initial commit

Signed-off-by: Oliver Upton <oliver.upton@linux.dev>

8 files changed, 235 insertions, 0 deletions

diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..d08ca4b
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,2 @@
+/memcpy
+/*.o
diff --git a/Makefile b/Makefile
new file mode 100644
index 0000000..e9557a3
--- /dev/null
+++ b/Makefile
@@ -0,0 +1,20 @@
+CC	:= aarch64-linux-gnu-gcc
+CFLAGS	+= -O3 -Wall -Werror
+
+OBJS	= main.o
+OBJS	+= memcpy_ldp_str.o
+OBJS	+= memcpy_ldp_stp.o
+OBJS	+= memcpy_ldr_stp.o
+
+memcpy: $(OBJS)
+	$(CC) $(CFLAGS) $(OBJS) -o $@
+
+%.o: %.S
+	$(CC) $(CFLAGS) -c $<
+
+%.o: %.c
+	$(CC) $(CFLAGS) -c $<
+
+clean:
+	rm -rf $(OBJS)
+	rm -rf memcpy
diff --git a/copy_template.h b/copy_template.h
new file mode 100644
index 0000000..3ed0124
--- /dev/null
+++ b/copy_template.h
@@ -0,0 +1,35 @@
+dst	.req	x0
+src	.req	x1
+count	.req	x2
+
+A_l	.req	x3
+A_h	.req	x4
+B_l	.req	x5
+B_h	.req	x6
+C_l	.req	x7
+C_h	.req	x8
+D_l	.req	x9
+D_h	.req	x10
+
+	ldp1	A_l, A_h, src, #16
+	ldp1	B_l, B_h, src, #16
+	ldp1	C_l, C_h, src, #16
+	ldp1	D_l, D_h, src, #16
+	sub	count, count, #64
+
+1:
+	stp1	A_l, A_h, dst, #16
+	ldp1	A_l, A_h, src, #16
+	stp1	B_l, B_h, dst, #16
+	ldp1	B_l, B_h, src, #16
+	stp1	C_l, C_h, dst, #16
+	ldp1	C_l, C_h, src, #16
+	stp1	D_l, D_h, dst, #16
+	ldp1	D_l, D_h, src, #16
+	subs	count, count, #64
+	b.ne	1b
+
+	stp1	A_l, A_h, dst, #16
+	stp1	B_l, B_h, dst, #16
+	stp1	C_l, C_h, dst, #16
+	stp1	D_l, D_h, dst, #16
diff --git a/main.c b/main.c
new file mode 100644
index 0000000..21452ea
--- /dev/null
+++ b/main.c
@@ -0,0 +1,128 @@
+#include <assert.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/mman.h>
+#include <unistd.h>
+
+#include "memcpy.h"
+
+#define KB	1024
+#define MB	(1024 * KB)
+#define GB	(1024 * MB)
+
+enum mode {
+	LDP_STR		= 0,
+	LDR_STP,
+	LDP_STP,
+
+	NR_MODES
+};
+
+static size_t test_size = 2UL * GB;
+static size_t test_iterations = 100;
+static enum mode test_mode;
+
+#define PR_MODE(mode)				\
+	printf("    %d: "#mode"\n", mode);
+
+static void pr_modes(void)
+{
+	PR_MODE(LDP_STR);
+	PR_MODE(LDR_STP);
+	PR_MODE(LDP_STP);
+}
+
+static void pr_help(const char *progname)
+{
+	printf("%s [OPTIONS]\n", progname);
+	printf("  -s SIZE		(default: %lu)\n", test_size);
+	printf("  -i ITERATIONS		(default: %lu)\n", test_iterations);
+	printf("  -m MODE		(default: %d)\n", test_mode);
+	pr_modes();
+	printf("  -h prints this message\n");
+}
+
+static void *setup_test_buffer(void)
+{
+	void *buf = mmap(0, test_size, PROT_READ|PROT_WRITE,
+			 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 0, 0);
+
+	assert(buf != MAP_FAILED);
+	memset(buf, 0xf, test_size);
+	return buf;
+}
+
+static void destroy_test_buffer(void *buf)
+{
+	munmap(buf, test_size);
+}
+
+static void do_memcpy(void *dst, const void *src)
+{
+	switch (test_mode) {
