multiplication of matrices in a finite ring

4 files changed, 389 insertions, 0 deletions

diff --git a/test/monniaux/xor_and_mat/Makefile b/test/monniaux/xor_and_mat/Makefile
new file mode 100644
index 00000000..627e2163
--- /dev/null
+++ b/test/monniaux/xor_and_mat/Makefile
@@ -0,0 +1,40 @@
+CFLAGS=-Wall -O3
+K1C_CC=k1-mbr-gcc
+K1C_CFLAGS=-Wall -O3 -std=c99
+K1C_CCOMP=../../../ccomp
+K1C_CCOMPFLAGS=-Wall -O3
+
+PRODUCTS=int_mat.host int_mat.gcc.k1c.out int_mat.ccomp.k1c.out int_mat.ccomp.k1c.s int_mat.gcc.k1c.s int_mat.gcc.k1c int_mat.ccomp.k1c
+
+all:	$(PRODUCTS)
+
+%.gcc.k1c.s: %.c
+	$(K1C_CC) $(K1C_CFLAGS) -S $< -o $@
+
+%.gcc.k1c.o: %.gcc.k1c.s
+	$(K1C_CC) $(K1C_CFLAGS) -c $< -o $@
+
+%.ccomp.k1c.s: %.c
+	$(K1C_CCOMP) $(K1C_CCOMPFLAGS) -S $< -o $@
+
+%.ccomp.k1c.o: %.ccomp.k1c.s
+	$(K1C_CCOMP) $(K1C_CCOMPFLAGS) -c $< -o $@
+
+int_mat.host: int_mat.c int_mat_run.c xor_and.h
+	$(CC) $(CFLAGS) int_mat.c int_mat_run.c -o $@
+
+int_mat.gcc.k1c.s int_mat.ccomp.k1c.s int_mat_run.gcc.k1c.s: xor_and.h
+
+int_mat.gcc.k1c: int_mat.gcc.k1c.o int_mat_run.gcc.k1c.o
+	$(K1C_CC) $(K1C_CFLAGS) $+ -o $@
+
+int_mat.ccomp.k1c: int_mat.ccomp.k1c.o int_mat_run.gcc.k1c.o
+	$(K1C_CCOMP) $(K1C_CCOMPFLAGS) $+ -o $@
+
+%.k1c.out: %.k1c
+	k1-cluster --cycle-based -- $< | tee $@
+
+clean:
+	$(RM) -f $(PRODUCTS) int_mat.gcc.k1c.o int_mat.ccomp.k1c.o int_mat_run.gcc.k1c.o
+
+.PHONY: clean
diff --git a/test/monniaux/xor_and_mat/int_mat.c b/test/monniaux/xor_and_mat/int_mat.c
new file mode 100644
index 00000000..0bfe16b9
--- /dev/null
+++ b/test/monniaux/xor_and_mat/int_mat.c
@@ -0,0 +1,210 @@
+#include "xor_and.h"
+
+#define ADD ^=
+#define MUL &
+
+void xor_and_mat_mul1(unsigned m, unsigned n, unsigned p,
+		     xor_and * restrict c, unsigned stride_c,
+		     const xor_and *a, unsigned stride_a,
+		     const xor_and *b, unsigned stride_b) {
+  for(unsigned i=0; i<m; i++) {
+    for(unsigned k=0; k<p; k++) {
+      c[i*stride_c+k] = 0;
+    }
+  }
+  for(unsigned i=0; i<m; i++) {
+    for(unsigned k=0; k<p; k++) {
+      for(unsigned j=0; j<n; j++) {
+	c[i*stride_c+k] ADD (a[i*stride_a+j] MUL b[j*stride_b+k]); 
+      }
+    }
+  }
+}
+
+void xor_and_mat_mul2(unsigned m, unsigned n, unsigned p,
+		     xor_and * restrict c, unsigned stride_c,
+		     const xor_and *a, unsigned stride_a,
+		     const xor_and *b, unsigned stride_b) {
+  for(unsigned i=0; i<m; i++) {
+    for(unsigned k=0; k<p; k++) {
+      xor_and total = 0;
+      for(unsigned j=0; j<n; j++) {
+	total ADD (a[i*stride_a + j] MUL b[j*stride_b + k]);
+      }
+      c[i*stride_c+k] = total;
+    }
+  }
+}
+
+void xor_and_mat_mul3(unsigned m, unsigned n, unsigned p,
+		     xor_and * restrict c, unsigned stride_c,
+		     const xor_and *a, unsigned stride_a,
+		     const xor_and *b, unsigned stride_b) {
+  for(unsigned i=0; i<m; i++) {
+    for(unsigned k=0; k<p; k++) {
