topic Re:An interesting problem on DPCPP in Intel® oneAPI DPC++/C++ Compiler

An interesting problem on DPCPP

shenuo — Tue, 17 Aug 2021 04:24:02 GMT

kernelfunc_1(sycl::queue&Q, int *ptr1,int*ptr2) {

image im3(ptr1, image_channel_order::rgba, image_channel_type::unsigned_int32, range{ M,N,L });

try
{
auto ev3 = Q.submit([&](handler &h) {

auto out = sycl::stream(1024, 1024, h);
accessor<int4, 2, access::mode::write, access::target::image_array> acs3_w(im3, h);
im3.set_final_data(ptr1);
h.parallel_for(nd_range{ range{ M ,N,L}, range{ 1,1,1 } }, [=](nd_item<3> it) {
int idx = it.get_global_linear_id();
if (idx == 0) {
for (int k = 0; k < L; k++) {
out << "layer:" << k << ":" << sycl::endl;
for (int i = 0; i < M; i++) {
for (int j = 0; j < N; j++)
{
if (i == 0 && j == 0 && k == 0) continue;
acs3_w[k].write(int2(i, j), int4(1, 2, 3, 4));
}
out << sycl::endl;
}
}

}

});
}
);
return ev3;
}
catch (const std::exception& e)
{
std::cout << "An exception is caught for vector add.\n" << e.what();
std::terminate();
}
}

auto kernelfunc_2(sycl::queue&Q, int *ptr1, int*ptr2) {
Q.wait();
buffer ptr(ptr1, range{ N*M*L*C });
ptr.set_final_data(ptr2);

try
{
auto ev2 = Q.submit([&](handler &h) {
ptr.set_final_data(ptr2);
accessor acs(ptr, h);
auto out = sycl::stream(1024, 1024, h);
h.parallel_for(nd_range{ range{ M ,N,L}, range{ 1,1,1 } }, [=](nd_item<3> it) {

int idx = it.get_global_linear_id();
if (idx == 0) {
for (int k = 0; k < L; k++) {
acs[0] = 999;
}
}
});
}
);
return ev2;
}
catch (const std::exception& e)
{
std::cout << "An exception is caught for vector add.\n" << e.what();
std::terminate();
}
}

int main(){

kernelfunc_1(Q, host_array3_2,host_array3_2);

kernelfunc_2(Q, shared_array3_2, shared_array3_2);

---------------------v1---------------------------------------------

kernelfunc_1(Q, host_array3_2,shared_array3_2);

kernelfunc_2(Q, shared_array3_2, shared_array3_2);

----------------------v2--------------------------------------------

The V1 can produce the correct result. Obviously, there is no relationship between the two functions.

However, the V2 REALLY confuses me a lot, as you can see in the kernelfunc_1, I didn't use the shared_array3_2, why the v2 will get stuck like this: Screenshot 2021-08-17 121435.png

my programming environment :

VS2017

one API DPC++ Compiler (2021.3.0)

windows 10.

can anybody tell me why?

}

Re:An interesting problem on DPCPP

ShanmukhS_Intel — Tue, 17 Aug 2021 11:26:38 GMT

Hi,

Thanks for reaching out to us.

We have tried executing your code in "Intel oneAPI command prompt for Intel 64 for Visual Studio 2019" and in Visual Studio 2019 (release mode).

We see that the code compiles fine with smooth exit under our system environment.

Our system environment :

VS2019-16.9.0

one API DPC++ Compiler (2021.3.0)

windows 10.

We'd like to request you to provide the Visual Studio version being used.

Best Regards,

Shanmukh.SS

Re: Re:An interesting problem on DPCPP

shenuo — Wed, 18 Aug 2021 01:27:43 GMT

Glad to see your reply!

You remind me that this problem may lay in the hardware environment. So I move my code to another PC(desktop), under the same software environment(VS2017(10.0.17763.0), one API DPC++ Compiler (2021.3.0), windows 10.), it works fine. I can tell you my laptop is core i7-10610U CPU with Graphics UHD620(driver version 27.20.100.9320)（my desktop is i7-8700 &UHD 630(30.0.100.9684) )

Re: An interesting problem on DPCPP

ShanmukhS_Intel — Wed, 18 Aug 2021 11:18:59 GMT

Hi,

Thanks for reaching out to us.

>>You remind me that this problem may lay in the hardware environment.

Below is the link for oneAPI Base Toolkit System Requirements.

https://software.intel.com/content/www/us/en/develop/articles/intel-oneapi-base-toolkit-system-requirements.html

In addition, Below is the link for list of versions of Visual Studio supported for latest oneAPI 2021.3

https://software.intel.com/content/www/us/en/develop/articles/intel-compilers-compatibility-with-microsoft-visual-studio-and-xcode.html

Best Regards,

Shanmukh.SS

Re:An interesting problem on DPCPP

ShanmukhS_Intel — Wed, 25 Aug 2021 05:42:49 GMT

Hi,

Reminder:

Has the information provided helped? Could you please confirm whether we could close this thread from our end.

Best Regards,

Shanmukh.SS

Re: Re:An interesting problem on DPCPP

shenuo — Thu, 26 Aug 2021 00:43:48 GMT

Yes , It really helps, and thank you, but I don't know how to mark it as acceptable.

Re:An interesting problem on DPCPP

ShanmukhS_Intel — Thu, 26 Aug 2021 10:26:10 GMT

Hi,

Thank you for the confirmation. To mark it as accepted, You can find "Accept as solution" option under our reply.

Best Regards,

Shanmukh.SS

Re:An interesting problem on DPCPP

ShanmukhS_Intel — Thu, 02 Sep 2021 06:28:24 GMT

Hi,

Thanks for the confirmation. If you need any additional information, please submit a new question as this thread will no longer be monitored.

Best Regards,

Shanmukh.SS