Composite Prime Moduli Sampling and Scaling Factor Calculations (#910 phase 2) #929

fdiasmor · 2025-01-03T02:35:34Z

No description provided.

…mode.

…ng mode.

…ree.

…ters.

src/pke/include/cryptocontext.h

src/pke/lib/scheme/ckksrns/ckksrns-parametergeneration.cpp

src/pke/lib/schemerns/rns-cryptoparameters.cpp

yspolyakov · 2025-01-23T21:14:13Z

src/pke/include/cryptocontext.h

@@ -2372,11 +2387,19 @@ class CryptoContextImpl : public Serializable {
    void ModReduceInPlace(Ciphertext<Element>& ciphertext) const {
        ValidateCiphertext(ciphertext);

-        GetScheme()->ModReduceInPlace(ciphertext, BASE_NUM_LEVELS_TO_DROP);
+        if (isCKKS(m_schemeId)) {


Any CKKS-specific code should be implemented inside the folder for the CKKS scheme (not in global cryptocontext code)

yspolyakov · 2025-01-23T21:21:11Z

src/pke/lib/schemebase/base-leveledshe.cpp

    Ciphertext<Element> ciphertext = EvalMult(ciphertext1, ciphertext2);
    algo->KeySwitchInPlace(ciphertext, evalKey);
-    ModReduceInPlace(ciphertext, BASE_NUM_LEVELS_TO_DROP);
+    uint32_t levelsToDrop = BASE_NUM_LEVELS_TO_DROP;
+    if (cc->getSchemeId() == CKKSRNS_SCHEME) {


Why is this code here rather then in CKKS files?

dsuponitskiy · 2025-02-05T13:34:00Z

src/pke/include/cryptocontext.h

@@ -66,6 +66,7 @@
 #include <algorithm>
 #include <unordered_map>
 #include <set>
+#include <iostream>


this is not what i meant. I would consider the code for DEBUG_KEY as legacy, and we prefer not to expand it.

Should I remove all the code under DEBUG_KEY? I only added the header because of pre-commit complaints. Let me know.

dsuponitskiy · 2025-02-05T13:40:44Z

src/pke/include/cryptocontext.h

+    * GetCompositeDegree: get composite degree of the current scheme crypto context.
+    * @return integer value corresponding to composite degree
+    */
+    uint32_t GetCompositeDegreeFromCtxt() const {


GetCompositeDegreeFromCtxt() should be in either protected or private section of CryptoContextImpl as it is used only inside the class

dsuponitskiy · 2025-02-05T13:51:45Z

src/pke/lib/scheme/ckksrns/ckksrns-parametergeneration.cpp

-        std::cout << __FUNCTION__ << "::" << __LINE__ << " numPrimes: " << numPrimes << " qBound: " << qBound
-                  << std::endl;
-    }
+    numPrimes *= compositeDegree;


Just a question: this multiplication numPrimes *= compositeDegree; used to be performed for scalTech == COMPOSITESCALINGAUTO || scalTech == COMPOSITESCALINGMANUAL only. Now it is done for any scalTech. Is this correct?

Since it is a CKKS class, yes, because on single scaling mode it will be multiplied by 1. So, I figured we don't really need the overhead of branch prediction.

dsuponitskiy · 2025-02-05T14:06:46Z

src/pke/lib/scheme/ckksrns/ckksrns-advancedshe.cpp

+    const auto cryptoParams = std::dynamic_pointer_cast<CryptoParametersRNS>(ciphertextVec[0]->GetCryptoParameters());
+    uint32_t levelsToDrop   = cryptoParams->GetCompositeDegree();
+
+    for (size_t i = 0; i < lim; i = i + 2) {


this for loop should be split: the logic to calculate ciphertextVec's index is complicated as there are 2 additional conditions to check in every iteration. a possible solution could look like this (we can discuss the changes):

size_t ctrIndex = 0; size_t i = 0; for (; i < (inSize - 1); i += 2) { ciphertextMultVec[ctrIndex] = algo->EvalMultAndRelinearize(ciphertextVec[i], ciphertextVec[i + 1], evalKeys); algo->ModReduceInPlace(ciphertextMultVec[ctrIndex++], levelsToDrop); } if(i < inSize) { ciphertextMultVec[ctrIndex] = algo->EvalMultAndRelinearize(ciphertextVec[i], ciphertextVec[0], evalKeys); algo->ModReduceInPlace(ciphertextMultVec[ctrIndex++], levelsToDrop); i += 2; } for (; i < lim; i += 2) { ciphertextMultVec[ctrIndex] = algo->EvalMultAndRelinearize(ciphertextMultVec[i - inSize], ciphertextMultVec[i + 1 - inSize], evalKeys); algo->ModReduceInPlace(ciphertextMultVec[ctrIndex++], levelsToDrop); }

That works except that the second operand inside the special case is ciphertextMultVec[i + 1 - inSize].

dsuponitskiy · 2025-02-05T14:10:47Z

src/pke/lib/scheme/ckksrns/ckksrns-advancedshe.cpp

+    size_t ctrIndex = 0;
+
+    auto algo               = ciphertextVec[0]->GetCryptoContext()->GetScheme();
+    const auto cc           = ciphertextVec[0]->GetCryptoContext();


cc is not used here

I thought I had removed that... Thanks for picking up. Compiler should have complained. Fixed it now.

dsuponitskiy · 2025-02-05T14:15:27Z

src/pke/lib/scheme/ckksrns/ckksrns-advancedshe.cpp

@@ -46,6 +46,32 @@ CKKS implementation. See https://eprint.iacr.org/2020/1118 for details.

 namespace lbcrypto {

+Ciphertext<DCRTPoly> AdvancedSHECKKSRNS::EvalMultMany(const std::vector<Ciphertext<DCRTPoly>>& ciphertextVec,
+                                                      const std::vector<EvalKey<DCRTPoly>>& evalKeys) const {
+    if (ciphertextVec.size() < 1)


if ciphertextVec.size() == 1 then the loop below is not executed. possibly if (ciphertextVec.size() < 2) would suffice? Also, return ciphertextMultVec.back(); will cause a segfault if ciphertextVec.size() == 1

This implementation is borrowed from the base-advancedshe.cpp. This fix will be propagated to the original implementation in base-advancedshe.cpp as follows:

if (ciphertextVec.size() < 2) OPENFHE_THROW("Input ciphertext vector size should be 1 or more"); if (ciphertextVec.size() == 1) return ciphertextVec[0];

dsuponitskiy · 2025-02-05T15:23:28Z

src/pke/lib/schemerns/rns-leveledshe.cpp

+    uint32_t levelsToDrop = BASE_NUM_LEVELS_TO_DROP;
+    if (cryptoParams->GetScalingTechnique() == COMPOSITESCALINGAUTO ||
+        cryptoParams->GetScalingTechnique() == COMPOSITESCALINGMANUAL) {
+        const auto cryptoRNSParams = std::dynamic_pointer_cast<CryptoParametersRNS>(cryptoParams);


cryptoRNSParams is not needed, you could re-use cryptoParams

Not sure what happened here. Also just above, the line const auto cc = ciphertext1->GetCryptoContext(); is not needed either.

fdiasmor added 14 commits December 11, 2024 17:51

Initial commit: adding composite scaling parameters and scaling tech …

3ed5909

…mode.

Initial commit: adding composite scaling parameters and scaling tech …

fb50d79

…mode.

Composite prime generation and adaptive scaling factors.

d3c1a58

Propagating composite degree to mod reduce calls.

e397a73

Handling overflow for scaling factors over 64 bits on composite scali…

3bf393a

…ng mode.

Bug fix of overflow handling for d > 2 and sc > 64.

