From ca5eb817689340c9e11f21f75b2eede73622ad2e Mon Sep 17 00:00:00 2001
From: jsing <>
Date: Mon, 27 Jan 2025 14:02:32 +0000
Subject: Mop up RC4_INDEX.

The RC4_INDEX define switches between base pointer indexing and per-byte
pointer increment. This supposedly made a huge difference to performance
on x86 at some point, however compilers have improved somewhat since then.
There is no change (or effectively no change) in generated assembly on
a the majority of LLVM platforms and even when there is some change
(e.g. aarch64), there is no noticable performance difference.

Simplify the (still messy) macros/code and mop up RC4_INDEX.

ok tb@
---
 src/lib/libcrypto/arch/aarch64/opensslconf.h   |  7 ----
 src/lib/libcrypto/arch/alpha/opensslconf.h     |  7 ----
 src/lib/libcrypto/arch/amd64/opensslconf.h     |  7 ----
 src/lib/libcrypto/arch/arm/opensslconf.h       |  7 ----
 src/lib/libcrypto/arch/hppa/opensslconf.h      |  7 ----
 src/lib/libcrypto/arch/i386/opensslconf.h      |  7 ----
 src/lib/libcrypto/arch/m88k/opensslconf.h      |  7 ----
 src/lib/libcrypto/arch/mips64/opensslconf.h    |  7 ----
 src/lib/libcrypto/arch/powerpc/opensslconf.h   |  7 ----
 src/lib/libcrypto/arch/powerpc64/opensslconf.h |  7 ----
 src/lib/libcrypto/arch/riscv64/opensslconf.h   |  7 ----
 src/lib/libcrypto/arch/sh/opensslconf.h        |  7 ----
 src/lib/libcrypto/arch/sparc64/opensslconf.h   |  7 ----
 src/lib/libcrypto/rc4/rc4.c                    | 44 +++++++++++---------------
 14 files changed, 19 insertions(+), 116 deletions(-)

diff --git a/src/lib/libcrypto/arch/aarch64/opensslconf.h b/src/lib/libcrypto/arch/aarch64/opensslconf.h
index fed209a008..e6e294a962 100644
--- a/src/lib/libcrypto/arch/aarch64/opensslconf.h
+++ b/src/lib/libcrypto/arch/aarch64/opensslconf.h
@@ -65,13 +65,6 @@
 #undef EIGHT_BIT
 #endif
 
-#if defined(HEADER_RC4_LOCL_H) && !defined(CONFIG_HEADER_RC4_LOCL_H)
-#define CONFIG_HEADER_RC4_LOCL_H
-/* if this is defined data[i] is used instead of *data, this is a %20
- * speedup on x86 */
-#undef RC4_INDEX
-#endif
-
 #if defined(HEADER_BF_LOCL_H) && !defined(CONFIG_HEADER_BF_LOCL_H)
 #define CONFIG_HEADER_BF_LOCL_H
 #undef BF_PTR
diff --git a/src/lib/libcrypto/arch/alpha/opensslconf.h b/src/lib/libcrypto/arch/alpha/opensslconf.h
index 1dbd8f22fa..1c033369ea 100644
--- a/src/lib/libcrypto/arch/alpha/opensslconf.h
+++ b/src/lib/libcrypto/arch/alpha/opensslconf.h
@@ -63,13 +63,6 @@
 #undef THIRTY_TWO_BIT
 #endif
 
-#if defined(HEADER_RC4_LOCL_H) && !defined(CONFIG_HEADER_RC4_LOCL_H)
-#define CONFIG_HEADER_RC4_LOCL_H
-/* if this is defined data[i] is used instead of *data, this is a %20
- * speedup on x86 */
-#undef RC4_INDEX
-#endif
-
 #if defined(HEADER_BF_LOCL_H) && !defined(CONFIG_HEADER_BF_LOCL_H)
 #define CONFIG_HEADER_BF_LOCL_H
 #define BF_PTR
diff --git a/src/lib/libcrypto/arch/amd64/opensslconf.h b/src/lib/libcrypto/arch/amd64/opensslconf.h
index 7615744a6b..4187a44c86 100644
--- a/src/lib/libcrypto/arch/amd64/opensslconf.h
+++ b/src/lib/libcrypto/arch/amd64/opensslconf.h
@@ -60,13 +60,6 @@
 #undef THIRTY_TWO_BIT
 #endif
 
-#if defined(HEADER_RC4_LOCL_H) && !defined(CONFIG_HEADER_RC4_LOCL_H)
-#define CONFIG_HEADER_RC4_LOCL_H
-/* if this is defined data[i] is used instead of *data, this is a %20
- * speedup on x86 */
-#undef RC4_INDEX
-#endif
-
 #if defined(HEADER_BF_LOCL_H) && !defined(CONFIG_HEADER_BF_LOCL_H)
 #define CONFIG_HEADER_BF_LOCL_H
 #undef BF_PTR
diff --git a/src/lib/libcrypto/arch/arm/opensslconf.h b/src/lib/libcrypto/arch/arm/opensslconf.h
index 3413abb672..98dcfee3a1 100644
--- a/src/lib/libcrypto/arch/arm/opensslconf.h
+++ b/src/lib/libcrypto/arch/arm/opensslconf.h
@@ -65,13 +65,6 @@
 #undef EIGHT_BIT
 #endif
 
-#if defined(HEADER_RC4_LOCL_H) && !defined(CONFIG_HEADER_RC4_LOCL_H)
-#define CONFIG_HEADER_RC4_LOCL_H
-/* if this is defined data[i] is used instead of *data, this is a %20
- * speedup on x86 */
-#define RC4_INDEX
-#endif
-
 #if defined(HEADER_BF_LOCL_H) && !defined(CONFIG_HEADER_BF_LOCL_H)
 #define CONFIG_HEADER_BF_LOCL_H
 #undef BF_PTR
diff --git a/src/lib/libcrypto/arch/hppa/opensslconf.h b/src/lib/libcrypto/arch/hppa/opensslconf.h
index 3413abb672..98dcfee3a1 100644
--- a/src/lib/libcrypto/arch/hppa/opensslconf.h
+++ b/src/lib/libcrypto/arch/hppa/opensslconf.h
@@ -65,13 +65,6 @@
 #undef EIGHT_BIT
 #endif
 
-#if defined(HEADER_RC4_LOCL_H) && !defined(CONFIG_HEADER_RC4_LOCL_H)
-#define CONFIG_HEADER_RC4_LOCL_H
-/* if this is defined data[i] is used instead of *data, this is a %20
- * speedup on x86 */
-#define RC4_INDEX
-#endif
-
 #if defined(HEADER_BF_LOCL_H) && !defined(CONFIG_HEADER_BF_LOCL_H)
 #define CONFIG_HEADER_BF_LOCL_H
 #undef BF_PTR
diff --git a/src/lib/libcrypto/arch/i386/opensslconf.h b/src/lib/libcrypto/arch/i386/opensslconf.h
index ac1ea99cc9..1ea367102b 100644
--- a/src/lib/libcrypto/arch/i386/opensslconf.h
+++ b/src/lib/libcrypto/arch/i386/opensslconf.h
@@ -65,13 +65,6 @@
 #undef EIGHT_BIT
 #endif
 
-#if defined(HEADER_RC4_LOCL_H) && !defined(CONFIG_HEADER_RC4_LOCL_H)
-#define CONFIG_HEADER_RC4_LOCL_H
-/* if this is defined data[i] is used instead of *data, this is a %20
- * speedup on x86 */
-#define RC4_INDEX
-#endif
-
 #if defined(HEADER_BF_LOCL_H) && !defined(CONFIG_HEADER_BF_LOCL_H)
 #define CONFIG_HEADER_BF_LOCL_H
 #undef BF_PTR
diff --git a/src/lib/libcrypto/arch/m88k/opensslconf.h b/src/lib/libcrypto/arch/m88k/opensslconf.h
index 3413abb672..98dcfee3a1 100644
--- a/src/lib/libcrypto/arch/m88k/opensslconf.h
+++ b/src/lib/libcrypto/arch/m88k/opensslconf.h
@@ -65,13 +65,6 @@
 #undef EIGHT_BIT
 #endif
 
-#if defined(HEADER_RC4_LOCL_H) && !defined(CONFIG_HEADER_RC4_LOCL_H)
-#define CONFIG_HEADER_RC4_LOCL_H
-/* if this is defined data[i] is used instead of *data, this is a %20
- * speedup on x86 */
-#define RC4_INDEX
-#endif
-
 #if defined(HEADER_BF_LOCL_H) && !defined(CONFIG_HEADER_BF_LOCL_H)
 #define CONFIG_HEADER_BF_LOCL_H
 #undef BF_PTR
diff --git a/src/lib/libcrypto/arch/mips64/opensslconf.h b/src/lib/libcrypto/arch/mips64/opensslconf.h
index e5b1b51ec5..ec35dc9946 100644
--- a/src/lib/libcrypto/arch/mips64/opensslconf.h
+++ b/src/lib/libcrypto/arch/mips64/opensslconf.h
@@ -65,13 +65,6 @@
 #undef EIGHT_BIT
 #endif
 
