IAR Information Center for 8051

• Important information
• New features
• Known problems
• Program corrections
• User guide corrections
• Miscellaneous
• Release history

Important information

• Backward compatibility issues
  The following information is relevant when upgrading from IAR Embedded Workbench versions prior to version 10.10.
  The option C/C++ Compiler>Optimizations>Disabled register banks cannot be used when linking existing libraries, built with earlier versions of IAR Embedded Workbench, that contain code which explicitly switches register banks. Switching register banks is done, for example, by using #pragma register_bank for interrupt functions or by setting the value of the symbol ?REGISTER_BANK to something else than 0.
  The linker issues a warning when such problems are detected, but because this control mechanism is located in the file cstartup.s51 it is important that users who have replaced this file with their own version adapt it using the file that comes with the product as a template.
• Some of the names defined for interrupt vectors in the I/O include files (8051\inc\io<device name>.h) have been changed. This might require a rebuild of the source code.
• If you are using multi-file compilation together with the banked memory model, you must make sure that the size of each multi-file compiled module does not exceed the size of the code bank on the device you are using.
• Please note that libraries built with versions of IAR Embedded Workbench for 8051, earlier than 8.30, will have to be rebuilt to link correctly with an application built with version 8.30.

New features

• None

Known Problems

• [EW8051-796, EW21366] If the banked code model and the option --place_constants=code are used, constants must be placed in the root bank and NOT in a code bank. This is regardless of whether the device you are working with supports the command line option --has_cobank or not.

Program Corrections

• None

User guide corrections

• None.

Miscellaneous

• None.

Release history

10.20.1, April 2018

Program corrections

• [EW8051-941] When casting a pdata pointer to an xdata or generic pointer, the compiler will read the value of the pdata bank register. This can cause problems when there are more than one sfr page.
• [EW8051-936] On optimization level High, variables that are written to by the library function memset can be treated incorrectly as unchanged.

10.10.3, October 2017

Program corrections

• [EW8051-929] The compiler sometimes fails to perform cross-call optimization in cases where it should be performed.

10.10.1, April 2017

New features

• It is now possible to control cross call optimization using the option C/C++ Compiler>Optimizations>Cross call.
• A new compiler option that disables the use of register banks is available, C/C++ Compiler>Optimizations>Disabled register banks. This can be used to create more efficient code if the application does not have to switch register banks.

Program corrections

• [EW26204]
  In some situations, using enum types together with multi-file compilation can cause an internal error in the compiler.
• [EW26416]
  Using another calling convention than the default when declaring a function can sometimes result in the default frame being accessed instead of the declared one.

9.30.3, October 2016

Program corrections

• [EW26203]
  Placing banked functions in user-defined segments can generate illegal calls when the optimization level is set to High.

9.30.2, February 2016

Program corrections

• [EW25851]
  In rare cases, the compiler can optimize away a clear carry instruction after a switch on short or long integers while using optimization level medium or higher.

9.20.1, May 2015

9.10.3, February 2015

Program corrections

• [EW25202]
  The code bank number is not saved by setjmp when using plain core 8051.

9.10.2, February 2015

Program corrections

• [EW25137]
  Optimizing conditional branches can in some circumstances cause an internal error.
• [EW25146]
  Taking the address of a variable on the XDATA stack can sometimes generate an internal error when using shadowed data pointer registers.

9.10.1, December 2014

Program corrections

• [EW24452]
  The volatile operands of certain expressions are referenced in an incorrect order.

8.30.4, July 2014

Program corrections

• [EW24838]
  In C++ a function that returns a bool and is declared __monitor will return an incorrect value.
• [EW24438]
  Some labels in the assembler support routines causes a misra error to occur at link stage.

8.30.1, December 2013

Program corrections

• [EW23770]
  The compiler warns about low_level_init when using the banked code model.
• [EW23812]
  The bool type used in the intrinsic function __tbac(bool) is not available when using the C89 dialect.
• [EW24083]
  Far function calls can incorrectly generate banked function calls instead.

8.20.2, February 2012

Program corrections

• [EW23759]
  A bug fix caused the compiler to generate more code when moving 8-bit data from __data to __data memory and from __idata to __data memory.

8.20.1, October 2012

New features

• The behaviour of the __monitor function keyword has been corrected to only restore the global interrupt enable bit at exit.

Program corrections

• [EW23540]
  Using a static anonymous union placed at an absolute address caused an internal error in the compiler.
• [EW23537]
  A long 16 bits shift with bit mask and following pointer update could on high optimization cause an Internal Error. This has been corrected.
• [EW23480]
  Inline assembler statements were always considered equal when optimizing.
• [EW23406]
  __set_interrupt_state now only restores the enable global interrupts bit.
• [EW23296]
  A loop that initialized an array larger than 255 members with a constant value could fail to initialize the members above 255.

8.11.4, May 2012

Program corrections

• [EW23188]
  Fixed a rare problem with sparse byte switches and 24-bit dptr.

8.11.3, April 2012

Program corrections

• [EW22878]
  A problem where an anonymous union could generate an error if it was also declared extern in the same module has been corrected.
• [EW23071]
  The compiler no longer generates object files with corrupt symbol name information for certain constructions involving nested templates.

8.11.1, February 2012

Program corrections

• [EW22648, EW22648, EW22742]
  Typos and omissions in the following device header files has been corrected:
  ioC8051F540.h, iop89LV51Rx2.h, ioHVC2xyzA.h and iop89LV51Rx2.h

8.10.4, December 2011

Program corrections

• [EW22876]
  16-bit equal operations could fail if the low byte had value zero.

8.10.3, September 2011

Program corrections

• [EW22647]
  The inclusion of PDATA_Z in ?RESET_PDATA_BANK was erroneously root'ed. It has now been given the attribute NOROOT. (The problem you might have seen as a consequence of this was the linker error: "Error[e16]: Segment PDATA_Z (size: 0 align: 0) is too long for segment definition. At least 0 more bytes needed.")
• [EW22656]
  Debug info could fail for interrupt functions in xdata or pdata reentrant mode causing wrong addresses shown for local variables in the interrupt function.
• [EW22697]
  Subtraction of two short variables could generate faulty code. This has been corrected.
• [EW22699]
  Comparing a short variable in idata memory against zero could fail to check the high byte.

8.10.1, May 2011

New features

• C99 compliance
  The IAR C/C++ Compiler for 8051 now supports the ISO/IEC 9899:1999 standard. Refer to the Embedded Workbench Migration Guide (v8) for more information on how this affects your application.
• Improved runtime library support for constants placed in code memory
  The CLIB runtime library has been extended with functions that accept constants placed in code memory as parameters. Please refer to the Compiler Reference Guide for more information on how to use this in your application.

Program corrections

• [EW20971]
  A range error could occur at link time if a large array was accessed with an immediate index that was larger than 0x7FFF. This has been corrected.
• [EW21354]
  In some circumstances, the compiler could generate an internal error when the __idata_overlay keyword was used on a function. This has been corrected.
• [EW21913]
  A signed division of an expression with a negative divider no longer gets optimized into using an unsigned division.
• [EW21981, EW21990]
  The compiler did not correctly preserve the accumulator for certain 8-bit comparison operations. This has been corrected.
• [EW21996]
  An internal error was issued in the compiler by expressions such as 'const volatile __code int a;'. This has been corrected.
• [EW22111]
  On the highest optimization level, the compiler could sometimes neglect to save and restore the content of the accumulator when the function inlining and the common subexpression elimination optimizations were used to optimize switch cases for bitfields. This has been corrected.
• [EW22158]
  Bitwise OR of a value to an output port register would lead to the compiler generating a superfluous MOV instruction. This has been corrected.
• [EW22280]
  Casting a __pdata address to a short generated faulty code. This has been corrected.
• [EW22284]
  Code could be displayed as data in the Disassembly window if it was immediately preceded in memory by a data object. This has been corrected.
• [EW21115, EW21180, EW21367, EW22039, EW22132, EW22253]
  The documentation has been updated in this release, correcting several errors from the previous edition.
• The library functions that update the XDATA stack pointer (XSP) and the augmented extended stack pointer (ESP) have been modified to ensure that stack pointer update operations always are carried out atomically.

