Memory issues in generated for loops

The simple Matlab code:

function y = fcn(u)
y = u;
for i = single(0:0.15:1000-1)
    y = y + (i+1)^u;
end

generates the following C:

loop.c:

 int32_T i;
  static const real32_T b[6661] = { 0.0F, 0.15F, 0.3F, 0.45F, 0.6F, 0.75F, 0.9F,
    1.05F, 1.2F, 1.35F, 1.5F, 1.65F, 1.8F, 1.95F, 2.1F, 2.25F, 2.4F, 2.55F, 2.7F,
    2.85F, 3.0F, 3.15F, 3.3F, 3.45F, 3.6F, 3.75F, 3.9F, 4.05F, 4.2F, 4.35F, 4.5F,
    4.65F, 4.8F, 4.95F, 5.1F, 5.25F, 5.4F, 5.55F, 5.7F, 5.85F, 6.0F, 6.15F, 6.3F,
    6.45F, 6.6F, 6.75F, 6.9F, 7.05F, 7.2F, 7.35F, 7.5F, 7.65F, 7.8F, 7.95F, 8.1F,
    8.25F, 8.4F, 8.55F, 8.7F, 8.85F, 9.0F, 9.15F, 9.3F, 9.45F, 9.6F, 9.75F, 9.9F,
    ...
    };

     for (i = 0; i < 6661; i++) {
    /* '<S1>:1:3' */
    /* '<S1>:1:4' */
    rtb_y += (real32_T)pow(b[i] + 1.0F, for_loop_U->In1);

    /* '<S1>:1:3' */
  }

The loop consumes a lot of memory due to the high number of iterations, but I can't find a simple way to force the generator not to use arrays (optimizing for ROM efficiency is already turned on). I should be able to reduce the memory footprint at the price of re-calculating i*0.15 with floating point hardware.

Another similar example:

function y = fcn(u)
y = u;
for i = single(0:0.15:1000000-1)
    y = y - u*(i+1);
end

loop.h:

typedef struct {
  real_T y[6666661];
} DW_for_loop_T;

loop.c:

   for_loop_DW->y[0] = 0.0;
  for_loop_DW->y[6666660] = 999999.0;
  for (k = 0; k < 3333329; k++) {
    kd = (((real_T)k) + 1.0) * 0.15;
    for_loop_DW->y[k + 1] = kd;
    for_loop_DW->y[6666659 - k] = 999999.0 - kd;
  }

for_loop_DW->y[3333330] = 499999.5;

/* '<S1>:1:3' */
for (k = 0; k < 6666661; k++) {
  /* '<S1>:1:3' */
  /* '<S1>:1:4' */
  rtb_y -= (((real32_T)for_loop_DW->y[k]) + 1.0F) * for_loop_U->In1;

    /* '<S1>:1:3' */
  }

This time, despite the fact that y is ultimately a single, the intermediate values (for_loop_DW->y and kd) are doubles, which consumes twice as much memory.

Answer 1

[Jonah Caplan]

The problem seems to be the single().

This works better:

function y = fcn(u)
y = u;
for i = 0:single(0.15):1000000-1
    y = y - u*(i+1);
end

gives

rtb_y = for_loop_U->In1;

/* '<S1>:1:3' */
for (i = 0; i < 6666661; i++) {
  /* '<S1>:1:3' */
  /* '<S1>:1:4' */
  rtb_y -= ((((real32_T)i) * 0.15F) + 1.0F) * for_loop_U->In1;

    /* '<S1>:1:3' */
  }