[GPU] Fix some conv and pooling unit tests for DG2 (#13016)

* change activation function for onednn
* adaptive tolerance
* higher tolerance for multiple eltwise test
* enable all eltwise_mode of postop eltwise

src/plugins/intel_gpu/src/graph/graph_optimizer/prepare_primitive_fusing.cpp

@@ -1034,12 +1034,7 @@ void prepare_primitive_fusing::fuse_simple_primitives(program &p) {
                 auto eltw_in_size = peer_node->get_output_layout();
                 if (eltw_in_size.is_dynamic())
                     return;
-                // Temporary disable mul fusion with full tensor as onednn doesn't support it
-                if (fused_node->is_type<convolution>() && prim->mode == eltwise_mode::prod &&
-                    (eltw_in_size.spatial(0) > 1 || eltw_in_size.spatial(1) > 1 || eltw_in_size.batch() > 1))
-                    return;
             }
-
             if (parent1->is_type<convolution>() && !conv_supports_fusings(parent1->as<convolution>()))
                 return;
 

src/plugins/intel_gpu/src/graph/program_node.cpp

@@ -986,8 +986,6 @@ void program_node::init_onednn_primitive_attributes() {
                     update_onednn_post_op_list(onednn_post_op_type::binary_add, dep_idx);
                 }
             } else {
-                if (in.spatial(0) > 1 || in.spatial(1) > 1 || in.batch() > 1)
-                    throw std::runtime_error("Unsupported eltwise mode for fused onednn op");
                 // convolution using post-op output scales can only be int8/uint8
                 if (idx == 0 && !has_out_scales(attrs) && !is_type<pooling>() && !is_type<reduce>() &&
                     !(is_type<convolution>() && data_type_traits::is_floating_point(output_layout.data_type))) {

src/plugins/intel_gpu/tests/fusions/convolution_fusion_test.cpp

@@ -448,6 +448,7 @@ TEST_P(conv_fp32_activation, basic) {
         reorder("reorder_bfyx", "activation", p.default_format, data_types::f32)
     );
 
