Fix linalg compilation, update compiler and fix rng for stock (#19)

* Fix linalg compilation error

* Replace icc with icx compiler

* Fix icx compile macro

* Fix icx compile error for constexpr

* Remove -std=c++11 compile option for linalg

* revert back -std=c++11 and change icx with icpx for linalg

* Print samples size and Python implementation

* Add arg for rng.py to choose RNG implementation

* Change args names for rng.py

* Switch to 2022.1 icx version

* Fix LDFLAGS for umath

* Replace COMMON-AVX512 with CORE-AVx512 in native impl

Author: DenisScherbakov

numpy/linalg/Makefile

@@ -2,9 +2,8 @@
 #
 # SPDX-License-Identifier: MIT
 
-CXX = icc
-CXXFLAGS = -O3 -g -xCORE-AVX2 -axCOMMON-AVX512 \
-	   -qopt-report=5 -qopt-report-phase=openmp,par,vec
+CXX = icpx
+CXXFLAGS = -O3 -g -xSSE4.2 -axCORE-AVX2,CORE-AVX512 -qopt-report
 LDFLAGS = -lmkl_rt
 
 TARGET = linalg

numpy/linalg/bench.h

@@ -15,14 +15,14 @@ using namespace std;
 #include "assert.h"
 #include "stdlib.h"
 
-#if defined(__INTEL_COMPILER)
+#if defined(__INTEL_LLVM_COMPILER)
 
 #include "mkl.h"
 
 class Random {
   private:
     enum { SEED = 77777 };
-    static double const d_zero = 0.0, d_one = 1.0;
+    static constexpr double const d_zero = 0.0, d_one = 1.0;
     VSLStreamStatePtr stream;
 
   public:

numpy/linalg/linalg.cc

@@ -18,6 +18,7 @@
 #include <getopt.h>
 #include <iostream>
 #include <map>
+#include <vector>
 
 static const struct option longopts[] = {
     {"size", required_argument, nullptr, 'n'},

numpy/random/Makefile

@@ -4,7 +4,7 @@
 
 BENCHMARKS = rng
 SOURCES = $(addsuffix .c,$(BENCHMARKS))
-CC = icc
+CC = icx
 CLANG_FORMAT = clang-format
 CFLAGS += -m64 -fPIC -fomit-frame-pointer -xSSE4.2 -axCORE-AVX2,CORE-AVX512 \
 	  -O3 -fp-model fast=2 -fimf-precision=high -prec-sqrt -prec-div \

numpy/random/rng.c

@@ -187,7 +187,7 @@ int main(void) {
                     if (times[i] < min_time)
                         min_time = times[i];
 
-                printf("Native-C,%s,%s,%.5f\n", brng_names[brng_idx],
+                printf("Native-C,%d,%s,%s,%.5f\n", sz,  brng_names[brng_idx],
                        dist_names[fn_idx], min_time);
             }
         }

numpy/random/rng.py

@@ -3,13 +3,8 @@
 # SPDX-License-Identifier: MIT
 
 import numpy as np
-try:
-    import mkl_random as rnd
-    mkl = True
-except (ImportError, ModuleNotFoundError):
-    import numpy.random as rnd
-    mkl = False
 import timeit
+import sys
 
 def sample_uniform(rs, sz):
     rs.uniform(-1, 1, size=sz)
@@ -46,31 +41,56 @@ def sample_hypergeom(rs, sz):
 
 OUTER_REPS=6
 INNER_REPS=512
+SEED=123
 
 def main():
     import itertools
+    import argparse
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--prefix',  required=False, default="IntelPython",     help="Print with each result")
+    parser.add_argument('--impl',  required=False, default="mkl_random", choices=['mkl_random', 'numpy'], help='RNG implementation\n'
+                                                                 'choices:\n'
+                                                                 'mkl_random: mkl_random.RandomState to be used\n'
+                                                                 'numpy: numpy.Generator to be used')
+
+    args = parser.parse_args()
+
+    if args.impl == 'mkl_random':
+        try:
+            import mkl_random as rnd
+            mkl = True
+        except (ImportError, ModuleNotFoundError) as e:
+            print(str(e))
+            print('mkl_random is chosen for benchmark, however it is not found in current environemnt')
+            sys.exit(1)
+    else:
+        import numpy.random as rnd
+        mkl = False
+
     if mkl:
         brngs = ['WH', 'PHILOX4X32X10', 'MT2203', 'MCG59', 'MCG31', 'MT19937', 'MRG32K3A', 'SFMT19937', 'R250']
     else:
         brngs = [np.random.MT19937, np.random.Philox]
+
     samplers = {'uniform': sample_uniform, 'normal': sample_normal, 'gamma': sample_gamma, 'beta': sample_beta,
                 'randint': sample_randint, 'poisson': sample_poisson, 'hypergeom': sample_hypergeom}
     multipliers = {'uniform': 10, 'normal': 2, 'gamma': 1, 'beta': 1, 'randint': 10, 'poisson': 5, 'hypergeom': 1}
+
     for brng_name, sfn in itertools.product(brngs, samplers.keys()):
         func = samplers[sfn]
         m = multipliers[sfn]
         times_list = []
         for __ in range(OUTER_REPS):
             if mkl:
-                rs = rnd.RandomState(123, brng=brng_name)
+                rs = rnd.RandomState(SEED, brng=brng_name)
             else:
-                rs = rnd.Generator(brng_name(seed=123))
+                rs = rnd.Generator(brng_name(seed=SEED))
             t0 = timeit.default_timer()
             for __ in range(INNER_REPS):
                 func(rs, (m*100, 1000))
             t1 = timeit.default_timer()
             times_list.append(t1-t0)
-        print("IntelPython,{brng_name},{dist_name},{time:.5f}".format(brng_name=brng_name, dist_name=sfn, time=min(times_list)))
+        print(f"{args.prefix},{m*100*1000},{brng_name if mkl else brng_name().__class__.__name__},{sfn},{min(times_list):.5f}")
 
 
 if __name__ == '__main__':

numpy/umath/Makefile

@@ -2,9 +2,10 @@
 #
 # SPDX-License-Identifier: MIT
 
-CC = icc
-CFLAGS = -qopenmp -xCORE-AVX2 -axCOMMON-AVX512 -O3 \
-	 -g -lmkl_rt -Wall -pedantic
+CC = icx
+CFLAGS = -qopenmp -xSSE4.2 -axCORE-AVX2,CORE-AVX512 -O3 \
+	 -g -Wall -pedantic
+LDFLAGS += -lmkl_rt
 
 ifneq ($(CONDA_PREFIX),)
 		CFLAGS += -I$(CONDA_PREFIX)/include
@@ -36,7 +37,7 @@ compile: $(TARGET)
 
 
 $(TARGET): umath_bench.c
-	$(CC) umath_bench.c $(CPPFLAGS) $(CFLAGS) -o $(TARGET)
+	$(CC) umath_bench.c $(CPPFLAGS) $(CFLAGS) $(LDFLAGS) -o $(TARGET)
 
 umath_bench.c: umath_bench.c.src
 	$(PYTHON) -m numpy.distutils.conv_template umath_bench.c.src