Use 8/4/1 bytes loads/stores for copyin/copyout/kcopy

Only use multiple bytes operations on properly aligned addresses, as
I have observed a 40x penalty for unaligned 8 bytes operations compared
to equivalent 1-byte loops on this Sifive Unmatched.  The speed gain is
small but significant.

Input & ok kettenis@

diff --git a/sys/arch/riscv64/riscv64/copy.S b/sys/arch/riscv64/riscv64/copy.S
index 075159f..e7eb3f1 100644 (file)
--- a/sys/arch/riscv64/riscv64/copy.S
+++ b/sys/arch/riscv64/riscv64/copy.S
@@ -1,4 +1,4 @@
-/*     $OpenBSD: copy.S,v 1.6 2021/06/28 18:53:10 deraadt Exp $        */
+/*     $OpenBSD: copy.S,v 1.7 2021/07/23 15:31:14 jca Exp $    */
 
 /*
  * Copyright (c) 2020 Brian Bamsch <bbamsch@google.com>
@@ -49,8 +49,38 @@ ENTRY(copyin)
        SWAP_FAULT_HANDLER(a3, a4, a5)
        ENTER_USER_ACCESS(a4)
 
-// XXX optimize?
 .Lcopyio:
+.Lcopy8:
+       li      a5, 8
+       bltu    a2, a5, .Lcopy4
+
+       or      a7, a0, a1
+       andi    a7, a7, 7
+       bnez    a7, .Lcopy4
+
+1:     ld      a4, 0(a0)
+       addi    a0, a0, 8
+       sd      a4, 0(a1)
+       addi    a1, a1, 8
+       addi    a2, a2, -8
+       bgeu    a2, a5, 1b
+
+.Lcopy4:
+       li      a5, 4
+       bltu    a2, a5, .Lcopy1
+
+       andi    a7, a7, 3
+       bnez    a7, .Lcopy1
+
+1:     lw      a4, 0(a0)
+       addi    a0, a0, 4
+       sw      a4, 0(a1)
+       addi    a1, a1, 4
+       addi    a2, a2, -4
+       bgeu    a2, a5, 1b
+
+.Lcopy1:
+       beqz    a2, .Lcopy0
 1:     lb      a4, 0(a0)
        addi    a0, a0, 1
        sb      a4, 0(a1)
@@ -58,6 +88,7 @@ ENTRY(copyin)
        addi    a2, a2, -1
        bnez    a2, 1b
 
+.Lcopy0:
        EXIT_USER_ACCESS(a4)
        SET_FAULT_HANDLER(a3, a4)
 .Lcopyiodone: