From 00083d13b6b01a3e72d2b688c9608d47e7255335 Mon Sep 17 00:00:00 2001
From: Megvii Engine Team <megengine@megvii.com>
Date: Thu, 20 May 2021 14:41:06 +0800
Subject: [PATCH] fix(dnn/cuda): fix recursive algo search for
 fallback_nchw_qs8

GitOrigin-RevId: 6be2991224bced3a38a17b6b888fd4f324d03f9f
---
 dnn/src/cuda/conv_bias/algo.h           |  5 ++++-
 dnn/src/cuda/conv_bias/conv_nchwqs8.cpp | 30 +++++++++++++++++++++++++
 2 files changed, 34 insertions(+), 1 deletion(-)

diff --git a/dnn/src/cuda/conv_bias/algo.h b/dnn/src/cuda/conv_bias/algo.h
index e4cd1a2a..869777b6 100644
--- a/dnn/src/cuda/conv_bias/algo.h
+++ b/dnn/src/cuda/conv_bias/algo.h
@@ -575,7 +575,10 @@ public:
         return AlgoAttribute::REPRODUCIBLE;
     }
     MEGDNN_DECL_ALGO_TYPE(CUDA_IMPLICIT_GEMM_NCHW4_DOTPROD_INT8)
-
+    std::vector<SearchItem> get_subopr_list(
+            const TensorLayoutArray& layouts,
+            const OperatorBase* opr) const override;
+ 
 private:
     void make_inner_layout(const SizeArgs& args, TensorLayout& inner_src_layout,
                            TensorLayout& inner_weight_layout,
diff --git a/dnn/src/cuda/conv_bias/conv_nchwqs8.cpp b/dnn/src/cuda/conv_bias/conv_nchwqs8.cpp
index 04a6697a..c8b395e6 100644
--- a/dnn/src/cuda/conv_bias/conv_nchwqs8.cpp
+++ b/dnn/src/cuda/conv_bias/conv_nchwqs8.cpp
@@ -69,6 +69,32 @@ void ConvBiasForwardImpl::AlgoFallbackNCHWQS8::make_inner_layout(
     }
 };
 
+std::vector<Algorithm::SearchItem>
+ConvBiasForwardImpl::AlgoFallbackNCHWQS8::get_subopr_list(
+        const TensorLayoutArray& layouts, const OperatorBase* opr) const {
+    const ConvBiasForwardImpl* o = static_cast<const ConvBiasForwardImpl*>(opr);
+    SizeArgs args(const_cast<ConvBiasForwardImpl*>(o), layouts[0], layouts[1],
+                  layouts[2], layouts[3], layouts[4], nullptr);
+    TensorLayout inner_src_layout;
+    TensorLayout inner_weight_layout;
+    TensorLayout inner_dst_layout;
+    TensorLayout inner_bias_layout;
+    TensorLayout inner_z_layout;
+    make_inner_layout(args, inner_src_layout, inner_weight_layout,
+                      inner_dst_layout, inner_bias_layout, inner_z_layout);
+
+    Param inner_conv_param = o->param();
+    inner_conv_param.format = Param::Format::NCHW4;
+ 
+    std::string param_str;
+    Algorithm::serialize_write_pod(inner_conv_param, param_str);
+
+    return {{Algorithm::OprType::CONVBIAS_FORWARD,
+             param_str,
+             {inner_src_layout, inner_weight_layout, inner_bias_layout,
+              inner_z_layout, inner_dst_layout}}};
+}
+
 bool ConvBiasForwardImpl::AlgoFallbackNCHWQS8::is_available(
         const SizeArgs& args) const {
     if (!args.src_layout->is_contiguous() ||
@@ -109,6 +135,8 @@ WorkspaceBundle ConvBiasForwardImpl::AlgoFallbackNCHWQS8::get_workspace_bundle(
     }
     auto opr = args.handle->create_operator<ConvBiasForward>();
     opr->param() = inner_conv_param;
+    set_execution_policy<ConvBiasForward, ConvBiasForward*>(args.opr,
+                                                            opr.get());
     return WorkspaceBundle(
             ptr,
             {inner_src_layout.span().dist_byte(),
@@ -164,6 +192,8 @@ void ConvBiasForwardImpl::AlgoFallbackNCHWQS8::exec(
     inner_conv_param.format =
             dst_float ? Param::Format::NCHW4_NCHW : Param::Format::NCHW4;
     auto inner_opr = args.handle->create_operator<ConvBiasForward>();
+    set_execution_policy<ConvBiasForward, ConvBiasForward*>(args.opr,
+                                                            inner_opr.get());
     inner_opr->param() = inner_conv_param;
 
     relayout_nchw_nchw4->exec(*args.src_tensor, inner_src, {});