Daniel Thornburgh
Daniel Thornburgh
For simple cases, it can be faster and smaller to directly emit a loop than to emit a call to a mem function. We should identify these cases and emit...
The current whole-program zero page allocator allocates only for size, not for speed. However, a (difficult to reproduce) real project by juj showed that the difference between zero page allocations...
We need a reasonably complete, standards compliant C++ standard library with a license that allows statically linking against it without virally requiring inclusion of a license disclaimer. There are really...
We rely heavily on MachineScheduler to order instructions to reduce register pressure, but it currently marks calls of any function as scheduling boundaries. That is, any operations that LLVM IR...
The optimization tracked in the reproducer test `inc-dec-phi.ll` has regressed, as the assembly output no longer contains the expected increment and decrement pattern. The other more specific test cases remain,...
This would allow folks to play with the semantics of a GAS-style assembler for the 6502, as well as providing a nice vehicle to demonstrate any assembler bugs.
The GNU assembler supports dependency tracking via the `-MD` command line option; we really want this for clang as well, since assembly files and `.incbin` in particular are very common...
Now that there are a number of appreciable code generation differences between the 6502 and 65c02, and since we have simulator support for the 65c02, we should add the -Os...
Our current tail call optimization is a bit of a hack compared to the one in other LLVM targets; it just checks if a JSR is immediately followed by an...
Whenever a symbol gets its value from an arithmetic expression involving symbols marked as `.zeropage`, it is sometimes possible to know statically that the result is also in the zero...