From 02daf9f8b5c1ae2267561ee05ce67b2d0393c12d Mon Sep 17 00:00:00 2001
From: Oliver Schmidt <ol.sc@web.de>
Date: Mon, 3 Apr 2017 23:20:26 +0200
Subject: [PATCH] So far the built-in inlining of several known standard
 function was always (!) enabled and the option -Os enabled additional,
 potentially unsafe inlining of some of those functions.

There were two aspects of this behavior that were considered undesirable:
- Although the safe inlining is in general desirable it should only be enabled if asked for it - like any other optimization.
- The option name -Os implies that it is a safe option, the potentially unsafe inlining should have a more explicit name.

So now:
- The option -Os enables the safe inlining.
- The new option --eagerly-inline-funcs enables the potentially unsafe inlining (including the safe inlining).

Additionally was added:
- The option --inline-stdfuncs that does like -Os enable the safe inlining but doesn't enable optimizations.
- The pragma inline-stdfuncs that works identical to --inline-stdfuncs.
- The pragma allow-eager-inline that enables the potentially unsafe inlining but doesn't include the safe inlining. That means that by itself it only marks code as safe for potentially unsafe inlining but doesn't actually enable any inlining.
---
 doc/cc65.sgml      |  296 ++++++----
 doc/cl65.sgml      |    4 +-
 include/ctype.h    |    4 +-
 src/cc65/compile.c |   19 +-
 src/cc65/global.c  |    3 +-
 src/cc65/global.h  |    3 +-
 src/cc65/main.c    |   91 +--
 src/cc65/pragma.c  |   62 +-
 src/cc65/stdfunc.c | 1379 +++++++++++++++++++++++---------------------
 9 files changed, 1013 insertions(+), 848 deletions(-)

diff --git a/doc/cc65.sgml b/doc/cc65.sgml
index 2a0fa0260..9322a2d44 100644
--- a/doc/cc65.sgml
+++ b/doc/cc65.sgml
@@ -58,7 +58,7 @@ Short options:
   -O                            Optimize code
   -Oi                           Optimize code, inline more code
   -Or                           Enable register variables
-  -Os                           Inline some known functions
+  -Os                           Inline some standard functions
   -T                            Include source as comment
   -V                            Print the compiler version number
   -W warning[,...]              Suppress warnings
@@ -88,9 +88,11 @@ Long options:
   --debug-opt name              Debug optimization steps
   --dep-target target           Use this dependency target
   --disable-opt name            Disable an optimization step
+  --eagerly-inline-funcs        Eagerly inline some known functions
   --enable-opt name             Enable an optimization step
   --help                        Help (this text)
   --include-dir dir             Set an include directory search path
+  --inline-stdfuncs             Inline some standard functions
   --list-opt-steps              List all optimizer steps and exit
   --list-warnings               List available warning types for -W
   --local-strings               Emit string literals immediately
@@ -219,11 +221,53 @@ Here is a description of all the command line options:
   symbols in a special section in the object file.
 
 
+  <label id="option-eagerly-inline-funcs">
+  <tag><tt>--eagerly-inline-funcs</tt></tag>
+
+  Have the compiler eagerly inline these functions from the C library:
+  <itemize>
+  <item><tt/memcpy()/
+  <item><tt/memset()/
+  <item><tt/strcmp()/
+  <item><tt/strcpy()/
+  <item><tt/strlen()/
+  <item>most of the functions declared in <tt/&lt;ctype.h&gt;/
+  </itemize>
+
+  Note: This has two consequences:
+  <itemize>
+  <item>You may not use names of standard C functions for your own functions.
+        If you do that, your program is not standard-compliant anyway; but,
+        using <tt/--eagerly-inline-funcs/ actually will break things.
+        <p>
+  <item>The inlined string and memory functions will not handle strings or
+        memory areas larger than 255 bytes.  Similarly, the inlined <tt/is..()/
+        functions will not work with values outside the char. range (such as
+        <tt/EOF/).
+        <p>
+  </itemize>
+
+  <tt/--eagerly-inline-funcs/ implies the <tt/<ref id="option-inline-stdfuncs"
+  name="--inline-stdfuncs"/ command line option.
+
+  See also <tt/<ref id="pragma-allow-eager-inline" name="#pragma&nbsp;allow-eager-inline">/.
+
+
   <tag><tt>-h, --help</tt></tag>
 
   Print the short option summary shown above.
 
 
+  <label id="option-inline-stdfuncs">
+  <tag><tt>--inline-stdfuncs</tt></tag>
+
+  Allow the compiler to inline some standard functions from the C library like
+  strlen.  This will not only remove the overhead for a function call, but will
+  make the code visible for the optimizer.  See also the <tt/<ref id="option-O"
+  name="-Os"/ command line option and <tt/<ref id="pragma-inline-stdfuncs"
+  name="#pragma&nbsp;inline-stdfuncs">/.
+
+
   <label id="option-list-warnings">
   <tag><tt>--list-warnings</tt></tag>
 
@@ -392,22 +436,22 @@ Here is a description of all the command line options:
   using
 
   <tscreen><verb>
-   	void f (void)
-   	{
-   	    unsigned a = 1;
-   	    ...
-   	}
+        void f (void)
+        {
+            unsigned a = 1;
+            ...
+        }
   </verb></tscreen>
 
   the variable <tt/a/ will always have the value <tt/1/ when entering the
   function and using <tt/-Cl/, while in
 
   <tscreen><verb>
-   	void f (void)
-   	{
-   	    static unsigned a = 1;
-   	    ....
-   	}
+        void f (void)
+        {
+            static unsigned a = 1;
+            ....
+        }
   </verb></tscreen>
 
   the variable <tt/a/ will have the value <tt/1/ only the first time that the
@@ -444,23 +488,13 @@ Here is a description of all the command line options:
   name="--register-vars">/ command line option, and the <ref
   id="register-vars" name="discussion of register variables"> below.
 
-  Using <tt/-Os/ will force the compiler to inline some known functions from
-  the C library like strlen. Note: This has two consequences:
-  <p>
-  <itemize>
-  <item>You may not use names of standard C functions in your own code. If you
-     	do that, your program is not standard compliant anyway, but using
-     	<tt/-Os/ will actually break things.
-     	<p>
-  <item>The inlined string and memory functions will not handle strings or
-	memory areas larger than 255 bytes. Similarly, the inlined <tt/is..()/
-	functions will not work with values outside the char. range (such as
-	<tt/EOF/).
-     	<p>
-  </itemize>
-  <p>
+  Using <tt/-Os/ will allow the compiler to inline some standard functions
+  from the C library like strlen.  This will not only remove the overhead
+  for a function call, but will make the code visible for the optimizer.
+  See also <tt/<ref id="option-inline-stdfuncs" name="--inline-stdfuncs"/.
+
   It is possible to concatenate the modifiers for <tt/-O/. For example, to
-  enable register variables and inlining of known functions, you may use
+  enable register variables and inlining of standard functions, you may use
   <tt/-Ors/.
 
 
@@ -518,6 +552,7 @@ Here is a description of all the command line options:
 </descrip><p>
 
 
+
 <sect>Input and output<p>
 
 The compiler will accept one C file per invocation and create a file with
@@ -556,21 +591,21 @@ and the one defined by the ISO standard:
 
 <itemize>
 
-<item>	The datatypes "float" and "double" are not available.
-      	<p>
-<item> 	C Functions may not return structs (or unions), and structs may not
+<item>  The datatypes "float" and "double" are not available.
+        <p>
+<item>  C Functions may not return structs (or unions), and structs may not
         be passed as parameters by value. However, struct assignment *is*
-      	possible.
-      	<p>
-<item>	Most of the C library is available with only the fastcall calling
-	convention (<ref id="extension-fastcall" name="see below">).  It means
-	that you must not mix pointers to those functions with pointers to
-	user-written, cdecl functions (the calling conventions are incompatible).
-	<p>
-<item>	The <tt/volatile/ keyword has almost no effect. That is not as bad
+        possible.
+        <p>
+<item>  Most of the C library is available with only the fastcall calling
+        convention (<ref id="extension-fastcall" name="see below">).  It means
+        that you must not mix pointers to those functions with pointers to
+        user-written, cdecl functions (the calling conventions are incompatible).
+        <p>
+<item>  The <tt/volatile/ keyword has almost no effect. That is not as bad
         as it sounds, since the 6502 has so few registers that it isn't
         possible to keep values in registers anyway.
-	<p>
+        <p>
 </itemize>
 
 There may be some more minor differences I'm currently not aware of. The
@@ -585,49 +620,48 @@ This cc65 version has some extensions to the ISO C standard.
 
 <itemize>
 
-<item>	The compiler allows to insert assembler statements into the output
-     	file. The syntax is
+<item>  The compiler allows to insert assembler statements into the output
+        file. The syntax is
 
-	<tscreen><verb>
-	asm [optional volatile] (&lt;string literal&gt;[, optional parameters]) ;
-     	</verb></tscreen>
-     	or
-	<tscreen><verb>
+        <tscreen><verb>
+        asm [optional volatile] (&lt;string literal&gt;[, optional parameters]) ;
+        </verb></tscreen>
+        or
+        <tscreen><verb>
         __asm__ [optional volatile] (&lt;string literal&gt;[, optional parameters]) ;
-     	</verb></tscreen>
+        </verb></tscreen>
 
-	The first form is in the user namespace; and, is disabled if the <tt/-A/
-     	switch is given.
+        The first form is in the user namespace; and, is disabled if the <tt/-A/
+        switch is given.
 
-	There is a whole section covering inline assembler statements,
-	<ref id="inline-asm" name="see there">.
-	<p>
+        There is a whole section covering inline assembler statements,
+        <ref id="inline-asm" name="see there">.
+        <p>
 
 <label id="extension-fastcall">
-<item>	The normal calling convention -- for non-variadic functions -- is
-	named "fastcall". The syntax for a function declaration that
-	<em/explicitly/ uses fastcall is
-
-	<tscreen><verb>
-	&lt;return type&gt; fastcall &lt;function name&gt; (&lt;parameter list&gt;)
-     	</verb></tscreen>
-     	or
-	<tscreen><verb>
-	&lt;return type&gt; __fastcall__ &lt;function name&gt; (&lt;parameter list&gt;)
-     	</verb></tscreen>
-	An example is
-	<tscreen><verb>
-	void __fastcall__ f (unsigned char c)
-     	</verb></tscreen>
-	The first form of the fastcall keyword is in the user namespace and can
-	therefore be disabled with the <tt><ref id="option--standard"
+<item>  The normal calling convention -- for non-variadic functions -- is
+        named "fastcall". The syntax for a function declaration that
+        <em/explicitly/ uses fastcall is
+
+        <tscreen><verb>
+        &lt;return type&gt; fastcall &lt;function name&gt; (&lt;parameter list&gt;)
+        </verb></tscreen>
+        or
+        <tscreen><verb>
+        &lt;return type&gt; __fastcall__ &lt;function name&gt; (&lt;parameter list&gt;)
+        </verb></tscreen>
+        An example is
+        <tscreen><verb>
+        void __fastcall__ f (unsigned char c)
+        </verb></tscreen>
+        The first form of the fastcall keyword is in the user namespace and can
+        therefore be disabled with the <tt><ref id="option--standard"
         name="--standard"></tt> command line option.
 
-	For functions that are <tt/fastcall/, the rightmost parameter is not
-	pushed on the stack but left in the primary register when the function
-	is called. That significantly reduces the cost of calling those functions.
-	<newline><newline>
-	<p>
+        For functions that are <tt/fastcall/, the rightmost parameter is not
+        pushed on the stack but left in the primary register when the function
+        is called. That significantly reduces the cost of calling those functions.
+        <p>
 
 <item>  There is another calling convention named "cdecl". Variadic functions
         (their prototypes have an ellipsis &lsqb;<tt/.../&rsqb;) always use that
@@ -652,40 +686,40 @@ This cc65 version has some extensions to the ISO C standard.
         For functions that are <tt/cdecl/, the rightmost parameter is pushed
         onto the stack before the function is called. That increases the cost
         of calling those functions, especially when they are called from many
-        places.<newline><newline>
+        places.
         <p>
 
-<item>	There are two pseudo variables named <tt/__AX__/ and <tt/__EAX__/.
-	Both refer to the primary register that is used by the compiler to
-	evaluate expressions or return function results. <tt/__AX__/ is of
-	type <tt/unsigned int/ and <tt/__EAX__/ of type <tt/long unsigned int/
-	respectively. The pseudo variables may be used as lvalue and rvalue as
-	every other variable. They are most useful together with short
-	sequences of assembler code. For example, the macro
+<item>  There are two pseudo variables named <tt/__AX__/ and <tt/__EAX__/.
+        Both refer to the primary register that is used by the compiler to
+        evaluate expressions or return function results. <tt/__AX__/ is of
+        type <tt/unsigned int/ and <tt/__EAX__/ of type <tt/long unsigned int/
+        respectively. The pseudo variables may be used as lvalue and rvalue as
+        every other variable. They are most useful together with short
+        sequences of assembler code. For example, the macro
 
-	<tscreen><verb>
-       	#define hi(x)           \
+        <tscreen><verb>
+        #define hi(x)           \
             (__AX__ = (x),      \
              asm ("txa"),       \
              asm ("ldx #$00"),  \
              __AX__)
-	</verb></tscreen>
+        </verb></tscreen>
 
-	will give the high byte of any unsigned value.
-	<p>
+        will give the high byte of any unsigned value.
+        <p>
 
-<item>	Inside a function, the identifier <tt/__func__/ gives the name of the
-	current function as a string. Outside of functions, <tt/__func__/ is
-	undefined.
-	Example:
+<item>  Inside a function, the identifier <tt/__func__/ gives the name of the
+        current function as a string. Outside of functions, <tt/__func__/ is
+        undefined.
+        Example:
 
-	<tscreen><verb>
-	#define PRINT_DEBUG(s)	printf ("%s: %s\n", __func__, s);
-	</verb></tscreen>
+        <tscreen><verb>
+        #define PRINT_DEBUG(s)  printf ("%s: %s\n", __func__, s);
+        </verb></tscreen>
 
-	The macro will print the name of the current function plus a given
-	string.
-	<p>
+        The macro will print the name of the current function plus a given
+        string.
+        <p>
 
 <item>  cc65 allows the initialization of <tt/void/ variables. This may be
         used to create arbitrary structures that are more compatible with
@@ -825,6 +859,11 @@ The compiler defines several macros at startup:
   This macro expands to the date of translation of the preprocessing
   translation unit in the form "Mmm dd yyyy".
 
