userxx

Linux OpenCL not working AMDGPU-PRO (max global size 0, CL_OUT_OF_HOST_MEMORY)

Discussion created by userxx on May 24, 2017
Latest reply on Feb 20, 2019 by dipak

Hello,

 

I tried to post in the OpenCL forum but I cannot create new discussions.

 

So I paste here my problem.

 

I have installed the amdgpu-pro driver just for opencl.

 

If I execute clinfo I get:

Number of platforms:                         
3
  Platform Profile:                          
FULL_PROFILE
  Platform Version:                          
OpenCL 2.0 AMD-APP (2348.3)
  Platform Name:                             
AMD Accelerated Parallel Processing
  Platform Vendor:                           
Advanced Micro Devices, Inc.
  Platform Extensions:                       
cl_khr_icd cl_amd_event_callback cl_amd_offline_devices
  Platform Profile:                          
FULL_PROFILE
  Platform Version:                          
OpenCL 1.1 Mesa 17.1.0
  Platform Name:                             
Clover
  Platform Vendor:                           
Mesa
  Platform Extensions:                       
cl_khr_icd
  Platform Profile:                          
FULL_PROFILE
  Platform Version:                          
OpenCL 1.2 LINUX
  Platform Name:                             
Intel(R) OpenCL
  Platform Vendor:                           
Intel(R) Corporation
  Platform Extensions:                       
cl_khr_icd cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store cl_khr_depth_images cl_khr_3d_image_writes cl_intel_exec_by_local_thread cl_khr_spir cl_khr_fp64



  Platform Name:                             
AMD Accelerated Parallel Processing
Number of devices:                           
2
  Device Type:                               
CL_DEVICE_TYPE_GPU
  Vendor ID:                                 
1002h
  Board name:                                
AMD Radeon (TM) R7 300 Series
  Device Topology:                           
PCI[ B#1, D#0, F#0 ]
  Max compute units:                         
4
  Max work items dimensions:                 
3

Max work items[0]:                       
256

Max work items[1]:                       
256

Max work items[2]:                       
256
  Max work group size:                       
256
  Preferred vector width char:               
4
  Preferred vector width short:              
2
  Preferred vector width int:                
1
  Preferred vector width long:               
1
  Preferred vector width float:              
1
  Preferred vector width double:             
1
  Native vector width char:                  
4
  Native vector width short:                 
2
  Native vector width int:                   
1
  Native vector width long:                  
1
  Native vector width float:                 
1
  Native vector width double:                
1
  Max clock frequency:                       
985Mhz
  Address bits:                              
32
  Max memory allocation:                     
1319064576
  Image support:                             
Yes
  Max number of images read arguments:       
128
  Max number of images write arguments:      
8
  Max image 2D width:                        
16384
  Max image 2D height:                       
16384
  Max image 3D width:                        
2048
  Max image 3D height:                       
2048
  Max image 3D depth:                        
2048
  Max samplers within kernel:                
16
  Max size of kernel argument:               
1024
  Alignment (bits) of base address:          
2048
  Minimum alignment (bytes) for any datatype:
128

  Single precision floating point capability

Denorms:                                 
No

Quiet NaNs:                              
Yes

Round to nearest even:                   
Yes

Round to zero:                           
Yes

Round to +ve and infinity:               
Yes

IEEE754-2008 fused multiply-add:         
Yes
  Cache type:                                
Read/Write
  Cache line size:                           
64
  Cache size:                                
16384
  Global memory size:                        
1971261440
  Constant buffer size:                      
65536
  Max number of constant args:               
8
  Local memory type:                         
Scratchpad
  Local memory size:                         
32768
  Max pipe arguments:                        
0
  Max pipe active reservations:              
0
  Max pipe packet size:                      
0
  Max global variable size:                  
0
  Max global variable preferred total size:  
0
  Max read/write image args:                 
0
  Max on device events:                      
0
  Queue on device max size:                  
0
  Max on device queues:                      
0
  Queue on device preferred size:            
0

  SVM capabilities:

Coarse grain buffer:                     
No

Fine grain buffer:                       
No

Fine grain system:                       
No

Atomics:                                 
No
  Preferred platform atomic alignment:       
0
  Preferred global atomic alignment:         
0
  Preferred local atomic alignment:          
0
  Kernel Preferred work group size multiple: 
64
  Error correction support:                  
0
  Unified memory for Host and Device:        
0
  Profiling timer resolution:                
1
  Device endianess:                          
Little
  Available:                                 
Yes
  Compiler available:                        
Yes

  Execution capabilities:

Execute OpenCL kernels:                  
Yes

Execute native function:                 
No

  Queue on Host properties:

Out-of-Order:                            
No

Profiling :                              
Yes

  Queue on Device properties:

Out-of-Order:                            
No

Profiling :                              
No
  Platform ID:                               
0x7f176ac9de98
  Name:                                      
Hainan
  Vendor:                                    
Advanced Micro Devices, Inc.
  Device OpenCL C version:                   
OpenCL C 1.2
  Driver version:                            
2348.3
  Profile:                                   
FULL_PROFILE
  Version:                                   
OpenCL 1.2 AMD-APP (2348.3)
  Extensions:                                
cl_khr_fp64 cl_amd_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_gl_sharing cl_amd_device_attribute_query cl_amd_vec3 cl_amd_printf cl_amd_media_ops cl_amd_media_ops2 cl_amd_popcnt cl_khr_image2d_from_buffer cl_khr_spir cl_khr_gl_event



  Device Type:                               
CL_DEVICE_TYPE_CPU
  Vendor ID:                                 
1002h

  Board name:
  Max compute units:                         
8
  Max work items dimensions:                 
3

