vulkan : fix Qantized Mat-Vec Mul on AMD GPUs for ncols < 64 (llama/8855)

* Fix Vulkan mul mat vec invalid results when ncols < warp size

* Only run backend ops mul mat vec block size test if block size not already covered

diff --git a/tests/test-backend-ops.cpp b/tests/test-backend-ops.cpp
index f5065f14515628d54633301ab30bf8f0a135a0d1..54cef05c3de3d376ce548dc4264498f325e5f561 100644 (file)
--- a/tests/test-backend-ops.cpp
+++ b/tests/test-backend-ops.cpp
@@ -2271,9 +2271,10 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op
 
     for (ggml_type type_a : other_types) {
         for (ggml_type type_b : {GGML_TYPE_F32}) {
-
-            test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, ggml_blck_size(type_a), { 1,  1}, {1, 1}));
-            test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, { 1,  1}, {1, 1}));
+            if (ggml_blck_size(type_a) != 256) {
+                test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, ggml_blck_size(type_a), {1,  1}, {1, 1}));
+            }
+            test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {1,  1}, {1, 1}));
         }
     }