GPU coder :basic Code Generation failed

Visual studio 2019

MATLAB 2022A

CUDA 11.6,When I test

coder.checkGpuInstall:
 
Compatible GPU           : PASSED 
CUDA Environment         : PASSED 
	Runtime   : PASSED 
	cuFFT     : PASSED 
	cuSOLVER  : PASSED 
	cuBLAS    : PASSED 
cuDNN Environment        : FAILED (Unable to find the 'NVIDIA_CUDNN' environment variable. Set 'NVIDIA_CUDNN' to point to the root directory of a NVIDIA cuDNN installation.)
Basic Code Generation    : FAILED (GPU code generation failed with the error 'emlc:compilationError'. View report for further information: View report)

ans = 

  包含以下字段的 struct:

                 gpu: 1
                cuda: 1
               cudnn: 0
            tensorrt: 0
        basiccodegen: 0
       basiccodeexec: 0
         deepcodegen: 0
        deepcodeexec: 0
    tensorrtdatatype: 0
           profiling: 0

If I type in following codes in the commander line

gpuEnvObj = coder.gpuEnvConfig;
gpuEnvObj.GpuId = 0;
gpuEnvObj.BasicCodegen = 1;
gpuEnvObj.BasicCodeexec = 1;
results = coder.checkGpuInstall(gpuEnvObj)

the results is

Compatible GPU           : FAILED (Invalid CUDA device id: 1. Select a device id from the range 0:0.)
CUDA Environment         : PASSED 
	Runtime   : PASSED 
	cuFFT     : PASSED 
	cuSOLVER  : PASSED 
	cuBLAS    : PASSED 
Basic Code Generation    : FAILED (GPU code generation failed with the error 'emlc:compilationError'. View report for further information: View report)

results = 

  包含以下字段的 struct:

                 gpu: 1
                cuda: 1
               cudnn: 0
            tensorrt: 0
        basiccodegen: 0
       basiccodeexec: 0
         deepcodegen: 0
        deepcodeexec: 0
    tensorrtdatatype: 0
           profiling: 0

>>mex -setup
MEX 配置为使用 'Microsoft Visual C++ 2019 (C)' 以进行 C 语言编译。

要选择不同的 C 编译器，请从以下选项中选择一种命令:
Microsoft Visual C++ 2019 (C)  mex -setup:C:\Users\sunla\AppData\Roaming\MathWorks\MATLAB\R2022a\mex_C_win64.xml C
Microsoft Visual C++ 2022 (C)  mex -setup:D:\MATLAB2022a\bin\win64\mexopts\msvc2022.xml C

要选择不同的语言，请从以下选项中选择一种命令:
 mex -setup C++ 
 mex -setup FORTRAN

I looked throught the answers about 'emlc:compilationError' and "Invalid CUDA device", but no answer could figures out my problem.

what wrong with my gpu coder?

[Chao Luo]

Hi Caiwei,

Could you also show us the error message in the report by clicking "View report"?

Thanks,

Chao

[Hariprasad Ravishankar]

Hi Caiwei,

From the error message (Invalid CUDA device id: 1. Select a device id from the range 0:0.) it seems that the GpuId might have been 1?

Can you try running

>> gpuDeviceCount

and

>> gpuDeviceTable

to see if MATLAB is able to identify your gpu?

Answer 1

[Walter Roberson]

I think Cuda Toolkit 11.6 is not supported in your release. I am having difficulty finding the toolkit version that is supported: 11.0 for the release before yours.

