Error from magmablas_dtrsm

[katayama]

Hi,
Thanks.
Please let me know if you could reproduce it.
Best

[pdgetrf]

I planted a printf in dtrsm so that when magma_dtrsm is called, we know it for sure.

This is what I'm seeing so far:

du:disco ~/disco-home/sandbox/bugfix> ./nan <- 11:08AM
[ [ 0.391137 , 0.37845 , 0.676826 , 0.828647 ];
[ 0.572013 , 1.29715 , 1.72296 , 3.94056 ];
[ 0.382288 , 0.33823 , 1.39019 , 2.54674 ];
[ 0.719967 , 1.69088 , 1.85096 , 1.95143 ] ]
[ [ 0.729483 , 0 , 0 , 0 ];
[ 0.551107 , 1.98647 , 0 , 0 ];
[ 0.946236 , 0.61067 , 2.91446 , 0 ];
[ 0.493657 , 0.600847 , 0.152826 , 3.33746 ] ]
in magmablas_dtrsm_tesla
[ [ 0.536184 , 0.0417602 , 0.0493978 , 0.159198 ];
[ 0.784135 , 0.43545 , 0.245352 , 0.975092 ];
[ 0.524053 , 0.0248786 , 0.301641 , 0.667271 ];
[ 0.986955 , 0.577387 , 0.193681 , 0.325904 ] ]

And this is the result after removing the #define:


du:disco ~/disco-home/sandbox/bugfix> !. <- 11:09AM
./nan
[ [ 0.391137 , 0.37845 , 0.676826 , 0.828647 ];
[ 0.572013 , 1.29715 , 1.72296 , 3.94056 ];
[ 0.382288 , 0.33823 , 1.39019 , 2.54674 ];
[ 0.719967 , 1.69088 , 1.85096 , 1.95143 ] ]
[ [ 0.729483 , 0 , 0 , 0 ];
[ 0.551107 , 1.98647 , 0 , 0 ];
[ 0.946236 , 0.61067 , 2.91446 , 0 ];
[ 0.493657 , 0.600847 , 0.152826 , 3.33746 ] ]
[ [ 0.536184 , 0.0417602 , 0.0493978 , 0.159198 ];
[ 0.784135 , 0.43545 , 0.245352 , 0.975092 ];
[ 0.524053 , 0.0248786 , 0.301641 , 0.667271 ];
[ 0.986955 , 0.577387 , 0.193681 , 0.325904 ] ]

One difference is that ourh C2050 machine is using CUDA driver 3.1. We are still looking into it.

[katayama]

Hi,
I also did the same thing. I copied other info. I am using 64bit OS and thus cula/lib64.

Best

[katayama@lb01 ~]$ uname -a
Linux lb01.kek.jp 2.6.18-194.26.1.el5 #1 SMP Tue Nov 9 12:54:20 EST 2010 x86_64 x86_64 x86_64 GNU/Linux
[katayama@lb01 ~]$ g++ -v
Using built-in specs.
Target: x86_64-unknown-linux-gnu
Configured with: ../configure --with-mpfr-include=/usr/local/include --with-mpfr-lib=/usr/local/lib
Thread model: posix
gcc version 4.4.4 (GCC)
[katayama@lb01 magma]$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2010 NVIDIA Corporation
Built on Wed_Nov__3_16:16:57_PDT_2010
Cuda compilation tools, release 3.2, V0.2.1221