-#if defined(HEADER_RC4_LOCL_H) && !defined(CONFIG_HEADER_RC4_LOCL_H)
-#define CONFIG_HEADER_RC4_LOCL_H
-/* if this is defined data[i] is used instead of *data, this is a %20
- * speedup on x86 */
-#undef RC4_INDEX
-#endif
-
 #if defined(HEADER_BF_LOCL_H) && !defined(CONFIG_HEADER_BF_LOCL_H)
 #define CONFIG_HEADER_BF_LOCL_H
 #define BF_PTR
diff --git a/src/lib/libcrypto/arch/powerpc/opensslconf.h b/src/lib/libcrypto/arch/powerpc/opensslconf.h
index 3413abb672..98dcfee3a1 100644
--- a/src/lib/libcrypto/arch/powerpc/opensslconf.h
+++ b/src/lib/libcrypto/arch/powerpc/opensslconf.h
@@ -65,13 +65,6 @@
 #undef EIGHT_BIT
 #endif
 
-#if defined(HEADER_RC4_LOCL_H) && !defined(CONFIG_HEADER_RC4_LOCL_H)
-#define CONFIG_HEADER_RC4_LOCL_H
-/* if this is defined data[i] is used instead of *data, this is a %20
- * speedup on x86 */
-#define RC4_INDEX
-#endif
-
 #if defined(HEADER_BF_LOCL_H) && !defined(CONFIG_HEADER_BF_LOCL_H)
 #define CONFIG_HEADER_BF_LOCL_H
 #undef BF_PTR
diff --git a/src/lib/libcrypto/arch/powerpc64/opensslconf.h b/src/lib/libcrypto/arch/powerpc64/opensslconf.h
index 7615744a6b..4187a44c86 100644
--- a/src/lib/libcrypto/arch/powerpc64/opensslconf.h
+++ b/src/lib/libcrypto/arch/powerpc64/opensslconf.h
@@ -60,13 +60,6 @@
 #undef THIRTY_TWO_BIT
 #endif
 
-#if defined(HEADER_RC4_LOCL_H) && !defined(CONFIG_HEADER_RC4_LOCL_H)
-#define CONFIG_HEADER_RC4_LOCL_H
-/* if this is defined data[i] is used instead of *data, this is a %20
- * speedup on x86 */
-#undef RC4_INDEX
-#endif
-
 #if defined(HEADER_BF_LOCL_H) && !defined(CONFIG_HEADER_BF_LOCL_H)
 #define CONFIG_HEADER_BF_LOCL_H
 #undef BF_PTR
diff --git a/src/lib/libcrypto/arch/riscv64/opensslconf.h b/src/lib/libcrypto/arch/riscv64/opensslconf.h
index fed209a008..e6e294a962 100644
--- a/src/lib/libcrypto/arch/riscv64/opensslconf.h
+++ b/src/lib/libcrypto/arch/riscv64/opensslconf.h
@@ -65,13 +65,6 @@
 #undef EIGHT_BIT
 #endif
 
-#if defined(HEADER_RC4_LOCL_H) && !defined(CONFIG_HEADER_RC4_LOCL_H)
-#define CONFIG_HEADER_RC4_LOCL_H
-/* if this is defined data[i] is used instead of *data, this is a %20
- * speedup on x86 */
-#undef RC4_INDEX
-#endif
-
 #if defined(HEADER_BF_LOCL_H) && !defined(CONFIG_HEADER_BF_LOCL_H)
 #define CONFIG_HEADER_BF_LOCL_H
 #undef BF_PTR
diff --git a/src/lib/libcrypto/arch/sh/opensslconf.h b/src/lib/libcrypto/arch/sh/opensslconf.h
index 3413abb672..98dcfee3a1 100644
--- a/src/lib/libcrypto/arch/sh/opensslconf.h
+++ b/src/lib/libcrypto/arch/sh/opensslconf.h
@@ -65,13 +65,6 @@
 #undef EIGHT_BIT
 #endif
 
-#if defined(HEADER_RC4_LOCL_H) && !defined(CONFIG_HEADER_RC4_LOCL_H)
-#define CONFIG_HEADER_RC4_LOCL_H
-/* if this is defined data[i] is used instead of *data, this is a %20
- * speedup on x86 */
-#define RC4_INDEX
-#endif
-
 #if defined(HEADER_BF_LOCL_H) && !defined(CONFIG_HEADER_BF_LOCL_H)
 #define CONFIG_HEADER_BF_LOCL_H
 #undef BF_PTR
diff --git a/src/lib/libcrypto/arch/sparc64/opensslconf.h b/src/lib/libcrypto/arch/sparc64/opensslconf.h
index e5b1b51ec5..ec35dc9946 100644
--- a/src/lib/libcrypto/arch/sparc64/opensslconf.h
+++ b/src/lib/libcrypto/arch/sparc64/opensslconf.h
@@ -65,13 +65,6 @@
 #undef EIGHT_BIT
 #endif
 
