Home / flashinfer-ai / flashinfer
AI Slop Code RIsk Monitor

flashinfer-ai/flashinfer

FlashInfer: Kernel Library for LLM Serving

19.5 Elevated pattern signal View on GitHub

Maintenance and Governance Review

This report summarizes scanner evidence for flashinfer-ai/flashinfer, a Cuda project with 6,586 GitHub stars. SynthScan examined 3,497,390 lines of code across 4895 source files, recording 38309 pattern matches distributed across 25 syntactic categories. The legacy adjusted pattern signal is 19.5 (Elevated pattern signal); it is not an authorship estimate.

Use the findings to ask whether maintainers can explain intent, verify dependencies, simplify unnecessary scaffolding, and focus review effort. Risk Payload v1 adds ownership, supply-chain, over-engineering, and review-effort vectors when available. These are prompts for human verification, not probabilities, security findings, or measured maintenance costs.

19.5
Adjusted Signal
19.5
Raw Pattern Score
100%
Time Factor
2026-10-10
Last Push
6.6K
Stars
Cuda
Language
3.5M
Lines of Code
4.9K
Files
38.3K
Pattern Hits
2026-10-12
Scan Date
1.04
HC Hit Rate

Risk Review · v3

Use each vector to guide manual verification of the underlying findings.

Cognitive Ownership Risk 22.4/100 · LOW

Can maintainers explain intent, assumptions, and trade-offs? 19192 finding(s).

Hallucinated Dependency & Supply Chain Risk 9.7/100 · LOW

Are imports declared and package/API names verified? 6932 finding(s).

Agent Over-Engineering Ratio 55.0/100 · HIGH

Are wrappers, guards, and repeated structures necessary? 26384 finding(s).

PR Review Velocity Impact 66.3/100 · HIGH

Which changes need focused human review? 38309 finding(s).

What These Metrics Mean

Adjusted Signal
Legacy pattern score normalized per 1,000 scanned lines and multiplied by the repository-level time factor. It is not an estimate of AI authorship.
Raw Pattern Score
Sum of severity- and context-weighted pattern findings before applying the time factor.
Time Factor
A coarse multiplier based on repository last-push metadata. It does not identify when individual code was written.
Pattern Hits
Count of configured pattern findings. Review each location and surrounding code before acting.
HC Hit Rate
High- and critical-severity findings per scanned file. Severity is a configured detector weight, not proof of a vulnerability or authorship.
Lines / Files
Scan coverage denominators used by the legacy normalized score.

Pattern-Signal History

This chart tracks the legacy adjusted pattern score across scan runs. Changes can reflect source changes, scan coverage, or detector/scoring versions; they do not measure AI adoption or prove that a team refactored code. Compare runs only when their scoring versions are compatible. The dashed line tracks the raw pattern-hit count.

Severity Breakdown

Groups findings by configured severity weight. Severity is not a confidence estimate, security rating, or proof that a package is missing. Verify imports against the project's manifests and inspect each finding in its local design context.

CRITICAL 84HIGH 5024MEDIUM 5908LOW 27293

Directory Score Breakdown

This horizontal bar chart shows where configured pattern scores accumulate across directories. Treat the bars as locations to inspect, not as evidence of authorship or code quality. Use the file-level findings to decide whether a directory needs a focused ownership, dependency, complexity, or review check.

Pattern Findings

The scanner identified 38309 distinct pattern matches across 25 syntactic categories. Each entry below represents a discrete location in the source code where the engine recorded a detector match to inspect in context. Expand any category row to inspect the individual file paths, line numbers, code snippets, and the lexical context (CODE, COMMENT, or STRING) in which each match was detected.

Reading the findings table: The Severity column indicates the configured finding weight (CRITICAL / HIGH / MEDIUM / LOW). The Context column identifies whether the match occurred inside executable code, an inline comment, or a string literal — legacy comment-context matches receive a ×1.5 score weight. The ⚡ bolt icon marks clustered matches: three or more patterns within a 10-line window, each receiving an additional ×1.5 density multiplier under the legacy scoring model.

Cross-File Repetition4824 hits · 24120 pts
SeverityFileLineSnippetContext
HIGHbuild_backend.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_dsa_indexer.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_kimi_k3_fp8_projection.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_nvfp4_sparse_mla_decode.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_mla_varq_dcp_decode.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_kimi_k3_attn_res.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_nvfp4_mla_decode.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_balanced_gqa_decode.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_nvfp4_attention.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_sm110_gqa_decode.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_kimi_k3_latent_moe.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_dsa_train.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_sm110_gqa_decode_prepared.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_lm_head_loss.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_vsa_sm90.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_experimental_api.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_msa_nvfp4_decode.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_kimi_k3_fused_router.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_kimi_k3_vision_tower.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGH…/experimental/test_cake_minimax_h3_varlen_attention.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_mxfp8_megamoe_ep16.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/experimental/test_cake_moe_grouped_gemm.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/attention/test_nvfp4_attention_sm120.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/attention/test_mla_auto_backend.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/attention/test_lse_base.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGH…tention/test_cake_sage_block_sparse_attention_sm100.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/attention/test_bidirectional_ranges_attention.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/attention/test_workspace_size.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/attention/test_auto_backend_upgrade.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/attention/test_cake_vsa.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/attention/test_cake_vsa_planning.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/trace/test_b12x_moe_trace.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/moe_ep/test_sm90_push_fp8_gemm.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/mamba/test_cake_ssd_combined.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/mamba/test_cake_ssd_combined_capture.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/comm/test_mixed_comm_checkpoint.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/comm/test_pcie_ipc_all_reduce.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/comm/test_ulysses_head_chunk.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/utils/test_collect_env.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/utils/test_gen_module_symlink_race_condition.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/utils/test_sampling_seed_offset.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/utils/test_cake_sampling_upstream.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/utils/test_gated_act_mxfp8.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/gemm/test_cake_mm_fp4.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/gemm/test_group_gemm_fp4.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/quantization/test_cake_nvfp4_quantize.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/gdn/test_replayssm_spec_fold.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/jit/test_bgmv_moe_jit.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/jit/test_aot_sm103_modules.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/jit/test_cutlass_fused_moe_aot.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/jit/test_cake_bgmv_moe_jit.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/jit/test_blackwell_msa_jit.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/moe/test_trtllm_moe_valid_dims.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/moe/test_b12x_w4a16_ep.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/moe/test_cake_bgmv_moe.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/moe/test_b12x_fused_moe.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/moe/test_unified_moe_mxfp4.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/moe/test_hash_topk.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/moe/test_alphamoe_fused_router.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
HIGHtests/moe/test_trtllm_gen_moe_autotune_tactics.py0\ /* * copyright (c) 2026, nvidia corporation. all rights reserved. * * licensed under the apache license, version 2.0 (STRING
4764 more matches not shown…
Decorative Section Separators5655 hits · 16960 pts
SeverityFileLineSnippetContext
MEDIUMcsrc/fused_moe/monomoe/src/ptx_utils.h14// ── cp.async scalar helpers (sm_80+) ──────────────────────────────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/ptx_utils.h35// ── Hopper WGMMA (sm_90a) helpers ─────────────────────────────────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/ptx_utils.h148// ── Hopper mbarrier helpers (sm_90a) ──────────────────────────────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/ptx_utils.h309// ── Hopper TMA bulk-tensor copy (sm_90a) ──────────────────────────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_tma.h214// ─── Device-side TMA load helpers ────────────────────────────────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_tma.h437// ─── Down-projection (Phase 4) TMA load helpers ──────────────────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_interface.h40// ── Single fixed-shape Dims (E=256, N=512, K=2048 block-wise FP8) ─────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h24// ── KernelConfig opt-in detection ───────────────────────────────────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h257 // ── Sentinel-based Phase 3 → 4 handoff state (tail fields) ────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h285 // ── Phase 4 → 5 readiness flags (replaces the site-#3 barrier) ────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h285 // ── Phase 4 → 5 readiness flags (replaces the site-#3 barrier) ────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h306// ── Sentinel handoff helpers (Phase 3 → 4) ────────────────────────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h409 // ── WGMMA tile geometry ───────────────────────────────────────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h427 // ── Down-proj outer K-step (K_STEP_DOWN tunable) ──────────────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h437 // ── Down-proj TMA pipeline depth (DOWN_PIPE_DEPTH tunable) ────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h451 // ── Up-proj outer K-step (K_STEP_UP tunable) ──────────────────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h462 // ── Up-proj weight-TMA lookahead (UP_W_SLOTS tunable) ─────────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h483 // ── Up-proj col-halves (M-axis 128-row atoms per block) ───────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h501 // ── Down-projection grid layout ───────────────────────────────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h665 // ── mbarriers (16-B alignment required by SM90 mbarrier PTX) ─────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h678 // ── Phase 3 → Phase 4 (expert, token) reorganization tables ──────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h730 // ── Common fields ────────────────────────────────────────────────────────COMMENT
MEDIUMcsrc/fused_moe/monomoe/src/moe_internal.h781// ── Warp identity helpers ───────────────────────────────────────────────────COMMENT
MEDIUMtests/test_minimax_h3_mxfp8.py244# --------------------------------------------------------------------------COMMENT
MEDIUMtests/test_minimax_h3_mxfp8.py246# --------------------------------------------------------------------------COMMENT
MEDIUMtests/test_minimax_h3_nvfp4.py296# ---------------------------------------------------------------------------COMMENT
MEDIUMtests/test_minimax_h3_nvfp4.py298# ---------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/test_minimax_h3_qkv_quantize_pack.py65# ---------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/test_minimax_h3_qkv_quantize_pack.py67# ---------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/test_minimax_h3_qkv_quantize_pack.py163# ---------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/test_minimax_h3_qkv_quantize_pack.py165# ---------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_kda_output_only.py493# =============================================================================COMMENT
MEDIUM⚡tests/kda/test_kda_output_only.py495# =============================================================================COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py752# ==============================================================================COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py754# ==============================================================================COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py1468# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py1470# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py1589# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py1591# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py1655# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py1657# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py1687# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py1689# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py2109# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py2111# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py2155# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py2157# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py2227# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py2229# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py2287# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py2289# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py2457# ==============================================================================COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py2459# ==============================================================================COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py2468# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py2470# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py2540# ------------------------------------------------------------------------------COMMENT
MEDIUM⚡tests/kda/test_recurrent_kda.py2542# ------------------------------------------------------------------------------COMMENT
MEDIUMtests/kda/test_recurrent_kda.py43# ==============================================================================COMMENT
MEDIUMtests/kda/test_recurrent_kda.py45# ==============================================================================COMMENT
MEDIUMtests/kda/test_recurrent_kda.py113# ==============================================================================COMMENT
5595 more matches not shown…
Hyper-Verbose Identifiers13344 hits · 11999 pts
SeverityFileLineSnippetContext
LOW⚡build_backend.py1153def get_requires_for_build_wheel(config_settings=None):CODE
LOW⚡build_backend.py1158def get_requires_for_build_sdist(config_settings=None):CODE
LOW⚡build_backend.py1163def get_requires_for_build_editable(config_settings=None):CODE
LOW⚡build_backend.py1168def prepare_metadata_for_build_wheel(metadata_directory, config_settings=None):CODE
LOW⚡build_backend.py1173def prepare_metadata_for_build_editable(metadata_directory, config_settings=None):CODE
LOWbuild_backend.py359def _synthesize_nccl_builddir(build: Path) -> None:CODE
LOWbuild_backend.py417def _check_nccl_version_drift(wheel: Path) -> None:CODE
LOWbuild_backend.py698def _install_nvep_runtime_wheels(built_nixl: bool) -> None:CODE
LOWbuild_backend.py792def _system_cuda_tile_compiler_available() -> bool:CODE
LOWbuild_backend.py801def _install_cuda_tile_compile_deps() -> None:CODE
LOWbuild_utils.py43def get_dependency_requirements(CODE
LOW⚡build_utils.py69def get_build_dependency_requirements(CODE
LOW⚡build_utils.py76def get_cuda_extra_dependency_requirements(CODE
LOW⚡build_utils.py83def get_ci_image_dependency_requirements(CODE
LOW⚡build_utils.py90def get_cuda_tile_compile_dependency_requirements() -> list[str]:CODE
LOW…03a/cake_mxfp4_situ_moe_636d7842ea994e777f9a_kernel.py112def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_6cb3a150103691dde0de_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_7f291b75fc80b9a071f0_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_6cfe5b1f49445aeb8696_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_0d3d1450452b45b8c833_kernel.py109def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_e08d1d673c6646be544d_kernel.py113def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_7dccce76ace7a05df2b3_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_72ece5210802b11d0db8_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_96ae6baf9f1ce5c68f64_kernel.py109def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_5e57c8286981e3b9b414_kernel.py112def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_b7aedaabbebd7af0495a_kernel.py109def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_0888b2ad92e1a620131a_kernel.py109def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_420577304e5b3471f743_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_335a69c359d728d6af44_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_63abec40d9d3c5d4938d_kernel.py112def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_b3ce0d3341158795cdc2_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_a5525fc77d36e6af0c95_kernel.py113def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_9efcd6af56b94dc0151d_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_172cacdf988f22944dc2_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_8620631fd416b458d166_kernel.py109def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_3c697485ac8a3394f2e2_kernel.py116def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_394ae8bc3951d3e66cbe_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_ff45ab57e3db34e1cda4_kernel.py109def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_18b1f0ff11480b38b2df_kernel.py109def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_26453ed3ca0313104c61_kernel.py109def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_4c4ded85112c2ae18068_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_b5fd05ebbb6897d40909_kernel.py109def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_90b3ce0b48c78aadf332_kernel.py109def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_4c432e76a8546fd1e754_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_d4e456c26269c9634d1b_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_bcf47b55163b9faca2df_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_ae801cbb2b9a5b773ba4_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_5f13d763df1e76b40af9_kernel.py109def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_652e013e44ce09397cc7_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_1f2050097a8b81a95b1f_kernel.py109def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_74f85ada980f4396e364_kernel.py112def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_48a8beddc50f3585743e_kernel.py102def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOW…03a/cake_mxfp4_situ_moe_b3397620198e1f92789c_kernel.py109def _cake_launch_cluster_spread(kernel_launcher, **kwargs):CODE
