From 95635418707e5a6ed270a341c02a743751631099 Mon Sep 17 00:00:00 2001
From: Polluks <polluks@sdf.lonestar.org>
Date: Sun, 29 Nov 2020 17:10:34 +0100
Subject: [PATCH 01/26] crt0 clean-up

---
 libsrc/atari5200/crt0.s   | 1 -
 libsrc/c16/crt0.s         | 1 -
 libsrc/c64/crt0.s         | 1 -
 libsrc/pet/crt0.s         | 1 -
 libsrc/plus4/crt0.s       | 2 --
 libsrc/supervision/crt0.s | 2 --
 libsrc/telestrat/crt0.s   | 1 -
 libsrc/vic20/crt0.s       | 1 -
 8 files changed, 10 deletions(-)

diff --git a/libsrc/atari5200/crt0.s b/libsrc/atari5200/crt0.s
index ee3d0de4f..9538d3bb4 100644
--- a/libsrc/atari5200/crt0.s
+++ b/libsrc/atari5200/crt0.s
@@ -13,7 +13,6 @@
         .import         zerobss, copydata
 
         .include        "zeropage.inc"
-        .include        "atari5200.inc"
 
 start:
 
diff --git a/libsrc/c16/crt0.s b/libsrc/c16/crt0.s
index bee81a113..1df1e5c62 100644
--- a/libsrc/c16/crt0.s
+++ b/libsrc/c16/crt0.s
@@ -13,7 +13,6 @@
         .importzp       ST
 
         .include        "zeropage.inc"
-        .include        "plus4.inc"
 
 ; ------------------------------------------------------------------------
 ; Startup code
diff --git a/libsrc/c64/crt0.s b/libsrc/c64/crt0.s
index 7bd294ca7..4e5c7c9d4 100644
--- a/libsrc/c64/crt0.s
+++ b/libsrc/c64/crt0.s
@@ -13,7 +13,6 @@
         .importzp       ST
 
         .include        "zeropage.inc"
-        .include        "c64.inc"
 
 
 ; ------------------------------------------------------------------------
diff --git a/libsrc/pet/crt0.s b/libsrc/pet/crt0.s
index 520a147f7..e56a7eca4 100644
--- a/libsrc/pet/crt0.s
+++ b/libsrc/pet/crt0.s
@@ -12,7 +12,6 @@
 
         .include        "zeropage.inc"
         .include        "pet.inc"
-        .include        "../cbm/cbm.inc"
 
 ; ------------------------------------------------------------------------
 ; Startup code
diff --git a/libsrc/plus4/crt0.s b/libsrc/plus4/crt0.s
index 2262b4c42..6b44a2b7e 100644
--- a/libsrc/plus4/crt0.s
+++ b/libsrc/plus4/crt0.s
@@ -198,5 +198,3 @@ irqcount:       .byte   0
 .segment        "INIT"
 
 zpsave:         .res    zpspace
-
-
diff --git a/libsrc/supervision/crt0.s b/libsrc/supervision/crt0.s
index ac61c8215..203c681b8 100644
--- a/libsrc/supervision/crt0.s
+++ b/libsrc/supervision/crt0.s
@@ -78,5 +78,3 @@ not_dma:
 .word   nmi
 .word   reset32kcode
 .word   irq
-
-
diff --git a/libsrc/telestrat/crt0.s b/libsrc/telestrat/crt0.s
index 59b1ea642..df75520ce 100644
--- a/libsrc/telestrat/crt0.s
+++ b/libsrc/telestrat/crt0.s
@@ -12,7 +12,6 @@
         .import         __MAIN_START__, __MAIN_SIZE__
 
         .include        "zeropage.inc"
-        .include        "telestrat.inc"
 
 ; ------------------------------------------------------------------------
 ; Place the startup code in a special segment.
diff --git a/libsrc/vic20/crt0.s b/libsrc/vic20/crt0.s
index 68ab3ed12..c5486063b 100644
--- a/libsrc/vic20/crt0.s
+++ b/libsrc/vic20/crt0.s
@@ -13,7 +13,6 @@
         .importzp       ST
 
         .include        "zeropage.inc"
-        .include        "vic20.inc"
 
 ; ------------------------------------------------------------------------
 ; Startup code

From b22d8c7441d29657fa772cd4da08b2ca5dd61946 Mon Sep 17 00:00:00 2001
From: Olli Savia <ops@iki.fi>
Date: Sat, 28 Nov 2020 13:03:48 +0200
Subject: [PATCH 02/26] Added STATUS (0) definition

---
 asminc/c128.inc  | 1 +
 asminc/c64.inc   | 1 +
 asminc/plus4.inc | 1 +
 asminc/vic20.inc | 1 +
 4 files changed, 4 insertions(+)

diff --git a/asminc/c128.inc b/asminc/c128.inc
index 2852631f3..749b4168c 100644
--- a/asminc/c128.inc
+++ b/asminc/c128.inc
@@ -7,6 +7,7 @@
 ; Zero page, Commodore stuff
 
 TXTPTR          := $3D          ; Pointer into BASIC source code
+STATUS          := $90          ; Kernal I/O completion status
 TIME            := $A0          ; 60HZ clock
 FNAM_LEN        := $B7          ; Length of filename
 SECADR          := $B9          ; Secondary address
diff --git a/asminc/c64.inc b/asminc/c64.inc
index 1d10f673d..2cfc50db4 100644
--- a/asminc/c64.inc
+++ b/asminc/c64.inc
@@ -9,6 +9,7 @@
 VARTAB          := $2D          ; Pointer to start of BASIC variables
 MEMSIZE         := $37          ; Pointer to highest BASIC RAM location (+1)
 TXTPTR          := $7A          ; Pointer into BASIC source code
+STATUS          := $90          ; Kernal I/O completion status
 TIME            := $A0          ; 60 HZ clock
 FNAM_LEN        := $B7          ; Length of filename
 SECADR          := $B9          ; Secondary address
diff --git a/asminc/plus4.inc b/asminc/plus4.inc
index 774722e93..6c6017a78 100644
--- a/asminc/plus4.inc
+++ b/asminc/plus4.inc
@@ -10,6 +10,7 @@ TMPPTR          := $22          ; Temporary ptr used by BASIC
 VARTAB          := $2D          ; Pointer to start of BASIC variables
 MEMSIZE         := $37          ; Pointer to highest BASIC RAM location (+1)
 TXTPTR          := $3B          ; Pointer into BASIC source code
+STATUS          := $90          ; Kernal I/O completion status
 TIME            := $A3          ; 60HZ clock
 FNAM_LEN        := $AB          ; Length of filename
 LFN             := $AC          ; Logical file number
diff --git a/asminc/vic20.inc b/asminc/vic20.inc
index b82874f56..fa9fdfa8d 100644
--- a/asminc/vic20.inc
+++ b/asminc/vic20.inc
@@ -9,6 +9,7 @@
 VARTAB          := $2D          ; Pointer to start of BASIC variables
 MEMSIZE         := $37          ; Pointer to highest BASIC RAM location (+1)
 TXTPTR          := $7A          ; Pointer into BASIC source code
+STATUS          := $90          ; Kernal I/O completion status
 TIME            := $A0          ; 60HZ clock
 FNAM_LEN        := $B7          ; Length of filename
 SECADR          := $B9          ; Secondary address

From 59c58acbe3229a4e860981e97d3d5f3a23adf2bc Mon Sep 17 00:00:00 2001
From: Greg King <gregdk@users.sf.net>
Date: Sat, 5 Dec 2020 23:04:48 -0500
Subject: [PATCH 03/26] Documented the address size argument of the bss-name,
 code-name, data-name, and rodata-name pragmas.

---
 doc/cc65.sgml | 143 +++++++++++++++++++++++++++++++++-----------------
 1 file changed, 96 insertions(+), 47 deletions(-)

diff --git a/doc/cc65.sgml b/doc/cc65.sgml
index 115f0a30c..e273ced9c 100644
--- a/doc/cc65.sgml
+++ b/doc/cc65.sgml
@@ -6,8 +6,9 @@
 <url url="mailto:gregdk@users.sf.net" name="Greg King">
 
 <abstract>
-cc65 is a C compiler for 6502 targets. It supports several 6502 based home
-computers like the Commodore and Atari machines, but it is easily retargetable.
+cc65 is a C compiler for 6502 targets. It supports several 6502-based home
+computers such as the Commodore and Atari machines, but it easily is
+retargetable.
 </abstract>
 
 <!-- Table of contents -->
@@ -1011,7 +1012,7 @@ The compiler defines several macros at startup:
   <tag><tt>__TELESTRAT__</tt></tag>
 
   This macro is defined if the target is the Telestrat (-t telestrat).
-  
+
   <tag><tt>__TIME__</tt></tag>
 
   This macro expands to the time of translation of the preprocessing
@@ -1045,26 +1046,39 @@ parameter with the <tt/#pragma/.
   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>
+<sect1><tt>#pragma bss-name ([push, ]&lt;name>[ ,&lt;addrsize>])</tt><label id="pragma-bss-name"><p>
 
-  This pragma changes the name used for the BSS segment (the BSS segment
-  is used to store variables with static storage duration and no explicit
-  initializer). The argument is a string enclosed in double quotes.
+  This pragma changes the name used for the BSS segment (the BSS segment is
+  used to store variables with static storage duration and no explicit
+  initializers). The <tt/name/ argument is a string enclosed in quotation
+  marks.
 
-  Note: The default linker configuration file does only map the standard
-  segments. If you use other segments, you have to create a new linker
-  configuration file.
+  <tt/addrsize/ is an optional string that gives a hint about where the
+  <tt/name/ segment will be put in the CPU's address space.  It describes the
+  width of address numbers that point into that segment.  Only words that
+  are known to ca65 are allowed:
+  <enum>
+  <item>"zp", "zeropage", "direct"
+  <item>"abs", "absolute", "near", "default"
+  <item>"far"
+  <item>"long", "dword"
+  </enum>
 
-  Beware: The startup code will zero only the default BSS segment. If you
-  use another BSS segment, you have to do that yourself, otherwise
-  variables with static storage duration and no explicit initializer will
-  not have the value zero.
+  Note: The default linker configuration file maps only the standard segments.
+  If you use other segments, you must create a new linker configuration file.
 
-  The <tt/#pragma/ understands the push and pop parameters as explained above.
+  Beware: The start-up code will zero only the default BSS segment. If you use
+  another BSS segment, then you must do that yourself; otherwise, variables
+  with static storage duration and no explicit initializer will not have the
+  value zero.
 
-  Example:
+  The <tt/#pragma/ understands the push and pop parameters, as explained above.
+
+  Examples:
   <tscreen><verb>
-        #pragma bss-name ("MyBSS")
+  #pragma bss-name ("MyBSS")
+  #pragma bss-name (push, "MyBSS")
+  #pragma bss-name ("MyBSS", "zp")
   </verb></tscreen>
 
 
@@ -1120,21 +1134,33 @@ 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>
+<sect1><tt>#pragma code-name ([push, ]&lt;name>[ ,&lt;addrsize>])</tt><label id="pragma-code-name"><p>
 
-  This pragma changes the name used for the CODE segment (the CODE segment
-  is used to store executable code). The argument is a string enclosed in
-  double quotes.
+  This pragma changes the name used for the CODE segment (the CODE segment is
+  used to store executable code). The <tt/name/ argument is a string enclosed
+  in quotation marks.
 
-  Note: The default linker configuration file does only map the standard
-  segments. If you use other segments, you have to create a new linker
-  configuration file.
+  <tt/addrsize/ is an optional string that gives a hint about where the
+  <tt/name/ segment will be put in the CPU's address space.  It describes the
+  width of address numbers that point into that segment.  Only words that
+  are known to ca65 are allowed:
+  <enum>
+  <item>"zp", "zeropage", "direct"
+  <item>"abs", "absolute", "near", "default"
+  <item>"far"
+  <item>"long", "dword"
+  </enum>
 
-  The <tt/#pragma/ understands the push and pop parameters as explained above.
+  Note: The default linker configuration file maps only the standard segments.
+  If you use other segments, you must create a new linker configuration file.
 
-  Example:
+  The <tt/#pragma/ understands the push and pop parameters, as explained above.
+
+  Examples:
   <tscreen><verb>
-        #pragma code-name ("MyCODE")
+  #pragma code-name ("MyCODE")
+  #pragma code-name (push, "MyCODE")
+  #pragma code-name (push, "MyCODE", "far")
   </verb></tscreen>
 
 
@@ -1148,21 +1174,33 @@ parameter with the <tt/#pragma/.
   The <tt/#pragma/ understands the push and pop parameters as explained above.
 
 
-<sect1><tt>#pragma data-name ([push,] &lt;name&gt;)</tt><label id="pragma-data-name"><p>
+<sect1><tt>#pragma data-name ([push, ]&lt;name>[ ,&lt;addrsize>])</tt><label id="pragma-data-name"><p>
 
-  This pragma changes the name used for the DATA segment (the DATA segment
-  is used to store initialized data). The argument is a string enclosed in
-  double quotes.
+  This pragma changes the name used for the DATA segment (the DATA segment is
+  used to store initialized data). The <tt/name/ argument is a string enclosed
+  in quotation marks.
 
-  Note: The default linker configuration file does only map the standard
-  segments. If you use other segments, you have to create a new linker
-  configuration file.
+  <tt/addrsize/ is an optional string that gives a hint about where the
+  <tt/name/ segment will be put in the CPU's address space.  It describes the
+  width of address numbers that point into that segment.  Only words that
+  are known to ca65 are allowed:
+  <enum>
+  <item>"zp", "zeropage", "direct"
+  <item>"abs", "absolute", "near", "default"
+  <item>"far"
+  <item>"long", "dword"
+  </enum>
 
-  The <tt/#pragma/ understands the push and pop parameters as explained above.
+  Note: The default linker configuration file maps only the standard segments.
+  If you use other segments, you must create a new linker configuration file.
 
-  Example:
+  The <tt/#pragma/ understands the push and pop parameters, as explained above.
+
+  Examples:
   <tscreen><verb>
-        #pragma data-name ("MyDATA")
+  #pragma data-name ("MyDATA")
+  #pragma data-name (push, "MyDATA")
+  #pragma data-name ("MyDATA", "zeropage")
   </verb></tscreen>
 
 
@@ -1224,21 +1262,32 @@ parameter with the <tt/#pragma/.
   The <tt/#pragma/ understands the push and pop parameters as explained above.
 
 
-<sect1><tt>#pragma rodata-name ([push,] &lt;name&gt;)</tt><label id="pragma-rodata-name"><p>
+<sect1><tt>#pragma rodata-name ([push, ]&lt;name>[ ,&lt;addrsize>])</tt><label id="pragma-rodata-name"><p>
 
-  This pragma changes the name used for the RODATA segment (the RODATA
-  segment is used to store readonly data). The argument is a string
-  enclosed in double quotes.
+  This pragma changes the name used for the RODATA segment (the RODATA segment
+  is used to store read-only data). The <tt/name/ argument is a string enclosed
+  in quotation marks.
 
-  Note: The default linker configuration file does only map the standard
-  segments. If you use other segments, you have to create a new linker
-  configuration file.
+  <tt/addrsize/ is an optional string that gives a hint about where the
+  <tt/name/ segment will be put in the CPU's address space.  It describes the
+  width of address numbers that point into that segment.  Only words that
+  are known to ca65 are allowed:
+  <enum>
+  <item>"zp", "zeropage", "direct"
+  <item>"abs", "absolute", "near", "default"
+  <item>"far"
+  <item>"long", "dword"
+  </enum>
 
-  The <tt/#pragma/ understands the push and pop parameters as explained above.
+  Note: The default linker configuration file maps only the standard segments.
+  If you use other segments, you must create a new linker configuration file.
 