-#if defined(HEADER_RC4_LOCL_H) && !defined(CONFIG_HEADER_RC4_LOCL_H)
-#define CONFIG_HEADER_RC4_LOCL_H
-/* if this is defined data[i] is used instead of *data, this is a %20
- * speedup on x86 */
-#undef RC4_INDEX
-#endif
-
 #if defined(HEADER_BF_LOCL_H) && !defined(CONFIG_HEADER_BF_LOCL_H)
 #define CONFIG_HEADER_BF_LOCL_H
 #define BF_PTR
diff --git a/src/lib/libcrypto/rc4/rc4.c b/src/lib/libcrypto/rc4/rc4.c
index 930d0d9bb3..56ed43cba7 100644
--- a/src/lib/libcrypto/rc4/rc4.c
+++ b/src/lib/libcrypto/rc4/rc4.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: rc4.c,v 1.12 2024/08/11 13:02:39 jsing Exp $ */
+/* $OpenBSD: rc4.c,v 1.13 2025/01/27 14:02:32 jsing Exp $ */
 /* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com)
  * All rights reserved.
  *
@@ -196,7 +196,7 @@ rc4_internal(RC4_KEY *key, size_t len, const unsigned char *indata,
 #endif
 	}
 #endif
-#define LOOP(in,out) \
+#define RC4_LOOP(in,out) \
 		x=((x+1)&0xff); \
 		tx=d[x]; \
 		y=(tx+y)&0xff; \
@@ -204,27 +204,21 @@ rc4_internal(RC4_KEY *key, size_t len, const unsigned char *indata,
 		d[y]=tx; \
 		(out) = d[(tx+ty)&0xff]^ (in);
 
-#ifndef RC4_INDEX
-#define RC4_LOOP(a,b,i)	LOOP(*((a)++),*((b)++))
-#else
-#define RC4_LOOP(a,b,i)	LOOP(a[i],b[i])
-#endif
-
 	i = len >> 3;
 	if (i) {
 		for (;;) {
-			RC4_LOOP(indata, outdata, 0);
-			RC4_LOOP(indata, outdata, 1);
-			RC4_LOOP(indata, outdata, 2);
-			RC4_LOOP(indata, outdata, 3);
-			RC4_LOOP(indata, outdata, 4);
-			RC4_LOOP(indata, outdata, 5);
-			RC4_LOOP(indata, outdata, 6);
-			RC4_LOOP(indata, outdata, 7);
-#ifdef RC4_INDEX
+			RC4_LOOP(indata[0], outdata[0]);
+			RC4_LOOP(indata[1], outdata[1]);
+			RC4_LOOP(indata[2], outdata[2]);
+			RC4_LOOP(indata[3], outdata[3]);
+			RC4_LOOP(indata[4], outdata[4]);
+			RC4_LOOP(indata[5], outdata[5]);
+			RC4_LOOP(indata[6], outdata[6]);
+			RC4_LOOP(indata[7], outdata[7]);
+
 			indata += 8;
 			outdata += 8;
-#endif
+
 			if (--i == 0)
 				break;
 		}
@@ -232,25 +226,25 @@ rc4_internal(RC4_KEY *key, size_t len, const unsigned char *indata,
 	i = len&0x07;
 	if (i) {
 		for (;;) {
-			RC4_LOOP(indata, outdata, 0);
+			RC4_LOOP(indata[0], outdata[0]);
 			if (--i == 0)
 				break;
-			RC4_LOOP(indata, outdata, 1);
+			RC4_LOOP(indata[1], outdata[1]);
 			if (--i == 0)
 				break;
-			RC4_LOOP(indata, outdata, 2);
+			RC4_LOOP(indata[2], outdata[2]);
 			if (--i == 0)
 				break;
-			RC4_LOOP(indata, outdata, 3);
+			RC4_LOOP(indata[3], outdata[3]);
 			if (--i == 0)
 				break;
-			RC4_LOOP(indata, outdata, 4);
+			RC4_LOOP(indata[4], outdata[4]);
 			if (--i == 0)
 				break;
-			RC4_LOOP(indata, outdata, 5);
+			RC4_LOOP(indata[5], outdata[5]);
 			if (--i == 0)
 				break;
-			RC4_LOOP(indata, outdata, 6);
+			RC4_LOOP(indata[6], outdata[6]);
 			if (--i == 0)
 				break;
 		}
-- 
cgit v1.2.3-55-g6feb