Send vector of bytes to BDPI functions

[nirajnsharma]

For comparing quire outputs between Melodica and SoftPosits, the BDPI functions need to be able to handle bit vectors longer than 64-bits.

• Figure out how to send a vector of bytes to a BDPI function
• Make changes to fm*, p2q and q2p testbenches

[nirajnsharma]

Here the helper function converts a 7-bit input to its one-hot notation, filling out a 128-bit vector. The important thing to remember is that C function returns void and the first argument is the unsigned int* to the return structure.

I tried to modify the fma16 checker, but got a little tied up in knots:

//               Quire-Result         Input-Quire        Posit-A          Posit-B
void fmaAdd16 (unsigned int* result, unsigned int* a, unsigned int c, unsigned int d) {
   posit16_t x,y; quire16_t qZ;
   x = castP16(c); y = castP16(d); qZ = q16_clr(qZ);
   qZ = castQ16(a[3], a[2], a[1], a[0]);  // IS THIS CORRECT?
   qZ = q16_fdp_add(qZ, x, y);
   for (idx=0; idx<4; idx++) result[idx] = qz.V[idx];  // IS THIS CORRECT?
   return;
}

[nirajnsharma]

@riyarjain
I have verified the new FMA testbench and the FMA softposit checked function with (quire, 0, 0) inputs as you suggested. The input is equal to the output. So the BDPI plumbing is working as expected. I have verified this for random input quire values.

However, when I replace the input posits with non-zero values, all hell breaks loose. My guess is that I have made a hash of the softposit code where I am handling the posits. Can you please take a look at softposit_wrappers::c_fmaAdd16 and see if anything rings a bell?

void c_fmaAdd16 (unsigned int* quire_out, unsigned int* quire_in, unsigned int posit1, unsigned posit2) {
   posit16_t x = castP16 ((uint16_t) posit1);
   posit16_t y = castP16 ((uint16_t) posit2);
   quire16_t qZ; q16_clr (qZ);

   // Form quire from constituents
   uint64_t q1 = (uint64_t) quire_in[3];
   uint64_t q0 = (uint64_t) quire_in[1];
   q1 = ((q1 << 32) | ((uint64_t) quire_in[2]));
   q0 = ((q0 << 32) | ((uint64_t) quire_in[0]));
   qZ = castQ16 (q0, q1);
   qZ = q16_fdp_add(qZ, x, y);

   // Break it back into its constituents
   q0 = qZ.v[0];
   q1 = qZ.v[1];

   quire_out[0] = (unsigned int) q0;
   quire_out[1] = (unsigned int) (q0 >> 32);
   quire_out[2] = (unsigned int) q1;
   quire_out[3] = (unsigned int) (q1 >> 32);
}