ERROR: [KernelCheck 83-114] in sdx 2017.4

[yoneoka.noboru]

Hello,

I have updated sdx version 2017.4 from 2017.2 on KCU1500, and
I have a compilation error as follows:

ERROR: [KernelCheck 83-114] Kernel 'compute' port 'M_AXI_GMEM' is not mapped from any kernel argument.

It seems that the error occurs when
sdx 2017.4 and
the kernel does not have any global memory arguments and
--xp 'param:compiler.version=31' is defined.

The design has many pipes and uses 'get_pipe_num_packets' functions for controlling kernel data flow.
The compiler option --xp 'param:compiler.version=31' is necessary because sdx 2017.4 reports an another error without this option.
Gcc: ***.cl:201:'get_pipe_num_packets' is not supported by the current compiler; please use 'xocc --xp param:compiler.version=31' to enable 'get_pipe_num_packets'

To reproduce this issue,
in SDAccel_Examples/getting_started/kernel_to_gmem/window_array_2d_ocl,
modify Makefile to add the option --xp 'param:compiler.version=31'
and
$ make all TARGETS=hw_emu

Is there any solution?

Regards,
yoneoka

[brucey]

Please update the newest SDAccel  examples in Github. 

Previous version of examples is not suited for 2017.4.

[yoneoka.noboru]

Hi brucey,

I retried the latest version of SDAccel examples but it causes the same error

when adding --xp 'param:compiler.version=31'  like (2).

*****

commit 82ec0bcc682b93c54cfa8f4ac64c672a30e2c3f0

--- a/getting_started/kernel_to_gmem/window_array_2d_ocl/Makefile
+++ b/getting_started/kernel_to_gmem/window_array_2d_ocl/Makefile

-window_array_2d_CLFLAGS=-I./src/                                                         (1) original
+window_array_2d_CLFLAGS=-I./src/ --xp 'param:compiler.version=31'       (2) modified

*****

Regards,

Yoneoka

[seanz · Accepted solution]

Hi @yoneoka.noboru

Please add a dummy pointer interface in the compute function and see if the error can be resolved.

void compute(__global int *dummy, int alpha) {

[yoneoka.noboru]

Hi seanz

With adding a dummy pointer, there is no error in compiling and running.

I hope this issue will be fixed in a future version.

[seanz]

Hi Yoneoka-san,

Thanks for reporting this issue to us. 

Xilinx developers were aware of this and are looking at it.

Also, if the solution help to resolve your issue, can you please mark the appropriate post as an answer "Accept as solution".

This will help other users in future looking for the same query with an accepted answer.

Regards,

Sean