 ] fmt[i32] qnt[ASM zp= 0, scale=0.000013]
D [print_tensor:146]out(0): id[ 34] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 68, scale=0.005292]
D [setup_node:441]Setup node id[20] uid[21] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 31] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.003384]
D [print_tensor:146]in(1) : id[ 32] vtl[0] const[1] shape[ 3, 3, 24, 24 ] fmt[u8 ] qnt[ASM zp=118, scale=0.009258]
D [print_tensor:146]in(2) : id[ 33] vtl[0] const[1] shape[ 24 ] fmt[i32] qnt[ASM zp= 0, scale=0.000031]
D [print_tensor:146]out(0): id[ 35] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 96, scale=0.013286]
D [setup_node:441]Setup node id[21] uid[22] op[ADD]
D [validate_op_io_types:165]Validate [ADD]
D [print_tensor:146]in(0) : id[ 35] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 96, scale=0.013286]
D [print_tensor:146]in(1) : id[ 34] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 68, scale=0.005292]
D [print_tensor:146]out(0): id[ 36] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp=100, scale=0.011393]
D [setup_node:441]Setup node id[22] uid[23] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 36] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp=100, scale=0.011393]
D [print_tensor:146]out(0): id[ 39] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.004456]
D [setup_node:441]Setup node id[23] uid[24] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 39] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.004456]
D [print_tensor:146]in(1) : id[ 38] vtl[0] const[1] shape[ 3, 3, 24, 24 ] fmt[u8 ] qnt[ASM zp=111, scale=0.006807]
D [print_tensor:146]in(2) : id[ 37] vtl[0] const[1] shape[ 24 ] fmt[i32] qnt[ASM zp= 0, scale=0.000030]
D [print_tensor:146]out(0): id[ 110] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.005790]
D [setup_node:441]Setup node id[24] uid[25] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 110] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.005790]
D [print_tensor:146]out(0): id[ 40] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.005790]
D [setup_node:441]Setup node id[25] uid[26] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 40] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.005790]
D [print_tensor:146]in(1) : id[ 41] vtl[0] const[1] shape[ 3, 3, 24, 24 ] fmt[u8 ] qnt[ASM zp=112, scale=0.006564]
D [print_tensor:146]in(2) : id[ 42] vtl[0] const[1] shape[ 24 ] fmt[i32] qnt[ASM zp= 0, scale=0.000038]
D [print_tensor:146]out(0): id[ 43] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp=123, scale=0.044463]
D [setup_node:441]Setup node id[26] uid[27] op[ADD]
D [validate_op_io_types:165]Validate [ADD]
D [print_tensor:146]in(0) : id[ 43] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp=123, scale=0.044463]
D [print_tensor:146]in(1) : id[ 36] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp=100, scale=0.011393]
D [print_tensor:146]out(0): id[ 44] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.021985]
D [setup_node:441]Setup node id[27] uid[28] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 44] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.021985]
D [print_tensor:146]out(0): id[ 47] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.021985]
D [setup_node:441]Setup node id[28] uid[29] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 47] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.021985]
D [print_tensor:146]in(1) : id[ 46] vtl[0] const[1] shape[ 3, 3, 24, 48 ] fmt[u8 ] qnt[ASM zp=106, scale=0.004273]
D [print_tensor:146]in(2) : id[ 45] vtl[0] const[1] shape[ 48 ] fmt[i32] qnt[ASM zp= 0, scale=0.000094]
D [print_tensor:146]out(0): id[ 111] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.029140]
D [setup_node:441]Setup node id[29] uid[30] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 111] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.029140]
D [print_tensor:146]out(0): id[ 50] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.029140]
D [setup_node:441]Setup node id[30] uid[31] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 47] vtl[1] const[0] shape[ 56, 64, 24, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.021985]
D [print_tensor:146]in(1) : id[ 48] vtl[0] const[1] shape[ 1, 1, 24, 48 ] fmt[u8 ] qnt[ASM zp=124, scale=0.005828]
D [print_tensor:146]in(2) : id[ 49] vtl[0] const[1] shape[ 48 ] fmt[i32] qnt[ASM zp= 0, scale=0.000128]
D [print_tensor:146]out(0): id[ 53] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp=141, scale=0.027526]
