Commit message (Collapse) | Author | Age | |
---|---|---|---|
* | More WIP on register allocation | Jason Gross | 2017-08-17 |
| | | | | | | | | | | The current allocation is terrible, probably because we are currently requiring that all instructions output to registers. My current guess at a decent thing to do is to make a pass, after register allocation, and eliminate all registers that simply get stored to memory, replacing the relevant instructions with the memory-using versions. Then we can re-register allocate, ignoring values that go straight to memory. | ||
* | Handle most of register allocation | Jason Gross | 2017-08-14 |
| | |||
* | Handle equality in parsing | Jason Gross | 2017-08-14 |
| | |||
* | Fixup header and footer | Jason Gross | 2017-08-14 |
| | |||
* | Update scheduler to know about implicit mulx arg | Jason Gross | 2017-08-14 |
| | | | | | Now it prefers putting together mulx with the same implicit arg (approximated as the same variable with the lower number). | ||
* | Use a more realistic processor model | Jason Gross | 2017-08-13 |
| | |||
* | Revert "Revert "Subset compiler differently"" | Jason Gross | 2017-08-13 |
| | | | | This reverts commit c8c82f2edf416b667ba487dfa2cff0795d37cbe6. | ||
* | Revert "Subset compiler differently" | Jason Gross | 2017-08-13 |
| | | | | This reverts commit 7038444e6515ec51a3b6d1cdfe972664d3f16c81. | ||
* | Subset compiler differently | Jason Gross | 2017-08-13 |
| | |||
* | Add heuristic search | Jason Gross | 2017-08-13 |