LOWdocker/test_ci_image.py72def _validate_cuda_tile_compiler(CODE
LOWdocker/test_ci_image.py139def _validate_cuda_tile_for_runtime(CODE
LOWdocker/test_ci_image.py147def _validate_cuda_runtime_distributions(CODE
LOWflashinfer-cubin/build_backend.py110def get_requires_for_build_wheel(config_settings=None):CODE
LOWflashinfer-cubin/build_backend.py118def get_requires_for_build_editable(config_settings=None):CODE
LOWci/validate_cuda_versions.py80def build_jit_cache_provider_matrix(config: dict[str, Any]) -> list[dict[str, str]]:CODE
LOWci/validate_cuda_versions.py99def build_jit_cache_shim_matrix(config: dict[str, Any]) -> list[dict[str, str]]:CODE
13284 more matches not shown…
Unused Imports6785 hits · 6136 pts
SeverityFileLineSnippetContext
LOW…03a/cake_mxfp4_situ_moe_f5a20e3662d9aa8a230a_kernel.py18CODE
LOW…03a/cake_mxfp4_situ_moe_f5a20e3662d9aa8a230a_kernel.py23CODE
LOW…03a/cake_mxfp4_situ_moe_f5a20e3662d9aa8a230a_kernel.py26CODE
LOW…03a/cake_mxfp4_situ_moe_f5a20e3662d9aa8a230a_kernel.py27CODE
LOW…03a/cake_mxfp4_situ_moe_f5a20e3662d9aa8a230a_kernel.py30CODE
LOW…03a/cake_mxfp4_situ_moe_f5a20e3662d9aa8a230a_kernel.py31CODE
LOW…03a/cake_mxfp4_situ_moe_f5a20e3662d9aa8a230a_kernel.py32CODE
LOW…03a/cake_mxfp4_situ_moe_f5a20e3662d9aa8a230a_kernel.py33CODE
LOW…03a/cake_mxfp4_situ_moe_f5a20e3662d9aa8a230a_kernel.py34CODE
LOW…03a/cake_mxfp4_situ_moe_f5a20e3662d9aa8a230a_kernel.py42CODE
LOW…03a/cake_mxfp4_situ_moe_f5a20e3662d9aa8a230a_kernel.py43CODE
LOW…03a/cake_mxfp4_situ_moe_de8253c4a7b8e18298dc_kernel.py18CODE
LOW…03a/cake_mxfp4_situ_moe_de8253c4a7b8e18298dc_kernel.py23CODE
LOW…03a/cake_mxfp4_situ_moe_de8253c4a7b8e18298dc_kernel.py26CODE
LOW…03a/cake_mxfp4_situ_moe_de8253c4a7b8e18298dc_kernel.py27CODE
LOW…03a/cake_mxfp4_situ_moe_de8253c4a7b8e18298dc_kernel.py30CODE
LOW…03a/cake_mxfp4_situ_moe_de8253c4a7b8e18298dc_kernel.py31CODE
LOW…03a/cake_mxfp4_situ_moe_de8253c4a7b8e18298dc_kernel.py32CODE
LOW…03a/cake_mxfp4_situ_moe_de8253c4a7b8e18298dc_kernel.py33CODE
LOW…03a/cake_mxfp4_situ_moe_de8253c4a7b8e18298dc_kernel.py34CODE
LOW…03a/cake_mxfp4_situ_moe_de8253c4a7b8e18298dc_kernel.py42CODE
LOW…03a/cake_mxfp4_situ_moe_de8253c4a7b8e18298dc_kernel.py43CODE
LOW…03a/cake_mxfp4_situ_moe_273def40246d5afa1a88_kernel.py18CODE
LOW…03a/cake_mxfp4_situ_moe_273def40246d5afa1a88_kernel.py23CODE
LOW…03a/cake_mxfp4_situ_moe_273def40246d5afa1a88_kernel.py26CODE
LOW…03a/cake_mxfp4_situ_moe_273def40246d5afa1a88_kernel.py27CODE
LOW…03a/cake_mxfp4_situ_moe_273def40246d5afa1a88_kernel.py30CODE
LOW…03a/cake_mxfp4_situ_moe_273def40246d5afa1a88_kernel.py31CODE
LOW…03a/cake_mxfp4_situ_moe_273def40246d5afa1a88_kernel.py32CODE
LOW…03a/cake_mxfp4_situ_moe_273def40246d5afa1a88_kernel.py33CODE
LOW…03a/cake_mxfp4_situ_moe_273def40246d5afa1a88_kernel.py34CODE
LOW…03a/cake_mxfp4_situ_moe_273def40246d5afa1a88_kernel.py42CODE
LOW…03a/cake_mxfp4_situ_moe_273def40246d5afa1a88_kernel.py43CODE
LOW…03a/cake_mxfp4_situ_moe_636d7842ea994e777f9a_kernel.py18CODE
LOW…03a/cake_mxfp4_situ_moe_636d7842ea994e777f9a_kernel.py26CODE
LOW…03a/cake_mxfp4_situ_moe_636d7842ea994e777f9a_kernel.py30CODE
LOW…03a/cake_mxfp4_situ_moe_636d7842ea994e777f9a_kernel.py31CODE
LOW…03a/cake_mxfp4_situ_moe_636d7842ea994e777f9a_kernel.py32CODE
LOW…03a/cake_mxfp4_situ_moe_636d7842ea994e777f9a_kernel.py33CODE
LOW…03a/cake_mxfp4_situ_moe_636d7842ea994e777f9a_kernel.py34CODE
LOW…03a/cake_mxfp4_situ_moe_636d7842ea994e777f9a_kernel.py42CODE
LOW…03a/cake_mxfp4_situ_moe_636d7842ea994e777f9a_kernel.py43CODE
LOW…03a/cake_mxfp4_situ_moe_cbb779f90aa8d79d2bd3_kernel.py18CODE
LOW…03a/cake_mxfp4_situ_moe_cbb779f90aa8d79d2bd3_kernel.py23CODE
LOW…03a/cake_mxfp4_situ_moe_cbb779f90aa8d79d2bd3_kernel.py26CODE
LOW…03a/cake_mxfp4_situ_moe_cbb779f90aa8d79d2bd3_kernel.py27CODE
LOW…03a/cake_mxfp4_situ_moe_cbb779f90aa8d79d2bd3_kernel.py30CODE
LOW…03a/cake_mxfp4_situ_moe_cbb779f90aa8d79d2bd3_kernel.py31CODE
LOW…03a/cake_mxfp4_situ_moe_cbb779f90aa8d79d2bd3_kernel.py32CODE
LOW…03a/cake_mxfp4_situ_moe_cbb779f90aa8d79d2bd3_kernel.py33CODE
LOW…03a/cake_mxfp4_situ_moe_cbb779f90aa8d79d2bd3_kernel.py34CODE
LOW…03a/cake_mxfp4_situ_moe_cbb779f90aa8d79d2bd3_kernel.py42CODE
LOW…03a/cake_mxfp4_situ_moe_cbb779f90aa8d79d2bd3_kernel.py43CODE
LOW…03a/cake_mxfp4_situ_moe_6cb3a150103691dde0de_kernel.py18CODE
LOW…03a/cake_mxfp4_situ_moe_6cb3a150103691dde0de_kernel.py26CODE
LOW…03a/cake_mxfp4_situ_moe_6cb3a150103691dde0de_kernel.py30CODE
LOW…03a/cake_mxfp4_situ_moe_6cb3a150103691dde0de_kernel.py31CODE
LOW…03a/cake_mxfp4_situ_moe_6cb3a150103691dde0de_kernel.py32CODE
LOW…03a/cake_mxfp4_situ_moe_6cb3a150103691dde0de_kernel.py33CODE
LOW…03a/cake_mxfp4_situ_moe_6cb3a150103691dde0de_kernel.py34CODE
6725 more matches not shown…
Deep Nesting2297 hits · 1984 pts
SeverityFileLineSnippetContext
LOWbuild_backend.py556CODE
LOW…03a/cake_mxfp4_situ_moe_f5a20e3662d9aa8a230a_kernel.py112CODE
LOW…03a/cake_mxfp4_situ_moe_de8253c4a7b8e18298dc_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_273def40246d5afa1a88_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_636d7842ea994e777f9a_kernel.py127CODE
LOW…03a/cake_mxfp4_situ_moe_cbb779f90aa8d79d2bd3_kernel.py112CODE
LOW…03a/cake_mxfp4_situ_moe_6cb3a150103691dde0de_kernel.py117CODE
LOW…03a/cake_mxfp4_situ_moe_7f291b75fc80b9a071f0_kernel.py117CODE
LOW…03a/cake_mxfp4_situ_moe_6cfe5b1f49445aeb8696_kernel.py117CODE
LOW…03a/cake_mxfp4_situ_moe_8d8cc3ba9be7f1289c42_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_0d3d1450452b45b8c833_kernel.py124CODE
LOW…03a/cake_mxfp4_situ_moe_3bf630a285d8c472322a_kernel.py109CODE
LOW…03a/cake_mxfp4_situ_moe_e08d1d673c6646be544d_kernel.py128CODE
LOW…03a/cake_mxfp4_situ_moe_7dccce76ace7a05df2b3_kernel.py117CODE
LOW…03a/cake_mxfp4_situ_moe_cbecac4b68e04db75fc5_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_47f1e2df2cedfd36e9f9_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_72ece5210802b11d0db8_kernel.py117CODE
LOW…03a/cake_mxfp4_situ_moe_53ac52d16cb499ebf2d0_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_066b521eaa2bc452965f_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_a8feec497e1f6c708202_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_d4eee9dcefe1606b0a09_kernel.py103CODE
LOW…03a/cake_mxfp4_situ_moe_6e331f731ce37a7118af_kernel.py109CODE
LOW…03a/cake_mxfp4_situ_moe_96ae6baf9f1ce5c68f64_kernel.py124CODE
LOW…03a/cake_mxfp4_situ_moe_4c6e3ca17f6f1d2dde40_kernel.py112CODE
LOW…03a/cake_mxfp4_situ_moe_5e57c8286981e3b9b414_kernel.py127CODE
LOW…03a/cake_mxfp4_situ_moe_fe500433242ee8099055_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_2eadcb42d9080b1e7367_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_e76fd2975e732d448c05_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_b7aedaabbebd7af0495a_kernel.py124CODE
LOW…03a/cake_mxfp4_situ_moe_6a671ba07a86320dc21c_kernel.py109CODE
LOW…03a/cake_mxfp4_situ_moe_4ef6547ae6f36697e657_kernel.py112CODE
LOW…03a/cake_mxfp4_situ_moe_0888b2ad92e1a620131a_kernel.py124CODE
LOW…03a/cake_mxfp4_situ_moe_ec004a67ba58a2b2cdfb_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_420577304e5b3471f743_kernel.py117CODE
LOW…03a/cake_mxfp4_situ_moe_335a69c359d728d6af44_kernel.py117CODE
LOW…03a/cake_mxfp4_situ_moe_63abec40d9d3c5d4938d_kernel.py127CODE
LOW…03a/cake_mxfp4_situ_moe_bae0ebdbe0d16b41f2df_kernel.py103CODE
LOW…03a/cake_mxfp4_situ_moe_a8065acfe85304720703_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_b3ce0d3341158795cdc2_kernel.py117CODE
LOW…03a/cake_mxfp4_situ_moe_a5525fc77d36e6af0c95_kernel.py128CODE
LOW…03a/cake_mxfp4_situ_moe_691dbc0c459ce444c506_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_51bf8596d88c9a466338_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_f80bd81a91f63f45931f_kernel.py109CODE
LOW…03a/cake_mxfp4_situ_moe_a264b052435345c4e167_kernel.py112CODE
LOW…03a/cake_mxfp4_situ_moe_0aacd570decd7c7afb3c_kernel.py109CODE
LOW…03a/cake_mxfp4_situ_moe_9efcd6af56b94dc0151d_kernel.py117CODE
LOW…03a/cake_mxfp4_situ_moe_172cacdf988f22944dc2_kernel.py117CODE
LOW…03a/cake_mxfp4_situ_moe_8620631fd416b458d166_kernel.py124CODE
LOW…03a/cake_mxfp4_situ_moe_5a776154a02b863cf8be_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_3c697485ac8a3394f2e2_kernel.py131CODE
LOW…03a/cake_mxfp4_situ_moe_607f85ee269af2314ed1_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_394ae8bc3951d3e66cbe_kernel.py117CODE
LOW…03a/cake_mxfp4_situ_moe_ff45ab57e3db34e1cda4_kernel.py124CODE
LOW…03a/cake_mxfp4_situ_moe_b5bf97b3c957dde50fc4_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_e92d7de3ac82b5fae8e7_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_18b1f0ff11480b38b2df_kernel.py124CODE
LOW…03a/cake_mxfp4_situ_moe_0f142b2ee1ef5e6258f7_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_b40bdd660109e3ed19f3_kernel.py110CODE
LOW…03a/cake_mxfp4_situ_moe_26453ed3ca0313104c61_kernel.py124CODE
LOW…03a/cake_mxfp4_situ_moe_072e760aeee15dfe0914_kernel.py109CODE
2237 more matches not shown…
Over-Commented Block2025 hits · 1773 pts
SeverityFileLineSnippetContext
LOWpytest.ini1[pytest]COMMENT
LOWrequirements.txt1apache-tvm-ffi>=0.1.11,<0.2COMMENT
LOWrequirements.txt21# pin nvidia-nccl-cuXX==2.30.7 exactly, which trips the torch-eviction problemCOMMENT
LOW.pre-commit-config.yaml1# To use:COMMENT
LOW.pre-commit-config.yaml21# upstream kernel drops that are re-synced periodicallyCOMMENT
LOWpyproject.toml1# Copyright (c) 2024 by FlashInfer team.COMMENT
LOWpyproject.toml181 "flashinfer/topk_varlen/kernels/filtered_topk_util.py",COMMENT
LOWbuild_backend.py501 shutil.copy(sopath, dst / soname)COMMENT
LOWcsrc/topk_transform_checks.h21// (radix_topk_page_table_transform and cub_topk_page_table_transform), so the backendsCOMMENT
LOWcsrc/fmha_v2/fused_multihead_attention.h201 int64_t softmax_stats_stride_in_bytes;COMMENT
LOW…/fmha_v2/fused_multihead_attention_kernel_4x1_hopper.h461 // Create buffer_full barriers with 1 arrive countCOMMENT
LOWcsrc/fmha_v2/fused_multihead_attention_utils.h801 int const* cu_mask_rows) {COMMENT
LOW…/fused_multihead_flash_attention_kernel_noloop_tiled.h241 float global_sum[Softmax::ROWS_PER_THREAD];COMMENT
LOWcsrc/fmha_v2/fused_multihead_attention_kernel.h21#include <fmha/smem_tile_o.h>COMMENT
LOWcsrc/fmha_v2/fused_multihead_attention_kernel.h41// - P = norm * K^T * Q , where norm is the normalization term (a scalar),COMMENT
LOW…c/fmha_v2/fused_multihead_attention_demo_bert_params.h101 // The stride between matrices of packed mask.COMMENT
LOW…c/fmha_v2/fused_multihead_attention_demo_bert_params.h181#ifdef SKIP_SOFTMAX_STATCOMMENT
LOWcsrc/fmha_v2/fmha/softmax.h1041 }COMMENT
LOWcsrc/fmha_v2/fmha/mask.h341 }COMMENT
LOWcsrc/fmha_v2/fmha/utils.h21#include <__clang_cuda_builtin_vars.h>COMMENT
LOWcsrc/fmha_v2/fmha/utils.h61 using Unpack_remaining = typename Unpack<Ns...>::Type;COMMENT
LOWcsrc/fmha_v2/fmha/utils.h81// (84): hereCOMMENT
LOWcsrc/fmha_v2/fmha/utils.h101struct Cond_inspect_ns {COMMENT
LOWcsrc/fmha_v2/fmha/utils.h801// asm volatile("fma.rn.f16x2.relu %0, %1, %2, %3;" : "=r"(c) : "r"(a), "r"(one),COMMENT
LOWcsrc/fmha_v2/fmha/utils.h821// return dst;COMMENT
LOWcsrc/fmha_v2/fmha/paged_kv_cache.h21struct Kv_block_array {COMMENT
LOWcsrc/fmha_v2/fmha/gmem_tile_o_packed.h1281 // The scale.COMMENT
LOWcsrc/fmha_v2/fmha/kernel_traits.h21#include <fmha/gmem_tile_qkv.h>COMMENT
LOWcsrc/fmha_v2/fmha/fragment.h141 f.reg(4 * ii + 0) = tmp.x;COMMENT
LOWcsrc/fmha_v2/fmha/smem_tile_v.h481 t[3] = y;COMMENT
LOWcsrc/fmha_v2/fmha/smem_tile_v.h601 // Jump by 8*16 rows per K but account for packing.COMMENT
LOWcsrc/fmha_v2/fmha/smem_tile_v.h921 uint2 tmp_0;COMMENT
LOWcsrc/fmha_v2/fmha/warpspec/dma.h341 q_steps,COMMENT
LOWcsrc/fmha_v2/fmha/warpspec/dma.h721 named_barrier_wait(SYNC_BARRIER, NUM_THREADS_IN_DMA_GROUP);COMMENT
LOWcsrc/fmha_v2/fmha/warpspec/dma.h741//COMMENT
LOWcsrc/fmha_v2/fmha/warpspec/compute.h581// results once. For example, STEP_KV=256, qgmma.K=32, then 256/32=8 MMAs are needs,COMMENT
LOWcsrc/fmha_v2/fmha/warpspec/kernel_traits.h41 int Q_BUFFERS_,COMMENT
LOWcsrc/fmha_v2/fmha/warpspec/kernel_traits.h201 // Apply the exp2f optimization (fuse bmm1_scale and -max into FMAs).COMMENT
LOWcsrc/fmha_v2/fmha/warpspec/kernel_traits.h421 // The step size in key/value sequence dimension (N of BMM1 and K of BMM2).COMMENT
LOWcsrc/fmha_v2/fmha/hopper/arrive_wait.h21COMMENT
LOWcsrc/fmha_v2/fmha/hopper/arrive_wait.h41#endifCOMMENT
LOWcsrc/fmha_v2/fmha/hopper/smem_tile.h1801 tmp[nii][ki].w); // PRMT 2+3COMMENT
LOWcsrc/fmha_v2/fmha/hopper/smem_tile.h1821 // [g0, g1] [o0, o1] => y: [g0, h0, o0, p0], i.e. col 0COMMENT
LOWcsrc/fmha_v2/fmha/hopper/smem_tile.h2001 // [00, 01]COMMENT
LOWcsrc/fmha_v2/fmha/hopper/smem_tile.h2121 BYTES_PER_STS, 1, 0, ROWS_PER_XOR_PATTERN, 1>;COMMENT
LOWcsrc/fmha_v2/fmha/hopper/smem_tile.h2321 // w: +24 (o)COMMENT
LOWcsrc/fmha_v2/fmha/hopper/gmem_tile_o_packed.h601 int64_t row_offset =COMMENT
LOWcsrc/fmha_v2/fmha/hopper/fragment.h281 Gmma_single_desc_b::TRANS_MODE == fmha::Gmma_descriptor_transpose::TRANS ? true : false,COMMENT
LOWcsrc/fused_moe/cake_mxfp4_situ_moe/ruff.toml1# The CuTe DSL units under cute/ are generated (rendered from the pinned CakeCOMMENT
LOW…3a/cake_mxfp4_situ_moe_51bf8596d88c9a466338_binding.py1#COMMENT
LOW…03a/cake_mxfp4_situ_moe_f5a20e3662d9aa8a230a_kernel.py1#COMMENT
LOW…3a/cake_mxfp4_situ_moe_47f1e2df2cedfd36e9f9_binding.py1#COMMENT
LOW…3a/cake_mxfp4_situ_moe_4c6e3ca17f6f1d2dde40_binding.py1#COMMENT
LOW…3a/cake_mxfp4_situ_moe_9efcd6af56b94dc0151d_binding.py1#COMMENT
LOW…3a/cake_mxfp4_situ_moe_5f13d763df1e76b40af9_binding.py1#COMMENT
LOW…3a/cake_mxfp4_situ_moe_5f80dbdb484492cbb643_binding.py1#COMMENT
LOW…3a/cake_mxfp4_situ_moe_de8253c4a7b8e18298dc_binding.py1#COMMENT
