feat(mgb): fastrun algo profile deduplication

GitOrigin-RevId: 0d1bed781d889e6e105268ea09dc5c96cd4df013

src/opr/impl/search_policy/algo_chooser.cpp

@@ -627,7 +627,7 @@ AlgoChooser<Opr>::AlgoChooserHelper::choose_by_profile(
 }
 
 template <typename Opr>
-typename AlgoChooser<Opr>::ImplAlgoDesc
+std::pair<typename AlgoChooser<Opr>::ImplAlgoDesc, Maybe<AlgoChooserProfileCache::Result>>
 AlgoChooser<Opr>::AlgoChooserHelper::get_profile_result_from_cache(
         const ExecutionStrategy& selected_strategy) const {
     MIDOUT_B(Opr, midout_iv(MGB_HASH_STR("get_profile_result_from_cache")))
@@ -639,11 +639,11 @@ AlgoChooser<Opr>::AlgoChooserHelper::get_profile_result_from_cache(
                                            &origin_param, sizeof(origin_param)};
     auto&& rst = cache.get(cache_key);
     if (!rst.valid())
-        return {};
+        return {{}, rst};
 
     auto&& prof = rst.val();
     if (prof.empty())
-        return {};
+        return {{}, rst};
 
     auto target_attr = extract_algo_attribute(selected_strategy);
     bool skip_by_negative = false;
@@ -657,7 +657,7 @@ AlgoChooser<Opr>::AlgoChooserHelper::get_profile_result_from_cache(
         if (contain_attr_all_positive) {
             if (!contain_attr_any_negative) {
                 Algorithm::Info::Desc algo_desc = deserialize_read_pod(i.algo);
-                return algo_desc;
+                return {algo_desc, rst};
             } else {
                 skip_by_negative = true;
             }
@@ -695,7 +695,7 @@ void AlgoChooser<Opr>::AlgoChooserHelper::construct_execution_policy(
     MIDOUT_B(Opr, midout_iv(MGB_HASH_STR("construct_execution_policy")))
     if (!policy.algo.valid()) {
         if (retrive_from_cache) {
-            policy.algo = get_profile_result_from_cache(selected_strategy);
+            policy.algo = get_profile_result_from_cache(selected_strategy).first;
             if (!policy.algo.valid()) {
                 if (allow_log) {
                     auto target_attr =
@@ -886,7 +886,8 @@ template <typename Opr>
 void AlgoChooser<Opr>::AlgoChooserHelper::profile(
         const ExecutionStrategy& selected_strategy) const {
     MIDOUT_B(Opr, midout_iv(MGB_HASH_STR("profile")))
-    if (get_profile_result_from_cache(selected_strategy).valid())
+    auto&& rst = get_profile_result_from_cache(selected_strategy);
+    if (rst.first.valid())
         return;
     AlgoChooserProfileCache::Result prof_rst;
 
@@ -898,7 +899,20 @@ void AlgoChooser<Opr>::AlgoChooserHelper::profile(
     auto workspace_limit = WorkspaceLimitGetter::get_workspace_limit(
             owner_graph(), m_cn, m_execution_policy.workspace_limit);
     RealTimer timer;
+    std::unordered_set<std::string> rst_algos;
+    if (rst.second.valid()) {
+        std::transform(rst.second.val().begin(), rst.second.val().end(),
+                       std::inserter(rst_algos, rst_algos.end()),
+                       [](const AlgoChooserProfileCache::ResultEntry& result) {
+                           return result.algo;
+                       });
+    }
     for (auto algo : get_all_candidates()) {
+        std::string desc;
+        serialize_write_pod(algo.desc, desc);
+        if (rst_algos.find(desc) != rst_algos.end()) {
+            continue;
+        }
         Maybe<AlgoChooserProfileCache::ResultEntry> cur_rst;
 
         ImplExecutionPolicy policy;
@@ -960,6 +974,9 @@ void AlgoChooser<Opr>::AlgoChooserHelper::profile(
             Algorithm::attribute_str(target_attr.second).c_str(),
             workspace_limit);
     mgb_assert(!prof_rst.empty(), "%s", msg.c_str());
+    if (rst.second.valid())
+        prof_rst.insert(prof_rst.end(), rst.second.val().begin(),
+                        rst.second.val().end());
 
     FixedTensorLayouts incache_layouts = m_incache_layouts;
     typename Opr::Param origin_param = m_dnn_opr->param();
@@ -1058,7 +1075,8 @@ AlgoChooser<Opr>::AlgoChooserHelper::extract_algo_attribute(
     AlgoChooser<megdnn::Opr>::AlgoChooserHelper::choose_by_profile(            \
             const ExecutionStrategy& select_strategy, bool enable_update)      \
             const;                                                             \
-    template typename AlgoChooser<megdnn::Opr>::ImplAlgoDesc                   \
+    template std::pair<typename AlgoChooser<megdnn::Opr>::ImplAlgoDesc,        \
+                       Maybe<AlgoChooserProfileCache::Result>>                 \
     AlgoChooser<megdnn::Opr>::AlgoChooserHelper::                              \
             get_profile_result_from_cache(                                     \
                     const ExecutionStrategy& select_strategy) const;           \

src/opr/include/megbrain/opr/search_policy/algo_chooser.h

@@ -131,7 +131,8 @@ public:
                 bool enable_update) const;
 
         //! get all profile algorithm from cache, return invalid if not exists
-        ImplAlgoDesc get_profile_result_from_cache(
+        std::pair<ImplAlgoDesc, Maybe<AlgoChooserProfileCache::Result>>
+        get_profile_result_from_cache(
                 const ExecutionStrategy& selected_strategy) const;
 
         /**