| Commit message (Collapse) | Author | Age |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Some help from jadep on BitwiseOr.
After | File Name | Before || Change
--------------------------------------------------------------------------------------
0m39.26s | Total | 0m36.03s || +0m03.23s
--------------------------------------------------------------------------------------
N/A | BoundedArithmetic/DoubleBoundedProofs | 0m21.20s || -0m21.19s
0m07.47s | BoundedArithmetic/Double/Proofs/Multiply | N/A || +0m07.46s
0m06.70s | BoundedArithmetic/Double/Proofs/SpreadLeftImmediate | N/A || +0m06.70s
0m05.14s | BoundedArithmetic/Double/Proofs/RippleCarryAddSub | N/A || +0m05.13s
0m02.75s | BoundedArithmetic/Double/Proofs/Decode | N/A || +0m02.75s
0m08.06s | BoundedArithmetic/ArchitectureToZLikeProofs | 0m08.05s || +0m00.00s
0m02.01s | Specific/FancyMachine256/Barrett | 0m02.13s || -0m00.12s
0m02.00s | Specific/FancyMachine256/Montgomery | 0m02.06s || -0m00.06s
0m01.75s | Specific/FancyMachine256/Core | 0m01.66s || +0m00.09s
0m00.90s | BoundedArithmetic/Double/Proofs/LoadImmediate | N/A || +0m00.90s
0m00.88s | BoundedArithmetic/Double/Proofs/BitwiseOr | N/A || +0m00.88s
0m00.58s | BoundedArithmetic/Double/Proofs/SelectConditional | N/A || +0m00.57s
0m00.53s | BoundedArithmetic/ArchitectureToZLike | 0m00.47s || +0m00.06s
0m00.49s | BoundedArithmetic/Double/Core | N/A || +0m00.49s
N/A | BoundedArithmetic/DoubleBounded | 0m00.46s || -0m00.46s
|
| |
|
|
|
|
|
| |
By being careful about building the expressions in the first place, we
no longer need it, and can rely on dead code elimination.
|
| |
|
|
|
|
|
|
| |
We still use CSE in fancy machine, because we want to lift the ldi's
above the rest of the code. However, on a quick inspection, the
algorithm no longer needs CSE to be duplicate-free.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
Now the rewrite strategy no longer relies on projections of anything
other than [decode], and the conversion to ZLike is simpler.
Modulo some annoyingly delicate arithmetic around things like [2^n * 2^n
= 2^(2*n)] and whether to factor [(decode (fst x) + decode (snd x) >> n)
>> b] as [decode x >> n] or as [shrd (fst x) (snd x) n], the proofs
bascially go by pulling/pushing decodes.
|
| |
|
|
|