aboutsummaryrefslogtreecommitdiff
path: root/src/Specific/IntegrationTestLadderstep130.v
Commit message (Collapse)AuthorAge
* Build curve-specific files from jsonGravatar Jason Gross2017-10-18
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The X25519 curves are now generated from `.json` files. This code only works in >= 8.7, because it makes use of the recently-merged-from-fiat `transparent_abstract` tactic to allow defining things in tactics without massive slowdown. The structure is as follows: 0. The module types and tactic definitions that set up the infrastructure live in `src/Specific/Framework/` 1. There are `.json` files in `src/Specific/CurveParameters/` that specify curve characteristics. A simple example is `x2555_130.json`, which is: ```json { "modulus" : "2^255-5", "base" : "130", "a24" : "121665 (* XXX TODO(andreser) FIXME? Is this right for this curve? *)", "sz" : "3", "bitwidth" : "128", "carry_chain1" : "default", "carry_chain2" : ["0", "1"], "coef_div_modulus" : "2", "operations" : ["ladderstep"] } ``` A more complicated example is `x25519_c64.json`: ```json { "modulus" : "2^255-19", "base" : "51", "a24" : "121665", "sz" : "5", "bitwidth" : "64", "carry_chain1" : "default", "carry_chain2" : ["0", "1"], "coef_div_modulus" : "2", "operations" : ["femul", "fesquare", "freeze", "ladderstep"], "extra_files" : ["X25519_C64/scalarmult.c"], "compiler" : "gcc -march=native -mtune=native -std=gnu11 -O3 -flto -fomit-frame-pointer -fwrapv -Wno-attributes", "mul_header" : "(* Micro-optimized form from curve25519-donna-c64 by Adam Langley (Google) and Daniel Bernstein. See <https://github.com/agl/curve25519-donna/blob/master/LICENSE.md>. *)", "mul_code" : " uint128_t t[5]; limb r0,r1,r2,r3,r4,s0,s1,s2,s3,s4,c; r0 = in[0]; r1 = in[1]; r2 = in[2]; r3 = in[3]; r4 = in[4]; s0 = in2[0]; s1 = in2[1]; s2 = in2[2]; s3 = in2[3]; s4 = in2[4]; t[0] = ((uint128_t) r0) * s0; t[1] = ((uint128_t) r0) * s1 + ((uint128_t) r1) * s0; t[2] = ((uint128_t) r0) * s2 + ((uint128_t) r2) * s0 + ((uint128_t) r1) * s1; t[3] = ((uint128_t) r0) * s3 + ((uint128_t) r3) * s0 + ((uint128_t) r1) * s2 + ((uint128_t) r2) * s1; t[4] = ((uint128_t) r0) * s4 + ((uint128_t) r4) * s0 + ((uint128_t) r3) * s1 + ((uint128_t) r1) * s3 + ((uint128_t) r2) * s2; r4 *= 19; r1 *= 19; r2 *= 19; r3 *= 19; t[0] += ((uint128_t) r4) * s1 + ((uint128_t) r1) * s4 + ((uint128_t) r2) * s3 + ((uint128_t) r3) * s2; t[1] += ((uint128_t) r4) * s2 + ((uint128_t) r2) * s4 + ((uint128_t) r3) * s3; t[2] += ((uint128_t) r4) * s3 + ((uint128_t) r3) * s4; t[3] += ((uint128_t) r4) * s4; ", "square_header" : "(* Micro-optimized form from curve25519-donna-c64 by Adam Langley (Google) and Daniel Bernstein. See <https://github.com/agl/curve25519-donna/blob/master/LICENSE.md>. *)", "square_code" : " uint128_t t[5]; limb r0,r1,r2,r3,r4,c; limb d0,d1,d2,d4,d419; r0 = in[0]; r1 = in[1]; r2 = in[2]; r3 = in[3]; r4 = in[4]; do { d0 = r0 * 2; d1 = r1 * 2; d2 = r2 * 2 * 19; d419 = r4 * 19; d4 = d419 * 2; t[0] = ((uint128_t) r0) * r0 + ((uint128_t) d4) * r1 + (((uint128_t) d2) * (r3 )); t[1] = ((uint128_t) d0) * r1 + ((uint128_t) d4) * r2 + (((uint128_t) r3) * (r3 * 19)); t[2] = ((uint128_t) d0) * r2 + ((uint128_t) r1) * r1 + (((uint128_t) d4) * (r3 )); t[3] = ((uint128_t) d0) * r3 + ((uint128_t) d1) * r2 + (((uint128_t) r4) * (d419 )); t[4] = ((uint128_t) d0) * r4 + ((uint128_t) d1) * r3 + (((uint128_t) r2) * (r2 )); " } ``` 3. The `src/Specific/CurveParameters/remake_curves.sh` script holds a list of curves to be made, what directories they should end up living in, and it invokes `src/Specific/Framework/make_curve.py` to transform these files into outputs. The Python script fills in a few defaults (such as computing `s` and `c` from the modulus, if you don't pass them explicitly), and does a lot of processing on the C code that is pasted verbatim from donna to get it to be in the right format for Coq. This Python script creates the files: - `CurveParameters.v` (the Coq-ified version of the json file, which instantiates an appropriate module type); - `Synthesis.v`, which instantiates a `MakeSynthesisTactics` with the curve parameter modules, invokes a tactic from the applied module functor to synthesize all of the relevant non-reflective bits (basically, what used to live in @jadephilipoom 's `ArithmeticSynthesisTest.v`), and then instantiates another module functor `PackageSynthesis` which defines notations via tactics in terms to access the names of the various fields defined by the synthesis tactic; - any other files you ask it for, such as `compiler.sh`, `femul.v`, `femulDisplay.v`. All of the `*Display.v` files are simple, and all the the operation synthesis files have a single `Definition` (with the appropriate type), and solve the definition by invoking a single tactic defined in `PackageSynthesis`, e.g., `synthesize_mul` or `synthesize_ladderstep`.