D [setup_node:441]Setup node id[31] uid[32] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 50] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.029140]
D [print_tensor:146]in(1) : id[ 52] vtl[0] const[1] shape[ 3, 3, 48, 48 ] fmt[u8 ] qnt[ASM zp=125, scale=0.004647]
D [print_tensor:146]in(2) : id[ 51] vtl[0] const[1] shape[ 48 ] fmt[i32] qnt[ASM zp= 0, scale=0.000135]
D [print_tensor:146]out(0): id[ 54] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp=134, scale=0.114629]
D [setup_node:441]Setup node id[32] uid[33] op[ADD]
D [validate_op_io_types:165]Validate [ADD]
D [print_tensor:146]in(0) : id[ 54] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp=134, scale=0.114629]
D [print_tensor:146]in(1) : id[ 53] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp=141, scale=0.027526]
D [print_tensor:146]out(0): id[ 55] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp=146, scale=0.112530]
D [setup_node:441]Setup node id[33] uid[34] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 55] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp=146, scale=0.112530]
D [print_tensor:146]out(0): id[ 58] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.044880]
D [setup_node:441]Setup node id[34] uid[35] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 58] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.044880]
D [print_tensor:146]in(1) : id[ 57] vtl[0] const[1] shape[ 3, 3, 48, 48 ] fmt[u8 ] qnt[ASM zp=105, scale=0.006360]
D [print_tensor:146]in(2) : id[ 56] vtl[0] const[1] shape[ 48 ] fmt[i32] qnt[ASM zp= 0, scale=0.000285]
D [print_tensor:146]out(0): id[ 112] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.086943]
D [setup_node:441]Setup node id[35] uid[36] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 112] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.086943]
D [print_tensor:146]out(0): id[ 59] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.086943]
D [setup_node:441]Setup node id[36] uid[37] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 59] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.086943]
D [print_tensor:146]in(1) : id[ 60] vtl[0] const[1] shape[ 3, 3, 48, 48 ] fmt[u8 ] qnt[ASM zp=164, scale=0.007681]
D [print_tensor:146]in(2) : id[ 61] vtl[0] const[1] shape[ 48 ] fmt[i32] qnt[ASM zp= 0, scale=0.000668]
D [print_tensor:146]out(0): id[ 62] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp=159, scale=0.370318]
D [setup_node:441]Setup node id[37] uid[38] op[ADD]
D [validate_op_io_types:165]Validate [ADD]
D [print_tensor:146]in(0) : id[ 62] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp=159, scale=0.370318]
D [print_tensor:146]in(1) : id[ 55] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp=146, scale=0.112530]
D [print_tensor:146]out(0): id[ 63] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.163684]
D [setup_node:441]Setup node id[38] uid[39] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 63] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.163684]
D [print_tensor:146]out(0): id[ 66] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.163684]
D [setup_node:441]Setup node id[41] uid[42] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 66] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.163684]
D [print_tensor:146]in(1) : id[ 67] vtl[0] const[1] shape[ 1, 1, 48, 64 ] fmt[u8 ] qnt[ASM zp=129, scale=0.003319]
D [print_tensor:146]in(2) : id[ 68] vtl[0] const[1] shape[ 64 ] fmt[i32] qnt[ASM zp= 0, scale=0.000543]
D [print_tensor:146]out(0): id[ 72] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp=103, scale=0.108540]
D [setup_node:441]Setup node id[39] uid[40] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 66] vtl[1] const[0] shape[ 28, 32, 48, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.163684]
D [print_tensor:146]in(1) : id[ 65] vtl[0] const[1] shape[ 3, 3, 48, 64 ] fmt[u8 ] qnt[ASM zp=129, scale=0.005332]
D [print_tensor:146]in(2) : id[ 64] vtl[0] const[1] shape[ 64 ] fmt[i32] qnt[ASM zp= 0, scale=0.000873]
D [print_tensor:146]out(0): id[ 113] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.231447]
D [setup_node:441]Setup node id[40] uid[41] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 113] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.231447]
D [print_tensor:146]out(0): id[ 69] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.231447]
D [setup_node:441]Setup node id[42] uid[43] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 69] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.231447]
D [print_tensor:146]in(1) : id[ 71] vtl[0] const[1] shape[ 3, 3, 64, 64 ] fmt[u8 ] qnt[ASM zp=134, scale=0.003610]