-  Example:
+  The <tt/#pragma/ understands the push and pop parameters, as explained above.
+
+  Examples:
   <tscreen><verb>
-        #pragma rodata-name ("MyRODATA")
+  #pragma rodata-name ("MyRODATA")
+  #pragma rodata-name (push, "MyRODATA")
   </verb></tscreen>
 
 

From 0f4cb443b4d34dfceef35c0f9b31abcf69d00c67 Mon Sep 17 00:00:00 2001
From: Oliver Schmidt <ol.sc@web.de>
Date: Sat, 19 Dec 2020 19:54:12 +0100
Subject: [PATCH 04/26] Improved device I/O under DOS 3.3

Certain scenarios (e.g. not running any Applesoft program at all since booting DOS 3.3) can make DOS 3.3 consider cc65 device input (e.g. getchar()) that reads a CR interpreting the command in the keyboard buffer. Setting the hibyte of the Applesoft currently executed line number to some value <> $FF (beside setting the input prompt to some value <> ']') makes DOS 3.3 understand that we're not in intermediate mode and that therefore I/O not preceded with ctrl-d mustn't be fiddled with (see DOS 3.3 routine at $A65E).
---
 asminc/apple2.inc    | 1 +
 libsrc/apple2/read.s | 9 ++++++---
 2 files changed, 7 insertions(+), 3 deletions(-)

diff --git a/asminc/apple2.inc b/asminc/apple2.inc
index 5ebf73164..528c463a1 100644
--- a/asminc/apple2.inc
+++ b/asminc/apple2.inc
@@ -15,6 +15,7 @@ PROMPT  :=      $33     ; Used by GETLN
 RNDL    :=      $4E     ; Random counter low
 RNDH    :=      $4F     ; Random counter high
 HIMEM   :=      $73     ; Highest available memory address+1
+CURLIN  :=      $75     ; Current line number being executed
 
 ;-----------------------------------------------------------------------------
 ; Vectors
diff --git a/libsrc/apple2/read.s b/libsrc/apple2/read.s
index 14c80b7e2..99c189cbd 100644
--- a/libsrc/apple2/read.s
+++ b/libsrc/apple2/read.s
@@ -19,11 +19,14 @@
         .segment        "ONCE"
 
 initprompt:
-        ; Set prompt <> ']' to let DOS 3.3 know that we're
-        ; not in Applesoft immediate mode and thus keep it
-        ; from scanning our device I/O for DOS commands.
+        ; Set prompt <> ']' and currently executed Applesoft
+        ; line number hibyte <> $FF to let DOS 3.3 (at $A65E)
+        ; know that we're not in Applesoft immediate mode and
+        ; thus keep it from scanning our device I/O for DOS
+        ; commands.
         lda     #$80            ; Same value used at $D52C
         sta     PROMPT
+        sta     CURLIN+1        ; Any value <> $FF will do
         rts
 
         .code

From b2c1a77bb389bf20b5043f796f2dee496c07fdbd Mon Sep 17 00:00:00 2001
From: Greg King <gregdk@users.sf.net>
Date: Thu, 24 Dec 2020 12:27:09 -0500
Subject: [PATCH 05/26] Fixed the cc65 code that optimizes 16-bit compares when
 the high bytes are known to be equal.

Only the low bytes are compared.  Originally, signed 16-bit compares were optimized into signed 8-bit compares.  But, the sign bits are in the high bytes; and, they're equal.  Therefore, the low bytes always must be compared as unsigned numbers.
Fixes #1348.
---
 src/cc65/coptstop.c          | 78 +++++++++++-------------------------
 test/misc/Makefile           |  8 ----
 test/{misc => val}/bug1348.c | 28 ++++++-------
 3 files changed, 37 insertions(+), 77 deletions(-)
 rename test/{misc => val}/bug1348.c (61%)

diff --git a/src/cc65/coptstop.c b/src/cc65/coptstop.c
index 59bbe6dc9..77b79281b 100644
--- a/src/cc65/coptstop.c
+++ b/src/cc65/coptstop.c
@@ -287,7 +287,7 @@ static unsigned Opt_tosshift (StackOpData* D, const char* Name)
         /* ldx */
         X = NewCodeEntry (OP65_LDX, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI);
         InsertEntry (D, X, D->IP++);
- 
+
         /* Lhs load entries can be removed if not used later */
         D->Lhs.X.Flags |= LI_REMOVE;
         D->Lhs.A.Flags |= LI_REMOVE;
@@ -1100,7 +1100,7 @@ static unsigned Opt_tosxorax (StackOpData* D)
 
 
 static unsigned Opt_a_toscmpbool (StackOpData* D, const char* BoolTransformer)
-/* Optimize the tos compare sequence with a bool transformer */
+/* Optimize the TOS compare sequence with a bool transformer */
 {
     CodeEntry* X;
     cmp_t      Cond;
@@ -1119,9 +1119,8 @@ static unsigned Opt_a_toscmpbool (StackOpData* D, const char* BoolTransformer)
         D->Rhs.A.Flags |= LI_REMOVE;
 
     } else if ((D->Lhs.A.Flags & LI_DIRECT) != 0) {
-
-        /* If the lhs is direct (but not stack relative), encode compares with lhs
-        ** effectively reverting the order (which doesn't matter for ==).
+        /* If the lhs is direct (but not stack relative), encode compares with lhs,
+        ** effectively reversing the order (which doesn't matter for == and !=).
         */
         Cond = FindBoolCmpCond (BoolTransformer);
         Cond = GetRevertedCond (Cond);
@@ -1138,7 +1137,6 @@ static unsigned Opt_a_toscmpbool (StackOpData* D, const char* BoolTransformer)
         D->Lhs.A.Flags |= LI_REMOVE;
 
     } else {
-
         /* We'll do reverse-compare */
         Cond = FindBoolCmpCond (BoolTransformer);
         Cond = GetRevertedCond (Cond);
@@ -1162,7 +1160,7 @@ static unsigned Opt_a_toscmpbool (StackOpData* D, const char* BoolTransformer)
     X = NewCodeEntry (OP65_JSR, AM65_ABS, BoolTransformer, 0, D->OpEntry->LI);
     InsertEntry (D, X, D->IP++);
 
-    /* Remove the push and the call to the tosgeax function */
+    /* Remove the push and the call to the TOS function */
     RemoveRemainders (D);
 
     /* We changed the sequence */
@@ -1179,22 +1177,6 @@ static unsigned Opt_a_toseq (StackOpData* D)
 
 
 
-static unsigned Opt_a_tosge (StackOpData* D)
-/* Optimize the tosgeax sequence */
-{
-    return Opt_a_toscmpbool (D, "boolge");
-}
-
-
-
-static unsigned Opt_a_tosgt (StackOpData* D)
-/* Optimize the tosgtax sequence */
-{
-    return Opt_a_toscmpbool (D, "boolgt");
-}
-
-
-
 static unsigned Opt_a_tosicmp (StackOpData* D)
 /* Replace tosicmp with CMP */
 {
@@ -1236,7 +1218,7 @@ static unsigned Opt_a_tosicmp (StackOpData* D)
                 }
                 InsertEntry (D, X, D->IP++);
 
-                /* cmp src,y OR cmp (sp),y*/
+                /* cmp src,y OR cmp (sp),y */
                 if (D->Rhs.A.LoadEntry->OPC == OP65_JSR) {
                     /* opc (sp),y */
                     X = NewCodeEntry (OP65_CMP, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
@@ -1268,18 +1250,18 @@ static unsigned Opt_a_tosicmp (StackOpData* D)
         InsertEntry (D, X, D->IP-3);
 
     } else {
-        /* Just clear A,Z,N and set C */
+        /* Just clear A,Z,N; and set C */
+        Arg = MakeHexArg (0);
         if ((RI = GetLastChangedRegInfo (D, &D->Lhs.A)) != 0 &&
             RegValIsKnown (RI->Out.RegA)                     &&
             (RI->Out.RegA & 0xFF) == 0) {
-            Arg = MakeHexArg (0);
-            X   = NewCodeEntry (OP65_CMP, AM65_IMM, Arg, 0, D->OpEntry->LI);
+
+            X = NewCodeEntry (OP65_CMP, AM65_IMM, Arg, 0, D->OpEntry->LI);
             InsertEntry (D, X, D->OpIndex + 1);
         } else {
-            Arg = MakeHexArg (0);
-            X   = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, D->OpEntry->LI);
+            X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, D->OpEntry->LI);
             InsertEntry (D, X, D->OpIndex + 1);
-            X   = NewCodeEntry (OP65_CMP, AM65_IMM, Arg, 0, D->OpEntry->LI);
+            X = NewCodeEntry (OP65_CMP, AM65_IMM, Arg, 0, D->OpEntry->LI);
             InsertEntry (D, X, D->OpIndex + 2);
         }
     }
@@ -1292,24 +1274,8 @@ static unsigned Opt_a_tosicmp (StackOpData* D)
 
 
 
-static unsigned Opt_a_tosle (StackOpData* D)
-/* Optimize the tosleax sequence */
-{
-    return Opt_a_toscmpbool (D, "boolle");
-}
-
-
-
-static unsigned Opt_a_toslt (StackOpData* D)
-/* Optimize the tosltax sequence */
-{
-    return Opt_a_toscmpbool (D, "boollt");
-}
-
-
-
 static unsigned Opt_a_tosne (StackOpData* D)
-/* Optimize the toseqax sequence */
+/* Optimize the tosneax sequence */
 {
     return Opt_a_toscmpbool (D, "boolne");
 }
@@ -1317,7 +1283,7 @@ static unsigned Opt_a_tosne (StackOpData* D)
 
 
 static unsigned Opt_a_tosuge (StackOpData* D)
-/* Optimize the tosugeax sequence */
+/* Optimize the tosgeax and tosugeax sequences */
 {
     return Opt_a_toscmpbool (D, "booluge");
 }
@@ -1325,7 +1291,7 @@ static unsigned Opt_a_tosuge (StackOpData* D)
 
 
 static unsigned Opt_a_tosugt (StackOpData* D)
-/* Optimize the tosugtax sequence */
+/* Optimize the tosgtax and tosugtax sequences */
 {
     return Opt_a_toscmpbool (D, "boolugt");
 }
@@ -1333,7 +1299,7 @@ static unsigned Opt_a_tosugt (StackOpData* D)
 
 
 static unsigned Opt_a_tosule (StackOpData* D)
-/* Optimize the tosuleax sequence */
+/* Optimize the tosleax and tosuleax sequences */
 {
     return Opt_a_toscmpbool (D, "boolule");
 }
@@ -1341,7 +1307,7 @@ static unsigned Opt_a_tosule (StackOpData* D)
 
 
 static unsigned Opt_a_tosult (StackOpData* D)
-/* Optimize the tosultax sequence */
+/* Optimize the tosltax and tosultax sequences */
 {
     return Opt_a_toscmpbool (D, "boolult");
 }
@@ -1354,6 +1320,8 @@ static unsigned Opt_a_tosult (StackOpData* D)
 
 
 
+/* The first column of these two tables must be sorted in lexical order */
+
 static const OptFuncDesc FuncTable[] = {
     { "__bzero",    Opt___bzero,   REG_NONE, OP_X_ZERO | OP_A_KNOWN                    },
     { "staspidx",   Opt_staspidx,  REG_NONE, OP_NONE                                   },
@@ -1379,11 +1347,11 @@ static const OptFuncDesc FuncTable[] = {
 
 static const OptFuncDesc FuncRegATable[] = {
     { "toseqax",    Opt_a_toseq,   REG_NONE, OP_NONE                                   },
-    { "tosgeax",    Opt_a_tosge,   REG_NONE, OP_NONE                                   },
-    { "tosgtax",    Opt_a_tosgt,   REG_NONE, OP_NONE                                   },
+    { "tosgeax",    Opt_a_tosuge,  REG_NONE, OP_NONE                                   },
+    { "tosgtax",    Opt_a_tosugt,  REG_NONE, OP_NONE                                   },
     { "tosicmp",    Opt_a_tosicmp, REG_NONE, OP_NONE                                   },
-    { "tosleax",    Opt_a_tosle,   REG_NONE, OP_NONE                                   },
-    { "tosltax",    Opt_a_toslt,   REG_NONE, OP_NONE                                   },
+    { "tosleax",    Opt_a_tosule,  REG_NONE, OP_NONE                                   },
+    { "tosltax",    Opt_a_tosult,  REG_NONE, OP_NONE                                   },
     { "tosneax",    Opt_a_tosne,   REG_NONE, OP_NONE                                   },
     { "tosugeax",   Opt_a_tosuge,  REG_NONE, OP_NONE                                   },
     { "tosugtax",   Opt_a_tosugt,  REG_NONE, OP_NONE                                   },
diff --git a/test/misc/Makefile b/test/misc/Makefile
index ad31e8554..81a09c693 100644
--- a/test/misc/Makefile
+++ b/test/misc/Makefile
@@ -117,14 +117,6 @@ $(WORKDIR)/endless.$1.$2.prg: endless.c | $(WORKDIR)
 	$(LD65) -t sim$2 -o $$@ $$(@:.prg=.o) sim$2.lib $(NULLERR)
 	$(NOT) $(SIM65) $(SIM65FLAGS) $$@ $(NULLOUT) $(NULLERR)
 
-# this one fails when optimization are enabled
-$(WORKDIR)/bug1348.$1.$2.prg: bug1348.c | $(WORKDIR)
-	$(if $(QUIET),echo misc/bug1348.$1.$2.prg)
-	$(CC65) -Osr -t sim$2 -$1 -o $$(@:.prg=.s) $$< $(NULLERR)
-	$(CA65) -t sim$2 -o $$(@:.prg=.o) $$(@:.prg=.s) $(NULLERR)
-	$(LD65) -t sim$2 -o $$@ $$(@:.prg=.o) sim$2.lib $(NULLERR)
-	$(NOT) $(SIM65) $(SIM65FLAGS) $$@ $(NULLOUT) $(NULLERR)
-
 # these need reference data that can't be generated by a host-compiled program,
 # in a useful way
 $(WORKDIR)/limits.$1.$2.prg: limits.c $(ISEQUAL) | $(WORKDIR)
diff --git a/test/misc/bug1348.c b/test/val/bug1348.c
similarity index 61%
rename from test/misc/bug1348.c
rename to test/val/bug1348.c
index 913849482..c40d9ac84 100644
--- a/test/misc/bug1348.c
+++ b/test/val/bug1348.c
@@ -1,27 +1,26 @@
-
-/* bug#1348, wrongly optimized integer promotion in comparison */
+/* bug #1348, wrongly optimized integer promotion in comparison */
 
 #include <stdio.h>
 
-int notrandtab[] = {
+static const int notrandtab[] = {
     0xffff, 0x7fff, 0x3fff, 0x1fff,
     0x0fff, 0x07ff, 0x03ff, 0x01ff,
     0x00ff, 0x007f, 0x003f, 0x001f,
     0x000f, 0x0007, 0x0003, 0x0001
 };
 
-unsigned char notrandcount = 0;
+static unsigned char notrandcount = 0;
 
-int notrand(void)
+static int notrand(void)
 {
     return notrandtab[notrandcount & 0x0f];
 }
 
-unsigned char n1, n2;
-unsigned char i, ii, s;
-unsigned char err = 0;
+static unsigned char n1, n2;
+static unsigned char i, ii, s;
+static unsigned char err = 0;
 
-unsigned char cmptab[] = {
+static const unsigned char cmptab[] = {
     0xff, 0x7f, 0x3f, 0x1f,
     0x0f, 0x07, 0x03, 0x01,
     0x80, 0x40, 0x20, 0x10,
@@ -40,13 +39,14 @@ int main(void)
             if ((notrand() & 0xffu) > s) {
                 n2 = 1;
             }
-            printf("%5d>%3d %d(%02x) %d(%02x)  %s\n",
-                   notrandtab[notrandcount & 0x0f], s,
+            printf("%5d > %3d %u(%02x) %u(%02x)  %s\n",
+                   notrandtab[i], s,
                    n1, (notrand() & 0xff),
                    n2, (notrand() & 0xffu),
-                   n1 == n2 ? "=" : "!="
-                  );
-            if (n1 != n2) err = 1;
+                   n1 == n2 ? "=" : "!=");
+            if (n1 != n2) {
+                err = 1;
+            }
             notrandcount++;
         }
     }

From dfd047ce6a693a2bc32a8778b06bd08c3d39aa0b Mon Sep 17 00:00:00 2001
From: Jesse Rosenstock <jmr@google.com>
Date: Fri, 13 Nov 2020 09:01:26 +0100
Subject: [PATCH 06/26] g_typeadjust: Use CF_CHAR for char args

If lhs and rhs are either both signed char or both unsigned char,
return flags for that type instead of (unsigned) int.  The flags
are used only for codegen.  Currently, this does nothing, since
codegen treats chars as ints unless CF_FORCECHAR is set, but it
allows more efficient char x char -> int codegen to be added in
the future.
---
 src/cc65/codegen.c | 23 +++++++++++++----------
 1 file changed, 13 insertions(+), 10 deletions(-)

diff --git a/src/cc65/codegen.c b/src/cc65/codegen.c
index f56abcd95..a58484cf1 100644
--- a/src/cc65/codegen.c
+++ b/src/cc65/codegen.c
@@ -1433,17 +1433,20 @@ unsigned g_typeadjust (unsigned lhs, unsigned rhs)
 
