Unable to build for MIPS

[bubiche]

Hi,

Hi, I have followed the instructions to build OpenBLAS for Android and made a GNU toolchain for Android with Fortran support following the guide here: https://github.com/buffer51/android-gfortran

When I use make TARGET=P5600 HOSTCC=gcc CC=mipsel-linux-android-gcc FC=mipsel-linux-android-gfortran, I get the error: http://paste.ubuntu.com/16884117/

Can you help me please. Thank you very much!

[brada4]

You need to add -march=mips32r5 to permit assembly instructions that your cpu has

[bubiche]

Thanks a lot! It works perfectly now

[brada4]

Let @xianyi close just in case this can/should be integrated in build system

[martin-frbg]

Build instructions for P5600 (and other MIPS targets) as provided by ksraste in #949 have been included in the faq since then, so closing now

[James6xie]

I build 0.3.3 on mips64el there had something wrong:

1.msaflags unrecognized:
rpmbuild/BUILD/openblas-0.3.3/Rblas'
gcc: error: unrecognized command line option '-msched-weight
2. test failed
gcc -O3 -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -Wp,-D_GLIBCXX_ASSERTIONS -fexceptions -fstack-protector-strong -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -mabi=64 -march=loongson3a -fno-delete-null-pointer-checks -Wa,-mno-fix-loongson3-loads -fPIC -DUTEST_CHECK -DSANITY_CHECK -DREFNAME=test_potrsf_ -DMAX_STACK_ALLOC=2048 -Wall -mabi=64 -march=loongson3a -DF_INTERFACE_GFORT -fPIC -DNO_WARMUP -DMAX_CPU_NUMBER=128 -DMAX_PARALLEL_NUMBER=1 -DUSE_SIMPLE_THREADED_LEVEL3 -DUSE_TLS -DASMNAME=test_potrs -DASMFNAME=test_potrs_ -DNAME=test_potrs_ -DCNAME=test_potrs -DCHAR_NAME="test_potrs_" -DCHAR_CNAME="test_potrs" -DNO_AFFINITY -I.. -c -o test_potrs.o test_potrs.c
gcc -O3 -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -Wp,-D_GLIBCXX_ASSERTIONS -fexceptions -fstack-protector-strong -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -mabi=64 -march=loongson3a -fno-delete-null-pointer-checks -Wa,-mno-fix-loongson3-loads -fPIC -DUTEST_CHECK -DSANITY_CHECK -DREFNAME=test_forkf_ -DMAX_STACK_ALLOC=2048 -Wall -mabi=64 -march=loongson3a -DF_INTERFACE_GFORT -fPIC -DNO_WARMUP -DMAX_CPU_NUMBER=128 -DMAX_PARALLEL_NUMBER=1 -DUSE_SIMPLE_THREADED_LEVEL3 -DUSE_TLS -DASMNAME=test_fork -DASMFNAME=test_fork_ -DNAME=test_fork_ -DCNAME=test_fork -DCHAR_NAME="test_fork_" -DCHAR_CNAME="test_fork" -DNO_AFFINITY -I.. -c -o test_fork.o test_fork.c
gcc -O3 -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -Wp,-D_GLIBCXX_ASSERTIONS -fexceptions -fstack-protector-strong -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -mabi=64 -march=loongson3a -fno-delete-null-pointer-checks -Wa,-mno-fix-loongson3-loads -fPIC -DUTEST_CHECK -DSANITY_CHECK -DREFNAME=f_ -DMAX_STACK_ALLOC=2048 -Wall -mabi=64 -march=loongson3a -DF_INTERFACE_GFORT -fPIC -DNO_WARMUP -DMAX_CPU_NUMBER=128 -DMAX_PARALLEL_NUMBER=1 -DUSE_SIMPLE_THREADED_LEVEL3 -DUSE_TLS -DASMNAME= -DASMFNAME=_ -DNAME=_ -DCNAME= -DCHAR_NAME="_" -DCHAR_CNAME="" -DNO_AFFINITY -I.. -Wl,-z,relro -Wl,-z,now -specs=/usr/lib/rpm/redhat/redhat-hardened-ld -o openblas_utest utest_main.o test_amax.o test_rotmg.o test_axpy.o test_dotu.o test_dsdot.o test_swap.o test_rot.o test_potrs.o test_fork.o ../libRblas_loongson3a-r0.3.3.a -lm -lgfortran -lm -lgfortran -L/usr/lib/gcc/mips64el-redhat-linux/8 -L/usr/lib/gcc/mips64el-redhat-linux/8/../../../../lib64 -L/lib64/../lib64 -L/usr/lib64/../lib64 -L/usr/lib/gcc/mips64el-redhat-linux/8/../../.. -L/lib64 -L/usr/lib64 -lgfortran -lm -lm -lc
./openblas_utest
TEST 1/23 amax:samax [OK]
TEST 2/23 drotmg:drotmg_D1_big_D2_big_flag_zero [OK]
TEST 3/23 drotmg:rotmg_D1eqD2_X1eqX2 [OK]
TEST 4/23 drotmg:rotmg_issue1452 [OK]
TEST 5/23 drotmg:rotmg [OK]
TEST 6/23 axpy:caxpy_inc_0 [OK]
TEST 7/23 axpy:saxpy_inc_0 [FAIL]
ERR: test_axpy.c:90 expected 4.000e+00, got 2.250e+00 (diff 1.750e+00, tol 1.000e-13)
TEST 8/23 axpy:zaxpy_inc_0 [OK]
TEST 9/23 axpy:daxpy_inc_0 [FAIL]
ERR: test_axpy.c:52 expected 4.000e+00, got 2.250e+00 (diff 1.750e+00, tol 1.000e-13)
TEST 10/23 zdotu:zdotu_offset_1 [OK]
TEST 11/23 zdotu:zdotu_n_1 [OK]
TEST 12/23 dsdot:dsdot_n_1 [OK]
TEST 13/23 swap:cswap_inc_0 [OK]
TEST 14/23 swap:sswap_inc_0 [OK]
TEST 15/23 swap:zswap_inc_0 [OK]
TEST 16/23 swap:dswap_inc_0 [OK]
TEST 17/23 rot:csrot_inc_0 [OK]
TEST 18/23 rot:srot_inc_0 [OK]
TEST 19/23 rot:zdrot_inc_0 [OK]
TEST 20/23 rot:drot_inc_0 [OK]
TEST 21/23 potrf:smoketest_trivial [OK]
TEST 22/23 potrf:bug_695 [OK]
TEST 23/23 fork:safety [OK]
RESULTS: 23 tests (21 ok, 2 failed, 0 skipped) ran in 1660 ms
make[1]: *** [Makefile:33: run_test] Error 2

