Archives Discussions

[kos]

I have something you may be interested in.

I have my own (written with my friend ) abstraction layer on Stream SDK, which can reduce number of code lines to run hellocal exemple in 5 times, and I know how to provide an ability to write kernels with some brook+ looking like language, but with much more clear IL generated. So who are interested in that please write me a private message.

[kos]

Here is a code example wich mostly equivalent to hellocal example in AMD Streamcomputing SDK, note that this is useGPU 1.0 it supports all CAL feautures exept AllocateRemote, which could be easily added. useGPU 2.0 are ready to use and provides only 2 classes to work with GPU - mem and kernel, this version have no multigpu support now. I'll post some exemples soon.

#include "useGPU.h"
#include "cal.h"
#include "calcl.h"
#include
#include

using namespace std;

std::string programIL =
 "il_ps_2_0\n"
 "dcl_input_interp(linear) v0.xy\n"
 "dcl_output_generic o0\n"
 "dcl_resource_id(0)_type(2d,unnorm)_fmtx(float)_fmty(float)_fmtz(float)_fmtw(float)\n"
 "sample_resource(0)_sampler(0) r0, v0.xy\n"
 "dcl_literal l0, 0x3F000000, 0x3F000000, 0x3F000000, 0x3F000000\n"
 "sub r2.x, r0.x, v0.x\n"
 "mov o0.x, r2.x\n"
 "ret_dyn\n"
 "end\n";

void main()
{
 useGPU g;
 float idata[256][256];
 for (int i = 0; i < 256; ++i)
  {
  for (int j = 0; j < 256; j++)
  {
  idata = 256;
  }
  }
 g.Allocate(0, idata , std::make_pair(256,256), CAL_FORMAT_FLOAT_1);
 float* odata = NULL;
 g.Allocate(0, odata , std::make_pair(256,256), CAL_FORMAT_FLOAT_1);
 g.CreateImage(programIL,0);
 vector < pair<int,string> > params;
 params.push_back( make_pair(0,"i0") );
 params.push_back( make_pair(1,"o0") );
 CALevent e = g.Execute(0,0,1,params);
 while ( g.Wait(0,e) == CAL_RESULT_PENDING);
 void* ret = g.GetResult(1);
 odata = reinterpret_cast(ret);
 for(int i = 0; i < 8; i++)
 {
  for(int j = 0; j < 8; j++)printf("%.1f ", odata[i*256+j]);
  printf("\n");
 }
 getchar();
}