LOW…3a/cake_mxfp4_situ_moe_bcf47b55163b9faca2df_binding.py1#COMMENT
LOW…03a/cake_mxfp4_situ_moe_de8253c4a7b8e18298dc_kernel.py1#COMMENT
LOW…3a/cake_mxfp4_situ_moe_542fb1c6733740f9f17d_binding.py1#COMMENT
1965 more matches not shown…
Modern Structural Boilerplate828 hits · 791 pts
SeverityFileLineSnippetContext
LOWflashinfer-cubin/flashinfer_cubin/__init__.py80__all__ = ["get_cubin_dir", "list_cubins", "get_cubin_path", "CUBIN_DIR"]CODE
LOW…che-provider/flashinfer_jit_cache_provider/__init__.py27__all__ = ["get_provider"]CODE
LOW…/experimental/b12x/attention/test_attention_mla_api.py77 def set_split_chunk_config(self, *, kv_chunk_size: int, num_chunks: int) -> None:CODE
LOW…/experimental/b12x/attention/test_attention_mla_api.py89 def set_decode_chunk_config(self, *, kv_chunk_size: int, num_chunks: int) -> None:CODE
LOWtests/experimental/deepgemm_common/__init__.py19__all__ = [CODE
LOWtests/attention/conftest.py24logger = logging.getLogger(__name__)CODE
LOWtests/comm/test_nvshmem_allreduce.py12logger = logging.getLogger(__name__)CODE
LOWtests/comm/test_vllm_custom_allreduce.py18logger = logging.getLogger(__name__)CODE
LOWtests/utils/test_topk.py896 def set_selected_scores(use_prefix: bool) -> None:CODE
LOWtests/gemm/test_cute_dsl_gemm_allreduce_two_shot.py36logger = logging.getLogger(__name__)CODE
LOWtests/test_helpers/cake_balanced_gqa_plan.py444__all__ = [CODE
LOWtests/test_helpers/cake_fmha_capability.py559__all__ = [CODE
LOWtests/test_helpers/cake_dsa_indexer_reference.py1510__all__ = [CODE
LOWflashinfer-jit-cache/flashinfer_jit_cache/__init__.py25logger = logging.getLogger(__name__)CODE
LOWflashinfer-jit-cache/flashinfer_jit_cache/__init__.py134__all__ = [CODE
LOWexamples/pytorch/wan/transformer_wan_flashinfer.py63logger = logging.getLogger(__name__)CODE
LOWexamples/pytorch/wan/transformer_wan_flashinfer.py303def set_ulysses_communicator(comm: Optional["UlyssesCommunicator"]) -> None:CODE
LOWbenchmarks/experimental/b12x/benchmark_iq2_xs_moe.py77def set_routes(inputs: Inputs, pattern: str, *, experts: int, offset: int = 0) -> None:CODE
LOWscripts/aot_memory_monitor.py265def update_stats(stats: MemoryStats, sample: dict[str, int]) -> None:CODE
LOWscripts/test_sharding/__init__.py5__all__ = ["ALGORITHM_VERSION", "SCHEMA_VERSION"]CODE
LOWflashinfer/cake_rmsnorm_train.py695__all__ = [STRING
LOWflashinfer/gated_act_mxfp8.py283__all__ = [STRING
LOWflashinfer/cake_dcp.py1139__all__ = [CODE
LOWflashinfer/fp8_quantization.py20__all__ = [CODE
LOWflashinfer/mega_moe_v3.py12__all__ = [CODE
LOWflashinfer/compilation_context.py25logger = logging.getLogger(__name__)CODE
LOWflashinfer/cake_fmha_request_ordered.py377__all__ = [CODE
LOWflashinfer/kimi_k3_latent_moe.py218__all__ = [STRING
LOWflashinfer/cake_minimax_h3.py560__all__ = [CODE
LOWflashinfer/cake_fused_qk_rope_append.py314__all__ = ["cake_fused_qk_rmsnorm_rope_append_paged_kv_cache", "launch_plan"]STRING
LOWflashinfer/sm110_xqa.py13__all__ = ["prepare", "attention"]CODE
LOWflashinfer/kda_prefill_cute.py592__all__ = [CODE
LOWflashinfer/kimi_k3_vision.py132__all__ = ["kimi_k3_vision_tower", "prepare_kimi_k3_vision_tower"]STRING
LOWflashinfer/cake_vsa_sm90.py1451__all__ = [CODE
LOWflashinfer/utils.py38logger = logging.getLogger(__name__)CODE
LOWflashinfer/utils.py864def set_log_level(lvl_str: str) -> None:CODE
LOWflashinfer/fp4_quantization.py45__all__ = [CODE
LOWflashinfer/cake_fmha.py2052__all__ = [CODE
LOWflashinfer/cake_sampling.py1910__all__ = [STRING
LOWflashinfer/artifacts.py368 def update_pbar_cb(_) -> None:CODE
LOWflashinfer/kimi_k3_tp12_tail.py172__all__ = [STRING
LOWflashinfer/cake_fused_qk_rope_fp8_append.py375__all__ = [STRING
LOWflashinfer/deep_gemm.py45logger = logging.getLogger(__name__)CODE
LOWflashinfer/mla_kv_pack.py39logger = logging.getLogger(__name__)CODE
LOWflashinfer/cake_vsa.py836__all__ = ["plan_cake_vsa", "run_cake_vsa"]CODE
LOWflashinfer/prims_ts/__init__.py37__all__ = [CODE
LOWflashinfer/prims_ts/gemm/kernel.py343def _set_io_dtypes(input_dtype_name: str, output_dtype_name: str) -> None:CODE
LOWflashinfer/prims_ts/gemm/__init__.py22__all__ = [CODE
LOWflashinfer/prims_ts/batched_gemm/tmem_sf_resources.py404 def _setup_common(self, context=None) -> None:CODE
LOW…hinfer/prims_ts/batched_gemm/batched_gemm_resources.py71__all__ = [CODE
LOWflashinfer/prims_ts/batched_gemm/smem_sf_resources.py469 def _setup_common(self, context=None) -> None:CODE
LOWflashinfer/prims_ts/batched_gemm/smem_sf_resources.py901 def _setup_common(self, context=None) -> None:CODE
LOWflashinfer/prims_ts/moe/runner.py463 def set_cache_key_static_extras(self, **kwargs) -> None:CODE
LOWflashinfer/prims_ts/moe/__init__.py26__all__ = [CODE
LOWflashinfer/experimental/__init__.py44__all__ = [CODE
LOWflashinfer/experimental/kda_prefill_wrapper/__init__.py31__all__ = ["RecurrentKDAPrefillPlanner"]CODE
LOWflashinfer/experimental/b12x/_api.py153__all__ = [CODE
LOW…er/experimental/b12x/attention/qsa/_sparse_gqa_cute.py12__all__ = [CODE
LOWflashinfer/experimental/b12x/attention/qsa/_tuning.py309__all__ = ["QsaConfig", "QsaQuery", "TUNING"]CODE
LOW…rimental/b12x/attention/qsa/_sparse_gqa_cute_config.py146__all__ = [CODE
768 more matches not shown…
Excessive Try-Catch Wrapping756 hits · 749 pts
SeverityFileLineSnippetContext
LOWbuild_backend.py142 except Exception:CODE
LOWbuild_backend.py221 except Exception:CODE
LOWbuild_backend.py236 except Exception:CODE
LOWbuild_backend.py355 except Exception:CODE
LOWbuild_backend.py432 except Exception:CODE
LOWbuild_backend.py588 except Exception:CODE
LOWbuild_backend.py1009 except Exception as e:CODE
LOWbuild_utils.py116 except Exception:CODE
LOWdocker/test_ci_image.py105 except Exception as error:CODE
LOWtests/kda/test_recurrent_kda_decode_export.py348 except Exception:STRING
LOWtests/kda/test_recurrent_kda_prefill_sm120.py1061 except Exception as exc: # noqa: BLE001 -- the type and text are the assertionCODE
LOWtests/kda/test_recurrent_kda_prefill_sm120.py1447 except Exception as exc: # noqa: BLE001 -- reported belowCODE
MEDIUMtests/kda/test_recurrent_kda_prefill_sm120.py1058def outcome(backend):CODE
LOWtests/prims_ts/test_moe_fp8_block_support.py71 except Exception:CODE
LOW⚡tests/prims_ts/test_moe_bf16_support.py88 except Exception:CODE
LOW…s/experimental/b12x/attention/test_qsa_program_keys.py112 except Exception as error:CODE
LOW…l/b12x/attention/test_attention_mla_sglang_boundary.py32 except Exception as exc: # pragma: no cover - environment-dependent import pathCODE
LOW…ttention/test_attention_dsa_indexer_sglang_boundary.py32 except Exception as exc: # pragma: no cover - environment-dependent import pathCODE
LOW…ttention/test_attention_nsa_backend_sglang_boundary.py22 except Exception as exc: # pragma: no cover - environment-dependent import pathCODE
LOWtests/attention/conftest.py456 except Exception as e:CODE
LOWtests/attention/conftest.py510 except Exception: # pragma: no cover - diagnostics must never fail the runCODE
LOWtests/attention/test_cudnn_decode.py1165 except Exception as exc: # noqa: BLE001 -- the frontend's error types varyCODE
LOWtests/attention/test_deepseek_mla.py106 except Exception as e:CODE
MEDIUMtests/attention/test_deepseek_mla.py46def warmup_jit():CODE
LOW…sts/trace/test_tgv_gemm_sm100_reference_correctness.py37 except Exception as exc:CODE
LOW…sts/trace/test_mxfp8_quantize_reference_correctness.py25 except Exception as exc:CODE
LOWtests/trace/test_mm_bf16_fp4_reference_correctness.py51 except Exception as exc:CODE
LOWtests/trace/test_single_decode_reference_correctness.py31 except Exception as exc:CODE
LOW…ize_fp8_append_paged_kv_cache_reference_correctness.py69 except Exception as exc:CODE
LOWtests/trace/test_bmm_fp8_reference_correctness.py38 except Exception as exc:CODE
LOW…/trace/test_cutlass_fused_moe_reference_correctness.py62 except Exception as exc:CODE
LOW…s/trace/test_segment_gemm_run_reference_correctness.py34 except Exception as exc:CODE
LOW…s/trace/test_block_sparse_run_reference_correctness.py58 except Exception as exc:CODE
LOW…s/trace/test_xqa_batch_decode_reference_correctness.py74 except Exception as exc:CODE
LOW…sts/trace/test_mxfp4_quantize_reference_correctness.py33 except Exception as exc:CODE
LOW…ace/test_var_block_sparse_run_reference_correctness.py67 except Exception as exc:CODE
LOWtests/trace/test_bmm_bf16_reference_correctness.py29 except Exception as exc:CODE
LOWtests/trace/test_mm_bf16_reference_correctness.py29 except Exception as exc:CODE
LOWtests/trace/test_fi_trace_template_consistency.py469 except Exception as exc: # noqa: BLE001CODE
LOW…ace/test_xqa_batch_decode_mla_reference_correctness.py72 except Exception as exc:CODE
LOW…trace/test_cudnn_batch_decode_reference_correctness.py80 except Exception as exc:CODE
LOW⚡tests/trace/example.py668except Exception:CODE
LOW⚡tests/trace/example.py679except Exception:CODE
LOW⚡tests/trace/example.py697except Exception:CODE
LOW⚡tests/trace/example.py707except Exception:CODE
LOWtests/trace/example.py646except Exception:CODE
LOWtests/trace/example.py725except Exception:CODE
LOWtests/trace/example.py739except Exception:CODE
LOWtests/trace/example.py751except Exception:CODE
LOWtests/trace/example.py1696except Exception:CODE
LOW…e/test_trtllm_fmha_v2_prefill_reference_correctness.py62 except Exception as exc:CODE
LOWtests/trace/test_concat_mla_k_reference_correctness.py30 except Exception as exc:CODE
LOW…/test_trtllm_batch_decode_mla_reference_correctness.py68 except Exception as exc:CODE
LOW…sts/trace/test_single_prefill_reference_correctness.py31 except Exception as exc:CODE
LOW…/test_multi_level_cascade_run_reference_correctness.py61 except Exception as exc:CODE
LOW…race/test_cudnn_batch_prefill_reference_correctness.py91 except Exception as exc:CODE
LOW…/test_fused_add_rmsnorm_quant_reference_correctness.py32 except Exception as exc:CODE
LOW…t_pod_with_paged_kv_cache_run_reference_correctness.py70 except Exception as exc:CODE
LOWtests/trace/test_rmsnorm_quant_reference_correctness.py27 except Exception as exc:CODE
LOW…sts/trace/test_nvfp4_quantize_reference_correctness.py26 except Exception as exc:CODE
696 more matches not shown…
Hallucination Indicators84 hits · 720 pts
SeverityFileLineSnippetContext
CRITICALflashinfer/fused_moe/runners.py4624 activation_name = self.config.activation.type.name.lower()CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py783 txl.ptx.st.release.gpu.global_.s32(CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py791 txl.ptx.fence.proxy.async_.shared__cta()CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py798 txl.ptx.tcgen05.wait__ld.sync.aligned()CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1018 txl.ptx.tcgen05.wait__ld.sync.aligned()CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1022 txl.ptx.tcgen05.wait__ld.sync.aligned()CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1026 txl.ptx.tcgen05.wait__st.sync.aligned()CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1075 txl.ptx.ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16(CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1080 txl.ptx.stmatrix.sync.aligned.m8n8.x4.shared.b16(CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1086 txl.ptx.movmatrix.sync.aligned.m8n8.trans.b16(dst[z], src[z])CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1094 txl.ptx.mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32(CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1233 txl.ptx.st.shared.v4.u32(CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1236 txl.ptx.st.shared.v4.u32(CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1239 txl.ptx.fence.proxy.async_.shared__cta()CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1441 txl.ptx.fence.proxy.async_.shared__cta()CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1449 txl.ptx.fence.proxy.async_.shared__cta()CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1451 txl.ptx.stmatrix.sync.aligned.m8n8.x4.shared.b16(CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1458 txl.ptx.fence.proxy.async_.shared__cta()CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1585 txl.ptx.ld.shared.v4.b32(CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1590 txl.ptx.st.shared.v4.b32(CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1596 txl.ptx.rcp.approx.ftz.f32(r, x)CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1653 txl.ptx.ld.global_.v4.f32(CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1660 txl.ptx.ld.global_.v4.f32(CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1667 txl.ptx.st.shared.v4.f32(CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1674 txl.ptx.st.shared.v4.f32(CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1684 txl.ptx.ld.shared.v4.f32(CODE
CRITICAL⚡flashinfer/kda_kernels/tirx/fused.py1687 txl.ptx.ld.shared.v4.f32(CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py241 txl.ptx.fence.proxy.async_.shared__cta()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py243 txl.ptx.fence.mbarrier_init.release.cluster()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py350 txl.ptx.mbarrier.try_wait.parity.shared.b64(CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py435 txl.ptx.cvt.rn.bf16x2.f32(r, hi, lo)CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py455 txl.ptx.ex2.approx.ftz.f32(r, x)CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py507 txl.ptx.mbarrier.arrive.expect_tx.shared.b64(CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py517 txl.ptx.fence.proxy.async_.shared__cta()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py534 txl.ptx.mbarrier.arrive.expect_tx.shared.b64(CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py695 txl.ptx.ld.global_.v4.f32(CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py707 txl.ptx.tcgen05.wait__st.sync.aligned()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py727 txl.ptx.ld.acquire.gpu.global_.s32(fl, flags.ptr_to([cta]))CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py746 txl.ptx.tcgen05.wait__ld.sync.aligned()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py755 txl.ptx.st.global_.v4.f32(CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py824 txl.ptx.stmatrix.sync.aligned.m8n8.x4.trans.shared.b16(CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py831 txl.ptx.fence.proxy.async_.shared__cta()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py851 txl.ptx.cp.async_.bulk.commit_group()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py852 txl.ptx.cp.async_.bulk.wait_group.read(0)CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py869 txl.ptx.ld.shared.v4.b32(CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py872 txl.ptx.st.global_.v4.b32(CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py894 txl.ptx.tcgen05.wait__ld.sync.aligned()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py897 txl.ptx.tcgen05.wait__st.sync.aligned()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py909 txl.ptx.tcgen05.wait__ld.sync.aligned()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py970 txl.ptx.tcgen05.wait__ld.sync.aligned()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py973 txl.ptx.tcgen05.wait__st.sync.aligned()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py990 txl.ptx.ld.shared.v4.f32(CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py1000 txl.ptx.mul.rn.ftz.f32x2(CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py1253 txl.ptx.ld.shared.v2.f32(CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py1266 txl.ptx.tcgen05.wait__ld.sync.aligned()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py1269 txl.ptx.ld.shared.v2.f32(CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py1307 txl.ptx.tcgen05.wait__ld.sync.aligned()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py1340 txl.ptx.stmatrix.sync.aligned.m8n8.x4.shared.b16(CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py1393 txl.ptx.fence.proxy.async_.shared__cta()CODE