     /* Note that this logic is largely duplicated by ArithmeticConvert. */
 
-    /* Before we apply the integral promotions, we check if both types are unsigned char.
-    ** If so, we return unsigned int, rather than int, which would be returned by the standard
-    ** rules.  This is only a performance optimization and does not affect correctness, as
-    ** the flags are only used for code generation, and not to determine types of other
-    ** expressions containing this one.  All unsigned char bit-patterns are valid as both int
-    ** and unsigned int and represent the same value, so either signed or unsigned int operations
-    ** can be used.  This special case part is not duplicated by ArithmeticConvert.
+    /* Before we apply the integral promotions, we check if both types are the same character type.
+    ** If so, we return that type, rather than int, which would be returned by the standard
+    ** rules.  This is only a performance optimization allowing the use of unsigned and/or char
+    ** operations; it does not affect correctness, as the flags are only used for code generation,
+    ** and not to determine types of other expressions containing this one.  For codgen, CF_CHAR
+    ** means the operands are char and the result is int (unless CF_FORCECHAR is also set, in
+    ** which case the result is char).  This special case part is not duplicated by
+    ** ArithmeticConvert.
     */
-    if ((lhs & CF_TYPEMASK) == CF_CHAR && (lhs & CF_UNSIGNED) &&
-        (rhs & CF_TYPEMASK) == CF_CHAR && (rhs & CF_UNSIGNED)) {
-        return const_flag | CF_UNSIGNED | CF_INT;
+    if ((lhs & CF_TYPEMASK) == CF_CHAR && (rhs & CF_TYPEMASK) == CF_CHAR &&
+        (lhs & CF_UNSIGNED) == (rhs & CF_UNSIGNED)) {
+        /* Signedness flags are the same, so just use one of them. */
+        const unsigned unsigned_flag = lhs & CF_UNSIGNED;
+        return const_flag | unsigned_flag | CF_CHAR;
     }
 
     /* Apply integral promotions for types char/short. */

From 1c72da490479b6876557c3cf75577e7ed55f5e15 Mon Sep 17 00:00:00 2001
From: baktrasf <zylon@post.cz>
Date: Mon, 21 Dec 2020 21:31:53 +0100
Subject: [PATCH 07/26] Add operating system symbols for the Atari 5200 target

---
 include/_atari5200os.h | 80 ++++++++++++++++++++++++++++++++++++++++++
 include/atari5200.h    |  4 +++
 2 files changed, 84 insertions(+)
 create mode 100644 include/_atari5200os.h

diff --git a/include/_atari5200os.h b/include/_atari5200os.h
new file mode 100644
index 000000000..77ccf14ed
--- /dev/null
+++ b/include/_atari5200os.h
@@ -0,0 +1,80 @@
+/*****************************************************************************/
+/*                                                                           */
+/*                  _atari5200os.h                                           */
+/*                                                                           */
+/*            Internal include file, do not use directly                     */
+/*                                                                           */
+/*                                                                           */
+/* This software is provided 'as-is', without any expressed or implied       */
+/* warranty.  In no event will the authors be held liable for any damages    */
+/* arising from the use of this software.                                    */
+/*                                                                           */
+/* Permission is granted to anyone to use this software for any purpose,     */
+/* including commercial applications, and to alter it and redistribute it    */
+/* freely, subject to the following restrictions:                            */
+/*                                                                           */
+/* 1. 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.                                       */
+/* 2. Altered source versions must be plainly marked as such, and must not   */
+/*    be misrepresented as being the original software.                      */
+/* 3. This notice may not be removed or altered from any source              */
+/*    distribution.                                                          */
+/*                                                                           */
+/*****************************************************************************/
+
+#ifndef __ATARI5200OS_H
+#define __ATARI5200OS_H
+
+
+struct __os {
+
+    /*Page zero*/
+    unsigned char pokmsk;            // = $00       System mask for POKEY IRQ enable 
+    unsigned char rtclok[2];         // = $01,$02   Real time clock
+    unsigned char critic;            // = $03       Critical section flag 
+    unsigned char atract;            // = $04       Attract mode counter
+    
+    union {
+        struct {
+            unsigned char sdlstl;    // = $05       Save display list LO
+            unsigned char sdlsth;    // = $06       Save display list HI
+        };
+        void*   sdlst;               // = $05,$06   Display list shadow
+    };  
+    
+    unsigned char sdmctl;            // = $07       DMACTL shadow
+    unsigned char pcolr0;            // = $08       PM color 0
+    unsigned char pcolr1;            // = $09       PM color 1
+    unsigned char pcolr2;            // = $0A       PM color 2
+    unsigned char pcolr3;            // = $0B       PM color 3
+    unsigned char color0;            // = $0C       PF Color 0
+    unsigned char color1;            // = $0D       PF Color 1
+    unsigned char color2;            // = $0E       PF Color 2
+    unsigned char color3;            // = $0F       PF Color 3
+    unsigned char color4;            // = $10       PF Color 4
+    unsigned char __filler[0xEF];    // = $11-$FF   Filler
+    
+    /*Stack*/
+    unsigned char stack[0x100];      // = $100-$1FF Stack
+    
+    /*Page 2 OS variables*/
+    void (*vinter)(void);            // = $200      Immediate IRQ vector
+    void (*vvblki)(void);            // = $202      Immediate VBI vector
+    void (*vvblkd)(void);            // = $204      Deferred VBI vector
+    void (*vdslst)(void);            // = $206      DLI vector
+    void (*vkeybd)(void);            // = $208      Keyboard IRQ vector
+    void (*vkeypd)(void);            // = $20A      Keypad continue vector
+    void (*vbrkky)(void);            // = $20C      Break key interrupt vector
+    void (*vbreak)(void);            // = $20E      Break instruction interrupt vector
+    void (*vserin)(void);            // = $210      Serial input ready vector
+    void (*vseror)(void);            // = $212      Serial output data needed vector
+    void (*vseroc)(void);            // = $214      Serial output completed vector
+    void (*vtimr1)(void);            // = $216      POKEY timer 1 IRQ vector
+    void (*vtimr2)(void);            // = $218      POKEY timer 2 IRQ vector
+    void (*vtimr4)(void);            // = $21A      POKEY timer 4 IRQ vector
+    
+};
+
+#endif
\ No newline at end of file
diff --git a/include/atari5200.h b/include/atari5200.h
index a18360c61..ff176c15b 100644
--- a/include/atari5200.h
+++ b/include/atari5200.h
@@ -65,6 +65,10 @@ extern void atr5200std_joy[];        /* referred to by joy_static_stddrv[] */
 #define AT_NTSC     0
 #define AT_PAL      1
 
+/* Define variables used by the OS*/
+#include <_atari5200os.h>
+#define OS (*(struct __os*)0x0000)
+
 /* define hardware */
 #include <_gtia.h>
 #define GTIA_READ  (*(struct __gtia_read*)0xC000)

From 2e9bada1f20b645ea94343adaf03d85da15a1d45 Mon Sep 17 00:00:00 2001
From: baktrasf <zylon@post.cz>
Date: Wed, 23 Dec 2020 23:35:09 +0100
Subject: [PATCH 08/26] Atari 5200 OS header refinements

---
 doc/atari5200.sgml     | 10 ++++++++++
 include/_atari5200os.h | 18 +++++++++---------
 2 files changed, 19 insertions(+), 9 deletions(-)

diff --git a/doc/atari5200.sgml b/doc/atari5200.sgml
index 032b0ef6c..2ec3d08f0 100644
--- a/doc/atari5200.sgml
+++ b/doc/atari5200.sgml
@@ -73,7 +73,17 @@ Special locations:
 
 Programs containing Atari 5200 specific code may use the <tt/atari5200.h/ header file.
 
+This also includes access to operating system locations (e.g. hardware shadow registers) by a structure called
+"<tt/OS/".
+The names are the usual ones you can find in system reference manuals. Example:
 
+<tscreen><verb>
+...
+    OS.sdmctl = 0x00;                       // screen off
+    OS.color4 = 14;                         // white frame
+... tics = OS.rtclok[1]                     // get ticks
+</verb></tscreen>
+ 
 <sect1>Atari 5200 specific functions<p>
 
 <itemize>
diff --git a/include/_atari5200os.h b/include/_atari5200os.h
index 77ccf14ed..db0f7f0c9 100644
--- a/include/_atari5200os.h
+++ b/include/_atari5200os.h
@@ -49,12 +49,12 @@ struct __os {
     unsigned char pcolr1;            // = $09       PM color 1
     unsigned char pcolr2;            // = $0A       PM color 2
     unsigned char pcolr3;            // = $0B       PM color 3
-    unsigned char color0;            // = $0C       PF Color 0
-    unsigned char color1;            // = $0D       PF Color 1
-    unsigned char color2;            // = $0E       PF Color 2
-    unsigned char color3;            // = $0F       PF Color 3
-    unsigned char color4;            // = $10       PF Color 4
-    unsigned char __filler[0xEF];    // = $11-$FF   Filler
+    unsigned char color0;            // = $0C       PF color 0
+    unsigned char color1;            // = $0D       PF color 1
+    unsigned char color2;            // = $0E       PF color 2
+    unsigned char color3;            // = $0F       PF color 3
+    unsigned char color4;            // = $10       PF color 4
+    unsigned char _free_1[0xEF];     // = $11-$FF   User space
     
     /*Stack*/
     unsigned char stack[0x100];      // = $100-$1FF Stack
@@ -65,9 +65,9 @@ struct __os {
     void (*vvblkd)(void);            // = $204      Deferred VBI vector
     void (*vdslst)(void);            // = $206      DLI vector
     void (*vkeybd)(void);            // = $208      Keyboard IRQ vector
-    void (*vkeypd)(void);            // = $20A      Keypad continue vector
+    void (*vkeypd)(void);            // = $20A      Keyboard continuation vector
     void (*vbrkky)(void);            // = $20C      Break key interrupt vector
-    void (*vbreak)(void);            // = $20E      Break instruction interrupt vector
+    void (*vbreak)(void);            // = $20E      BRK instruction interrupt vector
     void (*vserin)(void);            // = $210      Serial input ready vector
     void (*vseror)(void);            // = $212      Serial output data needed vector
     void (*vseroc)(void);            // = $214      Serial output completed vector
@@ -77,4 +77,4 @@ struct __os {
     
 };
 
-#endif
\ No newline at end of file
+#endif

From d67b955e528d032d287b1bdec450643b0d5a8a6c Mon Sep 17 00:00:00 2001
From: michael <zylon@post.cz>
Date: Wed, 23 Dec 2020 23:50:10 +0100
Subject: [PATCH 09/26] Fixed example of the OS struct usage for Atari 5200

---
 doc/atari5200.sgml | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/doc/atari5200.sgml b/doc/atari5200.sgml
index 2ec3d08f0..aff212b15 100644
--- a/doc/atari5200.sgml
+++ b/doc/atari5200.sgml
@@ -81,7 +81,8 @@ The names are the usual ones you can find in system reference manuals. Example:
 ...
     OS.sdmctl = 0x00;                       // screen off
     OS.color4 = 14;                         // white frame
-... tics = OS.rtclok[1]                     // get ticks
+    tics = OS.rtclok[1]                     // get ticks
+...
 </verb></tscreen>
  
 <sect1>Atari 5200 specific functions<p>

From ef258bdc1913b2bc97011e64affb7925b80c1e5a Mon Sep 17 00:00:00 2001
From: Christian Groessler <chris@groessler.org>
Date: Fri, 25 Dec 2020 07:16:26 +0100
Subject: [PATCH 10/26] remove TABs which again slipped in....

---
 src/cc65/expr.h    |    8 +-
 test/val/binlit.c  | 1044 ++++++++++++++++++++++----------------------
 test/val/bug1071.c |   40 +-
 test/val/rand.c    |  134 +++---
 4 files changed, 613 insertions(+), 613 deletions(-)

diff --git a/src/cc65/expr.h b/src/cc65/expr.h
index d0a9988af..71dbe8e0d 100644
--- a/src/cc65/expr.h
+++ b/src/cc65/expr.h
@@ -23,10 +23,10 @@
 
 
 
-#define SQP_KEEP_NONE	0x00
-#define SQP_KEEP_TEST	0x01U
-#define SQP_KEEP_EAX	0x02U
-#define SQP_KEEP_EXPR	0x03U	/* SQP_KEEP_TEST | SQP_KEEP_EAX */
+#define SQP_KEEP_NONE   0x00
+#define SQP_KEEP_TEST   0x01U
+#define SQP_KEEP_EAX    0x02U
+#define SQP_KEEP_EXPR   0x03U   /* SQP_KEEP_TEST | SQP_KEEP_EAX */
 
 
 
diff --git a/test/val/binlit.c b/test/val/binlit.c
index 47e7a196e..f89ca60fc 100644
--- a/test/val/binlit.c
+++ b/test/val/binlit.c
@@ -3,531 +3,531 @@ static unsigned big[256];
 
 int main() {
 
-	unsigned i;
+        unsigned i;
 
