| Commit message (Collapse) | Author | Age |
... | |
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
was causing spurious illegal instruction traps
|
| |
|
|
|
|
| |
instructions
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
in rv32 these will be replaced with loads and stores.
|
| |
|
| |
|
|
|
|
|
|
|
| |
- Added 5th rounding mode
- Removed MFCR/MTCR in favor of MFFSR/MTFSR (it was the only CR...)
- merged MTF.D with MTFLH.D; operation depends on RV32/RV64 mode
- made MFFL.D and MFFH.D illegal in RV64
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
now generic variants behave differently in RV32 and RV64.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
| |
Now, you can either use the RM in the FSR or specify it in the insn.
(Except for FP->int; no dynamic for that.)
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
this is symmetric with fp stores, so we only need one decoding pipe
|
|
|
|
| |
...to be used instead of mff.s when doing int -> DP FP moves on a 32-bit cpu
|
| |
|
|
|
|
| |
This will simplify control logic (since every branch has a logical inverse)
|
| |
|