public JittedKernel(JitContext ctx)
{
_cfg = ctx.Cfg.Clone();
_ptx = ctx.Ptx;
_hir = ctx.Hir.Fluent(blk => blk.FreezeForever());
var flds = ctx.Allocator.Fields;
_memcpyHostToDevice = kernel =>
{
var args = new KernelArguments();
flds.Keys.ForEach(fld =>
{
var value = fld.GetValue(kernel);
if (flds[fld] is SlotLayout)
{
args.Add(value.In());
}
else if (flds[fld] is ArrayLayout)
{
var arr = value.AssertCast <Array>();
args.Add(arr.InOut());
var rank = arr.GetType().GetArrayRank();
0.UpTo(rank - 1).ForEach(i => args.Add(arr.GetLength(i).In()));
}
else
{
throw AssertionHelper.Fail();
}
});
return(args);
};
_memcpyDeviceToHost = (result, kernel) =>
{
var idx = 0;
flds.Keys.ForEach(fld =>
{
Object value;
if (flds[fld] is SlotLayout)
{
value = result[idx];
idx += 1;
}
else if (flds[fld] is ArrayLayout)
{
value = result[idx];
idx += 3;
}
else
{
throw AssertionHelper.Fail();
}
fld.SetValue(kernel, value);
});
};
}
public JittedKernel(JitContext ctx)
{
_cfg = ctx.Cfg.Clone();
_ptx = ctx.Ptx;
_hir = ctx.Hir.Fluent(blk => blk.FreezeForever());
var flds = ctx.Allocator.Fields;
_memcpyHostToDevice = kernel =>
{
var args = new KernelArguments();
flds.Keys.ForEach(fld =>
{
var value = fld.GetValue(kernel);
if (flds[fld] is SlotLayout)
{
args.Add(value.In());
}
else if (flds[fld] is ArrayLayout)
{
var arr = value.AssertCast<Array>();
args.Add(arr.InOut());
var rank = arr.GetType().GetArrayRank();
0.UpTo(rank - 1).ForEach(i => args.Add(arr.GetLength(i).In()));
}
else
{
throw AssertionHelper.Fail();
}
});
return args;
};
_memcpyDeviceToHost = (result, kernel) =>
{
var idx = 0;
flds.Keys.ForEach(fld =>
{
Object value;
if (flds[fld] is SlotLayout)
{
value = result[idx];
idx += 1;
}
else if (flds[fld] is ArrayLayout)
{
value = result[idx];
idx += 3;
}
else
{
throw AssertionHelper.Fail();
}
fld.SetValue(kernel, value);
});
};
}
public KernelInvocation(JittedFunction function, IEnumerable<KernelArgument> args)
{
CudaDriver.Ensure();
Function = function.AssertNotNull();
Args = new KernelArguments(args);
}