Max work items[0]:                       
1024

Max work items[1]:                       
1024

Max work items[2]:                       
1024
  Max work group size:                       
1024
  Preferred vector width char:               
16
  Preferred vector width short:              
8
  Preferred vector width int:                
4
  Preferred vector width long:               
2
  Preferred vector width float:              
8
  Preferred vector width double:             
4
  Native vector width char:                  
16
  Native vector width short:                 
8
  Native vector width int:                   
4
  Native vector width long:                  
2
  Native vector width float:                 
8
  Native vector width double:                
4
  Max clock frequency:                       
3200Mhz
  Address bits:                              
64
  Max memory allocation:                     
4198982656
  Image support:                             
Yes
  Max number of images read arguments:       
128
  Max number of images write arguments:      
64
  Max image 2D width:                        
8192
  Max image 2D height:                       
8192
  Max image 3D width:                        
2048
  Max image 3D height:                       
2048
  Max image 3D depth:                        
2048
  Max samplers within kernel:                
16
  Max size of kernel argument:               
4096
  Alignment (bits) of base address:          
1024
  Minimum alignment (bytes) for any datatype:
128

  Single precision floating point capability

Denorms:                                 
Yes

Quiet NaNs:                              
Yes

Round to nearest even:                   
Yes

Round to zero:                           
Yes

Round to +ve and infinity:               
Yes

IEEE754-2008 fused multiply-add:         
Yes
  Cache type:                                
Read/Write
  Cache line size:                           
64
  Cache size:                                
16384
  Global memory size:                        
16795930624
  Constant buffer size:                      
65536
  Max number of constant args:               
8
  Local memory type:                         
Global
  Local memory size:                         
32768
  Max pipe arguments:                        
16
  Max pipe active reservations:              
16
  Max pipe packet size:                      
4198982656
  Max global variable size:                  
1879048192
  Max global variable preferred total size:  
1879048192
  Max read/write image args:                 
64
  Max on device events:                      
0
  Queue on device max size:                  
0
  Max on device queues:                      
0
  Queue on device preferred size:            
0

  SVM capabilities:

Coarse grain buffer:                     
No

Fine grain buffer:                       
No

Fine grain system:                       
No

Atomics:                                 
No
  Preferred platform atomic alignment:       
0
  Preferred global atomic alignment:         
0
  Preferred local atomic alignment:          
0
  Kernel Preferred work group size multiple: 
1
  Error correction support:                  
0
  Unified memory for Host and Device:        
1
  Profiling timer resolution:                
1
  Device endianess:                          
Little
  Available:                                 
Yes
  Compiler available:                        
Yes

  Execution capabilities:

Execute OpenCL kernels:                  
Yes

Execute native function:                 
Yes

  Queue on Host properties:

Out-of-Order:                            
No

Profiling :                              
Yes

  Queue on Device properties:

Out-of-Order:                            
No

Profiling :                              
No
  Platform ID:                               
0x7f176ac9de98
  Name:                                      
AMD FX-8320E Eight-Core Processor
  Vendor:                                    
AuthenticAMD
  Device OpenCL C version:                   
OpenCL C 1.2
  Driver version:                            
2348.3 (sse2,avx,fma4)
  Profile:                                   
FULL_PROFILE
  Version:                                   
OpenCL 1.2 AMD-APP (2348.3)
  Extensions:                                
cl_khr_fp64 cl_amd_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_gl_sharing cl_ext_device_fission cl_amd_device_attribute_query cl_amd_vec3 cl_amd_printf cl_amd_media_ops cl_amd_media_ops2 cl_amd_popcnt cl_khr_spir cl_khr_gl_event



If I execute a simple application to run a kernel in the GPU device I get this:

 

platforms: 3 (0)
platform name: AMD Accelerated Parallel Processing
num devices: 2 (0)
devices: 2 (0)
context (0)
device name: Hainan
device type: CL_DEVICE_TYPE_GPU
command queue (-6 CL_OUT_OF_HOST_MEMORY)

create buffer a (0)
create buffer b (0)
create buffer c (0)
create program (0)
build program (0)
create kernel1 (0)
gws {1024, 1, 1}
lws {128, 1, 1}
set arg 0 (0)
set arg 1 (0)
set arg 2 (0)
set arg 3 (0)
nd range kernel1 (-36)
create user event (0)
c_v 0x6e5030
about to read the c buffer
read buffer c_v - buf_c (-36)
about to check (first: 0)
Wrong checking: a_v[0] = 0, b_v[0] = 1, c_v[0] = 0

 

 

As you can see it fails when creating the command queue (CL_OUT_OF_HOST_MEMORY). Checking the clinfo I can see weird values related to 'Max global variable size' and similars (always is 0). On the other side, if you check the CPU is has non zero values and it runs correctly, as shown below:

 

platforms: 3 (0)
platform name: AMD Accelerated Parallel Processing
num devices: 2 (0)
devices: 2 (0)
context (0)
device name: AMD FX-8320E Eight-Core Processor
device type: CL_DEVICE_TYPE_CPU
command queue (0 CL_SUCCESS)
mode CPU
create buffer a (0)
create buffer b (0)
create buffer c (0)
create program (0)
build program (0)
create kernel1 (0)
gws {1024, 1, 1}
lws {128, 1, 1}
set arg 0 (0)
set arg 1 (0)
set arg 2 (0)
set arg 3 (0)
nd range kernel1 (0)
create user event (0)
c_v 0x1607030
about to read the c buffer
read buffer c_v - buf_c (0)
about to check (first: 1)

 

 

What is that error and how can I solve it? I need OpenCL for the GPU.

Outcomes