You might have to enable forward compatibility https://www.mathworks.com/help/parallel-computing/parallel.gpu.enablecudaforwardcompatibility.html#mw_f28510a0-2087-45eb-84d4-80a0b1248e37

[zhou caiwei]

I can run mexcuda and gpu mexw64 file. The fault occurs only when I use gpu coder

[zhou caiwei]

After I settle down the tensorrt and cudnn,the report becomes:

>>> gpuEnvObj = coder.gpuEnvConfig;
gpuEnvObj.GpuId = 0;
gpuEnvObj.DeepLibTarget = 'tensorrt';
gpuEnvObj.DeepCodegen = 1;
gpuEnvObj.DeepCodeexec = 1;
coder.checkGpuInstall(gpuEnvObj)

>>>
Runtime   : PASSED 
cuFFT     : PASSED 
cuSOLVER  : PASSED 
cuBLAS    : PASSED 
cuDNN Environment        : PASSED (Warning: Deep learning code generation has been tested with cuDNN v8.1. The provided cuDNN library v8.5 may not be fully compatible.)
TensorRT Environment     : PASSED (Warning: Deep learning code generation has been tested with TensorRT v7.2. The provided TensorRT library v8.4 may not be fully compatible.)
Deep Learning (TensorRT) Code Generation: FAILED (Test GPU code generation failed with the error 'emlc:compilationError'. View report for further information: View report)

[zhou caiwei]

So, how do you think if I install MATLAB 2022b? could MATLAB 2022b release solve my problem?

[zhou caiwei]

Name: 'NVIDIA GeForce RTX 3090'

Index: 1

ComputeCapability: '8.6'

SupportsDouble: 1

DriverVersion: 11.6000

ToolkitVersion: 11

MaxThreadsPerBlock: 1024

MaxShmemPerBlock: 49152

MaxThreadBlockSize: [1024 1024 64]

MaxGridSize: [2.1475e+09 65535 65535]

SIMDWidth: 32

TotalMemory: 2.5769e+10

AvailableMemory: 2.4283e+10

MultiprocessorCount: 82

ClockRateKHz: 1695000

ComputeMode: 'Default'

GPUOverlapsTransfers: 1

KernelExecutionTimeout: 1

CanMapHostMemory: 1

DeviceSupported: 1

DeviceAvailable: 1

DeviceSelected: 1

[Walter Roberson]

I think the 3090 is too new to be supported in your release unless you turn on forward compatibility.

[zhou caiwei]

definitely yes! I downloaded MATLAB2022a and all the test passed!

[Walter Roberson]

@zhou caiwei

Do you perhaps means it works in R2022b ??

[zhou caiwei]

MATLAB2022a is proved to be qualified

[Walter Roberson]

See also https://www.mathworks.com/matlabcentral/answers/1825838-gpu-support-for-rtx-4090#comment_2430013 which indicates that R2021a is enough to generally use the RTX 3090, but that using GPU Coder could have problems, with a work-around suggestion there.

Answer 2

[zhou caiwei]

However, I'm not sure if other error would happens

[wtg Wtg]

Hi zhou caiwei ,

Could you instruct me how you solved the problem ?

I always got the following error messages for the Deep code gen & exec tests :

>> coder.checkGpuInstall

Compatible GPU : PASSED

CUDA Environment : PASSED

Runtime : PASSED

cuFFT : PASSED

cuSOLVER : PASSED

cuBLAS : PASSED

cuDNN Environment : PASSED

Host Compiler : PASSED

Basic Code Generation : PASSED

Basic Code Execution : PASSED

gpu: 1

cuda: 1

cudnn: 1

tensorrt: 0

hostcompiler: 1

basiccodegen: 1

basiccodeexec: 1

deepcodegen: 0

tensorrtdatatype: 0

deepcodeexec: 0

Deep Learning (cuDNN) Code Generation: FAILED (GPU code generation failed with an error. View report for further information: View report

Report shows :

Calling cade generaion implementation function 'xxx' is from user function '... DeepLearningNetwork' is Not Supported.

....

Regards, Wtg

[Justin Hontz]

@wtg Wtg For better visibility and organization, please create a new MATLAB Answers thread for the issue you are encountering.

[wtg Wtg]

Hi Justin Hontz ,

I had created a new Ask with a subject named "GPU coder check fail for deep learning cuDNN code generation and execution".