CUDA: fix get_rows_back for tables with more than 65535 rows (grid-y clamp + stride) (#25103)

2026-06-30 05:16:24 -07:00
parent e495d1e748
commit 931eb37f8c
2 changed files with 18 additions and 14 deletions
--- a/tests/test-backend-ops.cpp
+++ b/tests/test-backend-ops.cpp
@@ -7759,6 +7759,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
    }

    test_cases.emplace_back(new test_get_rows_back(GGML_TYPE_F32, 1, 8, 2, 1, false));
+    test_cases.emplace_back(new test_get_rows_back(GGML_TYPE_F32, 1, 70000, 4, 1, false)); // row count > CUDA grid-y limit (65535)
    for (ggml_type type : all_types) {
        for (bool v : {false, true}) {
            test_cases.emplace_back(new test_get_rows_back(type, 256, 5, 4, 1, v));