7.60.1, June 2010

Program corrections

• [EW20949]
  The symbols _heap_of_memory and _top_of_heap are no longer implemented as variables but rather as macros that automatically find the beginning and end of the XDATA_HEAP segment.
• [EW20958]
  Realloc in Clib could fail if the pointer plus its size caused the pointer to wrap around. This has been corrected.
• [EW20976]
  The compiler now uses the command line option --dlib_config instead of -D_DLIB_CONFIG_FILE=FILE.
• [EW21103]
  Using a boolean type as prameter to a function would fail if the banked code model together with a ?CBANK_MASK other than 0xFF was used.
• [EW21104]
  In case banked functions were explicitly placed in a user defined segment (e.g using the #pragma location-directive) the compiler could at times generate code which triggered a range check error from the linker.
• [EW21138]
  Combining the options --code_model=far with --data_model=large and --place_constants=code made it impossible to build the CLIB library, this has been corrected.
• [EW21476]
  Initialization of global C++ objects in the banked code model has been corrected.
• [EW21574]
  The saving and restoration of DPTR1 in the function prologue and epilogue was incorrect if more than one DPTR, combined with a function declared with the __idata_overlay calling convention, was used. This has been corrected.
• [EW21629]
  The upper byte of the PDATA stack pointer was hardwired to the same address as the P2 register. This has been corrected.

7.51A, March 2009

Program corrections

• [EW20344]
  The _EXTENDED_STACK_END symbol was incorrectly defined. This has been corrected.
• [EW20509]
  The address of the FLSTAT register was wrong in the device io file for the Silabs C8051F127 device. This has been corrected.
• [EW20869]
  The address of the PLL0CN register was wrong in the device io file for the Silabs C8051F120 device. This has been corrected.
• [EW20965]
  Testing whether an unsigned character value was greater than or equal to 0xFF would lead to an erroneus error message. This has been fixed.
• [EW20870]
  Passing a variable placed with the keyword __data as a parameter when using calling convention pdata_reentrant and more than one dptr could make the compiler stop as in a uncontrolled termination.
• [EW20747, EW20748]
  When using multiple file compilation (--mfc), defining a variable with an absolute location in more than one translation unit was erroneously signalled as an error.

7.50C, September 2008

Program corrections

• [EW20134]
  Support has been added to handle the Silicon Laboratories C8051F12x-F13x devices COBANK-selection bits (part of the PSBANK register on address 0xB1) correctly within the compiler if one uses the banked code model. Please specify the command line option "--has_cobank" if you are using one of these devices.
• [EW20187]
  In C++ mode, the compiler could sometimes erroneously generate a copy assignment operator for memories that were too small for a particular class/struct/union, resulting in spurious 'type too large' errors.
• [EW20286]
  A shift combined with a bitwise OR of a SFR one bit bitfield could generate wrong code in specific cases.
• [EW20457]
  Empty while loops will no longer cause an internal error.

7.50B, July 2008

Program corrections

• [EW20274, EW20277, EW20293, EW20299]
  The compiler would fail when attempting to generate code for bitfields wider than 1 bit.

7.50A, June 2008

New features

• New options added to the compiler:
  
  • --mfc
    Use this option to compile several source files in one compilation. The advantage of multi-file compilation is that it gives the interprocedural optimizations a larger set of functions to work on.
    Syntax:
    
    icc8051 -mfc file1 {file2 file3 ...}
    
    The order in which the source files are given on the command line may affect the code size.
    
    
  • --discard_unused_publics
    Use this option together with the --mfc option if the compilation unit is the whole application apart from the library. This can improve the optimization further because the compiler will assume that there are no references to any symbol not included in the current batch of files being compiled (ie. that the compilation unit is the whole application apart from the library).

Program corrections

• [EW20035]
  The default memory attribute of an interrupt function when the banked memory model is used is now __near_func.
• [EW19424]
  It is now possible to declare volatile constants with the __code attribute set.
• [EW18278, EW19909]
  Placing a C++ variable that needed dynamic initialization in a non-writable memory caused an internal error. This is now treated as a regular error.
• [EW18243, EW19911]
  Instantiating a template function or a template class on a member of an anonymous union caused an internal error in the compiler.
• [EW19924, EW19979]
  In C, a bool cast at the top level of a Boolean controlling expression could result in an internal error in the compiler.

7.40A, February 2008

New features

• The storing and restoration of the original state of the Interrupt Enable register (IE) during an update of the XDATA or Extended Stack Pointer is now interrupt safe.

Program corrections

• [EW15883]
  Added peephole optimization to improve code generation for bit manipulations in the SFR area.
• [EW16816, EW18918]
  Converting a legal constant value to a pointer could earlier incorrectly generate the warning: Warning[Pe1053]: conversion from integer to smaller pointer.
• [EW18065]
  Attempts to create pointers to the SFR area now generates a correct error message.
  