+      xor_and total0 = 0, total1 = 0;
+      unsigned j;
+      for(j=0; j+1<n; j+=2) {
+	total0 ADD (a[i*stride_a + j] MUL b[j*stride_b + k]);
+	total1 ADD (a[i*stride_a + (j+1)] MUL b[(j+1)*stride_b + k]);
+      }
+      if (j < n) {
+	total0 ADD a[i*stride_a + j] MUL b[j*stride_b + k];
+      }
+      total0 ADD total1;
+      c[i*stride_c+k] = total0;
+    }
+  }
+}
+
+void xor_and_mat_mul4(unsigned m, unsigned n, unsigned p,
+		     xor_and * c, unsigned stride_c,
+		     const xor_and *a, unsigned stride_a,
+		     const xor_and *b, unsigned stride_b) {
+  const xor_and *pa_i = a;
+  xor_and * pc_i = c;
+  for(unsigned i=0; i<m; i++) {
+    for(unsigned k=0; k<p; k++) {
+      const xor_and *pb_j_k = b+k, *pa_i_j = pa_i;
+      xor_and total = 0;
+      for(unsigned j=0; j<n; j++) {
+	total ADD (*pa_i_j MUL *pb_j_k);
+	pa_i_j ++;
+	pb_j_k += stride_b;
+      }
+      pc_i[k] = total;
+    }
+    pa_i += stride_a;
+    pc_i += stride_c;
+  }
+}
+
+void xor_and_mat_mul5(unsigned m, unsigned n, unsigned p,
+		     xor_and * c, unsigned stride_c,
+		     const xor_and *a, unsigned stride_a,
+		     const xor_and *b, unsigned stride_b) {
+  const xor_and *pa_i = a;
+  xor_and * pc_i = c;
+  for(unsigned i=0; i<m; i++) {
+    for(unsigned k=0; k<p; k++) {
+      const xor_and *pb_j_k = b+k, *pa_i_j = pa_i;
+      xor_and total = 0;
+      for(unsigned j2=0, n2=n/2; j2<n2; j2++) {
+	xor_and p0 = *pa_i_j MUL *pb_j_k;
+	pa_i_j ++;
+	pb_j_k += stride_b;
+	xor_and p1 = *pa_i_j MUL *pb_j_k;
+	pa_i_j ++;
+	pb_j_k += stride_b;
+	total ADD p0;
+	total ADD p1;
+      }
+      if (n%2) {
+	total ADD *pa_i_j MUL *pb_j_k;
+      }
+      pc_i[k] = total;
+    }
+    pa_i += stride_a;
+    pc_i += stride_c;
+  }
+}
+
+#define CHUNK \
+	    total ADD (*pa_i_j MUL *pb_j_k); \
+	    pa_i_j ++; \
+	    pb_j_k += stride_b;
+
+void xor_and_mat_mul6(unsigned m, unsigned n, unsigned p,
+		     xor_and * c, unsigned stride_c,
+		     const xor_and *a, unsigned stride_a,
+		     const xor_and *b, unsigned stride_b) {
+  const xor_and *pa_i = a;
+  xor_and * pc_i = c;
+  for(unsigned i=0; i<m; i++) {
+    for(unsigned k=0; k<p; k++) {
+      const xor_and *pb_j_k = b+k, *pa_i_j = pa_i;
+      xor_and total = 0;
+      unsigned j2=0, n2=n/2;
+      if (n2 > 0) {
+	do {
+	  CHUNK
+	  CHUNK
+	  j2++;
+	} while (j2 < n2);
+      }
+      if (n%2) {
+	total ADD (*pa_i_j MUL *pb_j_k);
+      }
+      pc_i[k] = total;
+    }
+    pa_i += stride_a;
+    pc_i += stride_c;
+  }
+}
+
+#define UNROLL 4
+void xor_and_mat_mul7(unsigned m, unsigned n, unsigned p,
+		     xor_and * c, unsigned stride_c,
+		     const xor_and *a, unsigned stride_a,
+		     const xor_and *b, unsigned stride_b) {
+  const xor_and *pa_i = a;
+  xor_and * pc_i = c;
+  for(unsigned i=0; i<m; i++) {
+    for(unsigned k=0; k<p; k++) {
+      const xor_and *pb_j_k = b+k, *pa_i_j = pa_i;
