This is a summary of the latest features and changes introduced in Astrobe for ARM Cortex-M0.
The following details are of interest if you are using the Module and Application Disassemblers:
There are two instruction sets, called Thumb-1 and Thumb-2, used in Cortex-M processors Thumb-1 instructions are 16-bit and most can only access the low general purpose registers r0 .. r7 The equivalent 32-bit Thumb-2 instructions can be used to access additional general purpose registers r8 .. r11. In general Thumb-1 results in smaller code size and Thumb-2 results in better performance.
Most Thumb-1, but only a few Thumb-2, instructions can be used in Cortex-M0. All Thumb-1 and most Thumb-2 instructions can be used in Cortex-M3, M4 and M7.
Additional instructions are used in Cortex-M4 and M7 for the hardware floating point operations: add, multiply, divide and square root. These are implemented in software (the FPU library module) in Cortex-M0 and M3.
- The compiler allocates registers in sequence from r0 up to r7.
- Local variables in leaf procedures are not stored in registers due to the limited availabity of general purpose registers.
- The maximum number of parameters that can be passed to a procedure is eight. Some parameters, e.g. arrays and records, require two registers.
The frame pointer (FP = r12) is no longer required resulting in less overhead on procedure entry code. The register is reserved for future use.
Numeric Case Statements
- Cortex-M0: The Thumb-1 Branch and Exchange (BX) instruction is used. The size limit of the CASE statement is 32KB.
- M3, M4 and M7: The Thumb-2 Table Branch (TBH) instruction is used resulting in more compact code. The size limit of the CASE statement is 64KB.
- Strings now immediately follow the code that accesses them.
- Shifts involving two constants are evaluated at compile time.
The built-in procedure CLEAR initialises every word in a record or an array variable v to zero.
CLEAR(VAR v: <record or array type>)
It is designed to be used to simplify and optimise the code required to initialise complex data structures. Its use can be seen in the Performance example and procedure InitHeader in the HCFiler library module HCDir.
The expression which is passed as a parameter to ORD can be a BYTE expression as well as a CHAR, BOOLEAN or SET.
BITS will report an error if its actual parameter is a BYTE expression. However, ORD can be used to convert the expression to an INTEGER e.g.
setvar := BITS(ORD(bytevar))
ALIGN can be used to ensure that the following instruction is aligned on a word boundary. It will insert a NOP instruction at the current code location if necessary.
PROCEDURE EMITH(instruction: INTEGER)
EMITH (EMIT Halfword) inserts the constant value instruction, which represents a 16-bit ARM Thumb instruction, at the current code location. A compilation error results if the value is not a valid 16-bit positive integer.
The read-only restriction for SYSTEM.VAL has been removed e.g. the following (where x is a word-length variable of any type) is a valid statement :
SYSTEM.FP, SYSTEM.SP, SYSTEM.LNK and SYSTEM.PC have been removed. SYSTEM.LDREG and SYSTEM.REG with values 12, 13, 14 and 15 respectively, should be used instead.
Oberon Language Extensions
ARRAY OF BYTE
An existing Oberon language extension allows a procedure with a formal parameter which is an array of bytes with a fixed size to accept actual parameters of any type, except POINTER, whose size is the same number of bytes.
The converse situation is now also catered for i.e. a procedure with a formal parameter of any type, except POINTER, can accept an actual parameter which is an array of bytes of the same size.
These extensions are primarily designed to simplify and optimise the code required for tasks such as de-serialising / serialising complex data structures for Input / Output operations. An example of its use, ByteArrays.mod, is included with the General examples supplied with Astrobe.
Dynamic Local Array Variables
Eliminating the frame pointer resulted in the removal of support for dynamic local array variables. These can be replaced by local array variables with a fixed size.
Oberon Language Clarifications and Restrictions
- All parameters of exported procedures must be of exported data types.
- String constants cannot be exported. They can be assigned to exported variables which are read-only in modules that import them.
- Anonymous record and array variables can now be exported
- Pointer types can only point to named record types e.g.
Item = POINTER TO ItemDesc;
ItemDesc = RECORD value: INTEGER; next: Item END;
Item = POINTER TO RECORD value: INTEGER; next: Item END;
- Pointer types can only point to global record types.
- A warning is reported at compile-time when a pointer type, whose base type is a private record, is exported.
Item* = POINTER TO ItemDesc;
ItemDesc = RECORD value: INTEGER; next: Item END;
This is referred to as an opaque pointer.
- An error is reported at compile time if a type test is performed on an imported opaque pointer variable, or it is passed as parameter to the built-in function NEW.
Thumb Code is a new configuration option which directs the compiler to make more use of 16-bit Thumb-1 instructions instead of equivalent 32-bit Thumb-2 instructions when generating code. The typical result is a reduction in code size of about 30%. It is permanently checked on for Cortex-M0.
Configuration ID is a new configuration option which can be used for any purpose by the programmer. It is an INTEGER value that can be accessed within an application by reading the value of the variable LinkOptions.ConfigID.
The Line Numbers configuration option has been renamed to Stack Trace to avoid confusion with the Tools > Preferences > Line Numbers option.
The obsolete Crystal Frequency configuration option has been removed.
- The Module Disassembler has been rewritten to disassemble the code after it has been generated not while it is being generated. The resulting disassembly is more accurate as it can show resolved internal backward and forward references.
- The 16-bit SEV, WFE, WFI, and YIELD instructions are supported by the disassemblers.
Compile, Build and Rebuild
- The summaries include the number of lines compiled and the compilation time in milliseconds for each module and totals for all modules.
- The compiler reports an error if a constant parameter has a value which is out of range for the built-in functions LSL, LSR, ASR, ROR, INCL, EXCL, BFI, BFX and the SYSTEM functions LDREG, REG and BIT.
- The syntax diagram for the Type CASE statement has been added to the Oberon Quick Reference Help file.
Alerts are a new type of compiler message like Warnings and Errors. An Alert is the same as an Error except that an object file and executable is still produced. This is to allow the disassemblers to be used to diagnose the cause of the problem. The behaviour of the executable will be unpredictable until all alerts have been eliminated.
The options shown in the linker map for each module now include t+ or t- corresponding to whether the Thumb Code configuration option was checked on or off when that module was last compiled
Astrobe Terminal displays error messages instead of exceptions in the following situations:
- A serial port is not specified when Reset is pressed.
- A serial port is already in use when Reset is pressed.
- An attempt is made to write in the input area before Reset has been pressed.
The default Editor Font is Consolas, Size: 9.
Additional procedures Set and Reset are more efficient substitutes for Put with a constant state parameter equal to TRUE or FALSE respectively.
Info includes additional information related to the status of the various clocks and PLL.
ByteArrays is a new example which shows how structured variables can be interpreted as arrays of bytes and vice versa.
Supported Development Boards
Library modules, example source code and configuration files are included for use with the currently supported development boards.
Details of 3rd-party Oberon-07 libraries and tools which allow you to target the Raspberry Pi RP2040 microcontroller are on the Oberon RTK website.
Problems fixed in v9.1
- SYSTEM.BIT now generates correct code when the bit number is 31.