D [print_tensor:146]in(2) : id[ 70] vtl[0] const[1] shape[ 64 ] fmt[i32] qnt[ASM zp= 0, scale=0.000836]
D [print_tensor:146]out(0): id[ 73] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp=140, scale=0.829758]
D [setup_node:441]Setup node id[43] uid[44] op[ADD]
D [validate_op_io_types:165]Validate [ADD]
D [print_tensor:146]in(0) : id[ 73] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp=140, scale=0.829758]
D [print_tensor:146]in(1) : id[ 72] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp=103, scale=0.108540]
D [print_tensor:146]out(0): id[ 74] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp=138, scale=0.906703]
D [setup_node:441]Setup node id[44] uid[45] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 74] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp=138, scale=0.906703]
D [print_tensor:146]out(0): id[ 77] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.397434]
D [setup_node:441]Setup node id[45] uid[46] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 77] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.397434]
D [print_tensor:146]in(1) : id[ 76] vtl[0] const[1] shape[ 3, 3, 64, 64 ] fmt[u8 ] qnt[ASM zp= 94, scale=0.005339]
D [print_tensor:146]in(2) : id[ 75] vtl[0] const[1] shape[ 64 ] fmt[i32] qnt[ASM zp= 0, scale=0.002122]
D [print_tensor:146]out(0): id[ 114] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.446657]
D [setup_node:441]Setup node id[46] uid[47] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 114] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.446657]
D [print_tensor:146]out(0): id[ 78] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.446657]
D [setup_node:441]Setup node id[47] uid[48] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 78] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.446657]
D [print_tensor:146]in(1) : id[ 80] vtl[0] const[1] shape[ 3, 3, 64, 64 ] fmt[u8 ] qnt[ASM zp=124, scale=0.004056]
D [print_tensor:146]in(2) : id[ 79] vtl[0] const[1] shape[ 64 ] fmt[i32] qnt[ASM zp= 0, scale=0.001812]
D [print_tensor:146]out(0): id[ 81] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp=184, scale=2.196078]
D [setup_node:441]Setup node id[48] uid[49] op[ADD]
D [validate_op_io_types:165]Validate [ADD]
D [print_tensor:146]in(0) : id[ 81] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp=184, scale=2.196078]
D [print_tensor:146]in(1) : id[ 74] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp=138, scale=0.906703]
D [print_tensor:146]out(0): id[ 82] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.586414]
D [setup_node:441]Setup node id[49] uid[50] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 82] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.586414]
D [print_tensor:146]out(0): id[ 85] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.586414]
D [setup_node:441]Setup node id[52] uid[53] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 85] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.586414]
D [print_tensor:146]in(1) : id[ 86] vtl[0] const[1] shape[ 1, 1, 64, 128 ] fmt[u8 ] qnt[ASM zp=118, scale=0.002226]
D [print_tensor:146]in(2) : id[ 87] vtl[0] const[1] shape[ 128 ] fmt[i32] qnt[ASM zp= 0, scale=0.001305]
D [print_tensor:146]out(0): id[ 91] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp=138, scale=0.493036]
D [setup_node:441]Setup node id[50] uid[51] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 85] vtl[1] const[0] shape[ 14, 16, 64, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.586414]
D [print_tensor:146]in(1) : id[ 84] vtl[0] const[1] shape[ 3, 3, 64, 128 ] fmt[u8 ] qnt[ASM zp=124, scale=0.004449]
D [print_tensor:146]in(2) : id[ 83] vtl[0] const[1] shape[ 128 ] fmt[i32] qnt[ASM zp= 0, scale=0.002609]
D [print_tensor:146]out(0): id[ 115] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.594060]
D [setup_node:441]Setup node id[51] uid[52] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 115] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.594060]
D [print_tensor:146]out(0): id[ 88] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.594060]
D [setup_node:441]Setup node id[53] uid[54] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 88] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.594060]
D [print_tensor:146]in(1) : id[ 90] vtl[0] const[1] shape[ 3, 3, 128, 128 ] fmt[u8 ] qnt[ASM zp=129, scale=0.006039]
D [print_tensor:146]in(2) : id[ 89] vtl[0] const[1] shape[ 128 ] fmt[i32] qnt[ASM zp= 0, scale=0.003588]