CRITICALflashinfer/kda_kernels/tirx/fused.py1486 txl.ptx.rsqrt.approx.ftz.f32(norm, n0 + n1 + txl.float32(EPS))CODE
24 more matches not shown…
AI Structural Patterns855 hits · 656 pts
SeverityFileLineSnippetContext
LOWtests/kda/test_recurrent_kda_decode_export.py1333CODE
LOWtests/kda/test_recurrent_kda_cudnn_backend.py83CODE
LOWtests/prims_ts/test_batched_gemm_fp4_fc2.py84CODE
LOWtests/prims_ts/test_batched_gemm_bf16_fc2.py66CODE
LOWtests/prims_ts/test_batched_gemm_fp4_fc1.py97CODE
LOWtests/prims_ts/test_batched_gemm_bf16_fc1.py224CODE
LOWtests/prims_ts/test_batched_gemm_bf16_fc1.py329CODE
LOWtests/gdp/test_prefill_gdp_cudnn_backend.py62CODE
LOWtests/experimental/test_cake_lm_head_loss.py163CODE
LOW…al/b12x/attention/test_dsa_indexer_mxfp4_boundaries.py98CODE
LOW…mental/b12x/attention/test_attention_mla_compressed.py161CODE
LOW…s/experimental/b12x/gemm/test_dense_gemm_expected_m.py1263CODE
LOWtests/experimental/b12x/sequence/test_kda_prefill.py61CODE
LOWtests/experimental/b12x/sequence/test_gdn_decode_kda.py46CODE
LOWtests/experimental/b12x/sequence/test_gdn_decode.py78CODE
LOWtests/experimental/b12x/preparation/test_session.py186CODE
LOW…sts/experimental/b12x/moe/test_dynamic_w4a8_trellis.py51CODE
LOWtests/experimental/b12x/moe/test_route_pack_cache.py7CODE
LOWtests/experimental/b12x/moe/test_w4a8_dynamic_kernel.py100CODE
LOWtests/attention/test_modular_fmha_prefill_paged.py127CODE
LOWtests/attention/test_attention_ts_decode.py1674CODE
LOWtests/attention/test_attention_ts_decode.py6091CODE
LOWtests/attention/test_dcp_spec_fp8.py624CODE
LOWtests/attention/test_cake_fmha.py3076CODE
LOWtests/attention/test_attention_ts_block_sparse.py948CODE
LOWtests/attention/test_sparse_mla_blackwell_dsv4_nvfp4.py187CODE
LOWtests/attention/test_fmha_v2_prefill.py64CODE
LOWtests/attention/test_xqa.py124CODE
LOWtests/attention/test_bidirectional_ranges_attention.py139CODE
LOW…/test_attention_ts_q_token_kv_block_sparse_metadata.py76CODE
LOWtests/attention/test_cute_dsl_mla_decode.py327CODE
LOWtests/attention/test_trtllm_ragged_kv_stride.py40CODE
LOWtests/attention/test_trtllm_gen_attention_decode.py627CODE
LOWtests/attention/test_batch_attention.py95CODE
LOWtests/attention/test_cudnn_decode.py1320CODE
LOWtests/msa_ops/test_msa_nvfp4_decode_sm100.py578CODE
LOWtests/msa_ops/test_blackwell_msa_sm100.py31CODE
LOWtests/moe_ep/test_nvfp4_cutedsl_kernel_vs_reference.py68CODE
LOW…sts/moe_ep/test_moe_ep_sm90_pull_fp8_mega_multirank.py381CODE
LOW…sts/moe_ep/test_moe_ep_sm90_pull_fp8_mega_multirank.py431CODE
LOWtests/moe_ep/test_sm90_push_fp8_kernel.py109CODE
LOWtests/moe_ep/test_sm90_push_fp8_gemm_contract.py442CODE
LOWtests/diffusion_ops/test_minimax_h3_mlp.py1119CODE
LOWtests/mamba/test_chunk_scan_combined.py27CODE
LOWtests/mamba/test_cake_ssd_combined.py326CODE
LOWtests/mamba/utils.py16CODE
LOWtests/mamba/test_checkpointing_ssu.py184CODE
LOWtests/mamba/test_replayssm_materialize.py33CODE
LOW…amba/triton_reference/selective_state_update_varlen.py272CODE
LOWtests/mamba/triton_reference/selective_state_update.py431CODE
LOW…amba/triton_reference/replay_selective_state_update.py1976CODE
LOW…amba/triton_reference/replay_selective_state_update.py3653CODE
LOWtests/mamba/triton_reference/ssd_combined.py32CODE
LOWtests/mamba/triton_reference/ssd_combined.py197CODE
LOWtests/mamba/triton_reference/causal_conv1d_triton.py1004CODE
LOWtests/comm/test_all_gather_matmul.py45CODE
LOWtests/comm/test_mixed_comm.py36CODE
LOWtests/utils/test_concat_mla_kv_quant_fp8.py53CODE
LOWtests/mla/test_cake_dsv4_hardening.py863CODE
LOWtests/gemm/test_cute_dsl_gemm_allreduce_two_shot.py139CODE
795 more matches not shown…
Empty Error Handlers120 hits · 600 pts
SeverityFileLineSnippetContext
HIGHbuild_backend.py142except Exception:CODE
HIGHbuild_backend.py588except Exception:CODE
HIGHflashinfer-cubin/flashinfer_cubin/__init__.py55except ImportError:CODE
HIGHflashinfer-cubin/flashinfer_cubin/__init__.py72except (ImportError, AttributeError):CODE
HIGHtests/conftest.py208except ImportError:CODE
HIGHtests/experimental/b12x/moe/test_nvfp4_split_backend.py481except ValueError:CODE
HIGHtests/attention/conftest.py510except Exception: # pragma: no cover - diagnostics must never fail the runCODE
HIGHtests/attention/conftest.py294except ImportError: # pragma: no cover - fixture still builds theseCODE
HIGHtests/trace/example.py646except Exception:CODE
HIGHtests/trace/example.py668except Exception:CODE
HIGHtests/trace/example.py679except Exception:CODE
HIGHtests/trace/example.py697except Exception:CODE
HIGHtests/trace/example.py707except Exception:CODE
HIGHtests/trace/example.py725except Exception:CODE
HIGHtests/trace/example.py739except Exception:CODE
HIGHtests/trace/example.py751except Exception:CODE
HIGHtests/msa_ops/test_msa_nvfp4_decode_sm100.py527except _LeaveBlock:CODE
HIGHtests/moe_ep/test_moe_ep_compute_correctness.py129except ImportError:CODE
HIGHtests/comm/test_ulysses_communicator.py1263except Boom:CODE
HIGHtests/comm/test_ulysses_communicator.py1446except queue_mod.Empty:CODE
HIGHtests/comm/test_ulysses_topology.py378except queue_mod.Empty:CODE
HIGHtests/gemm/test_ragged_block_scaled_bmm_cutile.py111except Exception:CODE
HIGHbenchmarks/bench_cake_fused_kda_decode.py603except ValueError:CODE
HIGHbenchmarks/bench_cake_fused_kda_decode.py1276except ValueError:CODE
HIGHbenchmarks/bench_cudnn_wrapper_host.py53except (OSError, subprocess.CalledProcessError):CODE
HIGHbenchmarks/bench_cake_lm_head_loss.py143except (TypeError, RuntimeError):CODE
HIGHbenchmarks/bench_checkpointing_ssu.py1253except Exception:CODE
HIGHbenchmarks/experimental/b12x/benchmark_roce_oneshot.py74except OSError:CODE
HIGHbenchmarks/experimental/b12x/benchmark_roce_oneshot.py141except (OSError, subprocess.SubprocessError):CODE
HIGHbenchmarks/experimental/b12x/benchmark_mla.py841except Exception:CODE
HIGHbenchmarks/routines/attention.py18except ImportError:CODE
HIGHscripts/run_with_output_watchdog.py247except (OSError, subprocess.TimeoutExpired):CODE
HIGHscripts/unit_test_runner.py204except Exception:CODE
HIGHscripts/pr_checks/check_api_docs.py47except Exception:CODE
HIGHscripts/pr_checks/experimental_test_scope.py295except ValueError:CODE
HIGHscripts/pr_checks/experimental_test_scope.py308except ValueError:CODE
HIGHscripts/pr_checks/experimental_test_scope.py333except ValueError:CODE
HIGHscripts/pr_checks/experimental_test_scope.py258except ValueError:CODE
HIGHscripts/test_sharding/runner.py1011except queue.Empty:CODE
HIGHscripts/test_sharding/junit.py152except (OSError, AttributeError, json.JSONDecodeError):CODE
HIGHscripts/test_sharding/workers.py187except (OSError, ValueError):CODE
HIGHscripts/test_sharding/state.py43except FileExistsError:CODE
HIGHflashinfer/api_logging.py2515except Exception:CODE
HIGHflashinfer/collect_env.py461except Exception:CODE
HIGHflashinfer/__init__.py285except (ImportError, AttributeError):CODE
HIGHflashinfer/__init__.py432except Exception:CODE
HIGHflashinfer/fi_trace.py139except AttributeError:CODE
HIGHflashinfer/fi_trace.py164except Exception:CODE
HIGHflashinfer/experimental/b12x/_lib/compiler.py1573except Exception:CODE
HIGHflashinfer/experimental/b12x/_lib/compiler.py1766except (TypeError, ValueError):CODE
HIGHflashinfer/experimental/b12x/loader/_progress.py206except (AttributeError, OSError, ValueError):CODE
HIGHflashinfer/experimental/b12x/sequence/engram/_disk.py128except Exception:CODE
HIGHflashinfer/experimental/b12x/sequence/_shared/_gds.py156except Exception:CODE
HIGHflashinfer/experimental/b12x/sequence/_shared/_gds.py140except Exception:CODE
HIGHflashinfer/experimental/b12x/preparation/_progress.py378except (AttributeError, OSError, ValueError):CODE
HIGH…shinfer/experimental/cake_lm_head_loss/cake_backend.py3131except (TypeError, RuntimeError):CODE
HIGH…ion/prims_ts/kernels/fmha_decode/fmha_decode_config.py3760except (ImportError, RuntimeError):CODE
HIGH…ion/prims_ts/kernels/fmha_decode/fmha_decode_config.py4237except ValueError:CODE
HIGHflashinfer/trace/template.py744except Exception:CODE
HIGHflashinfer/moe_ep/core/validation/common.py174except Exception:CODE
60 more matches not shown…
AI Slop Vocabulary101 hits · 300 pts
SeverityFileLineSnippetContext
MEDIUMtests/experimental/test_cake_dsa_train.py68# for the iid top-k-2048 configuration and are enforced by the project harness, not per test shape.COMMENT
MEDIUMtests/experimental/test_cake_dsa_train.py72# harness values); rope parts at twice the latent gate, peaked attention at one common gate.COMMENT
MEDIUMtests/experimental/b12x/attention/test_fused_indexer.py347 # value-multiset parity (tie-robust); fp8 e4m3 dot in f32 matches the MMA.COMMENT
MEDIUMtests/experimental/b12x/attention/test_fused_indexer.py717 # returned index (rather than asserting an index order) keeps this robustCOMMENT
MEDIUM…ental/b12x/attention/test_attention_dsa_indexer_api.py1470 # tie-robust check is that the selected logit-VALUE multiset matches theCOMMENT
MEDIUMtests/experimental/b12x/moe/test_nvfp4_phase_kernels.py556 # direct torch chain must match the oracle too (sanity of the harness)COMMENT
MEDIUM⚡tests/msa_ops/test_msa_nvfp4_decode_sm100.py407# same numbers a consumer's capability probe and the kernel-level A/B harnessCOMMENT
MEDIUM⚡tests/msa_ops/test_msa_nvfp4_decode_sm100.py1856 # A warm launch is not a caller's call, and the A/B harness differencesCOMMENT
MEDIUMtests/autotuner/test_autotuner_distributed.py226# independent of how the harness imported the tests (a ``spawn`` worker thatCOMMENT
MEDIUMtests/mamba/test_philox_rounding.py794 """Software e4m3 SR matches hardware PTX path bitwise across a comprehensiveSTRING
LOWtests/mamba/triton_reference/causal_conv1d_triton.py171 # first chunk and does not have prior-token, so just set to 0COMMENT
MEDIUM⚡tests/gemm/test_unified_gemm_fuzz.py595# + tight tolerance. Same model as the MoE harness: with exact-grid inputs the oracle measures theCOMMENT
MEDIUMtests/gemm/test_unified_gemm_fuzz.py549 # layout_128x4 / mxfp4 default this harness prepares). Toggling it on the cutlass/cudnn/autoCOMMENT
MEDIUM⚡tests/gdn/test_decode_delta_rule.py1328# Test MTP kernel with FP32 state, cache ON, state update ON (comprehensive)COMMENT
LOWtests/test_helpers/alibi_reference.py78 # Since it's causal mask we can just use $[0, 1, \dots, N]$ too.STRING
MEDIUMtests/moe/test_trtllm_moe_valid_dims.py187 # harness / CUDA-only deps on machines that only run the unit-level checks above.COMMENT
MEDIUMtests/moe/test_trtllm_moe_valid_dims.py523# GEMM2 output width equal to hidden_size, which the harness FP4COMMENT
MEDIUMtests/moe/test_trtllm_moe_valid_dims.py705# The harness FP4 prepare_static_weights_for_kernel reshapes GEMM2 by a singleCOMMENT
MEDIUM⚡tests/moe/test_unified_moe_fuzz.py3441# different shape would produce a wrong answer here. (Shares the harness's snap/reference/prep.)COMMENT
MEDIUMtests/moe/test_unified_moe_fuzz.py1582# Covers the in-kernel-routing bug cluster the pre-routed harness structurally can't reach:COMMENT
MEDIUMtests/moe/test_unified_moe_fuzz.py1738 # separate validation, and EP collectives are out of scope for this single-GPU harness.COMMENT
MEDIUMtests/moe/test_unified_moe_fuzz.py1888 # In-kernel routing (FromLogits) headline cases the pre-routed harness can't reach:COMMENT
MEDIUMtests/moe/test_unified_moe_fuzz.py1585 RoutingMethodType.RenormalizeNaive, # == the harness's original host routingCODE
MEDIUM⚡benchmarks/bench_mamba_ssd_combined.py257# Layer 2: bench_mode — batched or varlen harnessCOMMENT
MEDIUMbenchmarks/bench_silu_and_mul_nvfp4_quantize.py132 # more comprehensive correctness coverage.COMMENT
MEDIUMbenchmarks/cake_warp_decode.py800 # ambiguous stale second release while the harness unwinds.COMMENT
MEDIUMbenchmarks/bench_recurrent_kda_prefill.py660 # harness while keeping route logging out of every timed call.COMMENT
MEDIUM…chmarks/experimental/b12x/benchmark_paged_attention.py2069 # copies introduced solely by the benchmark harness.COMMENT
MEDIUMbenchmarks/routines/kda.py329 # difference between the two APIs, not an artifact of this harness, and itCOMMENT
MEDIUMbenchmarks/routines/kda.py560 # published combination. They need a protocol this harness cannotCOMMENT
MEDIUMbenchmarks/routines/kda.py567 # what this harness is actually able to set up.COMMENT
MEDIUMscripts/task_test_moe_ep_smoke.sh4# Run the moe_ep smoke + multi-rank harness via torchrun.COMMENT
LOWflashinfer/api_logging.py1569 # The safest way is to just call the class constructor.COMMENT
MEDIUMflashinfer/aot.py350 # The holistic (persistent) batch-attention kernelCOMMENT
MEDIUMflashinfer/cake_sampling.py591# in the interleaved CUPTI harness, +0.8 / +2.4 % in two idle-node matrices at k = 10 / 50), while the 16 %-fullCOMMENT
MEDIUMflashinfer/autotune_cache.py104 # execution_mode. Used by the accuracy harness to A/BCOMMENT
MEDIUM⚡flashinfer/prims_ts/batched_gemm/batched_gemm_run.py3360# Perf-harness interface (``@pytest.mark.perf`` profiler sweep).COMMENT
MEDIUMflashinfer/experimental/b12x/comm/pcie/pcie_oneshot.py382 # `broadcast_object_list` is more robust than `all_gather_object` forCOMMENT
MEDIUM…shinfer/experimental/b12x/testing/reference/glm_ref.py174 # Random KV in the same regime as the DSV4 harness (randn/10 clamp(-1,1)).COMMENT
MEDIUM…fer/experimental/b12x/testing/reference/prefill_ref.py85# harness exercises the kernel's last-partial-tile masking.COMMENT
MEDIUM…fer/experimental/b12x/testing/reference/prefill_ref.py354 # Confirm the harness really produced non-multiple-of-64 lengths.COMMENT
MEDIUM…fer/experimental/b12x/testing/reference/prefill_ref.py381 # correct shapes, and the harness emits per-token non-mult-of-64 lengths.COMMENT
MEDIUMflashinfer/attention/_core.py181 # head_dim > 256 is for holistic (persistent) kernel.STRING
