Add workaround for NVIDIA HPC mishandling of the asm DOT kernels

kernel/arm64/KERNEL.CORTEXA57

@@ -70,10 +70,19 @@ DCOPYKERNEL  = copy.S
 CCOPYKERNEL  = copy.S
 ZCOPYKERNEL  = copy.S
 
+ifneq ($(C_COMPILER), PGI)
 SDOTKERNEL   = ../generic/dot.c
+else
+SDOTKERNEL   = dot.S
+endif
 DDOTKERNEL   = dot.S
+ifneq ($(C_COMPILER), PGI)
 CDOTKERNEL   = zdot.S
 ZDOTKERNEL   = zdot.S
+else
+CDOTKERNEL = ../arm/zdot.c
+ZDOTKERNEL = ../arm/zdot.c
+endif
 DSDOTKERNEL  = dot.S
 
 SNRM2KERNEL  = nrm2.S