Application Acceleration With FPGAs
Programmable Acceleration Cards (PACs), DCP, FPGA AI Suite, Software Stack, and Reference Designs
504 討論

Running HARP example design returns error value -54.

whan01
初學者
1,623 檢視

I modify the host code of example design.

The original part in launching the first kernel is

status = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, gsize, lsize, 0, NULL, NULL)

I modify this part to

// define NDRange Size size_t gsize[1]={(size_t) 10}; size_t lsize[1] = {(size_t) 2}; status = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, gsize, lsize, 0, NULL, NULL)

Then I run the code it returns error code -54. Why can't I increase the 1-d lsize to 2?

0 積分
1 解決方案
HRZ
傑出貢獻者 III
1,318 檢視

Is there any reqd_work_group_size or max_work_group_size used in the kernel?

在原始文章中檢視解決方案

6 回應
KhaiChein_Y_Intel
1,318 檢視

Hi,

 

May I know the software version and edition you are using? Could you share the error message and provide the design for investigation?

 

Thanks.

whan01
初學者
1,318 檢視

The software is quartus 16.0.2 pro with all update patches installed. My OpenCL is v2.0.

HRZ
傑出貢獻者 III
1,319 檢視

Is there any reqd_work_group_size or max_work_group_size used in the kernel?

whan01
初學者
1,318 檢視

You remind me. There's neither reqd_work_group_size nor max_work_group_size, but at the head of each kernel function there's a declaration __attribute__((task)) which implicitly indicates the max work group dimension is 0. The max group work items is 1. So here's problem. Thanks!

HRZ
傑出貢獻者 III
1,318 檢視

So the kernel is not even NDRange. 😅

Be careful that just removing that attribute will not convert the kernel to NDRange; you will need to rewrite the kernel and replace the loops in it with get_global_id()/get_local_id().

whan01
初學者
1,318 檢視

Thanks for the further reminder. I'm new to HARP2 and OpenCL and maybe I need to go back to the programming manual.

回覆