JIT compiler update: fix x86-64 alignemnt issue

git-svn-id: svn://vcs.exim.org/pcre/code/trunk@956 2f5784b3-3f2a-0410-8824-cb99058d5e15

3 files changed, 57 insertions, 65 deletions

diff --git a/sljit/sljitNativeX86_32.c b/sljit/sljitNativeX86_32.c
index 24ec6e0..e955825 100644
--- a/sljit/sljitNativeX86_32.c
+++ b/sljit/sljitNativeX86_32.c
@@ -149,14 +149,14 @@ SLJIT_API_FUNC_ATTRIBUTE int sljit_emit_enter(struct sljit_compiler *compiler, i
 #ifdef _WIN32
 	if (local_size > 1024) {
 		FAIL_IF(emit_do_imm(compiler, 0xb8 + reg_map[SLJIT_TEMPORARY_REG1], local_size));
-		FAIL_IF(sljit_emit_ijump(compiler, SLJIT_CALL1, SLJIT_IMM, SLJIT_FUNC_OFFSET(sljit_touch_stack)));
+		FAIL_IF(sljit_emit_ijump(compiler, SLJIT_CALL1, SLJIT_IMM, SLJIT_FUNC_OFFSET(sljit_grow_stack)));
 	}
 #endif
 
 	compiler->local_size = local_size;
-	if (local_size > 0)
-		return emit_non_cum_binary(compiler, 0x2b, 0x29, 0x5 << 3, 0x2d,
-			SLJIT_LOCALS_REG, 0, SLJIT_LOCALS_REG, 0, SLJIT_IMM, local_size);
+	SLJIT_ASSERT(local_size > 0);
+	return emit_non_cum_binary(compiler, 0x2b, 0x29, 0x5 << 3, 0x2d,
+		SLJIT_LOCALS_REG, 0, SLJIT_LOCALS_REG, 0, SLJIT_IMM, local_size);
 
 	return SLJIT_SUCCESS;
 }
@@ -199,9 +199,9 @@ SLJIT_API_FUNC_ATTRIBUTE int sljit_emit_return(struct sljit_compiler *compiler,
 	compiler->flags_saved = 0;
 	FAIL_IF(emit_mov_before_return(compiler, op, src, srcw));
 
-	if (compiler->local_size > 0)
-		FAIL_IF(emit_cum_binary(compiler, 0x03, 0x01, 0x0 << 3, 0x05,
-				SLJIT_LOCALS_REG, 0, SLJIT_LOCALS_REG, 0, SLJIT_IMM, compiler->local_size));
+	SLJIT_ASSERT(compiler->local_size > 0);
+	FAIL_IF(emit_cum_binary(compiler, 0x03, 0x01, 0x0 << 3, 0x05,
+		SLJIT_LOCALS_REG, 0, SLJIT_LOCALS_REG, 0, SLJIT_IMM, compiler->local_size));
 
 	size = 2 + (compiler->saveds <= 3 ? compiler->saveds : 3);
 #if (defined SLJIT_X86_32_FASTCALL && SLJIT_X86_32_FASTCALL)
diff --git a/sljit/sljitNativeX86_64.c b/sljit/sljitNativeX86_64.c
index 5b15ccf..480cebc 100644
--- a/sljit/sljitNativeX86_64.c
+++ b/sljit/sljitNativeX86_64.c
@@ -196,12 +196,11 @@ SLJIT_API_FUNC_ATTRIBUTE int sljit_emit_enter(struct sljit_compiler *compiler, i
 #endif
 	}
 
-	local_size = ((local_size + pushed_size + 16 - 1) & ~(16 - 1)) - pushed_size;
-#ifdef _WIN64
-	local_size += 4 * sizeof(sljit_w);
+	local_size = ((local_size + FIXED_LOCALS_OFFSET + pushed_size + 16 - 1) & ~(16 - 1)) - pushed_size;
 	compiler->local_size = local_size;
+#ifdef _WIN64
 	if (local_size > 1024) {
-		/* Allocate the stack for the function itself. */
+		/* Allocate stack for the callback, which grows the stack. */
 		buf = (sljit_ub*)ensure_buf(compiler, 1 + 4);
 		FAIL_IF(!buf);
 		INC_SIZE(4);
@@ -218,36 +217,29 @@ SLJIT_API_FUNC_ATTRIBUTE int sljit_emit_enter(struct sljit_compiler *compiler, i
 			local_size -= 4 * sizeof(sljit_w);
 		}
 		FAIL_IF(emit_load_imm64(compiler, SLJIT_TEMPORARY_REG1, local_size));
-		FAIL_IF(sljit_emit_ijump(compiler, SLJIT_CALL1, SLJIT_IMM, SLJIT_FUNC_OFFSET(sljit_touch_stack)));
+		FAIL_IF(sljit_emit_ijump(compiler, SLJIT_CALL1, SLJIT_IMM, SLJIT_FUNC_OFFSET(sljit_grow_stack)));
 	}
-#else
-	local_size += sizeof(sljit_w);
-	compiler->local_size = local_size;
-	if (local_size > 0) {
 #endif
-		/* In case of Win64, local_size is always > 4 * sizeof(sljit_w) */
-		if (local_size <= 127) {
-			buf = (sljit_ub*)ensure_buf(compiler, 1 + 4);
-			FAIL_IF(!buf);
-			INC_SIZE(4);
-			*buf++ = REX_W;
-			*buf++ = 0x83;
-			*buf++ = 0xc0 | (5 << 3) | 4;
-			*buf++ = local_size;
-		}
-		else {
-			buf = (sljit_ub*)ensure_buf(compiler, 1 + 7);
-			FAIL_IF(!buf);
-			INC_SIZE(7);
-			*buf++ = REX_W;
-			*buf++ = 0x81;
-			*buf++ = 0xc0 | (5 << 3) | 4;
-			*(sljit_hw*)buf = local_size;
-			buf += sizeof(sljit_hw);
-		}
-#ifndef _WIN64
+	SLJIT_ASSERT(local_size > 0);
+	if (local_size <= 127) {
+		buf = (sljit_ub*)ensure_buf(compiler, 1 + 4);
+		FAIL_IF(!buf);
+		INC_SIZE(4);
+		*buf++ = REX_W;
+		*buf++ = 0x83;
+		*buf++ = 0xc0 | (5 << 3) | 4;
+		*buf++ = local_size;
+	}
+	else {
+		buf = (sljit_ub*)ensure_buf(compiler, 1 + 7);
+		FAIL_IF(!buf);
+		INC_SIZE(7);
+		*buf++ = REX_W;
+		*buf++ = 0x81;
+		*buf++ = 0xc0 | (5 << 3) | 4;
+		*(sljit_hw*)buf = local_size;
+		buf += sizeof(sljit_hw);
 	}
-#endif
 
