[feature] add regular dgemm with custom alloc

Just to be able to test various matrix placements.

benchmarks/Makefile.am

@@ -9,5 +9,6 @@ noinst_PROGRAMS = stream_add_pth_st \
 		  stream_add_omp_st \
 		  stream_add_omp_mt \
 		  dgemm_vanilla \
+		  dgemm_mkl \
 		  dgemm_prefetch \
 		  dgemm_noprefetch

benchmarks/dgemm_mkl.c

+#include <aml.h>
+#include <assert.h>
+#include <errno.h>
+#include <mkl.h>
+#include <omp.h>
+#include <pthread.h>
+#include <stdio.h>
+#include <time.h>
+#include <math.h>
+#include <stdlib.h>
+
+
+int main(int argc, char *argv[])
+{
+	AML_ARENA_JEMALLOC_DECL(arena);
+	AML_AREA_LINUX_DECL(slow);
+	AML_AREA_LINUX_DECL(fast);
+	struct bitmask *slowb, *fastb;
+	struct timespec start, stop;
+	double *a, *b, *c;
+	aml_init(&argc, &argv);
+	assert(argc == 4);
+	fastb = numa_parse_nodestring_all(argv[1]);
+	slowb = numa_parse_nodestring_all(argv[2]);
+	long int N = atol(argv[3]);
+	unsigned long memsize = sizeof(double)*N*N;
+
+	assert(!aml_arena_jemalloc_init(&arena, AML_ARENA_JEMALLOC_TYPE_REGULAR));
+	assert(!aml_area_linux_init(&slow,
+				    AML_AREA_LINUX_MANAGER_TYPE_SINGLE,
+				    AML_AREA_LINUX_MBIND_TYPE_REGULAR,
+				    AML_AREA_LINUX_MMAP_TYPE_ANONYMOUS,
+				    &arena, MPOL_BIND, slowb->maskp));
+	assert(!aml_area_linux_init(&fast,
+				    AML_AREA_LINUX_MANAGER_TYPE_SINGLE,
+				    AML_AREA_LINUX_MBIND_TYPE_REGULAR,
+				    AML_AREA_LINUX_MMAP_TYPE_ANONYMOUS,
+				    &arena, MPOL_BIND, fastb->maskp));
+	a = aml_area_malloc(&slow, memsize);
+	b = aml_area_malloc(&slow, memsize);
+	c = aml_area_malloc(&fast, memsize);
+	assert(a != NULL && b != NULL && c != NULL);
+
+	double alpha = 1.0, beta = 1.0;
+	for(unsigned long i = 0; i < N*N; i++){
+		a[i] = (double)rand();
+		b[i] = (double)rand();
+		c[i] = 0.0;
+	}
+
+	clock_gettime(CLOCK_REALTIME, &start);
+	cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans, N, N, N, alpha, a, N, b, N, beta, c, N);
+	clock_gettime(CLOCK_REALTIME, &stop);
+	long long int time = 0;
+	time =  (stop.tv_nsec - start.tv_nsec) +
+                1e9* (stop.tv_sec - start.tv_sec);
+	double flops = (2.0*N*N*N)/(time/1e9);
+	/* print the flops in GFLOPS */
+	printf("dgemm-mkl: %llu %lld %lld %f\n", N, memsize, time, flops/1e9);
+	aml_area_free(&slow, a);
+	aml_area_free(&slow, b);
+	aml_area_free(&fast, c);
+	aml_area_linux_destroy(&slow);
+	aml_area_linux_destroy(&fast);
+	aml_finalize();
+	return 0;
+}