Bug Report for nppiNV32ToBGR_8u_P2C4R_Ctx and nppiNV21ToRGB_8u_P2C4R_Ctx

I’ve recently encountered a bug in CUDA 10.2 whereby any use of either of the two routines mentioned in the title (ie nppiNV32ToBGR_8u_P2C4R_Ctx and nppiNV21ToRGB_8u_P2C4R_Ctx), the resulting output has a corrupt blue channel.

I’m reasonably certain of this because when I first convert to NV12 and then use the equivalent (ie, nppiNV12ToBGR_8u_P2C4R_Ctx) then I get the correct output.

This is under Linux, if that matters.

Do you mind filing a bug?