Parameter Passing

Stony Brook Compilers do not guarantee the process of passingparameters to procedures, unless:

• The procedure is declared in a Definition module(Modula-2) or Package specification(Ada95).
• The procedure is declared as an external procedure.
• Parameters of the procedure are referenced by built-in assembler statements.

Two methods are used to pass parameters to procedures:

• Address parameters are passed by pushing the address of the data onto the stack.
• Value parameters are passed by pushing the actual data onto the stack.

The method used to pass a specific parameter is determined bythe data type of the parameter, and whether or not it is a reference parameter.  Thefollowing conditions determine the method used:

• Modula-2 VAR parameters, and Ada95 out and in out parameters are passed by address.
• Arrays and records are always passed by address. For value parameters the type may be copied locally by the called procedure.
• Sets larger than two CPU words are passed by address.
• All other data types are passed by value except where noted below.

Some languages (notably C) always pass parameters by value.  Others(like FORTRAN) always pass parameters by address.  When interfacing to suchlanguages, you must take this into account when declaring the procedure.

Passing a pointer by value is equivalent to passing data of thetype it refers to by address.  In C, the effect of Modula-2 VAR, or Ada out/inout parameters is achieved by passing pointers.  If you are interfacing witha C procedure, you should determine whether you really want to pass a pointer orwhether you should translate it to an address parameter type.

The CPU Word

In order to more clearly discuss parameter passing with 16-bitand 32-bit compilers, we introduce the concept of a CPU word.  In order to efficientlypass parameters (put things on the processor stack), all items are passed as a multipleof the CPU word size.  CPU word sizes are as follows:

• 16-bit CPU word : 2 bytes
• 32-bit CPU word : 4 bytes

Address Parameters

Address parameters of all types are passed the same way.  Theaddress passed can be either near or far addresses.

• Near addresses take one CPU word.
• Far addresses take two CPU words.  The first CPU word pushed is the segment and the second is the offset.

Value Parameters

Value parameters are passed by pushing onto the stack the actualvalue of the data that is being passed.  At least one CPU word is pushed foreach parameter passed. The size of the value parameter passed is determined by thetype of the formal parameter, and not by the type of the actual parameter.  Forexample, you may be able pass an 8 byte integer value to a procedure that definesa formal parameter as a 4 byte integer value.  Language assignment rules applyhere. In this case, the length of the 4 byte integer type is pushed onto the stack,not the length of 8 byte integer.  The 8 byte integer value is truncated toa 4 byte integer value.

Value parameters are passed as follows:

IA-32

• All Integer, enumeration and set types with a size less than or equal to a CPU word size are passed by pushing a single CPU word onto the stack.  If the object being passed takes less than a full CPU word, then the high byte(s) of the CPU word passed is undefined.
• All Integer types larger than the CPU word are passed by pushing the two CPU words they occupy (the high-order CPU word is pushed first, then the low-order CPU word).
• All Floating point types are passed by pushing the entire value on the stack.  The highest-order CPU word is pushed first, and the lowest-order CPU word last.
• All COMPLEX types are passed by pushing first the imaginary then real component of the number.  Each component is a REAL type, and is pushed as a REAL type.
• Near pointer types are passed by pushing a single CPU word onto the stack.
• Far pointer types are passed by pushing the segment CPU word first, and then the offset CPU word.

SPARC

• SPARC passes the first 6 CPU word parameters in the r8..r13 registers. The order of register assignment is left to right. Stack space for the register parameters is always reserved in case the parameter must be moved to memory. Once the 6 register parameter registers are exhausted, the parameters are passed on the stack. A parameter which requires two CPU words may be passed partly in a register and partly on the stack. The compiler code generation will move the register part to the stack at procedure entry so the parameter image is coherent.
• All COMPLEX types are passed by reference.

Modula-2 Open array parameters

Open array parameters have HIGH bounds passed along with thearray address. The HIGH bounds are passed by value, have a type of CARDINAL, andthe highest array dimension is passed first. This means that the first array dimensionHIGH bound is the last passed and has the lowest address on the stack and is passedjust before the array address. Therefore HIGH bounds are found above the array addresson the stack and the first array dimension HIGH bound has the lowest address.

For example with a left to right parameter pass.

PROCEDURE Demo(first : CARDINAL; twoDim : ARRAY OF ARRAY OF CARDINAL);

1. first <= highest stack address
2. HIGH of dimension 2 of twoDim
3. HIGH of dimension 1 of twoDim
4. twoDim Address <= lowest stack address

Ada95 unconstrained array parameters

Unconstrained array parameters have their 'First and 'Last boundspassed along with the array address. The bounds are passed by value as an Integer,'Last is passed before 'First, and the highest array dimension is passed first. Thismeans that the first array dimension bound is the last passed and has the lowestaddress on the stack and is passed just before the array address. Therefore boundsare found above the array address on the stack and the first array dimension bound'shave the lowest address, and 'First has a lower address than 'Last.

For example with a left to right parameter pass.

TYPE TwoDimArray is array (1..10, 1..10) of Natural;

Procedure Demo(first : natural; twoDim : TwoDimArray);

1. first <= highest stack address
2. 'Last of dimension 2 of twoDim
3. 'First of dimension 2 of twoDim
4. 'Last of dimension 2 of twoDim
5. 'First of dimension 2 of twoDim
6. twoDim Address <= lowest stack address

Order of parameter passing

By default the compiler uses the StonyBrook calling convention.See Procedure attributes for detailed information.

Register parameter passing (IA-32 only)

Using procedure attributes you canforce the compiler to pass parameters in specific registers. This is not allowedfor procedures implemented in Modula-2. You can only use this for procedures implementedin Assembly code, or external procedures implemented in some other language or system.You can register parameter pass to Assembler and PureAsm procedures, since theseare not Modula-2 procedures.

Returning Values

Function procedures are procedures that return a value to thecaller.  The value of a function is returned either in registers or on the stack,depending on the data type.

The compiler uses the following registers to return values ofother data types:

IA-32

• One-byte values are returned in register AL.
• Two-byte values are returned in register AX.
• For 16 bit compilers: Four-byte values are returned in AX and DX, with DX holding the high-order word and AX the low-order word.
• For 32 bit compilers: Four-byte values are returned in register EAX.
• For 32 bit compilers: Eight-byte values are returned in EAX and EDX, with EDX holding the high-order DWORD and EAX the low-order DWORD.
• Functions returning floating point types return the value on the floating point processor stack.

SPARC

• 8, 16 and 32-bit value are returned in r8
• 64-bit values are returned in r8:r9
• Floating point returns are in f0