ggml: update unit tests for the new vec_dot interface

Author: snadampal

pocs/vdot/q8dot.cpp

@@ -156,8 +156,11 @@ int main(int argc, char** argv) {
 
         t1 = std::chrono::high_resolution_clock::now();
         float fs;
-        if (type == 0) funcs.vec_dot(kVecSize * QK4_1, &fs, x40.data(), y.data());
-        else funcs.vec_dot(kVecSize * QK4_1, &fs, x41.data(), y.data());
+        const void * x40_data_l[1] = {x40.data()};
+        const void * x41_data_l[1] = {x41.data()};
+        const void * y_data_l[1] = {y.data()};
+        if (type == 0) funcs.vec_dot(kVecSize * QK4_1, &fs, x40_data_l, y_data_l, MAX_NUM_ROWS_COLS_DOT_PRODUCT);
+        else funcs.vec_dot(kVecSize * QK4_1, &fs, x41_data_l, y_data_l, MAX_NUM_ROWS_COLS_DOT_PRODUCT);
         t2 = std::chrono::high_resolution_clock::now();
         t = 1e-3*std::chrono::duration_cast<std::chrono::nanoseconds>(t2-t1).count();
         if (iloop > 3) ggml.addResult(fs, t);

pocs/vdot/vdot.cpp

@@ -285,8 +285,11 @@ int main(int argc, char** argv) {
         else {
             auto vdot = ggml_internal_get_type_traits(funcs.vec_dot_type);
             vdot.from_float(y1.data(), q8.data(), kVecSize);
-            if (useQ4_1) funcs.vec_dot(kVecSize, &result, q41.data(), q8.data());
-            else funcs.vec_dot(kVecSize, &result, q40.data(), q8.data());
+            const void * q40_data_l[1] = {q40.data()};
+            const void * q41_data_l[1] = {q41.data()};
+            const void * q8_data_l[1] = {q8.data()};
+            if (useQ4_1) funcs.vec_dot(kVecSize, &result, q41_data_l, q8_data_l, MAX_NUM_ROWS_COLS_DOT_PRODUCT);
+            else funcs.vec_dot(kVecSize, &result, q40_data_l, q8_data_l, MAX_NUM_ROWS_COLS_DOT_PRODUCT);
         }
         sumq += result;
         t2 = std::chrono::high_resolution_clock::now();

tests/test-quantize-fns.cpp

@@ -85,7 +85,9 @@ static float dot_product_error(
     vdot.from_float(test_data2, tmp_q2.data(), test_size);
 
     float result = INFINITY;
-    qfns.vec_dot(test_size, &result, tmp_q1.data(), tmp_q2.data());
+    const void * tmp_q1_data_l[1] = {tmp_q1.data()};
+    const void * tmp_q2_data_l[1] = {tmp_q2.data()};
+    qfns.vec_dot(test_size, &result, tmp_q1_data_l, tmp_q2_data_l, MAX_NUM_ROWS_COLS_DOT_PRODUCT);
 
     const float dot_ref = dot_product(test_data1, test_data2, test_size);
 

tests/test-quantize-perf.cpp

@@ -344,7 +344,9 @@ int main(int argc, char * argv[]) {
                     printf("    %zu values (%.2f MB)\n", size, 4*size/(float)(1024*1024));
                     auto quantize_fn = [&](void) -> float {
                         float result;
-                        qfns.vec_dot(size, &result, test_q1, test_q2);
+                        const void * test_q1_l[1] = {test_q1};
+                        const void * test_q2_l[1] = {test_q2};
+                        qfns.vec_dot(size, &result, test_q1_l, test_q2_l, MAX_NUM_ROWS_COLS_DOT_PRODUCT);
                         return result;
                     };
                     size_t quantized_size = ggml_row_size(type, size);