github/concrete
1
1
Fork 0
mirror of https://github.com/zama-ai/concrete.git synced 2026-02-09 03:55:04 -05:00
Code Issues Packages Projects Releases Wiki Activity
2,750 Commits 185 Branches 425 Tags
c236dffdd8f7431aeff43fca3d9337ed9f08b56e
Commit Graph

2750 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Pedro Alves
c236dffdd8 chore(backend/cuda): reduces memory consumption in the bootstrap benchmark 2023-04-03 17:49:33 +02:00
Agnes Leroy
d9652b8936 bench(ci): fix concrete-cuda benchmarks 2023-04-03 17:49:33 +02:00
rudy
9cacd4adff fix(compiler): accept signed & tensorized tlu 2023-04-03 13:23:20 +02:00
youben11
e77c87993d fix(ci): use OpenMP library from torch to avoid conflicts 
read: https://discuss.python.org/t/conflicting-binary-extensions-in-different-packages/25332/8
 2023-04-03 08:54:29 +02:00
youben11
4a2f552a08 feat(ci): use new MacOs machine on aws for concrete-python 2023-04-03 08:54:29 +02:00
Umut
87e8f5db02 feat(ci/frontend): add MacOS tests for concrete-python 2023-04-03 08:54:29 +02:00
Umut
e2ed5d1e59 fix(frontend/python): ignore errors during requirements artifact export 2023-04-03 08:54:29 +02:00
Umut
e9607395c4 fix(frontend/python): add bare concrete namespace to setup.py 2023-03-31 14:51:38 +02:00
Umut
8f90074958 feat(compiler): implement FHELinalg.round operation 2023-03-31 10:12:21 +02:00
Beka Barbakadze
adf7adffa3 feat(backend/cuda): add tests for fft in concrete-cuda. fix shared memory mode choosing bug, when compute capability was lower than 8. 2023-03-30 11:12:08 +02:00
rudy
86bec09225 fix(optimizer): dot better support for for broadcasting 
needed for rounded over tensors
 2023-03-30 09:48:58 +02:00
rudy
f596afb607 fix(compiler): more robust high global_p_error optimization 2023-03-30 09:31:08 +02:00
rudy
88ab4d0312 fix(optimizer): woppbs + global_p_error was not possible 2023-03-30 09:31:08 +02:00
David Testé
3e42b8b2fb chore(ci): add concrete-cuda benchmarks using slab 2023-03-29 15:18:01 +02:00
Pedro Alves
dab31433d9 fix(backend): fix ci after 08e8012 
Google benchmark is built twice due to the new bench infrastructure for
concrete-cuda, this commit fixes it by introducing
CONCRETE_CUDA_BUILD_TESTS and CONCRETE_CUDA_BUILD_BENCHMARKS options to skip
unecessary builds.
 2023-03-29 14:36:35 +02:00
Umut
b6f1eb3d2f chore(frontend/python): introduce nightly releases 2023-03-28 14:29:19 +02:00
Umut
4445985aab chore(frontend/python): allow setting python to use in the makefile 2023-03-28 14:29:19 +02:00
Umut
371f2076f7 fix(compiler/bindings): update runtime library lookup to consider concrete-python as well 2023-03-28 14:29:19 +02:00
Pedro Alves
08e8012061 bench(backend): add a benchmark tool for concrete-cuda 2023-03-28 10:19:57 +02:00
aPere3
2728046ae7 chore(compiler): allows unsecure keyset caches to be generated 
For debugging purpose, add a cmake variable that allows to generate
unsecure keycaches, that allows tracing ops to show the message in the
ciphertext body.
 2023-03-28 09:15:04 +02:00
Quentin Bourgerie
8b4dc7be82 chore(concrete-cpu): Use concrete-fft nightly when nightly features is enable to use avx512 2023-03-28 08:50:33 +02:00
Umut
0b9c2a0a3e chore(frontend/python): create nightly release workflow 2023-03-27 16:21:33 +02:00
Umut
5e2349839a chore(frontend/python): pin development dependencies to avoid future CI failures 2023-03-27 16:21:33 +02:00
Umut
18c2a894a5 fix(frontend/python): set extension modules in setup.py 2023-03-27 16:21:33 +02:00
Umut
43a9c499e1 chore(frontend/python): disable S311 lint in tests as it's a false positive 2023-03-27 16:21:33 +02:00
Umut
f0a1122d23 chore(frontend/python): allow filtering tests using markers in make pytest target 2023-03-27 16:21:33 +02:00
Umut
d9313f350a test(frontend/python): use dataflow marker in appropriate tests 2023-03-27 16:21:33 +02:00
Umut
382282736e chore(frontend/python): add dataflow marker to pytest 2023-03-27 16:21:33 +02:00
Umut
ee6da15732 fix(frontend/python): disable dataflow parallelization by default 2023-03-27 16:21:33 +02:00
Umut
805f3702d7 chore(frontend/python): add concurrency control to test linux test workflow 2023-03-27 16:21:33 +02:00
Mayeul@Zama
0b86167310 chore(ci): be more strict on commit messages 2023-03-27 11:39:23 +02:00
Agnes Leroy
df5f1837fc test(compiler): add test for N = 16384 on GPU 2023-03-24 15:29:43 +01:00
Andi Drebes
d1781e0901 refactor(compiler): Remove unused header files from TensorOpsToLinalg.cpp 2023-03-24 11:06:51 +01:00
Andi Drebes
488877ac0b test(compiler): Add tests for batching of TFHE operations 
Re-introduce the previously deleted batching tests for BConcrete as
tests for TFHE with the addition of a new test, checking that
non-batchable operands generated by pure operations are hoisted.
 2023-03-24 11:06:51 +01:00
Andi Drebes
fa5c09a52b feat(compiler): Run batching pass after conversion to TFHE 
With TFHE operations becoming batchable, the batching pass must now be
run after the conversion to TFHE,and TFHE parametrization, but before
any further lowering.
 2023-03-24 11:06:51 +01:00
Andi Drebes
9cd238db82 refactor(compiler): Separate TFHE parametrization into its own pipeline stage 2023-03-24 11:06:51 +01:00
Andi Drebes
fdb4594a2b feat(compiler): Add lowering path for batched TFHE operations to Concrete 2023-03-24 11:06:51 +01:00
Andi Drebes
b495f9dd5c feat(compiler): Make TFHE.keyswitch_glwe and TFHE.bootstrap_glwe batchable 2023-03-24 11:06:51 +01:00
Andi Drebes
b24709a1ec feat(compiler): Batching: Hoist non-batchable operands produced by pure ops 
The batching pass only creates a batched version of a batchable
operation if all of its non-batchable operands are defined out ouf the
outermost loop the iterating over the values of the batchable operand.