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -477,7 +478,7 @@ TEST_P(conv_fp32_scale, basic) {
         reorder("reorder_bfyx", "scale", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -507,7 +508,7 @@ TEST_P(conv_fp32_bias, basic) {
         reorder("reorder_bfyx", "add_bias", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -539,7 +540,7 @@ TEST_P(conv_fp32_double_bias, basic) {
         reorder("reorder_bfyx", "add_bias2", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -563,7 +564,7 @@ TEST_P(conv_fp32_prelu_eltwise, basic_sum) {
         reorder("reorder_bfyx", "eltwise", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -581,7 +582,7 @@ TEST_P(conv_fp32_prelu_eltwise, basic_sum_slope_2) {
         reorder("reorder_bfyx", "eltwise", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -599,7 +600,7 @@ TEST_P(conv_fp32_prelu_eltwise, basic_prod) {
         reorder("reorder_bfyx", "eltwise", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -617,7 +618,7 @@ TEST_P(conv_fp32_prelu_eltwise, basic_prod_slope_2) {
         reorder("reorder_bfyx", "eltwise", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -636,7 +637,7 @@ TEST_P(conv_fp32_prelu_eltwise, eltw_broadcast_sum) {
         reorder("reorder_bfyx", "eltwise", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -655,7 +656,7 @@ TEST_P(conv_fp32_prelu_eltwise, eltw_broadcast_sum_slope_2) {
         reorder("reorder_bfyx", "eltwise", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -674,7 +675,7 @@ TEST_P(conv_fp32_prelu_eltwise, eltw_broadcast_prod) {
         reorder("reorder_bfyx", "eltwise", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -693,7 +694,7 @@ TEST_P(conv_fp32_prelu_eltwise, eltw_broadcast_prod_slope_2) {
         reorder("reorder_bfyx", "eltwise", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -715,7 +716,7 @@ TEST_P(conv_fp32_prelu_eltwise, vector_ops) {
     implementation_desc conv_impl = { format::b_fs_yx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -736,7 +737,7 @@ TEST_P(conv_fp32_prelu_eltwise, vector_ops_slope_2) {
     implementation_desc conv_impl = { format::b_fs_yx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -758,7 +759,7 @@ TEST_P(conv_fp32_prelu_eltwise, vector_ops_mixed_types) {
     implementation_desc conv_impl = { format::b_fs_yx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -780,7 +781,7 @@ TEST_P(conv_fp32_prelu_eltwise, vector_ops_mixed_types_slope_2) {
     implementation_desc conv_impl = { format::b_fs_yx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -816,7 +817,7 @@ TEST_P(conv_fp32_multi_eltwise_2, basic) {
     implementation_desc conv_impl = { format::b_fs_yx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -852,7 +853,7 @@ TEST_P(conv_fp32_multi_eltwise_2_clamp, basic) {
     implementation_desc conv_impl = { format::b_fs_yx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -892,7 +893,7 @@ TEST_P(conv_fp32_multi_eltwise_4_clamp, basic) {
     implementation_desc conv_impl = { format::b_fs_yx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -931,7 +932,7 @@ TEST_P(conv_fp32_eltwise_fusing_extend_ops, pattern01_simple_sub) {
     implementation_desc conv_impl = { format::b_fs_yx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -959,7 +960,7 @@ TEST_P(conv_fp32_eltwise_fusing_extend_ops, pattern02_sub_scale) {
     implementation_desc conv_impl = { format::b_fs_yx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -988,7 +989,7 @@ TEST_P(conv_fp32_eltwise_fusing_extend_ops, pattern03_sub_div) {
     implementation_desc conv_impl = { format::b_fs_yx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1026,7 +1027,7 @@ TEST_P(conv_fp32_eltwise_fusing_2conv, basic) {
     implementation_desc conv_impl = { format::b_fs_yx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim0", conv_impl }, { "conv_prim", conv_impl }  }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1063,7 +1064,7 @@ TEST_P(conv_fp32_multi_eltwise_3_fusing, basic) {
     implementation_desc conv_impl = { format::b_fs_yx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1134,7 +1135,7 @@ TEST_P(conv_fp32_multi_eltwise_concat, basic) {
     implementation_desc conv_impl = { format::b_fs_yx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1165,7 +1166,7 @@ TEST_P(conv_fp32_eltwise_b_fs_zyx_fsv16, vector_ops) {
     implementation_desc conv_impl = { format::b_fs_zyx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1182,7 +1183,7 @@ TEST_P(conv_fp32_swish, basic) {
         reorder("reorder_bfyx", "mul", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1218,7 +1219,7 @@ TEST_P(conv_fp32_eltwise_b_fs_zyx_fsv16, splitted_vector_ops) {
     implementation_desc conv_impl = { format::b_fs_zyx_fsv16, "" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     //  commented because split mode is disabled
     //  execute(p);
 }
@@ -1257,7 +1258,7 @@ TEST_P(conv_fp32_quantize_u8_first_conv, basic) {
         reorder("reorder_bfyx", "quantize", p.default_format, data_types::f32)
     );
 
-    tolerance = 1.0f;
+    tolerance = 1.f;
     execute(p);
 }
 
@@ -1281,7 +1282,7 @@ TEST_P(conv_fp32_quantize_u8, basic) {
         reorder("reorder_bfyx", "quantize", p.default_format, data_types::f32)
     );
 
-    tolerance = 1.0f;
+    tolerance = 1.f;
     execute(p);
 }
 
@@ -1314,7 +1315,7 @@ TEST_P(conv_fp32_scale_quantize_i8, basic) {
     // Output elements are in range [-127, 127]
     // 1.0f difference is allowed, since quantize can return different values in ref and scale_shift kernels
     // due to big error of division (in ref kernel).
-    tolerance = 1.0f;
+    tolerance = 1.f;
     execute(p);
 }
 