#
# added printf
#

[katayama@lb01 magmablas]$ make
nvcc --compiler-options -fno-strict-aliasing -DUNIX -O3 -DADD_ -arch sm_20 -DGPUSHMEM=200 -I/usr/local/cuda/include -I../include -I../quark/include -c dtrsm_tesla.cu -o dtrsm_tesla.cu_o
ar cr ../lib/libmagmablas.a zauxiliary.cu_o zpermute.cu_o zpermute-v2.cu_o zswap.cu_o zswapblk.cu_o zlacpy.cu_o zlange.cu_o zlanhe.cu_o ztranspose.cu_o ztranspose-v2.cu_o zinplace_transpose.cu_o zclaswp.cu_o zcaxpycp.cu_o zlag2c.cu_o clag2z.cu_o zlat2c.cu_o cauxiliary.cu_o cpermute.cu_o cpermute-v2.cu_o cswap.cu_o cswapblk.cu_o clacpy.cu_o clange.cu_o clanhe.cu_o ctranspose.cu_o ctranspose-v2.cu_o cinplace_transpose.cu_o dauxiliary.cu_o dpermute.cu_o dpermute-v2.cu_o dswap.cu_o dswapblk.cu_o dlacpy.cu_o dlange.cu_o dlansy.cu_o dtranspose.cu_o dtranspose-v2.cu_o dinplace_transpose.cu_o dslaswp.cu_o dsaxpycp.cu_o dlag2s.cu_o slag2d.cu_o dlat2s.cu_o sauxiliary.cu_o spermute.cu_o spermute-v2.cu_o sswap.cu_o sswapblk.cu_o slacpy.cu_o slange.cu_o slansy.cu_o stranspose.cu_o stranspose-v2.cu_o sinplace_transpose.cu_o dgemv_MLU.cu_o sgemv_fermi.cu_o dgemv_fermi.cu_o sgemm_fermi.cu_o cgemv_fermi.cu_o zgemv_fermi.cu_o sgemm_fermi80.cu_o sgemm_fermi64.cu_o dgemm_fermi.cu_o ssymv_fermi.cu_o dsymv_fermi.cu_o zsymv_fermi.cu_o csymv_fermi.cu_o chemv_fermi.cu_o zhemv_fermi.cu_o strsm_tesla.cu_o dtrsm_tesla.cu_o ssyr2k.cu_o
ranlib ../lib/libmagmablas.a

[[katayama@lb01 magma]$ make
g++ -I/home/katayama/work/magma/magma_1.0.0-rc3/include -I/usr/local/cuda/include -O3 -c -o nan.o nan.cc
g++ -o nan nan.o -L/home/katayama/work/magma/magma_1.0.0-rc3/lib -L/usr/local/cuda/lib64 -lcuda -lmagma -lmagmablas -lcublas -lm

[katayama@lb01 magma]$ ./nan
[ [ 0.391137 , 0.37845 , 0.676826 , 0.828647 ];
[ 0.572013 , 1.29715 , 1.72296 , 3.94056 ];
[ 0.382288 , 0.33823 , 1.39019 , 2.54674 ];
[ 0.719967 , 1.69088 , 1.85096 , 1.95143 ] ]
[ [ 0.729483 , 0 , 0 , 0 ];
[ 0.551107 , 1.98647 , 0 , 0 ];
[ 0.946236 , 0.61067 , 2.91446 , 0 ];
[ 0.493657 , 0.600847 , 0.152826 , 3.33746 ] ]
magmablas_dtrsm_tesla/nk[ [ nan , nan , nan , nan ];
[ nan , nan , nan , nan ];
[ nan , nan , nan , nan ];
[ nan , nan , nan , nan ] ]

[pdgetrf]

here is a quick patch. please let me know if this helps:

du:disco ~/disco-home/sandbox/clean-svn/web/magma_1.0.0-rc3/magmablas> diff dtrsm_tesla.cu ../../../run/magmablas/dtrsm_tesla.cu
1823a1824
> if (i*2>=M) break;

This line is added at the end of the 'for (int i=BLOCK_SIZE; i<NB; i*=2)' loop

[katayama]

Yes. It does!
with the fix, both testing_dpotrf and testing_dpotrf_gpu work.
These are the ones I should test, right?
Thanks,
Nobu

[pdgetrf]

great! these should be good tests but if you see any other, feel free to let us know. thanks.