I asked the question of being able to offload compilation to the GPU since of course the type of data GPU's are good at doesn't fit today's GPU's almost at all, but the point is Windows NT was forward looking, and considering GPU's are becoming more and more general purpose with even being able to access shared memory; using forward thinking there may be a way to have the msbuild build system pre-process the code in a clever way so we can compile on our thousands of GPU cores. Right now they are tailored towards similar numeric data sure, but maybe with a tweak here or there we could finally use all or even a fration of that compute power to compile on GPU's instead of just doing all the compilation process on our CPU's. Id be a very happy and very impressed camper, since my GPU is being relatively idle while compiling anyway, and not like I have a huge compilation farm. But I rarely play a game AND compile at the same time since hey are both very resource hungry tasks so they are usually seperate tasks for me. So may as well try to offload some computation to the GPU during compilation process no? We need Dave Cutler joint with AMD/NVIDIA on the problem
Great panel Charles! Lets have a Going Native show, or perhaps an intra show segment that runs for a few episodes of GN that covers the work (and motivation) to add a AST to the C1XX front end compiler for C++. It was fascinating to hear Herb mention at BUILD that this was a relatively recent addition to the compilers implementation. Adding an AST at this stage in the game sounds like a complex and very likely insightful engineer journey.