FWIW, I started a home USB audio project a couple/three years ago, and all of the base code provided by TI was built using the Keil tools. After futzing with sdcc, which back then completely broken and it appeared as if development had been abandoned, I bit the bullet and bought a copy of Keil uVision 2 and the C compiler and assembler (no simulator etc). It was pricey, in fact more expensive than the PCB layout tool I also had to buy.

When I sold the design, the tools were easily paid for.

And I still have them. I haven't seen a reason to upgrade to the 8.xx version and uVision 3, as my current version happily works with all of the SiLabs chips I've used in the last buncha years. In this case, the investment (of my own personal $$$) has worked out very well in my favor. (About the only other software package I use that gets more regular work is Quicken, which I do upgrade every three or so years, and it's usually worth the $50.)

Now, maybe sdcc has matured to the point where it's as good as what Keil can do. A bonus for sdcc is that it runs on Macs (I run Keil in Parallels). But I haven't seen any reason to abandon my paid-for tools. Obviously if I started using another micro architecture (like ARM) I'd need new tools, and I strongly suspect I'd just choose the Keil stuff based on my experience with C51.

Having said that, my current micro designs are all Microblaze embedded in Xilinx chips, which I've gotta say is pretty frickin' cool, even though the Xilinx tools are typically diabolical.

(And because you'll ask: it turns out that for what we do, it ends up being cheaper to embed the Microblaze into the FPGA, now Spartan 3AN, instead of using an FPGA with an SiLabs part.)

-a