Writing efficient VBA UDFs (Part5) - UDF Array Formulas go faster!

Just in case you thought the previous posts on writing efficient VBA UDfs (Part1, Part2, Part3, Part 4) meant we had finished making UDFs run faster, think again – its time to explore UDF Array Formulas.

Single and Multi-Cell Array Formulas

Excel array formulas can do amazing things. They are like ordinary formulas except that you enter them with Control/Shift/Enter rather than just enter.

There are two kinds of array formulae:

• Single cell array formulae are entered into a single cell, loop through their arguments (which are often calculated arguments) and return a single answer.
• Multi-cell array formulae are entered into multiple cells and return an answer to each of the cells.

With this power comes a cost: because array formulae are doing a lot of work they can be slow to calculate (particularly single-cell array formulas).

UDF Multi-cell Array Formulas go Faster!

You can break down the time taken by a VBA UDF into these components:

• Overhead time to call the UDF.
• Time to fetch the data thats going to be used by the UDF.
• Time to do the calculations.
• Overhead time to return the answer(s).

In the post on Excel VBA Read/Write timeings you could see that there was quite a significant overhead on each VBA read and write call, so that its usually much faster to read and write large blocks of data at a time.

So it sounds like a good idea to make your VBA UDF read as much data as possible in a single block and return data to Excel in as large a block as possible.
Enter the Multi-cell array formula – it does exactly that – and also minimises the calling overhead – and often it can read the data once and re-use it lots of times.

So how do you make a Multi-Cell Array formula?

Lets create an array version of the AverageTolE function shown in the first Writing Efficient VBA UDFs post.
The scenario is that you want to find the Averages of the data excluding a number of different tolerances rather just one tolerance.
To keep things simple I am assuming that

• the tolerances are all in one row
• both the data and the tolerances will be supplied as ranges
• error-handling is largely ignored
• the function returns a row of answers that correspond to the row of tolerances.

Public Function AverageTolM(theRange As Range, theTols As Range) As Variant
Dim vArr As Variant
Dim vArrTols As Variant
Dim v As Variant
Dim d As Double
Dim r As Double
Dim k As Long
Dim vOut() As Variant
Dim dTol As Double
Dim lCount As Long
On Error GoTo FuncFail
vArr = theRange.Value2
vArrTols = theTols.Value2
ReDim vOut(1 To 1, 1 To UBound(vArrTols, 2))
On Error GoTo skip
For k = 1 To UBound(vArrTols, 2)
dTol = CDbl(vArrTols(1, k))
r = 0#
lCount = 0
For Each v In vArr
d = CDbl(v)
If Abs(d) > dTol Then
r = r + d
lCount = lCount + 1
End If
skip:
Next v
vOut(1, k) = r / lCount
Next k
AverageTolM = vOut
Exit Function
FuncFail:
AverageTolM = CVErr(xlErrNA)
End Function

The changes to the UDF are quite simple:

• theTols range is coerced into a variant array: vArrTols = theTols.Value2
• an output array of the same size is created: ReDim vOut(1 To 1, 1 To UBound(vArrTols, 2))
• The UDF loops on the tolerance array and populates the output array
• The output array is assigned to the function variable: AverageTolM = vOut

Note that the Function is declared as returning a variant (which will contain an array) rather than being declared as returning an array of variants.

Assuming that the data is in H27:AA27 then enter the array function with Ctrl/Shift/Enter as

{=AVERAGETOLM(Data!$A$1:$A$32000,$H$27:$AA$27)} into 20 rows (so we will get 20 x 20 = 400 cells of answers).

Calculating this 20 formulas takes 975 milliseconds.

Using the original AVERAGETOLE formula for the 400 cells takes 1660 milliseconds, an improvement factor of 1.7

Summary

• In many real-life cases using multi-cell array UDFs can be the fastest way to calculate.
• Converting a conventional UDF to a multi-cell array UDf is straightforward.

About these ads