-	small[0] = 0b0;
-	small[1] = 0b1;
-	small[2] = 0b10;
-	small[3] = 0b11;
-	small[4] = 0b100;
-	small[5] = 0b101;
-	small[6] = 0b110;
-	small[7] = 0b000111;
-	small[8] = 0b1000;
-	small[9] = 0b1001;
-	small[10] = 0b0001010;
-	small[11] = 0b0001011;
-	small[12] = 0b1100;
-	small[13] = 0b1101;
-	small[14] = 0b1110;
-	small[15] = 0b1111;
-	small[16] = 0b10000;
-	small[17] = 0b10001;
-	small[18] = 0b00010010;
-	small[19] = 0b00010011;
-	small[20] = 0b00010100;
-	small[21] = 0b00010101;
-	small[22] = 0b10110;
-	small[23] = 0b10111;
-	small[24] = 0b11000;
-	small[25] = 0b11001;
-	small[26] = 0b11010;
-	small[27] = 0b11011;
-	small[28] = 0b11100;
-	small[29] = 0b11101;
-	small[30] = 0b11110;
-	small[31] = 0b11111;
-	small[32] = 0b00000000100000;
-	small[33] = 0b00000000100001;
-	small[34] = 0b100010;
-	small[35] = 0b100011;
-	small[36] = 0b100100;
-	small[37] = 0b100101;
-	small[38] = 0b100110;
-	small[39] = 0b100111;
-	small[40] = 0b101000;
-	small[41] = 0b101001;
-	small[42] = 0b101010;
-	small[43] = 0b101011;
-	small[44] = 0b101100;
-	small[45] = 0b101101;
-	small[46] = 0b101110;
-	small[47] = 0b101111;
-	small[48] = 0b110000;
-	small[49] = 0b110001;
-	small[50] = 0b110010;
-	small[51] = 0b110011;
-	small[52] = 0b110100;
-	small[53] = 0b110101;
-	small[54] = 0b110110;
-	small[55] = 0b110111;
-	small[56] = 0b111000;
-	small[57] = 0b111001;
-	small[58] = 0b111010;
-	small[59] = 0b111011;
-	small[60] = 0b111100;
-	small[61] = 0b111101;
-	small[62] = 0b111110;
-	small[63] = 0b111111;
-	small[64] = 0b1000000;
-	small[65] = 0b1000001;
-	small[66] = 0b1000010;
-	small[67] = 0b1000011;
-	small[68] = 0b1000100;
-	small[69] = 0b1000101;
-	small[70] = 0b1000110;
-	small[71] = 0b1000111;
-	small[72] = 0b1001000;
-	small[73] = 0b1001001;
-	small[74] = 0b1001010;
-	small[75] = 0b1001011;
-	small[76] = 0b1001100;
-	small[77] = 0b1001101;
-	small[78] = 0b1001110;
-	small[79] = 0b1001111;
-	small[80] = 0b1010000;
-	small[81] = 0b1010001;
-	small[82] = 0b1010010;
-	small[83] = 0b1010011;
-	small[84] = 0b1010100;
-	small[85] = 0b1010101;
-	small[86] = 0b1010110;
-	small[87] = 0b1010111;
-	small[88] = 0b1011000;
-	small[89] = 0b1011001;
-	small[90] = 0b1011010;
-	small[91] = 0b1011011;
-	small[92] = 0b1011100;
-	small[93] = 0b1011101;
-	small[94] = 0b1011110;
-	small[95] = 0b1011111;
-	small[96] = 0b1100000;
-	small[97] = 0b1100001;
-	small[98] = 0b1100010;
-	small[99] = 0b1100011;
-	small[100] = 0b1100100;
-	small[101] = 0b1100101;
-	small[102] = 0b1100110;
-	small[103] = 0b1100111;
-	small[104] = 0b1101000;
-	small[105] = 0b1101001;
-	small[106] = 0b1101010;
-	small[107] = 0b1101011;
-	small[108] = 0b1101100;
-	small[109] = 0b1101101;
-	small[110] = 0b1101110;
-	small[111] = 0b1101111;
-	small[112] = 0b1110000;
-	small[113] = 0b1110001;
-	small[114] = 0b1110010;
-	small[115] = 0b1110011;
-	small[116] = 0b1110100;
-	small[117] = 0b1110101;
-	small[118] = 0b1110110;
-	small[119] = 0b1110111;
-	small[120] = 0b1111000;
-	small[121] = 0b1111001;
-	small[122] = 0b1111010;
-	small[123] = 0b1111011;
-	small[124] = 0b1111100;
-	small[125] = 0b1111101;
-	small[126] = 0b1111110;
-	small[127] = 0b1111111;
-	small[128] = 0b10000000;
-	small[129] = 0b10000001;
-	small[130] = 0b10000010;
-	small[131] = 0b10000011;
-	small[132] = 0b10000100;
-	small[133] = 0b10000101;
-	small[134] = 0b10000110;
-	small[135] = 0b10000111;
-	small[136] = 0b10001000;
-	small[137] = 0b10001001;
-	small[138] = 0b10001010;
-	small[139] = 0b10001011;
-	small[140] = 0b10001100;
-	small[141] = 0b10001101;
-	small[142] = 0b10001110;
-	small[143] = 0b10001111;
-	small[144] = 0b10010000;
-	small[145] = 0b10010001;
-	small[146] = 0b10010010;
-	small[147] = 0b10010011;
-	small[148] = 0b10010100;
-	small[149] = 0b10010101;
-	small[150] = 0b10010110;
-	small[151] = 0b10010111;
-	small[152] = 0b10011000;
-	small[153] = 0b10011001;
-	small[154] = 0b10011010;
-	small[155] = 0b10011011;
-	small[156] = 0b10011100;
-	small[157] = 0b10011101;
-	small[158] = 0b10011110;
-	small[159] = 0b10011111;
-	small[160] = 0b10100000;
-	small[161] = 0b10100001;
-	small[162] = 0b10100010;
-	small[163] = 0b10100011;
-	small[164] = 0b10100100;
-	small[165] = 0b10100101;
-	small[166] = 0b10100110;
-	small[167] = 0b10100111;
-	small[168] = 0b10101000;
-	small[169] = 0b10101001;
-	small[170] = 0b10101010;
-	small[171] = 0b10101011;
-	small[172] = 0b10101100;
-	small[173] = 0b10101101;
-	small[174] = 0b10101110;
-	small[175] = 0b10101111;
-	small[176] = 0b10110000;
-	small[177] = 0b10110001;
-	small[178] = 0b10110010;
-	small[179] = 0b10110011;
-	small[180] = 0b10110100;
-	small[181] = 0b10110101;
-	small[182] = 0b10110110;
-	small[183] = 0b10110111;
-	small[184] = 0b10111000;
-	small[185] = 0b10111001;
-	small[186] = 0b10111010;
-	small[187] = 0b10111011;
-	small[188] = 0b10111100;
-	small[189] = 0b10111101;
-	small[190] = 0b10111110;
-	small[191] = 0b10111111;
-	small[192] = 0b11000000;
-	small[193] = 0b11000001;
-	small[194] = 0b11000010;
-	small[195] = 0b11000011;
-	small[196] = 0b11000100;
-	small[197] = 0b11000101;
-	small[198] = 0b11000110;
-	small[199] = 0b11000111;
-	small[200] = 0b11001000;
-	small[201] = 0b11001001;
-	small[202] = 0b11001010;
-	small[203] = 0b11001011;
-	small[204] = 0b11001100;
-	small[205] = 0b11001101;
-	small[206] = 0b11001110;
-	small[207] = 0b11001111;
-	small[208] = 0b11010000;
-	small[209] = 0b11010001;
-	small[210] = 0b11010010;
-	small[211] = 0b11010011;
-	small[212] = 0b11010100;
-	small[213] = 0b11010101;
-	small[214] = 0b11010110;
-	small[215] = 0b11010111;
-	small[216] = 0b11011000;
-	small[217] = 0b11011001;
-	small[218] = 0b11011010;
-	small[219] = 0b11011011;
-	small[220] = 0b11011100;
-	small[221] = 0b11011101;
-	small[222] = 0b11011110;
-	small[223] = 0b11011111;
-	small[224] = 0b11100000;
-	small[225] = 0b11100001;
-	small[226] = 0b11100010;
-	small[227] = 0b11100011;
-	small[228] = 0b11100100;
-	small[229] = 0b11100101;
-	small[230] = 0b11100110;
-	small[231] = 0b11100111;
-	small[232] = 0b11101000;
-	small[233] = 0b11101001;
-	small[234] = 0b11101010;
-	small[235] = 0b11101011;
-	small[236] = 0b11101100;
-	small[237] = 0b11101101;
-	small[238] = 0b11101110;
-	small[239] = 0b11101111;
-	small[240] = 0b11110000;
-	small[241] = 0b11110001;
-	small[242] = 0b11110010;
-	small[243] = 0b11110011;
-	small[244] = 0b11110100;
-	small[245] = 0b11110101;
-	small[246] = 0b11110110;
-	small[247] = 0b11110111;
-	small[248] = 0b11111000;
-	small[249] = 0b11111001;
-	small[250] = 0b11111010;
-	small[251] = 0b11111011;
-	small[252] = 0b11111100;
-	small[253] = 0b11111101;
-	small[254] = 0b11111110;
-	small[255] = 0b11111111;
+        small[0] = 0b0;
+        small[1] = 0b1;
+        small[2] = 0b10;
+        small[3] = 0b11;
+        small[4] = 0b100;
+        small[5] = 0b101;
+        small[6] = 0b110;
+        small[7] = 0b000111;
+        small[8] = 0b1000;
+        small[9] = 0b1001;
+        small[10] = 0b0001010;
+        small[11] = 0b0001011;
+        small[12] = 0b1100;
+        small[13] = 0b1101;
+        small[14] = 0b1110;
+        small[15] = 0b1111;
+        small[16] = 0b10000;
+        small[17] = 0b10001;
+        small[18] = 0b00010010;
+        small[19] = 0b00010011;
+        small[20] = 0b00010100;
+        small[21] = 0b00010101;
+        small[22] = 0b10110;
+        small[23] = 0b10111;
+        small[24] = 0b11000;
+        small[25] = 0b11001;
+        small[26] = 0b11010;
+        small[27] = 0b11011;
+        small[28] = 0b11100;
+        small[29] = 0b11101;
+        small[30] = 0b11110;
+        small[31] = 0b11111;
+        small[32] = 0b00000000100000;
+        small[33] = 0b00000000100001;
+        small[34] = 0b100010;
+        small[35] = 0b100011;
+        small[36] = 0b100100;
+        small[37] = 0b100101;
+        small[38] = 0b100110;
+        small[39] = 0b100111;
+        small[40] = 0b101000;
+        small[41] = 0b101001;
+        small[42] = 0b101010;
+        small[43] = 0b101011;
+        small[44] = 0b101100;
+        small[45] = 0b101101;
+        small[46] = 0b101110;
+        small[47] = 0b101111;
+        small[48] = 0b110000;
+        small[49] = 0b110001;
+        small[50] = 0b110010;
+        small[51] = 0b110011;
+        small[52] = 0b110100;
+        small[53] = 0b110101;
+        small[54] = 0b110110;
+        small[55] = 0b110111;
+        small[56] = 0b111000;
+        small[57] = 0b111001;
+        small[58] = 0b111010;
+        small[59] = 0b111011;
+        small[60] = 0b111100;
+        small[61] = 0b111101;
+        small[62] = 0b111110;
+        small[63] = 0b111111;
+        small[64] = 0b1000000;
+        small[65] = 0b1000001;
+        small[66] = 0b1000010;
+        small[67] = 0b1000011;
+        small[68] = 0b1000100;
+        small[69] = 0b1000101;
+        small[70] = 0b1000110;
+        small[71] = 0b1000111;
+        small[72] = 0b1001000;
+        small[73] = 0b1001001;
+        small[74] = 0b1001010;
+        small[75] = 0b1001011;
+        small[76] = 0b1001100;
+        small[77] = 0b1001101;
+        small[78] = 0b1001110;
+        small[79] = 0b1001111;
+        small[80] = 0b1010000;
+        small[81] = 0b1010001;
+        small[82] = 0b1010010;
+        small[83] = 0b1010011;
+        small[84] = 0b1010100;
+        small[85] = 0b1010101;
+        small[86] = 0b1010110;
+        small[87] = 0b1010111;
+        small[88] = 0b1011000;
+        small[89] = 0b1011001;
+        small[90] = 0b1011010;
+        small[91] = 0b1011011;
+        small[92] = 0b1011100;
+        small[93] = 0b1011101;
+        small[94] = 0b1011110;
+        small[95] = 0b1011111;
+        small[96] = 0b1100000;
+        small[97] = 0b1100001;
+        small[98] = 0b1100010;
+        small[99] = 0b1100011;
+        small[100] = 0b1100100;
+        small[101] = 0b1100101;
+        small[102] = 0b1100110;
+        small[103] = 0b1100111;
+        small[104] = 0b1101000;
+        small[105] = 0b1101001;
+        small[106] = 0b1101010;
+        small[107] = 0b1101011;
+        small[108] = 0b1101100;
+        small[109] = 0b1101101;
+        small[110] = 0b1101110;
+        small[111] = 0b1101111;
+        small[112] = 0b1110000;
+        small[113] = 0b1110001;
+        small[114] = 0b1110010;
+        small[115] = 0b1110011;
+        small[116] = 0b1110100;
+        small[117] = 0b1110101;
+        small[118] = 0b1110110;
+        small[119] = 0b1110111;
+        small[120] = 0b1111000;
+        small[121] = 0b1111001;
+        small[122] = 0b1111010;
+        small[123] = 0b1111011;
+        small[124] = 0b1111100;
+        small[125] = 0b1111101;
+        small[126] = 0b1111110;
+        small[127] = 0b1111111;
+        small[128] = 0b10000000;
+        small[129] = 0b10000001;
+        small[130] = 0b10000010;
+        small[131] = 0b10000011;
+        small[132] = 0b10000100;
+        small[133] = 0b10000101;
+        small[134] = 0b10000110;
+        small[135] = 0b10000111;
+        small[136] = 0b10001000;
+        small[137] = 0b10001001;
+        small[138] = 0b10001010;
+        small[139] = 0b10001011;
+        small[140] = 0b10001100;
+        small[141] = 0b10001101;
+        small[142] = 0b10001110;
+        small[143] = 0b10001111;
+        small[144] = 0b10010000;
+        small[145] = 0b10010001;
+        small[146] = 0b10010010;
+        small[147] = 0b10010011;
+        small[148] = 0b10010100;
+        small[149] = 0b10010101;
+        small[150] = 0b10010110;
+        small[151] = 0b10010111;
+        small[152] = 0b10011000;
+        small[153] = 0b10011001;
+        small[154] = 0b10011010;
+        small[155] = 0b10011011;
+        small[156] = 0b10011100;
+        small[157] = 0b10011101;
+        small[158] = 0b10011110;
+        small[159] = 0b10011111;
+        small[160] = 0b10100000;
+        small[161] = 0b10100001;
+        small[162] = 0b10100010;
+        small[163] = 0b10100011;
+        small[164] = 0b10100100;
+        small[165] = 0b10100101;
+        small[166] = 0b10100110;
+        small[167] = 0b10100111;
+        small[168] = 0b10101000;
+        small[169] = 0b10101001;
+        small[170] = 0b10101010;
+        small[171] = 0b10101011;
+        small[172] = 0b10101100;
+        small[173] = 0b10101101;
+        small[174] = 0b10101110;
+        small[175] = 0b10101111;
+        small[176] = 0b10110000;
+        small[177] = 0b10110001;
+        small[178] = 0b10110010;
+        small[179] = 0b10110011;
+        small[180] = 0b10110100;
+        small[181] = 0b10110101;
+        small[182] = 0b10110110;
+        small[183] = 0b10110111;
+        small[184] = 0b10111000;
+        small[185] = 0b10111001;
+        small[186] = 0b10111010;
+        small[187] = 0b10111011;
+        small[188] = 0b10111100;
+        small[189] = 0b10111101;
+        small[190] = 0b10111110;
+        small[191] = 0b10111111;
+        small[192] = 0b11000000;
+        small[193] = 0b11000001;
+        small[194] = 0b11000010;
+        small[195] = 0b11000011;
+        small[196] = 0b11000100;
+        small[197] = 0b11000101;
+        small[198] = 0b11000110;
+        small[199] = 0b11000111;
+        small[200] = 0b11001000;
+        small[201] = 0b11001001;
+        small[202] = 0b11001010;
+        small[203] = 0b11001011;
+        small[204] = 0b11001100;
+        small[205] = 0b11001101;
+        small[206] = 0b11001110;
+        small[207] = 0b11001111;
+        small[208] = 0b11010000;
+        small[209] = 0b11010001;
+        small[210] = 0b11010010;
+        small[211] = 0b11010011;
+        small[212] = 0b11010100;
+        small[213] = 0b11010101;
+        small[214] = 0b11010110;
+        small[215] = 0b11010111;
+        small[216] = 0b11011000;
+        small[217] = 0b11011001;
+        small[218] = 0b11011010;
+        small[219] = 0b11011011;
+        small[220] = 0b11011100;
+        small[221] = 0b11011101;
+        small[222] = 0b11011110;
+        small[223] = 0b11011111;
+        small[224] = 0b11100000;
+        small[225] = 0b11100001;
+        small[226] = 0b11100010;
+        small[227] = 0b11100011;
+        small[228] = 0b11100100;
+        small[229] = 0b11100101;
+        small[230] = 0b11100110;
+        small[231] = 0b11100111;
+        small[232] = 0b11101000;
+        small[233] = 0b11101001;
+        small[234] = 0b11101010;
+        small[235] = 0b11101011;
+        small[236] = 0b11101100;
+        small[237] = 0b11101101;
+        small[238] = 0b11101110;
+        small[239] = 0b11101111;
+        small[240] = 0b11110000;
+        small[241] = 0b11110001;
+        small[242] = 0b11110010;
+        small[243] = 0b11110011;
+        small[244] = 0b11110100;
+        small[245] = 0b11110101;
+        small[246] = 0b11110110;
+        small[247] = 0b11110111;
+        small[248] = 0b11111000;
+        small[249] = 0b11111001;
+        small[250] = 0b11111010;
+        small[251] = 0b11111011;
+        small[252] = 0b11111100;
+        small[253] = 0b11111101;
+        small[254] = 0b11111110;
+        small[255] = 0b11111111;
 
-	for (i = 0; i < 256; i++) {
-		if (small[i] != i)
-			return 1;
-	}
+        for (i = 0; i < 256; i++) {
+                if (small[i] != i)
+                        return 1;
+        }
 