+      xor_and total = 0;
+      {
+	unsigned j4=0, n4=n/UNROLL;
+	if (n4 > 0) {
+	  do {
+            CHUNK
+	    CHUNK
+	    CHUNK
+	    CHUNK
+	    j4++;
+	  } while (j4 < n4);
+	}
+      }
+      {
+	unsigned j4=0, n4=n%UNROLL;
+	if (n4 > 0) {
+	  do {
+	    CHUNK
+	    j4++;
+	  } while (j4 < n4);
+	}
+      }
+      pc_i[k] = total;
+    }
+    pa_i += stride_a;
+    pc_i += stride_c;
+  }
+}
+
+xor_and xor_and_random(void) {
+  static uint64_t next = 1325997111;
+  next = next * 1103515249 + 12345;
+  return next;
+}
+
+void xor_and_mat_random(unsigned m,
+		       unsigned n,
+		       xor_and *a, unsigned stride_a) {
+  for(unsigned i=0; i<m; i++) {
+    for(unsigned j=0; j<n; j++) {
+      a[i*stride_a + j] = xor_and_random();
+    }
+  }
+}
+
+bool xor_and_mat_equal(unsigned m,
+		      unsigned n,
+		      const xor_and *a, unsigned stride_a,
+		      const xor_and *b, unsigned stride_b) {
+  for(unsigned i=0; i<m; i++) {
+    for(unsigned j=0; j<n; j++) {
+      if (a[i*stride_a + j] != b[i*stride_b + j]) return false;
+    }
+  }
+  return true;
+}
diff --git a/test/monniaux/xor_and_mat/int_mat_run.c b/test/monniaux/xor_and_mat/int_mat_run.c
new file mode 100644
index 00000000..225d68f3
--- /dev/null
+++ b/test/monniaux/xor_and_mat/int_mat_run.c
@@ -0,0 +1,89 @@
+#include <stdbool.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <inttypes.h>
+#include "xor_and.h"
+#include "../cycles.h"
+
+int main() {
+  const unsigned m = 40, n = 21, p = 30;
+  cycle_count_config();
+  xor_and *a = malloc(sizeof(xor_and) * m * n);
+  xor_and_mat_random(m, n, a, n);
+  xor_and *b = malloc(sizeof(xor_and) * n * p);
+  xor_and_mat_random(n, p, b, p);
+  
+  xor_and *c1 = malloc(sizeof(xor_and) * m * p);
+  cycle_t c1_time = get_cycle();
+  xor_and_mat_mul1(m, n, p, c1, p, a, n, b, p);
+  c1_time = get_cycle()-c1_time;
+  
+  xor_and *c2 = malloc(sizeof(xor_and) * m * p);
+  cycle_t c2_time = get_cycle();
+  xor_and_mat_mul2(m, n, p, c2, p, a, n, b, p);
+  c2_time = get_cycle()-c2_time;
+  
+  xor_and *c3 = malloc(sizeof(xor_and) * m * p);
+  cycle_t c3_time = get_cycle();
+  xor_and_mat_mul3(m, n, p, c3, p, a, n, b, p);
+  c3_time = get_cycle()-c3_time;
+  
+  xor_and *c4 = malloc(sizeof(xor_and) * m * p);
+  cycle_t c4_time = get_cycle();
+  xor_and_mat_mul4(m, n, p, c4, p, a, n, b, p);
+  c4_time = get_cycle()-c4_time;
+
+  xor_and *c5 = malloc(sizeof(xor_and) * m * p);
+  cycle_t c5_time = get_cycle();
+  xor_and_mat_mul5(m, n, p, c5, p, a, n, b, p);
+  c5_time = get_cycle()-c5_time;
+  
+  xor_and *c6 = malloc(sizeof(xor_and) * m * p);
+  cycle_t c6_time = get_cycle();
+  xor_and_mat_mul6(m, n, p, c6, p, a, n, b, p);
+  c6_time = get_cycle()-c6_time;
+  
+  xor_and *c7 = malloc(sizeof(xor_and) * m * p);
+  cycle_t c7_time = get_cycle();
+  xor_and_mat_mul7(m, n, p, c7, p, a, n, b, p);
+  c7_time = get_cycle()-c7_time;