@@ -1346,7 +1347,7 @@ TEST_P(conv_fp32_scale_activation_quantize_i8, basic) {
         reorder("reorder_bfyx", "quantize", p.default_format, data_types::f32)
     );
 
-    tolerance = 1.0f;
+    tolerance = 1.f;
     execute(p);
 }
 
@@ -1380,7 +1381,7 @@ TEST_P(conv_fp32_scale_activation_quantize_u8_eltwise_fp32, basic) {
         reorder("reorder_bfyx", "sum", p.default_format, data_types::f32)
     );
 
-    tolerance = 1.0f;
+    tolerance = 1.f;
     execute(p);
 }
 
@@ -1414,7 +1415,7 @@ TEST_P(conv_fp32_scale_activation_quantize_i8_activation, basic) {
         reorder("reorder_bfyx", "activation_quantize", p.default_format, data_types::f32)
     );
 
-    tolerance = 1.0f;
+    tolerance = 1.f;
     execute(p);
 }
 
@@ -1507,7 +1508,7 @@ TEST_P(conv_fp32_activation_eltwise_diff_sizes, basic) {
         reorder("reorder_bfyx", "sum", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1575,7 +1576,7 @@ TEST_P(conv_int8_scale, basic) {
         reorder("reorder_bfyx", "scale", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1591,7 +1592,7 @@ TEST_P(conv_int8_scale, fp16_scale_out) {
         reorder("reorder_bfyx", "scale", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1630,7 +1631,7 @@ TEST_P(conv_int8_eltwise, fp16_eltwise_out) {
         reorder("reorder_bfyx", "scale", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1675,7 +1676,8 @@ TEST_P(conv_int8_scale_shift_swish, basic) {
         reorder("reorder_bfyx", "mul", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-3f;
+    // high tolerance because many eltwise operations
+    tolerance = default_tolerance(p.default_type) * 10;
     execute(p);
 }
 
@@ -1716,7 +1718,7 @@ TEST_P(conv_int8_prelu_eltwise, basic) {
         reorder("reorder_bfyx", "eltwise", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1734,7 +1736,7 @@ TEST_P(conv_int8_prelu_eltwise, basic_slope_2) {
         reorder("reorder_bfyx", "eltwise", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1760,7 +1762,7 @@ TEST_P(conv_int8_prelu_eltwise, fsv16) {
         return;
     }
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1786,7 +1788,7 @@ TEST_P(conv_int8_prelu_eltwise, fsv16_slope_2) {
         return;
     }
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1923,7 +1925,7 @@ TEST_P(conv_int8_activation_eltwise, fsv16) {
     if (engine.get_device_info().supports_immad)
         p.expected_fused_primitives += 2;
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1951,7 +1953,7 @@ TEST_P(conv_int8_activation_eltwise, fsv32) {
     if (engine.get_device_info().supports_immad)
         p.expected_fused_primitives += 2;
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -1986,7 +1988,7 @@ TEST_P(conv_int8_quantize_u8, per_channel) {
         reorder("reorder_bfyx", "quantize", p.default_format, data_types::f32)
     );
 
-    tolerance = 1.0f;
+    tolerance = 1.f;
     execute(p);
 }
 
@@ -2005,7 +2007,7 @@ TEST_P(conv_int8_quantize_u8, per_tensor) {
         reorder("reorder_bfyx", "quantize", p.default_format, data_types::f32)
     );
 
-    tolerance = 1.0f;
+    tolerance = 1.f;
     execute(p);
 }
 
@@ -2052,7 +2054,7 @@ TEST_P(conv_int8_scale_quantize_i8, basic) {
     // Output elements are in range [-127, 127]
     // 1.0f difference is allowed, since quantize can return different values in ref and scale_shift kernels
     // due to big error of division (in ref kernel).
-    tolerance = 1.0f;
+    tolerance = 1.f;
     execute(p);
 }
 