MEDIUMflashinfer/msa_ops/_nvfp4_decode_sm100.py243# about callers: the kernel-level A/B harness attests dispatch by differencingCOMMENT
MEDIUMflashinfer/msa_ops/_nvfp4_prefill_sm100.py227# about callers: the kernel-level A/B harness attests dispatch by differencingCOMMENT
MEDIUM…ashinfer/msa_ops/cute_dsl/sparse_decode_nvfp4_sm100.py1266 # one loop body to hide a COLD round trip behind -- the harnessCOMMENT
MEDIUM…ashinfer/msa_ops/cute_dsl/sparse_decode_nvfp4_sm100.py1396 # cold L2 (the harness flushes it before every iteration) that roundCOMMENT
MEDIUMflashinfer/moe_ep/core/validation/common.py238# aborts the process from C++ with no Python traceback -- under a test harnessCOMMENT
MEDIUM…_megakernel/src/moe_hopper_fp8/run_functional_tests.sh2# Functional test harness for the FP8 GLU fused fc1+fc2 runner.COMMENT
MEDIUM…utedsl_megakernel/src/moe_hopper_fp8/run_mega_tests.sh2# Functional test harness for the FP8 distributed MegaMoE fused dispatch + fc1/fc2COMMENT
MEDIUM…cutedsl_megakernel/src/moe_hopper_fp8/run_perf_test.sh2# DSV4 performance harness for moe_hopper_fp8.COMMENT
MEDIUM⚡…l_megakernel/src/moe_nvfp4_swapab/epilogue_refactor.py944 _EpilogueTokenTileSize = 64 # Fundamentally the epi_tile_nCODE
MEDIUM⚡…l_megakernel/src/moe_nvfp4_swapab/epilogue_refactor.py945 _EpilogueFc1IntermediateGateUpTileSize = 128 # Fundamentally epi_tile_mCODE
MEDIUM⚡…l_megakernel/src/moe_nvfp4_swapab/epilogue_refactor.py946 _EpilogueFc1IntermediateDownTileSize = 64 # Fundamentally epi_tile_m // 2CODE
MEDIUM⚡…l_megakernel/src/moe_nvfp4_swapab/epilogue_refactor.py947 _EpilogueFc2HiddenTileSize = 128 # Fundamentally epi_tile_mCODE
MEDIUM…egakernel/src/moe_nvfp4_swapab/run_functional_tests.sh2# Functional test harness for the local fused fc1+fc2 MoE NVFP4 swap-AB runner.COMMENT
MEDIUM…edsl_megakernel/src/moe_nvfp4_swapab/run_mega_tests.sh2# Functional test harness for the distributed MegaMoE fused dispatch + fc1/fc2COMMENT
MEDIUM…cutedsl_megakernel/src/moe_nvfp4_swapab/mega_runner.py1981 # to avoid recomputing it the harness keeps the globalCOMMENT
MEDIUM⚡…ull_style_cutedsl_megakernel/src/src/inputs_process.py579# Correctness harness (GPU only; needs cutlass + torch fp4/e8m0 dtypes).COMMENT
MEDIUM…ull_style_cutedsl_megakernel/src/src/inputs_process.py728 except Exception as exc: # noqa: BLE001 -- smoke harness, report and continueCODE
41 more matches not shown…
Cross-Language Confusion59 hits · 290 pts
SeverityFileLineSnippetContext
HIGHtests/experimental/test_cake_lm_head_loss.py1735 # and in the fp32 fused cast; the bf16 cast keeps the 1-D raster; a pinned ``group_n`` (0 or null) keeps the defaultCOMMENT
HIGH⚡…ts/experimental/b12x/preparation/test_delta_prefill.py46 tensors["recurrent_state"][null].fill_(float("nan"))CODE
HIGH⚡…ts/experimental/b12x/preparation/test_delta_prefill.py47 tensors["initial_state_indices"].fill_(null)CODE
HIGH⚡…ts/experimental/b12x/preparation/test_delta_prefill.py53 expected, state = oracle(case, tensors, null_state_index=null)CODE
HIGHtests/attention/test_cake_fmha.py366 " CAKE_FMHA_HAS_SINK == 1 && CAKE_FMHA_HAS_WINDOW == 0 && "CODE
HIGHtests/attention/test_cake_fmha.py386 " CAKE_FMHA_HAS_SINK == 0 && CAKE_FMHA_HAS_WINDOW == 0 && "CODE
HIGH⚡tests/trace/test_paged_mqa_logits_checker.py76 With rnorm == 0 the relative error is undefined; substituting 0.0 madeSTRING
HIGHtests/mamba/test_cake_ssd_combined.py1700 state undefined) -- both documented in D12, both flagged. A clean callSTRING
HIGHtests/comm/test_trtllm_allreduce_reduction.py43 TORCH_CHECK(num_threads > 0 && num_threads <= 1024, "invalid block size");CODE
HIGHtests/comm/test_trtllm_mnnvl_allreduce.py306 ordinary data; a `val == 0.F && signbit(val)` check would FTZ-flushSTRING
HIGHtests/comm/test_mnnvl_moe_alltoall_fused_adapter.py649 assert "if (bid == 0 && blockIdx.y == 0)" in topk6_sourceCODE
HIGHtests/utils/test_jit_example.py189 if (qo_idx < qo_len && kv_idx < kv_len) {CODE
HIGHtests/gdn/test_decode_ucache.py57 source env.sh && pytest tests/gdn/test_decode_ucache.py -vSTRING
HIGHtests/jit/test_flash_kda_packed_t1_jit.py206 assert "minor == 0 || minor == 3 || minor == 7" in bindingCODE
HIGHtests/jit/test_cake_bgmv_moe_jit.py147 "major == CAKE_BGMV_MOE_CC_MAJOR && minor == CAKE_BGMV_MOE_CC_MINOR" in bindingCODE
HIGHtests/jit/test_cake_bgmv_moe_jit.py410 "major == CAKE_BGMV_MOE_CC_MAJOR && minor == CAKE_BGMV_MOE_CC_MINOR" in bindingCODE
HIGHtests/jit/test_flash_kda_decode_jit.py143 assert "else if (has_token && k_lane == 0)" in generated_bodyCODE
HIGH⚡tests/jit/test_flash_kda_decode_jit.py189 assert "major == 10 && (minor == 0 || minor == 3)" in commonCODE
HIGH⚡tests/jit/test_flash_kda_decode_jit.py190 assert "major == 10 && minor == expected_minor" in commonCODE
HIGH⚡tests/jit/test_flash_kda_decode_jit.py194 assert "GateKind == 0 || GateKind == 1 || GateKind == 2" in commonCODE
HIGHbenchmarks/bench_monomoe.py86 Uses mono_moe's `[gate || up]` fc1 half-ordering and the given (alreadySTRING
HIGHbenchmarks/bench_gdn_ucache_flush.py30 source env.sh && python benchmarks/bench_gdn_ucache_flush.py [--iters 200]STRING
HIGHbenchmarks/routines/attention.py4045 # Leave empty (null) when not explicitly overridden so legacyCOMMENT
HIGHflashinfer/collect_env.py156 _run("grep -sq -e docker -e containerd -e kubepods /proc/1/cgroup && echo 1")CODE
HIGHflashinfer/kda_decode.py1033 # slots are padding (the vLLM drop-in reserves slot 0 as null).STRING
HIGHflashinfer/prims_ts/moe/runner.py219 """Pick finalize weights. Match C++ ``ndim()==2 && size(0)>0``.STRING
HIGH…/experimental/b12x/attention/_shared/mla/prefill_mg.py181 XV path enables it (USE_WFP8_ROW_XOR = DUAL_CACHE && pbs_extra==2,STRING
HIGH…fer/experimental/b12x/testing/reference/prefill_ref.py38 if attn_sink != null:STRING
HIGH…imental/b12x/sequence/gdn_prefill/_parallel_kernels.py600 result = slot == Int64(self.null)CODE
HIGH…hinfer/experimental/b12x/sequence/gdn_prefill/_impl.py83 object.__setattr__(self, "null_state_index", null)CODE
HIGH…hinfer/experimental/b12x/sequence/kda_prefill/_impl.py66 object.__setattr__(self, "null_state_index", null)CODE
HIGH…er/experimental/b12x/sequence/kda_prefill/reference.py117 initial state from ``initial_state_indices[i]`` (zero when null), writes itsSTRING
HIGH…experimental/b12x/moe/_shared/kernels/w4a16/prepare.py2239 ``payload`` uses the reference container order ``record0 || record1`` andSTRING
HIGH…experimental/b12x/moe/_shared/kernels/w4a16/prepare.py2241 compact three-bpw byte count and record0 || record1 channel semantics. ForSTRING
HIGHflashinfer/trace/templates/attn_scores.py227 # An all-zero reference makes the relative error undefined; the oldCOMMENT
HIGHflashinfer/trace/templates/moe.py3440 description="Up/gate projection weights (FP8, [gate || up] rows).",CODE
HIGHflashinfer/trace/templates/moe.py3291 up-projection weights use the [gate || up] half ordering (monomoe'sSTRING
HIGH…90/pull_style_cutedsl_megakernel/src/src/flag_batch.py21 flag_addr: Int64 # per-lane counter-slot address (0 == null)CODE
HIGH…c/sm120/split_cutedsl_megakernel/src/src/flag_batch.py21 flag_addr: Int64 # per-lane counter-slot address (0 == null)CODE
HIGH…/sm120/swapab_cutedsl_megakernel/src/src/flag_batch.py21 flag_addr: Int64 # per-lane counter-slot address (0 == null)CODE
HIGH…kernel_src/sm100/cutedsl_megamoe/src/src/flag_batch.py21 flag_addr: Int64 # per-lane counter-slot address (0 == null)CODE
HIGH⚡…ashinfer/gdn_kernels/gdn_decode_bf16_wy_output_only.py1049 # _v7e_a/b regs are undefined), so they must keep log_alpha=beta=0. The causalCOMMENT
HIGH⚡…shinfer/gdn_kernels/gdn_decode_bf16_wy_ucache_flush.py1671 # _v7e_a/b regs are undefined), so they must keep log_alpha=beta=0. The causalCOMMENT
HIGH⚡flashinfer/gdn_kernels/gdn_decode_bf16_wy_ucache.py1139 # _v7e_a/b regs are undefined), so they must keep log_alpha=beta=0. The causalCOMMENT
HIGH⚡…hinfer/fused_moe/cake_cute/cake_mxfp4_situ_moe_plan.py3761 """K3 build-time variant: ``kPdl`` and whether the ``all`` list pointers are given (non-null)."""STRING
HIGHflashinfer/fused_moe/cute_dsl/moe_utils.py417 ``!enable_alltoall || ep_size <= top_k`` branch). FunctionallySTRING
HIGH⚡flashinfer/fused_moe/cake/cake_mxfp4_situ_moe_plan.py3761 """K3 build-time variant: ``kPdl`` and whether the ``all`` list pointers are given (non-null)."""STRING
HIGHflashinfer/mla/_core.py4897 Accepts the concatenated ``(q_nope || q_rope)`` query and ``(ckv || kpe)`` paged KVCODE
HIGH⚡flashinfer/topk_varlen/kernels/gvr2_topk_decode.py1221 HB=NBS; KBIG=(KPT>=2 && KPT*BLK>=2048); SCPB=(BLK>=1024)?(SPLIT?8192:16384)STRING
HIGHflashinfer/jit/attention/variants.py39 float log_sink = (kv_tile_idx == 0 && qo_head_idx < params.num_qo_heads) ? params.sink[qo_head_idx] * math::log2e : CODE
HIGHflashinfer/jit/attention/variants.py219 return causal || in_range;CODE
HIGHflashinfer/jit/attention/fmha_v2/utils.py543 code = "if (!{has_noloop} || (!force_unroll && (ignore_b1opt || b > {unroll_threshold})))".format(STRING
HIGHflashinfer/jit/attention/fmha_v2/utils.py547 code = "if (!{has_noloop} || (!force_unroll && (ignore_b1opt || b * h > {unroll_threshold})))".format(STRING
HIGHflashinfer/jit/attention/fmha_v2/generator_utils.py1543 USE_TMA_STORE && false,CODE
HIGHflashinfer/jit/attention/fmha_v2/generator_utils.py1564 USE_TMA_STORE && false,CODE
HIGHflashinfer/jit/attention/fmha_v2/generator_utils.py2424 code = "if (!{has_noloop} || (!force_unroll && (ignore_b1opt || b > {unroll_threshold})))".format(CODE
HIGHflashinfer/jit/attention/fmha_v2/generator_utils.py2428 code = "if (!{has_noloop} || (!force_unroll && (ignore_b1opt || b * h > {unroll_threshold})))".format(CODE
HIGHflashinfer/jit/attention/fmha_v2/fmha_library.py761 code = "if (!{has_noloop} || (!force_unroll && (ignore_b1opt || b > {unroll_threshold})))".format(CODE
HIGHflashinfer/jit/attention/fmha_v2/fmha_library.py765 code = "if (!{has_noloop} || (!force_unroll && (ignore_b1opt || b * h > {unroll_threshold})))".format(CODE
Structural Annotation Overuse195 hits · 286 pts
SeverityFileLineSnippetContext
LOWtests/attention/test_rope.py915 # Step 1: Pre-populate cache with existing tokensCOMMENT
LOWtests/attention/test_rope.py1126 # Step 2: Append new tokens to the pre-populated cacheCOMMENT
LOW…_quantize_fp8_append_paged_kv_cache_h8_kv2_rope64.json194 "reference": "from __future__ import annotations\nimport math\nimport torch\nimport torch.nn.functional as F\n\n@torchCODE
LOW…tize_fp8_append_paged_kv_cache_h8_kv2_rope64_ps16.json195 "reference": "from __future__ import annotations\nimport math\nimport torch\nimport torch.nn.functional as F\n\n@torchCODE
LOW⚡tests/utils/test_fp4_quantize.py315 # Step 1: Quantize with fp4_quantizeCOMMENT
LOW⚡tests/utils/test_fp4_quantize.py325 # Step 2: Dequantize with e2m1_and_ufp8sf_scale_to_floatCOMMENT
LOWtests/utils/test_fp4_quantize.py340 # Step 3: Compare resultsCOMMENT
LOW⚡tests/gdn/reference_delta_rule.py1081 # Step 1: Compute g = -exp(A_log) * softplus(a + dt_bias)COMMENT
LOW⚡tests/gdn/reference_delta_rule.py1136 # Step 1: Apply gating to hidden state: h *= exp(g)COMMENT
LOW⚡tests/gdn/reference_delta_rule.py1139 # Step 2: Delta rule: v -= k^T @ h ([B,H,K] x [B,H,K,V] -> [B,H,V])COMMENT
LOW⚡tests/gdn/reference_delta_rule.py1142 # Step 3: Apply beta gating: v *= betaCOMMENT
LOW⚡tests/gdn/reference_delta_rule.py1145 # Step 4: Update hidden state with outer product: h += k ⊗ v_newCOMMENT
LOW⚡tests/gdn/reference_delta_rule.py1148 # Step 5: Compute output: o = q^T @ hCOMMENT
LOWtests/gdn/reference_delta_rule.py1098 # Step 2: Compute beta = sigmoid(b)COMMENT
LOW.claude/skills/self-review/SKILL.md16## Step 1: Get the diffCOMMENT
LOW.claude/skills/self-review/SKILL.md29## Step 2: Apply the review guidanceCOMMENT
LOW.claude/skills/self-review/SKILL.md52## Step 3: Apply the PR rulesCOMMENT
LOW.claude/skills/self-review/SKILL.md71## Step 4: Report, then fixCOMMENT
LOW.claude/skills/benchmark-kernel/SKILL.md63### Step 1: Choose Your Test RoutineCOMMENT
LOW.claude/skills/benchmark-kernel/SKILL.md70### Step 2: Run a Single BenchmarkCOMMENT
LOW.claude/skills/benchmark-kernel/SKILL.md118### Step 3: Understand the OutputCOMMENT
LOW.claude/skills/benchmark-kernel/SKILL.md134### Step 4: Run Batch BenchmarksCOMMENT
LOW.claude/skills/benchmark-kernel/SKILL.md156### Step 5: Common FlagsCOMMENT
LOW.claude/skills/benchmark-kernel/SKILL.md174### Step 1: Write Your Benchmark ScriptCOMMENT
LOW.claude/skills/benchmark-kernel/SKILL.md210### Step 2: Run Your BenchmarkCOMMENT
LOW.claude/skills/benchmark-kernel/SKILL.md229### Step 3: Advanced OptionsCOMMENT
LOW.claude/skills/debug-cuda-crash/SKILL.md24## Step 1: Enable API LoggingCOMMENT
LOW.claude/skills/debug-cuda-crash/SKILL.md109## Step 2: Reproduce the CrashCOMMENT
LOW.claude/skills/debug-cuda-crash/SKILL.md146## Step 3: Common CUDA Errors and How to DebugCOMMENT
LOW.claude/skills/debug-cuda-crash/SKILL.md249## Step 4: Multi-Process DebuggingCOMMENT
LOW.claude/skills/debug-cuda-crash/SKILL.md268## Step 5: Advanced Debugging with compute-sanitizerCOMMENT
LOW.claude/skills/debug-cuda-crash/SKILL.md309## Step 6: Kernel-Level Debugging with printf()COMMENT
LOW.claude/skills/add-cuda-kernel/SKILL.md18## Step 1: Define CUDA Kernel in `include/`COMMENT
LOW.claude/skills/add-cuda-kernel/SKILL.md76## Step 2: Create Launcher in `csrc/`COMMENT
LOW.claude/skills/add-cuda-kernel/SKILL.md163## Step 3: Create TVM-FFI Binding in `csrc/`COMMENT
LOW.claude/skills/add-cuda-kernel/SKILL.md183## Step 4: Create JIT Generator (No Jinja for Simple Case)COMMENT
LOW.claude/skills/add-cuda-kernel/SKILL.md387## Step 5: Create Python API in `flashinfer/`COMMENT
LOW.claude/skills/add-cuda-kernel/SKILL.md628## Step 6: Add a Trace TemplateCOMMENT
LOW.claude/skills/add-cuda-kernel/SKILL.md776## Step 7: Write Tests in `tests/`COMMENT
LOW.claude/skills/add-cuda-kernel/SKILL.md846## Step 7: Register in AOTCOMMENT
LOW.claude/skills/add-cuda-kernel/SKILL.md872## Step 8: Export APICOMMENT
LOW.claude/skills/add-cuda-kernel/SKILL.md883## Step 9: Run and TestCOMMENT
LOW.claude/skills/add-cuda-kernel/SKILL.md893## Step 10: Add BenchmarkCOMMENT
LOWdocs/tutorials/jax_tvm_ffi/flashinfer_jax_tvm_ffi.py176# Step 1 - Compile and load the TVM FFI moduleCOMMENT
LOWdocs/tutorials/jax_tvm_ffi/flashinfer_jax_tvm_ffi.py197# Step 2 - Register as a JAX FFI targetCOMMENT
LOWdocs/tutorials/jax_tvm_ffi/flashinfer_jax_tvm_ffi.py240# Step 3 - Call as a regular JAX expressionCOMMENT
LOW⚡benchmarks/bench_cute_dsl_add_rmsnorm_fp4quant.py149 # Step 1: AddCOMMENT
LOW⚡benchmarks/bench_cute_dsl_add_rmsnorm_fp4quant.py151 # Step 2: RMSNormCOMMENT
LOW⚡benchmarks/bench_cute_dsl_add_rmsnorm_fp4quant.py153 # Step 3: FP4 quantize (with global_scale for NVFP4)COMMENT
LOW⚡benchmarks/gdn_triton_reference.py179 # Step 1: Apply decay to hidden state: h *= exp(g)COMMENT
LOW⚡benchmarks/gdn_triton_reference.py182 # Step 2: Delta rule: v -= sum(h * k, dim=0) = k @ hCOMMENT
