Can I run caldgemm on HD6990s?
Hi,
I am sorry that I cannot find the caldgemm forum so I post here and hopefully its developers read this message.
I have 4 HD6990s and I really like to see how they perform in GFLOPs, so I come across this tool, http://code.compeng.uni-frankfurt.de/projects/caldgemm/wiki
I have Ubuntu 10.10 x86_64 + AMD driver 11.5 + SDK 2.4 and I followed the instructions on the wiki, however I have to add "make TARGET=NEHALEM NO_MEMPOLICY=1 -j" in order to compile GotoBLAS2 since I have two E5620 CPUs.
I compile caldgemm, its outputs look fine.
But If I run it, it prompts error and hangs idle, can anyone help? Thanks!
rolly@rolly-X8DTG-QF:~/caldgemm$ make g++ -c caldgemm.cpp -Wfloat-equal -Wpointer-arith -DATI_OS_LINUX -g3 -ffor-scope -O3 -march=core2 -ftree-vectorize -msse3 -fkeep-inline-functions -fweb -frename-registers -minline-all-stringops -funit-at-a-time -mfpmath=sse -ftracer -finline-limit=1200 -fpeel-loops -D_NO_AMD_CPU -I ../GotoBLAS2 -I /home/rolly/AMD-APP-SDK-v2.4-lnx64/include/CAL caldgemm.cpp: In function ‘void* divide_wrapper(void*)’: caldgemm.cpp:1661: warning: format ‘%lld’ expects type ‘long long int’, but argument 4 has type ‘int’ caldgemm.cpp: In member function ‘int caldgemm::RunCALDGEMM(double*, double*, double*, double, double, size_t, size_t, size_t, size_t, size_t, size_t, CBLAS_ORDER, CBLAS_TRANSPOSE, CBLAS_TRANSPOSE, int)’: caldgemm.cpp:2260: warning: format ‘%lld’ expects type ‘long long int’, but argument 3 has type ‘size_t’ caldgemm.cpp:2269: warning: format ‘%lld’ expects type ‘long long int’, but argument 3 has type ‘size_t’ caldgemm.cpp:2403: warning: format ‘%lld’ expects type ‘long long int’, but argument 4 has type ‘size_t’ caldgemm.cpp:2421: warning: format ‘%lld’ expects type ‘long long int’, but argument 4 has type ‘size_t’ caldgemm.cpp: In member function ‘int caldgemm::DGEMM_prepare(size_t, int, unsigned int)’: caldgemm.cpp:2738: warning: format ‘%d’ expects type ‘int’, but argument 4 has type ‘size_t’ caldgemm.cpp:2761: warning: format ‘%d’ expects type ‘int’, but argument 4 has type ‘size_t’ g++ -c benchmark.cpp -Wfloat-equal -Wpointer-arith -DATI_OS_LINUX -g3 -ffor-scope -O3 -march=core2 -ftree-vectorize -msse3 -fkeep-inline-functions -fweb -frename-registers -minline-all-stringops -funit-at-a-time -mfpmath=sse -ftracer -finline-limit=1200 -fpeel-loops -D_NO_AMD_CPU -I ../GotoBLAS2 -I /home/rolly/AMD-APP-SDK-v2.4-lnx64/include/CAL g++ -o dgemm_bench caldgemm.o benchmark.o -lpthread -ldl -L/usr/X11R6/lib -laticalrt -laticalcl -lgfortran ../GotoBLAS2/libgoto2.a rolly@rolly-X8DTG-QF:~/caldgemm$ ./dgemm_bench -c Use -? for help Cannot use multiple devices without multithreading Segmentation fault rolly@rolly-X8DTG-QF:~/caldgemm$ ./dgemm_bench -z Use -? for help There was an error in allocating resources and binding them to memory Error initializing CALDGEMM rolly@rolly-X8DTG-QF:~/caldgemm$ ./dgemm_bench -g Use -? for help Cannot use multiple devices without multithreading Was able to allocate 21 bbuffers Initializing Data... ...alloc AERROR locking Pages ...alloc BERROR locking Pages ...alloc CERROR locking Pages Memory allocation error allocating matrices