* Reorgainze synthesis framework files into a Framework folderGravatar Jason Gross2017-10-18
|
* Fix more commented out alternativesGravatar Jason Gross2017-09-05
|
* Factor out some of the preglue synthesis codeGravatar Jason Gross2017-07-08
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This makes it a bit more uniform, and hopefully more automatable and packageable. Unfortunately, there's still no spec for this part of the pipeline, so the tactics simply aggregate common patterns. Alas, this also makes things a bit slower; I suspect that [Defined] is the place where things are slower. After | File Name | Before || Change --------------------------------------------------------------------------------------- 13m51.14s | Total | 12m59.29s || +0m51.84s --------------------------------------------------------------------------------------- 1m54.18s | Specific/IntegrationTestKaratsubaMul | 1m43.12s || +0m11.06s 1m38.97s | Specific/IntegrationTestLadderstep130 | 1m30.26s || +0m08.70s 2m19.75s | Specific/NISTP256/AMD64/femul | 2m14.08s || +0m05.66s 0m39.90s | Specific/IntegrationTestMontgomeryP256_128 | 0m34.21s || +0m05.68s 0m21.95s | Specific/NISTP256/AMD64/fesub | 0m19.23s || +0m02.71s 0m21.37s | Specific/NISTP256/AMD64/feadd | 0m18.82s || +0m02.55s 0m21.02s | Specific/X25519/C64/femul | 0m18.32s || +0m02.69s 0m20.53s | Specific/IntegrationTestFreeze | 0m23.26s || -0m02.73s 0m18.28s | Specific/IntegrationTestMontgomeryP256_128_Sub | 0m15.32s || +0m02.96s 0m18.20s | Specific/IntegrationTestMontgomeryP256_128_Add | 0m15.52s || +0m02.67s 0m16.35s | Specific/IntegrationTestMontgomeryP256_128_Opp | 0m13.52s || +0m02.83s 0m13.92s | Specific/IntegrationTestMontgomeryP256_128_Nonzero | 0m11.84s || +0m02.08s 0m18.23s | Specific/NISTP256/AMD64/MontgomeryP256 | 0m16.62s || +0m01.60s 0m15.54s | Specific/IntegrationTestSub | 0m14.53s || +0m01.00s 0m14.78s | Specific/X25519/C64/fesquare | 0m13.13s || +0m01.64s 0m13.71s | Specific/NISTP256/AMD64/fenz | 0m12.69s || +0m01.02s 3m14.34s | Specific/X25519/C64/ladderstep | 3m14.32s || +0m00.02s 0m16.54s | Specific/NISTP256/AMD64/feopp | 0m16.48s || +0m00.05s 0m12.21s | Specific/MontgomeryP256_128 | 0m12.70s || -0m00.48s 0m00.73s | Specific/IntegrationTestTemporaryMiscCommon | 0m00.72s || +0m00.01s 0m00.64s | Specific/IntegrationTestDisplayCommon | 0m00.60s || +0m00.04s
* Don't rely on autogenerated namesGravatar Jason Gross2017-06-05
| | | | | | This fixes all of the private-names warnings emitted by compiling fiat-crypto with https://github.com/coq/coq/pull/268 (minus the ones in coqprime, which I didn't touch).
* Inline a24_sig in ladderstepGravatar Jason Gross2017-04-17
| | | | | Update the pipeline to be powerful enough to take advantage of this change by doing more simplification.
* Update display of ladderstep130Gravatar Jason Gross2017-04-14
|
* Use 128/256 in ladderstep 130Gravatar Jason Gross2017-04-14
|
* Add 130-bit 3-register synthesisGravatar Jason Gross2017-04-07
This one has fewer operations, and so will be faster to play with