Release notes for 0.9.2-rc1

.github/workflows/linux.yml

@@ -30,6 +30,7 @@ jobs:
           git config --global user.name "ciuser" && \
           git config --global user.email "ci@openquantumsafe.org" && \
           export LIBOQS_DIR=`pwd` && \
+          git config --global --add safe.directory $LIBOQS_DIR && \
           cd scripts/copy_from_upstream && \
           ! pip3 install -r requirements.txt 2>&1 | grep ERROR && \
           python3 copy_from_upstream.py copy && \

.travis.yml

@@ -1,6 +1,6 @@
 language: c
 before_script:
-  - sudo apt -y install astyle cmake gcc ninja-build libssl-dev python3-pytest python3-pytest-xdist unzip xsltproc doxygen graphviz valgrind
+  - sudo apt update && sudo apt -y install astyle cmake gcc ninja-build libssl-dev python3-pytest python3-pytest-xdist unzip xsltproc doxygen graphviz valgrind
 jobs:
   include:
     - arch: ppc64le         # The IBM Power LXD container based build for OSS only

CMakeLists.txt

@@ -33,7 +33,7 @@ set(CMAKE_C_STANDARD 11)
 set(CMAKE_C_STANDARD_REQUIRED ON)
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 set(CMAKE_C_VISIBILITY_PRESET hidden)
-set(OQS_VERSION_TEXT "0.9.0")
+set(OQS_VERSION_TEXT "0.9.2-rc1")
 set(OQS_COMPILE_BUILD_TARGET "${CMAKE_SYSTEM_PROCESSOR}-${CMAKE_HOST_SYSTEM}")
 set(OQS_MINIMAL_GCC_VERSION "7.1.0")
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)

RELEASE.md

@@ -1,5 +1,5 @@
-liboqs version 0.9.0
-====================
+liboqs version 0.9.2-rc1
+========================
 
 About
 -----
@@ -28,78 +28,22 @@ liboqs can also be used in the following programming languages via language-spec
 Release notes
 =============
 
-This is version 0.9.0 of liboqs. It was released on October 12, 2023.
+This is release candidate 1 of version 0.9.2 of liboqs. It was released on January 11, 2024.
 
-This release features an update to the Classic McEliece KEM, bringing it in line with NIST Round 4. It also adds or updates ARM implementations for Kyber, Dilithium, and Falcon.
+This release is a security release which fixes potential non-constant-time behaviour in Kyber based on https://github.com/pq-crystals/kyber/commit/272125f6acc8e8b6850fd68ceb901a660ff48196
 
 What's New
 ----------
 
-This release continues from the 0.8.0 release of liboqs.
+This release continues from the 0.9.1 release of liboqs.
 
 ### Key encapsulation mechanisms
 
-- Classic McEliece: updated to Round 4 version.
-- Kyber: aarch64 implementation updated.
-
-### Digital signature schemes
-
-- Dilithium: aarch64 implementation updated.
-- Falcon: aarch64 implementation added.
-
-### Other changes
-
-- Update algorithm documentation
-- Support compilation for Windows on ARM64, Apple mobile, and Android platforms
-- Improve resilience of randombytes on Apple systems
-
-Release call
-============
-
-Users of liboqs are invited to join a webinar on Thursday, November 2, 2023, from 12-1pm US Eastern time for information on this release, plans for the next release cycle, and to provide feedback on OQS usage and features.  
-
-The Zoom link for the webinar is: https://uwaterloo.zoom.us/j/98288698086
-
----
+- Kyber: C, AVX2, and aarch64 implementation updated
 
 Detailed changelog
 ------------------
 
-* Fix libdir value in liboqs.pc by @vt-alt in https://github.com/open-quantum-safe/liboqs/pull/1496
-* update version and remove CCI triggers by @baentsch in https://github.com/open-quantum-safe/liboqs/pull/1498
-* create deb package and retain as artifact by @baentsch in https://github.com/open-quantum-safe/liboqs/pull/1501
-* README correction to docs path & additional gitignore to macos + vscode by @planetf1 in https://github.com/open-quantum-safe/liboqs/pull/1503
-* Trigger liboqs-python CI via GitHub API by @SWilson4 in https://github.com/open-quantum-safe/liboqs/pull/1507
-* Update Classic McEliece by @praveksharma in https://github.com/open-quantum-safe/liboqs/pull/1470
-* update BIKE documentation by @baentsch in https://github.com/open-quantum-safe/liboqs/pull/1509
-* kyber/dilithium aarch64 pull from pqclean + patches by @bhess in https://github.com/open-quantum-safe/liboqs/pull/1512
-* Pull Falcon updates from PQClean by @dstebila in https://github.com/open-quantum-safe/liboqs/pull/1523
-* Bump XCode by @baentsch in https://github.com/open-quantum-safe/liboqs/pull/1526
-* Update Classic McEliece supression files by @praveksharma in https://github.com/open-quantum-safe/liboqs/pull/1527
-* Bump gitpython from 3.1.30 to 3.1.32 in /scripts/copy_from_upstream by @dependabot in https://github.com/open-quantum-safe/liboqs/pull/1524
-* ci: add CI for android by @res0nance in https://github.com/open-quantum-safe/liboqs/pull/1531
-* re-enable armhf speed testing by @baentsch in https://github.com/open-quantum-safe/liboqs/pull/1535
-* Bump gitpython from 3.1.32 to 3.1.34 in /scripts/copy_from_upstream by @dependabot in https://github.com/open-quantum-safe/liboqs/pull/1538
-* Prefer arc4random on Apple platforms by @res0nance in https://github.com/open-quantum-safe/liboqs/pull/1544
-* Bump gitpython from 3.1.34 to 3.1.35 in /scripts/copy_from_upstream by @dependabot in https://github.com/open-quantum-safe/liboqs/pull/1551
-* Update Classic McEliece suppression files by @praveksharma in https://github.com/open-quantum-safe/liboqs/pull/1541
-* Pull Neon implementation of Falcon from PQClean by @SWilson4 in https://github.com/open-quantum-safe/liboqs/pull/1547
-* ci: add CI for apple mobile platforms by @res0nance in https://github.com/open-quantum-safe/liboqs/pull/1546
-* Add Windows ARM64 support by @res0nance in https://github.com/open-quantum-safe/liboqs/pull/1545
-* Document Falcon constant time errors by @praveksharma in https://github.com/open-quantum-safe/liboqs/pull/1552
-* ci: github actions CI for Windows x86 and x64 by @res0nance in https://github.com/open-quantum-safe/liboqs/pull/1554
-* build: Align VS test folder with all other Generators by @res0nance in https://github.com/open-quantum-safe/liboqs/pull/1557
-* Fix weekly.yml to skip McEliece by @praveksharma in https://github.com/open-quantum-safe/liboqs/pull/1562
-* Enable extensions in constant-time tests by @SWilson4 in https://github.com/open-quantum-safe/liboqs/pull/1567
-* Update Classic McEliece supression files by @praveksharma in https://github.com/open-quantum-safe/liboqs/pull/1568
-* liboqs 0.9.0 release candidate 1 by @SWilson4 in https://github.com/open-quantum-safe/liboqs/pull/1570
-* add community standard documentation [skip ci] by @baentsch in https://github.com/open-quantum-safe/liboqs/pull/1565
-* Bump gitpython from 3.1.35 to 3.1.37 in /scripts/copy_from_upstream by @dependabot in https://github.com/open-quantum-safe/liboqs/pull/1575
+* Pull Kyber division fixes from PQ-Crystals into dev-092 by @praveksharma in https://github.com/open-quantum-safe/liboqs/pull/1652
 
-## New Contributors
-* @planetf1 made their first contribution in https://github.com/open-quantum-safe/liboqs/pull/1503
-* @SWilson4 made their first contribution in https://github.com/open-quantum-safe/liboqs/pull/1507
-* @praveksharma made their first contribution in https://github.com/open-quantum-safe/liboqs/pull/1470
-* @res0nance made their first contribution in https://github.com/open-quantum-safe/liboqs/pull/1531
-
-**Full Changelog**: https://github.com/open-quantum-safe/liboqs/compare/0.8.0...0.9.0
+**Full Changelog**: https://github.com/open-quantum-safe/liboqs/compare/0.9.1...0.9.2-rc1

docs/algorithms/kem/classic_mceliece.md

@@ -14,7 +14,7 @@
 ## Advisories
 
 - Classic-McEliece-460896, Classic-McEliece-460896f, Classic-McEliece-6960119, and Classic-McEliece-6960119f parameter sets fail memory leak testing on x86-64 when building with ``clang`` using optimization level ``-O2`` and ``-O3``. Care is advised when using the algorithm at higher optimization levels, and any other compiler and architecture.
-- Current implementation of the algorithm may not be constant-time. Additionally, environment specific constant-time leaks may not be documented; please report potential constant-time leaks when found. 
+- Current implementation of the algorithm may not be constant-time. Additionally, environment specific constant-time leaks may not be documented; please report potential constant-time leaks when found.
 