+  
+  printf("c1==c2: %s\n"
+	 "c1==c3: %s\n"
+	 "c1==c4: %s\n"
+	 "c1==c5: %s\n"
+	 "c1==c6: %s\n"
+	 "c1==c7: %s\n"
+	 "c1_time = %" PRIu64 "\n"
+	 "c2_time = %" PRIu64 "\n"
+	 "c3_time = %" PRIu64 "\n"
+	 "c4_time = %" PRIu64 "\n"
+	 "c5_time = %" PRIu64 "\n"
+	 "c6_time = %" PRIu64 "\n"
+	 "c7_time = %" PRIu64 "\n",
+	
+	 xor_and_mat_equal(m, n, c1, p, c2, p)?"true":"false",
+	 xor_and_mat_equal(m, n, c1, p, c3, p)?"true":"false",
+	 xor_and_mat_equal(m, n, c1, p, c4, p)?"true":"false",
+	 xor_and_mat_equal(m, n, c1, p, c5, p)?"true":"false",
+	 xor_and_mat_equal(m, n, c1, p, c6, p)?"true":"false",
+	 xor_and_mat_equal(m, n, c1, p, c7, p)?"true":"false",
+	 
+	 c1_time,
+	 c2_time,
+	 c3_time,
+	 c4_time,
+	 c5_time,
+	 c6_time,
+	 c7_time);
+  
+  free(a);
+  free(b);
+  free(c1);
+  free(c2);
+  free(c3);
+  free(c4);
+  free(c5);
+  free(c6);
+  return 0;
+}
diff --git a/test/monniaux/xor_and_mat/xor_and.h b/test/monniaux/xor_and_mat/xor_and.h
new file mode 100644
index 00000000..956ad800
--- /dev/null
+++ b/test/monniaux/xor_and_mat/xor_and.h
@@ -0,0 +1,50 @@
+#include <stdint.h>
+#include <stdbool.h>
+
+typedef uint64_t xor_and;
+
+void xor_and_mat_mul1(unsigned m, unsigned n, unsigned p,
+		     xor_and * restrict c, unsigned stride_c,
+		     const xor_and *a, unsigned stride_a,
+		     const xor_and *b, unsigned stride_b);
+
+void xor_and_mat_mul2(unsigned m, unsigned n, unsigned p,
+		     xor_and * restrict c, unsigned stride_c,
+		     const xor_and *a, unsigned stride_a,
+		     const xor_and *b, unsigned stride_b);
+
+void xor_and_mat_mul3(unsigned m, unsigned n, unsigned p,
+		     xor_and * restrict c, unsigned stride_c,
+		     const xor_and *a, unsigned stride_a,
+		     const xor_and *b, unsigned stride_b);
+
+void xor_and_mat_mul4(unsigned m, unsigned n, unsigned p,
+		     xor_and * restrict c, unsigned stride_c,
+		     const xor_and *a, unsigned stride_a,
+		     const xor_and *b, unsigned stride_b);
+
+void xor_and_mat_mul5(unsigned m, unsigned n, unsigned p,
+		     xor_and * restrict c, unsigned stride_c,
+		     const xor_and *a, unsigned stride_a,
+		     const xor_and *b, unsigned stride_b);
+
+void xor_and_mat_mul6(unsigned m, unsigned n, unsigned p,
+		     xor_and * restrict c, unsigned stride_c,
+		     const xor_and *a, unsigned stride_a,
+		     const xor_and *b, unsigned stride_b);
+
+void xor_and_mat_mul7(unsigned m, unsigned n, unsigned p,
+		     xor_and * restrict c, unsigned stride_c,
+		     const xor_and *a, unsigned stride_a,
+		     const xor_and *b, unsigned stride_b);
+
+xor_and xor_and_random(void);
+
+void xor_and_mat_random(unsigned m,
+		       unsigned n,
+		       xor_and *a, unsigned stride_a);
+
+bool xor_and_mat_equal(unsigned m,
+		      unsigned n,
+		      const xor_and *a, unsigned stride_a,
+		      const xor_and *b, unsigned stride_b);