|
1 | | -# HERACLES Polynomial Instruction Set Architecture Tools |
2 | | -[](https://github.com/IntelLabs/hec-p-isa-tools/actions/workflows/codeql-analysis.yml) |
3 | | -[](https://scorecard.dev/viewer/?uri=github.com/IntelLabs/hec-p-isa-tools) |
| 1 | +# Encrypted Computing Software Development Kit |
| 2 | +[](https://github.com/IntelLabs/encrypted-computing-sdk/actions/workflows/codeql-analysis.yml) |
| 3 | +[](https://scorecard.dev/viewer/?uri=github.com/IntelLabs/encrypted-computing-sdk) |
4 | 4 | [](https://www.bestpractices.dev/projects/9276) |
5 | 5 |
|
6 | 6 | ## Overview |
7 | | -Intel’s [HERACLES accelerator technology](https://dl.acm.org/doi/10.1145/3560810.3565290) |
8 | | -aims at improving the computational performance of |
9 | | -[Fully Homomorphic Encryption (FHE)](https://en.wikipedia.org/wiki/Homomorphic_encryption). |
10 | | -FHE allows computation to be performed on |
11 | | -encrypted data without having to decrypt the data which brings in |
12 | | -profound and beneficial implications for data privacy and data confidentiality. |
13 | | -However, these benefits come with a significant performance cost which has so far |
14 | | -confined FHE based applications to specific use case archetypes primarily in use |
15 | | -by the regulated industries and government. |
| 7 | +[Fully Homomorphic Encryption (FHE)](https://en.wikipedia.org/wiki/Homomorphic_encryption) |
| 8 | +allows computation to be performed on encrypted data without having to decrypt |
| 9 | +the data which brings in profound and beneficial implications for data privacy |
| 10 | +and data confidentiality. However, these benefits come with a significant performance |
| 11 | +cost which has so far confined FHE based applications to specific use case archetypes |
| 12 | +primarily in use by the regulated industries and government. |
16 | 13 |
|
17 | 14 | Most of the FHE schemes of today perform the computation using very large |
18 | 15 | polynomial rings, thus requiring considerable compute power and data |
19 | | -movement between main memory and the CPU's registers. HERACLES improves |
20 | | -the performance of FHE by accelerating the computation over the large |
21 | | -polynomials and optimizing the data movement involved in the computation. |
| 16 | +movement between main memory and the CPU's registers, which introduces large |
| 17 | +computational overheds. New FHE hardware accelerator architectures aim at |
| 18 | +accelerate the computation over large polynomial rings while minimizing |
| 19 | +the data movement between the memory and the compute elements. |
| 20 | + |
| 21 | +The Encrypted Computing SDK introduces a multistage transformation |
| 22 | +(compiler) pipeline that breaks down the large polynomial computation into |
| 23 | +the various abstraction layers to make it easier for developers to target |
| 24 | +different hardware architectures as well as to develop new implementations |
| 25 | +of FHE schemes, and integrate with existing 3rd Party libraries, compilers |
| 26 | +and transpilers. The SDK is inspired by the |
| 27 | +[LLVM Compiler Infrastructure](https://llvm.org/), and adopts a |
| 28 | +modular approach based on language independent intermediate |
| 29 | +representations (IRs) that promotes the separation of concerns at each |
| 30 | +stage of the pipeline and allowing for dedicated transformations and |
| 31 | +optimizations. |
22 | 32 |
|
23 | | -HERACLES introduces a new Polynomial Data type which does not exist in |
24 | | -today's traditional CPUs. For this new polynomial data type, it supports |
25 | | -a new set of novel and fundamental instructions, the Polynomial Instructions |
| 33 | +Our first target hardware platform is Intel’s |
| 34 | +[HERACLES accelerator technology](https://dl.acm.org/doi/10.1145/3560810.3565290) |
| 35 | +which introduces a new Polynomial Data type which does not exist in today's |
| 36 | +traditional CPUs. For this new polynomial data type, it supports a |
| 37 | +new set of novel and fundamental instructions, the Polynomial Instructions |
26 | 38 | Set Architecture (P-ISA), that operates directly on large polynomials in |
27 | | -a SIMD fashion. We at Intel Labs are developing a new compiler pipeline, |
28 | | -the Encrypted Computing SDK, to make it easier for developers to develop |
29 | | -new implementations of FHE schemes and also integrate with existing libraries. |
| 39 | +a SIMD fashion. |
30 | 40 |
|
31 | | -<p> |
32 | | -<img src="docs/images/HERACLES_SDK_Integration_3rd_Party.png" align="left" width="600" /> |
33 | 41 |
|
34 | | -The Encrypted Computing SDK (or HERACLES SDK) will realize a multistage |
35 | | -transformation (compiler) pipeline, inspired by the |
36 | | -[LLVM Compiler Infrastructure](https://llvm.org/). We have adopted a |
37 | | -modular approach based on language independent intermediate |
38 | | -representations (IR) that promotes the separation of concerns at each |
39 | | -stage of the pipeline and allowing for dedicated transformations and |
40 | | -optimizations. |
| 42 | +<p> |
| 43 | +<img src="docs/images/SDK_Integration_3rd_party.png" align="left" width="600" /> |
41 | 44 |
|
42 | | -This approach also allows for integration with 3rd Party compilers and |
43 | | -transpilers. |
44 | 45 | <br clear="left"> |
45 | 46 | <br/> |
46 | 47 | <br/> |
47 | 48 | </p> |
48 | 49 |
|
49 | | -## HERACLES SDK Roadmap: Phased Approach |
| 50 | +## Encrypted Computing SDK Roadmap: Phased Approach |
50 | 51 | <p> |
51 | | -<img src="docs/images/HERACLES_SDK_Phased_Approach.png" width=80% height=80% /> |
| 52 | +<img src="docs/images/SDK_Roadmap.png" width=80% height=80% /> |
52 | 53 | <br/> |
53 | 54 | <br/> |
54 | 55 | <br/> |
55 | 56 | </p> |
56 | 57 |
|
57 | | -### HERACLES SDK Phase 1: Components and Tasks |
| 58 | +### Encrypted Computing SDK Phase 1: Components and Tasks |
58 | 59 | <p> |
59 | | -<img src="docs/images/HERACLES_SDK_Phase_1.png" width=80% height=80% /> |
| 60 | +<img src="docs/images/SDK_Phase_1.png" width=80% height=80% /> |
60 | 61 | <br> |
61 | 62 | <br/> |
62 | 63 | <br/> |
63 | 64 | </p> |
64 | 65 |
|
65 | | -We are currently at Phase 1, more specifically developing the P-ISA Tools |
66 | | -component which comprises three main tools, a) Kernel Generator, b) Program |
67 | | -Mapper, and c) Functional Modeler Simulator. |
| 66 | +We are currently at Phase 1, developing the P-ISA tools and Assembler tools |
| 67 | +component which comprises a) Kernel Generator, b) Program |
| 68 | +Mapper, c) Functional Modeler Simulator, and d) HERACLES Assembler. |
68 | 69 | Each tool in this repo is self contained and has its own local README. |
69 | 70 |
|
70 | | -Current development is focussed on the Kernel Generator. |
71 | | -Follow the instructions [here](./kerngen) to start experimenting with it. |
72 | | - |
73 | 71 | # Contributing |
74 | 72 | Intel P-ISA Tools project welcomes external contributions through pull |
75 | 73 | requests to the `main` branch. |
|
0 commit comments