From 2fd29e2235376162ba15fb690b780bf2cf36e6f3 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Tue, 1 Sep 2026 22:59:29 +0300
Subject: [PATCH] tests : run fusion test from ci instead of ctest

The fusion test needs Metal and generates a lot of dummy models, so it
does not belong in the generic ctest suite. Move it to ci/run.sh as
gg_run_test_fusion, gated on GG_BUILD_METAL like
gg_run_test_llama_archs_tensor_split: it generates the dummy models with
test-llama-archs -o and then validates the fusion counts against the
committed baseline. test-fusion.cpp is still built (llama_build) but no
longer registered as a ctest.

Assisted-by: pi:llama.cpp/DeepSeek-V4-Flash-0731
---
 ci/run.sh            | 29 +++++++++++++++++++++++++++++
 tests/CMakeLists.txt |  7 -------
 2 files changed, 29 insertions(+), 7 deletions(-)

diff --git a/ci/run.sh b/ci/run.sh
index 3909283d12..0851e20aab 100755
--- a/ci/run.sh
+++ b/ci/run.sh
@@ -334,6 +334,34 @@ function gg_sum_test_llama_archs_tensor_split {
     gg_printf '```\n'
 }
 
+# test_fusion
+
+function gg_run_test_fusion {
+    cd ${SRC}
+
+    set -e
+
+    if [ ! -z ${GG_BUILD_METAL} ]; then
+        rm -rf build-ci-fusion-models && mkdir -p build-ci-fusion-models
+
+        # generate the dummy models and run the fusion regression test
+        ./build-ci-release/bin/test-llama-archs -o build-ci-fusion-models 2>&1
+        ./build-ci-release/bin/test-fusion --models build-ci-fusion-models --device MTL0 --check tests/fusion/MTL.tsv 2>&1
+    fi
+
+    set +e
+}
+
+function gg_sum_test_fusion {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Runs test-fusion against the Metal baseline\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '```\n'
+    gg_printf '%s\n' "$(cat $OUT/${ci}.log)"
+    gg_printf '```\n'
+}
+
 # test_scripts
 
 function gg_run_test_scripts {
@@ -791,6 +819,7 @@ test $ret -eq 0 && gg_run ctest_debug
 test $ret -eq 0 && gg_run ctest_release
 
 test $ret -eq 0 && gg_run test_llama_archs_tensor_split
+test $ret -eq 0 && gg_run test_fusion
 
 if [ ! -z ${GG_BUILD_HIGH_PERF} ]; then
     test $ret -eq 0 && gg_run test_backend_ops_cpu
diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txt
index 905833ffa1..531a690cb5 100644
--- a/tests/CMakeLists.txt
+++ b/tests/CMakeLists.txt
@@ -247,14 +247,7 @@ if (NOT WIN32 OR NOT BUILD_SHARED_LIBS)
     )
     set_tests_properties(test-save-load-state PROPERTIES FIXTURES_REQUIRED generate-models)
 
-    # Test Metal fusion against a per-device baseline, using the generated dummy models
     llama_build(test-fusion.cpp)
-    llama_test(
-        test-fusion
-        LABEL main
-        ARGS --models "${MODEL_DIR}" --device MTL0 --check "${PROJECT_SOURCE_DIR}/tests/fusion/MTL.tsv"
-    )
-    set_tests_properties(test-fusion PROPERTIES FIXTURES_REQUIRED generate-models)
 endif()
 
 llama_build_and_test(test-chat-peg-parser.cpp peg-parser/simple-tokenize.cpp)