 ## Parameter set summary
 
@@ -35,8 +35,8 @@
 
 |       Implementation source       | Identifier in upstream   | Supported architecture(s)   | Supported operating system(s)   | CPU extension(s) used   | No branching-on-secrets claimed?   | No branching-on-secrets checked by valgrind?   | Large stack usage?‡   |
 |:---------------------------------:|:-------------------------|:----------------------------|:--------------------------------|:------------------------|:-----------------------------------|:-----------------------------------------------|:----------------------|
-| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | True                               | True                                           | True                  |
-| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT             | False                              | True                                           | True                  |
+| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | False                              | False                                          | True                  |
+| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT             | False                              | False                                          | True                  |
 
 Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
@@ -46,8 +46,8 @@ Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
 |       Implementation source       | Identifier in upstream   | Supported architecture(s)   | Supported operating system(s)   | CPU extension(s) used   | No branching-on-secrets claimed?   | No branching-on-secrets checked by valgrind?   | Large stack usage?   |
 |:---------------------------------:|:-------------------------|:----------------------------|:--------------------------------|:------------------------|:-----------------------------------|:-----------------------------------------------|:---------------------|
-| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | True                               | True                                           | True                 |
-| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT,BMI1        | False                              | True                                           | True                 |
+| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | False                              | False                                          | True                 |
+| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT,BMI1        | False                              | False                                          | True                 |
 
 Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
@@ -55,8 +55,8 @@ Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
 |       Implementation source       | Identifier in upstream   | Supported architecture(s)   | Supported operating system(s)   | CPU extension(s) used   | No branching-on-secrets claimed?   | No branching-on-secrets checked by valgrind?   | Large stack usage?   |
 |:---------------------------------:|:-------------------------|:----------------------------|:--------------------------------|:------------------------|:-----------------------------------|:-----------------------------------------------|:---------------------|
-| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | True                               | True                                           | True                 |
-| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT             | False                              | True                                           | True                 |
+| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | False                              | False                                          | True                 |
+| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT             | False                              | False                                          | True                 |
 
 Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
@@ -64,8 +64,8 @@ Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
 |       Implementation source       | Identifier in upstream   | Supported architecture(s)   | Supported operating system(s)   | CPU extension(s) used   | No branching-on-secrets claimed?   | No branching-on-secrets checked by valgrind?   | Large stack usage?   |
 |:---------------------------------:|:-------------------------|:----------------------------|:--------------------------------|:------------------------|:-----------------------------------|:-----------------------------------------------|:---------------------|
-| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | True                               | True                                           | True                 |
-| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT,BMI1        | False                              | True                                           | True                 |
+| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | False                              | False                                          | True                 |
+| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT,BMI1        | False                              | False                                          | True                 |
 
 Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
@@ -73,8 +73,8 @@ Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
 |       Implementation source       | Identifier in upstream   | Supported architecture(s)   | Supported operating system(s)   | CPU extension(s) used   | No branching-on-secrets claimed?   | No branching-on-secrets checked by valgrind?   | Large stack usage?   |
 |:---------------------------------:|:-------------------------|:----------------------------|:--------------------------------|:------------------------|:-----------------------------------|:-----------------------------------------------|:---------------------|
-| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | True                               | True                                           | True                 |
-| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT             | False                              | True                                           | True                 |
+| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | False                              | False                                          | True                 |
+| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT             | False                              | False                                          | True                 |
 
 Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
@@ -82,8 +82,8 @@ Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
 |       Implementation source       | Identifier in upstream   | Supported architecture(s)   | Supported operating system(s)   | CPU extension(s) used   | No branching-on-secrets claimed?   | No branching-on-secrets checked by valgrind?   | Large stack usage?   |
 |:---------------------------------:|:-------------------------|:----------------------------|:--------------------------------|:------------------------|:-----------------------------------|:-----------------------------------------------|:---------------------|
-| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | True                               | True                                           | True                 |
-| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT,BMI1        | False                              | True                                           | True                 |
+| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | False                              | False                                          | True                 |
+| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT,BMI1        | False                              | False                                          | True                 |
 
 Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
@@ -91,8 +91,8 @@ Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
 |       Implementation source       | Identifier in upstream   | Supported architecture(s)   | Supported operating system(s)   | CPU extension(s) used   | No branching-on-secrets claimed?   | No branching-on-secrets checked by valgrind?   | Large stack usage?   |
 |:---------------------------------:|:-------------------------|:----------------------------|:--------------------------------|:------------------------|:-----------------------------------|:-----------------------------------------------|:---------------------|
-| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | True                               | True                                           | True                 |
-| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT             | False                              | True                                           | True                 |
+| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | False                              | False                                          | True                 |
+| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT             | False                              | False                                          | True                 |
 
 Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
@@ -100,8 +100,8 @@ Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
 |       Implementation source       | Identifier in upstream   | Supported architecture(s)   | Supported operating system(s)   | CPU extension(s) used   | No branching-on-secrets claimed?   | No branching-on-secrets checked by valgrind?   | Large stack usage?   |
 |:---------------------------------:|:-------------------------|:----------------------------|:--------------------------------|:------------------------|:-----------------------------------|:-----------------------------------------------|:---------------------|
-| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | True                               | True                                           | True                 |
-| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT,BMI1        | False                              | True                                           | True                 |
+| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | False                              | False                                          | True                 |
+| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT,BMI1        | False                              | False                                          | True                 |
 
 Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
@@ -109,8 +109,8 @@ Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
 |       Implementation source       | Identifier in upstream   | Supported architecture(s)   | Supported operating system(s)   | CPU extension(s) used   | No branching-on-secrets claimed?   | No branching-on-secrets checked by valgrind?   | Large stack usage?   |
 |:---------------------------------:|:-------------------------|:----------------------------|:--------------------------------|:------------------------|:-----------------------------------|:-----------------------------------------------|:---------------------|
-| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | True                               | True                                           | True                 |
-| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT             | False                              | True                                           | True                 |
+| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | False                              | False                                          | True                 |
+| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT             | False                              | False                                          | True                 |
 
 Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
@@ -118,8 +118,8 @@ Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
 |       Implementation source       | Identifier in upstream   | Supported architecture(s)   | Supported operating system(s)   | CPU extension(s) used   | No branching-on-secrets claimed?   | No branching-on-secrets checked by valgrind?   | Large stack usage?   |
 |:---------------------------------:|:-------------------------|:----------------------------|:--------------------------------|:------------------------|:-----------------------------------|:-----------------------------------------------|:---------------------|
-| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | True                               | True                                           | True                 |
-| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT,BMI1        | False                              | True                                           | True                 |
+| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | False                              | False                                          | True                 |
+| [Primary Source](#primary-source) | avx2                     | x86\_64                     | Linux,Darwin                    | AVX2,POPCNT,BMI1        | False                              | False                                          | True                 |
 
 Are implementations chosen based on runtime CPU feature detection? **Yes**.
 

