From 98a99ac340518d66c81df28f514232d930610500 Mon Sep 17 00:00:00 2001
From: Camden Dixie O'Brien <cdo@wip.sh>
Date: Fri, 25 Nov 2022 00:13:19 +0000
Subject: [PATCH] Run 8 solving threads in parallel

---
 Makefile |  6 +++++-
 main.c   | 42 +++++++++++++++++++++++++++++++++++-------
 2 files changed, 40 insertions(+), 8 deletions(-)

diff --git a/Makefile b/Makefile
index a7ba07d..47b33ea 100644
--- a/Makefile
+++ b/Makefile
@@ -22,12 +22,16 @@ CFLAGS += -D_XOPEN_SOURCE=700
 CFLAGS += -march=native
 CFLAGS += -static
 
+LDFLAGS += -lpthread
+
 # For profiling
 # CFLAGS += -pg
 
 SRC = main.c sud.c solve.c
 OBJ = $(SRC:.c=.o)
 
+default: sudoku eval
+
 sudoku: $(OBJ)
 	$(CC) $(CFLAGS) $(LDFLAGS) $(OBJ) -o $@
 
@@ -55,4 +59,4 @@ run: sudoku eval puzzles solns results
 	time ./sudoku
 	./eval
 
-.PHONY: clean run
+.PHONY: default clean run
diff --git a/main.c b/main.c
index f5a047f..78e0884 100644
--- a/main.c
+++ b/main.c
@@ -21,6 +21,7 @@
 #include "sud.h"
 
 #include <fcntl.h>
+#include <pthread.h>
 #include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -31,6 +32,29 @@
 #define OFNAME "results"
 #define FSIZE (NPUZZ * NCELLS)
 
+#define NTHREADS 8U
+#define PUZZPT (NPUZZ / NTHREADS)
+#define CELLPT (PUZZPT * NCELLS)
+
+static char *ibuf;
+static char *obuf;
+
+void *threadproc(void *arg)
+{
+	unsigned start = CELLPT * (uintptr_t)arg;
+	const char *libuf = ibuf + start;
+	char *lobuf = obuf + start;
+
+	struct sudoku sud;
+	for (unsigned o = 0; o < CELLPT; o += NCELLS) {
+		load(&sud, libuf + o);
+		solve(&sud);
+		save(&sud, lobuf + o);
+	}
+
+	return NULL;
+}
+
 int main(void)
 {
 	int fd;
@@ -40,7 +64,7 @@ int main(void)
 		fputs("Failed to open puzzles file\n", stderr);
 		return EXIT_FAILURE;
 	}
-	char *ibuf = mmap(NULL, FSIZE, PROT_READ, MAP_PRIVATE, fd, 0);
+	ibuf = mmap(NULL, FSIZE, PROT_READ, MAP_PRIVATE, fd, 0);
 	if (ibuf == MAP_FAILED) {
 		fputs("Failed to mmap() puzzles file\n", stderr);
 		return EXIT_FAILURE;
@@ -52,18 +76,22 @@ int main(void)
 		fputs("Failed to open results file\n", stderr);
 		return EXIT_FAILURE;
 	}
-	char *obuf = mmap(NULL, FSIZE, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
+	obuf = mmap(NULL, FSIZE, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 	if (obuf == MAP_FAILED) {
 		fputs("Failed to mmap() results file\n", stderr);
 		return EXIT_FAILURE;
 	}
 	close(fd);
 
-	struct sudoku sud;
-	for (unsigned o = 0; o < FSIZE; o += NCELLS) {
-		load(&sud, ibuf + o);
-		solve(&sud);
-		save(&sud, obuf + o);
+	pthread_t pool[NTHREADS];
+	uintptr_t i;
+	for (i = 0; i < NTHREADS; ++i) {
+		if (pthread_create(&pool[i], NULL, threadproc, (void *)i) != 0)
+			fprintf(stderr, "Failed to create thread #%lu\n", i);
+	}
+	for (i = 0; i < NTHREADS; ++i) {
+		if (pthread_join(pool[i], NULL) != 0)
+			fprintf(stderr, "Failed to join thread #%lu\n", i);
 	}
 
 	munmap(ibuf, FSIZE);