• [EW18258]
  The compiler now generates a proper error message when the parameter passed to __tbac is not declared __bit. In previous versions of the compiler an internal error was generated.
• [EW18277]
  A struct declared in SFR memory cannot be used for copying because it is not possible to have a pointer to the SFR area. A more descriptive error message has been added for this situation.
• [EW18821]
  Located variables can now be defined in other modules than the ones where they are used.
• [EW18911]
  When the result of a multiplication was used as adress for @ and the result overflowed to a non-negative (signed) integer constant an Internal Error was generated. This has been replaced with a more descriptive error message.
• [EW18914, EW18915, EW19041, EW19045, EW19618]
  Calling an __idata_reentrant function from an __xdata_reentrant function could sometimes generate an Internal Error. The same Internal Error would also be triggered when a struct of size 256 to 511 was passed as a parameter to an __xdata function.

7.30B, September 2007

New features

• New design of banked code model
  The banked code model was redesigned in this release.

Program corrections

• [EW19423]
  Declaring a pdata reentrant interrupt function would generate an internal error.

7.30A, August 2007

Program corrections

• [EW16031, EW16043]
  The use of a __bit bool variable in a switch statement generated an internal error.
• [EW17380]
  Including an I/O definition header and referencing more than one SFR from C++ code resulted in an incorrect warning when linking.
• [EW17385]
  Multiple-line, inline assembler used to generate an internal error from the compiler.
• [EW17388]
  Under certain specific circumstances the loop unrolling optimization would cause an internal error
• [EW17389]
  If a label ending with ':' was used inside an inline assembler function an internal error was given.
• [EW17531]
  An empty conditional loop is no longer incorrectly removed as dead code.
• [EW17573]
  An internal error occured when multiplying signed values between 0x4000 and 0x7FFF by two.
• [EW18359]
  Conditional expressions using the ? operator would generate illegal code if the contitions involved manipulation of SFR registers.
• [EW19257]
  A problem that occured when attempting to allocate more memory than 0x1F bytes on the XDATA heap when using banked code has been resolved.

7.20H, November 2006

Program corrections

• [EW18238]
  The compiler failed to properly preserve the carry flag for functions that return a bool value and have more than three bytes of auto area allocated. This resulted in the return value being corrupted.
• [EW18018]
  The code that generated text for a constant only worked for 8-bit constants. This only affects CALL instructions which uses a special output mode.
• [EW18406]
  The code generator would misstakenly try to generate MOV instructions for all accesses to located variables where the address was less than 256 and the memory attribute was not __pdata or __generic. This was especially bad for __xdata variables which should have used MOVX. This affected all located __xdata, __ixdata, and __xdata_rom variables.
• [EW18443]
  The code generator did not clear the accumulator before accessing the second byte of a 16-bit pointer located in __code, __far_code or __huge_code memory when doing pointer arithmetics. Only expressions of the kind "p + i" where p is a 16-bit pointer stored in the code memory and i is any integer expression are affected (independent if the pointer expression is used as an lvalue or an rvalue).
• [EW17854]
  The assemble source created by icc8051 compiler no longer generates "*" comments.
• [EW17770]
  The optimizer could sometimes mistakenly combine two if-else statments into one common set.
• [EW14809]
  An error in the instruction level analysis phase could cause the optimizer to erronously remove a load instruction. A support routine used the wrong target address (off by one) when tracking memory accesses in conjunction to pushing values onto the stack.
• [EW18050]
  A MISRA error in sysmac.h has been fixed.

7.20D-7.20G

• These versions were never publically released.

7.20C, January 2006

Program corrections

• [EW17693, EW17694]
  Accesses to volatile declared variables could under certain circumstances be removed at optimization level medium or high (6 or 9). Now the volatile information is correctly handled and operations on volatile declared variables will not be removed at higher optimization levels.