-	big[0] = 0b1111111100000000;
-	big[1] = 0b1111111100000001;
-	big[2] = 0b1111111100000010;
-	big[3] = 0b1111111100000011;
-	big[4] = 0b1111111100000100;
-	big[5] = 0b1111111100000101;
-	big[6] = 0b1111111100000110;
-	big[7] = 0b1111111100000111;
-	big[8] = 0b1111111100001000;
-	big[9] = 0b1111111100001001;
-	big[10] = 0b1111111100001010;
-	big[11] = 0b1111111100001011;
-	big[12] = 0b1111111100001100;
-	big[13] = 0b1111111100001101;
-	big[14] = 0b1111111100001110;
-	big[15] = 0b1111111100001111;
-	big[16] = 0b1111111100010000;
-	big[17] = 0b1111111100010001;
-	big[18] = 0b1111111100010010;
-	big[19] = 0b1111111100010011;
-	big[20] = 0b1111111100010100;
-	big[21] = 0b1111111100010101;
-	big[22] = 0b1111111100010110;
-	big[23] = 0b1111111100010111;
-	big[24] = 0b1111111100011000;
-	big[25] = 0b1111111100011001;
-	big[26] = 0b1111111100011010;
-	big[27] = 0b1111111100011011;
-	big[28] = 0b1111111100011100;
-	big[29] = 0b1111111100011101;
-	big[30] = 0b1111111100011110;
-	big[31] = 0b1111111100011111;
-	big[32] = 0b1111111100100000;
-	big[33] = 0b1111111100100001;
-	big[34] = 0b1111111100100010;
-	big[35] = 0b1111111100100011;
-	big[36] = 0b1111111100100100;
-	big[37] = 0b1111111100100101;
-	big[38] = 0b1111111100100110;
-	big[39] = 0b1111111100100111;
-	big[40] = 0b1111111100101000;
-	big[41] = 0b1111111100101001;
-	big[42] = 0b1111111100101010;
-	big[43] = 0b1111111100101011;
-	big[44] = 0b1111111100101100;
-	big[45] = 0b1111111100101101;
-	big[46] = 0b1111111100101110;
-	big[47] = 0b1111111100101111;
-	big[48] = 0b1111111100110000;
-	big[49] = 0b1111111100110001;
-	big[50] = 0b1111111100110010;
-	big[51] = 0b1111111100110011;
-	big[52] = 0b1111111100110100;
-	big[53] = 0b1111111100110101;
-	big[54] = 0b1111111100110110;
-	big[55] = 0b1111111100110111;
-	big[56] = 0b1111111100111000;
-	big[57] = 0b1111111100111001;
-	big[58] = 0b1111111100111010;
-	big[59] = 0b1111111100111011;
-	big[60] = 0b1111111100111100;
-	big[61] = 0b1111111100111101;
-	big[62] = 0b1111111100111110;
-	big[63] = 0b1111111100111111;
-	big[64] = 0b1111111101000000;
-	big[65] = 0b1111111101000001;
-	big[66] = 0b1111111101000010;
-	big[67] = 0b1111111101000011;
-	big[68] = 0b1111111101000100;
-	big[69] = 0b1111111101000101;
-	big[70] = 0b1111111101000110;
-	big[71] = 0b1111111101000111;
-	big[72] = 0b1111111101001000;
-	big[73] = 0b1111111101001001;
-	big[74] = 0b1111111101001010;
-	big[75] = 0b1111111101001011;
-	big[76] = 0b1111111101001100;
-	big[77] = 0b1111111101001101;
-	big[78] = 0b1111111101001110;
-	big[79] = 0b1111111101001111;
-	big[80] = 0b1111111101010000;
-	big[81] = 0b1111111101010001;
-	big[82] = 0b1111111101010010;
-	big[83] = 0b1111111101010011;
-	big[84] = 0b1111111101010100;
-	big[85] = 0b1111111101010101;
-	big[86] = 0b1111111101010110;
-	big[87] = 0b1111111101010111;
-	big[88] = 0b1111111101011000;
-	big[89] = 0b1111111101011001;
-	big[90] = 0b1111111101011010;
-	big[91] = 0b1111111101011011;
-	big[92] = 0b1111111101011100;
-	big[93] = 0b1111111101011101;
-	big[94] = 0b1111111101011110;
-	big[95] = 0b1111111101011111;
-	big[96] = 0b1111111101100000;
-	big[97] = 0b1111111101100001;
-	big[98] = 0b1111111101100010;
-	big[99] = 0b1111111101100011;
-	big[100] = 0b1111111101100100;
-	big[101] = 0b1111111101100101;
-	big[102] = 0b1111111101100110;
-	big[103] = 0b1111111101100111;
-	big[104] = 0b1111111101101000;
-	big[105] = 0b1111111101101001;
-	big[106] = 0b1111111101101010;
-	big[107] = 0b1111111101101011;
-	big[108] = 0b1111111101101100;
-	big[109] = 0b1111111101101101;
-	big[110] = 0b1111111101101110;
-	big[111] = 0b1111111101101111;
-	big[112] = 0b1111111101110000;
-	big[113] = 0b1111111101110001;
-	big[114] = 0b1111111101110010;
-	big[115] = 0b1111111101110011;
-	big[116] = 0b1111111101110100;
-	big[117] = 0b1111111101110101;
-	big[118] = 0b1111111101110110;
-	big[119] = 0b1111111101110111;
-	big[120] = 0b1111111101111000;
-	big[121] = 0b1111111101111001;
-	big[122] = 0b1111111101111010;
-	big[123] = 0b1111111101111011;
-	big[124] = 0b1111111101111100;
-	big[125] = 0b1111111101111101;
-	big[126] = 0b1111111101111110;
-	big[127] = 0b1111111101111111;
-	big[128] = 0b1111111110000000;
-	big[129] = 0b1111111110000001;
-	big[130] = 0b1111111110000010;
-	big[131] = 0b1111111110000011;
-	big[132] = 0b1111111110000100;
-	big[133] = 0b1111111110000101;
-	big[134] = 0b1111111110000110;
-	big[135] = 0b1111111110000111;
-	big[136] = 0b1111111110001000;
-	big[137] = 0b1111111110001001;
-	big[138] = 0b1111111110001010;
-	big[139] = 0b1111111110001011;
-	big[140] = 0b1111111110001100;
-	big[141] = 0b1111111110001101;
-	big[142] = 0b1111111110001110;
-	big[143] = 0b1111111110001111;
-	big[144] = 0b1111111110010000;
-	big[145] = 0b1111111110010001;
-	big[146] = 0b1111111110010010;
-	big[147] = 0b1111111110010011;
-	big[148] = 0b1111111110010100;
-	big[149] = 0b1111111110010101;
-	big[150] = 0b1111111110010110;
-	big[151] = 0b1111111110010111;
-	big[152] = 0b1111111110011000;
-	big[153] = 0b1111111110011001;
-	big[154] = 0b1111111110011010;
-	big[155] = 0b1111111110011011;
-	big[156] = 0b1111111110011100;
-	big[157] = 0b1111111110011101;
-	big[158] = 0b1111111110011110;
-	big[159] = 0b1111111110011111;
-	big[160] = 0b1111111110100000;
-	big[161] = 0b1111111110100001;
-	big[162] = 0b1111111110100010;
-	big[163] = 0b1111111110100011;
-	big[164] = 0b1111111110100100;
-	big[165] = 0b1111111110100101;
-	big[166] = 0b1111111110100110;
-	big[167] = 0b1111111110100111;
-	big[168] = 0b1111111110101000;
-	big[169] = 0b1111111110101001;
-	big[170] = 0b1111111110101010;
-	big[171] = 0b1111111110101011;
-	big[172] = 0b1111111110101100;
-	big[173] = 0b1111111110101101;
-	big[174] = 0b1111111110101110;
-	big[175] = 0b1111111110101111;
-	big[176] = 0b1111111110110000;
-	big[177] = 0b1111111110110001;
-	big[178] = 0b1111111110110010;
-	big[179] = 0b1111111110110011;
-	big[180] = 0b1111111110110100;
-	big[181] = 0b1111111110110101;
-	big[182] = 0b1111111110110110;
-	big[183] = 0b1111111110110111;
-	big[184] = 0b1111111110111000;
-	big[185] = 0b1111111110111001;
-	big[186] = 0b1111111110111010;
-	big[187] = 0b1111111110111011;
-	big[188] = 0b1111111110111100;
-	big[189] = 0b1111111110111101;
-	big[190] = 0b1111111110111110;
-	big[191] = 0b1111111110111111;
-	big[192] = 0b1111111111000000;
-	big[193] = 0b1111111111000001;
-	big[194] = 0b1111111111000010;
-	big[195] = 0b1111111111000011;
-	big[196] = 0b1111111111000100;
-	big[197] = 0b1111111111000101;
-	big[198] = 0b1111111111000110;
-	big[199] = 0b1111111111000111;
-	big[200] = 0b1111111111001000;
-	big[201] = 0b1111111111001001;
-	big[202] = 0b1111111111001010;
-	big[203] = 0b1111111111001011;
-	big[204] = 0b1111111111001100;
-	big[205] = 0b1111111111001101;
-	big[206] = 0b1111111111001110;
-	big[207] = 0b1111111111001111;
-	big[208] = 0b1111111111010000;
-	big[209] = 0b1111111111010001;
-	big[210] = 0b1111111111010010;
-	big[211] = 0b1111111111010011;
-	big[212] = 0b1111111111010100;
-	big[213] = 0b1111111111010101;
-	big[214] = 0b1111111111010110;
-	big[215] = 0b1111111111010111;
-	big[216] = 0b1111111111011000;
-	big[217] = 0b1111111111011001;
-	big[218] = 0b1111111111011010;
-	big[219] = 0b1111111111011011;
-	big[220] = 0b1111111111011100;
-	big[221] = 0b1111111111011101;
-	big[222] = 0b1111111111011110;
-	big[223] = 0b1111111111011111;
-	big[224] = 0b1111111111100000;
-	big[225] = 0b1111111111100001;
-	big[226] = 0b1111111111100010;
-	big[227] = 0b1111111111100011;
-	big[228] = 0b1111111111100100;
-	big[229] = 0b1111111111100101;
-	big[230] = 0b1111111111100110;
-	big[231] = 0b1111111111100111;
-	big[232] = 0b1111111111101000;
-	big[233] = 0b1111111111101001;
-	big[234] = 0b1111111111101010;
-	big[235] = 0b1111111111101011;
-	big[236] = 0b1111111111101100;
-	big[237] = 0b1111111111101101;
-	big[238] = 0b1111111111101110;
-	big[239] = 0b1111111111101111;
-	big[240] = 0b1111111111110000;
-	big[241] = 0b1111111111110001;
-	big[242] = 0b1111111111110010;
-	big[243] = 0b1111111111110011;
-	big[244] = 0b1111111111110100;
-	big[245] = 0b1111111111110101;
-	big[246] = 0b1111111111110110;
-	big[247] = 0b1111111111110111;
-	big[248] = 0b1111111111111000;
-	big[249] = 0b1111111111111001;
-	big[250] = 0b1111111111111010;
-	big[251] = 0b1111111111111011;
-	big[252] = 0b1111111111111100;
-	big[253] = 0b1111111111111101;
-	big[254] = 0b1111111111111110;
-	big[255] = 0b1111111111111111;
+        big[0] = 0b1111111100000000;
+        big[1] = 0b1111111100000001;
+        big[2] = 0b1111111100000010;
+        big[3] = 0b1111111100000011;
+        big[4] = 0b1111111100000100;
+        big[5] = 0b1111111100000101;
+        big[6] = 0b1111111100000110;
+        big[7] = 0b1111111100000111;
+        big[8] = 0b1111111100001000;
+        big[9] = 0b1111111100001001;
+        big[10] = 0b1111111100001010;
+        big[11] = 0b1111111100001011;
+        big[12] = 0b1111111100001100;
+        big[13] = 0b1111111100001101;
+        big[14] = 0b1111111100001110;
+        big[15] = 0b1111111100001111;
+        big[16] = 0b1111111100010000;
+        big[17] = 0b1111111100010001;
+        big[18] = 0b1111111100010010;
+        big[19] = 0b1111111100010011;
+        big[20] = 0b1111111100010100;
+        big[21] = 0b1111111100010101;
+        big[22] = 0b1111111100010110;
+        big[23] = 0b1111111100010111;
+        big[24] = 0b1111111100011000;
+        big[25] = 0b1111111100011001;
+        big[26] = 0b1111111100011010;
+        big[27] = 0b1111111100011011;
+        big[28] = 0b1111111100011100;
+        big[29] = 0b1111111100011101;
+        big[30] = 0b1111111100011110;
+        big[31] = 0b1111111100011111;
+        big[32] = 0b1111111100100000;
+        big[33] = 0b1111111100100001;
+        big[34] = 0b1111111100100010;
+        big[35] = 0b1111111100100011;
+        big[36] = 0b1111111100100100;
+        big[37] = 0b1111111100100101;
+        big[38] = 0b1111111100100110;
+        big[39] = 0b1111111100100111;
+        big[40] = 0b1111111100101000;
+        big[41] = 0b1111111100101001;
+        big[42] = 0b1111111100101010;
+        big[43] = 0b1111111100101011;
+        big[44] = 0b1111111100101100;
+        big[45] = 0b1111111100101101;
+        big[46] = 0b1111111100101110;
+        big[47] = 0b1111111100101111;
+        big[48] = 0b1111111100110000;
+        big[49] = 0b1111111100110001;
+        big[50] = 0b1111111100110010;
+        big[51] = 0b1111111100110011;
+        big[52] = 0b1111111100110100;
+        big[53] = 0b1111111100110101;
+        big[54] = 0b1111111100110110;
+        big[55] = 0b1111111100110111;
+        big[56] = 0b1111111100111000;
+        big[57] = 0b1111111100111001;
+        big[58] = 0b1111111100111010;
+        big[59] = 0b1111111100111011;
+        big[60] = 0b1111111100111100;
+        big[61] = 0b1111111100111101;
+        big[62] = 0b1111111100111110;
+        big[63] = 0b1111111100111111;
+        big[64] = 0b1111111101000000;
+        big[65] = 0b1111111101000001;
+        big[66] = 0b1111111101000010;
+        big[67] = 0b1111111101000011;
+        big[68] = 0b1111111101000100;
+        big[69] = 0b1111111101000101;
+        big[70] = 0b1111111101000110;
+        big[71] = 0b1111111101000111;
+        big[72] = 0b1111111101001000;
+        big[73] = 0b1111111101001001;
+        big[74] = 0b1111111101001010;
+        big[75] = 0b1111111101001011;
+        big[76] = 0b1111111101001100;
+        big[77] = 0b1111111101001101;
+        big[78] = 0b1111111101001110;
+        big[79] = 0b1111111101001111;
+        big[80] = 0b1111111101010000;
+        big[81] = 0b1111111101010001;
+        big[82] = 0b1111111101010010;
+        big[83] = 0b1111111101010011;
+        big[84] = 0b1111111101010100;
+        big[85] = 0b1111111101010101;
+        big[86] = 0b1111111101010110;
+        big[87] = 0b1111111101010111;
+        big[88] = 0b1111111101011000;
+        big[89] = 0b1111111101011001;
+        big[90] = 0b1111111101011010;
+        big[91] = 0b1111111101011011;
+        big[92] = 0b1111111101011100;
+        big[93] = 0b1111111101011101;
+        big[94] = 0b1111111101011110;
+        big[95] = 0b1111111101011111;
+        big[96] = 0b1111111101100000;
+        big[97] = 0b1111111101100001;
+        big[98] = 0b1111111101100010;
+        big[99] = 0b1111111101100011;
+        big[100] = 0b1111111101100100;
+        big[101] = 0b1111111101100101;
+        big[102] = 0b1111111101100110;
+        big[103] = 0b1111111101100111;
+        big[104] = 0b1111111101101000;
+        big[105] = 0b1111111101101001;
+        big[106] = 0b1111111101101010;
+        big[107] = 0b1111111101101011;
+        big[108] = 0b1111111101101100;
+        big[109] = 0b1111111101101101;
+        big[110] = 0b1111111101101110;
+        big[111] = 0b1111111101101111;
+        big[112] = 0b1111111101110000;
+        big[113] = 0b1111111101110001;
+        big[114] = 0b1111111101110010;
+        big[115] = 0b1111111101110011;
+        big[116] = 0b1111111101110100;
+        big[117] = 0b1111111101110101;
+        big[118] = 0b1111111101110110;
+        big[119] = 0b1111111101110111;
+        big[120] = 0b1111111101111000;
+        big[121] = 0b1111111101111001;
+        big[122] = 0b1111111101111010;
+        big[123] = 0b1111111101111011;
+        big[124] = 0b1111111101111100;
+        big[125] = 0b1111111101111101;
+        big[126] = 0b1111111101111110;
+        big[127] = 0b1111111101111111;
+        big[128] = 0b1111111110000000;
+        big[129] = 0b1111111110000001;
+        big[130] = 0b1111111110000010;
+        big[131] = 0b1111111110000011;
+        big[132] = 0b1111111110000100;
+        big[133] = 0b1111111110000101;
+        big[134] = 0b1111111110000110;
+        big[135] = 0b1111111110000111;
+        big[136] = 0b1111111110001000;
+        big[137] = 0b1111111110001001;
+        big[138] = 0b1111111110001010;
+        big[139] = 0b1111111110001011;
+        big[140] = 0b1111111110001100;
+        big[141] = 0b1111111110001101;
+        big[142] = 0b1111111110001110;
+        big[143] = 0b1111111110001111;
+        big[144] = 0b1111111110010000;
+        big[145] = 0b1111111110010001;
+        big[146] = 0b1111111110010010;
+        big[147] = 0b1111111110010011;
+        big[148] = 0b1111111110010100;
+        big[149] = 0b1111111110010101;
+        big[150] = 0b1111111110010110;
+        big[151] = 0b1111111110010111;
+        big[152] = 0b1111111110011000;
+        big[153] = 0b1111111110011001;
+        big[154] = 0b1111111110011010;
+        big[155] = 0b1111111110011011;
+        big[156] = 0b1111111110011100;
+        big[157] = 0b1111111110011101;
+        big[158] = 0b1111111110011110;
+        big[159] = 0b1111111110011111;
+        big[160] = 0b1111111110100000;
+        big[161] = 0b1111111110100001;
+        big[162] = 0b1111111110100010;
+        big[163] = 0b1111111110100011;
+        big[164] = 0b1111111110100100;
+        big[165] = 0b1111111110100101;
+        big[166] = 0b1111111110100110;
+        big[167] = 0b1111111110100111;
+        big[168] = 0b1111111110101000;
+        big[169] = 0b1111111110101001;
+        big[170] = 0b1111111110101010;
+        big[171] = 0b1111111110101011;
+        big[172] = 0b1111111110101100;
+        big[173] = 0b1111111110101101;
+        big[174] = 0b1111111110101110;
+        big[175] = 0b1111111110101111;
+        big[176] = 0b1111111110110000;
+        big[177] = 0b1111111110110001;
+        big[178] = 0b1111111110110010;
+        big[179] = 0b1111111110110011;
+        big[180] = 0b1111111110110100;
+        big[181] = 0b1111111110110101;
+        big[182] = 0b1111111110110110;
+        big[183] = 0b1111111110110111;
+        big[184] = 0b1111111110111000;
+        big[185] = 0b1111111110111001;
+        big[186] = 0b1111111110111010;
+        big[187] = 0b1111111110111011;
+        big[188] = 0b1111111110111100;
+        big[189] = 0b1111111110111101;
+        big[190] = 0b1111111110111110;
+        big[191] = 0b1111111110111111;
+        big[192] = 0b1111111111000000;
+        big[193] = 0b1111111111000001;
+        big[194] = 0b1111111111000010;
+        big[195] = 0b1111111111000011;
+        big[196] = 0b1111111111000100;
+        big[197] = 0b1111111111000101;
+        big[198] = 0b1111111111000110;
+        big[199] = 0b1111111111000111;
+        big[200] = 0b1111111111001000;
+        big[201] = 0b1111111111001001;
+        big[202] = 0b1111111111001010;
+        big[203] = 0b1111111111001011;
+        big[204] = 0b1111111111001100;
+        big[205] = 0b1111111111001101;
+        big[206] = 0b1111111111001110;
+        big[207] = 0b1111111111001111;
+        big[208] = 0b1111111111010000;
+        big[209] = 0b1111111111010001;
+        big[210] = 0b1111111111010010;
+        big[211] = 0b1111111111010011;
+        big[212] = 0b1111111111010100;
+        big[213] = 0b1111111111010101;
+        big[214] = 0b1111111111010110;
+        big[215] = 0b1111111111010111;
+        big[216] = 0b1111111111011000;
+        big[217] = 0b1111111111011001;
+        big[218] = 0b1111111111011010;
+        big[219] = 0b1111111111011011;
+        big[220] = 0b1111111111011100;
+        big[221] = 0b1111111111011101;
+        big[222] = 0b1111111111011110;
+        big[223] = 0b1111111111011111;
+        big[224] = 0b1111111111100000;
+        big[225] = 0b1111111111100001;
+        big[226] = 0b1111111111100010;
+        big[227] = 0b1111111111100011;
+        big[228] = 0b1111111111100100;
+        big[229] = 0b1111111111100101;
+        big[230] = 0b1111111111100110;
+        big[231] = 0b1111111111100111;
+        big[232] = 0b1111111111101000;
+        big[233] = 0b1111111111101001;
+        big[234] = 0b1111111111101010;
+        big[235] = 0b1111111111101011;
+        big[236] = 0b1111111111101100;
+        big[237] = 0b1111111111101101;
+        big[238] = 0b1111111111101110;
+        big[239] = 0b1111111111101111;
+        big[240] = 0b1111111111110000;
+        big[241] = 0b1111111111110001;
+        big[242] = 0b1111111111110010;
+        big[243] = 0b1111111111110011;
+        big[244] = 0b1111111111110100;
+        big[245] = 0b1111111111110101;
+        big[246] = 0b1111111111110110;
+        big[247] = 0b1111111111110111;
+        big[248] = 0b1111111111111000;
+        big[249] = 0b1111111111111001;
+        big[250] = 0b1111111111111010;
+        big[251] = 0b1111111111111011;
+        big[252] = 0b1111111111111100;
+        big[253] = 0b1111111111111101;
+        big[254] = 0b1111111111111110;
+        big[255] = 0b1111111111111111;
 
