From a4cdbc3086708450ae8163a2e2c93ba9b0f631fa Mon Sep 17 00:00:00 2001
From: Alex Van Damme <atv@google.com>
Date: Wed, 27 Mar 2024 07:13:37 +0000
Subject: [PATCH] Add memcpy-kelvin.c

---
 newlib/libc/machine/riscv/Makefile.am     |  2 +-
 newlib/libc/machine/riscv/Makefile.in     | 10 ++++-
 newlib/libc/machine/riscv/memcpy-asm.S    |  2 +-
 newlib/libc/machine/riscv/memcpy-kelvin.c | 46 +++++++++++++++++++++++
 newlib/libc/machine/riscv/memcpy.c        |  2 +-
 5 files changed, 57 insertions(+), 5 deletions(-)
 create mode 100644 newlib/libc/machine/riscv/memcpy-kelvin.c

diff --git a/newlib/libc/machine/riscv/Makefile.am b/newlib/libc/machine/riscv/Makefile.am
index 017b4be2e..c18f728a1 100644
--- a/newlib/libc/machine/riscv/Makefile.am
+++ b/newlib/libc/machine/riscv/Makefile.am
@@ -8,7 +8,7 @@ AM_CCASFLAGS = $(INCLUDES)
 
 noinst_LIBRARIES = lib.a
 
-lib_a_SOURCES = memmove.S memmove-stub.c memset.S memcpy-asm.S memcpy.c strlen.c \
+lib_a_SOURCES = memmove.S memmove-stub.c memset.S memcpy-asm.S memcpy.c memcpy-kelvin.c strlen.c \
 	strcpy.c strcmp.S setjmp.S ieeefp.c ffs.c
 lib_a_CCASFLAGS=$(AM_CCASFLAGS)
 lib_a_CFLAGS=$(AM_CFLAGS)
diff --git a/newlib/libc/machine/riscv/Makefile.in b/newlib/libc/machine/riscv/Makefile.in
index e6dee8763..0eefef21d 100644
--- a/newlib/libc/machine/riscv/Makefile.in
+++ b/newlib/libc/machine/riscv/Makefile.in
@@ -71,7 +71,7 @@ lib_a_AR = $(AR) $(ARFLAGS)
 lib_a_LIBADD =
 am_lib_a_OBJECTS = lib_a-memmove.$(OBJEXT) \
 	lib_a-memmove-stub.$(OBJEXT) lib_a-memset.$(OBJEXT) \
-	lib_a-memcpy-asm.$(OBJEXT) lib_a-memcpy.$(OBJEXT) \
+	lib_a-memcpy-asm.$(OBJEXT) lib_a-memcpy.$(OBJEXT) lib_a-memcpy-kelvin.$(OBJEXT) \
 	lib_a-strlen.$(OBJEXT) lib_a-strcpy.$(OBJEXT) \
 	lib_a-strcmp.$(OBJEXT) lib_a-setjmp.$(OBJEXT) \
 	lib_a-ieeefp.$(OBJEXT) lib_a-ffs.$(OBJEXT)
@@ -200,7 +200,7 @@ AUTOMAKE_OPTIONS = cygnus
 INCLUDES = $(NEWLIB_CFLAGS) $(CROSS_CFLAGS) $(TARGET_CFLAGS)
 AM_CCASFLAGS = $(INCLUDES)
 noinst_LIBRARIES = lib.a
-lib_a_SOURCES = memmove.S memmove-stub.c memset.S memcpy-asm.S memcpy.c strlen.c \
+lib_a_SOURCES = memmove.S memmove-stub.c memset.S memcpy-asm.S memcpy.c memcpy-kelvin.c strlen.c \
 	strcpy.c strcmp.S setjmp.S ieeefp.c ffs.c
 
 lib_a_CCASFLAGS = $(AM_CCASFLAGS)
@@ -313,6 +313,12 @@ lib_a-memcpy.o: memcpy.c
 lib_a-memcpy.obj: memcpy.c
 	$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-memcpy.obj `if test -f 'memcpy.c'; then $(CYGPATH_W) 'memcpy.c'; else $(CYGPATH_W) '$(srcdir)/memcpy.c'; fi`
 
+lib_a-memcpy-kelvin.o: memcpy-kelvin.c
+	$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-memcpy-kelvin.o `test -f 'memcpy-kelvin.c' || echo '$(srcdir)/'`memcpy-kelvin.c
+
+lib_a-memcpy-kelvin.obj: memcpy-kelvin.c
+	$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-memcpy-kelvin.obj `if test -f 'memcpy-kelvin.c'; then $(CYGPATH_W) 'memcpy-kelvin.c'; else $(CYGPATH_W) '$(srcdir)/memcpy-kelvin.c'; fi`
+
 lib_a-strlen.o: strlen.c
 	$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-strlen.o `test -f 'strlen.c' || echo '$(srcdir)/'`strlen.c
 
diff --git a/newlib/libc/machine/riscv/memcpy-asm.S b/newlib/libc/machine/riscv/memcpy-asm.S
index 5571e4704..2a271b095 100644
--- a/newlib/libc/machine/riscv/memcpy-asm.S
+++ b/newlib/libc/machine/riscv/memcpy-asm.S
@@ -9,7 +9,7 @@
    http://www.opensource.org/licenses.
 */
 
