From a5296875f101923b76d383d57a383bfa16406fbe Mon Sep 17 00:00:00 2001
From: gbeauche <>
Date: Sun, 9 Jul 2006 15:18:08 +0000
Subject: [PATCH] Optimize alignment routine for x86 & x86_64.

---
 .../src/kpx_cpu/src/cpu/jit/basic-dyngen.cpp  | 93 +++++++++++++++++++
 .../src/kpx_cpu/src/cpu/jit/basic-dyngen.hpp  |  8 --
 2 files changed, 93 insertions(+), 8 deletions(-)

diff --git a/SheepShaver/src/kpx_cpu/src/cpu/jit/basic-dyngen.cpp b/SheepShaver/src/kpx_cpu/src/cpu/jit/basic-dyngen.cpp
index 19153af2..7c8d863e 100644
--- a/SheepShaver/src/kpx_cpu/src/cpu/jit/basic-dyngen.cpp
+++ b/SheepShaver/src/kpx_cpu/src/cpu/jit/basic-dyngen.cpp
@@ -78,3 +78,96 @@ DEFINE_INVOKE(CPU_im_im, (void (*func)(dyngen_cpu_base, uint32, uint32), uint32
 DEFINE_INVOKE(CPU_A0_ret_A0, (void *(*func)(dyngen_cpu_base)), (funcptr));
 
 #undef DEFINE_INVOKE
+
+uint8 *
+basic_dyngen::gen_align(int align)
+{
+	int nbytes = align - (((uintptr)code_ptr()) % align);
+	if (nbytes == 0)
+		return code_ptr();
+
+#if defined(__i386__) || defined(__x86_64__)
+	/* Source: GNU Binutils 2.12.90.0.15 */
+	/* Various efficient no-op patterns for aligning code labels.
+	   Note: Don't try to assemble the instructions in the comments.
+	   0L and 0w are not legal.  */
+	static const uint8 f32_1[] =
+		{0x90};									/* nop					*/
+	static const uint8 f32_2[] =
+		{0x89,0xf6};							/* movl %esi,%esi		*/
+	static const uint8 f32_3[] =
+		{0x8d,0x76,0x00};						/* leal 0(%esi),%esi	*/
+	static const uint8 f32_4[] =
+		{0x8d,0x74,0x26,0x00};					/* leal 0(%esi,1),%esi	*/
+	static const uint8 f32_5[] =
+		{0x90,									/* nop					*/
+		 0x8d,0x74,0x26,0x00};					/* leal 0(%esi,1),%esi	*/
+	static const uint8 f32_6[] =
+		{0x8d,0xb6,0x00,0x00,0x00,0x00};		/* leal 0L(%esi),%esi	*/
+	static const uint8 f32_7[] =
+		{0x8d,0xb4,0x26,0x00,0x00,0x00,0x00};	/* leal 0L(%esi,1),%esi */
+	static const uint8 f32_8[] =
+		{0x90,									/* nop					*/
+		 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00};	/* leal 0L(%esi,1),%esi */
+	static const uint8 f32_9[] =
+		{0x89,0xf6,								/* movl %esi,%esi		*/
+		 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00};	/* leal 0L(%edi,1),%edi */
+	static const uint8 f32_10[] =
+		{0x8d,0x76,0x00,						/* leal 0(%esi),%esi	*/
+		 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00};	/* leal 0L(%edi,1),%edi */
+	static const uint8 f32_11[] =
+		{0x8d,0x74,0x26,0x00,					/* leal 0(%esi,1),%esi	*/
+		 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00};	/* leal 0L(%edi,1),%edi */
+	static const uint8 f32_12[] =
+		{0x8d,0xb6,0x00,0x00,0x00,0x00,			/* leal 0L(%esi),%esi	*/
+		 0x8d,0xbf,0x00,0x00,0x00,0x00};		/* leal 0L(%edi),%edi	*/
+	static const uint8 f32_13[] =
+		{0x8d,0xb6,0x00,0x00,0x00,0x00,			/* leal 0L(%esi),%esi	*/
+		 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00};	/* leal 0L(%edi,1),%edi */
+	static const uint8 f32_14[] =
+		{0x8d,0xb4,0x26,0x00,0x00,0x00,0x00,	/* leal 0L(%esi,1),%esi */
+		 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00};	/* leal 0L(%edi,1),%edi */
+	static const uint8 f32_15[] =
+		{0xeb,0x0d,0x90,0x90,0x90,0x90,0x90,	/* jmp .+15; lotsa nops	*/
+		 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
+	static const uint8 f32_16[] =
+		{0xeb,0x0d,0x90,0x90,0x90,0x90,0x90,	/* jmp .+15; lotsa nops	*/
+		 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
+	static const uint8 *const f32_patt[] = {
+		f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
+		f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
+	};
+	static const uint8 prefixes[4] = { 0x66, 0x66, 0x66, 0x66 };
+
+#if defined(__x86_64__)
+	/* The recommended way to pad 64bit code is to use NOPs preceded by
+	   maximally four 0x66 prefixes.  Balance the size of nops.  */
+	int i;
+	int nnops = (nbytes + 3) / 4;
+	int len = nbytes / nnops;
+	int remains = nbytes - nnops * len;
+
+	for (i = 0; i < remains; i++) {
+		emit_block(prefixes, len);
+		emit_8(0x90); // NOP
+	}
+	for (; i < nnops; i++) {
+		emit_block(prefixes, len - 1);
+		emit_8(0x90); // NOP
+	}
+#else
+	int nloops = nbytes / 16;
+	while (nloops-- > 0)
+		emit_block(f32_16, sizeof(f32_16));
+
+	nbytes %= 16;
+	if (nbytes)
+		emit_block(f32_patt[nbytes - 1], nbytes);
+#endif
+#else
+#warning "FIXME: tune for your target"
+	for (int i = 0; i < nbytes; i++)
+		emit_8(0);
+#endif
+	return code_ptr();
+}
diff --git a/SheepShaver/src/kpx_cpu/src/cpu/jit/basic-dyngen.hpp b/SheepShaver/src/kpx_cpu/src/cpu/jit/basic-dyngen.hpp
index f3a56237..b0d5831c 100644
--- a/SheepShaver/src/kpx_cpu/src/cpu/jit/basic-dyngen.hpp
+++ b/SheepShaver/src/kpx_cpu/src/cpu/jit/basic-dyngen.hpp
@@ -290,14 +290,6 @@ basic_dyngen::direct_call_possible(uintptr target) const
 	return false;
 }
 
-inline uint8 *
-basic_dyngen::gen_align(int align)
-{
-	while ((uintptr)code_ptr() & (align - 1))
-		inc_code_ptr(1);
-	return code_ptr();
-}
-
 inline uint8 *
 basic_dyngen::gen_start()
 {