-	for (i = 0; i < 256; i++) {
-		if (big[i] != i + 65280U)
-			return 1;
-	}
+        for (i = 0; i < 256; i++) {
+                if (big[i] != i + 65280U)
+                        return 1;
+        }
 
-	return 0;
+        return 0;
 }
diff --git a/test/val/bug1071.c b/test/val/bug1071.c
index 02b069de0..66e298b25 100644
--- a/test/val/bug1071.c
+++ b/test/val/bug1071.c
@@ -7,24 +7,24 @@
 
 struct ImageStruct
 {
-	uint8_t _imageData;
-	#if !defined(NO_COLOR)
-		uint8_t _color;
-	#endif
+        uint8_t _imageData;
+        #if !defined(NO_COLOR)
+                uint8_t _color;
+        #endif
 };
 
 typedef struct ImageStruct Image;
 
 struct CharacterStruct
 {
-	// character coordinates
-	uint8_t _x;
-	uint8_t _y;
+        // character coordinates
+        uint8_t _x;
+        uint8_t _y;
 
-	// _status decides whether the character is active
-	uint8_t _status;
+        // _status decides whether the character is active
+        uint8_t _status;
 
-	Image* _imagePtr;
+        Image* _imagePtr;
 };
 
 typedef struct CharacterStruct Character;
@@ -53,20 +53,20 @@ Character bombs[BOMBS_NUMBER];
 
 uint16_t test1(void)
 {
-	if((loop<MAX_GHOST_LOOP) && (ghostLevel<MAX_GHOST_LEVEL))
-	{
-		return INITIAL_GHOST_SLOWDOWN-(uint16_t)level*256-ghostLevel*8;
-	}
-	return GHOST_MIN_SLOWDOWN;
+        if((loop<MAX_GHOST_LOOP) && (ghostLevel<MAX_GHOST_LEVEL))
+        {
+                return INITIAL_GHOST_SLOWDOWN-(uint16_t)level*256-ghostLevel*8;
+        }
+        return GHOST_MIN_SLOWDOWN;
 }
 
 uint16_t test2(void)
 {
-	if((loop<MAX_GHOST_LOOP) && (ghostLevel<MAX_GHOST_LEVEL))
-	{
-		return INITIAL_GHOST_SLOWDOWN-(uint16_t)level*256-ghostLevel*16;
-	}
-	return GHOST_MIN_SLOWDOWN;
+        if((loop<MAX_GHOST_LOOP) && (ghostLevel<MAX_GHOST_LEVEL))
+        {
+                return INITIAL_GHOST_SLOWDOWN-(uint16_t)level*256-ghostLevel*16;
+        }
+        return GHOST_MIN_SLOWDOWN;
 }
 
 uint16_t res = 0;
diff --git a/test/val/rand.c b/test/val/rand.c
index f2f604449..ac2ab0d17 100644
--- a/test/val/rand.c
+++ b/test/val/rand.c
@@ -28,83 +28,83 @@ static uint32_t seed;
 
 int ref_rand()
 {
-	uint16_t output;
-	/* seed follows the LCG sequence * 0x01010101 + 0xB3B3B3B3 */
-	seed = seed * 0x01010101UL + 0xB3B3B3B3UL;
-	/* output uses the top two bytes (reversed) XOR with bottom two bytes */
-	{
-		uint16_t s0 = (seed >>  0) & 0xFF;
-		uint16_t s1 = (seed >>  8) & 0xFF;
-		uint16_t s2 = (seed >> 16) & 0xFF;
-		uint16_t s3 = (seed >> 24) & 0xFF;
-		uint16_t o0 = s3 ^ s1;
-		uint16_t o1 = s2 ^ s0;
-		output = o0 | (o1 << 8);
-	}
-	return (int)(output & 0x7FFF);
+        uint16_t output;
+        /* seed follows the LCG sequence * 0x01010101 + 0xB3B3B3B3 */
+        seed = seed * 0x01010101UL + 0xB3B3B3B3UL;
+        /* output uses the top two bytes (reversed) XOR with bottom two bytes */
+        {
+                uint16_t s0 = (seed >>  0) & 0xFF;
+                uint16_t s1 = (seed >>  8) & 0xFF;
+                uint16_t s2 = (seed >> 16) & 0xFF;
+                uint16_t s3 = (seed >> 24) & 0xFF;
+                uint16_t o0 = s3 ^ s1;
+                uint16_t o1 = s2 ^ s0;
+                output = o0 | (o1 << 8);
+        }
+        return (int)(output & 0x7FFF);
 }
 
 void ref_srand(int ax)
 {
-	uint32_t s = (unsigned int)ax;
-	seed = s | (s << 16); /* low 16 bits is convenient filler for high 16 bits */
-	ref_rand(); /* one pre-call "shuffles" the first rand() result so it isn't too predictable */
+        uint32_t s = (unsigned int)ax;
+        seed = s | (s << 16); /* low 16 bits is convenient filler for high 16 bits */
+        ref_rand(); /* one pre-call "shuffles" the first rand() result so it isn't too predictable */
 }
 
 int main(void)
 {
-	unsigned int i,j;
-	int a,b;
+        unsigned int i,j;
+        int a,b;
 
-	/* test that startup state is equivalent to srand(1) */
-	{
-		//srand(1); // implied
-		ref_srand(1);
-		for (j=0; j<SUBTESTS; ++j)
-		{
-			a = rand();
-			b = ref_rand();
-			if (a != b)
-			{
-				printf("failed startup seed at test %d. rand()=%d reference=%d\n",j,a,b);
-				return EXIT_FAILURE;
-			}
-		}
-	}
+        /* test that startup state is equivalent to srand(1) */
+        {
+                //srand(1); // implied
+                ref_srand(1);
+                for (j=0; j<SUBTESTS; ++j)
+                {
+                        a = rand();
+                        b = ref_rand();
+                        if (a != b)
+                        {
+                                printf("failed startup seed at test %d. rand()=%d reference=%d\n",j,a,b);
+                                return EXIT_FAILURE;
+                        }
+                }
+        }
 
-	/* test every power of 2 seed */
-	for (i = 0; i < 16; ++i)
-	{
-		srand(1<<i);
-		ref_srand(1<<i);
-		for (j=0; j<SUBTESTS; ++j)
-		{
-			a = rand();
-			b = ref_rand();
-			if (a != b)
-			{
-				printf("failed seed %d at test %d. rand()=%d reference=%d\n",(1<<i),j,a,b);
-				return EXIT_FAILURE;
-			}
-		}
-	}
+        /* test every power of 2 seed */
+        for (i = 0; i < 16; ++i)
+        {
+                srand(1<<i);
+                ref_srand(1<<i);
+                for (j=0; j<SUBTESTS; ++j)
+                {
+                        a = rand();
+                        b = ref_rand();
+                        if (a != b)
+                        {
+                                printf("failed seed %d at test %d. rand()=%d reference=%d\n",(1<<i),j,a,b);
+                                return EXIT_FAILURE;
+                        }
+                }
+        }
 
-	/* test a sampling of seeds*/
-	for (i = 0; i < 32768UL; i += TESTINC)
-	{
-		srand(i);
-		ref_srand(i);
-		for (j=0; j<SUBTESTS; ++j)
-		{
-			a = rand();
-			b = ref_rand();
-			if (a != b)
-			{
-				printf("failed seed %d at test %d. rand()=%d reference=%d\n",(1<<i),j,a,b);
-				return EXIT_FAILURE;
-			}
-		}
-	}
+        /* test a sampling of seeds*/
+        for (i = 0; i < 32768UL; i += TESTINC)
+        {
+                srand(i);
+                ref_srand(i);
+                for (j=0; j<SUBTESTS; ++j)
+                {
+                        a = rand();
+                        b = ref_rand();
+                        if (a != b)
+                        {
+                                printf("failed seed %d at test %d. rand()=%d reference=%d\n",(1<<i),j,a,b);
+                                return EXIT_FAILURE;
+                        }
+                }
+        }
 
-	return EXIT_SUCCESS;
+        return EXIT_SUCCESS;
 }

From e0c12c90cd7b729d416f6dd76e1f9185a339b088 Mon Sep 17 00:00:00 2001
From: Jesse Rosenstock <jmr@google.com>
Date: Sun, 15 Nov 2020 23:03:01 +0100
Subject: [PATCH 11/26] g_asr, g_asl: Use ROL/ROR for char shifts by >= 6

Instead of `val` right (left) shifts, we can also do `9 - val` left (right)
rotates and a mask.  This saves 3 bytes and 8 cycles for `val == 7` and
1 byte and 4 cycles for `val == 6`.
---
 src/cc65/codegen.c | 41 +++++++++++++++++++++++++++++++++++------
 1 file changed, 35 insertions(+), 6 deletions(-)

diff --git a/src/cc65/codegen.c b/src/cc65/codegen.c
index a58484cf1..3ca9d81e6 100644
--- a/src/cc65/codegen.c
+++ b/src/cc65/codegen.c
@@ -3114,9 +3114,26 @@ void g_asr (unsigned flags, unsigned long val)
         switch (flags & CF_TYPEMASK) {
             case CF_CHAR:
                 if (flags & CF_FORCECHAR) {
-                    if ((flags & CF_UNSIGNED) != 0 && val < 8) {
-                        while (val--) {
-                            AddCodeLine ("lsr a");
+                    val &= 7;
+                    if ((flags & CF_UNSIGNED) != 0) {
+                        /* Instead of `val` right shifts, we can also do `9 - val` left rotates
+                        ** and a mask.  This saves 3 bytes and 8 cycles for `val == 7` and
+                        ** 1 byte and 4 cycles for `val == 6`.
+                        */
+                        if (val < 6) {
+                            while (val--) {
+                                AddCodeLine ("lsr a");  /* 1 byte, 2 cycles */
+                            }
+                        } else {
+                            unsigned i;
+                            /* The first ROL shifts in garbage and sets carry to the high bit.
+                            ** The garbage is cleaned up by the mask.
+                            */
+                            for (i = val; i < 9; ++i) {
+                                AddCodeLine ("rol a");  /* 1 byte,  2 cycles */
+                            }
+                            /* 2 bytes, 2 cycles */
+                            AddCodeLine ("and #$%02X", 0xFF >> val);
                         }
                         return;
                     } else if (val <= 2) {
@@ -3270,9 +3287,21 @@ void g_asl (unsigned flags, unsigned long val)
     if (flags & CF_CONST) {
         switch (flags & CF_TYPEMASK) {
             case CF_CHAR:
-                if ((flags & CF_FORCECHAR) != 0 && val <= 6) {
-                    while (val--) {
-                        AddCodeLine ("asl a");
+                if ((flags & CF_FORCECHAR) != 0) {
+                    val &= 7;
+                    /* Large shifts are faster and smaller with ROR.  See g_asr for detailed
+                    ** byte and cycle counts.
+                    */
+                    if (val < 6) {
+                        while (val--) {
+                            AddCodeLine ("asl a");
+                        }
+                    } else {
+                        unsigned i;
+                        for (i = val; i < 9; ++i) {
+                            AddCodeLine ("ror a");
+                        }
+                        AddCodeLine ("and #$%02X", (~0U << val) & 0xFF);
                     }
                     return;
                 }

From 39573109500d06b415026af2d07de4b3c8b52a47 Mon Sep 17 00:00:00 2001
From: Sven Michael Klose <pixel@hugbox.org>
Date: Sun, 25 Oct 2020 14:27:26 +0100
Subject: [PATCH 12/26] Knock off two bytes from getcwd(), cbm_read() and
 cbm_write().

---
 libsrc/cbm/cbm_read.s  | 14 +++++++-------
 libsrc/cbm/cbm_write.s | 14 +++++++-------
 libsrc/common/getcwd.s | 16 ++++++++--------
 3 files changed, 22 insertions(+), 22 deletions(-)

diff --git a/libsrc/cbm/cbm_read.s b/libsrc/cbm/cbm_read.s
index b010966a3..a30983302 100644
--- a/libsrc/cbm/cbm_read.s
+++ b/libsrc/cbm/cbm_read.s
@@ -45,11 +45,11 @@
 
 
 _cbm_read:
-        eor     #$FF
-        sta     ptr1
-        txa
-        eor     #$FF
-        sta     ptr1+1          ; Save -size-1
+        inx
+        stx     ptr1+1
+        tax
+        inx
+        stx     ptr1            ; Save size with each byte incremented.
 
         jsr     popax
         sta     ptr2
@@ -92,9 +92,9 @@ _cbm_read:
         bne     @L3
         inc     ptr3+1          ; ++bytesread;
 
-@L3:    inc     ptr1
+@L3:    dec     ptr1
         bne     @L1
-        inc     ptr1+1
+        dec     ptr1+1
         bne     @L1
 
 @L4:    jsr     CLRCH
diff --git a/libsrc/cbm/cbm_write.s b/libsrc/cbm/cbm_write.s
index 2d932d04a..a199c1d90 100644
--- a/libsrc/cbm/cbm_write.s
+++ b/libsrc/cbm/cbm_write.s
@@ -39,11 +39,11 @@
 _cbm_write:
         sta     ptr3
         stx     ptr3+1          ; Save size
-        eor     #$FF
-        sta     ptr1
-        txa
-        eor     #$FF
-        sta     ptr1+1          ; Save -size-1
+        inx
+        stx     ptr1+1
+        tax
+        inx
+        stx     ptr1            ; Save size with each byte incremented
 
         jsr     popax
         sta     ptr2
@@ -69,9 +69,9 @@ _cbm_write:
 
 @L2:    jsr     BSOUT           ; cbm_k_bsout (A);
 
-@L3:    inc     ptr1            ; --size;
+@L3:    dec     ptr1            ; --size;
         bne     @L1
-        inc     ptr1+1
+        dec     ptr1+1
         bne     @L1
 
         jsr     CLRCH
diff --git a/libsrc/common/getcwd.s b/libsrc/common/getcwd.s
index 22b6ceded..c0a1c3634 100644
--- a/libsrc/common/getcwd.s
+++ b/libsrc/common/getcwd.s
@@ -17,22 +17,22 @@
 
 .proc   _getcwd
 
-; Remember -size-1 because this simplifies the following loop
+; Remember size with each byte incremented because this simplifies the following loop
 
-        eor     #$FF
-        sta     ptr2
-        txa
-        eor     #$FF
-        sta     ptr2+1
+        inx
+        stx     ptr2+1
+        tax
+        inx
+        stx     ptr2
 
         jsr     popptr1         ; Get buf to ptr1
 
 ; Copy __cwd to the given buffer checking the length
 
         ; ldy     #$00          is guaranteed by popptr1
-loop:   inc     ptr2
+loop:   dec     ptr2
         bne     @L1
-        inc     ptr2+1
+        dec     ptr2+1
         beq     overflow
 
 ; Copy one character, end the loop if the zero terminator is reached. We

From 99c0815cdb6b55ee89a1945e58674a5eefbb8126 Mon Sep 17 00:00:00 2001
From: Sven Michael Klose <pixel@hugbox.org>
Date: Tue, 27 Oct 2020 06:08:04 +0100
Subject: [PATCH 13/26] Clear up comments a bit.