+  <tag><tt>__EAGERLY_INLINE_FUNCS__</tt></tag>
+
+  Is defined if the compiler was called with the <tt/<ref id="option-eagerly-inline-funcs"
+  name="--eagerly-inline-funcs"/ command line option.
+
   <tag><tt>__FILE__</tt></tag>
 
   This macro expands to a string containing the name of the C source file.
@@ -912,6 +951,7 @@ The compiler defines several macros at startup:
 </descrip>
 
 
+
 <sect>&num;pragmas<label id="pragmas"><p>
 
 The compiler understands some pragmas that may be used to change code
@@ -920,6 +960,19 @@ If the first parameter is <tt/push/, the old value is saved onto a stack
 before changing it. The value may later be restored by using the <tt/pop/
 parameter with the <tt/#pragma/.
 
+
+<sect1><tt>#pragma allow-eager-inline ([push,] on|off)</tt><label id="pragma-allow-eager-inline"><p>
+
+  Allow eager inlining of known functions.  If the argument is "off", eager
+  inlining is disabled, otherwise it is enabled. Please note that (in contrast
+  to the <tt/<ref id="option-eagerly-inline-funcs" name="--eagerly-inline-funcs"/
+  command line option) this pragma does not imply the <tt/<ref id="option-inline-stdfuncs"
+  name="--inline-stdfuncs"/ command line option.  Rather it marks code to be safe for
+  eager inlining of known functions if inlining of standard functions is enabled.
+
+  The <tt/#pragma/ understands the push and pop parameters as explained above.
+
+
 <sect1><tt>#pragma bss-name (&lsqb;push,&rsqb; &lt;name&gt;)</tt><label id="pragma-bss-name"><p>
 
   This pragma changes the name used for the BSS segment (the BSS segment
@@ -938,7 +991,7 @@ parameter with the <tt/#pragma/.
 
   Example:
   <tscreen><verb>
-       	#pragma bss-name ("MyBSS")
+        #pragma bss-name ("MyBSS")
   </verb></tscreen>
 
 
@@ -993,6 +1046,7 @@ parameter with the <tt/#pragma/.
 
   The <tt/#pragma/ understands the push and pop parameters as explained above.
 
+
 <sect1><tt>#pragma code-name ([push,] &lt;name&gt;)</tt><label id="pragma-code-name"><p>
 
   This pragma changes the name used for the CODE segment (the CODE segment
@@ -1007,7 +1061,7 @@ parameter with the <tt/#pragma/.
 
   Example:
   <tscreen><verb>
-       	#pragma code-name ("MyCODE")
+        #pragma code-name ("MyCODE")
   </verb></tscreen>
 
 
@@ -1035,10 +1089,21 @@ parameter with the <tt/#pragma/.
 
   Example:
   <tscreen><verb>
-       	#pragma data-name ("MyDATA")
+        #pragma data-name ("MyDATA")
   </verb></tscreen>
 
 
+<sect1><tt>#pragma inline-stdfuncs ([push,] on|off)</tt><label id="pragma-inline-stdfuncs"><p>
+
+  Allow the compiler to inline some standard functions from the C library like
+  strlen.  If the argument is "off", inlining is disabled, otherwise it is enabled.
+
+  See also the the <tt/<ref id="option-inline-stdfuncs" name="--inline-stdfuncs"/
+  command line option.
+
+  The <tt/#pragma/ understands the push and pop parameters as explained above.
+
+
 <sect1><tt>#pragma local-strings ([push,] on|off)</tt><label id="pragma-local-strings"><p>
 
   When "on", emit string literals to the data segment when they're encountered
@@ -1083,7 +1148,7 @@ parameter with the <tt/#pragma/.
 
   Example:
   <tscreen><verb>
-       	#pragma rodata-name ("MyRODATA")
+        #pragma rodata-name ("MyRODATA")
   </verb></tscreen>
 
 
@@ -1105,9 +1170,9 @@ parameter with the <tt/#pragma/.
 
   Example:
   <tscreen><verb>
-       	#pragma regvaraddr(on) 	/* Allow taking the address
-       		      	       	 * of register variables
-       		      	       	 */
+        #pragma regvaraddr(on)  /* Allow taking the address
+                                 * of register variables
+                                 */
   </verb></tscreen>
 
 
@@ -1154,7 +1219,7 @@ parameter with the <tt/#pragma/.
   Example:
   <tscreen><verb>
         /* Don't warn about the unused parameter in function func */
-      	#pragma warn (unused-param, push, off)
+        #pragma warn (unused-param, push, off)
         static int func (int unused)
         {
             return 0;
@@ -1187,13 +1252,12 @@ parameter with the <tt/#pragma/.
 
   Example:
   <tscreen><verb>
-      	extern int foo;
-      	#pragma zpsym ("foo");	/* foo is in the zeropage */
+        extern int foo;
+        #pragma zpsym ("foo");  /* foo is in the zeropage */
   </verb></tscreen>
 
 
 
-
 <sect>Register variables<label id="register-vars"><p>
 
 The runtime for all supported platforms has 6 bytes of zero page space
@@ -1450,14 +1514,14 @@ including commercial applications, and to alter it and redistribute it
 freely, subject to the following restrictions:
 
 <enum>
-<item> 	The origin of this software must not be misrepresented; you must not
-       	claim that you wrote the original software. If you use this software
-       	in a product, an acknowledgment in the product documentation would be
-	appreciated but is not required.
-<item>	Altered source versions must be plainly marked as such, and must not
-	be misrepresented as being the original software.
-<item>	This notice may not be removed or altered from any source
-	distribution.
+<item>  The origin of this software must not be misrepresented; you must not
+        claim that you wrote the original software. If you use this software
+        in a product, an acknowledgment in the product documentation would be
+        appreciated but is not required.
+<item>  Altered source versions must be plainly marked as such, and must not
+        be misrepresented as being the original software.
+<item>  This notice may not be removed or altered from any source
+        distribution.
 </enum>
 
 </article>
diff --git a/doc/cl65.sgml b/doc/cl65.sgml
index eef6a12a3..ffeba2321 100644
--- a/doc/cl65.sgml
+++ b/doc/cl65.sgml
@@ -54,9 +54,9 @@ Short options:
   -L path                       Specify a library search path
   -Ln name                      Create a VICE label file
   -O                            Optimize code
-  -Oi                           Optimize code, inline functions
+  -Oi                           Optimize code, inline more code
   -Or                           Optimize code, honour the register keyword
-  -Os                           Optimize code, inline known C funtions
+  -Os                           Optimize code, inline standard funtions
   -S                            Compile but don't assemble and link
   -T                            Include source as comment
   -V                            Print the version number
diff --git a/include/ctype.h b/include/ctype.h
index b440bfb70..17cbafe9f 100644
--- a/include/ctype.h
+++ b/include/ctype.h
@@ -89,9 +89,9 @@ unsigned char __fastcall__ toascii (unsigned char c);
 ** #undef'ing the macroes.
 ** Please note that the following macroes do NOT handle EOF correctly, as
 ** stated in the manual. If you need correct behaviour for EOF, don't
-** use -Os, or #undefine the following macroes.
+** use --eagerly-inline-funcs, or #undefine the following macroes.
 */
-#ifdef __OPT_s__
+#ifdef __EAGERLY_INLINE_FUNCS__
 
 #define isalnum(c)  (__AX__ = (c),                      \
                     __asm__ ("tay"),                    \
diff --git a/src/cc65/compile.c b/src/cc65/compile.c
index 4425b6aad..3c1042154 100644
--- a/src/cc65/compile.c
+++ b/src/cc65/compile.c
@@ -335,17 +335,22 @@ void Compile (const char* FileName)
     ** changes using #pragma later.
     */
     if (IS_Get (&Optimize)) {
-        long CodeSize = IS_Get (&CodeSizeFactor);
         DefineNumericMacro ("__OPT__", 1);
+    }
+    {
+        long CodeSize = IS_Get (&CodeSizeFactor);
         if (CodeSize > 100) {
             DefineNumericMacro ("__OPT_i__", CodeSize);
         }
-        if (IS_Get (&EnableRegVars)) {
-            DefineNumericMacro ("__OPT_r__", 1);
-        }
-        if (IS_Get (&InlineStdFuncs)) {
-            DefineNumericMacro ("__OPT_s__", 1);
-        }
+    }
+    if (IS_Get (&EnableRegVars)) {
+        DefineNumericMacro ("__OPT_r__", 1);
+    }
+    if (IS_Get (&InlineStdFuncs)) {
+        DefineNumericMacro ("__OPT_s__", 1);
+    }
+    if (IS_Get (&EagerlyInlineFuncs)) {
+        DefineNumericMacro ("__EAGERLY_INLINE_FUNCS__", 1);
     }
 
     /* __TIME__ and __DATE__ macros */
diff --git a/src/cc65/global.c b/src/cc65/global.c
index dbdd72f3c..a337549fe 100644
--- a/src/cc65/global.c
+++ b/src/cc65/global.c
@@ -53,7 +53,8 @@ unsigned      RegisterSpace     = 6;    /* Space available for register vars */
 /* Stackable options */
 IntStack WritableStrings    = INTSTACK(0);  /* Literal strings are r/w */
 IntStack LocalStrings       = INTSTACK(0);  /* Emit string literals immediately */
-IntStack InlineStdFuncs     = INTSTACK(0);  /* Inline some known functions */
+IntStack InlineStdFuncs     = INTSTACK(0);  /* Inline some standard functions */
+IntStack EagerlyInlineFuncs = INTSTACK(0);  /* Eagerly inline some known functions */
 IntStack EnableRegVars      = INTSTACK(0);  /* Enable register variables */
 IntStack AllowRegVarAddr    = INTSTACK(0);  /* Allow taking addresses of register vars */
 IntStack RegVarsToCallStack = INTSTACK(0);  /* Save reg variables on call stack */
diff --git a/src/cc65/global.h b/src/cc65/global.h
index 8b0af5a83..4ffc84a39 100644
--- a/src/cc65/global.h
+++ b/src/cc65/global.h
@@ -61,7 +61,8 @@ extern unsigned         RegisterSpace;          /* Space available for register
 /* Stackable options */
 extern IntStack         WritableStrings;        /* Literal strings are r/w */
 extern IntStack         LocalStrings;           /* Emit string literals immediately */
-extern IntStack         InlineStdFuncs;         /* Inline some known functions */
+extern IntStack         InlineStdFuncs;         /* Inline some standard functions */
+extern IntStack         EagerlyInlineFuncs;     /* Eagerly inline some known functions */
 extern IntStack         EnableRegVars;          /* Enable register variables */
 extern IntStack         AllowRegVarAddr;        /* Allow taking addresses of register vars */
 extern IntStack         RegVarsToCallStack;     /* Save reg variables on call stack */
diff --git a/src/cc65/main.c b/src/cc65/main.c
index 9ed19b90c..d3d298876 100644
--- a/src/cc65/main.c
+++ b/src/cc65/main.c
@@ -88,7 +88,7 @@ static void Usage (void)
             "  -O\t\t\t\tOptimize code\n"
             "  -Oi\t\t\t\tOptimize code, inline more code\n"
             "  -Or\t\t\t\tEnable register variables\n"
-            "  -Os\t\t\t\tInline some known functions\n"
+            "  -Os\t\t\t\tInline some standard functions\n"
             "  -T\t\t\t\tInclude source as comment\n"
             "  -V\t\t\t\tPrint the compiler version number\n"
             "  -W warning[,...]\t\tSuppress warnings\n"
@@ -118,9 +118,11 @@ static void Usage (void)
             "  --debug-opt name\t\tDebug optimization steps\n"
             "  --dep-target target\t\tUse this dependency target\n"
             "  --disable-opt name\t\tDisable an optimization step\n"
+            "  --eagerly-inline-funcs\t\tEagerly inline some known functions\n"
             "  --enable-opt name\t\tEnable an optimization step\n"
             "  --help\t\t\tHelp (this text)\n"
             "  --include-dir dir\t\tSet an include directory search path\n"
+            "  --inline-stdfuncs\t\tInline some standard functions\n"
             "  --list-opt-steps\t\tList all optimizer steps and exit\n"
             "  --list-warnings\t\tList available warning types for -W\n"
             "  --local-strings\t\tEmit string literals immediately\n"
@@ -581,6 +583,16 @@ static void OptDisableOpt (const char* Opt attribute ((unused)), const char* Arg
 
 
 
+static void OptEagerlyInlineFuncs (const char* Opt attribute((unused)),
+                                   const char* Arg attribute((unused)))
+/* Eagerly inline some known functions */
+{
+    IS_Set (&InlineStdFuncs, 1);
+    IS_Set (&EagerlyInlineFuncs, 1);
+}
+
+
+
 static void OptEnableOpt (const char* Opt attribute ((unused)), const char* Arg)
 /* Enable an optimization step */
 {
@@ -608,6 +620,15 @@ static void OptIncludeDir (const char* Opt attribute ((unused)), const char* Arg
 
