int clPeak::runComputeDP(cl::CommandQueue &queue, cl::Program &prog, device_info_t &devInfo)
{
float timed, gflops;
cl_uint workPerWI;
cl::NDRange globalSize, localSize;
cl_double A = 1.3f;
int iters = devInfo.computeIters;
if(!isComputeDP)
return 0;
if(!devInfo.doubleSupported)
{
cout << NEWLINE TAB TAB "No double precision support! Skipped" << endl;
return 0;
}
try
{
cl::Context ctx = queue.getInfo<CL_QUEUE_CONTEXT>();
uint globalWIs = (devInfo.numCUs) * (devInfo.computeWgsPerCU) * (devInfo.maxWGSize);
uint t = MIN((globalWIs * sizeof(cl_double)), devInfo.maxAllocSize);
t = roundToPowOf2(t);
globalWIs = t / sizeof(cl_double);
cl::Buffer outputBuf = cl::Buffer(ctx, CL_MEM_WRITE_ONLY, (globalWIs * sizeof(cl_double)));
globalSize = globalWIs;
localSize = devInfo.maxWGSize;
cl::Kernel kernel_v1(prog, "compute_dp_v1");
kernel_v1.setArg(0, outputBuf), kernel_v1.setArg(1, A);
cl::Kernel kernel_v2(prog, "compute_dp_v2");
kernel_v2.setArg(0, outputBuf), kernel_v2.setArg(1, A);
cl::Kernel kernel_v4(prog, "compute_dp_v4");
kernel_v4.setArg(0, outputBuf), kernel_v4.setArg(1, A);
cl::Kernel kernel_v8(prog, "compute_dp_v8");
kernel_v8.setArg(0, outputBuf), kernel_v8.setArg(1, A);
cl::Kernel kernel_v16(prog, "compute_dp_v16");
kernel_v16.setArg(0, outputBuf), kernel_v16.setArg(1, A);
cout << NEWLINE TAB TAB "Double-precision compute (GFLOPS)" << endl;
cout << setprecision(2) << fixed;
///////////////////////////////////////////////////////////////////////////
// Vector width 1
cout << TAB TAB TAB "double : "; cout.flush();
workPerWI = 4096; // Indicates flops executed per work-item
timed = run_kernel(queue, kernel_v1, globalSize, localSize, iters);
gflops = ((float)globalWIs * workPerWI) / timed / 1e3f;
cout << gflops << endl;
///////////////////////////////////////////////////////////////////////////
// Vector width 2
cout << TAB TAB TAB "double2 : "; cout.flush();
workPerWI = 4096;
timed = run_kernel(queue, kernel_v2, globalSize, localSize, iters);
gflops = ((float)globalWIs * workPerWI) / timed / 1e3f;
cout << gflops << endl;
///////////////////////////////////////////////////////////////////////////
// Vector width 4
cout << TAB TAB TAB "double4 : "; cout.flush();
workPerWI = 4096;
timed = run_kernel(queue, kernel_v4, globalSize, localSize, iters);
gflops = ((float)globalWIs * workPerWI) / timed / 1e3f;
cout << gflops << endl;
///////////////////////////////////////////////////////////////////////////
// Vector width 8
cout << TAB TAB TAB "double8 : "; cout.flush();
workPerWI = 4096;
timed = run_kernel(queue, kernel_v8, globalSize, localSize, iters);
gflops = ((float)globalWIs * workPerWI) / timed / 1e3f;
cout << gflops << endl;
///////////////////////////////////////////////////////////////////////////
// Vector width 16
cout << TAB TAB TAB "double16 : "; cout.flush();
workPerWI = 4096;
timed = run_kernel(queue, kernel_v16, globalSize, localSize, iters);
gflops = ((float)globalWIs * workPerWI) / timed / 1e3f;
cout << gflops << endl;
///////////////////////////////////////////////////////////////////////////
}
catch(cl::Error error)
{
cerr << error.what() << "(" << error.err() << ")" << endl;
cerr << TAB TAB TAB "Tests skipped" << endl;
return -1;
}
return 0;
}
int clPeak::runComputeInteger(cl::CommandQueue &queue, cl::Program &prog, device_info_t &devInfo)
{
float timed, gflops;
cl_uint workPerWI;
cl::NDRange globalSize, localSize;
cl_int A = 4;
