metal : fix binary ops for ne10 % 4 != 0

Author: ggerganov

ggml-metal.m

@@ -1080,7 +1080,7 @@ void ggml_metal_graph_compute(
 
                             int64_t nb = ne00;
 
-                            if (ggml_nelements(src1) == ne10 && ggml_is_contiguous(src1) && ne00 % 4 == 0) {
+                            if (ggml_nelements(src1) == ne10 && ggml_is_contiguous(src1) && ne00 % 4 == 0 && ne10 % 4 == 0) {
                                 GGML_ASSERT(ggml_is_contiguous(src0));
 
                                 // src1 is a row

tests/test-backend-ops.cpp

@@ -1272,6 +1272,7 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op
     };
 
     add_test_bin_bcast(GGML_TYPE_F32, {1, 1, 8, 1}, {1, 1, 1, 1});
+    add_test_bin_bcast(GGML_TYPE_F32, {1, 1, 1, 1}, {32, 1, 1, 1});
     add_test_bin_bcast(GGML_TYPE_F32, {1, 1, 320, 320}, {1, 1, 1, 1});
     add_test_bin_bcast(GGML_TYPE_F32, {16, 10, 1, 1}, {1, 1, 1, 1});
     add_test_bin_bcast(GGML_TYPE_F32, {16, 10, 10, 1}, {1, 1, 1, 1});