[martin-frbg]

Can you try with current 0.3.7 please ? I believe I fixed the axpy failure some time ago. (And I think you need a toolchain with gcc 4.9 or newer for the -msched-weight option to be recognized ).
Is that a Loongson3A cpu ?

[James6xie]

Can you try with current 0.3.7 please ? I believe I fixed the axpy failure some time ago. (And I think you need a toolchain with gcc 4.9 or newer for the -msched-weight option to be recognized ).
Is that a Loongson3A cpu ?

thanks a lot..
yes 0.3.7 can PASS AXPY TEST;
yes I build on Loongson3A cpu;

there is failed test on cblas_cherk :
cblas_cherk PASSED THE TESTS OF ERROR-EXITS

******* FATAL ERROR - COMPUTED RESULT IS LESS THAN HALF ACCURATE *******
EXPECTED RESULT COMPUTED RESULT
1 ( 0.456543 , 0.00000 ) ( 0.456543 , -0.100000E+11)
2 ( 0.186813 , -0.929071E-01) ( 0.186813 , -0.929071E-01)
3 ( -0.382617 , 0.498502 ) ( -0.382617 , 0.498502 )
4 ( -0.252747 , 0.429570E-01) ( -0.252747 , 0.429570E-01)
5 ( 0.106893 , -0.492507 ) ( 0.106893 , -0.492507 )
THESE ARE THE RESULTS FOR COLUMN 1
******* cblas_cherk FAILED ON CALL NUMBER:
911: cblas_cherk ( CblasColMajor, CblasLower, CblasNoTrans,
5, 1, 0.0, A, 6, 1.0, C, 6).
******* cblas_cherk FAILED ON CALL NUMBER:
217: cblas_cherk ( CblasRowMajor, CblasUpper, CblasNoTrans,
1, 0, 0.0, A, 2, 0.0, C, 2).

******* FATAL ERROR - TESTS ABANDONED *******

[brada4]

I think test is faulty, it is just one bit change in first i

[martin-frbg]

@brada4 test is the same on all platforms, and every bit counts (could be an overflow or register overwrite)
@zhongtianxie I expect you see a corresponding failure in CHERK as well ? Any other "LESS THAN HALF ACCURATE" errors ?

[martin-frbg]

Hmmm. I have to admit now that the error does look suspiciously similar to #46 - @zhongtianxie could you please check if adding

RALS=RONE
RBETS=RONE

after line 1505 in ctest/c_cblat3.f (as was done in test/cblat3.f years ago) makes the error go away ? (This would probably be needed in the same location in c_zblat3.f as well to avoid false errors for cblas_zherk)