---
 libsrc/cbm/cbm_read.s  | 2 +-
 libsrc/cbm/cbm_write.s | 2 +-
 libsrc/common/getcwd.s | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/libsrc/cbm/cbm_read.s b/libsrc/cbm/cbm_read.s
index a30983302..29e0e1f19 100644
--- a/libsrc/cbm/cbm_read.s
+++ b/libsrc/cbm/cbm_read.s
@@ -49,7 +49,7 @@ _cbm_read:
         stx     ptr1+1
         tax
         inx
-        stx     ptr1            ; Save size with each byte incremented.
+        stx     ptr1            ; Save size with both bytes incremented separately.
 
         jsr     popax
         sta     ptr2
diff --git a/libsrc/cbm/cbm_write.s b/libsrc/cbm/cbm_write.s
index a199c1d90..5ac07209c 100644
--- a/libsrc/cbm/cbm_write.s
+++ b/libsrc/cbm/cbm_write.s
@@ -43,7 +43,7 @@ _cbm_write:
         stx     ptr1+1
         tax
         inx
-        stx     ptr1            ; Save size with each byte incremented
+        stx     ptr1            ; Save size with both bytes incremented separately
 
         jsr     popax
         sta     ptr2
diff --git a/libsrc/common/getcwd.s b/libsrc/common/getcwd.s
index c0a1c3634..4bfc0a5b6 100644
--- a/libsrc/common/getcwd.s
+++ b/libsrc/common/getcwd.s
@@ -23,7 +23,7 @@
         stx     ptr2+1
         tax
         inx
-        stx     ptr2
+        stx     ptr2            ; Save size with each byte incremented separately
 
         jsr     popptr1         ; Get buf to ptr1
 

From f59cb9af0660e5401f18fe9e86329b2d0fd6b739 Mon Sep 17 00:00:00 2001
From: Sven Michael Klose <pixel@hugbox.org>
Date: Sun, 1 Nov 2020 22:59:07 +0100
Subject: [PATCH 14/26] Use more compact loops.

---
 libsrc/atmos/read.s                     | 15 ++++++++-------
 libsrc/atmos/write.s                    | 14 +++++++-------
 libsrc/cbm/read.s                       |  4 ++--
 libsrc/cbm/rwcommon.s                   | 10 +++++-----
 libsrc/cbm/write.s                      |  4 ++--
 libsrc/common/memcmp.s                  | 15 +++++++--------
 libsrc/common/strncat.s                 | 22 +++++++++++-----------
 libsrc/common/strncpy.s                 | 18 +++++++++---------
 libsrc/common/strnicmp.s                | 20 +++++++-------------
 libsrc/conio/vcprintf.s                 | 16 ++++++++--------
 libsrc/geos-common/drivers/fio_module.s | 15 ++++++++-------
 libsrc/osic1p/bootstrap.s               | 17 +++++++++--------
 libsrc/telestrat/write.s                | 14 +++++++-------
 13 files changed, 90 insertions(+), 94 deletions(-)

diff --git a/libsrc/atmos/read.s b/libsrc/atmos/read.s
index 83aa8024e..3f22d8d0e 100644
--- a/libsrc/atmos/read.s
+++ b/libsrc/atmos/read.s
@@ -20,18 +20,19 @@
 
         sta     ptr3
         stx     ptr3+1          ; save count as result
-        eor     #$FF
-        sta     ptr2
-        txa
-        eor     #$FF
-        sta     ptr2+1          ; Remember -count-1
+
+        inx
+        stx     ptr2+1
+        tax
+        inx
+        stx     ptr2            ; save count with each byte incremented separately
 
         jsr     popptr1         ; get buf
         jsr     popax           ; get fd and discard
 
-L1:     inc     ptr2
+L1:     dec     ptr2
         bnz     L2
-        inc     ptr2+1
+        dec     ptr2+1
         bze     L9              ; no more room in buf
 
 ; If there are no more characters in BASIC's input buffer, then get a line from
diff --git a/libsrc/atmos/write.s b/libsrc/atmos/write.s
index 7865b5698..4a68994ec 100644
--- a/libsrc/atmos/write.s
+++ b/libsrc/atmos/write.s
@@ -17,17 +17,17 @@
         sta     ptr3
         stx     ptr3+1          ; save count as result
 
-        eor     #$FF
-        sta     ptr2
-        txa
-        eor     #$FF
-        sta     ptr2+1          ; Remember -count-1
+        inx
+        stx     ptr2+1
+        tax
+        inx
+        stx     ptr2            ; save count with each byte incremented separately
 
         jsr     popptr1         ; get buf
         jsr     popax           ; get fd and discard
-L1:     inc     ptr2
+L1:     dec     ptr2
         bne     L2
-        inc     ptr2+1
+        dec     ptr2+1
         beq     L9
 L2:     ldy     #0
         lda     (ptr1),y
diff --git a/libsrc/cbm/read.s b/libsrc/cbm/read.s
index ee01596aa..fb0bb3171 100644
--- a/libsrc/cbm/read.s
+++ b/libsrc/cbm/read.s
@@ -106,9 +106,9 @@
 
 ; Decrement the count
 
-@L3:    inc     ptr2
+@L3:    dec     ptr2
         bne     @L0
-        inc     ptr2+1
+        dec     ptr2+1
         bne     @L0
         beq     done            ; Branch always
 
diff --git a/libsrc/cbm/rwcommon.s b/libsrc/cbm/rwcommon.s
index a1f92be8c..d13c478b5 100644
--- a/libsrc/cbm/rwcommon.s
+++ b/libsrc/cbm/rwcommon.s
@@ -21,11 +21,11 @@
 
 .proc   rwcommon
 
-        eor     #$FF
-        sta     ptr2
-        txa
-        eor     #$FF
-        sta     ptr2+1          ; Remember -count-1
+        inx
+        stx     ptr2+1
+        tax
+        inx
+        stx     ptr2            ; Save count with each byte incremented separately
 
         jsr     popptr1         ; Get buf to ptr1, Y=0 by call
 
diff --git a/libsrc/cbm/write.s b/libsrc/cbm/write.s
index 7a27f0044..ebc44a0ac 100644
--- a/libsrc/cbm/write.s
+++ b/libsrc/cbm/write.s
@@ -83,9 +83,9 @@
 
 ; Decrement count
 
-@L2:    inc     ptr2
+@L2:    dec     ptr2
         bne     @L0
-        inc     ptr2+1
+        dec     ptr2+1
         bne     @L0
 
 ; Wrote all chars or disk full. Close the output channel
diff --git a/libsrc/common/memcmp.s b/libsrc/common/memcmp.s
index 93a2c28dc..5879a433d 100644
--- a/libsrc/common/memcmp.s
+++ b/libsrc/common/memcmp.s
@@ -13,11 +13,11 @@ _memcmp:
 ; Calculate (-count-1) and store it into ptr3. This is some overhead here but
 ; saves time in the compare loop
 
-        eor     #$FF
-        sta     ptr3
-        txa
-        eor     #$FF
-        sta     ptr3+1
+        inx
+        stx     ptr3+1
+        tax
+        inx
+        stx     ptr3            ; Save count with each byte incremented separately
 
 ; Get the pointer parameters
 
@@ -33,7 +33,7 @@ _memcmp:
 
 ; Head of compare loop: Test for the end condition
 
-Loop:   inx                     ; Bump low byte of (-count-1)
+Loop:   dex                     ; Bump low byte of (-count-1)
         beq     BumpHiCnt       ; Jump on overflow
 
 ; Do the compare
@@ -53,7 +53,7 @@ Comp:   lda     (ptr1),y
 ; Entry on low counter byte overflow
 
 BumpHiCnt:
-        inc     ptr3+1          ; Bump high byte of (-count-1)
+        dec     ptr3+1          ; Bump high byte of (-count-1)
         bne     Comp            ; Jump if not done
         jmp     return0         ; Count is zero, areas are identical
 
@@ -67,4 +67,3 @@ NotEqual:
 Greater:
         ldx     #$01            ; Make result positive
         rts
-
diff --git a/libsrc/common/strncat.s b/libsrc/common/strncat.s
index caa6804fc..060378442 100644
--- a/libsrc/common/strncat.s
+++ b/libsrc/common/strncat.s
@@ -5,17 +5,17 @@
 ; char* strncat (char* dest, const char* src, size_t n);
 ;
 
-        .export         _strncat
-        .import         popax, popptr1
-        .importzp       ptr1, ptr2, ptr3, tmp1, tmp2
-        .macpack        cpu
+.export         _strncat
+.import         popax, popptr1
+.importzp       ptr1, ptr2, ptr3, tmp1, tmp2
+.macpack        cpu
         
 _strncat:
-    eor #$FF        ; one's complement to count upwards
-    sta tmp1
-    txa
-    eor #$FF
-    sta tmp2
+    inx
+    stx tmp2
+    tax
+    inx
+    stx tmp1        ; save count with each byte incremented separately
     
     jsr popptr1     ; get src
 
@@ -49,9 +49,9 @@ L2: sty ptr2
 L3: ldy #0
     ldx tmp1        ; low counter byte
 
-L4: inx
+L4: dex
     bne L5
-    inc tmp2
+    dec tmp2
     beq L6          ; jump if done
 L5: lda (ptr1),y
     sta (ptr2),y
diff --git a/libsrc/common/strncpy.s b/libsrc/common/strncpy.s
index 56387f880..49bd90b32 100644
--- a/libsrc/common/strncpy.s
+++ b/libsrc/common/strncpy.s
@@ -10,11 +10,11 @@
 
 .proc   _strncpy
 
-        eor     #$FF
-        sta     tmp1
-        txa
-        eor     #$FF
-        sta     tmp2            ; Store -size - 1
+        inx
+        stx     tmp2
+        tax
+        inx
+        stx     tmp1            ; save count with each byte incremented separately
 
         jsr     popptr1         ; get src
         jsr     popax           ; get dest
@@ -26,9 +26,9 @@
 
         ldx     tmp1            ; Load low byte of ones complement of size
         ldy     #$00
-L1:     inx
+L1:     dex
         bne     L2
-        inc     tmp2
+        dec     tmp2
         beq     L9
 
 L2:     lda     (ptr1),y        ; Copy one character
@@ -42,7 +42,7 @@ L2:     lda     (ptr1),y        ; Copy one character
 
 ; Fill the remaining bytes.
 
-L3:     inx                     ; Counter low byte
+L3:     dex                     ; Counter low byte
         beq     L6              ; Branch on overflow
 L4:     sta     (ptr2),y        ; Clear one byte
 L5:     iny                     ; Bump pointer
@@ -52,7 +52,7 @@ L5:     iny                     ; Bump pointer
 
 ; Bump the counter high byte
 
-L6:     inc     tmp2
+L6:     dec     tmp2
         bne     L4
 
 ; Done, return dest
diff --git a/libsrc/common/strnicmp.s b/libsrc/common/strnicmp.s
index 43d6d0d50..f447b9683 100644
--- a/libsrc/common/strnicmp.s
+++ b/libsrc/common/strnicmp.s
@@ -15,17 +15,11 @@
 _strnicmp:
 _strncasecmp:
 
-; Convert the given counter value in a/x from a downward counter into an
-; upward counter, so we can increment the counter in the loop below instead
-; of decrementing it. This adds some overhead now, but is cheaper than
-; executing a more complex test in each iteration of the loop. We do also
-; correct the value by one, so we can do the test on top of the loop.
-
-        eor     #$FF
-        sta     ptr3
-        txa
-        eor     #$FF
-        sta     ptr3+1
+        inx
+        stx     ptr3+1
+        tax
+        inx
+        stx     ptr3        ; save count with each byte incremented separately
 
 ; Get the remaining arguments
 
@@ -40,7 +34,7 @@ _strncasecmp:
 
 ; Start of compare loop. Check the counter.
 
-Loop:   inc     ptr3
+Loop:   dec     ptr3
         beq     IncHi           ; increment high byte
 
 ; Compare a byte from the strings
@@ -79,7 +73,7 @@ L2:     ldx     tmp1
 
 ; Increment hi byte
 
-IncHi:  inc     ptr3+1
+IncHi:  dec     ptr3+1
         bne     Comp            ; jump if counter not zero
 
 ; Exit code if strings are equal. a/x not set
diff --git a/libsrc/conio/vcprintf.s b/libsrc/conio/vcprintf.s
index 3a9ddf9d7..084efe089 100644
--- a/libsrc/conio/vcprintf.s
+++ b/libsrc/conio/vcprintf.s
@@ -47,12 +47,12 @@ outdesc:                        ; Static outdesc structure
 
 out:    jsr     popax           ; count
         sta     ptr2
-        eor     #$FF
-        sta     outdesc+6
-        txa
-        sta     ptr2+1
-        eor     #$FF
-        sta     outdesc+7
+        stx     ptr2+1
+        inx
+        stx     outdesc+7
+        tax
+        inx
+        stx     outdesc+6
 
         jsr     popptr1         ; buf
 
@@ -74,7 +74,7 @@ out:    jsr     popax           ; count
 
 ; Loop outputting characters
 
-@L1:    inc     outdesc+6
+@L1:    dec     outdesc+6
         beq     @L4
 @L2:    ldy     tmp1
         lda     (ptr1),y
@@ -85,7 +85,7 @@ out:    jsr     popax           ; count
         jsr     _cputc
         jmp     @L1
 
-@L4:    inc     outdesc+7
+@L4:    dec     outdesc+7
         bne     @L2
         rts
 
diff --git a/libsrc/geos-common/drivers/fio_module.s b/libsrc/geos-common/drivers/fio_module.s
index 0be5015a7..937ef292e 100644
--- a/libsrc/geos-common/drivers/fio_module.s
+++ b/libsrc/geos-common/drivers/fio_module.s
@@ -94,11 +94,12 @@ _read:
     ; popax - fd, must be == to the above one
     ; return -1+__oserror or number of bytes read
 
-        eor #$ff
-        sta ptr1
-        txa
-        eor #$ff
-        sta ptr1+1              ; -(# of bytes to read)-1
+        inx
+        stx ptr1+1
+        tax
+        inx
+        stx ptr1                ; save count with each byte incremented separately
+
         jsr popax
         sta ptr2
         stx ptr2+1              ; buffer ptr
@@ -152,9 +153,9 @@ _read:
         beq @done               ; yes, we're done
         jmp __mappederrno       ; no, we're screwed
 
-@L3:    inc ptr1                ; decrement the count
+@L3:    dec ptr1                ; decrement the count
         bne @L0
-        inc ptr1+1
+        dec ptr1+1
         bne @L0
 
 @done:
diff --git a/libsrc/osic1p/bootstrap.s b/libsrc/osic1p/bootstrap.s
index ed2ade222..efb480cb8 100644
--- a/libsrc/osic1p/bootstrap.s
+++ b/libsrc/osic1p/bootstrap.s
@@ -50,16 +50,17 @@ LINEDIST        =       $20             ; Offset in video RAM between two lines
         ldx     #>load_addr
         sta     load
         stx     load+1
-        lda     #<load_size
-        eor     #$FF
-        sta     count                   ; store (-size - 1)
-        lda     #>load_size
-        eor     #$FF
-        sta     count+1
 
-L1:     inc     count                   ; pre-count one's-complement upwards
+        ldx     #<load_size
+        inx
+        stx     count
+        ldx     #>load_size
+        inx
+        stx     count+1                 ; save size with each byte incremented separately
+
+L1:     dec     count                   ; pre-count one's-complement upwards
         bnz     L2
-        inc     count+1
+        dec     count+1
         bze     L3
 L2:     jsr     GETCHAR                 ; (doesn't change .Y)
         sta     (load),y
diff --git a/libsrc/telestrat/write.s b/libsrc/telestrat/write.s
index 68aef42d6..215db3e52 100644
--- a/libsrc/telestrat/write.s
+++ b/libsrc/telestrat/write.s
@@ -13,11 +13,11 @@
         sta     ptr3
         stx     ptr3+1          ; save count as result
 
-        eor     #$FF
-        sta     ptr2
-        txa
-        eor     #$FF
-        sta     ptr2+1          ; remember -count-1
+        inx
+        stx     ptr2+1
+        tax
+        inx
+        stx     ptr2            ; save count with each byte incremented separately
 
         jsr     popptr1         ; get buf
         jsr     popax           ; get fd and discard
@@ -51,9 +51,9 @@ next:
         rts
 
 
-L1:     inc     ptr2
+L1:     dec     ptr2
         bne     L2
-        inc     ptr2+1
+        dec     ptr2+1
         beq     L9
 L2:     ldy     #0
         lda     (ptr1),y

From 6201300816b3c6bcd9cd688f40c0f34b10d3584e Mon Sep 17 00:00:00 2001
From: Sven Michael Klose <pixel@hugbox.org>
Date: Tue, 3 Nov 2020 11:52:58 +0100
Subject: [PATCH 15/26] Fold constant calculation.