+	case LDP_STR:
+		memcpy_ldp_str(dst, src, test_size);
+		break;
+	case LDR_STP:
+		memcpy_ldr_stp(dst, src, test_size);
+		break;
+	case LDP_STP:
+		memcpy_ldp_stp(dst, src, test_size);
+		break;
+	default:
+		assert(0);
+	}
+}
+
+static void run_test(void)
+{
+	void *dst = setup_test_buffer();
+	void *src = setup_test_buffer();
+	size_t i;
+
+	printf("Iters:	%lu\n", test_iterations);
+	printf("Mode:	%d\n", test_mode);
+	printf("Source:	[%p, %p)\n", src, src + test_size);
+	printf("Dest:	[%p, %p)\n", dst, dst + test_size);
+
+	for (i = 0; i < test_iterations; i++)
+		do_memcpy(dst, src);
+
+	destroy_test_buffer(dst);
+	destroy_test_buffer(src);
+}
+
+int main(int argc, char **argv)
+{
+	int c;
+
+	while ((c = getopt(argc, argv, "hs:i:m:")) != -1) {
+		switch (c) {
+		case 's':
+			test_size = strtoul(optarg, NULL, 0);
+			if ((test_size % 64 != 0) || (test_size < 128)) {
+				pr_help(argv[0]);
+				return 1;
+			}
+			break;
+		case 'i':
+			test_iterations = strtoul(optarg, NULL, 0);
+			break;
+		case 'm':
+			test_mode = strtoul(optarg, NULL, 0);
+			if (test_mode >= NR_MODES) {
+				pr_help(argv[0]);
+				return 1;
+			}
+			break;
+		case 'h':
+			pr_help(argv[0]);
+			return 0;
+		}
+	}
+
+	run_test();
+	return 0;
+}
diff --git a/memcpy.h b/memcpy.h
new file mode 100644
index 0000000..f510c8f
--- /dev/null
+++ b/memcpy.h
@@ -0,0 +1,10 @@
+#ifndef __MEMCPY_H__
+#define __MEMCPY_H__
+
+#include <stdint.h>
+
+void memcpy_ldp_str(void *dst, const void *src, size_t count);
+void memcpy_ldr_stp(void *dst, const void *src, size_t count);
+void memcpy_ldp_stp(void *dst, const void *src, size_t count);
+
+#endif  /* __MEMCPY_H__ */
diff --git a/memcpy_ldp_stp.S b/memcpy_ldp_stp.S
new file mode 100644
index 0000000..a9851f5
--- /dev/null
+++ b/memcpy_ldp_stp.S
@@ -0,0 +1,12 @@
+.macro	ldp1, Xt1, Xt2, Xn, imm
+	ldp	\Xt1, \Xt2, [\Xn], \imm
+.endm
+
+.macro	stp1, Xt1, Xt2, Xn, imm
+	stp	\Xt1, \Xt2, [\Xn], \imm
+.endm
+
+.globl memcpy_ldp_stp
+memcpy_ldp_stp:
+	#include "copy_template.h"
+	ret
diff --git a/memcpy_ldp_str.S b/memcpy_ldp_str.S
new file mode 100644
index 0000000..0b51165
--- /dev/null
+++ b/memcpy_ldp_str.S
@@ -0,0 +1,14 @@
+.macro	ldp1, Xt1, Xt2, Xn, imm
+	ldp	\Xt1, \Xt2, [\Xn], \imm
+.endm
+
+.macro	stp1, Xt1, Xt2, Xn, imm
+	str	\Xt1, [\Xn]
+	str	\Xt2, [\Xn, #8]
+	add	\Xn, \Xn, \imm
+.endm
+
+.globl memcpy_ldp_str
+memcpy_ldp_str:
+	#include "copy_template.h"
+	ret
diff --git a/memcpy_ldr_stp.S b/memcpy_ldr_stp.S
new file mode 100644
index 0000000..78aed07
--- /dev/null
+++ b/memcpy_ldr_stp.S
@@ -0,0 +1,14 @@
+.macro	ldp1, Xt1, Xt2, Xn, imm
+	ldr	\Xt1, [\Xn]
+	ldr	\Xt2, [\Xn, #8]
+	add	\Xn, \Xn, \imm
+.endm
+
+.macro	stp1, Xt1, Xt2, Xn, imm
+	stp	\Xt1, \Xt2, [\Xn], \imm
+.endm
+
+.globl memcpy_ldr_stp
+memcpy_ldr_stp:
+	#include "copy_template.h"
+	ret