| Age | Commit message (Collapse) | Author | |
|---|---|---|---|
| 21 hours | fix example compilationalloca | Kimplul | |
| 21 hours | resolve relocations asap | Kimplul | |
| + Should help with compile times on platforms that use literal pools, might also improve code quality | |||
| 42 hours | armhf seems to work | Kimplul | |
| 42 hours | fix regalloc | Kimplul | |
| + Had some small issues with ending lifetimes a bit too late | |||
| 42 hours | improve build system a bit | Kimplul | |
| 42 hours | fix some warnings with clang | Kimplul | |
| 42 hours | fix big endian compilation | Kimplul | |
| + Code used some assumptions about type aliasing that might not hold for all systems | |||
| 3 days | powerpc64le seems to work | Kimplul | |
| 3 days | mips64el seems to work | Kimplul | |
| 3 days | expose sqrt | Kimplul | |
| + Requires linking with libm in some cases, which is fine I suppose, but kind of annoying | |||
| 3 days | mipsel seems to work | Kimplul | |
| 3 days | aarch64 linux seems to work | Kimplul | |
| 3 days | somewhat improved register allocator | Kimplul | |
| 9 days | remember to clean libejit.a | Kimplul | |
| 9 days | handle immediates a bit better | Kimplul | |
| + Passing floats as immediate values is not supported in lightening, but I might have a go at adding it since it seems like a useful feature at some point | |||
| 2025-03-15 | experiment with allocating regs on stack in interp | Kimplul | |
| + Avoids having to lug around an execution context, arguably simplified things but now there's no real way to detect when we run out memory for regs. | |||
| 2025-03-14 | protect jit pages | Kimplul | |
| 2025-03-14 | start supporting 32bit arches | Kimplul | |
| 2025-03-13 | improve register allocation | Kimplul | |
| + Still linear, but orders regs by some kind of priority + Use all registers available, not just callee-save | |||
| 2025-03-07 | make code a bit more robust | Kimplul | |
| + Should be more difficult to make mistakes in the future, ejit can now automatically keep track of how many register slots are used and if 64 bit mode is required. Slight runtime overhead, but not too bad. | |||
| 2025-03-05 | bump lightening | Kimplul | |
| 2025-03-05 | allow direct calls between compiled functions | Kimplul | |
| 2025-01-05 | fix some assertions | Kimplul | |
| 2025-01-04 | add LTO compilation flag | Kimplul | |
| 2024-12-12 | add todo about minimizing mmap syscalls | Kimplul | |
| 2024-10-22 | move interp into run_interp | Kimplul | |
| + This allows us to skip a potential extra function call | |||
| 2024-10-22 | use type-specific vectors instead of generic ones | Kimplul | |
| + An attempt at speeding up function calls in the interpreted mode | |||
| 2024-08-08 | remove extra file | Kimplul | |
| 2024-08-07 | add readme skeleton | Kimplul | |
| 2024-07-15 | remember to clean up properly | Kimplul | |
| 2024-07-14 | add relative args | Kimplul | |
| 2024-07-14 | jit tests pass (kind of) on x86_64 | Kimplul | |
| 2024-07-14 | add float + double instead of just double | Kimplul | |
| 2024-07-13 | implement some more jit instructions | Kimplul | |
| 2024-07-13 | make tests only compile once | Kimplul | |
| + Select between jit/bytecode with a runtime parameter | |||
| 2024-07-13 | implement some float JIT handling | Kimplul | |
| 2024-07-13 | bytecode tests pass | Kimplul | |
| 2024-06-30 | work through loads and stores | Kimplul | |
| 2024-06-30 | continue working through test cases | Kimplul | |
| + Remove overflow and more complicated floating point tests for now | |||
| 2024-06-29 | remove some tests that won't get implemented | Kimplul | |
| 2024-06-29 | work through branching instructions | Kimplul | |
| 2024-06-29 | continue working through bytecode ops | Kimplul | |
| 2024-06-29 | add flag for using 64 bit values on 32 machines | Kimplul | |
| + JIT backends needs to still trigger an assertion if we detect that the user lied | |||
| 2024-06-29 | clear up some semantics | Kimplul | |
| 2024-06-29 | start adding tests | Kimplul | |
| 2024-06-28 | optimize interpreter a bit | Kimplul | |
| + Use a gpr/fpr/arg stack shared between different functions to minimize amount of memory allocations done at runtime while still allowing interleaving bytecode/machine code functions during call, just with a bit of extra setup in between | |||
| 2024-06-28 | implement enough for fibonacci | Kimplul | |
| 2024-06-28 | some posthaste fixes | Kimplul | |
| + Interesting that the comparison thing only failed when compiled with RELEASE=1 | |||
| 2024-06-28 | implement more jit instructions | Kimplul | |
| 2024-06-26 | allow toggling between jit and bytecode | Kimplul | |
 |
index : ejit | |
| Hybrid JIT/bytecode built on top of Lightening | kimi.h.kuparinen+ejit@gmail.com |
| aboutsummaryrefslogtreecommitdiff |