3.4.7.1.7 : Produit de Hadamard NVC++ sur MUST


Commençons par quelques petites erreurs que l'on peut rencontrer sur le coin de la route noteDu moment que l'on est sur une route qui a des coins bien sûr. :

Comme nous compilons en C++17, nous ne pouvons pas utiliser un compilateur antique (voir ici). Et dans le cas, même G++7 est trop vieux noteNous sommes d'accord qu'il fourni du binaire relativement bien optimisé, mais son C++17 ne fonctione pas.. Dans mon cas, G++9 fonctionne très bien.

D'un manière générale, si vous obtenez des erreurs du genre :

"/opt/nvidia/hpc_sdk/Linux_x86_64/21.9/compilers/include-stdpar/thrust/mr/new.h", line 45: error: namespace "std" has no member "align_val_t"
          return ::operator new(bytes, std::align_val_t(alignment));
                                            ^


"/opt/nvidia/hpc_sdk/Linux_x86_64/21.9/compilers/include-stdpar/thrust/mr/new.h", line 67: error: namespace "std" has no member "align_val_t"
          ::operator delete(p, bytes, std::align_val_t(alignment));
                                           ^


C'est que votre compilateur C++ ne gère pas C++17.

Cela dit, si vous demandez à nvc++20.11-0 de compiler en C++20, il n'y arrive pas non plus. Mais la version de 2022 sera sensée le faire.


Une autre arnaque est que le mode d'exécution std::execution::par n'est pas gérer pour les GPU ayant des Compute Capabilities supérieures ou égales à 7.0. Évidemment, la P6000 est en 6.1. C'est pour cela que l'on utilise std::execution::par_unseq :

"hadamard.cpp", line 23: error: Calls to function
	"std::transform(_EP &&, _FIt1, _FIt1, _FIt2, _FIt3, _BF) [with _EP=std::execution::parallel_policy &, _FIt1=const float *, _FIt2=const float *,
	_FIt3=float *, _BF=lambda [](float, float)->float]" with execution policy std::execution::par will run sequentially when compiled for a
	compute capability less than cc70; only std::execution::par_unseq can be run in parallel on such GPUs
  	std::transform(std::execution::par, tabX, tabX + nbElement, tabY, tabRes,
  	              ^



condor_submit submit.condor
Submitting job(s).
1 job(s) submitted to cluster 9435.



