Skip to content

Fix linker error for tests - #570

Closed
kirillsurkov wants to merge 1 commit into
ggml-org:masterfrom
kirillsurkov:master
Closed

kirillsurkov wants to merge 1 commit into
ggml-org:masterfrom
kirillsurkov:master

Conversation

@kirillsurkov

Copy link
Copy Markdown

Link tests with ggml to avoid these errors:

[build] test-quantize.c.obj : error LNK2019: unresolved external symbol ggml_quantize_q4_0 referenced in function main
[build] test-quantize.c.obj : error LNK2019: unresolved external symbol ggml_quantize_q4_1 referenced in function main

@anzz1

anzz1 commented Mar 28, 2023 •

Copy link
Copy Markdown
Contributor

I reckon you are using MinGW ? This PR should fix it:

@ggerganov ggerganov closed this Mar 30, 2023
bong-water-water-bong added a commit to 1bit-MONSTER/llama.cpp that referenced this pull request Oct 25, 2026
… to the pin (engine ggml-org#444, ggml-org#445) (#148)

op_params_json() had no branch for FillParams (#140), so a graph snapshot dropped the FILL constant and the tree did not build under GCC; the constant is now written and read back, with a round-trip case. run_qwen_matmul_dispatch_checks leaked GGML_HRX_DISABLE_QWEN_DISPATCH across cases in one process; scoped. Expectation tables of hrx-backend-test and test-hrx-ops brought to the e7e5cd0 pin: both suites run to the end on HRX0, green except rows pointing at open issues (ggml-org#533 via ggml-org#570, ggml-org#534 via #149).

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
bong-water-water-bong added a commit to 1bit-MONSTER/llama.cpp that referenced this pull request Oct 25, 2026
…(engine ggml-org#533, ggml-org#570) (#150)

Registry-level fusion-order guard (dispatch-fusion-order.{h,cpp}): a match whose non-write-only binding reads transient data that no earlier dispatch (in the plan or in the match) wrote is refused and falls through to the next registration; GGML_HRX_FUSION_ORDER_GUARD=0 disables, GGML_HRX_LOG_FUSION_ORDER=1 logs. Also: routed-ffn gate_up_swiglu_f16_wmma only when every GLU consumer is a Q4_K/Q6_K routed down projection (a real defect: other down weights read an unwritten f32 value). hrx-backend-test host 60/0/6 -> 63/0/3, device 11/11, test-backend-ops 314/314, KLD identical on Qwen3.5-4B / ZAYA1-8B / Qwen3.8-27B with 0 refusals, 27B draft-mtp completions identical, hw-gate PASS. Supersedes #133 and #146.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants