Re: Emulation of "asiam_option" example appears hangs

Altera_Forum · ‎03-02-2017

Hi,

I want to learn how to write an application composed of multiple kernels communicated through channels.

For that purpose, I am using the "asian_option" example downloaded from:

https://www.altera.com/support/support-resources/design-examples/design-software/opencl/black-scholes.html

This is the compiler version:


>>aocl version
aocl 16.1.0.196 (Intel(R) FPGA SDK for OpenCL(TM), Version 16.1.0 Build 196, Copyright (C) 2016 Intel Corporation)

This is how I compiled the very original code (both host and device)


>>make
>>aoc -march=emulator -v --board a10gx device/asian_option.cl -o bin/asian_option.aocx

Emulation:


>>CL_CONTEXT_EMULATOR_DEVICE_ALTERA=1 ./host 
Querying platform for info:
==========================
CL_PLATFORM_NAME                         = Intel(R) FPGA SDK for OpenCL(TM)
CL_PLATFORM_VENDOR                       = Altera Corporation
CL_PLATFORM_VERSION                      = OpenCL 1.0 Intel(R) FPGA SDK for OpenCL(TM), Version 16.1
Programming Device(s)
Using AOCX: asian_option.aocx
Starting Computations

Here is where NO OUTPUT appears. Checking the corresponding htop output for the emulation, I saw three processes utilizing almost 100% of CPU, and one process using almost 300% of CPU.

Using gdb, I discovered the program hangs on an specific API call:

main.cpp: 337, inside get_result():


double get_result(int device_id)
{# if USE_SVM_API == 0
   // Read back the single result from the kernel
   status = clEnqueueReadBuffer(accumulate_queue, kernel_result, CL_TRUE, 0, sizeof(cl_double), &X, 0, NULL, NULL);
...

Commenting that part, let the emulation complete, but obviously the "Resulting Price" value is wrong:


CL_CONTEXT_EMULATOR_DEVICE_ALTERA=1 ./host 
Querying platform for info:
==========================
CL_PLATFORM_NAME                         = Intel(R) FPGA SDK for OpenCL(TM)
CL_PLATFORM_VENDOR                       = Altera Corporation
CL_PLATFORM_VERSION                      = OpenCL 1.0 Intel(R) FPGA SDK for OpenCL(TM), Version 16.1
Programming Device(s)
Using AOCX: asian_option.aocx
Starting Computations
DEVICE 0: r=0.08 sigma=0.30 T=1.0 S0=30.0 K=29.0 : Resulting Price is 0.000000
1 Devices ran a total of 2.09715e+11 Simulations
Throughput = 275295.56 Billion Simulations / second

I do not understand the reason of this. The ran before the "hello_world" example without problems. But I am always facing hanging-related problem when using designs composed of multiple kernels and channels.

Could anyone please give some tips to solve this?

Regards,

Leo

Altera_Forum · ‎03-03-2017

The emulator is extremely slow (it is an emulator after all). In reality it might not be stuck at all and still computing the output. Try changing the input settings to lower values to see if you get any output.