Compiler Explorer

Source code

#define float_v float2
#define uint_v uint2
#define ushort_v ushort2
#define CONVERT_FLOAT_V convert_float2
#define CONVERT_UINT_V convert_uint2
#define CONVERT_USHORT_V convert_ushort2
#define AS_UINT_V as_uint2

#define BUILTIN_CONVERT_FLOAT_V(x) __builtin_convertvector((x), float_v)
#define BUILTIN_CONVERT_UINT_V(x) __builtin_convertvector((x), uint_v)

#define BOOL_AS_UINT_V AS_UINT_V
#define limb16_v ushort_v

typedef struct _int75_v
{
  uint_v d0,d1,d2,d3,d4;
}int75_v;

typedef struct _int150_v
{
  uint_v d0,d1,d2,d3,d4,d5,d6,d7,d8,d9;
}int150_v;

void mod_simple_75(int75_v * const res, const int75_v q, const int75_v n, const float_v nf
)
/*
res = q mod n
used for refinement in barrett modular multiplication
assumes q < 6n (6n includes "optional mul 2")
*/
{
  __private float_v qf;
  __private uint_v qi;
  __private int75_v nn;

qf = CONVERT_FLOAT_V(mad24(q.d4, 32768u, q.d3));
  qf = qf * 32768.0f;
  
  qi = CONVERT_UINT_V(qf*nf);

nn.d0  = mul24(n.d0, qi);
  nn.d1  = mad24(n.d1, qi, n.d0 >> 15);
  nn.d2  = mad24(n.d2, qi, n.d1 >> 15);
  nn.d3  = mad24(n.d3, qi, n.d2 >> 15);
  nn.d4  = mad24(n.d4, qi, n.d3 >> 15);
  nn.d0 &= 0x7FFF;
  nn.d1 &= 0x7FFF;
  nn.d2 &= 0x7FFF;
  nn.d3 &= 0x7FFF;

res->d0 = q.d0 - nn.d0;
  res->d1 = q.d1 - nn.d1 + AS_UINT_V(res->d0 > 0x7FFF);
  res->d2 = q.d2 - nn.d2 + AS_UINT_V(res->d1 > 0x7FFF);
  res->d3 = q.d3 - nn.d3 + AS_UINT_V(res->d2 > 0x7FFF);
  res->d4 = q.d4 - nn.d4 + AS_UINT_V(res->d3 > 0x7FFF);
  res->d0 &= 0x7FFF;
  res->d1 &= 0x7FFF;
  res->d2 &= 0x7FFF;
  res->d3 &= 0x7FFF;

}