Why “for (FfxInt32 row = 0; row < 3; row++) for (FfxInt32 col = 0; col < 3; col++)”

I am reading the source code of FSR 2.0.
I found that in the function “ComputeUpsampledColorAndWeight” in "ffx_fsr2_upsample.h", there are 16 samples to be sampled, 
but the real code is "for (FfxInt32 row = 0; row < 3; row++)  for (FfxInt32 col = 0; col < 3; col++) ", which if I am right, there will be only 9 samples to be sampled.
Did I misunderstand anything?
