Hello,
I am trying to post an issue on the forum but this message "This message was flagged as spam and has been denied." continues to appear. Am I the only one experiencing this?
Dear all,
I've recently discovered that, by specifing "Kepler" as GPU_TARGET, MAGMA compiles itself only for GPUs with compute capability 3.5
- Code: Select all
ifeq (${GPU_TARGET}, Tesla)
NVOPTS += -DGPUSHMEM=130 -arch sm_13 \
-gencode arch=compute_13,code=compute_13 \
-gencode arch=compute_10,code=compute_10
OPTS += -DGPUSHMEM=130
LDOPTS += -DGPUSHMEM=130
else ifeq (${GPU_TARGET}, Fermi)
NVOPTS += -DGPUSHMEM=200 -arch sm_20
OPTS += -DGPUSHMEM=200
LDOPTS += -DGPUSHMEM=200
else ifeq (${GPU_TARGET}, Kepler)
NVOPTS += -DGPUSHMEM=300 -arch sm_35
OPTS += -DGPUSHMEM=300
LDOPTS += -DGPUSHMEM=300
else
$(error GPU_TARGET, currently ${GPU_TARGET}, must be one of Tesla, Fermi, or Kepler. Please edit your make.inc file)
endif
Is there a specific reason why "sm_30" is not considered? Does MAGMA work with GK104? May I eventually suggest to add to nvcc also "-gencode arch=compute_30,code=compute_30" like you did for "Tesla"?
mgates3 wrote:We recently upgraded a spam filter due to a large amount of spam coming through the forum. It looks like it is overly aggressive, and we are looking into why that is. Our apologies.
-mark
ifeq (${GPU_TARGET}, Tesla)
NVOPTS += -DGPUSHMEM=130 -arch sm_13 \
-gencode arch=compute_13,code=compute_13 \
-gencode arch=compute_10,code=compute_10
OPTS += -DGPUSHMEM=130
LDOPTS += -DGPUSHMEM=130
else ifeq (${GPU_TARGET}, Fermi)
NVOPTS += -DGPUSHMEM=200 -gencode arch=compute_20,code=\"sm_20,sm_21\"
OPTS += -DGPUSHMEM=200
LDOPTS += -DGPUSHMEM=200
else ifeq (${GPU_TARGET}, Kepler)
NVOPTS += -DGPUSHMEM=300 -gencode arch=compute_30,code=\"sm_30,sm_35\"
OPTS += -DGPUSHMEM=300
LDOPTS += -DGPUSHMEM=300
else
$(error GPU_TARGET, currently ${GPU_TARGET}, must be one of Tesla, Fermi, or Kepler. Please edit your make.inc file)
endifUsers browsing this forum: No registered users and 1 guest