Which invoke call are you referring to? There are 8 in total in the example you posted.

     cinvoke printf, _values, float qword [test_floats],\
                              float qword [test_floats + 8],\
                              float qword [test_floats + 16],\
                              float qword [test_floats + 24],\
                              float qword [test_floats + 32],\
                              float qword [test_floats + 40],\
                              float qword [test_floats + 48],\
                              float qword [test_floats + 56]    

If parameters are passed via varargs (for example, ellipsis arguments), then essentially the normal parameter passing applies including spilling the fifth and subsequent arguments. It is again the callee's responsibility to dump arguments that have their address taken. For floating-point values only, both the integer and the floating-point register will contain the float value in case the callee expects the value in the integer registers.

MS has this to say:

http://msdn.microsoft.com/en-us/library/dd2wa36c.aspx wrote:

If parameters are passed via varargs (for example, ellipsis arguments), then essentially the normal parameter passing applies including spilling the fifth and subsequent arguments. It is again the callee's responsibility to dump arguments that have their address taken. For floating-point values only, both the integer and the floating-point register will contain the float value in case the callee expects the value in the integer registers.

So actually the printf function is using the integer registers and ignores the xmm registers. So for the cinvoke fasm should be placing the data in both the xmm and integer registers.

Of course as an interim fix you can just remove the 'float' keyword.

For my code the only time I use cinvoke is with vararg calls so "fixing" cinvoke would be a suitable solution. But perhaps others use cinvoke also for other calls?