docs/algorithms/kem/classic_mceliece.yml

@@ -26,7 +26,9 @@ advisories:
   building with ``clang`` using optimization level ``-O2`` and ``-O3``. Care is advised
   when using the algorithm at higher optimization levels, and any other compiler and
   architecture.
-- Current implementation of the algorithm may not be constant-time. Additionally, environment specific constant-time leaks may not be documented; please report potential constant-time leaks when found. 
+- Current implementation of the algorithm may not be constant-time. Additionally,
+  environment specific constant-time leaks may not be documented; please report potential
+  constant-time leaks when found.
 parameter-sets:
 - name: Classic-McEliece-348864
   claimed-nist-level: 1

docs/algorithms/kem/kyber.md

@@ -7,9 +7,9 @@
 - **Authors' website**: https://pq-crystals.org/
 - **Specification version**: NIST Round 3 submission.
 - **Primary Source**<a name="primary-source"></a>:
-  - **Source**: https://github.com/pq-crystals/kyber/commit/518de2414a85052bb91349bcbcc347f391292d5b with copy_from_upstream patches
+  - **Source**: https://github.com/pq-crystals/kyber/commit/b628ba78711bc28327dc7d2d5c074a00f061884e with copy_from_upstream patches
   - **Implementation license (SPDX-Identifier)**: CC0-1.0 or Apache-2.0
-- **Optimized Implementation sources**: https://github.com/pq-crystals/kyber/commit/518de2414a85052bb91349bcbcc347f391292d5b with copy_from_upstream patches
+- **Optimized Implementation sources**: https://github.com/pq-crystals/kyber/commit/b628ba78711bc28327dc7d2d5c074a00f061884e with copy_from_upstream patches
   - **pqclean-aarch64**:<a name="pqclean-aarch64"></a>
       - **Source**: https://github.com/PQClean/PQClean/commit/8e220a87308154d48fdfac40abbb191ac7fce06a with copy_from_upstream patches
       - **Implementation license (SPDX-Identifier)**: CC0-1.0 and (CC0-1.0 or Apache-2.0) and (CC0-1.0 or MIT) and MIT

docs/algorithms/kem/kyber.yml

@@ -17,7 +17,7 @@ website: https://pq-crystals.org/
 nist-round: 3
 spec-version: NIST Round 3 submission
 primary-upstream:
-  source: https://github.com/pq-crystals/kyber/commit/518de2414a85052bb91349bcbcc347f391292d5b
+  source: https://github.com/pq-crystals/kyber/commit/b628ba78711bc28327dc7d2d5c074a00f061884e
     with copy_from_upstream patches
   spdx-license-identifier: CC0-1.0 or Apache-2.0
 optimized-upstreams:

docs/algorithms/sig/falcon.md

@@ -22,7 +22,7 @@
 
 |       Implementation source       | Identifier in upstream   | Supported architecture(s)   | Supported operating system(s)   | CPU extension(s) used   | No branching-on-secrets claimed?   | No branching-on-secrets checked by valgrind?   | Large stack usage?‡   |
 |:---------------------------------:|:-------------------------|:----------------------------|:--------------------------------|:------------------------|:-----------------------------------|:-----------------------------------------------|:----------------------|
-| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | False                              | False                                          | False                 |
+| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | True                               | True                                           | False                 |
 | [Primary Source](#primary-source) | avx2                     | x86\_64                     | All                             | AVX2                    | False                              | False                                          | False                 |
 | [Primary Source](#primary-source) | aarch64                  | ARM64\_V8                   | Linux,Darwin                    | None                    | False                              | False                                          | False                 |
 
@@ -34,7 +34,7 @@ Are implementations chosen based on runtime CPU feature detection? **Yes**.
 
 |       Implementation source       | Identifier in upstream   | Supported architecture(s)   | Supported operating system(s)   | CPU extension(s) used   | No branching-on-secrets claimed?   | No branching-on-secrets checked by valgrind?   | Large stack usage?   |
 |:---------------------------------:|:-------------------------|:----------------------------|:--------------------------------|:------------------------|:-----------------------------------|:-----------------------------------------------|:---------------------|
-| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | False                              | False                                          | False                |
+| [Primary Source](#primary-source) | clean                    | All                         | All                             | None                    | True                               | True                                           | False                |
 | [Primary Source](#primary-source) | avx2                     | x86\_64                     | All                             | AVX2                    | False                              | False                                          | False                |
 | [Primary Source](#primary-source) | aarch64                  | ARM64\_V8                   | Linux,Darwin                    | None                    | False                              | False                                          | False                |
 

scripts/copy_from_upstream/copy_from_upstream.py

@@ -611,8 +611,6 @@ def copy_from_upstream():
     for t in ["kem", "sig"]:
         with open(os.path.join(os.environ['LIBOQS_DIR'], 'tests', 'KATs', t, 'kats.json'), "w") as f:
             json.dump(kats[t], f, indent=2, sort_keys=True)
-    if not keepdata:
-        shutil.rmtree('repos')
 
     update_upstream_alg_docs.do_it(os.environ['LIBOQS_DIR'])
 
@@ -622,6 +620,10 @@ def copy_from_upstream():
     update_docs_from_yaml.do_it(os.environ['LIBOQS_DIR'])
     update_cbom.update_cbom_if_algs_not_changed(os.environ['LIBOQS_DIR'], "git")
 
+    if not keepdata:
+        shutil.rmtree('repos')
+
+
 def verify_from_upstream():
     instructions = load_instructions()
     basedir = "verify_from_upstream"

scripts/copy_from_upstream/copy_from_upstream.yml

@@ -8,13 +8,13 @@ upstreams:
     sig_meta_path: 'crypto_sign/{pqclean_scheme}/META.yml'
     kem_scheme_path: 'crypto_kem/{pqclean_scheme}'
     sig_scheme_path: 'crypto_sign/{pqclean_scheme}'
-    patches: [pqclean-sphincs.patch, pqclean-dilithium-arm-randomized-signing.patch, pqclean-kyber-armneon-shake-fixes.patch, pqclean-kyber-armneon-768-1024-fixes.patch]
+    patches: [pqclean-sphincs.patch, pqclean-dilithium-arm-randomized-signing.patch, pqclean-kyber-armneon-shake-fixes.patch, pqclean-kyber-armneon-768-1024-fixes.patch, pqclean-kyber-armneon-variable-timing-fix.patch]
     ignore: pqclean_sphincs-shake-256s-simple_aarch64, pqclean_sphincs-shake-256s-simple_aarch64, pqclean_sphincs-shake-256f-simple_aarch64, pqclean_sphincs-shake-192s-simple_aarch64, pqclean_sphincs-shake-192f-simple_aarch64, pqclean_sphincs-shake-128s-simple_aarch64, pqclean_sphincs-shake-128f-simple_aarch64
   -
     name: pqcrystals-kyber
     git_url: https://github.com/pq-crystals/kyber.git
     git_branch: master
-    git_commit: 518de2414a85052bb91349bcbcc347f391292d5b
+    git_commit: b628ba78711bc28327dc7d2d5c074a00f061884e
     kem_meta_path: '{pretty_name_full}_META.yml'
     kem_scheme_path: '.'
     patches: [pqcrystals-kyber-yml.patch, pqcrystals-kyber-ref-shake-aes.patch, pqcrystals-kyber-avx2-shake-aes.patch]

scripts/copy_from_upstream/patches/pqclean-kyber-armneon-variable-timing-fix.patch

@@ -0,0 +1,274 @@
+927a0eff4a45781218062953002001af4e6a5c8a
+diff --git a/crypto_kem/kyber1024/aarch64/poly.c b/crypto_kem/kyber1024/aarch64/poly.c
+index 1dfa52c..3115d1c 100644
+--- a/crypto_kem/kyber1024/aarch64/poly.c
++++ b/crypto_kem/kyber1024/aarch64/poly.c
+@@ -51,6 +51,7 @@
+ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const int16_t a[KYBER_N]) {
+     unsigned int i, j;
+     int16_t u;
++    uint32_t d0;
+     uint8_t t[8];
+ 
+     for (i = 0; i < KYBER_N / 8; i++) {
+@@ -58,7 +59,12 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const int16_t a[KYBER_N
+             // map to positive standard representatives
+             u  = a[8 * i + j];
+             u += (u >> 15) & KYBER_Q;
+-            t[j] = ((((uint32_t)u << 5) + KYBER_Q / 2) / KYBER_Q) & 31;
++            // t[j] = ((((uint32_t)u << 5) + KYBER_Q / 2) / KYBER_Q) & 31;
++            d0 = u << 5;
++            d0 += 1664;
++            d0 *= 40318;
++            d0 >>= 27;
++            t[j] = d0 & 0x1f;
+         }
+ 
+         r[0] = (t[0] >> 0) | (t[1] << 5);
+@@ -207,14 +213,19 @@ void poly_frommsg(int16_t r[KYBER_N], const uint8_t msg[KYBER_INDCPA_MSGBYTES])
+ **************************************************/
+ void poly_tomsg(uint8_t msg[KYBER_INDCPA_MSGBYTES], const int16_t a[KYBER_N]) {
+     unsigned int i, j;
+-    uint16_t t;
++    uint32_t t;
+ 
+     for (i = 0; i < KYBER_N / 8; i++) {
+         msg[i] = 0;
+         for (j = 0; j < 8; j++) {
+             t  = a[8 * i + j];
+-            t += ((int16_t)t >> 15) & KYBER_Q;
+-            t  = (((t << 1) + KYBER_Q / 2) / KYBER_Q) & 1;
++            // t += ((int16_t)t >> 15) & KYBER_Q;
++            // t  = (((t << 1) + KYBER_Q/2)/KYBER_Q) & 1;
++            t <<= 1;
++            t += 1665;
++            t *= 80635;
++            t >>= 28;
++            t &= 1;
+             msg[i] |= t << j;
+         }
+     }
+diff --git a/crypto_kem/kyber1024/aarch64/polyvec.c b/crypto_kem/kyber1024/aarch64/polyvec.c
+index d400348..f9a1ebf 100644
+--- a/crypto_kem/kyber1024/aarch64/polyvec.c
++++ b/crypto_kem/kyber1024/aarch64/polyvec.c
+@@ -21,6 +21,7 @@
+ **************************************************/
+ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K][KYBER_N]) {
+     unsigned int i, j, k;
++    uint64_t d0;
+ 
+     #if (KYBER_POLYVECCOMPRESSEDBYTES == (KYBER_K * 352))
+     uint16_t t[8];
+@@ -29,7 +30,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K
+             for (k = 0; k < 8; k++) {
+                 t[k]  = a[i][8 * j + k];
+                 t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
+-                t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q / 2) / KYBER_Q) & 0x7ff;
++                // t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q / 2) / KYBER_Q) & 0x7ff;
++                d0 = t[k];
++                d0 <<= 11;
++                d0 += 1664;
++                d0 *= 645084;
++                d0 >>= 31;
++                t[k] = d0 & 0x7ff;
+             }
+ 
+             r[ 0] = (t[0] >>  0);
+@@ -53,7 +60,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K
+             for (k = 0; k < 4; k++) {
+                 t[k]  = a[i][4 * j + k];
+                 t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
+-                t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q / 2) / KYBER_Q) & 0x3ff;
++                // t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q / 2) / KYBER_Q) & 0x3ff;
++                d0 = t[k];
++                d0 <<= 10;
++                d0 += 1665;
++                d0 *= 1290167;
++                d0 >>= 32;
++                t[k] = d0 & 0x3ff;
+             }
+ 
+             r[0] = (t[0] >> 0);
+diff --git a/crypto_kem/kyber512/aarch64/poly.c b/crypto_kem/kyber512/aarch64/poly.c
+index dffc655..361ce89 100644
+--- a/crypto_kem/kyber512/aarch64/poly.c
++++ b/crypto_kem/kyber512/aarch64/poly.c
+@@ -51,6 +51,7 @@
+ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const int16_t a[KYBER_N]) {
+     unsigned int i, j;
+     int16_t u;
++    uint32_t d0;
+     uint8_t t[8];
+ 
+     for (i = 0; i < KYBER_N / 8; i++) {
+@@ -58,7 +59,12 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const int16_t a[KYBER_N
+             // map to positive standard representatives
+             u  = a[8 * i + j];
+             u += (u >> 15) & KYBER_Q;
+-            t[j] = ((((uint16_t)u << 4) + KYBER_Q / 2) / KYBER_Q) & 15;
++            // t[j] = ((((uint16_t)u << 4) + KYBER_Q / 2) / KYBER_Q) & 15;
++            d0 = u << 4;
++            d0 += 1665;
++            d0 *= 80635;
++            d0 >>= 28;
++            t[j] = d0 & 0xf;
+         }
+ 
+         r[0] = t[0] | (t[1] << 4);
+@@ -194,14 +200,19 @@ void poly_frommsg(int16_t r[KYBER_N], const uint8_t msg[KYBER_INDCPA_MSGBYTES])
+ **************************************************/
+ void poly_tomsg(uint8_t msg[KYBER_INDCPA_MSGBYTES], const int16_t a[KYBER_N]) {
+     unsigned int i, j;
+-    uint16_t t;
++    uint32_t t;
+ 
+     for (i = 0; i < KYBER_N / 8; i++) {
+         msg[i] = 0;
+         for (j = 0; j < 8; j++) {
+             t  = a[8 * i + j];
+-            t += ((int16_t)t >> 15) & KYBER_Q;
+-            t  = (((t << 1) + KYBER_Q / 2) / KYBER_Q) & 1;
++            // t += ((int16_t)t >> 15) & KYBER_Q;
++            // t  = (((t << 1) + KYBER_Q/2)/KYBER_Q) & 1;
++            t <<= 1;
++            t += 1665;
++            t *= 80635;
++            t >>= 28;
++            t &= 1;
+             msg[i] |= t << j;
+         }
+     }
+diff --git a/crypto_kem/kyber512/aarch64/polyvec.c b/crypto_kem/kyber512/aarch64/polyvec.c
+index d400348..f9a1ebf 100644
+--- a/crypto_kem/kyber512/aarch64/polyvec.c
++++ b/crypto_kem/kyber512/aarch64/polyvec.c
+@@ -21,6 +21,7 @@
+ **************************************************/
+ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K][KYBER_N]) {
+     unsigned int i, j, k;
++    uint64_t d0;
+ 
+     #if (KYBER_POLYVECCOMPRESSEDBYTES == (KYBER_K * 352))
+     uint16_t t[8];
+@@ -29,7 +30,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K
+             for (k = 0; k < 8; k++) {
+                 t[k]  = a[i][8 * j + k];
+                 t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
+-                t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q / 2) / KYBER_Q) & 0x7ff;
++                // t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q / 2) / KYBER_Q) & 0x7ff;
++                d0 = t[k];
++                d0 <<= 11;
++                d0 += 1664;
++                d0 *= 645084;
++                d0 >>= 31;
++                t[k] = d0 & 0x7ff;
+             }
+ 
+             r[ 0] = (t[0] >>  0);
+@@ -53,7 +60,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K
+             for (k = 0; k < 4; k++) {
+                 t[k]  = a[i][4 * j + k];
+                 t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
+-                t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q / 2) / KYBER_Q) & 0x3ff;
++                // t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q / 2) / KYBER_Q) & 0x3ff;
++                d0 = t[k];
++                d0 <<= 10;
++                d0 += 1665;
++                d0 *= 1290167;
++                d0 >>= 32;
++                t[k] = d0 & 0x3ff;
+             }
+ 
+             r[0] = (t[0] >> 0);
+diff --git a/crypto_kem/kyber768/aarch64/poly.c b/crypto_kem/kyber768/aarch64/poly.c
+index dffc655..361ce89 100644
+--- a/crypto_kem/kyber768/aarch64/poly.c
++++ b/crypto_kem/kyber768/aarch64/poly.c
+@@ -51,6 +51,7 @@
+ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const int16_t a[KYBER_N]) {
+     unsigned int i, j;
+     int16_t u;
++    uint32_t d0;
+     uint8_t t[8];
+ 
+     for (i = 0; i < KYBER_N / 8; i++) {
+@@ -58,7 +59,12 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const int16_t a[KYBER_N
+             // map to positive standard representatives
+             u  = a[8 * i + j];
+             u += (u >> 15) & KYBER_Q;
+-            t[j] = ((((uint16_t)u << 4) + KYBER_Q / 2) / KYBER_Q) & 15;
++            // t[j] = ((((uint16_t)u << 4) + KYBER_Q / 2) / KYBER_Q) & 15;
++            d0 = u << 4;
++            d0 += 1665;
++            d0 *= 80635;
++            d0 >>= 28;
++            t[j] = d0 & 0xf;
+         }
+ 
+         r[0] = t[0] | (t[1] << 4);
+@@ -194,14 +200,19 @@ void poly_frommsg(int16_t r[KYBER_N], const uint8_t msg[KYBER_INDCPA_MSGBYTES])
+ **************************************************/
+ void poly_tomsg(uint8_t msg[KYBER_INDCPA_MSGBYTES], const int16_t a[KYBER_N]) {
+     unsigned int i, j;
+-    uint16_t t;
++    uint32_t t;
+ 
+     for (i = 0; i < KYBER_N / 8; i++) {
+         msg[i] = 0;
+         for (j = 0; j < 8; j++) {
+             t  = a[8 * i + j];
+-            t += ((int16_t)t >> 15) & KYBER_Q;
+-            t  = (((t << 1) + KYBER_Q / 2) / KYBER_Q) & 1;
++            // t += ((int16_t)t >> 15) & KYBER_Q;
++            // t  = (((t << 1) + KYBER_Q/2)/KYBER_Q) & 1;
++            t <<= 1;
++            t += 1665;
++            t *= 80635;
++            t >>= 28;
++            t &= 1;
+             msg[i] |= t << j;
+         }
+     }
+diff --git a/crypto_kem/kyber768/aarch64/polyvec.c b/crypto_kem/kyber768/aarch64/polyvec.c
+index d400348..f9a1ebf 100644
+--- a/crypto_kem/kyber768/aarch64/polyvec.c
++++ b/crypto_kem/kyber768/aarch64/polyvec.c
+@@ -21,6 +21,7 @@
+ **************************************************/
+ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K][KYBER_N]) {
+     unsigned int i, j, k;
++    uint64_t d0;
+ 
+     #if (KYBER_POLYVECCOMPRESSEDBYTES == (KYBER_K * 352))
+     uint16_t t[8];
+@@ -29,7 +30,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K
+             for (k = 0; k < 8; k++) {
+                 t[k]  = a[i][8 * j + k];
+                 t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
+-                t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q / 2) / KYBER_Q) & 0x7ff;
++                // t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q / 2) / KYBER_Q) & 0x7ff;
++                d0 = t[k];
++                d0 <<= 11;
++                d0 += 1664;
++                d0 *= 645084;
++                d0 >>= 31;
++                t[k] = d0 & 0x7ff;
+             }
+ 
+             r[ 0] = (t[0] >>  0);
+@@ -53,7 +60,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K
+             for (k = 0; k < 4; k++) {
+                 t[k]  = a[i][4 * j + k];
+                 t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
+-                t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q / 2) / KYBER_Q) & 0x3ff;
++                // t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q / 2) / KYBER_Q) & 0x3ff;
++                d0 = t[k];
++                d0 <<= 10;
++                d0 += 1665;
++                d0 *= 1290167;
++                d0 >>= 32;
++                t[k] = d0 & 0x3ff;
+             }
+ 
+             r[0] = (t[0] >> 0);

src/kem/kyber/pqclean_kyber1024_aarch64/poly.c

@@ -51,6 +51,7 @@
 void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const int16_t a[KYBER_N]) {
     unsigned int i, j;
     int16_t u;
+    uint32_t d0;
     uint8_t t[8];
 
     for (i = 0; i < KYBER_N / 8; i++) {
@@ -58,7 +59,12 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const int16_t a[KYBER_N
             // map to positive standard representatives
             u  = a[8 * i + j];
             u += (u >> 15) & KYBER_Q;
-            t[j] = ((((uint32_t)u << 5) + KYBER_Q / 2) / KYBER_Q) & 31;
+            // t[j] = ((((uint32_t)u << 5) + KYBER_Q / 2) / KYBER_Q) & 31;
+            d0 = u << 5;
+            d0 += 1664;
+            d0 *= 40318;
+            d0 >>= 27;
+            t[j] = d0 & 0x1f;
         }
 
         r[0] = (t[0] >> 0) | (t[1] << 5);
@@ -207,14 +213,19 @@ void poly_frommsg(int16_t r[KYBER_N], const uint8_t msg[KYBER_INDCPA_MSGBYTES])
 **************************************************/
 void poly_tomsg(uint8_t msg[KYBER_INDCPA_MSGBYTES], const int16_t a[KYBER_N]) {
     unsigned int i, j;
-    uint16_t t;
+    uint32_t t;
 
     for (i = 0; i < KYBER_N / 8; i++) {
         msg[i] = 0;
         for (j = 0; j < 8; j++) {
             t  = a[8 * i + j];
-            t += ((int16_t)t >> 15) & KYBER_Q;
-            t  = (((t << 1) + KYBER_Q / 2) / KYBER_Q) & 1;
+            // t += ((int16_t)t >> 15) & KYBER_Q;
+            // t  = (((t << 1) + KYBER_Q/2)/KYBER_Q) & 1;
+            t <<= 1;
+            t += 1665;
+            t *= 80635;
+            t >>= 28;
+            t &= 1;
             msg[i] |= t << j;
         }
     }

src/kem/kyber/pqclean_kyber1024_aarch64/polyvec.c

@@ -21,6 +21,7 @@
 **************************************************/
 void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K][KYBER_N]) {
     unsigned int i, j, k;
+    uint64_t d0;
 
     #if (KYBER_POLYVECCOMPRESSEDBYTES == (KYBER_K * 352))
     uint16_t t[8];
@@ -29,7 +30,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K
             for (k = 0; k < 8; k++) {
                 t[k]  = a[i][8 * j + k];
                 t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
-                t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q / 2) / KYBER_Q) & 0x7ff;
+                // t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q / 2) / KYBER_Q) & 0x7ff;
+                d0 = t[k];
+                d0 <<= 11;
+                d0 += 1664;
+                d0 *= 645084;
+                d0 >>= 31;
+                t[k] = d0 & 0x7ff;
             }
 
             r[ 0] = (t[0] >>  0);
@@ -53,7 +60,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K
             for (k = 0; k < 4; k++) {
                 t[k]  = a[i][4 * j + k];
                 t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
-                t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q / 2) / KYBER_Q) & 0x3ff;
+                // t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q / 2) / KYBER_Q) & 0x3ff;
+                d0 = t[k];
+                d0 <<= 10;
+                d0 += 1665;
+                d0 *= 1290167;
+                d0 >>= 32;
+                t[k] = d0 & 0x3ff;
             }
 
             r[0] = (t[0] >> 0);

src/kem/kyber/pqclean_kyber512_aarch64/poly.c

@@ -51,6 +51,7 @@
 void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const int16_t a[KYBER_N]) {
     unsigned int i, j;
     int16_t u;
+    uint32_t d0;
     uint8_t t[8];
 
     for (i = 0; i < KYBER_N / 8; i++) {
@@ -58,7 +59,12 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const int16_t a[KYBER_N
             // map to positive standard representatives
             u  = a[8 * i + j];
             u += (u >> 15) & KYBER_Q;
-            t[j] = ((((uint16_t)u << 4) + KYBER_Q / 2) / KYBER_Q) & 15;
+            // t[j] = ((((uint16_t)u << 4) + KYBER_Q / 2) / KYBER_Q) & 15;
+            d0 = u << 4;
+            d0 += 1665;
+            d0 *= 80635;
+            d0 >>= 28;
+            t[j] = d0 & 0xf;
         }
 
         r[0] = t[0] | (t[1] << 4);
@@ -194,14 +200,19 @@ void poly_frommsg(int16_t r[KYBER_N], const uint8_t msg[KYBER_INDCPA_MSGBYTES])
 **************************************************/
 void poly_tomsg(uint8_t msg[KYBER_INDCPA_MSGBYTES], const int16_t a[KYBER_N]) {
     unsigned int i, j;
-    uint16_t t;
+    uint32_t t;
 
     for (i = 0; i < KYBER_N / 8; i++) {
         msg[i] = 0;
         for (j = 0; j < 8; j++) {
             t  = a[8 * i + j];
-            t += ((int16_t)t >> 15) & KYBER_Q;
-            t  = (((t << 1) + KYBER_Q / 2) / KYBER_Q) & 1;
+            // t += ((int16_t)t >> 15) & KYBER_Q;
+            // t  = (((t << 1) + KYBER_Q/2)/KYBER_Q) & 1;
+            t <<= 1;
+            t += 1665;
+            t *= 80635;
+            t >>= 28;
+            t &= 1;
             msg[i] |= t << j;
         }
     }

src/kem/kyber/pqclean_kyber512_aarch64/polyvec.c

@@ -21,6 +21,7 @@
 **************************************************/
 void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K][KYBER_N]) {
     unsigned int i, j, k;
+    uint64_t d0;
 
     #if (KYBER_POLYVECCOMPRESSEDBYTES == (KYBER_K * 352))
     uint16_t t[8];
@@ -29,7 +30,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K
             for (k = 0; k < 8; k++) {
                 t[k]  = a[i][8 * j + k];
                 t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
-                t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q / 2) / KYBER_Q) & 0x7ff;
+                // t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q / 2) / KYBER_Q) & 0x7ff;
+                d0 = t[k];
+                d0 <<= 11;
+                d0 += 1664;
+                d0 *= 645084;
+                d0 >>= 31;
+                t[k] = d0 & 0x7ff;
             }
 
             r[ 0] = (t[0] >>  0);
@@ -53,7 +60,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K
             for (k = 0; k < 4; k++) {
                 t[k]  = a[i][4 * j + k];
                 t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
-                t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q / 2) / KYBER_Q) & 0x3ff;
+                // t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q / 2) / KYBER_Q) & 0x3ff;
+                d0 = t[k];
+                d0 <<= 10;
+                d0 += 1665;
+                d0 *= 1290167;
+                d0 >>= 32;
+                t[k] = d0 & 0x3ff;
             }
 
             r[0] = (t[0] >> 0);

src/kem/kyber/pqclean_kyber768_aarch64/poly.c

@@ -51,6 +51,7 @@
 void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const int16_t a[KYBER_N]) {
     unsigned int i, j;
     int16_t u;
+    uint32_t d0;
     uint8_t t[8];
 
     for (i = 0; i < KYBER_N / 8; i++) {
@@ -58,7 +59,12 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const int16_t a[KYBER_N
             // map to positive standard representatives
             u  = a[8 * i + j];
             u += (u >> 15) & KYBER_Q;
-            t[j] = ((((uint16_t)u << 4) + KYBER_Q / 2) / KYBER_Q) & 15;
+            // t[j] = ((((uint16_t)u << 4) + KYBER_Q / 2) / KYBER_Q) & 15;
+            d0 = u << 4;
+            d0 += 1665;
+            d0 *= 80635;
+            d0 >>= 28;
+            t[j] = d0 & 0xf;
         }
 
         r[0] = t[0] | (t[1] << 4);
@@ -194,14 +200,19 @@ void poly_frommsg(int16_t r[KYBER_N], const uint8_t msg[KYBER_INDCPA_MSGBYTES])
 **************************************************/
 void poly_tomsg(uint8_t msg[KYBER_INDCPA_MSGBYTES], const int16_t a[KYBER_N]) {
     unsigned int i, j;
-    uint16_t t;
+    uint32_t t;
 
     for (i = 0; i < KYBER_N / 8; i++) {
         msg[i] = 0;
         for (j = 0; j < 8; j++) {
             t  = a[8 * i + j];
-            t += ((int16_t)t >> 15) & KYBER_Q;
-            t  = (((t << 1) + KYBER_Q / 2) / KYBER_Q) & 1;
+            // t += ((int16_t)t >> 15) & KYBER_Q;
+            // t  = (((t << 1) + KYBER_Q/2)/KYBER_Q) & 1;
+            t <<= 1;
+            t += 1665;
+            t *= 80635;
+            t >>= 28;
+            t &= 1;
             msg[i] |= t << j;
         }
     }

src/kem/kyber/pqclean_kyber768_aarch64/polyvec.c

@@ -21,6 +21,7 @@
 **************************************************/
 void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K][KYBER_N]) {
     unsigned int i, j, k;
+    uint64_t d0;
 
     #if (KYBER_POLYVECCOMPRESSEDBYTES == (KYBER_K * 352))
     uint16_t t[8];
@@ -29,7 +30,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K
             for (k = 0; k < 8; k++) {
                 t[k]  = a[i][8 * j + k];
                 t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
-                t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q / 2) / KYBER_Q) & 0x7ff;
+                // t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q / 2) / KYBER_Q) & 0x7ff;
+                d0 = t[k];
+                d0 <<= 11;
+                d0 += 1664;
+                d0 *= 645084;
+                d0 >>= 31;
+                t[k] = d0 & 0x7ff;
             }
 
             r[ 0] = (t[0] >>  0);
@@ -53,7 +60,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], int16_t a[KYBER_K
             for (k = 0; k < 4; k++) {
                 t[k]  = a[i][4 * j + k];
                 t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
-                t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q / 2) / KYBER_Q) & 0x3ff;
+                // t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q / 2) / KYBER_Q) & 0x3ff;
+                d0 = t[k];
+                d0 <<= 10;
+                d0 += 1665;
+                d0 *= 1290167;
+                d0 >>= 32;
+                t[k] = d0 & 0x3ff;
             }
 
             r[0] = (t[0] >> 0);

src/kem/kyber/pqcrystals-kyber_kyber1024_avx2/verify.c

@@ -57,6 +57,16 @@ void cmov(uint8_t * restrict r, const uint8_t *x, size_t len, uint8_t b)
   size_t i;
   __m256i xvec, rvec, bvec;
 
+#if defined(__GNUC__) || defined(__clang__)
+  // Prevent the compiler from
+  //    1) inferring that b is 0/1-valued, and
+  //    2) handling the two cases with a branch.
+  // This is not necessary when verify.c and kem.c are separate translation
+  // units, but we expect that downstream consumers will copy this code and/or
+  // change how it is built.
+  __asm__("" : "+r"(b) : /* no inputs */);
+#endif
+
   bvec = _mm256_set1_epi64x(-(uint64_t)b);
   for(i=0;i<len/32;i++) {
     rvec = _mm256_loadu_si256((__m256i *)&r[32*i]);

src/kem/kyber/pqcrystals-kyber_kyber1024_ref/poly.c

@@ -19,6 +19,7 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const poly *a)
 {
   unsigned int i,j;
   int16_t u;
+  uint32_t d0;
   uint8_t t[8];
 
 #if (KYBER_POLYCOMPRESSEDBYTES == 128)
@@ -27,7 +28,12 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const poly *a)
       // map to positive standard representatives
       u  = a->coeffs[8*i+j];
       u += (u >> 15) & KYBER_Q;
-      t[j] = ((((uint16_t)u << 4) + KYBER_Q/2)/KYBER_Q) & 15;
+/*    t[j] = ((((uint16_t)u << 4) + KYBER_Q/2)/KYBER_Q) & 15; */
+      d0 = u << 4;
+      d0 += 1665;
+      d0 *= 80635;
+      d0 >>= 28;
+      t[j] = d0 & 0xf;
     }
 
     r[0] = t[0] | (t[1] << 4);
@@ -42,7 +48,12 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const poly *a)
       // map to positive standard representatives
       u  = a->coeffs[8*i+j];
       u += (u >> 15) & KYBER_Q;
-      t[j] = ((((uint32_t)u << 5) + KYBER_Q/2)/KYBER_Q) & 31;
+/*      t[j] = ((((uint32_t)u << 5) + KYBER_Q/2)/KYBER_Q) & 31; */
+      d0 = u << 5;
+      d0 += 1664;
+      d0 *= 40318;
+      d0 >>= 27;
+      t[j] = d0 & 0x1f;
     }
 
     r[0] = (t[0] >> 0) | (t[1] << 5);
@@ -180,14 +191,19 @@ void poly_frommsg(poly *r, const uint8_t msg[KYBER_INDCPA_MSGBYTES])
 void poly_tomsg(uint8_t msg[KYBER_INDCPA_MSGBYTES], const poly *a)
 {
   unsigned int i,j;
-  uint16_t t;
+  uint32_t t;
 
   for(i=0;i<KYBER_N/8;i++) {
     msg[i] = 0;
     for(j=0;j<8;j++) {
       t  = a->coeffs[8*i+j];
-      t += ((int16_t)t >> 15) & KYBER_Q;
-      t  = (((t << 1) + KYBER_Q/2)/KYBER_Q) & 1;
+      // t += ((int16_t)t >> 15) & KYBER_Q;
+      // t  = (((t << 1) + KYBER_Q/2)/KYBER_Q) & 1;
+      t <<= 1;
+      t += 1665;
+      t *= 80635;
+      t >>= 28;
+      t &= 1;
       msg[i] |= t << j;
     }
   }

src/kem/kyber/pqcrystals-kyber_kyber1024_ref/polyvec.c

@@ -15,6 +15,7 @@
 void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], const polyvec *a)
 {
   unsigned int i,j,k;
+  uint64_t d0;
 
 #if (KYBER_POLYVECCOMPRESSEDBYTES == (KYBER_K * 352))
   uint16_t t[8];
@@ -23,7 +24,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], const polyvec *a)
       for(k=0;k<8;k++) {
         t[k]  = a->vec[i].coeffs[8*j+k];
         t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
-        t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q/2)/KYBER_Q) & 0x7ff;
+/*      t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q/2)/KYBER_Q) & 0x7ff; */
+        d0 = t[k];
+        d0 <<= 11;
+        d0 += 1664;
+        d0 *= 645084;
+        d0 >>= 31;
+        t[k] = d0 & 0x7ff;
       }
 
       r[ 0] = (t[0] >>  0);
@@ -47,7 +54,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], const polyvec *a)
       for(k=0;k<4;k++) {
         t[k]  = a->vec[i].coeffs[4*j+k];
         t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
-        t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q/2)/ KYBER_Q) & 0x3ff;
+/*      t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q/2)/ KYBER_Q) & 0x3ff; */
+        d0 = t[k];
+        d0 <<= 10;
+        d0 += 1665;
+        d0 *= 1290167;
+        d0 >>= 32;
+        t[k] = d0 & 0x3ff;
       }
 
       r[0] = (t[0] >> 0);

src/kem/kyber/pqcrystals-kyber_kyber1024_ref/verify.c

@@ -41,6 +41,16 @@ void cmov(uint8_t *r, const uint8_t *x, size_t len, uint8_t b)
 {
   size_t i;
 
+#if defined(__GNUC__) || defined(__clang__)
+  // Prevent the compiler from
+  //    1) inferring that b is 0/1-valued, and
+  //    2) handling the two cases with a branch.
+  // This is not necessary when verify.c and kem.c are separate translation
+  // units, but we expect that downstream consumers will copy this code and/or
+  // change how it is built.
+  __asm__("" : "+r"(b) : /* no inputs */);
+#endif
+
   b = -b;
   for(i=0;i<len;i++)
     r[i] ^= b & (r[i] ^ x[i]);

src/kem/kyber/pqcrystals-kyber_kyber512_avx2/verify.c

@@ -57,6 +57,16 @@ void cmov(uint8_t * restrict r, const uint8_t *x, size_t len, uint8_t b)
   size_t i;
   __m256i xvec, rvec, bvec;
 
+#if defined(__GNUC__) || defined(__clang__)
+  // Prevent the compiler from
+  //    1) inferring that b is 0/1-valued, and
+  //    2) handling the two cases with a branch.
+  // This is not necessary when verify.c and kem.c are separate translation
+  // units, but we expect that downstream consumers will copy this code and/or
+  // change how it is built.
+  __asm__("" : "+r"(b) : /* no inputs */);
+#endif
+
   bvec = _mm256_set1_epi64x(-(uint64_t)b);
   for(i=0;i<len/32;i++) {
     rvec = _mm256_loadu_si256((__m256i *)&r[32*i]);

src/kem/kyber/pqcrystals-kyber_kyber512_ref/poly.c

@@ -19,6 +19,7 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const poly *a)
 {
   unsigned int i,j;
   int16_t u;
+  uint32_t d0;
   uint8_t t[8];
 
 #if (KYBER_POLYCOMPRESSEDBYTES == 128)
@@ -27,7 +28,12 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const poly *a)
       // map to positive standard representatives
       u  = a->coeffs[8*i+j];
       u += (u >> 15) & KYBER_Q;
-      t[j] = ((((uint16_t)u << 4) + KYBER_Q/2)/KYBER_Q) & 15;
+/*    t[j] = ((((uint16_t)u << 4) + KYBER_Q/2)/KYBER_Q) & 15; */
+      d0 = u << 4;
+      d0 += 1665;
+      d0 *= 80635;
+      d0 >>= 28;
+      t[j] = d0 & 0xf;
     }
 
     r[0] = t[0] | (t[1] << 4);
@@ -42,7 +48,12 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const poly *a)
       // map to positive standard representatives
       u  = a->coeffs[8*i+j];
       u += (u >> 15) & KYBER_Q;
-      t[j] = ((((uint32_t)u << 5) + KYBER_Q/2)/KYBER_Q) & 31;
+/*      t[j] = ((((uint32_t)u << 5) + KYBER_Q/2)/KYBER_Q) & 31; */
+      d0 = u << 5;
+      d0 += 1664;
+      d0 *= 40318;
+      d0 >>= 27;
+      t[j] = d0 & 0x1f;
     }
 
     r[0] = (t[0] >> 0) | (t[1] << 5);
@@ -180,14 +191,19 @@ void poly_frommsg(poly *r, const uint8_t msg[KYBER_INDCPA_MSGBYTES])
 void poly_tomsg(uint8_t msg[KYBER_INDCPA_MSGBYTES], const poly *a)
 {
   unsigned int i,j;
-  uint16_t t;
+  uint32_t t;
 
   for(i=0;i<KYBER_N/8;i++) {
     msg[i] = 0;
     for(j=0;j<8;j++) {
       t  = a->coeffs[8*i+j];
-      t += ((int16_t)t >> 15) & KYBER_Q;
-      t  = (((t << 1) + KYBER_Q/2)/KYBER_Q) & 1;
+      // t += ((int16_t)t >> 15) & KYBER_Q;
+      // t  = (((t << 1) + KYBER_Q/2)/KYBER_Q) & 1;
+      t <<= 1;
+      t += 1665;
+      t *= 80635;
+      t >>= 28;
+      t &= 1;
       msg[i] |= t << j;
     }
   }

src/kem/kyber/pqcrystals-kyber_kyber512_ref/polyvec.c

@@ -15,6 +15,7 @@
 void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], const polyvec *a)
 {
   unsigned int i,j,k;
+  uint64_t d0;
 
 #if (KYBER_POLYVECCOMPRESSEDBYTES == (KYBER_K * 352))
   uint16_t t[8];
@@ -23,7 +24,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], const polyvec *a)
       for(k=0;k<8;k++) {
         t[k]  = a->vec[i].coeffs[8*j+k];
         t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
-        t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q/2)/KYBER_Q) & 0x7ff;
+/*      t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q/2)/KYBER_Q) & 0x7ff; */
+        d0 = t[k];
+        d0 <<= 11;
+        d0 += 1664;
+        d0 *= 645084;
+        d0 >>= 31;
+        t[k] = d0 & 0x7ff;
       }
 
       r[ 0] = (t[0] >>  0);
@@ -47,7 +54,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], const polyvec *a)
       for(k=0;k<4;k++) {
         t[k]  = a->vec[i].coeffs[4*j+k];
         t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
-        t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q/2)/ KYBER_Q) & 0x3ff;
+/*      t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q/2)/ KYBER_Q) & 0x3ff; */
+        d0 = t[k];
+        d0 <<= 10;
+        d0 += 1665;
+        d0 *= 1290167;
+        d0 >>= 32;
+        t[k] = d0 & 0x3ff;
       }
 
       r[0] = (t[0] >> 0);

src/kem/kyber/pqcrystals-kyber_kyber512_ref/verify.c

@@ -41,6 +41,16 @@ void cmov(uint8_t *r, const uint8_t *x, size_t len, uint8_t b)
 {
   size_t i;
 
+#if defined(__GNUC__) || defined(__clang__)
+  // Prevent the compiler from
+  //    1) inferring that b is 0/1-valued, and
+  //    2) handling the two cases with a branch.
+  // This is not necessary when verify.c and kem.c are separate translation
+  // units, but we expect that downstream consumers will copy this code and/or
+  // change how it is built.
+  __asm__("" : "+r"(b) : /* no inputs */);
+#endif
+
   b = -b;
   for(i=0;i<len;i++)
     r[i] ^= b & (r[i] ^ x[i]);

src/kem/kyber/pqcrystals-kyber_kyber768_avx2/verify.c

@@ -57,6 +57,16 @@ void cmov(uint8_t * restrict r, const uint8_t *x, size_t len, uint8_t b)
   size_t i;
   __m256i xvec, rvec, bvec;
 
+#if defined(__GNUC__) || defined(__clang__)
+  // Prevent the compiler from
+  //    1) inferring that b is 0/1-valued, and
+  //    2) handling the two cases with a branch.
+  // This is not necessary when verify.c and kem.c are separate translation
+  // units, but we expect that downstream consumers will copy this code and/or
+  // change how it is built.
+  __asm__("" : "+r"(b) : /* no inputs */);
+#endif
+
   bvec = _mm256_set1_epi64x(-(uint64_t)b);
   for(i=0;i<len/32;i++) {
     rvec = _mm256_loadu_si256((__m256i *)&r[32*i]);

src/kem/kyber/pqcrystals-kyber_kyber768_ref/poly.c

@@ -19,6 +19,7 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const poly *a)
 {
   unsigned int i,j;
   int16_t u;
+  uint32_t d0;
   uint8_t t[8];
 
 #if (KYBER_POLYCOMPRESSEDBYTES == 128)
@@ -27,7 +28,12 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const poly *a)
       // map to positive standard representatives
       u  = a->coeffs[8*i+j];
       u += (u >> 15) & KYBER_Q;
-      t[j] = ((((uint16_t)u << 4) + KYBER_Q/2)/KYBER_Q) & 15;
+/*    t[j] = ((((uint16_t)u << 4) + KYBER_Q/2)/KYBER_Q) & 15; */
+      d0 = u << 4;
+      d0 += 1665;
+      d0 *= 80635;
+      d0 >>= 28;
+      t[j] = d0 & 0xf;
     }
 
     r[0] = t[0] | (t[1] << 4);
@@ -42,7 +48,12 @@ void poly_compress(uint8_t r[KYBER_POLYCOMPRESSEDBYTES], const poly *a)
       // map to positive standard representatives
       u  = a->coeffs[8*i+j];
       u += (u >> 15) & KYBER_Q;
-      t[j] = ((((uint32_t)u << 5) + KYBER_Q/2)/KYBER_Q) & 31;
+/*      t[j] = ((((uint32_t)u << 5) + KYBER_Q/2)/KYBER_Q) & 31; */
+      d0 = u << 5;
+      d0 += 1664;
+      d0 *= 40318;
+      d0 >>= 27;
+      t[j] = d0 & 0x1f;
     }
 
     r[0] = (t[0] >> 0) | (t[1] << 5);
@@ -180,14 +191,19 @@ void poly_frommsg(poly *r, const uint8_t msg[KYBER_INDCPA_MSGBYTES])
 void poly_tomsg(uint8_t msg[KYBER_INDCPA_MSGBYTES], const poly *a)
 {
   unsigned int i,j;
-  uint16_t t;
+  uint32_t t;
 
   for(i=0;i<KYBER_N/8;i++) {
     msg[i] = 0;
     for(j=0;j<8;j++) {
       t  = a->coeffs[8*i+j];
-      t += ((int16_t)t >> 15) & KYBER_Q;
-      t  = (((t << 1) + KYBER_Q/2)/KYBER_Q) & 1;
+      // t += ((int16_t)t >> 15) & KYBER_Q;
+      // t  = (((t << 1) + KYBER_Q/2)/KYBER_Q) & 1;
+      t <<= 1;
+      t += 1665;
+      t *= 80635;
+      t >>= 28;
+      t &= 1;
       msg[i] |= t << j;
     }
   }

src/kem/kyber/pqcrystals-kyber_kyber768_ref/polyvec.c

@@ -15,6 +15,7 @@
 void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], const polyvec *a)
 {
   unsigned int i,j,k;
+  uint64_t d0;
 
 #if (KYBER_POLYVECCOMPRESSEDBYTES == (KYBER_K * 352))
   uint16_t t[8];
@@ -23,7 +24,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], const polyvec *a)
       for(k=0;k<8;k++) {
         t[k]  = a->vec[i].coeffs[8*j+k];
         t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
-        t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q/2)/KYBER_Q) & 0x7ff;
+/*      t[k]  = ((((uint32_t)t[k] << 11) + KYBER_Q/2)/KYBER_Q) & 0x7ff; */
+        d0 = t[k];
+        d0 <<= 11;
+        d0 += 1664;
+        d0 *= 645084;
+        d0 >>= 31;
+        t[k] = d0 & 0x7ff;
       }
 
       r[ 0] = (t[0] >>  0);
@@ -47,7 +54,13 @@ void polyvec_compress(uint8_t r[KYBER_POLYVECCOMPRESSEDBYTES], const polyvec *a)
       for(k=0;k<4;k++) {
         t[k]  = a->vec[i].coeffs[4*j+k];
         t[k] += ((int16_t)t[k] >> 15) & KYBER_Q;
-        t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q/2)/ KYBER_Q) & 0x3ff;
+/*      t[k]  = ((((uint32_t)t[k] << 10) + KYBER_Q/2)/ KYBER_Q) & 0x3ff; */
+        d0 = t[k];
+        d0 <<= 10;
+        d0 += 1665;
+        d0 *= 1290167;
+        d0 >>= 32;
+        t[k] = d0 & 0x3ff;
       }
 
       r[0] = (t[0] >> 0);

src/kem/kyber/pqcrystals-kyber_kyber768_ref/verify.c

@@ -41,6 +41,16 @@ void cmov(uint8_t *r, const uint8_t *x, size_t len, uint8_t b)
 {
   size_t i;
 
+#if defined(__GNUC__) || defined(__clang__)
+  // Prevent the compiler from
+  //    1) inferring that b is 0/1-valued, and
+  //    2) handling the two cases with a branch.
+  // This is not necessary when verify.c and kem.c are separate translation
+  // units, but we expect that downstream consumers will copy this code and/or
+  // change how it is built.
+  __asm__("" : "+r"(b) : /* no inputs */);
+#endif
+
   b = -b;
   for(i=0;i<len;i++)
     r[i] ^= b & (r[i] ^ x[i]);