LOW⚡benchmarks/gdn_triton_reference.py187 # Step 3: Apply beta gating: v *= betaCOMMENT
LOW⚡benchmarks/gdn_triton_reference.py190 # Step 4: Update hidden state: h += outer(k, v) = k[:, None] * v[None, :]COMMENT
LOW⚡benchmarks/gdn_triton_reference.py193 # Step 5: Compute output: o = q @ h = sum(q[:, None] * h, dim=0)COMMENT
LOW⚡benchmarks/gdn_triton_reference.py361 # Step 1: Apply decay: h *= exp(g)COMMENT
LOW⚡benchmarks/gdn_triton_reference.py364 # Step 2: Delta rule: v -= h @ k (h is [BV, BK], k is [BK])COMMENT
LOW⚡benchmarks/gdn_triton_reference.py369 # Step 3: Apply beta gatingCOMMENT
LOW⚡benchmarks/gdn_triton_reference.py372 # Step 4: Update state: h += outer(v, k) = v[:, None] * k[None, :]COMMENT
LOW⚡benchmarks/gdn_triton_reference.py376 # Step 5: Compute output: o = h @ q = sum(h * q[None, :], axis=1) -> [BV]COMMENT
LOW⚡benchmarks/gdn_triton_reference.py633 # Step 1: Apply decay to hidden state: h *= exp(g)COMMENT
135 more matches not shown…
Verbosity Indicators168 hits · 248 pts
SeverityFileLineSnippetContext
LOW…mha_v2/fused_multihead_flash_attention_kernel_noloop.h286 // Do we need to check if there are negative inf for softmax row_max ?COMMENT
LOW…/fused_multihead_flash_attention_kernel_noloop_tiled.h92 // Do we need to check if there are negative inf for softmax row_max ?COMMENT
LOWcsrc/fmha_v2/fmha/warpspec/epilogue.h73 // Whether we need to check if local_max could be -inf or not.COMMENT
LOWcsrc/fmha_v2/fmha/warpspec/epilogue.h543 // Whether we need to check if local_max could be -inf or not.COMMENT
LOWcsrc/fmha_v2/fmha/warpspec/epilogue.h734 // Whether we need to check if local_max could be -inf or not.COMMENT
LOWtests/attention/test_rope.py915 # Step 1: Pre-populate cache with existing tokensCOMMENT
LOWtests/attention/test_rope.py1126 # Step 2: Append new tokens to the pre-populated cacheCOMMENT
LOW…_quantize_fp8_append_paged_kv_cache_h8_kv2_rope64.json194 "reference": "from __future__ import annotations\nimport math\nimport torch\nimport torch.nn.functional as F\n\n@torchCODE
LOW…tize_fp8_append_paged_kv_cache_h8_kv2_rope64_ps16.json195 "reference": "from __future__ import annotations\nimport math\nimport torch\nimport torch.nn.functional as F\n\n@torchCODE
LOW⚡tests/utils/test_fp4_quantize.py315 # Step 1: Quantize with fp4_quantizeCOMMENT
LOW⚡tests/utils/test_fp4_quantize.py325 # Step 2: Dequantize with e2m1_and_ufp8sf_scale_to_floatCOMMENT
LOWtests/utils/test_fp4_quantize.py340 # Step 3: Compare resultsCOMMENT
LOW⚡tests/gdn/reference_delta_rule.py1081 # Step 1: Compute g = -exp(A_log) * softplus(a + dt_bias)COMMENT
LOW⚡tests/gdn/reference_delta_rule.py1136 # Step 1: Apply gating to hidden state: h *= exp(g)COMMENT
LOW⚡tests/gdn/reference_delta_rule.py1139 # Step 2: Delta rule: v -= k^T @ h ([B,H,K] x [B,H,K,V] -> [B,H,V])COMMENT
LOW⚡tests/gdn/reference_delta_rule.py1142 # Step 3: Apply beta gating: v *= betaCOMMENT
LOW⚡tests/gdn/reference_delta_rule.py1145 # Step 4: Update hidden state with outer product: h += k ⊗ v_newCOMMENT
LOW⚡tests/gdn/reference_delta_rule.py1148 # Step 5: Compute output: o = q^T @ hCOMMENT
LOWtests/gdn/reference_delta_rule.py1098 # Step 2: Compute beta = sigmoid(b)COMMENT
LOW⚡benchmarks/bench_cute_dsl_add_rmsnorm_fp4quant.py149 # Step 1: AddCOMMENT
LOW⚡benchmarks/bench_cute_dsl_add_rmsnorm_fp4quant.py151 # Step 2: RMSNormCOMMENT
LOW⚡benchmarks/bench_cute_dsl_add_rmsnorm_fp4quant.py153 # Step 3: FP4 quantize (with global_scale for NVFP4)COMMENT
LOW⚡benchmarks/gdn_triton_reference.py179 # Step 1: Apply decay to hidden state: h *= exp(g)COMMENT
LOW⚡benchmarks/gdn_triton_reference.py182 # Step 2: Delta rule: v -= sum(h * k, dim=0) = k @ hCOMMENT
LOW⚡benchmarks/gdn_triton_reference.py187 # Step 3: Apply beta gating: v *= betaCOMMENT
LOW⚡benchmarks/gdn_triton_reference.py190 # Step 4: Update hidden state: h += outer(k, v) = k[:, None] * v[None, :]COMMENT
LOW⚡benchmarks/gdn_triton_reference.py193 # Step 5: Compute output: o = q @ h = sum(q[:, None] * h, dim=0)COMMENT
LOW⚡benchmarks/gdn_triton_reference.py361 # Step 1: Apply decay: h *= exp(g)COMMENT
LOW⚡benchmarks/gdn_triton_reference.py364 # Step 2: Delta rule: v -= h @ k (h is [BV, BK], k is [BK])COMMENT
LOW⚡benchmarks/gdn_triton_reference.py369 # Step 3: Apply beta gatingCOMMENT
LOW⚡benchmarks/gdn_triton_reference.py372 # Step 4: Update state: h += outer(v, k) = v[:, None] * k[None, :]COMMENT
LOW⚡benchmarks/gdn_triton_reference.py376 # Step 5: Compute output: o = h @ q = sum(h * q[None, :], axis=1) -> [BV]COMMENT
LOW⚡benchmarks/gdn_triton_reference.py633 # Step 1: Apply decay to hidden state: h *= exp(g)COMMENT
LOW⚡benchmarks/gdn_triton_reference.py636 # Step 2: Delta rule: v -= h @ k = sum(h * k[None, :], axis=1)COMMENT
LOW⚡benchmarks/gdn_triton_reference.py640 # Step 3: Apply beta gating: v *= betaCOMMENT
LOW⚡benchmarks/gdn_triton_reference.py643 # Step 4: Update hidden state: h += outer(v, k) = v[:, None] * k[None, :]COMMENT
LOW⚡benchmarks/gdn_triton_reference.py646 # Step 5: Compute output: o = h @ q = sum(h * q[None, :], axis=1)COMMENT
LOW⚡benchmarks/bench_cute_dsl_rmsnorm_fp4quant.py157 # Step 1: RMSNormCOMMENT
LOW⚡benchmarks/bench_cute_dsl_rmsnorm_fp4quant.py159 # Step 2: FP4 quantize (with global_scale for NVFP4)COMMENT
LOWbenchmarks/routines/norm.py1335 # Step 2: RMSNorm on residualCOMMENT
LOW⚡scripts/pr_checks/check_api_docs.py63# Step 1: collect all @flashinfer_api-decorated function names via ASTCOMMENT
LOW⚡scripts/pr_checks/check_api_docs.py91# Step 2: collect documented symbols from docs/api/*.rstCOMMENT
LOW⚡scripts/pr_checks/check_api_docs.py162# Step 3: compare and reportCOMMENT
LOWflashinfer/trace/templates/rope.py1018 # Step 3: append into paged cache (GQA/MHA) — materialize the quantizedCOMMENT
LOW⚡…e_cutedsl_megakernel/src/moe_hopper_fp8/mega_runner.py196 # Step 1: deterministic FP8 input + weight generationCOMMENT
LOW⚡…e_cutedsl_megakernel/src/moe_hopper_fp8/mega_runner.py625 # Step 2: FP8 referenceCOMMENT
LOW⚡…e_cutedsl_megakernel/src/moe_hopper_fp8/mega_runner.py696 # Step 5: FP8 validationCOMMENT
LOW⚡…e_cutedsl_megakernel/src/moe_hopper_fp8/mega_runner.py794 # Step 4: FP8 kernel launchCOMMENT
LOW…le_cutedsl_megakernel/src/moe_nvfp4_swapab/epilogue.py871 # Step 1: absmax over the weighted regs.COMMENT
LOW…l_megakernel/src/moe_nvfp4_swapab/epilogue_refactor.py1407 # Step 0: load tmemCOMMENT
LOW…l_megakernel/src/moe_nvfp4_swapab/epilogue_refactor.py1454 # Step 1: perform swiglu on the first part, interleave with the second's 32x32 tmem transpose.COMMENT
LOW…l_megakernel/src/moe_nvfp4_swapab/epilogue_refactor.py1482 # Step 2: QuantCOMMENT
LOW…l_megakernel/src/moe_nvfp4_swapab/epilogue_refactor.py1493 # Step 3: TMASTGCOMMENT
LOW⚡…cutedsl_megakernel/src/moe_nvfp4_swapab/mega_runner.py774 # Step 1: deterministic input + weight generationCOMMENT
LOW⚡…cutedsl_megakernel/src/moe_nvfp4_swapab/mega_runner.py984 # Step 2: referenceCOMMENT
LOW⚡…cutedsl_megakernel/src/moe_nvfp4_swapab/mega_runner.py1015 # Step 3: workspace allocationCOMMENT
LOW⚡…cutedsl_megakernel/src/moe_nvfp4_swapab/mega_runner.py1053 # Step 4: kernel launchCOMMENT
LOW⚡…cutedsl_megakernel/src/moe_nvfp4_swapab/mega_runner.py1466 # Step 5: validationCOMMENT
LOW⚡…cutedsl_megakernel/src/moe_nvfp4_swapab/kernel_fc12.py1913 # Step 1: coarse-grain spin on ``fc1_done_counter[slot]``COMMENT
LOW…cutedsl_megakernel/src/moe_nvfp4_swapab/kernel_fc12.py2014 # Step 3: K-loop with 2x cute.copy per tile (B +COMMENT
108 more matches not shown…
Self-Referential Comments75 hits · 197 pts
SeverityFileLineSnippetContext
MEDIUMtests/conftest.py309 # Create a dict with module info for reportingCOMMENT
MEDIUMtests/kda/test_recurrent_kda_decode_export.py38# This file is the deprecated facade's regression suite, so its own deprecationCOMMENT
MEDIUMtests/attention/test_attention_sink.py297 # Create a fragmented page allocation patternCOMMENT
MEDIUMtests/attention/test_trtllm_gen_attention_decode.py1976 # Create a larger tensor with 2*D in the last dimensionCOMMENT
MEDIUM⚡tests/trace/test_fi_trace.py1103 # Create a dummy instance — we don't call run(), only fi_trace()COMMENT
MEDIUMtests/moe_ep/nccl_ep/test_handle_mock.py803 # Create the handle at 222 rows on a fleet sized for 128, so theCOMMENT
MEDIUMtests/autotuner/test_autotuner_configs.py772 # Create a cache file with a known configCOMMENT
MEDIUMtests/autotuner/test_autotuner_configs.py963 # Create a cache fileCOMMENT
MEDIUMtests/mamba/utils.py266 # Create a simple linear chain structure by defaultCOMMENT
MEDIUMtests/comm/test_trtllm_alltoall.py114 # Create a random input tensorCOMMENT
MEDIUMtests/comm/test_trtllm_alltoall.py177 # Create a random input tensorCOMMENT
MEDIUMtests/comm/test_trtllm_moe_alltoall.py332 # Create a random input tensorCOMMENT
MEDIUMtests/comm/test_trtllm_moe_alltoall.py711 # Create a random input tensorCOMMENT
MEDIUMtests/comm/test_trtllm_moe_alltoall.py1383 # Create a random input tensorCOMMENT
MEDIUMtests/comm/test_mnnvl_memory.py135 # Create a random input tensorCOMMENT
MEDIUMtests/comm/test_quantized_allreduce.py213 # This file is single-node spawn-managed; use a file store to avoidCOMMENT
MEDIUMtests/utils/test_fp4_quantize.py253 # Create a test scale factors tensor with uint8 dtypeCOMMENT
MEDIUMtests/utils/test_fp4_quantize.py303 # Create a reasonable test tensorCOMMENT
MEDIUMtests/utils/test_logging.py520 # Create a tensor with specific propertiesCOMMENT
MEDIUMtests/utils/test_logging_replay.py1131 # Create a mock dump with safetensors filesCOMMENT
MEDIUMtests/utils/test_norm.py927 # Create a JIT spec for norm module without ENABLE_FP8 flagCOMMENT
MEDIUMtests/utils/test_load_cubin_compile_race_condition.py40 # Define the target file - same for all processesCOMMENT
MEDIUMtests/utils/test_load_cubin_compile_race_condition.py75 # Create a temporary directory for FLASHINFER_CUBIN_DIRCOMMENT
MEDIUMtests/cli/test_cli_cmds.py536 # Create a temporary JIT directory with some dummy cache filesSTRING
MEDIUMtests/cli/test_cli_cmds.py579 # Create a temporary cubin directory with some dummy cubin filesSTRING
MEDIUMtests/test_helpers/rope_reference.py113# The following code is from the vLLM's implementation of RoPE.COMMENT
MEDIUMtests/moe/test_dpsk_fused_moe_fp8.py596 # Create a mock MoE implementation for skip_checksCOMMENT
MEDIUMdocs/tutorials/jax_tvm_ffi/flashinfer_jax_tvm_ffi.py1# This file is the canonical Sphinx-Gallery source for the tutorial.COMMENT
MEDIUMdocs/tutorials/jax_tvm_ffi/gemma3_flashinfer_jax.py1# This file is the canonical Sphinx-Gallery source for the tutorial.COMMENT
MEDIUMexamples/pytorch/wan/transformer_wan_flashinfer.py4# This file is adapted from the diffusers WanTransformer3DModel implementation:COMMENT
MEDIUMscripts/task_test_multi_gpu_comm_kernels.sh22# Define the specific test files for multi-GPU comm tests (single-node)COMMENT
MEDIUMscripts/task_test_multi_node_comm_kernels.sh25# Define the specific test files for multi-node comm testsCOMMENT
MEDIUMscripts/test_utils.sh3# This file is meant to be sourced by test runner scriptsCOMMENT
MEDIUMscripts/test_utils.sh1024 # Create a bash array with the node IDsCOMMENT
MEDIUMscripts/test_utils.sh1194 # Create a temporary directory for parallel job stateCOMMENT
MEDIUMscripts/test_utils.sh1209 # Create a results file for each testCOMMENT
MEDIUMflashinfer/experimental/b12x/_lib/utils.py368 # Create a numpy arraySTRING
MEDIUMflashinfer/experimental/b12x/_lib/dense_gemm.py29# This file is ported from the CUTLASS dense block-scaled GEMM exampleCOMMENT
MEDIUMflashinfer/experimental/deepseek_v41/hca_v41.py2253 # This function is reached only by the dequant warp interval, soCOMMENT
MEDIUM…e_cutedsl_megakernel/src/moe_nvfp4_swapab/moe_utils.py221 # Create a new cute.Pointer with generic address space, preserving alignmentCOMMENT
MEDIUM…e_cutedsl_megakernel/src/moe_nvfp4_swapab/moe_utils.py249 # Create a new cute.Pointer with gmem address space, preserving alignmentCOMMENT
MEDIUM…b_cutedsl_megakernel/src/moe_nvfp4_swapab/moe_utils.py220 # Create a new cute.Pointer with generic address space, preserving alignmentCOMMENT
MEDIUM…b_cutedsl_megakernel/src/moe_nvfp4_swapab/moe_utils.py248 # Create a new cute.Pointer with gmem address space, preserving alignmentCOMMENT
MEDIUM…dsl_megakernel/src/moe_sm120_mxfp8_swapab/moe_utils.py260 # Create a new cute.Pointer with generic address space, preserving alignmentCOMMENT
MEDIUM…dsl_megakernel/src/moe_sm120_mxfp8_swapab/moe_utils.py288 # Create a new cute.Pointer with gmem address space, preserving alignmentCOMMENT
MEDIUM…m100/cutedsl_megamoe/src/moe_nvfp4_swapab/moe_utils.py229 # Create a new cute.Pointer with generic address space, preserving alignmentCOMMENT
MEDIUM…m100/cutedsl_megamoe/src/moe_nvfp4_swapab/moe_utils.py257 # Create a new cute.Pointer with gmem address space, preserving alignmentCOMMENT
MEDIUMflashinfer/comm/dlpack_utils.py224 # Create a new capsule each timeSTRING
MEDIUMflashinfer/fused_moe/core.py586 # Create the outputCOMMENT
MEDIUMflashinfer/fused_moe/cute_dsl/rubin/custom_pipeline.py44# This file is copied and modified from https://github.com/NVIDIA/cutlass/tree/main/python/CuTeDSL/cutlass/pipelineCOMMENT
MEDIUMflashinfer/fused_moe/cute_dsl/rubin/utils.py146 # Create a numpy arraySTRING
MEDIUM…hinfer/fused_moe/cute_dsl/blackwell/custom_pipeline.py44# This file is copied and modified from https://github.com/NVIDIA/cutlass/tree/main/python/CuTeDSL/cutlass/pipelineCOMMENT
MEDIUMflashinfer/fused_moe/cute_dsl/blackwell/utils.py44# This file is copied and modified from cutlass https://github.com/NVIDIA/cutlass/blob/main/python/CuTeDSL/cutlass/cute/COMMENT
MEDIUM…blockscaled_contiguous_grouped_gemm_finalize_fusion.py317 """This class implements batched matrix multiplication (C = A x SFA x B x SFB) with support for various data typesSTRING
MEDIUM…ockscaled_contiguous_gather_grouped_gemm_act_fusion.py332 """This class implements contiguous grouped matrix multiplication with gather operation and SwiGLU fusionSTRING
MEDIUMflashinfer/fused_moe/cute_dsl/common/kernel_utils.py44# This file is copied and modified from cutlass https://github.com/NVIDIA/cutlass/blob/main/python/CuTeDSL/cutlass/cute/COMMENT
MEDIUMflashinfer/gemm/kernels/dense_blockscaled_gemm_sm103.py29# This file is ported from CUTLASS's sm103_dense_blockscaled_gemm_persistent.pyCOMMENT
MEDIUMflashinfer/gemm/kernels/dense_blockscaled_gemm_sm103.py54 """This class implements batched matrix multiplication (C = A x SFA x B x SFB) with support for FP4 data typesSTRING
MEDIUMflashinfer/gemm/kernels/epilogue_utils.py4# This file contains modified versions of epilogue utility functions fromCOMMENT
MEDIUMflashinfer/gemm/kernels/dense_blockscaled_gemm_sm107.py29# This file is ported from CUTLASS's Rubin dense_blockscaled_gemm_persistent.pyCOMMENT
15 more matches not shown…
Redundant / Tautological Comments67 hits · 100 pts