@@ -2102,7 +2104,7 @@ TEST_P(conv_int8_scale_quantize_i8_conv_b_fs_yx_fsv4_int8, basic) {
     implementation_desc conv_impl = { format::b_fs_yx_fsv4, "convolution_gpu_b_fs_yx_fsv4_int8" };
     bo_fused.set_option(build_option::force_implementations({ { "conv_prim", conv_impl } }));
 
-    tolerance = 1.0f;
+    tolerance = 1.f;
     execute(p);
 }
 
@@ -2129,7 +2131,7 @@ TEST_P(conv_int8_relu_quantize, i8) {
     // Output elements are in range [-127, 127]
     // 1.0f difference is allowed, since quantize can return different values in ref and scale_shift kernels
     // due to big error of division (in ref kernel).
-    tolerance = 1.0f;
+    tolerance = 1.f;
     execute(p);
 }
 
@@ -2149,7 +2151,7 @@ TEST_P(conv_int8_relu_quantize, u8) {
         reorder("reorder_bfyx", "quantize", p.default_format, data_types::f32)
     );
 
-    tolerance = 1.0f;
+    tolerance = 1.f;
     execute(p);
 }
 
@@ -2669,7 +2671,7 @@ TEST_P(conv_i8_activation_eltwise_diff_sizes, basic) {
         reorder("reorder_bfyx", "sum", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -2718,7 +2720,7 @@ TEST_P(conv_fp16_scale, basic) {
         reorder("reorder_bfyx", "scale", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-5f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -2947,19 +2949,19 @@ TEST_P(conv_int8_eltwise_onednn, u8_eltwise_prod_out) {
 }
 
 INSTANTIATE_TEST_SUITE_P(fusings_gpu, conv_int8_eltwise_onednn, ::testing::ValuesIn(std::vector<convolution_test_params>{
-    convolution_test_params{ CASE_CONV_U8S8_1, 2, 4 },
-    convolution_test_params{ CASE_CONV_U8S8_2, 2, 4 },
-    convolution_test_params{ CASE_CONV_U8S8_3, 2, 4 },
-    convolution_test_params{ CASE_CONV_S8S8_1, 2, 4 },
-    convolution_test_params{ CASE_CONV_S8S8_2, 2, 4 },
-    convolution_test_params{ CASE_CONV_S8S8_3, 2, 4 },
+    convolution_test_params{ CASE_CONV_U8S8_1, 3, 4 },
+    convolution_test_params{ CASE_CONV_U8S8_2, 3, 4 },
+    convolution_test_params{ CASE_CONV_U8S8_3, 3, 4 },
+    convolution_test_params{ CASE_CONV_S8S8_1, 3, 4 },
+    convolution_test_params{ CASE_CONV_S8S8_2, 3, 4 },
+    convolution_test_params{ CASE_CONV_S8S8_3, 3, 4 },
 
-    convolution_test_params{ CASE_CONV_U8S8_11, 2, 4 },
-    convolution_test_params{ CASE_CONV_U8S8_12, 2, 4 },
-    convolution_test_params{ CASE_CONV_U8S8_13, 2, 4 },
-    convolution_test_params{ CASE_CONV_S8S8_12, 2, 4 },
-    convolution_test_params{ CASE_CONV_S8S8_13, 2, 4 },
-    convolution_test_params{ CASE_CONV_S8S8_14, 2, 4 },
+    convolution_test_params{ CASE_CONV_U8S8_11, 3, 4 },
+    convolution_test_params{ CASE_CONV_U8S8_12, 3, 4 },
+    convolution_test_params{ CASE_CONV_U8S8_13, 3, 4 },
+    convolution_test_params{ CASE_CONV_S8S8_12, 3, 4 },
+    convolution_test_params{ CASE_CONV_S8S8_13, 3, 4 },
+    convolution_test_params{ CASE_CONV_S8S8_14, 3, 4 },
 