51db3ca

Update impl of Rescale and Mod/LevelReduce methods with composite deg…

44b206b

…ree.

Minor bug fix: else case missing.

3d2c220

Update EvalPoly functions to use composite degree.

a9c9978

Update Compress function to raise error.

c03c8f9

Update/Add new composite scaling examples.

470cd11

Recalculate sizeP and update EstimateLogP interface.

6369eaa

Update composite-prime generation function.

5d54a9f

Remove comments from prime gen function.

a5a2b9c

fdiasmor requested a review from yspolyakov January 3, 2025 02:35

fdiasmor self-assigned this Jan 3, 2025

fdiasmor linked an issue Jan 3, 2025 that may be closed by this pull request

Add CKKS Composite Scaling #910

Open

fdiasmor changed the base branch from main to dev January 3, 2025 02:35

fdiasmor added 7 commits January 3, 2025 17:51

Update composite scaling unittests and fix unittest runtime issue.

9a7ad26

Some review changes.

c8c76a8

Update composite scaling unittests and fix unittest runtime issue.

4b721d9

Disable composite scaling set methods for non-CKKS schemes.

cf685de

Move COMPOSITESCALING support error handling.

a5a443a

Disable composite scaling set methods for non-CKKS schemes.

b14472d

Removing additional constructor that accepts composite scaling parame…

6dd35c0

…ters.

fdiasmor changed the title ~~#910 Composite Prime Moduli Sampling and Composite Scaling Factor Calculations~~ Composite Prime Moduli Sampling and Scaling Factor Calculations (#910 phase 2) Jan 10, 2025

fdiasmor added 2 commits January 17, 2025 17:30

Merge/rebase review changes from (#928).

e9cdffa

Merge branch 'dev' into 910-composite-prime-gen

80fe03f

fdiasmor requested a review from dsuponitskiy January 18, 2025 01:39

dsuponitskiy mentioned this pull request Jan 22, 2025

Code to be corrected #681

Open

dsuponitskiy reviewed Jan 23, 2025

View reviewed changes

yspolyakov reviewed Jan 23, 2025

View reviewed changes

Review changes for PR 910 phase 2.

c63d0fe

fdiasmor requested review from yspolyakov and dsuponitskiy February 5, 2025 01:47

dsuponitskiy reviewed Feb 5, 2025

View reviewed changes

Review updates and fixes.

2ec7adf

fdiasmor requested a review from dsuponitskiy February 6, 2025 01:33

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Composite Prime Moduli Sampling and Scaling Factor Calculations (#910 phase 2) #929

Composite Prime Moduli Sampling and Scaling Factor Calculations (#910 phase 2) #929

fdiasmor commented Jan 3, 2025

yspolyakov Jan 23, 2025

fdiasmor Feb 5, 2025

yspolyakov Jan 23, 2025

fdiasmor Feb 5, 2025

dsuponitskiy Feb 5, 2025

fdiasmor Feb 5, 2025

dsuponitskiy Feb 5, 2025

fdiasmor Feb 5, 2025

dsuponitskiy Feb 5, 2025

fdiasmor Feb 5, 2025

dsuponitskiy Feb 5, 2025

fdiasmor Feb 6, 2025

dsuponitskiy Feb 5, 2025

fdiasmor Feb 5, 2025

dsuponitskiy Feb 5, 2025 •

edited

Loading

fdiasmor Feb 6, 2025

dsuponitskiy Feb 5, 2025

fdiasmor Feb 6, 2025

Composite Prime Moduli Sampling and Scaling Factor Calculations (#910 phase 2) #929

Are you sure you want to change the base?

Composite Prime Moduli Sampling and Scaling Factor Calculations (#910 phase 2) #929

Conversation

fdiasmor commented Jan 3, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dsuponitskiy Feb 5, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dsuponitskiy Feb 5, 2025 •

edited

Loading