 
 
+static void OptInlineStdFuncs (const char* Opt attribute((unused)),
+                               const char* Arg attribute((unused)))
+/* Inline some standard functions */
+{
+    IS_Set (&InlineStdFuncs, 1);
+}
+
+
+
 static void OptListOptSteps (const char* Opt attribute ((unused)),
                              const char* Arg attribute ((unused)))
 /* List all optimizer steps */
@@ -819,39 +840,41 @@ int main (int argc, char* argv[])
 {
     /* Program long options */
     static const LongOpt OptTab[] = {
-        { "--add-source",       0,      OptAddSource            },
-        { "--all-cdecl",        0,      OptAllCDecl             },
-        { "--bss-name",         1,      OptBssName              },
-        { "--check-stack",      0,      OptCheckStack           },
-        { "--code-name",        1,      OptCodeName             },
-        { "--codesize",         1,      OptCodeSize             },
-        { "--cpu",              1,      OptCPU                  },
-        { "--create-dep",       1,      OptCreateDep            },
-        { "--create-full-dep",  1,      OptCreateFullDep        },
-        { "--data-name",        1,      OptDataName             },
-        { "--debug",            0,      OptDebug                },
-        { "--debug-info",       0,      OptDebugInfo            },
-        { "--debug-opt",        1,      OptDebugOpt             },
-        { "--debug-opt-output", 0,      OptDebugOptOutput       },
-        { "--dep-target",       1,      OptDepTarget            },
-        { "--disable-opt",      1,      OptDisableOpt           },
-        { "--enable-opt",       1,      OptEnableOpt            },
-        { "--help",             0,      OptHelp                 },
-        { "--include-dir",      1,      OptIncludeDir           },
-        { "--list-opt-steps",   0,      OptListOptSteps         },
-        { "--list-warnings",    0,      OptListWarnings         },
-        { "--local-strings",    0,      OptLocalStrings         },
-        { "--memory-model",     1,      OptMemoryModel          },
-        { "--register-space",   1,      OptRegisterSpace        },
-        { "--register-vars",    0,      OptRegisterVars         },
-        { "--rodata-name",      1,      OptRodataName           },
-        { "--signed-chars",     0,      OptSignedChars          },
-        { "--standard",         1,      OptStandard             },
-        { "--static-locals",    0,      OptStaticLocals         },
-        { "--target",           1,      OptTarget               },
-        { "--verbose",          0,      OptVerbose              },
-        { "--version",          0,      OptVersion              },
-        { "--writable-strings", 0,      OptWritableStrings      },
+        { "--add-source",           0,      OptAddSource            },
+        { "--all-cdecl",            0,      OptAllCDecl             },
+        { "--bss-name",             1,      OptBssName              },
+        { "--check-stack",          0,      OptCheckStack           },
+        { "--code-name",            1,      OptCodeName             },
+        { "--codesize",             1,      OptCodeSize             },
+        { "--cpu",                  1,      OptCPU                  },
+        { "--create-dep",           1,      OptCreateDep            },
+        { "--create-full-dep",      1,      OptCreateFullDep        },
+        { "--data-name",            1,      OptDataName             },
+        { "--debug",                0,      OptDebug                },
+        { "--debug-info",           0,      OptDebugInfo            },
+        { "--debug-opt",            1,      OptDebugOpt             },
+        { "--debug-opt-output",     0,      OptDebugOptOutput       },
+        { "--dep-target",           1,      OptDepTarget            },
+        { "--disable-opt",          1,      OptDisableOpt           },
+        { "--eagerly-inline-funcs", 0,      OptEagerlyInlineFuncs   },
+        { "--enable-opt",           1,      OptEnableOpt            },
+        { "--help",                 0,      OptHelp                 },
+        { "--include-dir",          1,      OptIncludeDir           },
+        { "--inline-stdfuncs",      0,      OptInlineStdFuncs       },
+        { "--list-opt-steps",       0,      OptListOptSteps         },
+        { "--list-warnings",        0,      OptListWarnings         },
+        { "--local-strings",        0,      OptLocalStrings         },
+        { "--memory-model",         1,      OptMemoryModel          },
+        { "--register-space",       1,      OptRegisterSpace        },
+        { "--register-vars",        0,      OptRegisterVars         },
+        { "--rodata-name",          1,      OptRodataName           },
+        { "--signed-chars",         0,      OptSignedChars          },
+        { "--standard",             1,      OptStandard             },
+        { "--static-locals",        0,      OptStaticLocals         },
+        { "--target",               1,      OptTarget               },
+        { "--verbose",              0,      OptVerbose              },
+        { "--version",              0,      OptVersion              },
+        { "--writable-strings",     0,      OptWritableStrings      },
     };
 
     unsigned I;
diff --git a/src/cc65/pragma.c b/src/cc65/pragma.c
index 86739ce22..707546e1d 100644
--- a/src/cc65/pragma.c
+++ b/src/cc65/pragma.c
@@ -64,6 +64,7 @@
 typedef enum {
     PRAGMA_ILLEGAL = -1,
     PRAGMA_ALIGN,
+    PRAGMA_ALLOW_EAGER_INLINE,
     PRAGMA_BSS_NAME,
     PRAGMA_BSSSEG,                                      /* obsolete */
     PRAGMA_CHARMAP,
@@ -74,6 +75,7 @@ typedef enum {
     PRAGMA_CODESIZE,
     PRAGMA_DATA_NAME,
     PRAGMA_DATASEG,                                     /* obsolete */
+    PRAGMA_INLINE_STDFUNCS,
     PRAGMA_LOCAL_STRINGS,
     PRAGMA_OPTIMIZE,
     PRAGMA_REGVARADDR,
@@ -96,31 +98,33 @@ static const struct Pragma {
     const char* Key;            /* Keyword */
     pragma_t    Tok;            /* Token */
 } Pragmas[PRAGMA_COUNT] = {
-    { "align",                  PRAGMA_ALIGN            },
-    { "bss-name",               PRAGMA_BSS_NAME         },
-    { "bssseg",                 PRAGMA_BSSSEG           },      /* obsolete */
-    { "charmap",                PRAGMA_CHARMAP          },
-    { "check-stack",            PRAGMA_CHECK_STACK      },
-    { "checkstack",             PRAGMA_CHECKSTACK       },      /* obsolete */
-    { "code-name",              PRAGMA_CODE_NAME        },
-    { "codeseg",                PRAGMA_CODESEG          },      /* obsolete */
-    { "codesize",               PRAGMA_CODESIZE         },
-    { "data-name",              PRAGMA_DATA_NAME        },
-    { "dataseg",                PRAGMA_DATASEG          },      /* obsolete */
-    { "local-strings",          PRAGMA_LOCAL_STRINGS    },
-    { "optimize",               PRAGMA_OPTIMIZE         },
-    { "register-vars",          PRAGMA_REGISTER_VARS    },
-    { "regvaraddr",             PRAGMA_REGVARADDR       },
-    { "regvars",                PRAGMA_REGVARS          },      /* obsolete */
-    { "rodata-name",            PRAGMA_RODATA_NAME      },
-    { "rodataseg",              PRAGMA_RODATASEG        },      /* obsolete */
-    { "signed-chars",           PRAGMA_SIGNED_CHARS     },
-    { "signedchars",            PRAGMA_SIGNEDCHARS      },      /* obsolete */
-    { "static-locals",          PRAGMA_STATIC_LOCALS    },
-    { "staticlocals",           PRAGMA_STATICLOCALS     },      /* obsolete */
-    { "warn",                   PRAGMA_WARN             },
-    { "writable-strings",       PRAGMA_WRITABLE_STRINGS },
-    { "zpsym",                  PRAGMA_ZPSYM            },
+    { "align",                  PRAGMA_ALIGN              },
+    { "allow-eager-inline",     PRAGMA_ALLOW_EAGER_INLINE },
+    { "bss-name",               PRAGMA_BSS_NAME           },
+    { "bssseg",                 PRAGMA_BSSSEG             },      /* obsolete */
+    { "charmap",                PRAGMA_CHARMAP            },
+    { "check-stack",            PRAGMA_CHECK_STACK        },
+    { "checkstack",             PRAGMA_CHECKSTACK         },      /* obsolete */
+    { "code-name",              PRAGMA_CODE_NAME          },
+    { "codeseg",                PRAGMA_CODESEG            },      /* obsolete */
+    { "codesize",               PRAGMA_CODESIZE           },
+    { "data-name",              PRAGMA_DATA_NAME          },
+    { "dataseg",                PRAGMA_DATASEG            },      /* obsolete */
+    { "inline-stdfuncs",        PRAGMA_INLINE_STDFUNCS    },
+    { "local-strings",          PRAGMA_LOCAL_STRINGS      },
+    { "optimize",               PRAGMA_OPTIMIZE           },
+    { "register-vars",          PRAGMA_REGISTER_VARS      },
+    { "regvaraddr",             PRAGMA_REGVARADDR         },
+    { "regvars",                PRAGMA_REGVARS            },      /* obsolete */
+    { "rodata-name",            PRAGMA_RODATA_NAME        },
+    { "rodataseg",              PRAGMA_RODATASEG          },      /* obsolete */
+    { "signed-chars",           PRAGMA_SIGNED_CHARS       },
+    { "signedchars",            PRAGMA_SIGNEDCHARS        },      /* obsolete */
+    { "static-locals",          PRAGMA_STATIC_LOCALS      },
+    { "staticlocals",           PRAGMA_STATICLOCALS       },      /* obsolete */
+    { "warn",                   PRAGMA_WARN               },
+    { "writable-strings",       PRAGMA_WRITABLE_STRINGS   },
+    { "zpsym",                  PRAGMA_ZPSYM              },
 };
 
 /* Result of ParsePushPop */
@@ -703,6 +707,10 @@ static void ParsePragma (void)
             IntPragma (&B, &DataAlignment, 1, 4096);
             break;
 
+        case PRAGMA_ALLOW_EAGER_INLINE:
+            FlagPragma (&B, &EagerlyInlineFuncs);
+            break;
+
         case PRAGMA_BSSSEG:
             Warning ("#pragma bssseg is obsolete, please use #pragma bss-name instead");
             /* FALLTHROUGH */
@@ -739,6 +747,10 @@ static void ParsePragma (void)
             SegNamePragma (&B, SEG_DATA);
             break;
 
+        case PRAGMA_INLINE_STDFUNCS:
+            FlagPragma (&B, &InlineStdFuncs);
+            break;
+
         case PRAGMA_LOCAL_STRINGS:
             FlagPragma (&B, &LocalStrings);
             break;
diff --git a/src/cc65/stdfunc.c b/src/cc65/stdfunc.c
index 720e6db15..7a0450146 100644
--- a/src/cc65/stdfunc.c
+++ b/src/cc65/stdfunc.c
@@ -261,262 +261,277 @@ static void StdFunc_memcpy (FuncDesc* F attribute ((unused)), ExprDesc* Expr)
         goto ExitPoint;
     }
 
-    /* We've generated the complete code for the function now and know the
-    ** types of all parameters. Check for situations where better code can
-    ** be generated. If such a situation is detected, throw away the
-    ** generated, and emit better code.
-    */
-    if (ED_IsConstAbsInt (&Arg3.Expr) && Arg3.Expr.IVal <= 256 &&
-        ((ED_IsRVal (&Arg2.Expr) && ED_IsLocConst (&Arg2.Expr)) ||
-         (ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr))) &&
-        ((ED_IsRVal (&Arg1.Expr) && ED_IsLocConst (&Arg1.Expr)) ||
-         (ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr)))) {
+    if (IS_Get (&InlineStdFuncs)) {
 
-        int Reg1 = ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr);
-        int Reg2 = ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr);
+        /* We've generated the complete code for the function now and know the
+        ** types of all parameters. Check for situations where better code can
+        ** be generated. If such a situation is detected, throw away the
+        ** generated, and emit better code.
+        */
+        if (ED_IsConstAbsInt (&Arg3.Expr) && Arg3.Expr.IVal <= 256 &&
+            ((ED_IsRVal (&Arg2.Expr) && ED_IsLocConst (&Arg2.Expr)) ||
+             (ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr))) &&
+            ((ED_IsRVal (&Arg1.Expr) && ED_IsLocConst (&Arg1.Expr)) ||
+             (ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr)))) {
 
-        /* Drop the generated code */
-        RemoveCode (&Arg1.Expr.Start);
+            int Reg1 = ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr);
+            int Reg2 = ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr);
 
-        /* We need a label */
-        Label = GetLocalLabel ();
+            /* Drop the generated code */
+            RemoveCode (&Arg1.Expr.Start);
 
-        /* Generate memcpy code */
-        if (Arg3.Expr.IVal <= 127) {
+            /* We need a label */
+            Label = GetLocalLabel ();
 
-            AddCodeLine ("ldy #$%02X", (unsigned char) (Arg3.Expr.IVal-1));
-            g_defcodelabel (Label);
-            if (Reg2) {
-                AddCodeLine ("lda (%s),y", ED_GetLabelName (&Arg2.Expr, 0));
-            } else {
-                AddCodeLine ("lda %s,y", ED_GetLabelName (&Arg2.Expr, 0));
-            }
-            if (Reg1) {
-                AddCodeLine ("sta (%s),y", ED_GetLabelName (&Arg1.Expr, 0));
-            } else {
-                AddCodeLine ("sta %s,y", ED_GetLabelName (&Arg1.Expr, 0));
-            }
-            AddCodeLine ("dey");
-            AddCodeLine ("bpl %s", LocalLabelName (Label));
+            /* Generate memcpy code */
+            if (Arg3.Expr.IVal <= 127) {
 
