Alessandro's answer is correct, but perhaps a little cryptic if you are not familar with the Tex landscape.
There are two main ways of creating output from Tex &c documents, call them paths: the Web2c path, which outputs dvi, and the Pdftex path, which outputs pdf. The Web2c path includes graphics in dvi files using Postscript specials, and cannot embed pdf graphics. Since you are being asked for a dvi file, that means you have to convert your pdf graphics to eps.
The other mechanism, the Pdftex path, is to embed graphics as pdf. Postscript can't be directly embedded in this way, because pdf is essentially a computationally crippled form of Postscript (with bells and whistles), and so the programming structures of Postscript can't be translated into pdf without running the Postscript program, which pdftex doesn't support. But the pdf and svg formats can be embedded, together with many bitmapped formats such as jpg.
Three futher points: firstly, the \includegraphics
code is different for the two paths: it looks for different file extensions depending on whether we are, and acts differently upon them. Second, the pdftex
program can create either pdfs or dvis: it looks to see how it is invoked, as either (for Latex) pdflatex
or latex
. Third, there is another, third, path, for Xetex, which targets a slightly different format than dvi, its xdv format, which handles fonts and character sets differently, but is otherwise like dvi.
So your problem has nothing to do with dvi's bounding boxes, but to answer that anyway: the dvi format does not have a concept of bounding box, and indeed dvi files don't have enough information to calculate one, since they show where to place characters from fonts, but don't include their font metrics: you have to look at the font metric files for that information. Additionally, with Postscript specials, one has to run a Postscript engine to find the extent of Postscript graphics.