D [print_tensor:146]out(0): id[ 92] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp=155, scale=1.879793]
D [setup_node:441]Setup node id[54] uid[55] op[ADD]
D [validate_op_io_types:165]Validate [ADD]
D [print_tensor:146]in(0) : id[ 92] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp=155, scale=1.879793]
D [print_tensor:146]in(1) : id[ 91] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp=138, scale=0.493036]
D [print_tensor:146]out(0): id[ 93] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp=153, scale=1.886373]
D [setup_node:441]Setup node id[55] uid[56] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 93] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp=153, scale=1.886373]
D [print_tensor:146]out(0): id[ 96] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.714904]
D [setup_node:441]Setup node id[56] uid[57] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 96] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.714904]
D [print_tensor:146]in(1) : id[ 95] vtl[0] const[1] shape[ 3, 3, 128, 128 ] fmt[u8 ] qnt[ASM zp=178, scale=0.010573]
D [print_tensor:146]in(2) : id[ 94] vtl[0] const[1] shape[ 128 ] fmt[i32] qnt[ASM zp= 0, scale=0.007559]
D [print_tensor:146]out(0): id[ 116] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=1.100026]
D [setup_node:441]Setup node id[57] uid[58] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 116] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=1.100026]
D [print_tensor:146]out(0): id[ 97] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=1.100026]
D [setup_node:441]Setup node id[58] uid[59] op[CONV2D]
D [validate_op_io_types:165]Validate [CONV2D]
D [print_tensor:146]in(0) : id[ 97] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=1.100026]
D [print_tensor:146]in(1) : id[ 99] vtl[0] const[1] shape[ 3, 3, 128, 128 ] fmt[u8 ] qnt[ASM zp=163, scale=0.008378]
D [print_tensor:146]in(2) : id[ 98] vtl[0] const[1] shape[ 128 ] fmt[i32] qnt[ASM zp= 0, scale=0.009216]
D [print_tensor:146]out(0): id[ 100] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp=197, scale=4.976527]
D [setup_node:441]Setup node id[59] uid[60] op[ADD]
D [validate_op_io_types:165]Validate [ADD]
D [print_tensor:146]in(0) : id[ 100] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp=197, scale=4.976527]
D [print_tensor:146]in(1) : id[ 93] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp=153, scale=1.886373]
D [print_tensor:146]out(0): id[ 101] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.811229]
D [setup_node:441]Setup node id[60] uid[61] op[RELU]
D [validate_op_io_types:165]Validate [RELU]
D [print_tensor:146]in(0) : id[ 101] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.811229]
D [print_tensor:146]out(0): id[ 102] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.811229]
D [setup_node:441]Setup node id[61] uid[62] op[POOL]
D [validate_op_io_types:165]Validate [POOL]
D [print_tensor:146]in(0) : id[ 102] vtl[1] const[0] shape[ 7, 8, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.811229]
D [print_tensor:146]out(0): id[ 103] vtl[1] const[0] shape[ 3, 4, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.811229]
D [setup_node:441]Setup node id[62] uid[63] op[RESHAPE]
D [print_tensor:146]in(0) : id[ 103] vtl[1] const[0] shape[ 3, 4, 128, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.811229]
D [print_tensor:146]out(0): id[ 104] vtl[1] const[0] shape[ 1, 1, 1536, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.811229]
D [setup_node:441]Setup node id[63] uid[64] op[L2_NORMALIZE]
D [validate_op_io_types:165]Validate [L2_NORMALIZE]
D [print_tensor:146]in(0) : id[ 104] vtl[1] const[0] shape[ 1, 1, 1536, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.811229]
D [print_tensor:146]out(0): id[ 0] vtl[0] const[0] shape[ 1, 1, 1536, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.000651]
D [optimize_node:385]Backward optimize neural network
D [op_optimize:113]Optimize RESHAPE, uid 63
D [op_optimize:113]Optimize RESHAPE, uid 11
D [optimize_node:392]Forward optimize neural network
D [op_optimize:113]Optimize RESHAPE, uid 11
D [op_optimize:113]Optimize RESHAPE, uid 63
I [compute_node:327]Create vx node
D [compute_node:350]Instance node[0] "POOL" ...
D [compute_node:350]Instance node[1] "CONV2D" ...
D [compute_node:350]Instance node[2] "RELU" ...
D [compute_node:350]Instance node[3] "POOL" ...
D [compute_node:350]Instance node[4] "CONV2D" ...
D [compute_node:350]Instance node[5] "RELU" ...
D [compute_node:350]Instance node[6] "POOL" ...
D [compute_node:350]Instance node[7] "CONV2D" ...
D [compute_node:350]Instance node[8] "RELU" ...