-        } else {
+                AddCodeLine ("ldy #$%02X", (unsigned char) (Arg3.Expr.IVal-1));
+                g_defcodelabel (Label);
+                if (Reg2) {
+                    AddCodeLine ("lda (%s),y", ED_GetLabelName (&Arg2.Expr, 0));
+                } else {
+                    AddCodeLine ("lda %s,y", ED_GetLabelName (&Arg2.Expr, 0));
+                }
+                if (Reg1) {
+                    AddCodeLine ("sta (%s),y", ED_GetLabelName (&Arg1.Expr, 0));
+                } else {
+                    AddCodeLine ("sta %s,y", ED_GetLabelName (&Arg1.Expr, 0));
+                }
+                AddCodeLine ("dey");
+                AddCodeLine ("bpl %s", LocalLabelName (Label));
 
-            AddCodeLine ("ldy #$00");
-            g_defcodelabel (Label);
-            if (Reg2) {
-                AddCodeLine ("lda (%s),y", ED_GetLabelName (&Arg2.Expr, 0));
-            } else {
-                AddCodeLine ("lda %s,y", ED_GetLabelName (&Arg2.Expr, 0));
-            }
-            if (Reg1) {
-                AddCodeLine ("sta (%s),y", ED_GetLabelName (&Arg1.Expr, 0));
             } else {
-                AddCodeLine ("sta %s,y", ED_GetLabelName (&Arg1.Expr, 0));
-            }
-            AddCodeLine ("iny");
-            AddCmpCodeIfSizeNot256 ("cpy #$%02X", Arg3.Expr.IVal);
-            AddCodeLine ("bne %s", LocalLabelName (Label));
 
-        }
+                AddCodeLine ("ldy #$00");
+                g_defcodelabel (Label);
+                if (Reg2) {
+                    AddCodeLine ("lda (%s),y", ED_GetLabelName (&Arg2.Expr, 0));
+                } else {
+                    AddCodeLine ("lda %s,y", ED_GetLabelName (&Arg2.Expr, 0));
+                }
+                if (Reg1) {
+                    AddCodeLine ("sta (%s),y", ED_GetLabelName (&Arg1.Expr, 0));
+                } else {
+                    AddCodeLine ("sta %s,y", ED_GetLabelName (&Arg1.Expr, 0));
+                }
+                AddCodeLine ("iny");
+                AddCmpCodeIfSizeNot256 ("cpy #$%02X", Arg3.Expr.IVal);
+                AddCodeLine ("bne %s", LocalLabelName (Label));
 
-        /* memcpy returns the address, so the result is actually identical
-        ** to the first argument.
-        */
-        *Expr = Arg1.Expr;
+            }
 
-    } else if (ED_IsConstAbsInt (&Arg3.Expr) && Arg3.Expr.IVal <= 256 &&
-               ED_IsRVal (&Arg2.Expr) && ED_IsLocConst (&Arg2.Expr) &&
-               ED_IsRVal (&Arg1.Expr) && ED_IsLocStack (&Arg1.Expr) &&
-               (Arg1.Expr.IVal - StackPtr) + Arg3.Expr.IVal < 256) {
-
-        /* It is possible to just use one index register even if the stack
-        ** offset is not zero, by adjusting the offset to the constant
-        ** address accordingly. But we cannot do this if the data in
-        ** question is in the register space or at an absolute address less
-        ** than 256. Register space is zero page, which means that the
-        ** address calculation could overflow in the linker.
-        */
-        int AllowOneIndex = !ED_IsLocRegister (&Arg2.Expr) &&
-                            !(ED_IsLocAbs (&Arg2.Expr) && Arg2.Expr.IVal < 256);
+            /* memcpy returns the address, so the result is actually identical
+            ** to the first argument.
+            */
+            *Expr = Arg1.Expr;
 
-        /* Calculate the real stack offset */
-        Offs = ED_GetStackOffs (&Arg1.Expr, 0);
+            /* Bail out, no need for further processing */
+            goto ExitPoint;
+        }
 
-        /* Drop the generated code */
-        RemoveCode (&Arg1.Expr.Start);
+        if (ED_IsConstAbsInt (&Arg3.Expr) && Arg3.Expr.IVal <= 256 &&
+            ED_IsRVal (&Arg2.Expr) && ED_IsLocConst (&Arg2.Expr) &&
+            ED_IsRVal (&Arg1.Expr) && ED_IsLocStack (&Arg1.Expr) &&
+            (Arg1.Expr.IVal - StackPtr) + Arg3.Expr.IVal < 256) {
+
+            /* It is possible to just use one index register even if the stack
+            ** offset is not zero, by adjusting the offset to the constant
+            ** address accordingly. But we cannot do this if the data in
+            ** question is in the register space or at an absolute address less
+            ** than 256. Register space is zero page, which means that the
+            ** address calculation could overflow in the linker.
+            */
+            int AllowOneIndex = !ED_IsLocRegister (&Arg2.Expr) &&
+                                !(ED_IsLocAbs (&Arg2.Expr) && Arg2.Expr.IVal < 256);
 
-        /* We need a label */
-        Label = GetLocalLabel ();
+            /* Calculate the real stack offset */
+            Offs = ED_GetStackOffs (&Arg1.Expr, 0);
 
-        /* Generate memcpy code */
-        if (Arg3.Expr.IVal <= 127 && !AllowOneIndex) {
+            /* Drop the generated code */
+            RemoveCode (&Arg1.Expr.Start);
+
+            /* We need a label */
+            Label = GetLocalLabel ();
+
+            /* Generate memcpy code */
+            if (Arg3.Expr.IVal <= 127 && !AllowOneIndex) {
+
+                if (Offs == 0) {
+                    AddCodeLine ("ldy #$%02X", (unsigned char) (Offs + Arg3.Expr.IVal - 1));
+                    g_defcodelabel (Label);
+                    AddCodeLine ("lda %s,y", ED_GetLabelName (&Arg2.Expr, -Offs));
+                    AddCodeLine ("sta (sp),y");
+                    AddCodeLine ("dey");
+                    AddCodeLine ("bpl %s", LocalLabelName (Label));
+                } else {
+                    AddCodeLine ("ldx #$%02X", (unsigned char) (Arg3.Expr.IVal-1));
+                    AddCodeLine ("ldy #$%02X", (unsigned char) (Offs + Arg3.Expr.IVal - 1));
+                    g_defcodelabel (Label);
+                    AddCodeLine ("lda %s,x", ED_GetLabelName (&Arg2.Expr, 0));
+                    AddCodeLine ("sta (sp),y");
+                    AddCodeLine ("dey");
+                    AddCodeLine ("dex");
+                    AddCodeLine ("bpl %s", LocalLabelName (Label));
+                }
 
-            if (Offs == 0) {
-                AddCodeLine ("ldy #$%02X", (unsigned char) (Offs + Arg3.Expr.IVal - 1));
-                g_defcodelabel (Label);
-                AddCodeLine ("lda %s,y", ED_GetLabelName (&Arg2.Expr, -Offs));
-                AddCodeLine ("sta (sp),y");
-                AddCodeLine ("dey");
-                AddCodeLine ("bpl %s", LocalLabelName (Label));
             } else {
-                AddCodeLine ("ldx #$%02X", (unsigned char) (Arg3.Expr.IVal-1));
-                AddCodeLine ("ldy #$%02X", (unsigned char) (Offs + Arg3.Expr.IVal - 1));
-                g_defcodelabel (Label);
-                AddCodeLine ("lda %s,x", ED_GetLabelName (&Arg2.Expr, 0));
-                AddCodeLine ("sta (sp),y");
-                AddCodeLine ("dey");
-                AddCodeLine ("dex");
-                AddCodeLine ("bpl %s", LocalLabelName (Label));
-            }
 
-        } else {
+                if (Offs == 0 || AllowOneIndex) {
+                    AddCodeLine ("ldy #$%02X", (unsigned char) Offs);
+                    g_defcodelabel (Label);
+                    AddCodeLine ("lda %s,y", ED_GetLabelName (&Arg2.Expr, -Offs));
+                    AddCodeLine ("sta (sp),y");
+                    AddCodeLine ("iny");
+                    AddCmpCodeIfSizeNot256 ("cpy #$%02X", Offs + Arg3.Expr.IVal);
+                    AddCodeLine ("bne %s", LocalLabelName (Label));
+                } else {
+                    AddCodeLine ("ldx #$00");
+                    AddCodeLine ("ldy #$%02X", (unsigned char) Offs);
+                    g_defcodelabel (Label);
+                    AddCodeLine ("lda %s,x", ED_GetLabelName (&Arg2.Expr, 0));
+                    AddCodeLine ("sta (sp),y");
+                    AddCodeLine ("iny");
+                    AddCodeLine ("inx");
+                    AddCmpCodeIfSizeNot256 ("cpx #$%02X", Arg3.Expr.IVal);
+                    AddCodeLine ("bne %s", LocalLabelName (Label));
+                }
 
-            if (Offs == 0 || AllowOneIndex) {
-                AddCodeLine ("ldy #$%02X", (unsigned char) Offs);
-                g_defcodelabel (Label);
-                AddCodeLine ("lda %s,y", ED_GetLabelName (&Arg2.Expr, -Offs));
-                AddCodeLine ("sta (sp),y");
-                AddCodeLine ("iny");
-                AddCmpCodeIfSizeNot256 ("cpy #$%02X", Offs + Arg3.Expr.IVal);
-                AddCodeLine ("bne %s", LocalLabelName (Label));
-            } else {
-                AddCodeLine ("ldx #$00");
-                AddCodeLine ("ldy #$%02X", (unsigned char) Offs);
-                g_defcodelabel (Label);
-                AddCodeLine ("lda %s,x", ED_GetLabelName (&Arg2.Expr, 0));
-                AddCodeLine ("sta (sp),y");
-                AddCodeLine ("iny");
-                AddCodeLine ("inx");
-                AddCmpCodeIfSizeNot256 ("cpx #$%02X", Arg3.Expr.IVal);
-                AddCodeLine ("bne %s", LocalLabelName (Label));
             }
 
+            /* memcpy returns the address, so the result is actually identical
+            ** to the first argument.
+            */
+            *Expr = Arg1.Expr;
+
+            /* Bail out, no need for further processing */
+            goto ExitPoint;
         }
 
-        /* memcpy returns the address, so the result is actually identical
-        ** to the first argument.
-        */
-        *Expr = Arg1.Expr;
+        if (ED_IsConstAbsInt (&Arg3.Expr) && Arg3.Expr.IVal <= 256 &&
+            ED_IsRVal (&Arg2.Expr) && ED_IsLocStack (&Arg2.Expr) &&
+            (Arg2.Expr.IVal - StackPtr) + Arg3.Expr.IVal < 256 &&
+            ED_IsRVal (&Arg1.Expr) && ED_IsLocConst (&Arg1.Expr)) {
+
+            /* It is possible to just use one index register even if the stack
+            ** offset is not zero, by adjusting the offset to the constant
+            ** address accordingly. But we cannot do this if the data in
+            ** question is in the register space or at an absolute address less
+            ** than 256. Register space is zero page, which means that the
+            ** address calculation could overflow in the linker.
+            */
+            int AllowOneIndex = !ED_IsLocRegister (&Arg1.Expr) &&
+                                !(ED_IsLocAbs (&Arg1.Expr) && Arg1.Expr.IVal < 256);
 
-    } else if (ED_IsConstAbsInt (&Arg3.Expr) && Arg3.Expr.IVal <= 256 &&
-               ED_IsRVal (&Arg2.Expr) && ED_IsLocStack (&Arg2.Expr) &&
-               (Arg2.Expr.IVal - StackPtr) + Arg3.Expr.IVal < 256 &&
-               ED_IsRVal (&Arg1.Expr) && ED_IsLocConst (&Arg1.Expr)) {
-
-        /* It is possible to just use one index register even if the stack
-        ** offset is not zero, by adjusting the offset to the constant
-        ** address accordingly. But we cannot do this if the data in
-        ** question is in the register space or at an absolute address less
-        ** than 256. Register space is zero page, which means that the
-        ** address calculation could overflow in the linker.
-        */
-        int AllowOneIndex = !ED_IsLocRegister (&Arg1.Expr) &&
-                            !(ED_IsLocAbs (&Arg1.Expr) && Arg1.Expr.IVal < 256);
+            /* Calculate the real stack offset */
+            Offs = ED_GetStackOffs (&Arg2.Expr, 0);
+
+            /* Drop the generated code */
+            RemoveCode (&Arg1.Expr.Start);
+
+            /* We need a label */
+            Label = GetLocalLabel ();
+
+            /* Generate memcpy code */
+            if (Arg3.Expr.IVal <= 127 && !AllowOneIndex) {
+
+                if (Offs == 0) {
+                    AddCodeLine ("ldy #$%02X", (unsigned char) (Arg3.Expr.IVal - 1));
+                    g_defcodelabel (Label);
+                    AddCodeLine ("lda (sp),y");
+                    AddCodeLine ("sta %s,y", ED_GetLabelName (&Arg1.Expr, 0));
+                    AddCodeLine ("dey");
+                    AddCodeLine ("bpl %s", LocalLabelName (Label));
+                } else {
+                    AddCodeLine ("ldx #$%02X", (unsigned char) (Arg3.Expr.IVal-1));
+                    AddCodeLine ("ldy #$%02X", (unsigned char) (Offs + Arg3.Expr.IVal - 1));
+                    g_defcodelabel (Label);
+                    AddCodeLine ("lda (sp),y");
+                    AddCodeLine ("sta %s,x", ED_GetLabelName (&Arg1.Expr, 0));
+                    AddCodeLine ("dey");
+                    AddCodeLine ("dex");
+                    AddCodeLine ("bpl %s", LocalLabelName (Label));
+                }
+
+            } else {
+
+                if (Offs == 0 || AllowOneIndex) {
+                    AddCodeLine ("ldy #$%02X", (unsigned char) Offs);
+                    g_defcodelabel (Label);
+                    AddCodeLine ("lda (sp),y");
+                    AddCodeLine ("sta %s,y", ED_GetLabelName (&Arg1.Expr, -Offs));
+                    AddCodeLine ("iny");
+                    AddCmpCodeIfSizeNot256 ("cpy #$%02X", Offs + Arg3.Expr.IVal);
+                    AddCodeLine ("bne %s", LocalLabelName (Label));
+                } else {
+                    AddCodeLine ("ldx #$00");
+                    AddCodeLine ("ldy #$%02X", (unsigned char) Offs);
+                    g_defcodelabel (Label);
+                    AddCodeLine ("lda (sp),y");
+                    AddCodeLine ("sta %s,x", ED_GetLabelName (&Arg1.Expr, 0));
+                    AddCodeLine ("iny");
+                    AddCodeLine ("inx");
+                    AddCmpCodeIfSizeNot256 ("cpx #$%02X", Arg3.Expr.IVal);
+                    AddCodeLine ("bne %s", LocalLabelName (Label));
+                }
 