---
 libsrc/osic1p/bootstrap.s | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/libsrc/osic1p/bootstrap.s b/libsrc/osic1p/bootstrap.s
index efb480cb8..031c4f651 100644
--- a/libsrc/osic1p/bootstrap.s
+++ b/libsrc/osic1p/bootstrap.s
@@ -51,11 +51,9 @@ LINEDIST        =       $20             ; Offset in video RAM between two lines
         sta     load
         stx     load+1
 
-        ldx     #<load_size
-        inx
+        ldx     #(<load_size) + 1
         stx     count
-        ldx     #>load_size
-        inx
+        ldx     #(>load_size) + 1
         stx     count+1                 ; save size with each byte incremented separately
 
 L1:     dec     count                   ; pre-count one's-complement upwards

From db312049503dc0effa872d66467b4f36de11831b Mon Sep 17 00:00:00 2001
From: Sven Michael Klose <pixel@hugbox.org>
Date: Tue, 3 Nov 2020 11:53:56 +0100
Subject: [PATCH 16/26] Remove stale comment.

---
 libsrc/osic1p/bootstrap.s | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/libsrc/osic1p/bootstrap.s b/libsrc/osic1p/bootstrap.s
index 031c4f651..e88e257fd 100644
--- a/libsrc/osic1p/bootstrap.s
+++ b/libsrc/osic1p/bootstrap.s
@@ -56,7 +56,7 @@ LINEDIST        =       $20             ; Offset in video RAM between two lines
         ldx     #(>load_size) + 1
         stx     count+1                 ; save size with each byte incremented separately
 
-L1:     dec     count                   ; pre-count one's-complement upwards
+L1:     dec     count
         bnz     L2
         dec     count+1
         bze     L3

From 9d62abb7ac19e2e117d7d162cb4d73482bca06ca Mon Sep 17 00:00:00 2001
From: Sven Michael Klose <pixel@hugbox.org>
Date: Tue, 3 Nov 2020 11:54:50 +0100
Subject: [PATCH 17/26] Fix comment.

---
 libsrc/common/strnicmp.s | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/libsrc/common/strnicmp.s b/libsrc/common/strnicmp.s
index f447b9683..477aa3d66 100644
--- a/libsrc/common/strnicmp.s
+++ b/libsrc/common/strnicmp.s
@@ -34,8 +34,8 @@ _strncasecmp:
 
 ; Start of compare loop. Check the counter.
 
-Loop:   dec     ptr3
-        beq     IncHi           ; increment high byte
+Loop:   dec     ptr3            ; decrement high byte
+        beq     IncHi
 
 ; Compare a byte from the strings
 

From 9800555bbb399decb5aaf50260a831076517332d Mon Sep 17 00:00:00 2001
From: Sven Michael Klose <pixel@hugbox.org>
Date: Fri, 20 Nov 2020 03:50:19 +0100
Subject: [PATCH 18/26] Remove stale comments.

---
 libsrc/common/strncpy.s | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/libsrc/common/strncpy.s b/libsrc/common/strncpy.s
index 49bd90b32..138413ecb 100644
--- a/libsrc/common/strncpy.s
+++ b/libsrc/common/strncpy.s
@@ -24,7 +24,7 @@
 
 ; Copy src -> dest up to size bytes
 
-        ldx     tmp1            ; Load low byte of ones complement of size
+        ldx     tmp1
         ldy     #$00
 L1:     dex
         bne     L2
@@ -43,7 +43,7 @@ L2:     lda     (ptr1),y        ; Copy one character
 ; Fill the remaining bytes.
 
 L3:     dex                     ; Counter low byte
-        beq     L6              ; Branch on overflow
+        beq     L6
 L4:     sta     (ptr2),y        ; Clear one byte
 L5:     iny                     ; Bump pointer
         bne     L3

From d90cd112125a0a51c286b019c4e76f2131499683 Mon Sep 17 00:00:00 2001
From: Greg King <gregdk@users.sf.net>
Date: Sun, 27 Dec 2020 18:22:12 -0500
Subject: [PATCH 19/26] Fixed outdated comments.

---
 libsrc/common/memcmp.s | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

diff --git a/libsrc/common/memcmp.s b/libsrc/common/memcmp.s
index 5879a433d..d651a3cfc 100644
--- a/libsrc/common/memcmp.s
+++ b/libsrc/common/memcmp.s
@@ -10,8 +10,8 @@
 
 _memcmp:
 
-; Calculate (-count-1) and store it into ptr3. This is some overhead here but
-; saves time in the compare loop
+; Calculate a special count, and store it into ptr3. That is some overhead here,
+; but saves time in the compare loop
 
         inx
         stx     ptr3+1
@@ -29,12 +29,12 @@ _memcmp:
 ; Loop initialization
 
         ;ldy     #$00           ; Initialize pointer (Y=0 guaranteed by popptr1)
-        ldx     ptr3            ; Load low counter byte into X
+        ldx     ptr3            ; Load inner counter byte into .X
 
 ; Head of compare loop: Test for the end condition
 
-Loop:   dex                     ; Bump low byte of (-count-1)
-        beq     BumpHiCnt       ; Jump on overflow
+Loop:   dex
+        beq     BumpHiCnt       ; Jump on end of inner count
 
 ; Do the compare
 
@@ -50,10 +50,10 @@ Comp:   lda     (ptr1),y
         inc     ptr2+1
         bne     Loop            ; Branch always (pointer wrap is illegal)
 
-; Entry on low counter byte overflow
+; Entry on inner loop end
 
 BumpHiCnt:
-        dec     ptr3+1          ; Bump high byte of (-count-1)
+        dec     ptr3+1
         bne     Comp            ; Jump if not done
         jmp     return0         ; Count is zero, areas are identical
 

From a9b71b6207c837833aacda2e5252dac4fd8fa8e5 Mon Sep 17 00:00:00 2001
From: Rocky <rocky@AvBook.local>
Date: Fri, 1 Jan 2021 15:39:23 +0100
Subject: [PATCH 20/26] return-type - new warning suppression type added

---
 src/cc65/error.c    | 2 ++
 src/cc65/error.h    | 1 +
 src/cc65/function.c | 4 ++--
 3 files changed, 5 insertions(+), 2 deletions(-)

diff --git a/src/cc65/error.c b/src/cc65/error.c
index 132bf331d..feb43565d 100644
--- a/src/cc65/error.c
+++ b/src/cc65/error.c
@@ -74,6 +74,7 @@ IntStack WarnUnknownPragma  = INTSTACK(1);  /* - unknown #pragmas */
 IntStack WarnUnusedLabel    = INTSTACK(1);  /* - unused labels */
 IntStack WarnUnusedParam    = INTSTACK(1);  /* - unused parameters */
 IntStack WarnUnusedVar      = INTSTACK(1);  /* - unused variables */
+IntStack WarnReturnType     = INTSTACK(1);  /* - control reaches end of non-void function */
 
 /* Map the name of a warning to the intstack that holds its state */
 typedef struct WarnMapEntry WarnMapEntry;
@@ -92,6 +93,7 @@ static WarnMapEntry WarnMap[] = {
     { &WarnUnusedLabel,         "unused-label"          },
     { &WarnUnusedParam,         "unused-param"          },
     { &WarnUnusedVar,           "unused-var"            },
+    { &WarnReturnType,          "return-type"           },
 };
 
 Collection DiagnosticStrBufs;
diff --git a/src/cc65/error.h b/src/cc65/error.h
index a443aeff8..898793651 100644
--- a/src/cc65/error.h
+++ b/src/cc65/error.h
@@ -71,6 +71,7 @@ extern IntStack WarnUnknownPragma;      /* - unknown #pragmas */
 extern IntStack WarnUnusedLabel;        /* - unused labels */
 extern IntStack WarnUnusedParam;        /* - unused parameters */
 extern IntStack WarnUnusedVar;          /* - unused variables */
+extern IntStack WarnReturnType;         /* - control reaches end of non-void function */
 
 /* Forward */
 struct StrBuf;
diff --git a/src/cc65/function.c b/src/cc65/function.c
index 00755ae65..9269d4c62 100644
--- a/src/cc65/function.c
+++ b/src/cc65/function.c
@@ -654,8 +654,8 @@ void NewFunc (SymEntry* Func, FuncDesc* D)
     ** environment returning int, output a warning if we didn't see a return
     ** statement.
     */
-    if (!F_HasVoidReturn (CurrentFunc) && !F_HasReturn (CurrentFunc) && !C99MainFunc) {
-        Warning ("Control reaches end of non-void function");
+    if (!F_HasVoidReturn (CurrentFunc) && !F_HasReturn (CurrentFunc) && !C99MainFunc && IS_Get (&WarnReturnType)) {
+        Warning ("Control reaches end of non-void function [-Wreturn-type]");
     }
 
     /* If this is the main function in a C99 environment returning an int, let

From 96624699577c7497fda3592088f3928867dff381 Mon Sep 17 00:00:00 2001
From: Piotr Kaczorowski <piotr.kaczorowski@gmail.com>
Date: Mon, 4 Jan 2021 22:05:55 +0100
Subject: [PATCH 21/26] Return-type warning and pseudo variable __A__
 documentation added.

---
 doc/cc65.sgml | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/doc/cc65.sgml b/doc/cc65.sgml
index e273ced9c..211aa21b3 100644
--- a/doc/cc65.sgml
+++ b/doc/cc65.sgml
@@ -549,6 +549,8 @@ Here is a description of all the command line options:
   <tag><tt/remap-zero/</tag>
         Warn about a <tt/<ref id="pragma-charmap" name="#pragma charmap()">/
         that changes a character's code number from/to 0x00.
+  <tag><tt/return-type/</tag>
+        Warn about no return statement in function returning non-void..
   <tag><tt/struct-param/</tag>
         Warn when passing structs by value.
   <tag><tt/unknown-pragma/</tag>
@@ -726,6 +728,29 @@ This cc65 version has some extensions to the ISO C standard.
         will give the high byte of any unsigned value.
         <p>
 
+<item>  There is pseudo variable named <tt/__A__/ that indicates accumulator register. 
+        It can be used to highlight the returned variable in non-void
+        function where result is hidden in the <tt/asm/ section.
+        every other variable. They are most useful together with short
+        sequences of assembler code. 
+
+    For example, in the function below
+
+        <tscreen><verb>
+        BYTE get_key_via_atarixl_romcall_device_k(void)      \
+            asm("LDA $E425");   \
+            asm("PHA");         \
+            asm("LDA $E424");   \
+            asm("PHA");         \
+            asm("RTA");         \
+            return __A__;
+        </verb></tscreen>
+
+        <tt/return __A__;/ will indicate that result is stored in the accumulator
+        and suppress warning that function has no return statement. Instead, you can
+        also use <tt/#pragma warn (return-type, off) ... pragma warn (return-type, on)/
+        <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.

From 68e1faa2c815696c5cebdb033593b3dcdbb9fff3 Mon Sep 17 00:00:00 2001
From: Piotr Kaczorowski <piotr.kaczorowski@gmail.com>
Date: Mon, 4 Jan 2021 22:10:16 +0100
Subject: [PATCH 22/26] Fix typo in cc65.sgml

---
 doc/cc65.sgml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/doc/cc65.sgml b/doc/cc65.sgml
index 211aa21b3..0b36812d4 100644
--- a/doc/cc65.sgml
+++ b/doc/cc65.sgml
@@ -550,7 +550,7 @@ Here is a description of all the command line options:
         Warn about a <tt/<ref id="pragma-charmap" name="#pragma charmap()">/
         that changes a character's code number from/to 0x00.
   <tag><tt/return-type/</tag>
-        Warn about no return statement in function returning non-void..
+        Warn about no return statement in function returning non-void.
   <tag><tt/struct-param/</tag>
         Warn when passing structs by value.
   <tag><tt/unknown-pragma/</tag>

From 601e6283436e00314b00fc0b2fb9a549535200ea Mon Sep 17 00:00:00 2001
From: Piotr Kaczorowski <piotr.kaczorowski@gmail.com>
Date: Mon, 4 Jan 2021 22:16:07 +0100
Subject: [PATCH 23/26] Another fix typo in cc65.sgml

---
 doc/cc65.sgml | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/doc/cc65.sgml b/doc/cc65.sgml
index 0b36812d4..82da74f49 100644
--- a/doc/cc65.sgml
+++ b/doc/cc65.sgml
@@ -731,10 +731,7 @@ This cc65 version has some extensions to the ISO C standard.
 <item>  There is pseudo variable named <tt/__A__/ that indicates accumulator register. 
         It can be used to highlight the returned variable in non-void
         function where result is hidden in the <tt/asm/ section.
-        every other variable. They are most useful together with short
-        sequences of assembler code. 
-
-    For example, in the function below
+        For example, in the function below
 
         <tscreen><verb>
         BYTE get_key_via_atarixl_romcall_device_k(void)      \

From e3fa247012a5664e6aa113dc8dd9d245bb51bf38 Mon Sep 17 00:00:00 2001
From: Piotr Kaczorowski <piotr.kaczorowski@gmail.com>
Date: Tue, 5 Jan 2021 17:48:46 +0100
Subject: [PATCH 24/26] Yet another fix typo in cc65.sgml

---
 doc/cc65.sgml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/doc/cc65.sgml b/doc/cc65.sgml
index 82da74f49..36b6d1071 100644
--- a/doc/cc65.sgml
+++ b/doc/cc65.sgml
@@ -739,7 +739,7 @@ This cc65 version has some extensions to the ISO C standard.
             asm("PHA");         \
             asm("LDA $E424");   \
             asm("PHA");         \
-            asm("RTA");         \
+            asm("RTS");         \
             return __A__;
         </verb></tscreen>
 

From 0884278ae3583f43c80611b06aae618cac688122 Mon Sep 17 00:00:00 2001
From: Piotr Kaczorowski <piotr.kaczorowski@gmail.com>
Date: Wed, 6 Jan 2021 16:02:25 +0100
Subject: [PATCH 25/26] Correction in documentation regarding __A__ pseudo
 variable

---
 doc/cc65.sgml | 22 +---------------------
 1 file changed, 1 insertion(+), 21 deletions(-)

diff --git a/doc/cc65.sgml b/doc/cc65.sgml
index 36b6d1071..f2fd69eba 100644
--- a/doc/cc65.sgml
+++ b/doc/cc65.sgml
@@ -709,7 +709,7 @@ This cc65 version has some extensions to the ISO C standard.
         places.
         <p>
 
-<item>  There are two pseudo variables named <tt/__AX__/ and <tt/__EAX__/.
+<item>  There are two pseudo variables named <tt/__A__, __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/
@@ -728,26 +728,6 @@ This cc65 version has some extensions to the ISO C standard.
         will give the high byte of any unsigned value.
         <p>
 
-<item>  There is pseudo variable named <tt/__A__/ that indicates accumulator register. 
-        It can be used to highlight the returned variable in non-void
-        function where result is hidden in the <tt/asm/ section.
-        For example, in the function below
-
-        <tscreen><verb>
-        BYTE get_key_via_atarixl_romcall_device_k(void)      \
-            asm("LDA $E425");   \
-            asm("PHA");         \
-            asm("LDA $E424");   \
-            asm("PHA");         \
-            asm("RTS");         \
-            return __A__;
-        </verb></tscreen>
-
-        <tt/return __A__;/ will indicate that result is stored in the accumulator
-        and suppress warning that function has no return statement. Instead, you can
-        also use <tt/#pragma warn (return-type, off) ... pragma warn (return-type, on)/
-        <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.

From a861d84011df82ac5f81a8c87ae3181fee393993 Mon Sep 17 00:00:00 2001
From: Oliver Schmidt <ol.sc@web.de>
Date: Wed, 6 Jan 2021 16:09:55 +0100
Subject: [PATCH 26/26] Fixed recent addition of __A__.

---
 doc/cc65.sgml | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)

diff --git a/doc/cc65.sgml b/doc/cc65.sgml
index f2fd69eba..24f4422f4 100644
--- a/doc/cc65.sgml
+++ b/doc/cc65.sgml
@@ -709,12 +709,13 @@ This cc65 version has some extensions to the ISO C standard.
         places.
         <p>
 
-<item>  There are two pseudo variables named <tt/__A__, __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
+<item>  There are three pseudo variables named <tt/__A__/, <tt/__AX__/ and
+        <tt/__EAX__/. They all refer to the primary register that is used
+        by the compiler to evaluate expressions or return function results.
+        <tt/__A__/ is of type <tt/unsigned char/, <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>