Browse Source

Add workaround for NVIDIA HPC mishandling of the asm DOT kernels

tags/v0.3.14^2
Martin Kroeker GitHub 4 years ago
parent
commit
c2a8ebfe69
No known key found for this signature in database GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 9 additions and 0 deletions
  1. +9
    -0
      kernel/arm64/KERNEL.ARMV8

+ 9
- 0
kernel/arm64/KERNEL.ARMV8 View File

@@ -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


Loading…
Cancel
Save