take a synthetic CPU core design and place it at the core of the FPGA design, take your software code written in a form as understood by the FPGA device you want to run there ,run it through a translation and convert it to a form that maps it to the elements of the FPGA simulator to verify that it works , output the binary (at 100 Kbytes on average) and load it into the real FPGA, and run it.
http://www.fpga4fun.com/FPGAsoftware4.html

hence instead of running your slow software code on the CPU, you are running that code now converted to pure hardware on the FPGA so it becomes a LOT faster.

their great things, but the embedded people usually use a FPGA to prototype a mass produced end product, then have that FPGA schematic converted to a cheaper ASIC for mass production…. and that’s as far as they usually think about it, and so FPGA tools and generic FPGA use is very crude and expensive development and antiquated business model , but the actual FPGA can be cheap, but not as cheap as a mass produced ASIC today.

there’s the thing though if they just lowered the prices and sold far more FPGA of various types then made their money on bulk and selling their IP then have all the PCB vendors use them, you could have things like so called mass ‘Reconfigurable computing’ loading an FPGA binary as required to perform a given task at a given time.

for instance you want to encode a video, load up your FPGA on the fly with that H.264 Encoder IP block you just downloaded the a FPGA online store, and send your HD cams native video directly to the input of that FPGA and have that save out the H.264.mp4, want to play it back after your done ,load up the H.264 decoder and related IP block into the same FPGA and have it do that for you…

the possibility are endless as long as someone has made the FPGA block you require and made it available to you for free or payware etc….