5 1.1 Q Can I find useful paper about GotoBLAS2?
7 A You may check following URL.
9 http://www.cs.utexas.edu/users/flame/Publications/index.htm
11 11. Kazushige Goto and Robert A. van de Geijn, " Anatomy of
12 High-Performance Matrix Multiplication," ACM Transactions on
13 Mathematical Software, accepted.
15 15. Kazushige Goto and Robert van de Geijn, "High-Performance
16 Implementation of the Level-3 BLAS." ACM Transactions on
17 Mathematical Software, submitted.
20 1.2 Q Does GotoBLAS2 work with Hyperthread (SMT)?
22 A Yes, it will work. GotoBLAS2 detects Hyperthread and
23 avoid scheduling on the same core.
26 1.3 Q When I type "make", following error occured. What's wrong?
29 "./Makefile.rule", line 58: Missing dependency operator
30 "./Makefile.rule", line 61: Need an operator
33 A This error occurs because you didn't use GNU make. Some binary
34 packages install GNU make as "gmake" and it's worth to try.
37 1.4 Q Function "xxx" is slow. Why?
39 A Generally GotoBLAS2 has many well optimized functions, but it's
40 far and far from perfect. Especially Level 1/2 function
41 performance depends on how you call BLAS. You should understand
42 what happends between your function and GotoBLAS2 by using profile
43 enabled version or hardware performance counter. Again, please
44 don't regard GotoBLAS2 as a black box.
47 1.5 Q I have a commercial C compiler and want to compile GotoBLAS2 with
50 A All function that affects performance is written in assembler
51 and C code is just used for wrapper of assembler functions or
52 complicated functions. Also I use many inline assembler functions,
53 unfortunately most of commercial compiler can't handle inline
54 assembler. Therefore you should use gcc.
57 1.6 Q I use OpenMP compiler. How can I use GotoBLAS2 with it?
59 A Please understand that OpenMP is a compromised method to use
60 thread. If you want to use OpenMP based code with GotoBLAS2, you
61 should enable "USE_OPENMP=1" in Makefile.rule.
64 1.7 Q Could you tell me how to use profiled library?
66 A You need to build and link your application with -pg
67 option. After executing your application, "gmon.out" is
68 generated in your current directory.
70 $shell> gprof <your application name> gmon.out
72 Each sample counts as 0.01 seconds.
73 % cumulative self self total
74 time seconds seconds calls Ks/call Ks/call name
75 89.86 975.02 975.02 79317 0.00 0.00 .dgemm_kernel
76 4.19 1020.47 45.45 40 0.00 0.00 .dlaswp00N
77 2.28 1045.16 24.69 2539 0.00 0.00 .dtrsm_kernel_LT
78 1.19 1058.03 12.87 79317 0.00 0.00 .dgemm_otcopy
79 1.05 1069.40 11.37 4999 0.00 0.00 .dgemm_oncopy
82 I think profiled BLAS library is really useful for your
83 research. Please find bottleneck of your application and
86 1.8 Q Is number of thread limited?
88 A Basically, there is no limitation about number of threads. You
89 can specify number of threads as many as you want, but larger
90 number of threads will consume extra resource. I recommend you to
91 specify minimum number of threads.
93 1.9 Q I have segfaults when I compile with USE_OPENMP=1. What's wrong?
95 A This may be related to a bug in the Linux kernel 2.6.32. Try applying
96 the patch segaults.patch using
98 patch < segfaults.patch
100 and see if the crashes persist. Note that this patch will lead to many
103 2. Architecture Specific issue or Implementation
105 2.1 Q GotoBLAS2 seems to support any combination with OS and
106 architecture. Is it possible?
108 A Combination is limited by current OS and architecture. For
109 examble, the combination OSX with SPARC is impossible. But it
110 will be possible with slight modification if these combination
111 appears in front of us.
114 2.2 Q I have POWER architecture systems. Do I need extra work?
116 A Although POWER architecture defined special instruction
117 like CPUID to detect correct architecture, it's privileged
118 and can't be accessed by user process. So you have to set
119 the architecture that you have manually in getarch.c.
122 2.3 Q I can't create DLL on Cygwin (Error 53). What's wrong?
124 A You have to make sure if lib.exe and mspdb80.dll are in Microsoft
125 Studio PATH. The easiest way is to use 'which' command.
127 $shell> which lib.exe
128 /cygdrive/c/Program Files/Microsoft Visual Studio/VC98/bin/lib.exe