This change also allows for operations to be batched if the
non-batachable operands are generated by operations, which are pure
and thus hoistable out of the outermost loop.
 2023-03-24 11:06:51 +01:00
Andi Drebes
3309615d7b fix(compiler): Batching: Bail out if batchable operand is not produced by an op 
An early test for a batchable operation checks whether the batchable
operand is produced by a `tensor.extract` operation and bails out if
this is not the case. However, the use of `llvm::dyn_cast<T>()` directly
on the defining operation of the batchable operand causes an attempt
to cast a null value for an operand which is not produced by an
operation (e.g., block arguments).

Using `llvm::dyn_cast_or_null<T>()` fixes this issue.
 2023-03-24 11:06:51 +01:00
Andi Drebes
a0e5628a88 refactor(compiler): Remove implementation of batching interface from all Concrete operations 2023-03-24 11:06:51 +01:00
Andi Drebes
dafb33414d refactor(compiler): Mark TFHE operations as side-effect free 2023-03-24 11:06:51 +01:00
Quentin Bourgerie
67e2518df9 chore(compiler): Build rust libraries with target-feature 2023-03-23 15:54:30 +01:00
youben11
d891ba2181 ci(compiler): enable minimal tests on PR (macos) 2023-03-22 18:55:26 +01:00
youben11
9e79c4750d test(compiler): disable exception tests on macos temporarily 2023-03-22 18:55:26 +01:00
youben11
568d0e2912 fix(compiler): use value in range of the i5 type 
it's actually weird that the test was passing all this time, and only
failed since we tested on MacOS M1
 2023-03-22 18:55:26 +01:00
youben11
2a81f2e333 chore(compiler): disable SDFG in SDFG tests for macos 2023-03-22 18:55:26 +01:00
youben11
529d96f564 fix(compiler): update mlir-c usage in rust bindings 2023-03-22 18:55:26 +01:00
tmontaigu
2cdf166b96 feat(rust): change build.rs to work on arm64 targets 2023-03-22 18:55:26 +01:00
youben11
afaaf70287 ci(compiler): run tests on M1 2023-03-22 18:55:26 +01:00