Repository navigation

[TRTLLM-17005][refactor] Component-local C++ Tests (Part 3) - #20886

Draft

Funatiq wants to merge 37 commits into

NVIDIA:mainfrom

Funatiq:dev/refactor/component-local-tests-3

Draft

[TRTLLM-17005][refactor] Component-local C++ Tests (Part 3)#20886
Funatiq wants to merge 37 commits into
NVIDIA:mainfrom
Funatiq:dev/refactor/component-local-tests-3

Commits on Oct 25, 2026
- [None][build] make GoogleTest discovery prefixes optional
  Funatiq
  committed
  e937d2a
- build: reject monolithic linkage without explicit FULL_STACK
  Funatiq
  committed
  eeca770
- [None][build] reject full-library interface links in component tests
  Funatiq
  committed
  38f147a
- [None][build] defer component-test linkage checks until directory end
  Funatiq
  committed
  f4dd409
- build: extract the logits kernel family and its test
  Funatiq
  committed
  0593d85
- build: extract the NVFP4 cold-page kernel family
  Funatiq
  committed
  3a5aefa
- build: extract RoPE and QKV preprocessing kernels
  Funatiq
  committed
  d9aa553
- build: extract sparse-attention page-gather kernels
  Funatiq
  committed
  7e66034
- build: extract MLA preprocessing kernels
  Funatiq
  committed
  8072719
- build: extract MLA chunked-prefill kernels
  Funatiq
  committed
  951f480
- build: extract fused MoE communication kernels
  Funatiq
  committed
  f1e51f8
- build: extract MoE load-balancing kernels
  Funatiq
  committed
  34c834b
- build: extract MoE LoRA preparation kernels
  Funatiq
  committed
  4c244cb
- build: extract custom attention-mask preparation
  Funatiq
  committed
  15b8b02
- build: extract CUDA core GEMM kernels
  Funatiq
  committed
  ea56f20
- build: extract shared CUTLASS GEMM host utilities
  Funatiq
  committed
  5c75e41
- build: migrate fused gated GEMM tests to focused targets
  Funatiq
  committed
  96c2931
- build: extract SmoothQuant and INT8 GEMM kernels
  Funatiq
  committed
  c09d877
- build: extract weight-only GEMV kernels and migrate their test
  Funatiq
  committed
  9f70a7a
- build: extract the complete MoE routing kernel family
  Funatiq
  committed
  984b91d
- build: extract block-scale MoE activation and finalization kernels
  Funatiq
  committed
  aad6db4
- build: extract cascade attention and its MMHA reference
  Funatiq
  committed
  cab9008
- build: expose generated FMHA as a focused static component
  Funatiq
  committed
  c309a36
- build: extract cuBLAS matrix-wrapper support for kernel leaves
  Funatiq
  committed
  a50d82d
- build: extract LoRA adapter and grouped-GEMM kernel support
  Funatiq
  committed
  b82f50c
- build: extract shared CUDA quantization kernels
  Funatiq
  committed
  c597220
- build: separate shared pre-quantization scaling kernels
  Funatiq
  committed
  65e0fb5
- build: expose the FP8 block-scale GEMM component
  Funatiq
  committed
  39078c3
- build: migrate CUTLASS MoE tests to explicit kernel components
  Funatiq
  committed
  527d7e9
- [None][fix] link internal MoE component to FP8 block-scale GEMM
  Funatiq
  committed
  70ea185
- build: link whole-archive components in a foreach loop
  Funatiq
  committed
  92d2423
- [None][fix] keep CUDA payloads beyond host sections on AArch64
  Funatiq
  committed
  0b16b45
- [None][fix] keep CUDA payloads beyond host sections on x86-64
  Funatiq
  committed
  dfcf501
- [None][fix] apply CUDA payload layout to decoder libraries
  Funatiq
  committed
  90ba442
- [None][build] simplify Linux CUDA payload layout guard
  Funatiq
  committed
  68cc086
- [None][fix] place CUDA payloads before host sections for BOLT
  Funatiq
  committed
  04000c3
- [None][test] remove GPU resource locks added by kernel test migration
  Funatiq
  committed
  6d84133

Commits

Commits on Oct 25, 2026