-        /* Calculate the real stack offset */
-        Offs = ED_GetStackOffs (&Arg2.Expr, 0);
+            }
 
-        /* Drop the generated code */
-        RemoveCode (&Arg1.Expr.Start);
+            /* memcpy returns the address, so the result is actually identical
+            ** to the first argument.
+            */
+            *Expr = Arg1.Expr;
 
-        /* We need a label */
-        Label = GetLocalLabel ();
+            /* Bail out, no need for further processing */
+            goto ExitPoint;
+        }
 
-        /* Generate memcpy code */
-        if (Arg3.Expr.IVal <= 127 && !AllowOneIndex) {
+        if (ED_IsConstAbsInt (&Arg3.Expr) && Arg3.Expr.IVal <= 256   &&
+            ED_IsRVal (&Arg2.Expr) && ED_IsLocStack (&Arg2.Expr)     &&
+            (Offs = ED_GetStackOffs (&Arg2.Expr, 0)) == 0) {
 
-            if (Offs == 0) {
+            /* Drop the generated code but leave the load of the first argument*/
+            RemoveCode (&Arg1.Push);
+
+            /* We need a label */
+            Label = GetLocalLabel ();
+
+            /* Generate memcpy code */
+            AddCodeLine ("sta ptr1");
+            AddCodeLine ("stx ptr1+1");
+            if (Arg3.Expr.IVal <= 127) {
                 AddCodeLine ("ldy #$%02X", (unsigned char) (Arg3.Expr.IVal - 1));
                 g_defcodelabel (Label);
                 AddCodeLine ("lda (sp),y");
-                AddCodeLine ("sta %s,y", ED_GetLabelName (&Arg1.Expr, 0));
+                AddCodeLine ("sta (ptr1),y");
                 AddCodeLine ("dey");
                 AddCodeLine ("bpl %s", LocalLabelName (Label));
             } else {
-                AddCodeLine ("ldx #$%02X", (unsigned char) (Arg3.Expr.IVal-1));
-                AddCodeLine ("ldy #$%02X", (unsigned char) (Offs + Arg3.Expr.IVal - 1));
+                AddCodeLine ("ldy #$00");
                 g_defcodelabel (Label);
                 AddCodeLine ("lda (sp),y");
-                AddCodeLine ("sta %s,x", ED_GetLabelName (&Arg1.Expr, 0));
-                AddCodeLine ("dey");
-                AddCodeLine ("dex");
-                AddCodeLine ("bpl %s", LocalLabelName (Label));
-            }
-
-        } else {
-
-            if (Offs == 0 || AllowOneIndex) {
-                AddCodeLine ("ldy #$%02X", (unsigned char) Offs);
-                g_defcodelabel (Label);
-                AddCodeLine ("lda (sp),y");
-                AddCodeLine ("sta %s,y", ED_GetLabelName (&Arg1.Expr, -Offs));
+                AddCodeLine ("sta (ptr1),y");
                 AddCodeLine ("iny");
-                AddCmpCodeIfSizeNot256 ("cpy #$%02X", Offs + Arg3.Expr.IVal);
-                AddCodeLine ("bne %s", LocalLabelName (Label));
-            } else {
-                AddCodeLine ("ldx #$00");
-                AddCodeLine ("ldy #$%02X", (unsigned char) Offs);
-                g_defcodelabel (Label);
-                AddCodeLine ("lda (sp),y");
-                AddCodeLine ("sta %s,x", ED_GetLabelName (&Arg1.Expr, 0));
-                AddCodeLine ("iny");
-                AddCodeLine ("inx");
-                AddCmpCodeIfSizeNot256 ("cpx #$%02X", Arg3.Expr.IVal);
+                AddCmpCodeIfSizeNot256 ("cpy #$%02X", Arg3.Expr.IVal);
                 AddCodeLine ("bne %s", LocalLabelName (Label));
             }
 
-        }
+            /* Reload result - X hasn't changed by the code above */
+            AddCodeLine ("lda ptr1");
 
-        /* memcpy returns the address, so the result is actually identical
-        ** to the first argument.
-        */
-        *Expr = Arg1.Expr;
-
-    } else if (ED_IsConstAbsInt (&Arg3.Expr) && Arg3.Expr.IVal <= 256   &&
-               ED_IsRVal (&Arg2.Expr) && ED_IsLocStack (&Arg2.Expr)     &&
-               (Offs = ED_GetStackOffs (&Arg2.Expr, 0)) == 0) {
-
-        /* Drop the generated code but leave the load of the first argument*/
-        RemoveCode (&Arg1.Push);
+            /* The function result is an rvalue in the primary register */
+            ED_MakeRValExpr (Expr);
+            Expr->Type = GetFuncReturn (Expr->Type);
 
-        /* We need a label */
-        Label = GetLocalLabel ();
-
-        /* Generate memcpy code */
-        AddCodeLine ("sta ptr1");
-        AddCodeLine ("stx ptr1+1");
-        if (Arg3.Expr.IVal <= 127) {
-            AddCodeLine ("ldy #$%02X", (unsigned char) (Arg3.Expr.IVal - 1));
-            g_defcodelabel (Label);
-            AddCodeLine ("lda (sp),y");
-            AddCodeLine ("sta (ptr1),y");
-            AddCodeLine ("dey");
-            AddCodeLine ("bpl %s", LocalLabelName (Label));
-        } else {
-            AddCodeLine ("ldy #$00");
-            g_defcodelabel (Label);
-            AddCodeLine ("lda (sp),y");
-            AddCodeLine ("sta (ptr1),y");
-            AddCodeLine ("iny");
-            AddCmpCodeIfSizeNot256 ("cpy #$%02X", Arg3.Expr.IVal);
-            AddCodeLine ("bne %s", LocalLabelName (Label));
+            /* Bail out, no need for further processing */
+            goto ExitPoint;
         }
-
-        /* Reload result - X hasn't changed by the code above */
-        AddCodeLine ("lda ptr1");
-
-        /* The function result is an rvalue in the primary register */
-        ED_MakeRValExpr (Expr);
-        Expr->Type = GetFuncReturn (Expr->Type);
-
-    } else {
-
-        /* The function result is an rvalue in the primary register */
-        ED_MakeRValExpr (Expr);
-        Expr->Type = GetFuncReturn (Expr->Type);
-
     }
 
+    /* The function result is an rvalue in the primary register */
+    ED_MakeRValExpr (Expr);
+    Expr->Type = GetFuncReturn (Expr->Type);
+
 ExitPoint:
     /* We expect the closing brace */
     ConsumeRParen ();
@@ -595,140 +610,151 @@ static void StdFunc_memset (FuncDesc* F attribute ((unused)), ExprDesc* Expr)
         goto ExitPoint;
     }
 
-    /* We've generated the complete code for the function now and know the
-    ** types of all parameters. Check for situations where better code can
-    ** be generated. If such a situation is detected, throw away the
-    ** generated, and emit better code.
-    ** Note: Lots of improvements would be possible here, but I will
-    ** concentrate on the most common case: memset with arguments 2 and 3
-    ** being constant numerical values. Some checks have shown that this
-    ** covers nearly 90% of all memset calls.
-    */
-    if (ED_IsConstAbsInt (&Arg3.Expr) && Arg3.Expr.IVal <= 256 &&
-        ED_IsConstAbsInt (&Arg2.Expr) &&
-        ((ED_IsRVal (&Arg1.Expr) && ED_IsLocConst (&Arg1.Expr)) ||
-         (ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr)))) {
+    if (IS_Get (&InlineStdFuncs)) {
 
-        int Reg = ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr);
+        /* We've generated the complete code for the function now and know the
+        ** types of all parameters. Check for situations where better code can
+        ** be generated. If such a situation is detected, throw away the
+        ** generated, and emit better code.
+        ** Note: Lots of improvements would be possible here, but I will
+        ** concentrate on the most common case: memset with arguments 2 and 3
+        ** being constant numerical values. Some checks have shown that this
+        ** covers nearly 90% of all memset calls.
+        */
+        if (ED_IsConstAbsInt (&Arg3.Expr) && Arg3.Expr.IVal <= 256 &&
+            ED_IsConstAbsInt (&Arg2.Expr) &&
+            ((ED_IsRVal (&Arg1.Expr) && ED_IsLocConst (&Arg1.Expr)) ||
+             (ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr)))) {
 
-        /* Drop the generated code */
-        RemoveCode (&Arg1.Expr.Start);
+            int Reg = ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr);
 
-        /* We need a label */
-        Label = GetLocalLabel ();
+            /* Drop the generated code */
+            RemoveCode (&Arg1.Expr.Start);
 
-        /* Generate memset code */
-        if (Arg3.Expr.IVal <= 127) {
+            /* We need a label */
+            Label = GetLocalLabel ();
 
-            AddCodeLine ("ldy #$%02X", (unsigned char) (Arg3.Expr.IVal-1));
-            AddCodeLine ("lda #$%02X", (unsigned char) Arg2.Expr.IVal);
-            g_defcodelabel (Label);
-            if (Reg) {
-                AddCodeLine ("sta (%s),y", ED_GetLabelName (&Arg1.Expr, 0));
-            } else {
-                AddCodeLine ("sta %s,y", ED_GetLabelName (&Arg1.Expr, 0));
-            }
-            AddCodeLine ("dey");
-            AddCodeLine ("bpl %s", LocalLabelName (Label));
+            /* Generate memset code */
+            if (Arg3.Expr.IVal <= 127) {
 
-        } else {
+                AddCodeLine ("ldy #$%02X", (unsigned char) (Arg3.Expr.IVal-1));
+                AddCodeLine ("lda #$%02X", (unsigned char) Arg2.Expr.IVal);
+                g_defcodelabel (Label);
+                if (Reg) {
+                    AddCodeLine ("sta (%s),y", ED_GetLabelName (&Arg1.Expr, 0));
+                } else {
+                    AddCodeLine ("sta %s,y", ED_GetLabelName (&Arg1.Expr, 0));
+                }
+                AddCodeLine ("dey");
+                AddCodeLine ("bpl %s", LocalLabelName (Label));
 
-            AddCodeLine ("ldy #$00");
-            AddCodeLine ("lda #$%02X", (unsigned char) Arg2.Expr.IVal);
-            g_defcodelabel (Label);
-            if (Reg) {
-                AddCodeLine ("sta (%s),y", ED_GetLabelName (&Arg1.Expr, 0));
             } else {
-                AddCodeLine ("sta %s,y", ED_GetLabelName (&Arg1.Expr, 0));
-            }
-            AddCodeLine ("iny");
-            AddCmpCodeIfSizeNot256 ("cpy #$%02X", Arg3.Expr.IVal);
-            AddCodeLine ("bne %s", LocalLabelName (Label));
 
-        }
-
-        /* memset returns the address, so the result is actually identical
-        ** to the first argument.
-        */
-        *Expr = Arg1.Expr;
-
-    } else if (ED_IsConstAbsInt (&Arg3.Expr) && Arg3.Expr.IVal <= 256 &&
-               ED_IsConstAbsInt (&Arg2.Expr) &&
-               ED_IsRVal (&Arg1.Expr) && ED_IsLocStack (&Arg1.Expr) &&
-               (Arg1.Expr.IVal - StackPtr) + Arg3.Expr.IVal < 256) {
-
-        /* Calculate the real stack offset */
-        int Offs = ED_GetStackOffs (&Arg1.Expr, 0);
+                AddCodeLine ("ldy #$00");
+                AddCodeLine ("lda #$%02X", (unsigned char) Arg2.Expr.IVal);
+                g_defcodelabel (Label);
+                if (Reg) {
+                    AddCodeLine ("sta (%s),y", ED_GetLabelName (&Arg1.Expr, 0));
+                } else {
+                    AddCodeLine ("sta %s,y", ED_GetLabelName (&Arg1.Expr, 0));
+                }
+                AddCodeLine ("iny");
+                AddCmpCodeIfSizeNot256 ("cpy #$%02X", Arg3.Expr.IVal);
+                AddCodeLine ("bne %s", LocalLabelName (Label));
 