D [compute_node:350]Instance node[9] "POOL" ...
D [compute_node:350]Instance node[10] "RESHAPE" ...
D [compute_node:350]Instance node[11] "FCL2" ...
D [compute_node:350]Instance node[12] "RELU" ...
D [compute_node:350]Instance node[13] "FCL2" ...
D [compute_node:350]Instance node[14] "SPATIAL_TRANSFORMER" ...
Kernel "com.vivantecorp.extension.vxcTransform_setupThres_F16toF16" does not existI [vsi_nn_RegisterClientKernelAndNewNode:421]Register client kernel com.vivantecorp.extension.vxcTransform_setupThres_F16toF16 successfully.
Kernel "com.vivantecorp.extension.vxcTransform_Gemm_F16toF16" does not existI [vsi_nn_RegisterClientKernelAndNewNode:421]Register client kernel com.vivantecorp.extension.vxcTransform_Gemm_F16toF16 successfully.
Kernel "com.vivantecorp.extension.vxcTransform_InterP_F16toF16" does not existI [vsi_nn_RegisterClientKernelAndNewNode:421]Register client kernel com.vivantecorp.extension.vxcTransform_InterP_F16toF16 successfully.
D [compute_node:350]Instance node[15] "CONV2D" ...
D [compute_node:350]Instance node[16] "RELU" ...
D [compute_node:350]Instance node[17] "CONV2D" ...
D [compute_node:350]Instance node[18] "RELU" ...
D [compute_node:350]Instance node[19] "CONV2D" ...
D [compute_node:350]Instance node[20] "CONV2D" ...
D [compute_node:350]Instance node[21] "ADD" ...
D [vsi_nn_kernel_selector:970]Instance OPENVX node with kernel "add"
D [compute_node:350]Instance node[22] "RELU" ...
D [compute_node:350]Instance node[23] "CONV2D" ...
D [compute_node:350]Instance node[24] "RELU" ...
D [compute_node:350]Instance node[25] "CONV2D" ...
D [compute_node:350]Instance node[26] "ADD" ...
D [vsi_nn_kernel_selector:970]Instance OPENVX node with kernel "add"
D [compute_node:350]Instance node[27] "RELU" ...
D [compute_node:350]Instance node[28] "CONV2D" ...
D [compute_node:350]Instance node[29] "RELU" ...
D [compute_node:350]Instance node[30] "CONV2D" ...
D [compute_node:350]Instance node[31] "CONV2D" ...
D [compute_node:350]Instance node[32] "ADD" ...
D [vsi_nn_kernel_selector:970]Instance OPENVX node with kernel "add"
D [compute_node:350]Instance node[33] "RELU" ...
D [compute_node:350]Instance node[34] "CONV2D" ...
D [compute_node:350]Instance node[35] "RELU" ...
D [compute_node:350]Instance node[36] "CONV2D" ...
D [compute_node:350]Instance node[37] "ADD" ...
D [vsi_nn_kernel_selector:970]Instance OPENVX node with kernel "add"
D [compute_node:350]Instance node[38] "RELU" ...
D [compute_node:350]Instance node[41] "CONV2D" ...
D [compute_node:350]Instance node[39] "CONV2D" ...
D [compute_node:350]Instance node[40] "RELU" ...
D [compute_node:350]Instance node[42] "CONV2D" ...
D [compute_node:350]Instance node[43] "ADD" ...
D [vsi_nn_kernel_selector:970]Instance OPENVX node with kernel "add"
D [compute_node:350]Instance node[44] "RELU" ...
D [compute_node:350]Instance node[45] "CONV2D" ...
D [compute_node:350]Instance node[46] "RELU" ...
D [compute_node:350]Instance node[47] "CONV2D" ...
D [compute_node:350]Instance node[48] "ADD" ...
D [vsi_nn_kernel_selector:970]Instance OPENVX node with kernel "add"
D [compute_node:350]Instance node[49] "RELU" ...
D [compute_node:350]Instance node[52] "CONV2D" ...
D [compute_node:350]Instance node[50] "CONV2D" ...
D [compute_node:350]Instance node[51] "RELU" ...
D [compute_node:350]Instance node[53] "CONV2D" ...
D [compute_node:350]Instance node[54] "ADD" ...
D [vsi_nn_kernel_selector:970]Instance OPENVX node with kernel "add"
D [compute_node:350]Instance node[55] "RELU" ...