     convolution_test_params{ CASE_CONV3D_U8S8_1, 3, 4 },
     convolution_test_params{ CASE_CONV3D_U8S8_2, 3, 4 },
@@ -2983,7 +2985,7 @@ TEST_P(conv_fp32_activation_abs_onednn, basic) {
         reorder("reorder_bfyx", "activation", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-2f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -3006,7 +3008,7 @@ TEST_P(conv_fp32_activation_mish_onednn, basic) {
         reorder("reorder_bfyx", "activation", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-2f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -3029,7 +3031,7 @@ TEST_P(conv_fp32_activation_swish_onednn, basic) {
         reorder("reorder_bfyx", "activation", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-2f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -3052,7 +3054,7 @@ TEST_P(conv_fp32_activation_hswish_onednn, basic) {
         reorder("reorder_bfyx", "activation", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-2f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -3075,7 +3077,7 @@ TEST_P(conv_fp32_activation_exp_onednn, basic) {
         reorder("reorder_bfyx", "activation", p.default_format, data_types::f32)
     );
 
-    tolerance = 1e-2f;
+    tolerance = default_tolerance(p.default_type);
     execute(p);
 }
 
@@ -3102,7 +3104,7 @@ TEST_P(conv_int8_quantize_u8_onednn, per_channel) {
         reorder("reorder_bfyx", "quantize", p.default_format, data_types::f32)
     );
 
-    tolerance = 1.0f;
+    tolerance = 1.f;
     execute(p);
 }
 
@@ -3121,7 +3123,7 @@ TEST_P(conv_int8_quantize_u8_onednn, per_tensor) {
         reorder("reorder_bfyx", "quantize", p.default_format, data_types::f32)
     );
 
-    tolerance = 1.0f;
+    tolerance = 1.f;
     execute(p);
 }
 
@@ -3649,6 +3651,8 @@ struct convolution_eltw_sum_test_params {
 class EltwiseSumFusingTestOneDNN : public BaseFusingTest<convolution_eltw_sum_test_params> {
 public:
     void execute(convolution_eltw_sum_test_params& p) {
+        if (!engine.get_device_info().supports_immad)
+            return;
         auto input_prim = p.data_type == data_types::u8 ? get_mem(get_input_layout(p), 0, 10) : get_mem(get_input_layout(p));
 
         network network_not_fused(this->engine, this->topology_non_fused, bo_not_fused);
@@ -3665,7 +3669,7 @@ public:
             return false;
         });
 
-        if (info_fused != pi_fused.end() && engine.get_device_info().supports_immad) {
+        if (info_fused != pi_fused.end()) {
             std::cout << "kernel: " << info_fused->kernel_id << std::endl;
             EXPECT_TRUE(info_fused->kernel_id.find("jit:ir") != std::string::npos);
         }
@@ -3707,8 +3711,8 @@ TEST_P(onednn_binary_add_full_tensor, basic) {
 }
 
 // in_shape; out_shape; kernel; stride; pad; dilation; groups; data_type; input_format; weights_type; weights_format; eltw_type; eltw_format; out_type; out_format; default_type; default_format;
-#define CASE_CONV_ELTW_SUM_BINARY_ADD_1 { 1, 32, 4, 4 }, { 1, 16, 4, 4 }, { 1, 1, 3, 3 }, { 1, 1 }, { 1, 1 }, { 1, 1 }, 1, data_types::u8, format::b_fs_yx_fsv32, data_types::i8, format::bfyx, data_types::u8, format::b_fs_yx_fsv32, data_types::f16, format::b_fs_yx_fsv16, data_types::f32, format::bfyx
-#define CASE_CONV_ELTW_SUM_SUM_1        { 1, 32, 4, 4 }, { 1, 16, 4, 4 }, { 1, 1, 3, 3 }, { 1, 1 }, { 1, 1 }, { 1, 1 }, 1, data_types::u8, format::b_fs_yx_fsv32, data_types::i8, format::bfyx, data_types::u8, format::b_fs_yx_fsv32, data_types::u8, format::b_fs_yx_fsv32, data_types::f32, format::bfyx
+#define CASE_CONV_ELTW_SUM_BINARY_ADD_1     { 1, 32, 4, 4 }, { 1, 16, 4, 4 }, { 1, 1, 3, 3 }, { 1, 1 }, { 1, 1 }, { 1, 1 }, 1, data_types::u8, format::b_fs_yx_fsv32, data_types::i8, format::bfyx, data_types::u8, format::b_fs_yx_fsv32, data_types::f16, format::b_fs_yx_fsv16, data_types::f32, format::bfyx
+#define CASE_CONV_ELTW_SUM_SUM_1            { 1, 32, 4, 4 }, { 1, 16, 4, 4 }, { 1, 1, 3, 3 }, { 1, 1 }, { 1, 1 }, { 1, 1 }, 1, data_types::u8, format::b_fs_yx_fsv32, data_types::i8, format::bfyx, data_types::u8, format::b_fs_yx_fsv32, data_types::u8, format::b_fs_yx_fsv32, data_types::f32, format::bfyx
 #define CASE_CONV_ELTW_SUM_SUM_DIFF_DTYPE_1 { 1, 32, 4, 4 }, { 1, 16, 4, 4 }, { 1, 1, 3, 3 }, { 1, 1 }, { 1, 1 }, { 1, 1 }, 1, data_types::u8, format::b_fs_yx_fsv32, data_types::i8, format::bfyx, data_types::i8, format::b_fs_yx_fsv32, data_types::u8, format::b_fs_yx_fsv32, data_types::f32, format::bfyx
 
 INSTANTIATE_TEST_SUITE_P(eltwise_sum_fusings_gpu, onednn_binary_add_full_tensor, ::testing::ValuesIn(std::vector<convolution_eltw_sum_test_params>{

src/plugins/intel_gpu/tests/fusions/pooling_fusion_test.cpp

@@ -304,13 +304,10 @@ TEST_P(pooling_scale_activation_quantize, per_channel) {
         data("scale_data", get_mem(get_per_channel_layout(p), 1.0f / 16.0f)),
         pooling("pooling", "input", "", p.pool_mode, kernel, stride, pad),
         eltwise("scale", { "pooling", "scale_data" }, eltwise_mode::prod, p.default_type),
-        activation("activation", "scale", activation_func::atan),
+        activation("activation", "scale", activation_func::hyperbolic_tan),
         quantize("quantize", "activation", "in_lo", "in_hi", "out_lo", "out_hi", 255, data_types::u8),
         reorder("output_reorder", "quantize", p.default_format, data_types::f32)
     );
-    // Activation won't be fused because onednn doesn't support atan activation
-    if (engine.get_device_info().supports_immad)
-        p.expected_fused_primitives++;
 
     tolerance = 1.0f;
     execute(p);

src/plugins/intel_gpu/tests/test_utils/test_utils.cpp

@@ -373,6 +373,21 @@ std::vector<cldnn::data_types> generic_test::test_data_types() {
     return result;
 }
 
+double default_tolerance(data_types dt) {
+    switch (dt) {
+    case data_types::f16:
+        return 1e-3;
+    case data_types::f32:
+        return 1e-5;
+    case data_types::i8:
+    case data_types::u8:
+        return 1.;
+    default:
+        IE_THROW() << "Unknown";
+    }
+    IE_THROW() << "Unknown";
+}
+
 std::vector<cldnn::format> generic_test::test_input_formats = { cldnn::format::bfyx , cldnn::format::yxfb, cldnn::format::fyxb, cldnn::format::byxf };
 std::vector<int32_t> generic_test::test_batch_sizes = { 1, 2 };// 4, 8, 16};
 std::vector<int32_t> generic_test::test_feature_sizes = { 1, 2 };// , 3, 15};

src/plugins/intel_gpu/tests/test_utils/test_utils.h

@@ -565,6 +565,8 @@ T div_up(const T a, const U b) {
     return (a + b - 1) / b;
 }
 
+double default_tolerance(data_types dt);
+
 // inline void print_bin_blob(cldnn::memory& mem, std::string name)
 // {
 //     auto&& size = mem.get_layout().get_tensor();