-        /* Drop the generated code */
-        RemoveCode (&Arg1.Expr.Start);
+            }
 
-        /* We need a label */
-        Label = GetLocalLabel ();
+            /* memset returns the address, so the result is actually identical
+            ** to the first argument.
+            */
+            *Expr = Arg1.Expr;
 
-        /* Generate memset code */
-        AddCodeLine ("ldy #$%02X", (unsigned char) Offs);
-        AddCodeLine ("lda #$%02X", (unsigned char) Arg2.Expr.IVal);
-        g_defcodelabel (Label);
-        AddCodeLine ("sta (sp),y");
-        AddCodeLine ("iny");
-        AddCmpCodeIfSizeNot256 ("cpy #$%02X", Offs + Arg3.Expr.IVal);
-        AddCodeLine ("bne %s", LocalLabelName (Label));
+            /* Bail out, no need for further processing */
+            goto ExitPoint;
+        }
 
-        /* memset returns the address, so the result is actually identical
-        ** to the first argument.
-        */
-        *Expr = Arg1.Expr;
+        if (ED_IsConstAbsInt (&Arg3.Expr) && Arg3.Expr.IVal <= 256 &&
+            ED_IsConstAbsInt (&Arg2.Expr) &&
+            ED_IsRVal (&Arg1.Expr) && ED_IsLocStack (&Arg1.Expr) &&
+            (Arg1.Expr.IVal - StackPtr) + Arg3.Expr.IVal < 256) {
 
-    } else if (ED_IsConstAbsInt (&Arg3.Expr) && Arg3.Expr.IVal <= 256 &&
-               ED_IsConstAbsInt (&Arg2.Expr) &&
-               (Arg2.Expr.IVal != 0 || IS_Get (&CodeSizeFactor) > 200)) {
+            /* Calculate the real stack offset */
+            int Offs = ED_GetStackOffs (&Arg1.Expr, 0);
 
-        /* Remove all of the generated code but the load of the first
-        ** argument.
-        */
-        RemoveCode (&Arg1.Push);
+            /* Drop the generated code */
+            RemoveCode (&Arg1.Expr.Start);
 
-        /* We need a label */
-        Label = GetLocalLabel ();
+            /* We need a label */
+            Label = GetLocalLabel ();
 
-        /* Generate code */
-        AddCodeLine ("sta ptr1");
-        AddCodeLine ("stx ptr1+1");
-        if (Arg3.Expr.IVal <= 127) {
-            AddCodeLine ("ldy #$%02X", (unsigned char) (Arg3.Expr.IVal-1));
+            /* Generate memset code */
+            AddCodeLine ("ldy #$%02X", (unsigned char) Offs);
             AddCodeLine ("lda #$%02X", (unsigned char) Arg2.Expr.IVal);
             g_defcodelabel (Label);
-            AddCodeLine ("sta (ptr1),y");
-            AddCodeLine ("dey");
-            AddCodeLine ("bpl %s", LocalLabelName (Label));
-        } else {
-            AddCodeLine ("ldy #$00");
-            AddCodeLine ("lda #$%02X", (unsigned char) Arg2.Expr.IVal);
-            g_defcodelabel (Label);
-            AddCodeLine ("sta (ptr1),y");
+            AddCodeLine ("sta (sp),y");
             AddCodeLine ("iny");
-            AddCmpCodeIfSizeNot256 ("cpy #$%02X", Arg3.Expr.IVal);
+            AddCmpCodeIfSizeNot256 ("cpy #$%02X", Offs + Arg3.Expr.IVal);
             AddCodeLine ("bne %s", LocalLabelName (Label));
+
+            /* memset returns the address, so the result is actually identical
+            ** to the first argument.
+            */
+            *Expr = Arg1.Expr;
+
+            /* Bail out, no need for further processing */
+            goto ExitPoint;
         }
 
-        /* Load the function result pointer into a/x (x is still valid). This
-        ** code will get removed by the optimizer if it is not used later.
-        */
-        AddCodeLine ("lda ptr1");
+        if (ED_IsConstAbsInt (&Arg3.Expr) && Arg3.Expr.IVal <= 256 &&
+            ED_IsConstAbsInt (&Arg2.Expr) &&
+            (Arg2.Expr.IVal != 0 || IS_Get (&CodeSizeFactor) > 200)) {
 
-        /* The function result is an rvalue in the primary register */
-        ED_MakeRValExpr (Expr);
-        Expr->Type = GetFuncReturn (Expr->Type);
+            /* Remove all of the generated code but the load of the first
+            ** argument.
+            */
+            RemoveCode (&Arg1.Push);
 
-    } else {
+            /* We need a label */
+            Label = GetLocalLabel ();
+
+            /* Generate code */
+            AddCodeLine ("sta ptr1");
+            AddCodeLine ("stx ptr1+1");
+            if (Arg3.Expr.IVal <= 127) {
+                AddCodeLine ("ldy #$%02X", (unsigned char) (Arg3.Expr.IVal-1));
+                AddCodeLine ("lda #$%02X", (unsigned char) Arg2.Expr.IVal);
+                g_defcodelabel (Label);
+                AddCodeLine ("sta (ptr1),y");
+                AddCodeLine ("dey");
+                AddCodeLine ("bpl %s", LocalLabelName (Label));
+            } else {
+                AddCodeLine ("ldy #$00");
+                AddCodeLine ("lda #$%02X", (unsigned char) Arg2.Expr.IVal);
+                g_defcodelabel (Label);
+                AddCodeLine ("sta (ptr1),y");
+                AddCodeLine ("iny");
+                AddCmpCodeIfSizeNot256 ("cpy #$%02X", Arg3.Expr.IVal);
+                AddCodeLine ("bne %s", LocalLabelName (Label));
+            }
 
-        /* The function result is an rvalue in the primary register */
-        ED_MakeRValExpr (Expr);
-        Expr->Type = GetFuncReturn (Expr->Type);
+            /* Load the function result pointer into a/x (x is still valid). This
+            ** code will get removed by the optimizer if it is not used later.
+            */
+            AddCodeLine ("lda ptr1");
+
+            /* The function result is an rvalue in the primary register */
+            ED_MakeRValExpr (Expr);
+            Expr->Type = GetFuncReturn (Expr->Type);
 
+            /* Bail out, no need for further processing */
+            goto ExitPoint;
+        }
     }
 
+    /* The function result is an rvalue in the primary register */
+    ED_MakeRValExpr (Expr);
+    Expr->Type = GetFuncReturn (Expr->Type);
+
 ExitPoint:
     /* We expect the closing brace */
     ConsumeRParen ();
@@ -790,142 +816,142 @@ static void StdFunc_strcmp (FuncDesc* F attribute ((unused)), ExprDesc* Expr)
         ECount1 = ECount2;
     }
 
-    /* If the second argument is the empty string literal, we can generate
-    ** more efficient code.
-    */
-    if (ED_IsLocLiteral (&Arg2.Expr) &&
-        IS_Get (&WritableStrings) == 0 &&
-        GetLiteralSize (Arg2.Expr.LVal) == 1 &&
-        GetLiteralStr (Arg2.Expr.LVal)[0] == '\0') {
+    if (IS_Get (&InlineStdFuncs)) {
 
-        /* Drop the generated code so we have the first argument in the
-        ** primary
+        /* If the second argument is the empty string literal, we can generate
+        ** more efficient code.
         */
-        RemoveCode (&Arg1.Push);
+        if (ED_IsLocLiteral (&Arg2.Expr) &&
+            IS_Get (&WritableStrings) == 0 &&
+            GetLiteralSize (Arg2.Expr.LVal) == 1 &&
+            GetLiteralStr (Arg2.Expr.LVal)[0] == '\0') {
 
-        /* We don't need the literal any longer */
-        ReleaseLiteral (Arg2.Expr.LVal);
-
-        /* We do now have Arg1 in the primary. Load the first character from
-        ** this string and cast to int. This is the function result.
-        */
-        IsArray = IsTypeArray (Arg1.Type) && ED_IsRVal (&Arg1.Expr);
-        if (IsArray && ED_IsLocStack (&Arg1.Expr) &&
-            (Offs = ED_GetStackOffs (&Arg1.Expr, 0) < 256)) {
-            /* Drop the generated code */
-            RemoveCode (&Arg1.Load);
+            /* Drop the generated code so we have the first argument in the
+            ** primary
+            */
+            RemoveCode (&Arg1.Push);
 
-            /* Generate code */
-            AddCodeLine ("ldy #$%02X", Offs);
-            AddCodeLine ("ldx #$00");
-            AddCodeLine ("lda (sp),y");
-        } else if (IsArray && ED_IsLocConst (&Arg1.Expr)) {
-            /* Drop the generated code */
-            RemoveCode (&Arg1.Load);
+            /* We don't need the literal any longer */
+            ReleaseLiteral (Arg2.Expr.LVal);
 
-            /* Generate code */
-            AddCodeLine ("ldx #$00");
-            AddCodeLine ("lda %s", ED_GetLabelName (&Arg1.Expr, 0));
-        } else {
-            /* Drop part of the generated code so we have the first argument
-            ** in the primary
+            /* We do now have Arg1 in the primary. Load the first character from
+            ** this string and cast to int. This is the function result.
             */
-            RemoveCode (&Arg1.Push);
+            IsArray = IsTypeArray (Arg1.Type) && ED_IsRVal (&Arg1.Expr);
+            if (IsArray && ED_IsLocStack (&Arg1.Expr) &&
+                (Offs = ED_GetStackOffs (&Arg1.Expr, 0) < 256)) {
+                /* Drop the generated code */
+                RemoveCode (&Arg1.Load);
+
+                /* Generate code */
+                AddCodeLine ("ldy #$%02X", Offs);
+                AddCodeLine ("ldx #$00");
+                AddCodeLine ("lda (sp),y");
+            } else if (IsArray && ED_IsLocConst (&Arg1.Expr)) {
+                /* Drop the generated code */
+                RemoveCode (&Arg1.Load);
 
-            /* Fetch the first char */
-            g_getind (CF_CHAR | CF_UNSIGNED, 0);
-        }
+                /* Generate code */
+                AddCodeLine ("ldx #$00");
+                AddCodeLine ("lda %s", ED_GetLabelName (&Arg1.Expr, 0));
+            } else {
+                /* Drop part of the generated code so we have the first argument
+                ** in the primary
+                */
+                RemoveCode (&Arg1.Push);
 
-    } else if ((IS_Get (&CodeSizeFactor) >= 165) &&
-               ((ED_IsRVal (&Arg2.Expr) && ED_IsLocConst (&Arg2.Expr)) ||
-                (ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr))) &&
-               ((ED_IsRVal (&Arg1.Expr) && ED_IsLocConst (&Arg1.Expr)) ||
-                (ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr))) &&
-               (IS_Get (&InlineStdFuncs) || (ECount1 > 0 && ECount1 < 256))) {
+                /* Fetch the first char */
+                g_getind (CF_CHAR | CF_UNSIGNED, 0);
+            }
 
+        } else if ((IS_Get (&CodeSizeFactor) >= 165) &&
+                   ((ED_IsRVal (&Arg2.Expr) && ED_IsLocConst (&Arg2.Expr)) ||
+                    (ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr))) &&
+                   ((ED_IsRVal (&Arg1.Expr) && ED_IsLocConst (&Arg1.Expr)) ||
+                    (ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr))) &&
+                   (IS_Get (&EagerlyInlineFuncs) || (ECount1 > 0 && ECount1 < 256))) {
 
-        unsigned    Entry, Loop, Fin;   /* Labels */
-        const char* Load;
-        const char* Compare;
+            unsigned    Entry, Loop, Fin;   /* Labels */
+            const char* Load;
+            const char* Compare;
 
-        if (ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr)) {
-            Load = "lda (%s),y";
-        } else {
-            Load = "lda %s,y";
-        }
-        if (ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr)) {
-            Compare = "cmp (%s),y";
-        } else {
-            Compare = "cmp %s,y";
-        }
+            if (ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr)) {
+                Load = "lda (%s),y";
+            } else {
+                Load = "lda %s,y";
+            }
+            if (ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr)) {
+                Compare = "cmp (%s),y";
+            } else {
+                Compare = "cmp %s,y";
+            }
 
-        /* Drop the generated code */
-        RemoveCode (&Arg1.Expr.Start);
-
-        /* We need labels */
-        Entry = GetLocalLabel ();
-        Loop  = GetLocalLabel ();
-        Fin   = GetLocalLabel ();
-
-        /* Generate strcmp code */
-        AddCodeLine ("ldy #$00");
-        AddCodeLine ("beq %s", LocalLabelName (Entry));
-        g_defcodelabel (Loop);
-        AddCodeLine ("tax");
-        AddCodeLine ("beq %s", LocalLabelName (Fin));
-        AddCodeLine ("iny");
-        g_defcodelabel (Entry);
-        AddCodeLine (Load, ED_GetLabelName (&Arg1.Expr, 0));
-        AddCodeLine (Compare, ED_GetLabelName (&Arg2.Expr, 0));
-        AddCodeLine ("beq %s", LocalLabelName (Loop));
-        AddCodeLine ("ldx #$01");
-        AddCodeLine ("bcs %s", LocalLabelName (Fin));
-        AddCodeLine ("ldx #$FF");
-        g_defcodelabel (Fin);
-
-    } else if ((IS_Get (&CodeSizeFactor) > 190) &&
-               ((ED_IsRVal (&Arg2.Expr) && ED_IsLocConst (&Arg2.Expr)) ||
-                (ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr))) &&
-               (IS_Get (&InlineStdFuncs) || (ECount1 > 0 && ECount1 < 256))) {
-
-
-        unsigned    Entry, Loop, Fin;   /* Labels */
-        const char* Compare;
-
-        if (ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr)) {
-            Compare = "cmp (%s),y";
-        } else {
-            Compare = "cmp %s,y";
-        }
+            /* Drop the generated code */
+            RemoveCode (&Arg1.Expr.Start);
 