uint iters = devInfo.computeIters;
if(!isComputeInt)
return 0;
try
{
log->print(NEWLINE TAB TAB "Integer compute (GIOPS)" NEWLINE);
log->xmlOpenTag("integer_compute");
log->xmlAppendAttribs("unit", "gflops");
cl::Context ctx = queue.getInfo<CL_QUEUE_CONTEXT>();
uint64_t globalWIs = (devInfo.numCUs) * (devInfo.computeWgsPerCU) * (devInfo.maxWGSize);
uint64_t t = MIN((globalWIs * sizeof(cl_int)), devInfo.maxAllocSize) / sizeof(cl_int);
globalWIs = roundToMultipleOf(t, devInfo.maxWGSize);
cl::Buffer outputBuf = cl::Buffer(ctx, CL_MEM_WRITE_ONLY, (globalWIs * sizeof(cl_int)));
globalSize = globalWIs;
localSize = devInfo.maxWGSize;
cl::Kernel kernel_v1(prog, "compute_integer_v1");
kernel_v1.setArg(0, outputBuf), kernel_v1.setArg(1, A);
cl::Kernel kernel_v2(prog, "compute_integer_v2");
kernel_v2.setArg(0, outputBuf), kernel_v2.setArg(1, A);
cl::Kernel kernel_v4(prog, "compute_integer_v4");
kernel_v4.setArg(0, outputBuf), kernel_v4.setArg(1, A);
cl::Kernel kernel_v8(prog, "compute_integer_v8");
kernel_v8.setArg(0, outputBuf), kernel_v8.setArg(1, A);
cl::Kernel kernel_v16(prog, "compute_integer_v16");
kernel_v16.setArg(0, outputBuf), kernel_v16.setArg(1, A);
///////////////////////////////////////////////////////////////////////////
// Vector width 1
log->print(TAB TAB TAB "int : ");
workPerWI = 2048; // Indicates integer operations executed per work-item
timed = run_kernel(queue, kernel_v1, globalSize, localSize, iters);
gflops = (static_cast<float>(globalWIs) * static_cast<float>(workPerWI)) / timed / 1e3f;
log->print(gflops); log->print(NEWLINE);
log->xmlRecord("int", gflops);
///////////////////////////////////////////////////////////////////////////
// Vector width 2
log->print(TAB TAB TAB "int2 : ");
workPerWI = 2048;
timed = run_kernel(queue, kernel_v2, globalSize, localSize, iters);
gflops = (static_cast<float>(globalWIs) * static_cast<float>(workPerWI)) / timed / 1e3f;
log->print(gflops); log->print(NEWLINE);
log->xmlRecord("int2", gflops);
///////////////////////////////////////////////////////////////////////////
// Vector width 4
log->print(TAB TAB TAB "int4 : ");
workPerWI = 2048;
timed = run_kernel(queue, kernel_v4, globalSize, localSize, iters);
gflops = (static_cast<float>(globalWIs) * static_cast<float>(workPerWI)) / timed / 1e3f;
log->print(gflops); log->print(NEWLINE);
log->xmlRecord("int4", gflops);
///////////////////////////////////////////////////////////////////////////
// Vector width 8
log->print(TAB TAB TAB "int8 : ");
workPerWI = 2048;
timed = run_kernel(queue, kernel_v8, globalSize, localSize, iters);
gflops = (static_cast<float>(globalWIs) * static_cast<float>(workPerWI)) / timed / 1e3f;
log->print(gflops); log->print(NEWLINE);
log->xmlRecord("int8", gflops);
///////////////////////////////////////////////////////////////////////////
// Vector width 16
log->print(TAB TAB TAB "int16 : ");
workPerWI = 2048;
timed = run_kernel(queue, kernel_v16, globalSize, localSize, iters);
gflops = (static_cast<float>(globalWIs) * static_cast<float>(workPerWI)) / timed / 1e3f;
log->print(gflops); log->print(NEWLINE);
log->xmlRecord("int16", gflops);
///////////////////////////////////////////////////////////////////////////
log->xmlCloseTag(); // integer_compute
}
catch(cl::Error &error)
{
stringstream ss;
ss << error.what() << " (" << error.err() << ")" NEWLINE
<< TAB TAB TAB "Tests skipped" NEWLINE;
log->print(ss.str());
return -1;
}
return 0;
}
