The weight variable in DHS surveys is normalized, which means that mean weight is 1, or, equivalently, the weighted total is equal to the unweighted total. HOWEVER, in order to get rid of a decimal point, the calculated weight is multiplied by one million and then rounded to the nearest integer.

When you use iweights you need to divide by one million. When you use pweights, as in a statistical model, you CAN divide by one million but you don't need to, because Stata automatically normalizes with pweights.

If you want inflation weights, which will give estimates of the population counts, you need to define new weights, and to do that you have to provide the population total. Say that the total number of households or persons in households or women 15-49 (depending on whether you are using the HR or PR or IR file) is N. You then multiply the weight by N/n, where n is the sample size. You also divide by one million and round to the nearest integer and then use fweight. Note that you cannot use standard errors, confidence intervals, or tests with fweight, only with pweight.

DHS advises against inflation weights. It can be difficult to find what is the population total on the date of the survey (the median date of interview, say). The sample includes only the household population and census estimates may be larger than the household population. It's unusual to see any use of inflation weights, but if you really want them, that's how you calculate them.