MATLAB: How to avoid a redundant for-loop in generated code that computes the virtually unused quantity ‘trueCount’

Question

For the following lines:

stopEdgeIdx = stopEdgeIdx(stopEdgeIdx <= globalMaxIndx);
if ~isempty(stopEdgeIdx)
<< do something >>
end

The generated C code contains two loops, which seems redundant:

k = trueCount - 1;
trueCount = 0;
for (b_i = 0; b_i <= k; b_i++) {
 if (stopEdgeIdx_data[b_i] <= globalMaxIndx) {
  trueCount++;
 }
}
partialTrueCount = 0;
for (b_i = 0; b_i <= k; b_i++) {
 if (stopEdgeIdx_data[b_i] <= globalMaxIndx) {
  stopEdgeIdx_data[partialTrueCount] = stopEdgeIdx_data[b_i];
  partialTrueCount++;
 }
}

Is there a way to have only one loop instead of two?

Answer 1

Development is aware of this issue and will consider fixing it in a future release of MATLAB. In the meantime, you could generate a dynamic array to eliminate one of the for-loops, but there is a memory tradeoff. Here is a contrived example:

function out = Example1(stopEdgeIdx, globalMaxIndx)
  stopEdgeIdxTemp = [];
  for elem = stopEdgeIdx
    if(elem <= globalMaxIndx)
      stopEdgeIdxTemp = [stopEdgeIdxTemp elem];
    end
  end
    if ~isempty(stopEdgeIdxTemp)
      % <do something> region

      out = sum(stopEdgeIdxTemp); 
    else
      out = 0;
    end
  end

The generated code has a single for-loop and stopEdgeIdxTemp is a dynamically sized array:

...
for (vlen = 0; vlen < 6; vlen++) {
 if (stopEdgeIdx[vlen] <= globalMaxIndx) {
  k = stopEdgeIdxTemp->size[1];
  i = stopEdgeIdxTemp->size[0] * stopEdgeIdxTemp->size[1];
  stopEdgeIdxTemp->size[1]++;
  emxEnsureCapacity_real_T(stopEdgeIdxTemp, i);
  stopEdgeIdxTemp->data[k] = stopEdgeIdx[vlen];
 }
}
...

The call to 'emxEnsureCapacity_real_T' during each loop iteration may be expensive, however. To avoid this, you can try pre-allocating with coder.nullcopy:

function out = Example2(stopEdgeIdx, globalMaxIndx)  
  StopEdgeIdxTmp = coder.nullcopy(zeros(1, length(stopEdgeIdx)));
  k = int32(1);
  for elem = stopEdgeIdx
    if elem <= globalMaxIndx
      StopEdgeIdxTmp(k) = elem;
      k = k + 1;
    end
  end
  StopEdgeIdxTmp = StopEdgeIdxTmp(1:k);
    if ~isempty(StopEdgeIdxTmp)
      % <do something> region
      out = sum(StopEdgeIdxTmp); 
    else
      out = 0;
    end
  end

The generated code will depend on whether the size of stopEdgeIdx is known at compile time. For the example above, then the generated C code has only 1 for-loop (plus the '<do something>' region) with no dynamic arrays:

...
double StopEdgeIdxTmp[6];
k = 1;
for (elem = 0; elem < 6; elem++) {
 if (stopEdgeIdx[elem] <= globalMaxIndx) {
  StopEdgeIdxTmp[k - 1] = stopEdgeIdx[elem];
  k++;
 }
}
/* <do something> region */
elem = (signed char)k;
out = StopEdgeIdxTmp[0];
for (k = 2; k <= elem; k++) {
 out += StopEdgeIdxTmp[k - 1];
}
...