7.20A, October 2005

Program corrections

• [EW15870]
  Now an error message is generated when the INTVEC segment is not located at the address zero. This is because the INTVEC segment contains the reset vector which should normally be located at the address zero. Previously no warning or error message was given.
• [EW16044, EW17131, EW17343]
  Mixing switch-case and if-else constructions can generate internal error: [CoreUtil/General]: 1: bad byte count for graph ... when the file is compiled. This error can often be avoided if the nested code statements are rewritten.
• [EW17157]
  The compiler could in certain cases generate a JBC instruction instead of a JNB instruction. This has now been corrected.
• [EW17241]
  If the same interrupt vector was used for different interrupt functions, located in different files, no error or warning message was generated and only one of the interrupt functions was placed in code memory, while the others were removed. Now a warning message is generated and all interrupt functions using the same interrupt vector are located in code memory.
• [EW17344]
  Accessing a structure member from a >struct located in the __sfr memory could generate the internal error: Internal error: [CoreUtil/General]: Illegal state. This problem has now been corrected.
• [EW17345]
  Data placed at an absolute location, with the @ operator or the #pragma=location could previously be placed in the wrong memory type. This could result in the same memory address being used for more than one variable, which generated the XLINK error message: Error[e24]: Segment ... overlaps segment ....

7.10A, July 2005

Program corrections

• [EW15834]
  Absolute located variables are now allowed in bdata memory, e.g., __no_init __bdata unsigned char foo @ 0x22.
• [EW15994]
  The error message generated by the compiler, when too much code is generated in the Baseline and Kickstart editions, has been corrected. The correct limits are 8192 and 4096 bytes.
• [EW16020]
  Cast from a signed char to a __far or __huge pointer generated an internal error. Now these cast operations are correctly generated.
• [EW16094]
  The inline register restore sequence performed at function return could be erroneously removed. This most likely occurred for interrupt functions using a separate register bank. This has now been corrected.
• [EW16675]
  Older versions of ubrof (ubrof5 or ubrof6) can now properly be generated from the compiler. Registers will be correctly displayed in third-party tools reading older ubrof versions.

6.11A-P04314, November 2004

Program corrections

• [EW16045]
  Incorrect __sfr variable declarations could previously generate an internal error. Now an error message is instead generated.
• [EW16096]
  Access to an __sfr declared variable could previously erroneously be performed with indirect addressing and thus accessing IDATA memory instead of SFR memory. This has now been corrected.

6.11A, October 2004

Program corrections

• [EW15795]
  The compiler now always clean the stack before an infinite loop. Earlier you could get the internal error: [CoreUtil/General]: Different paths give different backtrace info. The compiler removed the stack cleaning code because it was dead depending on the infinite loop. However, this assumtion was incorrect as an interrupt can occur.
• [EW15775]
  Interrupt functions that performs in-line saving and restoring of callee-saved registers now restores the saved registers correctly, independent of the optimization level.
• [EW15770]
  The internal error: [CoreUtil/General]: Illegal unspillreg, has been fixed. The error occured when a bool variable that had been spilled from the VB register to the B register was restored (unspilled) to the VB register.
• [EW15333]
  The __low_level_init function, used to initialize I/O registers or to omit the default initialization made by CStartup, is now declared as __root and will always be included in the linked application.
• [EW15311]
  #pragma dataseg and #pragma location now work as intended. The produced code is now located in the specified segment.
• [EW15047, EW15080]
  The __task keyword is now acknowledged.
• [EW15004]
  Variables located in a specific, user-created, segment with #pragma location="NAME" or @ "NAME" now ends up in this segment.
• [EW11851]
  Version 6 of 8051 IAR C/EC++ Compiler is totally new compared to version 5. All problems exsisting in version 5 have been corrected in version 6. There is for instance no longer a problem to combine banked code with high optimization.

6.10C, June 2004

Program corrections

• [EW15564]
  Operations on __bit declared variables and 1 bit bitfields have been improved. The direct bit addressing mode is used more often for instructions supporting this addressing mode.
• [EW15514]
  The volatile keyword was ignored and the generated code could be corrupt for a volatile variable.
• [EW15332]
  The not operation applied on bit fields now works correctly, e.g. P1_bit.b6 = !P1_bit.b6. This operation and lot of other operations on bitfields now also produce more efficient code.
• [EW15264, EW15573]
  Bit operations combined with xdata access could generate an internal error: [CgDriver]: Coloring failed.
• The following bit optimizations were implemented
  • Assignments (bytes in bit-addressable space)
    • byte.bit = constant
    • byte.bit = byte.bit
    • byte.bit &= constant
    • byte.bit |= constant
    • byte.bit ^= expr
  • Assignments (bytes in non-bit-addressable space)
    • byte.bit = expr
  • Compare (bytes in bit-addressable space)
    • if(byte & aMask)
    • if(byte.bit) and if(!byte.bit)

6.10A, February 2004

• Initial release.