SeverityFileLineSnippetContext
LOWbuild_backend.py1059 # Check if we're in a git repositoryCOMMENT
LOWflashinfer-cubin/build_backend.py75 # Check if we're in a git repositoryCOMMENT
LOWtests/attention/test_attention_sink.py928 # Check if any request has qo_len > kv_len for causal caseCOMMENT
LOW…s/comm/test_trtllm_allreduce_fusion_group_fp8_quant.py218 # Check if current workspace is large enough; recreate if notCOMMENT
LOWtests/utils/test_fp8_quantize.py832 # Check if the bits of paddings are zero.COMMENT
LOWtests/utils/test_load_cubin_compile_race_condition.py44 # Read the file from FLASHINFER_CUBIN_DIRCOMMENT
LOWtests/model_optimizations/test_dsv3_fused_routing.py276 # Check if kernel's selection matches expected experts (exact or tied)COMMENT
LOWtests/test_helpers/sink_attention_reference.py95 # Check if variable length mode is indicated by presence of indptrCOMMENT
LOWflashinfer-jit-cache/build_backend.py89 # Check if we're in a git repositoryCOMMENT
LOWbenchmarks/bench_tgv_gemm.py184 # Check if results are close enoughCOMMENT
LOWbenchmarks/bench_logging_overhead.py224 # Check if CUDA is availableCOMMENT
LOWbenchmarks/bench_dcp_alltoall.py297 # Print results (rank 0 only)COMMENT
LOWscripts/xfails_tracker.py52 # Check if this is a test function (starts with "test_" or has @pytest marks)COMMENT
LOWscripts/xfails_tracker.py70 # Check if this is a pytest.param callCOMMENT
LOWscripts/xfails_tracker.py77 # Check if this is a runtime pytest.xfail() callCOMMENT
LOWscripts/test_utils.sh1674 # Check if parallel execution is enabledCOMMENT
LOWscripts/codeowner_analyzer.py89 # Check if gh CLI is availableCOMMENT
LOWscripts/codeowner_analyzer.py114 # Check if it's already in our cacheCOMMENT
LOWscripts/codeowner_analyzer.py220 # Check if username is in allowed list (case-insensitive)COMMENT
LOWscripts/codeowner_analyzer.py290 # Check if it's a relevant code fileCOMMENT
LOWscripts/codeowner_analyzer.py420 # Check if contributor is in allowed users listCOMMENT
LOWscripts/codeowner_analyzer.py629 # Write file-level overrides LAST (CODEOWNERS uses last-match-wins)COMMENT
LOW.github/workflows/update-xfails-report.yml43 # Check if xfails report file is new (unstaged) or has changesCOMMENT
LOW.github/workflows/update-codeowners.yml50 # Check if CODEOWNERS file is new (unstaged) or has changesCOMMENT
LOW.github/workflows/issue-claim.yml86 # Check if issue already has assigneesCOMMENT
LOW.github/workflows/issue-claim.yml301 # Check if user is now a collaborator (accepted invitation)COMMENT
LOWflashinfer/api_logging.py2105 # Check if this parameter was providedCOMMENT
LOWflashinfer/utils.py357 # Check if it's a CUDA deviceCOMMENT
LOWflashinfer/prefill.py638 # Check if FP8 by presence of scale tensorsCOMMENT
LOWflashinfer/__main__.py1106 # Check if this is a single dump or a session / sequence rootCOMMENT
LOWflashinfer/cuda_utils.py20 # Check if cuda.cudart module is available and import accordinglyCOMMENT
LOW…ention/prims_ts/kernels/fmha_context/fmha_resources.py5364 # Check if we should skip correction (when old_max == new_max)COMMENT
LOW…utedsl_megamoe/src/moe_bf16_glu/mega_reference_bf16.py566 # Check if input data types are compatible with MMA instructionCOMMENT
LOWflashinfer/autotuner/autotuner.py2528 # Set time_measured to inf to notify the failure of the tactic. This can happen whenCOMMENT
LOWflashinfer/triton/sm_constraint_gemm.py66 # Set num_sms to be 100% of the available SMsCOMMENT
LOWflashinfer/comm/mnnvl.py757 # Check if cuda.cudart is available and import accordinglyCOMMENT
LOWflashinfer/comm/mnnvl.py794 # Check if device supports multicastingCOMMENT
LOW…hinfer/fused_moe/cute_dsl/blackwell/custom_pipeline.py357 # Set mask to None if not using clusters (i.e. 1CTA kernels)COMMENT
LOW…hinfer/fused_moe/cute_dsl/blackwell/custom_pipeline.py363 # Set mask to None if not using 2CTA instructionsCOMMENT
LOWflashinfer/gemm/gemm_base.py9472 # Check if dimensions are aligned for cute-dsl (16-byte alignment)COMMENT
LOWflashinfer/gemm/kernels/dense_blockscaled_gemm_sm103.py363 # Check if input data types are compatible with MMA instructionCOMMENT
LOWflashinfer/gemm/kernels/dense_blockscaled_gemm_sm103.py998 # Check if we're not at the last buffer of the last k_tileCOMMENT
LOWflashinfer/gemm/kernels/dense_blockscaled_gemm_sm103.py1083 # Check if we're not at the last stage of the last k_tileCOMMENT
LOWflashinfer/gemm/kernels/epilogue_utils.py241 # Check if tile_sched is StaticPersistentTileScheduler or any subclass inheriting from itCOMMENT
LOWflashinfer/gemm/kernels/dense_blockscaled_gemm_sm107.py754 # Check if prefetch is enabled (prefetch_dist > 0)COMMENT
LOWflashinfer/gemm/kernels/dense_blockscaled_gemm_sm107.py2069 # Check if the current combination is validCOMMENT
LOWflashinfer/gemm/kernels/dense_blockscaled_gemm_sm107.py2794 # Check if prefetch is enabled (prefetch_dist > 0)COMMENT
LOWflashinfer/gemm/kernels/grouped_gemm_masked_rubin.py694 # Check if input data types are compatible with MMA instructionCOMMENT
LOWflashinfer/gemm/kernels/grouped_gemm_masked_rubin.py1950 # Check if the current combination is validCOMMENT
LOWflashinfer/gemm/kernels/bmm_fp8_rubin.py547 # Check if we should use Bkeep-Breuse patternCOMMENT
LOWflashinfer/gemm/kernels/bmm_fp8_blackwell.py310 # Check if input data types are compatible with MMA instructionCOMMENT
LOW…ashinfer/gemm/kernels/grouped_gemm_masked_blackwell.py907 # Check if input data types are compatible with MMA instructionCOMMENT
LOW…er/gemm/kernels/dense_blockscaled_gemm_sm100_common.py102 # Check if input data types are compatible with MMA instructionCOMMENT
LOW…nfer/gemm/kernels/grouped_gemm_contiguous_blackwell.py551 # Check if input data types are compatible with MMA instructionCOMMENT
LOWflashinfer/gemm/kernels/cutile/ragged_bmm_cutile.py780 # Check if autotune is enabledCOMMENT
LOW…ashinfer/gemm/kernels/cutile/gemm_alpha_beta_cutile.py411 # Check if autotune is requestedCOMMENT
LOWflashinfer/quantization/kernels/mxfp8_quantize.py739 # Check if this thread's row is validCOMMENT
LOWflashinfer/cute_dsl/gemm_allreduce_two_shot.py519 # Check if input data types are compatible with MMA instructionCOMMENT
LOWflashinfer/testing/utils.py871 # Check if args are provided (determines how we call fn)COMMENT
LOWflashinfer/testing/utils.py1168 # Check if args are provided (determines how we call fn)COMMENT
7 more matches not shown…
Docstring Block Structure17 hits · 85 pts
SeverityFileLineSnippetContext
HIGHflashinfer/green_ctx.py129 Split the device into multiple `green contexts <https://docs.nvidia.com/cuda/cuda-driver-api/group__CUDA__GREEN__COSTRING
HIGHflashinfer/green_ctx.py199 Split the device into multiple green contexts, each with a fixed number of SMs, return the corresponding streamSTRING
HIGHflashinfer/utils.py341 Get GPU memory bandwidth in GB/s for the specified CUDA device. Args: device: torch.device object, e.gSTRING
HIGHflashinfer/autotuner/autotuner.py105Map *x* to the nearest bucket using floor or ceil semantics. Args: x: The value to map. buckets: BuSTRING
HIGHflashinfer/autotuner/autotuner.py149Create a mapper function for :class:`DynamicTensorSpec.map_to_tuning_buckets`. The returned callable maps any integSTRING
HIGHflashinfer/autotuner/autotuner.py843Context manager for autotuning with optional file-based caching. Controls how FlashInfer profiles and selects the bSTRING
HIGHflashinfer/autotuner/autotuner.py3678Load autotuner configs from a JSON file. Populates the internal config lookup table so that ``search_cache()`` STRING
HIGH…blockscaled_contiguous_grouped_gemm_finalize_fusion.py391Blockscaled contiguous grouped GEMM for MoE GEMM2 workloads. Fused mode applies routing weights and atomically reduSTRING
HIGH…ockscaled_contiguous_gather_grouped_gemm_act_fusion.py487Blockscaled contiguous gather grouped GEMM with fused FC1 activation. Performs grouped matrix multiplication with fSTRING
HIGHflashinfer/attn_scores/attn_scores.py1610FP8 paged MQA logits for Blackwell (SM100/SM103) and Rubin (SM107). Computes weighted per-head ReLU attention scoreSTRING
HIGHflashinfer/attn_scores/attn_scores.py2127FP4 (MXFP4) paged MQA logits for Blackwell (SM100/SM103) and Rubin (SM107). Computes weighted per-head ReLU attentiSTRING
HIGHflashinfer/gemm/gemm_bf16_fp4.py265Prepare FP4 weights for the bf16 x fp4 GEMM, for a specific backend. The caller is expected to start with weights iSTRING
HIGHflashinfer/quantization/fp4_quantization.py474Quantize a batched tensor to FP4 (E2M1x2) with per-block scale factors. This function converts a float/bfloat16STRING
HIGHflashinfer/quantization/fp4_quantization.py675Quantize a silu and matmul with masked batched tensor to FP4 (E2M1x2) with per-block scale factors. This functiSTRING
HIGHflashinfer/topk_varlen/kernels/filtered_topk_util.py114Return the filtered top-k resource policy for an architecture. Args: compute_capability: A ``(major, minor)STRING
HIGHflashinfer/parallel_attention/utils.py356Create a device mesh and return the Ring and Ulysses process groups. Builds a ``DeviceMesh`` with up to three dimenSTRING
HIGHflashinfer/kda_kernels/kda_decode_wy_output_only.py2070Output-only (frozen-state) KDA decode over 1..16 tokens per sequence. Computes the KDA (Kimi Delta Attention) outpuSTRING
Modern AI Meta-Vocabulary14 hits · 46 pts
SeverityFileLineSnippetContext
MEDIUMtests/autotuner/test_autotuner_core.py1914 # TODO: Let AutoTuner orchestrate multiple same-shaped value samples within one tuning transaction.COMMENT
MEDIUM⚡…cutedsl_megakernel/src/moe_nvfp4_swapab/mega_runner.py2414 # Top-level orchestrationCOMMENT
MEDIUM…cutedsl_megakernel/src/moe_nvfp4_swapab/mega_runner.py2658 # orchestration above is the part being smoke-tested for now.COMMENT
MEDIUM⚡…cutedsl_megakernel/src/moe_nvfp4_swapab/mega_runner.py2405 # Top-level orchestrationCOMMENT
MEDIUM…cutedsl_megakernel/src/moe_nvfp4_swapab/mega_runner.py2648 # orchestration above is the part being smoke-tested for now.COMMENT
MEDIUM⚡…l_megakernel/src/moe_sm120_mxfp8_swapab/mega_runner.py2151 # Top-level orchestrationCOMMENT
MEDIUM…l_megakernel/src/moe_sm120_mxfp8_swapab/mega_runner.py2378 # orchestration above is the part being smoke-tested for now.COMMENT
MEDIUM⚡…00/cutedsl_megamoe/src/moe_nvfp4_swapab/mega_runner.py2539 # Top-level orchestrationCOMMENT
MEDIUM…00/cutedsl_megamoe/src/moe_nvfp4_swapab/mega_runner.py2804 # orchestration above is the part being smoke-tested for now.COMMENT
MEDIUMflashinfer/fused_moe/da_runtime.py770 # Strip orchestration-owned keywords before delegating dtype-specific allocation andCOMMENT
MEDIUMflashinfer/mla/cake_dsv4.py1757# Launch orchestration #COMMENT
MEDIUM⚡flashinfer/cute_dsl/attention/roles/epilogue.py74 # Prefill orchestration — proven-correct inline implementationCOMMENT
MEDIUM⚡flashinfer/cute_dsl/attention/roles/loader_tma.py282 # Prefill orchestration — proven-correct inline implementationCOMMENT
MEDIUM⚡flashinfer/cute_dsl/attention/roles/mma.py198 # Prefill orchestration — uses primitives for GEMMs and TMEM lifecycleCOMMENT
Example Usage Blocks17 hits · 26 pts
SeverityFileLineSnippetContext
LOWcsrc/fmha_v2/fmha/utils.h66// Example usage:COMMENT
LOWscripts/list_apis.sh10# Usage:COMMENT
LOW…_megakernel/src/moe_hopper_fp8/run_functional_tests.sh13# Usage:COMMENT
LOW…utedsl_megakernel/src/moe_hopper_fp8/run_mega_tests.sh21# Usage:COMMENT
LOW…cutedsl_megakernel/src/moe_hopper_fp8/run_perf_test.sh18# Usage:COMMENT
LOW…egakernel/src/moe_nvfp4_swapab/run_functional_tests.sh10# Usage:COMMENT
LOW…edsl_megakernel/src/moe_nvfp4_swapab/run_mega_tests.sh17# Usage:COMMENT
LOW…egakernel/src/moe_nvfp4_swapab/run_functional_tests.sh10# Usage:COMMENT
LOW…edsl_megakernel/src/moe_nvfp4_swapab/run_mega_tests.sh17# Usage:COMMENT
LOW…l_megakernel/src/moe_mxfp8_glu/run_functional_tests.sh18# Usage:COMMENT
LOW…cutedsl_megakernel/src/moe_mxfp8_glu/run_mega_tests.sh22# Usage:COMMENT
LOW…l_megamoe/src/moe_nvfp4_swapab/run_functional_tests.sh10# Usage:COMMENT
LOW…cutedsl_megamoe/src/moe_nvfp4_swapab/run_mega_tests.sh17# Usage:COMMENT
LOW…edsl_megamoe/src/moe_mxfp8_glu/run_functional_tests.sh18# Usage:COMMENT
LOW…00/cutedsl_megamoe/src/moe_mxfp8_glu/run_mega_tests.sh22# Usage:COMMENT
LOW…tedsl_megamoe/src/moe_bf16_glu/run_functional_tests.sh18# Usage:COMMENT
LOW…100/cutedsl_megamoe/src/moe_bf16_glu/run_mega_tests.sh33# Usage:COMMENT
AI Response Leakage3 hits · 22 pts
SeverityFileLineSnippetContext
HIGHdocs/tutorials/jax_tvm_ffi/flashinfer_jax_tvm_ffi.py304# In this example, you should learn how minimal the JAX-TVM bridge can be when the function signature and ``arg_spec`` aCOMMENT
HIGHdocs/tutorials/jax_tvm_ffi/flashinfer_jax_tvm_ffi.py427# In this example, you should learn how to handle two new complications that were absent in Example 1: a TVM function thCOMMENT
HIGHdocs/tutorials/jax_tvm_ffi/flashinfer_jax_tvm_ffi.py644# In this example, you should learn about three patterns that appear together for the first time: a kernel **compiled seCOMMENT
Overly Generic Function Names13 hits · 11 pts
SeverityFileLineSnippetContext
LOWtests/utils/test_decorators.py17 def my_function(x, y):CODE
LOWtests/utils/test_logging.py186 def test_function(x, y):CODE
LOWtests/utils/test_logging.py212 def test_function(tensor, value):CODE
LOWtests/utils/test_logging.py250 def test_function(tensor):CODE
LOWtests/utils/test_logging.py278 def test_function(mode: TestEnum, strategy: StringEnum):CODE
LOWtests/utils/test_logging.py308 def test_function(x, y=10, z=20, mode=TestEnum.OPTION_A):CODE
LOWtests/utils/test_logging.py339 def test_function(x, y=10, z=20):CODE
LOWtests/utils/test_logging.py479 def test_function(CODE
LOWtests/utils/test_logging.py517 def test_function(tensor):CODE
LOWtests/utils/test_logging.py550 def test_function(nested):CODE
LOWtests/utils/test_logging.py615 def test_function(x):CODE
LOWtests/utils/test_logging.py640 def test_function(a, b, c):CODE
LOWflashinfer/api_logging.py2451 ... def my_function(x, y):STRING
Synthetic Comment Markers1 hit · 5 pts
SeverityFileLineSnippetContext
HIGH.github/workflows/update-codeowners.yml76 Auto-generated CODEOWNERS update based on commit activity over the last ${{ env.DAYS_BACK }} days.CODE
Dead Code2 hits · 4 pts
SeverityFileLineSnippetContext
MEDIUMtests/experimental/b12x/preparation/test_session.py62CODE
MEDIUM…infer/experimental/b12x/comm/pcie/pcie_vocab_argmax.py300CODE
Fake / Example Data4 hits · 3 pts
SeverityFileLineSnippetContext
LOW⚡flashinfer/gdn_kernels/gdn_decode_mtp.py3104 dummy_key = (q.device, h0_source.dtype)CODE
LOW⚡flashinfer/gdn_kernels/gdn_decode_mtp.py3105 if dummy_key not in dummy_intermediate_states:CODE
LOW⚡flashinfer/gdn_kernels/gdn_decode_mtp.py3106 dummy_intermediate_states[dummy_key] = torch.empty(CODE
LOW⚡flashinfer/gdn_kernels/gdn_decode_mtp.py3109 intermediate_states = dummy_intermediate_states[dummy_key]CODE