64-Bit Simulation Kernel
On Linux platforms (and some Windows platforms), a 64-bit address space is available for both the sequential and parallel kernels.
When compiling for 64-bit address space, the symbol VOSD_64BIT_SPACE is defined. This symbol can be used by model code to use different data types based on the address space.
Data Type Sizes
The 64-bit address space has some impact on various data types. When using the 64-bit kernel:
intremains a 32-bit storage area.longbecomes a 64-bit storage area on Linux (it remains 32-bit on Windows).long longbecomes a 128-bit storage area on Linux and 64-bit on Windows.- Addresses (pointers) are 64-bit values. (Any code that stores pointers into
intvariables is likely to generate incorrect results or crash when run in the 64-bit kernel.) doubleis not affected.
To allow models to behave similarly with various OSs and pointer sizes, Modeler provides a set of types with well-defined sizes in all situations:
Table 132: Address Pointer Data Types
| Data Type | Definition |
|---|---|
| OpT_Int32 | 32-bit signed integer |
| OpT_uInt32 | 32-bit unsigned integer |
| OpT_Int64 | 64-bit signed integer |
| OpT_uInt64 | 64-bit unsigned integer |
Modeler also provides a set of macros to define special values and formatting information used in printf-like APIs:
Table 133: Print Macros for 64-Bit Numbers
| Macro | Used For |
|---|---|
| OPC_INT64_MAX | maximum 64-bit signed integer value (2^63 – 1) |
| OPC_INT64_MIN | minimum 64-bit signed integer value (–2^63) |
| OPC_UINT64_MAX | maximum 64-bit unsigned integer value (2^64 – 1) |
| OPC_INT64_FMT | basic format string to display a 64-bit signed integer |
| OPC_UINT64_FMT | basic format string to display a 64-bit unsigned integer |
| OPC_INT64_PREFIX | 64-bit prefix string to include in more complex format strings |
Print macro example:
OpT_Int64 i64;
OpT_uInt64 u64;
printf ("64 bit signed = " OPC_INT64_FMT "\n", i64);
printf ("64 bit unsigned = " OPC_UINT64_FMT "\n", u64);
printf ("64 bit signed on 12 char width = %12" OPC_INT64_PREFIX "d \n", i64);
printf ("64 bit unsigned on 12 char width = %12" OPC_INT64_PREFIX "u \n", i64);
Event Codes
The code associated with an event becomes a 64-bit integer in the parallel kernel. The event-creating KPs that take a code argument use OpT_Int64 as the argument type. Code that passes a 32-bit integer does not have to be modified; the 32-bit values are converted automatically to 64-bit.
The KP op_ev_code() returns a 64-bit integer value. If this value is assigned to a 32-bit variable, the value is downcast.
Note: If the event code is used to pass pointers, then retrieving that pointer using op_ev_code() and assigning it to a 32-bit value will fail in a 64-bit address space.
TDAs
TDA integer values are stored as 64-bit numbers. The KPs to set integer TDAs have an OpT_Int64 argument (with appropriate casting of 32-bit values). The KP op_td_get_int() returns an OpT_Int64.
Packet and ICI Fields
Fields of type integer are 32-bit values, regardless of the address space. Use the int64 field type to store 64-bit values. When obtaining the value of int64 fields, model code should use OpT_Int64 variables.
Specific field types exist to represent packet ids and object ids; these correspond to 64- and 32-bit integers, respectively. When obtaining the value of a packet id field, use a variable of type OpT_Packet_Id.