1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123

cat hadamard_product_nvcpp.output
Used machine is Linux lapp-wngpu005.in2p3.fr 3.10.0-1160.42.2.el7.x86_64 #1 SMP Tue Sep 7 14:49:57 UTC 2021 x86_64 x86_64 x86_64 GNU/Linux
Remove existing directory build
-- The C compiler identification is GNU 4.8.5
-- The CXX compiler identification is PGI 21.9.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc - works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /opt/nvidia/hpc_sdk/Linux_x86_64/21.9/compilers/bin/nvc++
-- Check for working CXX compiler: /opt/nvidia/hpc_sdk/Linux_x86_64/21.9/compilers/bin/nvc++ - works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Program HadamardProductNvcpp version 0.1.0
-- SELF_TESTS_MODE = yes
-- OptionParser not found
-- Find project OptionParser in /lapp_data/cta/paubert/TestCondor/COURS/HadamardProductNvcpp/tmp_project/OptionParser
-- Program OptionParser version 1.7.9
-- StringUtils not found
-- Find project StringUtils in /lapp_data/cta/paubert/TestCondor/COURS/HadamardProductNvcpp/tmp_project/OptionParser/tmp_project/StringUtils
-- Program StringUtils version 1.7.9
-- MicroBenchmark not found
-- Find project MicroBenchmark in /lapp_data/cta/paubert/TestCondor/COURS/HadamardProductNvcpp/tmp_project/MicroBenchmark
-- Program MicroBenchmark version 1.7.9
-- Activate mode to avoid performance test refreshing : NO_PERF_REFRESH = yes
-- EXTRA_DEPENDENCIES = 
-- TensorAlloc not found
-- Find project TensorAlloc in /lapp_data/cta/paubert/TestCondor/COURS/HadamardProductNvcpp/tmp_project/TensorAlloc
-- Program TensorAlloc version 1.7.9
-- Automatic searching for architecture flags
-- Get LINUX extensions
-- Find SSSE3
-- Find SSE4
-- Find AVX
-- Find AVX2
-- Find AVX512F
-- tmp FLAG_VECTORIZED_COMPILATION = -mavx512f
-- global PHOENIX_FLAG_VECTORIZED_COMPILATION = -mavx512f
-- Automatic searching for architecture flags
-- Get LINUX extensions
-- Find SSSE3
-- Find SSE4
-- Find AVX
-- Find AVX2
-- Find AVX512F
-- CMAKE_VERSION = 3.17.5, MODE_NUNBER = '6'
-- CPU_MODEL_NAME =  Intel(R) Xeon(R) Silver 4108 CPU @ 1.80GHz
-- CPU_SIBLINGS =  16
-- NB_CORE = 8
-- CACHE_L1_DATA = 0 B
-- CACHE_L1_INSTRUCTION = 0 B
-- CACHE_L2 = 0 B
-- CACHE_L3 = 0 B
-- ENDIANESS = LittleEndian
-- DataStream not found
-- Find project DataStream in /lapp_data/cta/paubert/TestCondor/COURS/HadamardProductNvcpp/tmp_project/TensorAlloc/tmp_project/DataStream
-- Program DataStream version 1.7.9
-- TensorAlloc PHOENIX_FLAG_VECTORIZED_COMPILATION = -mavx512f
-- enable GPU mode : GPU_MODE = yes
-- Use nvc++ compiler at /opt/nvidia/hpc_sdk/Linux_x86_64/21.9/compilers/bin/nvc++
-- Configuring done
-- Generating done
-- Build files have been written to: /lapp_data/cta/paubert/TestCondor/COURS/HadamardProductNvcpp/build
Scanning dependencies of target string_utils
[  0%] Building CXX object tmp_project/OptionParser/tmp_project/StringUtils/src/CMakeFiles/string_utils.dir/PStream.cpp.o
[  0%] Building CXX object tmp_project/OptionParser/tmp_project/StringUtils/src/CMakeFiles/string_utils.dir/PString.cpp.o
[  1%] Building CXX object tmp_project/OptionParser/tmp_project/StringUtils/src/CMakeFiles/string_utils.dir/ProgressBarr.cpp.o
Scanning dependencies of target tensor_alloc
[  2%] Building CXX object tmp_project/TensorAlloc/src/CMakeFiles/tensor_alloc.dir/PStride.cpp.o
[  2%] Building CXX object tmp_project/TensorAlloc/src/CMakeFiles/tensor_alloc.dir/PTensor.cpp.o
[  2%] Building CXX object tmp_project/TensorAlloc/src/CMakeFiles/tensor_alloc.dir/alignement_type.cpp.o
[  2%] Building CXX object tmp_project/TensorAlloc/src/CMakeFiles/tensor_alloc.dir/pallocAlignedVector.cpp.o
[  4%] Building CXX object tmp_project/TensorAlloc/src/CMakeFiles/tensor_alloc.dir/template_alloc.cpp.o
[  4%] Linking CXX shared library libtensor_alloc.so
[  4%] Built target tensor_alloc
Scanning dependencies of target micro_benchmark
[  6%] Building CXX object tmp_project/MicroBenchmark/src/CMakeFiles/micro_benchmark.dir/micro_benchmark_common.cpp.o
[  6%] Building CXX object tmp_project/MicroBenchmark/src/CMakeFiles/micro_benchmark.dir/micro_benchmark_ns.cpp.o
[  6%] Building CXX object tmp_project/MicroBenchmark/src/CMakeFiles/micro_benchmark.dir/phoenix_timer.cpp.o
[  6%] Building CXX object tmp_project/MicroBenchmark/src/CMakeFiles/micro_benchmark.dir/pin_thread_to_core.cpp.o
[  9%] Linking CXX shared library libmicro_benchmark.so
[  9%] Built target micro_benchmark
Scanning dependencies of target data_stream
[  9%] Building CXX object tmp_project/TensorAlloc/tmp_project/DataStream/src/CMakeFiles/data_stream.dir/data_stream_file_simple_type.cpp.o
[  9%] Building CXX object tmp_project/TensorAlloc/tmp_project/DataStream/src/CMakeFiles/data_stream.dir/data_stream.cpp.o
[  9%] Building CXX object tmp_project/TensorAlloc/tmp_project/DataStream/src/CMakeFiles/data_stream.dir/data_stream_file.cpp.o
[ 11%] Building CXX object tmp_project/TensorAlloc/tmp_project/DataStream/src/CMakeFiles/data_stream.dir/data_stream_isSimpleType.cpp.o
[ 11%] Building CXX object tmp_project/TensorAlloc/tmp_project/DataStream/src/CMakeFiles/data_stream.dir/data_stream_message.cpp.o
[ 11%] Building CXX object tmp_project/TensorAlloc/tmp_project/DataStream/src/CMakeFiles/data_stream.dir/data_stream_size.cpp.o
[ 11%] Linking CXX shared library libdata_stream.so
[ 11%] Built target data_stream
Scanning dependencies of target perf_hadamard_gpu_stdpar_par_vectorize_O3
[ 11%] Building CXX object src/CMakeFiles/perf_hadamard_gpu_stdpar_par_vectorize_O3.dir/hadamard.cpp.o
[paubert@lappui7c HadamardProductNvcpp]$ cat hadamard_product_nvcpp.error
lscpu : option invalide -- 'B'
Try 'lscpu --help' for more information.
lscpu : option invalide -- 'B'
Try 'lscpu --help' for more information.
lscpu : option invalide -- 'B'
Try 'lscpu --help' for more information.
lscpu : option invalide -- 'B'
Try 'lscpu --help' for more information.
"/opt/nvidia/hpc_sdk/Linux_x86_64/21.9/compilers/include-stdpar/thrust/mr/new.h", line 45: error: namespace "std" has no member "align_val_t"
          return ::operator new(bytes, std::align_val_t(alignment));
                                            ^

