From c2a8ebfe695fda904ce2ae2153680d0c3810f2ce Mon Sep 17 00:00:00 2001 From: Martin Kroeker Date: Tue, 12 Jan 2021 16:38:51 +0100 Subject: [PATCH] Add workaround for NVIDIA HPC mishandling of the asm DOT kernels --- kernel/arm64/KERNEL.ARMV8 | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/kernel/arm64/KERNEL.ARMV8 b/kernel/arm64/KERNEL.ARMV8 index 603e47d87..c8a53c86b 100644 --- a/kernel/arm64/KERNEL.ARMV8 +++ b/kernel/arm64/KERNEL.ARMV8 @@ -97,9 +97,18 @@ CNRM2KERNEL = znrm2.S ZNRM2KERNEL = znrm2.S DDOTKERNEL = dot.S +ifneq ($(C_COMPILER), PGI) SDOTKERNEL = ../generic/dot.c +else +SDOTKERNEL = dot.S +endif +ifneq ($(C_COMPILER), PGI) CDOTKERNEL = zdot.S ZDOTKERNEL = zdot.S +else +CDOTKERNEL = ../arm/zdot.c +ZDOTKERNEL = ../arm/zdot.c +endif DSDOTKERNEL = dot.S DGEMM_BETA = dgemm_beta.S