-        /* Drop the generated code */
-        RemoveCode (&Arg1.Push);
+            /* We need labels */
+            Entry = GetLocalLabel ();
+            Loop  = GetLocalLabel ();
+            Fin   = GetLocalLabel ();
+
+            /* Generate strcmp code */
+            AddCodeLine ("ldy #$00");
+            AddCodeLine ("beq %s", LocalLabelName (Entry));
+            g_defcodelabel (Loop);
+            AddCodeLine ("tax");
+            AddCodeLine ("beq %s", LocalLabelName (Fin));
+            AddCodeLine ("iny");
+            g_defcodelabel (Entry);
+            AddCodeLine (Load, ED_GetLabelName (&Arg1.Expr, 0));
+            AddCodeLine (Compare, ED_GetLabelName (&Arg2.Expr, 0));
+            AddCodeLine ("beq %s", LocalLabelName (Loop));
+            AddCodeLine ("ldx #$01");
+            AddCodeLine ("bcs %s", LocalLabelName (Fin));
+            AddCodeLine ("ldx #$FF");
+            g_defcodelabel (Fin);
+
+        } else if ((IS_Get (&CodeSizeFactor) > 190) &&
+                   ((ED_IsRVal (&Arg2.Expr) && ED_IsLocConst (&Arg2.Expr)) ||
+                    (ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr))) &&
+                   (IS_Get (&EagerlyInlineFuncs) || (ECount1 > 0 && ECount1 < 256))) {
 
-        /* We need labels */
-        Entry = GetLocalLabel ();
-        Loop  = GetLocalLabel ();
-        Fin   = GetLocalLabel ();
-
-        /* Store Arg1 into ptr1 */
-        AddCodeLine ("sta ptr1");
-        AddCodeLine ("stx ptr1+1");
-
-        /* Generate strcmp code */
-        AddCodeLine ("ldy #$00");
-        AddCodeLine ("beq %s", LocalLabelName (Entry));
-        g_defcodelabel (Loop);
-        AddCodeLine ("tax");
-        AddCodeLine ("beq %s", LocalLabelName (Fin));
-        AddCodeLine ("iny");
-        g_defcodelabel (Entry);
-        AddCodeLine ("lda (ptr1),y");
-        AddCodeLine (Compare, ED_GetLabelName (&Arg2.Expr, 0));
-        AddCodeLine ("beq %s", LocalLabelName (Loop));
-        AddCodeLine ("ldx #$01");
-        AddCodeLine ("bcs %s", LocalLabelName (Fin));
-        AddCodeLine ("ldx #$FF");
-        g_defcodelabel (Fin);
+            unsigned    Entry, Loop, Fin;   /* Labels */
+            const char* Compare;
 
+            if (ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr)) {
+                Compare = "cmp (%s),y";
+            } else {
+                Compare = "cmp %s,y";
+            }
+
+            /* Drop the generated code */
+            RemoveCode (&Arg1.Push);
+
+            /* We need labels */
+            Entry = GetLocalLabel ();
+            Loop  = GetLocalLabel ();
+            Fin   = GetLocalLabel ();
+
+            /* Store Arg1 into ptr1 */
+            AddCodeLine ("sta ptr1");
+            AddCodeLine ("stx ptr1+1");
+
+            /* Generate strcmp code */
+            AddCodeLine ("ldy #$00");
+            AddCodeLine ("beq %s", LocalLabelName (Entry));
+            g_defcodelabel (Loop);
+            AddCodeLine ("tax");
+            AddCodeLine ("beq %s", LocalLabelName (Fin));
+            AddCodeLine ("iny");
+            g_defcodelabel (Entry);
+            AddCodeLine ("lda (ptr1),y");
+            AddCodeLine (Compare, ED_GetLabelName (&Arg2.Expr, 0));
+            AddCodeLine ("beq %s", LocalLabelName (Loop));
+            AddCodeLine ("ldx #$01");
+            AddCodeLine ("bcs %s", LocalLabelName (Fin));
+            AddCodeLine ("ldx #$FF");
+            g_defcodelabel (Fin);
+        }
     }
 
     /* The function result is an rvalue in the primary register */
@@ -983,142 +1009,154 @@ static void StdFunc_strcpy (FuncDesc* F attribute ((unused)), ExprDesc* Expr)
     /* Get the element count of argument 1 if it is an array */
     ECount = ArrayElementCount (&Arg1);
 
-    /* We've generated the complete code for the function now and know the
-    ** types of all parameters. Check for situations where better code can
-    ** be generated. If such a situation is detected, throw away the
-    ** generated, and emit better code.
-    */
-    if (((ED_IsRVal (&Arg2.Expr) && ED_IsLocConst (&Arg2.Expr)) ||
-         (ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr))) &&
-        ((ED_IsRVal (&Arg1.Expr) && ED_IsLocConst (&Arg1.Expr)) ||
-         (ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr))) &&
-        (IS_Get (&InlineStdFuncs) ||
-        (ECount != UNSPECIFIED && ECount < 256))) {
-
-        const char* Load;
-        const char* Store;
-        if (ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr)) {
-            Load = "lda (%s),y";
-        } else {
-            Load = "lda %s,y";
-        }
-        if (ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr)) {
-            Store = "sta (%s),y";
-        } else {
-            Store = "sta %s,y";
-        }
+    if (IS_Get (&InlineStdFuncs)) {
 
-        /* Drop the generated code */
-        RemoveCode (&Arg1.Expr.Start);
+        /* We've generated the complete code for the function now and know the
+        ** types of all parameters. Check for situations where better code can
+        ** be generated. If such a situation is detected, throw away the
+        ** generated, and emit better code.
+        */
+        if (((ED_IsRVal (&Arg2.Expr) && ED_IsLocConst (&Arg2.Expr)) ||
+             (ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr))) &&
+            ((ED_IsRVal (&Arg1.Expr) && ED_IsLocConst (&Arg1.Expr)) ||
+             (ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr))) &&
+            (IS_Get (&EagerlyInlineFuncs) ||
+            (ECount != UNSPECIFIED && ECount < 256))) {
+
+            const char* Load;
+            const char* Store;
+            if (ED_IsLVal (&Arg2.Expr) && ED_IsLocRegister (&Arg2.Expr)) {
+                Load = "lda (%s),y";
+            } else {
+                Load = "lda %s,y";
+            }
+            if (ED_IsLVal (&Arg1.Expr) && ED_IsLocRegister (&Arg1.Expr)) {
+                Store = "sta (%s),y";
+            } else {
+                Store = "sta %s,y";
+            }
 
-        /* We need labels */
-        L1 = GetLocalLabel ();
+            /* Drop the generated code */
+            RemoveCode (&Arg1.Expr.Start);
 
-        /* Generate strcpy code */
-        AddCodeLine ("ldy #$FF");
-        g_defcodelabel (L1);
-        AddCodeLine ("iny");
-        AddCodeLine (Load, ED_GetLabelName (&Arg2.Expr, 0));
-        AddCodeLine (Store, ED_GetLabelName (&Arg1.Expr, 0));
-        AddCodeLine ("bne %s", LocalLabelName (L1));
+            /* We need labels */
+            L1 = GetLocalLabel ();
 
-        /* strcpy returns argument #1 */
-        *Expr = Arg1.Expr;
+            /* Generate strcpy code */
+            AddCodeLine ("ldy #$FF");
+            g_defcodelabel (L1);
+            AddCodeLine ("iny");
+            AddCodeLine (Load, ED_GetLabelName (&Arg2.Expr, 0));
+            AddCodeLine (Store, ED_GetLabelName (&Arg1.Expr, 0));
+            AddCodeLine ("bne %s", LocalLabelName (L1));
 
-    } else if (ED_IsRVal (&Arg2.Expr) && ED_IsLocStack (&Arg2.Expr) &&
-               StackPtr >= -255 &&
-               ED_IsRVal (&Arg1.Expr) && ED_IsLocConst (&Arg1.Expr)) {
+            /* strcpy returns argument #1 */
+            *Expr = Arg1.Expr;
 
-        /* It is possible to just use one index register even if the stack
-        ** offset is not zero, by adjusting the offset to the constant
-        ** address accordingly. But we cannot do this if the data in
-        ** question is in the register space or at an absolute address less
-        ** than 256. Register space is zero page, which means that the
-        ** address calculation could overflow in the linker.
-        */
-        int AllowOneIndex = !ED_IsLocRegister (&Arg1.Expr) &&
-                            !(ED_IsLocAbs (&Arg1.Expr) && Arg1.Expr.IVal < 256);
+            /* Bail out, no need for further processing */
+            goto ExitPoint;
+        }
 
-        /* Calculate the real stack offset */
-        int Offs = ED_GetStackOffs (&Arg2.Expr, 0);
+        if (ED_IsRVal (&Arg2.Expr) && ED_IsLocStack (&Arg2.Expr) &&
+            StackPtr >= -255 &&
+            ED_IsRVal (&Arg1.Expr) && ED_IsLocConst (&Arg1.Expr)) {
 
-        /* Drop the generated code */
-        RemoveCode (&Arg1.Expr.Start);
+            /* It is possible to just use one index register even if the stack
+            ** offset is not zero, by adjusting the offset to the constant
+            ** address accordingly. But we cannot do this if the data in
+            ** question is in the register space or at an absolute address less
+            ** than 256. Register space is zero page, which means that the
+            ** address calculation could overflow in the linker.
+            */
+            int AllowOneIndex = !ED_IsLocRegister (&Arg1.Expr) &&
+                                !(ED_IsLocAbs (&Arg1.Expr) && Arg1.Expr.IVal < 256);
 
-        /* We need labels */
-        L1 = GetLocalLabel ();
+            /* Calculate the real stack offset */
+            int Offs = ED_GetStackOffs (&Arg2.Expr, 0);
 
-        /* Generate strcpy code */
-        AddCodeLine ("ldy #$%02X", (unsigned char) (Offs - 1));
-        if (Offs == 0 || AllowOneIndex) {
-            g_defcodelabel (L1);
-            AddCodeLine ("iny");
-            AddCodeLine ("lda (sp),y");
-            AddCodeLine ("sta %s,y", ED_GetLabelName (&Arg1.Expr, -Offs));
-        } else {
-            AddCodeLine ("ldx #$FF");
-            g_defcodelabel (L1);
-            AddCodeLine ("iny");
-            AddCodeLine ("inx");
-            AddCodeLine ("lda (sp),y");
-            AddCodeLine ("sta %s,x", ED_GetLabelName (&Arg1.Expr, 0));
-        }
-        AddCodeLine ("bne %s", LocalLabelName (L1));
+            /* Drop the generated code */
+            RemoveCode (&Arg1.Expr.Start);
 
-        /* strcpy returns argument #1 */
-        *Expr = Arg1.Expr;
+            /* We need labels */
+            L1 = GetLocalLabel ();
 
-    } else if (ED_IsRVal (&Arg2.Expr) && ED_IsLocConst (&Arg2.Expr) &&
-               ED_IsRVal (&Arg1.Expr) && ED_IsLocStack (&Arg1.Expr) &&
-               StackPtr >= -255) {
+            /* Generate strcpy code */
+            AddCodeLine ("ldy #$%02X", (unsigned char) (Offs - 1));
+            if (Offs == 0 || AllowOneIndex) {
+                g_defcodelabel (L1);
+                AddCodeLine ("iny");
+                AddCodeLine ("lda (sp),y");
+                AddCodeLine ("sta %s,y", ED_GetLabelName (&Arg1.Expr, -Offs));
+            } else {
+                AddCodeLine ("ldx #$FF");
+                g_defcodelabel (L1);
+                AddCodeLine ("iny");
+                AddCodeLine ("inx");
+                AddCodeLine ("lda (sp),y");
+                AddCodeLine ("sta %s,x", ED_GetLabelName (&Arg1.Expr, 0));
+            }
+            AddCodeLine ("bne %s", LocalLabelName (L1));
 
-        /* It is possible to just use one index register even if the stack
-        ** offset is not zero, by adjusting the offset to the constant
-        ** address accordingly. But we cannot do this if the data in
-        ** question is in the register space or at an absolute address less
-        ** than 256. Register space is zero page, which means that the
-        ** address calculation could overflow in the linker.
-        */
-        int AllowOneIndex = !ED_IsLocRegister (&Arg2.Expr) &&
-                            !(ED_IsLocAbs (&Arg2.Expr) && Arg2.Expr.IVal < 256);
+            /* strcpy returns argument #1 */
+            *Expr = Arg1.Expr;
 
-        /* Calculate the real stack offset */
-        int Offs = ED_GetStackOffs (&Arg1.Expr, 0);
+            /* Bail out, no need for further processing */
+            goto ExitPoint;
+        }
 
-        /* Drop the generated code */
-        RemoveCode (&Arg1.Expr.Start);
+        if (ED_IsRVal (&Arg2.Expr) && ED_IsLocConst (&Arg2.Expr) &&
+            ED_IsRVal (&Arg1.Expr) && ED_IsLocStack (&Arg1.Expr) &&
+            StackPtr >= -255) {
 
-        /* We need labels */
-        L1 = GetLocalLabel ();
+            /* It is possible to just use one index register even if the stack
+            ** offset is not zero, by adjusting the offset to the constant
+            ** address accordingly. But we cannot do this if the data in
+            ** question is in the register space or at an absolute address less
+            ** than 256. Register space is zero page, which means that the
+            ** address calculation could overflow in the linker.
+            */
+            int AllowOneIndex = !ED_IsLocRegister (&Arg2.Expr) &&
+                                !(ED_IsLocAbs (&Arg2.Expr) && Arg2.Expr.IVal < 256);
 
