Skip to content
STIMSMITH

Accelerator Value Spectra

Concept

Accelerator Value Spectra are FPGA-relevant runtime measurements tracked from synthesized HLS kernels (e.g., actual variable value ranges, accessed memory offsets, used stack size, loop iteration counts, and FIFO queue sizes) used by HeteroFuzz as guidance feedback to detect platform-dependent divergence between CPU and FPGA execution.

First seen 7/26/2026
Last seen 7/26/2026
Evidence 10 chunks
Wiki v1

WIKI

Accelerator Value Spectra

Definition

Accelerator Value Spectra are FPGA-relevant runtime value measurements collected from synthesized hardware kernels produced by High-Level Synthesis (HLS). Because HLS implements branching as a pipeline of hardware logic—where each branch executes in parallel as long as prior signals are ready, but untaken branches simply produce unused outputs—branch coverage cannot be derived directly from the synthesized hardware. To overcome this limitation, HeteroFuzz analyzes inserted HLS pragmas and traces their associated FPGA-relevant spectra as accelerator feedback [chunk:58e8345a]. Accelerator Value Spectra are described as a multi-dimensional guidance feedback signal, used alongside host-code branch coverage, to detect platform-dependent divergence between CPU and FPGA execution [chunk:309a4e09].

READ FULL ARTICLE →

NEIGHBORHOOD

No graph connections found for this entity yet. It may appear in future ingestion runs.

explore full graph →

RELATIONSHIPS

2 connections
HeteroFuzz ← implements 100% 2e
HeteroFuzz monitors accelerator-relevant value spectra as a guidance mechanism.
FIFO Queue ← part of 100% 2e
FIFO queue size is one of the accelerator value spectra tracked by HeteroFuzz.

LINKED ENTITIES

2 links
HeteroFuzz IMPLEMENTS Extracted graph relationship
FIFO Queue PART_OF Extracted graph relationship

CITATIONS

13 sources
13 citations — click to expand
[1] Accelerator Value Spectra are FPGA-relevant runtime measurements tracked from synthesized HLS kernels and used as guidance feedback for detecting platform-dependent divergence. HeteroFuzz: Fuzz Testing to Detect Platform Dependent Divergence for Heterogeneous Applications
[2] HeteroFuzz currently supports five kinds of value spectra, with the HLS-pragma-to-spectra mapping being user-extensible via a configuration file. HeteroFuzz: Fuzz Testing to Detect Platform Dependent Divergence for Heterogeneous Applications
[3] Table 2 maps HLS pragma categories (Data Type, Memory, Recursion, Parallelization, Loop, Interface/Configuration) to specific FPGA-relevant Accelerator Value Spectra, including actual variable values, accessed offsets, used stack size, iteration counts, and FIFO queue size. HeteroFuzz: Fuzz Testing to Detect Platform Dependent Divergence for Heterogeneous Applications
[4] Branch coverage cannot be derived directly from synthesized hardware because HLS implements branching as a pipeline of hardware logic where untaken branches simply produce unused outputs. HeteroFuzz: Fuzz Testing to Detect Platform Dependent Divergence for Heterogeneous Applications
[5] Inputs with the same branch coverage in the host code can still have distinct impacts on hardware characteristics, motivating accelerator value spectra as an additional guidance signal. HeteroFuzz: Fuzz Testing to Detect Platform Dependent Divergence for Heterogeneous Applications
[6] Each test execution initializes an acc_feedback array with entries of form <value spectra type, name, min value, max value>, and total_feedback tracks cumulative spectra of FPGA-safe inputs. HeteroFuzz: Fuzz Testing to Detect Platform Dependent Divergence for Heterogeneous Applications
[7] For the accumulate kernel, HeteroFuzz records the value range of fpga_data, sum, and i, the set of accessed offsets in data_fpga, and the actual loop iteration count. HeteroFuzz: Fuzz Testing to Detect Platform Dependent Divergence for Heterogeneous Applications
[8] Mutation activation probabilities are updated by adding α=0.05 when a mutation increases spectra and subtracting α/(l−1) otherwise, with initial probability 1/l. HeteroFuzz: Fuzz Testing to Detect Platform Dependent Divergence for Heterogeneous Applications
[9] Recording only the range of integer-valued kernel variables is sufficient because FPGA synthesis maps integer/fixed-point variables to contiguous value ranges. HeteroFuzz: Fuzz Testing to Detect Platform Dependent Divergence for Heterogeneous Applications
[10] Monitoring accelerator spectra produces 7.78× more divergence-inducing inputs (3.7× more unique divergence symptoms) than branch-coverage-only fuzzing. HeteroFuzz: Fuzz Testing to Detect Platform Dependent Divergence for Heterogeneous Applications
[11] Across seven applications, WithoutSpectra takes 21 hours to find 10 unique symptoms while HeteroFuzz takes 0.52 hours—a 40× speed-up. HeteroFuzz: Fuzz Testing to Detect Platform Dependent Divergence for Heterogeneous Applications
[12] On average, 56.8% of HeteroFuzz-generated inputs are divergence-inducing, versus 6.48% for WithoutSpectra. HeteroFuzz: Fuzz Testing to Detect Platform Dependent Divergence for Heterogeneous Applications
[13] FIFO queue size is the FPGA-relevant value spectra associated with #pragma HLS dataflow when a feedback path exists. HeteroFuzz: Fuzz Testing to Detect Platform Dependent Divergence for Heterogeneous Applications