add q8_0 q4_0 tests

Author: JohannesGaessler

tests/test-backend-ops.cpp

@@ -1540,16 +1540,18 @@ struct test_flash_attn_ext : public test_case {
 
     const float max_bias; // ALiBi
 
+    const ggml_type type_KV;
+
     std::string vars() override {
-        return VARS_TO_STR6(hs, nh, kv, nb, mask, max_bias);
+        return VARS_TO_STR7(hs, nh, kv, nb, mask, max_bias, type_KV);
     }
 
     double max_nmse_err() override {
         return 5e-4;
     }
 
-    test_flash_attn_ext(int64_t hs = 128, int64_t nh = 32, int64_t kv = 96, int64_t nb = 8, bool mask = true, float max_bias = 0.0f)
-        : hs(hs), nh(nh), kv(kv), nb(nb), mask(mask), max_bias(max_bias) {}
+    test_flash_attn_ext(int64_t hs = 128, int64_t nh = 32, int64_t kv = 96, int64_t nb = 8, bool mask = true, float max_bias = 0.0f, ggml_type type_KV = GGML_TYPE_F16)
+        : hs(hs), nh(nh), kv(kv), nb(nb), mask(mask), max_bias(max_bias), type_KV(type_KV) {}
 
     ggml_tensor * build_graph(ggml_context * ctx) override {
         ggml_tensor * q = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, hs, nb, nh, 1);
@@ -2238,7 +2240,9 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op
                 for (int nh : { 32, }) {
                     for (int kv : { 512, 1024, }) {
                         for (int nb : { 1, 2, 4, 8, }) {
-                            test_cases.emplace_back(new test_flash_attn_ext(hs, nh, kv, nb, mask, max_bias));
+                            for (ggml_type type_KV : {GGML_TYPE_F16, GGML_TYPE_Q8_0, GGML_TYPE_Q4_0}) {
+                                test_cases.emplace_back(new test_flash_attn_ext(hs, nh, kv, nb, mask, max_bias, type_KV));
+                            }
                         }
                     }
                 }