Using a Custom Bitfile in C Code

From Cpre584
Jump to: navigation, search

In the simplest form, running a routine on the coprocessor requires two parts:

  • A call to the cny_get_signature function to get the signature of the custom personality (bitfile) you created
  • A coprocessor funcation call (ex: l_copcall_fmt, d_copcall_fmt, etc)

If you want to pass parameters to your function call, you use an assembly file to marshal registers back and forth between the processor's registers (starting at A8) and the coprocessor's AEG registers. If you must pass a lot of data, it's better to pass a pointer to the data and allocate memory on the coprocessor board.

Reusing Code

Don't copy and paste your code to use with Convey; instead, use the __CONVEY preprocessor directive (the cnycc compiler always defines __CONVEY). You can then compile the same source code with both gcc and cnycc. For example:

  #ifdef __CONVEY
    int *bigarray = cny_cp_malloc(10000);
    int *bigarray = malloc(10000);

Getting the Signature

The standard way to get a signature is using:

  cny_image_t sig1, sig2;
  int stat;

  if (cny_get_signature)
    cny_get_signature("your personality name", &sig1, &sig2, &stat);
    fprintf(stderr, "ERROR:  cny_get_signature not found\n");

  • stat returns 0 on success
  • sig1 is required, and contains a 64 bit signature for your custom bitfile/personality/AE.
  • sig2 is not used, but may have use in the future.

Allocated Memory on the Coprocessor Board

System memory and memory used for the coprocessor are physically separate. In the example C file, function calls such as cny_cp_malloc and ny_cp_posix_memalign are used to allocate memory on the coprocessor.

See: Convey Programmers Guide (.pdf) - Chapter 9

Making a Coprocessor Call

The vector adder example uses:

act_sum = l_copcall_fmt(sig1, cpVadd, "AAAA", a1, a2, a3, size);

The first two arguments are always (1) the bitfile signature and (2) a assembly function name. The third argument lists the type and number of optional arguments that are passed and whether the get stored in the application hub's A or S registers. For example, "AAAA" means there are 4 long (64 bit) variables that are stored starting in register A8; suitable for the address of a location in the coprocessor's memory system.

In this case, the 'l' at the beginning of l_copcall_fmt means the return type is a long (64 bits). Other return types can be used:

  • extern float f_copcall_fmt(cny_image_t,float(*func)(void),char*, ...);
  • extern double d_copcall_fmt(cny_image_t,double(*func)(void),char*,...);
  • extern int i_copcall_fmt(cny_image_t,int (*func)(void),char*, ...);
  • extern long l_copcall_fmt(cny_image_t,long (*func)(void),char*, ...);
  • extern void copcall_fmt(cny_image_t,void (*func)(void),char*, ...);
  • extern void* v_copcall_fmt(cny_image_t,void*(*func)(void),char*, ...);

In addition to using A's for the third argument, you may use:

  • 'a' pass a 32 bit quantity in an A register
  • 'A' pass a 64 bit quantity in an A register
  • 's' pass a 32 bit float quantity in an S register
  • 'S' pass a 64 bit double quantity in an S register
  • 'l' pass a 32 bit quantity in an S register (lower-case L)
  • 'L' pass a 64 bit quantity in an S register

The first value loaded into an A register is loaded into A8, the second into A9, …

Similarly, the first value loaded into an S register is loaded into S1, the second into S2, …

See page 122 (Appendix G) of the Programmer's Guide.