diff options
Diffstat (limited to 'algorithm.tex')
| -rw-r--r-- | algorithm.tex | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/algorithm.tex b/algorithm.tex index 90e7644..727e067 100644 --- a/algorithm.tex +++ b/algorithm.tex @@ -80,7 +80,7 @@ RTL is the existing intermediate language in CompCert that was chosen to transfo %% Describe RTL RTL function definitions are a sequence of instructions encoded in a control-flow graph, with each instruction linking to the next instruction that should be executed. RTL consists of ten separate instructions in total, which makes this a good candidate for high-level synthesis translation, as these instructions translate quite directly to a hardware equivalent. This structure makes it ideal for performing optimisations on, as performing static analysis and transformations on this language is much simpler than doing the same optimisation for the C abstract syntax tree instead. However, even though RTL is quite architecture independent and good for software optimisations, it does not model hardware well and is therefore not suitable for hardware optimisations. We therefore need a new intermediate language that translates directly into hardware and which can also be used as an intermediate language for more hardware specific optimisations. -However, before this translation can be performed, all functions need to be inlined. This means that recursive function calls cannot be supported, however, these are difficult to translate to hardware as they require dynamic allocation of the stack. Dynamic allocation cannot be performed in hardware, because the memory size has to be set statically and then translated to a physical memory in the hardware. Inlining is already an optimisation that is performed at the RTL stage in CompCert, however, to get RTL that is ready to be translated to hardware, this pass was modified so that is forced all functions to be inlined. This eliminates all the function calls from the RTL, as long as there are no recursive function calls, which means that do not have to be handled by the translation to hardware. +However, before this translation can be performed, all functions need to be inlined. This means that recursive function calls cannot be supported, however, these are difficult to translate to hardware as they require dynamic allocation of the stack. Dynamic allocation cannot be performed in hardware, because the memory size has to be set statically and then translated to a physical memory in the hardware. \JP{I think we should explain this a little bit more; HW allocation is possible in principle.} Inlining is already an optimisation that is performed at the RTL stage in CompCert, however, to get RTL that is ready to be translated to hardware, this pass was modified so that is forced all functions to be inlined. This eliminates all the function calls from the RTL, as long as there are no recursive function calls, which means that do not have to be handled by the translation to hardware. \subsection{HTL} |