Repository navigation

@FIR-2216 - llama.cpp: expose ggml perf-summary symbols across ollama's split-binary backend boundary - #158

Merged

akapoor3518 merged 8 commits into

add-ollama-perf-profile-support

Aug 17, 2026

Merged

@FIR-2216 - llama.cpp: expose ggml perf-summary symbols across ollama's split-binary backend boundary#158
akapoor3518 merged 8 commits into
masterfrom
add-ollama-perf-profile-support

Commits on Aug 15, 2026
- Expose ggml_perf_accumulate and backend profile hook for ollama's split-binary build
  Anoop Kapoor
  committed
  faaad8b
- Fix data race in tsi_init_per_txe_state_once() lazy allocation
  Anoop Kapoor
  committed
  ed70fd4
- Harden call_triton_matmul_full_packed() against concurrent reuse
  Anoop Kapoor
  committed
  441cc4f
- Expose ggml_perf_log_open/write_detailed_csv/ggml_backend_type for ollama
  Anoop Kapoor
  committed
  aacf5ea
- Address cubic/qodo review findings on PR #158
  Anoop Kapoor
  committed
  aea1a75
- Address cubic's second-pass findings on aea1a75fa
  Anoop Kapoor
  committed
  2eb4d95
- Clear packed_args/scalar_*_args vectors on teardown, not just device_free
  Anoop Kapoor
  committed
  52eee52
Commits on Aug 16, 2026
- Guard ggml_perf_log_open/write_detailed_csv behind GGML_PERF_DETAIL
  Anoop Kapoor
  committed
  bcb7c7e