D [compute_node:350]Instance node[56] "CONV2D" ...
D [compute_node:350]Instance node[57] "RELU" ...
D [compute_node:350]Instance node[58] "CONV2D" ...
D [compute_node:350]Instance node[59] "ADD" ...
D [vsi_nn_kernel_selector:970]Instance OPENVX node with kernel "add"
D [compute_node:350]Instance node[60] "RELU" ...
D [compute_node:350]Instance node[61] "POOL" ...
D [compute_node:350]Instance node[62] "RESHAPE" ...
D [compute_node:350]Instance node[63] "L2_NORMALIZE" ...
---------------------------Begin VerifyTiling -------------------------@nightingalei ok, now i can see the "Register client kernel com.vivantecorp.extension.vxcTransform_setupThres_F16toF16 successfully", thanks; the unexpected result may comes from somewhere else;
nightingalei
commented
3 years ago @sky-fun The network seems to be success. Could you please make a unit test for SpatialTransformer from Tengine and make sure it's correct? The SpatialTransformer may have different implementations from different NN engine, TIM-VX uses this link(https://github.com/christopher5106/last_caffe_with_stn/blob/master/src/caffe/layers/st_layer.cpp).
sky-fun
commented
3 years ago @nightingalei sure， i will work on this to find the reason why i got wrong results。
sky-fun
commented
3 years ago @nightingalei , after check out the mxnet version SpatialTransformer , i find that there is some difference in the SpatialTransformer implementation: the order of GridData in vsi_nn_op_spatial_transformer.c is (y,x,1), the order of gemm result is (y,x); while in our mxnet model, GridData is (x,y,1), gemm result is (x,y); so after i manually change the theta data from (t0,t1,t2,t3,t4,t5) to (t4,t3,t5,t1,t0,t2), the SpatialTransformer result is close to the origin model.
thezha
commented
3 years ago @sky-fun Could you clarify, did you modify the implementation in tim-vx or the implementation in mxnet?
sky-fun
commented
3 years ago @thezha In fact, i add tim::vx::ops::Split and tim::vx::ops::Concat to change the order of second input data(the theta data) before i use the tim::vx::ops::SpatialTransformer. So i modify the implementation where tengine-lite add tim-vx node.
thezha
commented
3 years ago @sky-fun It's great that you can work around this. Let us know if you see any performance bottleneck caused by the additional channel shuffle operations.
sky-fun
commented
3 years ago @thezha i found that there is still some difference between the mxnet version and the caffe version： The grid data in mxnet is coordinate divide (W-1)and(H-1) (https://github.com/apache/incubator-mxnet/blob/8fd17cef2ee854239c6e66f2dd3b9467aa222f79/src/operator/spatial_transformer-inl.h#L99), while caffe version is coordinate divide W and H (https://github.com/christopher5106/last_caffe_with_stn/blob/037482abdfa801f9c606ae878ed8cd3dd2bec675/src/caffe/layers/st_layer.cpp#L118); And of course they are different in interp step when they transform gemm result to origin coordinate;
So is there anything i can do to fix this deviation?
thezha
commented
3 years ago In the next version, we will add align_corners attribute to spatial transform operation. It will be updated end of this month. Thanks!
rniranjan93
commented
3 years ago I have a .tflite file , I want to build a graph with Delegate class and want to return the graph using the following method
std::shared_ptr<tim::vx::Graph>& GetGraph() { return graph_; }
could you provide an example , i.e, taking .tflite as input -> returning a builded tim::vx::Graph
sunshinemyson
commented
3 years ago I have a .tflite file , I want to build a graph with Delegate class and want to return the graph using the following method
std::shared_ptr<tim::vx::Graph>& GetGraph() { return graph_; }could you provide an example , i.e, taking .tflite as input -> returning a builded tim::vx::Graph
What's the purpose for this usage? It's quite strange to me if you want use tflite interpreter to extract model but don't want to inference it with tflite interpreter.
thezha
commented
3 years ago @sky-fun align_corners option is added to SpatialTransformer 6a949bb315b7d711cf6f73056f1022ae23d32579
sky-fun
commented
3 years ago @thezha thanks
I try to run SpatialTransformer(mxnet) operator on rv1126 using Tengine, but the Status of "SpatialTransformer" operator in tim-vx is "InternalOnly", and no TIM-VX API implementation.So i can't find a way to add SpatialTransformer npu support to Tengine, i wondered that：