-        /* Generate strcpy code */
-        AddCodeLine ("ldy #$%02X", (unsigned char) (Offs - 1));
-        if (Offs == 0 || AllowOneIndex) {
-            g_defcodelabel (L1);
-            AddCodeLine ("iny");
-            AddCodeLine ("lda %s,y", ED_GetLabelName (&Arg2.Expr, -Offs));
-            AddCodeLine ("sta (sp),y");
-        } else {
-            AddCodeLine ("ldx #$FF");
-            g_defcodelabel (L1);
-            AddCodeLine ("iny");
-            AddCodeLine ("inx");
-            AddCodeLine ("lda %s,x", ED_GetLabelName (&Arg2.Expr, 0));
-            AddCodeLine ("sta (sp),y");
-        }
-        AddCodeLine ("bne %s", LocalLabelName (L1));
+            /* Calculate the real stack offset */
+            int Offs = ED_GetStackOffs (&Arg1.Expr, 0);
 
-        /* strcpy returns argument #1 */
-        *Expr = Arg1.Expr;
+            /* Drop the generated code */
+            RemoveCode (&Arg1.Expr.Start);
 
-    } else {
+            /* We need labels */
+            L1 = GetLocalLabel ();
+
+            /* Generate strcpy code */
+            AddCodeLine ("ldy #$%02X", (unsigned char) (Offs - 1));
+            if (Offs == 0 || AllowOneIndex) {
+                g_defcodelabel (L1);
+                AddCodeLine ("iny");
+                AddCodeLine ("lda %s,y", ED_GetLabelName (&Arg2.Expr, -Offs));
+                AddCodeLine ("sta (sp),y");
+            } else {
+                AddCodeLine ("ldx #$FF");
+                g_defcodelabel (L1);
+                AddCodeLine ("iny");
+                AddCodeLine ("inx");
+                AddCodeLine ("lda %s,x", ED_GetLabelName (&Arg2.Expr, 0));
+                AddCodeLine ("sta (sp),y");
+            }
+            AddCodeLine ("bne %s", LocalLabelName (L1));
 
-        /* The function result is an rvalue in the primary register */
-        ED_MakeRValExpr (Expr);
-        Expr->Type = GetFuncReturn (Expr->Type);
+            /* strcpy returns argument #1 */
+            *Expr = Arg1.Expr;
 
+            /* Bail out, no need for further processing */
+            goto ExitPoint;
+        }
     }
 
+    /* The function result is an rvalue in the primary register */
+    ED_MakeRValExpr (Expr);
+    Expr->Type = GetFuncReturn (Expr->Type);
+
+ExitPoint:
     /* We expect the closing brace */
     ConsumeRParen ();
 }
@@ -1142,8 +1180,6 @@ static void StdFunc_strlen (FuncDesc* F attribute ((unused)), ExprDesc* Expr)
     long        ECount;
     unsigned    L;
 
-
-
     /* Setup the argument type string */
     ArgType[1].C = GetDefaultChar () | T_QUAL_CONST;
 
@@ -1175,125 +1211,148 @@ static void StdFunc_strlen (FuncDesc* F attribute ((unused)), ExprDesc* Expr)
     /* Do type conversion */
     TypeConversion (&Arg, ArgType);
 
-    /* If the expression is a literal, and if string literals are read
-    ** only, we can calculate the length of the string and remove it
-    ** from the literal pool. Otherwise we have to calculate the length
-    ** at runtime.
-    */
-    if (ED_IsLocLiteral (&Arg) && IS_Get (&WritableStrings) == 0) {
+    if (IS_Get (&Optimize)) {
 
-        /* Constant string literal */
-        ED_MakeConstAbs (Expr, GetLiteralSize (Arg.LVal) - 1, type_size_t);
+        /* If the expression is a literal, and if string literals are read
+        ** only, we can calculate the length of the string and remove it
+        ** from the literal pool. Otherwise we have to calculate the length
+        ** at runtime.
+        */
+        if (ED_IsLocLiteral (&Arg) && IS_Get (&WritableStrings) == 0) {
 
-        /* We don't need the literal any longer */
-        ReleaseLiteral (Arg.LVal);
+            /* Constant string literal */
+            ED_MakeConstAbs (Expr, GetLiteralSize (Arg.LVal) - 1, type_size_t);
 
-    /* We will inline strlen for arrays with constant addresses, if either the
-    ** inlining was forced on the command line, or the array is smaller than
-    ** 256, so the inlining is considered safe.
-    */
-    } else if (ED_IsLocConst (&Arg) && IsArray &&
-               (IS_Get (&InlineStdFuncs) || IsByteIndex)) {
-
-        /* Generate the strlen code */
-        L = GetLocalLabel ();
-        AddCodeLine ("ldy #$FF");
-        g_defcodelabel (L);
-        AddCodeLine ("iny");
-        AddCodeLine ("lda %s,y", ED_GetLabelName (&Arg, 0));
-        AddCodeLine ("bne %s", LocalLabelName (L));
-        AddCodeLine ("tax");
-        AddCodeLine ("tya");
-
-        /* The function result is an rvalue in the primary register */
-        ED_MakeRValExpr (Expr);
-        Expr->Type = type_size_t;
-
-    /* We will inline strlen for arrays on the stack, if the array is
-    ** completely within the reach of a byte sized index register.
-    */
-    } else if (ED_IsLocStack (&Arg) && IsArray && IsByteIndex &&
-               (Arg.IVal - StackPtr) + ECount < 256) {
-
-        /* Calculate the true stack offset */
-        int Offs = ED_GetStackOffs (&Arg, 0);
-
-        /* Generate the strlen code */
-        L = GetLocalLabel ();
-        AddCodeLine ("ldx #$FF");
-        AddCodeLine ("ldy #$%02X", (unsigned char) (Offs-1));
-        g_defcodelabel (L);
-        AddCodeLine ("inx");
-        AddCodeLine ("iny");
-        AddCodeLine ("lda (sp),y");
-        AddCodeLine ("bne %s", LocalLabelName (L));
-        AddCodeLine ("txa");
-        AddCodeLine ("ldx #$00");
-
-        /* The function result is an rvalue in the primary register */
-        ED_MakeRValExpr (Expr);
-        Expr->Type = type_size_t;
-
-    /* strlen for a string that is pointed to by a register variable will only
-    ** get inlined if requested on the command line, since we cannot know how
-    ** big the buffer actually is, so inlining is not always safe.
-    */
-    } else if (ED_IsLocRegister (&Arg) && ED_IsLVal (&Arg) && IsPtr &&
-               IS_Get (&InlineStdFuncs)) {
-
-        /* Generate the strlen code */
-        L = GetLocalLabel ();
-        AddCodeLine ("ldy #$FF");
-        g_defcodelabel (L);
-        AddCodeLine ("iny");
-        AddCodeLine ("lda (%s),y", ED_GetLabelName (&Arg, 0));
-        AddCodeLine ("bne %s", LocalLabelName (L));
-        AddCodeLine ("tax");
-        AddCodeLine ("tya");
-
-        /* The function result is an rvalue in the primary register */
-        ED_MakeRValExpr (Expr);
-        Expr->Type = type_size_t;
-
-    /* Last check: We will inline a generic strlen routine if inlining was
-    ** requested on the command line, and the code size factor is more than
-    ** 400 (code is 13 bytes vs. 3 for a jsr call).
-    */
-    } else if (IS_Get (&CodeSizeFactor) > 400 && IS_Get (&InlineStdFuncs)) {
-
-        /* Load the expression into the primary */
-        LoadExpr (CF_NONE, &Arg);
-
-        /* Inline the function */
-        L = GetLocalLabel ();
-        AddCodeLine ("sta ptr1");
-        AddCodeLine ("stx ptr1+1");
-        AddCodeLine ("ldy #$FF");
-        g_defcodelabel (L);
-        AddCodeLine ("iny");
-        AddCodeLine ("lda (ptr1),y");
-        AddCodeLine ("bne %s", LocalLabelName (L));
-        AddCodeLine ("tax");
-        AddCodeLine ("tya");
-
-        /* The function result is an rvalue in the primary register */
-        ED_MakeRValExpr (Expr);
-        Expr->Type = type_size_t;
+            /* We don't need the literal any longer */
+            ReleaseLiteral (Arg.LVal);
 
-    } else {
+            /* Bail out, no need for further improvements */
+            goto ExitPoint;
+        }
+    }
+
+    if (IS_Get (&InlineStdFuncs)) {
+
+        /* We will inline strlen for arrays with constant addresses, if either
+        ** requested on the command line, or the array is smaller than 256,
+        ** so the inlining is considered safe.
+        */
+        if (ED_IsLocConst (&Arg) && IsArray &&
+            (IS_Get (&EagerlyInlineFuncs) || IsByteIndex)) {
+
+            /* Generate the strlen code */
+            L = GetLocalLabel ();
+            AddCodeLine ("ldy #$FF");
+            g_defcodelabel (L);
+            AddCodeLine ("iny");
+            AddCodeLine ("lda %s,y", ED_GetLabelName (&Arg, 0));
+            AddCodeLine ("bne %s", LocalLabelName (L));
+            AddCodeLine ("tax");
+            AddCodeLine ("tya");
+
+            /* The function result is an rvalue in the primary register */
+            ED_MakeRValExpr (Expr);
+            Expr->Type = type_size_t;
+
+            /* Bail out, no need for further processing */
+            goto ExitPoint;
+        }
+
+        /* We will inline strlen for arrays on the stack, if the array is
+        ** completely within the reach of a byte sized index register.
+        */
+        if (ED_IsLocStack (&Arg) && IsArray && IsByteIndex &&
+            (Arg.IVal - StackPtr) + ECount < 256) {
+
+            /* Calculate the true stack offset */
+            int Offs = ED_GetStackOffs (&Arg, 0);
+
+            /* Generate the strlen code */
+            L = GetLocalLabel ();
+            AddCodeLine ("ldx #$FF");
+            AddCodeLine ("ldy #$%02X", (unsigned char) (Offs-1));
+            g_defcodelabel (L);
+            AddCodeLine ("inx");
+            AddCodeLine ("iny");
+            AddCodeLine ("lda (sp),y");
+            AddCodeLine ("bne %s", LocalLabelName (L));
+            AddCodeLine ("txa");
+            AddCodeLine ("ldx #$00");
+
+            /* The function result is an rvalue in the primary register */
+            ED_MakeRValExpr (Expr);
+            Expr->Type = type_size_t;
+
+            /* Bail out, no need for further processing */
+            goto ExitPoint;
+        }
 
-        /* Load the expression into the primary */
-        LoadExpr (CF_NONE, &Arg);
+        /* strlen for a string that is pointed to by a register variable will only
+        ** get inlined if requested on the command line, since we cannot know how
+        ** big the buffer actually is, so inlining is not always safe.
+        */
+        if (ED_IsLocRegister (&Arg) && ED_IsLVal (&Arg) && IsPtr &&
+            IS_Get (&EagerlyInlineFuncs)) {
 
-        /* Call the strlen function */
-        AddCodeLine ("jsr _%s", Func_strlen);
+            /* Generate the strlen code */
+            L = GetLocalLabel ();
+            AddCodeLine ("ldy #$FF");
+            g_defcodelabel (L);
+            AddCodeLine ("iny");
+            AddCodeLine ("lda (%s),y", ED_GetLabelName (&Arg, 0));
+            AddCodeLine ("bne %s", LocalLabelName (L));
+            AddCodeLine ("tax");
+            AddCodeLine ("tya");
+
+            /* The function result is an rvalue in the primary register */
+            ED_MakeRValExpr (Expr);
+            Expr->Type = type_size_t;
+
+            /* Bail out, no need for further processing */
+            goto ExitPoint;
+        }
+
+        /* Last check: We will inline a generic strlen routine if inlining was
+        ** requested on the command line, and the code size factor is more than
+        ** 400 (code is 13 bytes vs. 3 for a jsr call).
+        */
+        if (IS_Get (&CodeSizeFactor) > 400 && IS_Get (&EagerlyInlineFuncs)) {
 
-        /* The function result is an rvalue in the primary register */
-        ED_MakeRValExpr (Expr);
-        Expr->Type = type_size_t;
+            /* Load the expression into the primary */
+            LoadExpr (CF_NONE, &Arg);
 
+            /* Inline the function */
+            L = GetLocalLabel ();
+            AddCodeLine ("sta ptr1");
+            AddCodeLine ("stx ptr1+1");
+            AddCodeLine ("ldy #$FF");
+            g_defcodelabel (L);
+            AddCodeLine ("iny");
+            AddCodeLine ("lda (ptr1),y");
+            AddCodeLine ("bne %s", LocalLabelName (L));
+            AddCodeLine ("tax");
+            AddCodeLine ("tya");
+
+            /* The function result is an rvalue in the primary register */
+            ED_MakeRValExpr (Expr);
+            Expr->Type = type_size_t;
+
+            /* Bail out, no need for further processing */
+            goto ExitPoint;
+        }
     }
 
+    /* Load the expression into the primary */
+    LoadExpr (CF_NONE, &Arg);
+
+    /* Call the strlen function */
+    AddCodeLine ("jsr _%s", Func_strlen);
+
+    /* The function result is an rvalue in the primary register */
+    ED_MakeRValExpr (Expr);
+    Expr->Type = type_size_t;
+
+ExitPoint:
     /* We expect the closing brace */
     ConsumeRParen ();
 }
-- 
2.39.5