 	return SLJIT_SUCCESS;
 }
@@ -271,12 +263,7 @@ SLJIT_API_FUNC_ATTRIBUTE void sljit_set_context(struct sljit_compiler *compiler,
 	if (temporaries >= 5)
 		pushed_size += sizeof(sljit_w);
 #endif
-	compiler->local_size = ((local_size + pushed_size + 16 - 1) & ~(16 - 1)) - pushed_size;
-#ifdef _WIN64
-	compiler->local_size += 4 * sizeof(sljit_w);
-#else
-	compiler->local_size += sizeof(sljit_w);
-#endif
+	compiler->local_size = ((local_size + FIXED_LOCALS_OFFSET + pushed_size + 16 - 1) & ~(16 - 1)) - pushed_size;
 }
 
 SLJIT_API_FUNC_ATTRIBUTE int sljit_emit_return(struct sljit_compiler *compiler, int op, int src, sljit_w srcw)
@@ -291,25 +278,24 @@ SLJIT_API_FUNC_ATTRIBUTE int sljit_emit_return(struct sljit_compiler *compiler,
 	compiler->flags_saved = 0;
 	FAIL_IF(emit_mov_before_return(compiler, op, src, srcw));
 
-	if (compiler->local_size > 0) {
-		if (compiler->local_size <= 127) {
-			buf = (sljit_ub*)ensure_buf(compiler, 1 + 4);
-			FAIL_IF(!buf);
-			INC_SIZE(4);
-			*buf++ = REX_W;
-			*buf++ = 0x83;
-			*buf++ = 0xc0 | (0 << 3) | 4;
-			*buf = compiler->local_size;
-		}
-		else {
-			buf = (sljit_ub*)ensure_buf(compiler, 1 + 7);
-			FAIL_IF(!buf);
-			INC_SIZE(7);
-			*buf++ = REX_W;
-			*buf++ = 0x81;
-			*buf++ = 0xc0 | (0 << 3) | 4;
-			*(sljit_hw*)buf = compiler->local_size;
-		}
+	SLJIT_ASSERT(compiler->local_size > 0);
+	if (compiler->local_size <= 127) {
+		buf = (sljit_ub*)ensure_buf(compiler, 1 + 4);
+		FAIL_IF(!buf);
+		INC_SIZE(4);
+		*buf++ = REX_W;
+		*buf++ = 0x83;
+		*buf++ = 0xc0 | (0 << 3) | 4;
+		*buf = compiler->local_size;
+	}
+	else {
+		buf = (sljit_ub*)ensure_buf(compiler, 1 + 7);
+		FAIL_IF(!buf);
+		INC_SIZE(7);
+		*buf++ = REX_W;
+		*buf++ = 0x81;
+		*buf++ = 0xc0 | (0 << 3) | 4;
+		*(sljit_hw*)buf = compiler->local_size;
 	}
 
 	size = 1 + compiler->saveds;
diff --git a/sljit/sljitNativeX86_common.c b/sljit/sljitNativeX86_common.c
index 5370d10..49b9089 100644
--- a/sljit/sljitNativeX86_common.c
+++ b/sljit/sljitNativeX86_common.c
@@ -457,11 +457,17 @@ static SLJIT_INLINE int emit_restore_flags(struct sljit_compiler *compiler, int
 #ifdef _WIN32
 #include <malloc.h>
 
-static void SLJIT_CALL sljit_touch_stack(sljit_w local_size)
+static void SLJIT_CALL sljit_grow_stack(sljit_w local_size)
 {
-	/* Workaround for calling _chkstk. */
+	/* Workaround for calling the internal _chkstk() function on Windows.
+	This function touches all 4k pages belongs to the requested stack space,
+	which size is passed in local_size. This is necessary on Windows where
+	the stack can only grow in 4k steps. However, this function just burn
+	CPU cycles if the stack is large enough, but you don't know it in advance.
+	I think this is a bad design even if it has some reasons. */
 	alloca(local_size);
 }
+
 #endif
 
 #if (defined SLJIT_CONFIG_X86_32 && SLJIT_CONFIG_X86_32)