fix fc_fuse pass (#37694)

* fix fc_fuse

* modify cmake notest,test=windows_ci

* retrigger all the ci

paddle/fluid/framework/ir/fc_fuse_pass.cc

@@ -130,6 +130,32 @@ int FCFusePass::ApplyFCPattern(Graph* graph, bool with_relu) const {
     GET_IR_NODE_FROM_SUBGRAPH(mul, mul, fc_pattern);
     GET_IR_NODE_FROM_SUBGRAPH(elementwise_add, elementwise_add, fc_pattern);
     GET_IR_NODE_FROM_SUBGRAPH(mul_out, mul_out, fc_pattern);
+
+    // Only support 2D-Tensor as weight for FC
+    std::vector<int64_t> w_shape = w->Var()->GetShape();
+    size_t w_rank = w_shape.size();
+    if (w_rank != 2) return;
+
+    // axis of elementwise_add should be -1 or x_num_col_dims
+    auto x_num_col_dims =
+        BOOST_GET_CONST(int, mul->Op()->GetAttr("x_num_col_dims"));
+    auto axis = BOOST_GET_CONST(int, elementwise_add->Op()->GetAttr("axis"));
+    if (axis != -1 && axis != x_num_col_dims) return;
+
+    // Shape of bias should be [1, out_size] or [out_size]
+    std::vector<int64_t> b_shape = bias->Var()->GetShape();
+    if (b_shape.size() == 1) {
+      if (b_shape[0] != w_shape[1]) {
+        return;
+      }
+    } else if (b_shape.size() == 2) {
+      if (b_shape[0] != 1 || b_shape[1] != w_shape[1]) {
+        return;
+      }
+    } else {
+      return;
+    }
+
     Node* relu = nullptr;
     Node* relu_out = nullptr;
     if (with_relu) {

paddle/fluid/framework/ir/fc_fuse_pass_tester.cc

@@ -55,14 +55,14 @@ TEST(FCFusePass, basic) {
   auto* bias_0 = layers.data("conv2d_bias_0", {}, true);
   auto* conv2d_out = layers.conv2d(a, filters_0, bias_0, false);
   auto* relu_out_0 = layers.relu(conv2d_out);
-  auto* weights_0 = layers.data("weights_0", {}, true);
+  auto* weights_0 = layers.data("weights_0", {5, 4}, true);
   auto* mul_out_0 = layers.mul(relu_out_0, weights_0);
-  auto* bias_1 = layers.data("bias_1", {}, true);
+  auto* bias_1 = layers.data("bias_1", {4}, true);
   auto* add_out_0 = layers.elementwise_add(mul_out_0, bias_1, nullptr, 1);
   auto* relu_out_1 = layers.relu(add_out_0);
-  auto* weights_1 = layers.data("weights_1", {}, true);
+  auto* weights_1 = layers.data("weights_1", {8, 9}, true);
   auto* mul_out_1 = layers.mul(relu_out_1, weights_1);
-  auto* bias_2 = layers.data("bias_2", {}, true);
+  auto* bias_2 = layers.data("bias_2", {1, 9}, true);
   auto* add_out_1 = layers.elementwise_add(mul_out_1, bias_2, nullptr, 1);
   VLOG(4) << add_out_1;
 

python/paddle/fluid/tests/unittests/ir/inference/CMakeLists.txt

@@ -71,8 +71,11 @@ set_tests_properties(test_trt_matmul_quant_dequant PROPERTIES TIMEOUT 100)
 set_tests_properties(test_trt_conv3d_op PROPERTIES TIMEOUT 60)
 set_tests_properties(test_trt_conv3d_transpose_op PROPERTIES TIMEOUT 60)
 set_tests_properties(test_trt_nearest_interp_v2_op PROPERTIES TIMEOUT 30)
+
+if (WITH_MKLDNN AND TENSORRT_FOUND AND WITH_GPU)
 set_tests_properties(test_emb_eltwise_layernorm_fuse_pass PROPERTIES TIMEOUT 120)
-set_tests_properties(test_fc_fuse_pass PROPERTIES TIMEOUT 120)
+set_tests_properties(test_fc_fuse_pass PROPERTIES TIMEOUT 240)
+endif()
 
 if (WITH_MKLDNN)
   set_tests_properties(test_mkldnn_prelu_op PROPERTIES TIMEOUT 300)

python/paddle/fluid/tests/unittests/ir/inference/test_fc_fuse_pass.py

@@ -46,6 +46,17 @@ class TestFcFusePass(PassAutoScanTest):
         config = self.create_inference_config(use_gpu=True)
         yield config, ["fc"], (1e-5, 1e-5)
 
+        # trt static_shape
+        config = self.create_trt_inference_config()
+        config.enable_tensorrt_engine(
+            max_batch_size=8,
+            workspace_size=102400,
+            min_subgraph_size=0,
+            precision_mode=paddle_infer.PrecisionType.Float32,
+            use_static=False,
+            use_calib_mode=False)
+        yield config, ['fc'], (1e-5, 1e-5)
+
     def add_ignore_pass_case(self):
         # Here we put some skip rules to avoid known bugs
         def teller1(program_config, predictor_config):
@@ -53,14 +64,22 @@ class TestFcFusePass(PassAutoScanTest):
             x_shape = list(program_config.inputs["mul_x"].shape)
             y_shape = list(program_config.weights["mul_y"].shape)
             bias_shape = program_config.weights["bias"].shape
-            if (bias_shape != [y_shape[-1], ] and
-                    bias_shape != [1, y_shape[-1]]):
+            bias_shape = list(program_config.weights["bias"].shape)
+
+            if predictor_config.tensorrt_engine_enabled():
+                # TensorRT cann't handle all the situation of elementwise_add
+                # disable it until this problem fixed
+                predictor_config.exp_disable_tensorrt_ops(["elementwise_add"])
+
+            if bias_shape != [y_shape[-1]] and bias_shape != [1, y_shape[-1]]:
                 return True
             return False
 
         def teller2(program_config, predictor_config):
             # TODO fuse has bug while axis != -1
-            if program_config.ops[1].attrs["axis"] != -1:
+            axis = program_config.ops[1].attrs["axis"]
+            if axis != -1 and axis != program_config.ops[0].attrs[
+                    "x_num_col_dims"]:
                 return True
             return False
 
@@ -164,7 +183,7 @@ class TestFcFusePass(PassAutoScanTest):
 
     def test(self):
         self.run_and_statis(
-            quant=False, max_examples=300, passes=["fc_fuse_pass"])
+            quant=False, max_examples=500, passes=["fc_fuse_pass"])
 
 
 if __name__ == "__main__":