-#if defined(PREFER_SIZE_OVER_SPEED) || defined(__OPTIMIZE_SIZE__)
+#if (defined(PREFER_SIZE_OVER_SPEED) || defined(__OPTIMIZE_SIZE__)) && !defined(__KELVIN__)
 .text
 .global memcpy
 .type	memcpy, @function
diff --git a/newlib/libc/machine/riscv/memcpy-kelvin.c b/newlib/libc/machine/riscv/memcpy-kelvin.c
new file mode 100644
index 000000000..6873b56c4
--- /dev/null
+++ b/newlib/libc/machine/riscv/memcpy-kelvin.c
@@ -0,0 +1,46 @@
+#if defined(__KELVIN__)
+
+#define __volatile_always__ volatile
+#define ARGS_F_A(FN, A0) FN " " #A0 "\n"
+#define ARGS_F_A_A(FN, A0, A1) FN " " #A0 ", " #A1 "\n"
+#define ARGS_F_A_A_A(FN, A0, A1, A2) FN " " #A0 ", " #A1 ", " #A2 "\n"
+#define ARGS_F_A_A_A_A(FN, A0, A1, A2, A3) \
+  FN " " #A0 ", " #A1 ", " #A2 ", " #A3 "\n"
+#define getvl_b_x_m(d, s)           __asm__ __volatile__(ARGS_F_A_A("getvl.b.x.m", %0, %1) : "=r"(d) : "r"(s))
+#define vld_b_lp_xx_m(Vd, s, t)     __asm__ __volatile_always__(ARGS_F_A_A_A("vld.b.lp.xx.m", Vd, %0, %1) : "=r"(s) : "r"(t), "0"(s) : "memory")
+#define vst_b_lp_xx_m(Vd, s, t)     __asm__ __volatile_always__(ARGS_F_A_A_A("vst.b.lp.xx.m", Vd, %0, %1) : "=r"(s) : "r"(t), "0"(s) : "memory")
+#define vld_w_x_m(Vd, s)            __asm__ __volatile_always__(ARGS_F_A_A("vld.w.x.m", Vd, %0) : : "r"(s) : "memory")
+#define vst_w_x_m(Vd, s)            __asm__ __volatile_always__(ARGS_F_A_A("vst.w.x.m", Vd, %0) : : "r"(s) : "memory")
+
+#include <stddef.h>
+#include <stdint.h>
+
+void *
+memcpy(void *__restrict dst, const void *__restrict src, size_t n)
+{
+  const uint8_t *s = (const uint8_t *)(src);
+  uint8_t *d = (uint8_t *)(dst);
+  int vl;
+  // Storage for caller's v0-v7
+  uint32_t spill[8 * 8];
+  vst_w_x_m(v0, spill);
+  vst_w_x_m(v4, spill + 32);
+  while (1) {
+    if (n <= 0) break;
+    getvl_b_x_m(vl, n);
+    n -= vl;
+    vld_b_lp_xx_m(v0, s, vl);
+    vst_b_lp_xx_m(v0, d, vl);
+
+    if (n <= 0) break;
+    getvl_b_x_m(vl, n);
+    n -= vl;
+    vld_b_lp_xx_m(v4, s, vl);
+    vst_b_lp_xx_m(v4, d, vl);
+  }
+  vld_w_x_m(v0, spill);
+  vld_w_x_m(v4, spill + 32);
+  return dst;
+}
+
+#endif  // defined(__KELVIN__)
diff --git a/newlib/libc/machine/riscv/memcpy.c b/newlib/libc/machine/riscv/memcpy.c
index 4098f3ab1..4e6c55071 100644
--- a/newlib/libc/machine/riscv/memcpy.c
+++ b/newlib/libc/machine/riscv/memcpy.c
@@ -9,7 +9,7 @@
    http://www.opensource.org/licenses.
 */
 
-#if defined(PREFER_SIZE_OVER_SPEED) || defined(__OPTIMIZE_SIZE__)
+#if defined(PREFER_SIZE_OVER_SPEED) || defined(__OPTIMIZE_SIZE__) || defined(__KELVIN__)
 //memcpy defined in memcpy-asm.S
 #else
 
-- 
2.44.0.396.g6e790dbe36-goog