You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I had the error below when running resnet18 on pytorch.
This only happens for a specific batch size.
To reproduce the error one can execute the scripts provided fp16util.py.txt micro_benchmarking_pytorch.py.txt
with the following arguments python micro_benchmarking_pytorch.py --network resnet18 --batch-size 64 --iterations 20 --fp16 1
<stdin>:216:3: error: instruction not supported on this GPU
v_add_u32 v3, s8, v0
^
<stdin>:230:5: error: instruction not supported on this GPU
v_add_u32 v6, 0+MIO_BN_CHW, v3
^
<stdin>:236:3: error: instruction not supported on this GPU
v_add_co_u32 v8, s[4:5], s32, v8
^
<stdin>:238:3: error: instruction not supported on this GPU
v_addc_co_u32 v9, s[4:5], v4, v9, s[4:5]
^
<stdin>:242:3: error: instruction not supported on this GPU
v_add_co_u32 v6, s[4:5], s32, v6
^
<stdin>:243:3: error: instruction not supported on this GPU
v_addc_co_u32 v7, s[4:5], v10, v7, s[4:5]
^
<stdin>:245:3: error: instruction not supported on this GPU
global_load_ushort v4, v[6:7], off
^
<stdin>:246:3: error: instruction not supported on this GPU
global_load_ushort v6, v[8:9], off
^
<stdin>:249:3: error: instruction not supported on this GPU
v_add_u32 v11, 4, v5
^
<stdin>:251:3: error: instruction not supported on this GPU
v_add_u32 v5, 4, v5
^
<stdin>:256:9: error: instruction not supported on this GPU
v_add_u32 v3, 0+MIO_BN_CHW*2, v3
^
<stdin>:357:3: error: instruction not supported on this GPU
v_add_u32 v3, 4, v3
^
<stdin>:402:3: error: instruction not supported on this GPU
v_add_u32 v7, s8, v0
^
<stdin>:404:3: error: instruction not supported on this GPU
v_add_u32 v0, 4, v4
^
<stdin>:413:3: error: instruction not supported on this GPU
v_add_u32 v4, 2, v4
^
<stdin>:420:3: error: instruction not supported on this GPU
v_add_co_u32 v8, s[4:5], s24, v8
^
<stdin>:422:3: error: instruction not supported on this GPU
v_add_u32 v7, 0+MIO_BN_CHW, v7
^
<stdin>:426:3: error: instruction not supported on this GPU
v_addc_co_u32 v9, s[4:5], v10, v9, s[4:5]
^
<stdin>:446:3: error: instruction not supported on this GPU
global_store_short v[8:9], v0, off
^
<stdin>:475:6: error: instruction not supported on this GPU
global_store_dword v[4:5], v1, off
^
<stdin>:478:6: error: instruction not supported on this GPU
global_store_dword v[6:7], v3, off
^
<stdin>:490:6: error: instruction not supported on this GPU
global_load_dword v8, v[3:4], off
^
<stdin>:492:6: error: instruction not supported on this GPU
global_load_dword v9, v[5:6], off
^
<stdin>:522:6: error: instruction not supported on this GPU
global_store_dword v[3:4], v8, off
^
<stdin>:524:6: error: instruction not supported on this GPU
global_store_dword v[5:6], v2, off
^
MIOpen(HIP): Warning [AmdgcnAssemble] -x assembler -target amdgcn--amdhsa -mno-code-object-v3 -Wa,-defsym,MIOPEN_USE_FP16=0 -Wa,-defsym,MIOPEN_USE_FP32=0 -Wa,-defsym,MIOPEN_USE_FPMIX=1 -Wa,-defsym,MIO_SAVE_MEAN_VARIANCE=1 -Wa,-defsym,MIO_RUNNING_RESULT=1 -Wa,-defsym,MIO_BN_N=64 -Wa,-defsym,MIO_BN_C=128 -Wa,-defsym,MIO_BN_HW=784 -Wa,-defsym,MIO_BN_NHW=50176 -Wa,-defsym,MIO_BN_NHW_DIV=1065353383 -Wa,-defsym,MIO_BN_CHW=100352 -Wa,-defsym,MIO_BN_NCHW=6422528 -Wa,-defsym,MIO_BN_LDS_SIZE=832 -Wa,-defsym,MIO_BN_LDSGCN_SIZE=13 -Wa,-defsym,MIO_BN_VARIANT=3 -Wa,-defsym,MIO_BN_GRP0=832 -Wa,-defsym,MIO_BN_GRP1=1 -Wa,-defsym,MIO_BN_GRP2=1 -mcpu=gfx803 - -o /tmp/miopen-tmp-90cd-a69e-6095-dc78/amdgcn-asm-out-XXXXXX
