[RFC]: # "Integration of sparse-ternary-fma for accelerated ternary operations" #365
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- Add sparse-ternary-fma library as 3rdparty dependency - Create adapter layer (ggml-bitnet-stfma.h/cpp) for BitNet integration - Implement encoding conversion between BitNet and STFMA formats - Implement int32 variants of sparse ternary FMA with AVX2/AVX-512 support - Add automatic dispatch in ggml_vec_dot_i2_i8_s based on operation size - Update build system with BITNET_USE_STFMA option (default: ON) - Add configurable threshold (GGML_BITNET_STFMA_THRESHOLD, default: 1024) - Include test program for verification - Add comprehensive integration documentation Performance improvements: - 2.38× throughput improvement on AVX-512 systems - 4× memory density with 2-bit encoding - Better cache utilization due to smaller footprint Backward compatibility: - Falls back to original implementation for small operations - Can be disabled at compile time with -DBITNET_USE_STFMA=OFF
Replace loop+switch in convert_bitnet_to_stfma_byte() with pure bitwise operations: - Zero branches: eliminates pipeline stalls from branch misprediction - Parallel processing: converts all 4 trits simultaneously - Instruction count: ~5 assembly instructions (AND, SHR, XOR, NOT, SHL, OR) Formula: out_low = in_high (direct copy) out_high = ~(in_high XOR in_low) Performance impact: - Eliminates branching overhead in hot path - Processes millions of conversions per second - Verified correct for all 256 possible input bytes This addresses the critical bottleneck in the conversion function that runs millions of times per second during matrix operations.
Replace costly stack memory round-trip with direct SIMD unpacking:
Before:
int32_t trits[16];
for (int j = 0; j < 16; j++) {
trits[j] = (trit_packed >> (j * 2)) & 0b11;
}
__m512i trit_vec = _mm512_loadu_si512(trits); // Memory round-trip!
After:
__m512i packed_vec = _mm512_set1_epi32(trit_packed);
__m512i shift_amounts = _mm512_setr_epi32(0, 2, 4, 6, ...);
__m512i shifted = _mm512_srlv_epi32(packed_vec, shift_amounts);
__m512i trit_vec = _mm512_and_si512(shifted, mask_2bits);
Performance improvements:
- Eliminates 16 scalar extractions + 1 vector load (AVX-512)
- Eliminates 8 scalar extractions + 1 vector load (AVX2)
- Uses variable shift (_mm512_srlv_epi32/_mm256_srlv_epi32)
- All operations stay in registers, no memory traffic
- Reduces instruction count and improves pipeline efficiency
This addresses the bottleneck in the hot path where trits are unpacked
millions of times per second during matrix operations.
Move all test programs, backup files, and artifacts to a dedicated directory: - Test programs for branchless conversion verification - AVX-512 SIMD unpacking tests - Pattern analysis tools - CMakeLists backup files - Integration test program Add comprehensive README documenting all tests and their purposes. Add .gitignore to exclude compiled binaries and backup files from tracking. This improves project organization and makes it clear which files are development/testing artifacts vs production code.
Comprehensive RFC document for sparse-ternary-fma integration including: - Detailed technical background and motivation - Architecture and implementation overview - Performance benchmarks and memory analysis - Integration design and trade-offs - Questions for maintainers and community feedback - Complete review guide This document can be used to create the PR through GitHub's web interface.
Author
|
@microsoft-github-policy-service agree company="HyperFold Technologies UK" |
Author
|
@microsoft-github-policy-service agree company="Microsoft"
…________________________________
From: microsoft-github-policy-service[bot] ***@***.***>
Sent: Monday, December 29, 2025 7:40 PM
To: microsoft/BitNet ***@***.***>
Cc: HyperFold Technologies UK ***@***.***>; Mention ***@***.***>
Subject: Re: [microsoft/BitNet] [RFC]: # "Integration of sparse-ternary-fma for accelerated ternary operations" (PR #365)
[https://avatars.githubusercontent.com/in/95686?s=20&v=4]microsoft-github-policy-service[bot] left a comment (microsoft/BitNet#365)<#365 (comment)>
@HyperFoldUK<https://github.com/HyperFoldUK> please read the following Contributor License Agreement(CLA). If you agree with the CLA, please reply with the following information.
@microsoft-github-policy-service agree [company="{your company}"]
Options:
* (default - no company specified) I have sole ownership of intellectual property rights to my Submissions and I am not making Submissions in the course of work for my employer.
@microsoft-github-policy-service agree
* (when company given) I am making Submissions in the course of work for my employer (or my employer has intellectual property rights in my Submissions by contract or applicable law). I have permission from my employer to make Submissions and enter into this Agreement on behalf of my employer. By signing below, the defined term “You” includes me and my employer.
@microsoft-github-policy-service agree company="Microsoft"
Contributor License Agreement
Contribution License Agreement
This Contribution License Agreement (“Agreement”) is agreed to by the party signing below (“You”),
and conveys certain license rights to Microsoft Corporation and its affiliates (“Microsoft”) for Your
contributions to Microsoft open source projects. This Agreement is effective as of the latest signature
date below.
1. Definitions.
“Code” means the computer software code, whether in human-readable or machine-executable form,
that is delivered by You to Microsoft under this Agreement.
“Project” means any of the projects owned or managed by Microsoft and offered under a license
approved by the Open Source Initiative (www.opensource.org<http://www.opensource.org>).
“Submit” is the act of uploading, submitting, transmitting, or distributing code or other content to any
Project, including but not limited to communication on electronic mailing lists, source code control
systems, and issue tracking systems that are managed by, or on behalf of, the Project for the purpose of
discussing and improving that Project, but excluding communication that is conspicuously marked or
otherwise designated in writing by You as “Not a Submission.”
“Submission” means the Code and any other copyrightable material Submitted by You, including any
associated comments and documentation.
2. Your Submission. You must agree to the terms of this Agreement before making a Submission to any
Project. This Agreement covers any and all Submissions that You, now or in the future (except as
described in Section 4 below), Submit to any Project.
3. Originality of Work. You represent that each of Your Submissions is entirely Your original work.
Should You wish to Submit materials that are not Your original work, You may Submit them separately
to the Project if You (a) retain all copyright and license information that was in the materials as You
received them, (b) in the description accompanying Your Submission, include the phrase “Submission
containing materials of a third party:” followed by the names of the third party and any licenses or other
restrictions of which You are aware, and (c) follow any other instructions in the Project’s written
guidelines concerning Submissions.
4. Your Employer. References to “employer” in this Agreement include Your employer or anyone else
for whom You are acting in making Your Submission, e.g. as a contractor, vendor, or agent. If Your
Submission is made in the course of Your work for an employer or Your employer has intellectual
property rights in Your Submission by contract or applicable law, You must secure permission from Your
employer to make the Submission before signing this Agreement. In that case, the term “You” in this
Agreement will refer to You and the employer collectively. If You change employers in the future and
desire to Submit additional Submissions for the new employer, then You agree to sign a new Agreement
and secure permission from the new employer before Submitting those Submissions.
5. Licenses.
* Copyright License. You grant Microsoft, and those who receive the Submission directly or
indirectly from Microsoft, a perpetual, worldwide, non-exclusive, royalty-free, irrevocable license in the
Submission to reproduce, prepare derivative works of, publicly display, publicly perform, and distribute
the Submission and such derivative works, and to sublicense any or all of the foregoing rights to third
parties.
* Patent License. You grant Microsoft, and those who receive the Submission directly or
indirectly from Microsoft, a perpetual, worldwide, non-exclusive, royalty-free, irrevocable license under
Your patent claims that are necessarily infringed by the Submission or the combination of the
Submission with the Project to which it was Submitted to make, have made, use, offer to sell, sell and
import or otherwise dispose of the Submission alone or with the Project.
* Other Rights Reserved. Each party reserves all rights not expressly granted in this Agreement.
No additional licenses or rights whatsoever (including, without limitation, any implied licenses) are
granted by implication, exhaustion, estoppel or otherwise.
1. Representations and Warranties. You represent that You are legally entitled to grant the above
licenses. You represent that each of Your Submissions is entirely Your original work (except as You may
have disclosed under Section 3). You represent that You have secured permission from Your employer to
make the Submission in cases where Your Submission is made in the course of Your work for Your
employer or Your employer has intellectual property rights in Your Submission by contract or applicable
law. If You are signing this Agreement on behalf of Your employer, You represent and warrant that You
have the necessary authority to bind the listed employer to the obligations contained in this Agreement.
You are not expected to provide support for Your Submission, unless You choose to do so. UNLESS
REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING, AND EXCEPT FOR THE WARRANTIES
EXPRESSLY STATED IN SECTIONS 3, 4, AND 6, THE SUBMISSION PROVIDED UNDER THIS AGREEMENT IS
PROVIDED WITHOUT WARRANTY OF ANY KIND, INCLUDING, BUT NOT LIMITED TO, ANY WARRANTY OF
NONINFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
2. Notice to Microsoft. You agree to notify Microsoft in writing of any facts or circumstances of which
You later become aware that would make Your representations in this Agreement inaccurate in any
respect.
3. Information about Submissions. You agree that contributions to Projects and information about
contributions may be maintained indefinitely and disclosed publicly, including Your name and other
information that You submit with Your Submission.
4. Governing Law/Jurisdiction. This Agreement is governed by the laws of the State of Washington, and
the parties consent to exclusive jurisdiction and venue in the federal courts sitting in King County,
Washington, unless no federal subject matter jurisdiction exists, in which case the parties consent to
exclusive jurisdiction and venue in the Superior Court of King County, Washington. The parties waive all
defenses of lack of personal jurisdiction and forum non-conveniens.
5. Entire Agreement/Assignment. This Agreement is the entire agreement between the parties, and
supersedes any and all prior agreements, understandings or communications, written or oral, between
the parties relating to the subject matter hereof. This Agreement may be assigned by Microsoft.
—
Reply to this email directly, view it on GitHub<#365 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/B3WGWDGSW3ULBV7VVAUEKPD4EF7RRAVCNFSM6AAAAACQIJ5J66VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMOJXGM2DMMJWGQ>.
You are receiving this because you were mentioned.Message ID: ***@***.***>
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
[RFC] Integration of sparse-ternary-fma for accelerated ternary operations
Pull Request Type: Request for Comment (RFC)
Target Repository: microsoft/BitNet
Source Branch: HyperFoldUK:main
Target Branch: microsoft:main
Author: HyperFoldUK maurice.wilson@hyperfold-technologies.com
Purpose
This RFC proposes the integration of the sparse-ternary-fma library into BitNet to significantly accelerate ternary matrix operations through optimized 2-bit encoding and SIMD instructions (AVX2/AVX-512).
Background & Principle
BitNet's 1.58-bit quantization represents weights as ternary values {-1, 0, +1}, enabling extreme model compression while maintaining competitive accuracy. However, the current implementation faces efficiency constraints:
The sparse-ternary-fma library addresses these limitations through:
Why This Matters
Ternary quantization is fundamentally different from traditional quantization. The presence of explicit zeros creates opportunities for sparsity-aware computation that standard quantization approaches cannot exploit. By using 2-bit encoding and SIMD operations, we can:
This Implementation
This fork demonstrates a clean integration:
Architecture
Key Optimizations
1. Branchless Encoding Conversion
Replaces loop+switch with pure bitwise operations:
Impact: Zero branches, processes 4 trits in parallel, ~5 assembly instructions
2. SIMD Trit Unpacking
Eliminates stack round-trip by unpacking directly in registers:
Impact: Eliminates 16 scalar operations + 1 memory load, stays in registers
3. Thread-Local Buffer Pooling
Impact: Zero allocations in hot path after warmup
4. Threshold-Based Dispatch
Impact: Automatic selection based on operation size
Integration Points
Modified Files:
src/ggml-bitnet-mad.cpp- Added automatic dispatch logicNew Files:
include/ggml-bitnet-stfma.h- Adapter layer APIsrc/ggml-bitnet-stfma.cpp- Adapter layer implementation3rdparty/sparse-ternary-fma/- Vendored library (Apache 2.0 licensed)Build System:
CMakeLists.txt- Added sparse-ternary-fma configurationsrc/CMakeLists.txt- Added adapter source filesPerformance
Based on sparse-ternary-fma benchmarks on Intel Xeon with AVX-512:
Benchmark Details
Memory
Encoding Efficiency
Runtime Overhead
Memory Access Pattern
Design
Backward Compatibility
No breaking changes
-DBITNET_USE_STFMA=OFFConfigurability
CMake Options:
Runtime Behavior:
n < threshold: Use original implementationn >= threshold: Use sparse-ternary-fmaTesting
Test Suite Location:
tests/stfma_integration/Coverage:
Test Results:
Code Quality
-Wall -Wextra -WpedanticFull Documentation
Complete technical documentation is available in:
We are seeking feedback from the maintainers and community on:
1. The technical approach and integration design
Questions:
Trade-offs:
2. Performance characteristics on diverse hardware
Needed benchmarks:
Questions:
3. The potential path to upstream adoption
Integration options:
Option A: Optional Feature (Current Approach)
Option B: Native Encoding Change
Option C: Hybrid Approach
Questions:
The code is complete, tested, and ready for review.
We believe this addresses a fundamental efficiency ceiling for ternary computation. By leveraging 2-bit encoding and SIMD acceleration, we can unlock significant performance gains for BitNet models while maintaining full backward compatibility.
What's Included
Complete implementation with all optimizations
Comprehensive test suite with 100% pass rate
Full documentation including integration guide
Backward compatibility with existing code
Configurable behavior via CMake options
Clean commit history with detailed messages
Commit Summary
Integrate sparse-ternary-fma for optimized ternary matrix operations
Optimize encoding conversion with branchless bitwise logic
Optimize AVX2/AVX-512 trit unpacking to eliminate stack round-trip
Organize test files and artifacts into tests/stfma_integration directory
All commits are authored by HyperFoldUK maurice.wilson@hyperfold-technologies.com
Related Work
How to Review
Quick Start
Clone the fork:
git clone https://github.com/HyperFoldUK/BitNet.git cd BitNetBuild with integration:
Run tests:
Detailed Review
STFMA_INTEGRATION_README.mdfor design overviewsrc/ggml-bitnet-stfma.cppfor adapter layersrc/ggml-bitnet-mad.cppfor dispatch logictests/stfma_integration/for verificationContact
For questions or discussions:
We look forward to your feedback and are happy to make adjustments based on maintainer preferences.