From software threads to parallel hardware with LegUp high-level synthesis
Bibliographic record
Abstract
High-level synthesis (HLS) can automatically synthesize software to hardware. With the design specification in software, HLS can reduce the lengthy design cycles of hardware, and make the performance and energy-efficiency benefits of hardware accessible to those without hardware skills.\n Since the introduction of the first C-based HLS tools more than a decade ago, however, the\nadaption of the technology has been slow by both software and hardware engineers. We attribute this\nto two key factors: 1) For hardware engineers, there is still a gap between HLS-generated hardware and\nhuman-designed hardware, partly due to the inability of HLS tools to fully exploit hardware parallelism,\nand 2) for software engineers, HLS remains to be a difficult endeavour, as many parts of the design, such\nas system integration, largely remain a manual process.\n This dissertation provides an HLS framework, LegUp, which seeks to address both issues. LegUp\ncan compile an entire software program to hardware to produce a hardware-only system, or it can\nalso automatically partition the program to generate a processor-accelerator hybrid system, wherein the\ncompute-intensive program segments are accelerated by hardware, with the remaining segments executed in software on a processor. In both cases, a complete system is generated, including necessary\nmemories and interconnect. To allow one to easily exploit hardware parallelism, we provide HLS support for synthesizing parallel software to parallel hardware. In particular, we support automatically\ncompiling a multi-threaded program with Pthreads and OpenMP to parallel hardware accelerators that\noperate concurrently within a hardware-only or a processor-accelerator hybrid system. In the context\nof parallel hardware, we investigate architectural and memory optimizations that help to improve circuit performance and area, and discuss a method of using the producer-consumer pattern in software\nto infer a streaming circuit in hardware. With these techniques, we show that LegUp can produce\nhigh-performance hardware that can be competitive to circuits that are generated by commercial HLS\ntools, and demonstrate that LegUp-generated circuits can also outperform software executing on x86\nprocessors.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.002 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.002 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".