MIOpen Error: /root/driver/MLOpen/src/ocl/gcn_asm_utils.cpp:228: Assembly error(1)
Traceback (most recent call last):
File "/home/setepenre/dev/bin/convnets", line 11, in <module>
load_entry_point('MixedPrecision==0.0.0', 'console_scripts', 'convnets')()
File "/home/setepenre/dev/lib/python3.6/site-packages/MixedPrecision-0.0.0-py3.6.egg/MixedPrecision/pytorch/convnet.py", line 246, in main
File "/home/setepenre/dev/lib/python3.6/site-packages/MixedPrecision-0.0.0-py3.6.egg/MixedPrecision/pytorch/convnet.py", line 237, in generic_main
File "/home/setepenre/dev/lib/python3.6/site-packages/MixedPrecision-0.0.0-py3.6.egg/MixedPrecision/pytorch/convnet.py", line 108, in train
File "/home/setepenre/rocm_pytorch/torch/nn/modules/module.py", line 494, in __call__
result = self.forward(*input, **kwargs)
File "/home/setepenre/rocm_pytorch/torch/nn/modules/container.py", line 97, in forward
input = module(input)
File "/home/setepenre/rocm_pytorch/torch/nn/modules/module.py", line 494, in __call__
result = self.forward(*input, **kwargs)
File "/home/setepenre/dev/lib/python3.6/site-packages/torchvision/models/resnet.py", line 156, in forward
x = self.layer2(x)
File "/home/setepenre/rocm_pytorch/torch/nn/modules/module.py", line 494, in __call__
result = self.forward(*input, **kwargs)
File "/home/setepenre/rocm_pytorch/torch/nn/modules/container.py", line 97, in forward
input = module(input)
File "/home/setepenre/rocm_pytorch/torch/nn/modules/module.py", line 494, in __call__
result = self.forward(*input, **kwargs)
File "/home/setepenre/dev/lib/python3.6/site-packages/torchvision/models/resnet.py", line 46, in forward
out = self.bn1(out)
File "/home/setepenre/rocm_pytorch/torch/nn/modules/module.py", line 494, in __call__
result = self.forward(*input, **kwargs)
File "/home/setepenre/rocm_pytorch/torch/nn/modules/batchnorm.py", line 83, in forward
exponential_average_factor, self.eps)
File "/home/setepenre/rocm_pytorch/torch/nn/functional.py", line 1691, in batch_norm
training, momentum, eps, torch.backends.cudnn.enabled
RuntimeError: miopenStatusUnknownError
With a batch size of 256 this does not happen.
python micro_benchmarking_pytorch.py --network resnet18 --batch-size 256 --iterations 20 --fp16 1
INFO: running forward and backward for warmup.
INFO: running the benchmark..
OK: finished running benchmark..
--------------------SUMMARY--------------------------
Microbenchmark for network : resnet18
Mini batch size [img] : 256
Time per mini-batch : 1.5049312353134154
Throughput [img/sec] : 170.10744012279451
The text was updated successfully, but these errors were encountered:
I had the error below when running resnet18 on pytorch.
This only happens for a specific batch size.
To reproduce the error one can execute the scripts provided
fp16util.py.txt
micro_benchmarking_pytorch.py.txt
with the following arguments
python micro_benchmarking_pytorch.py --network resnet18 --batch-size 64 --iterations 20 --fp16 1
With a batch size of 256 this does not happen.
The text was updated successfully, but these errors were encountered: