fix(mgb/gopt): fix convert batchnorm to elemwise pass issue

GitOrigin-RevId: eda7f1ab95f9f15448cdcafa2197fa1edd3e7946

src/gopt/impl/inference.cpp

@@ -1592,7 +1592,8 @@ void ConvertBatchNormToElemwisePass::apply(OptState& state) const {
                 SymbolVar bias = {rewriter.get_var(bn->input(2))};
                 SymbolVar mean = {rewriter.get_var(bn->input(3))};
                 SymbolVar variance = {rewriter.get_var(bn->input(4))};
-                SymbolVar invsqrt_variance = opr::PowC::make(variance, {-0.5});
+                SymbolVar invsqrt_variance = opr::PowC::make(variance
+                        + variance.make_scalar_dt(float(bn->param().epsilon)), {-0.5});
                 auto res = scale * (x - mean) * invsqrt_variance + bias;
                 rewriter.replace_var(
                         opr->output(4), res.node(),

src/gopt/test/inference.cpp

@@ -1404,7 +1404,7 @@ TEST(TestGoptInference, ConvertBatchNormPass) {
     auto func = graph->compile({make_callback_copy(y, host_y),
                                 make_callback_copy(y_opt, host_y_opt)});
     func->execute();
-    MGB_ASSERT_TENSOR_NEAR(host_y, host_y_opt, 1e-2);
+    MGB_ASSERT_TENSOR_NEAR(host_y, host_y_opt, 1e-5);
 }
 
 TEST(TestGoptInference, ConvBiasNonlinearityFusePass) {