"/opt/nvidia/hpc_sdk/Linux_x86_64/21.9/compilers/include-stdpar/thrust/mr/new.h", line 67: error: namespace "std" has no member "align_val_t"
          ::operator delete(p, bytes, std::align_val_t(alignment));
                                           ^

"hadamard.cpp", line 23: error: Calls to function "std::transform(_EP &&, _FIt1, _FIt1, _FIt2, _FIt3, _BF) [with _EP=std::execution::parallel_policy &, _FIt1=const float *, _FIt2=const float *, _FIt3=float *, _BF=lambda [](float, float)->float]" with execution policy std::execution::par will run sequentially when compiled for a compute capability less than cc70; only std::execution::par_unseq can be run in parallel on such GPUs
  	std::transform(std::execution::par, tabX, tabX + nbElement, tabY, tabRes,
  	              ^

2 errors detected in the compilation of "hadamard.cpp".
make[3]: *** [src/CMakeFiles/perf_hadamard_gpu_stdpar_par_vectorize_O3.dir/hadamard.cpp.o] Erreur 2
make[2]: *** [src/CMakeFiles/perf_hadamard_gpu_stdpar_par_vectorize_O3.dir/all] Erreur 2
make[1]: *** [tmp_project/MicroBenchmark/CMakeFiles/run_all.dir/rule] Erreur 2
make: *** [run_all] Erreur 2


Bon, là nvc++ n'est pas coopératif.

condor_submit submit.condor 
Submitting job(s).
1 job(s) submitted to cluster 9438.