Enable scheduler for 1 stream (#399)

* Enable scheduler for 1 stream

* Formatting

* Improve performance of sorting

* Formatting

* Adjust the weight calculation

* Formatting

* Simplify formula

* Formatting

* Avoid division by zero

* Fix scheduler test

* Check for either 1 or 2

* Check for waits when order may change

* Formatting

src/env.cpp

@@ -21,11 +21,11 @@ bool disabled(const char* name)
     return contains({"0", "disable", "disabled", "no", "false"}, e.front());
 }
 
-std::size_t value_of(const char* name)
+std::size_t value_of(const char* name, std::size_t fallback)
 {
     auto e = env(name);
     if(e.empty())
-        return 0;
+        return fallback;
     return std::stoul(e.front());
 }
 

src/include/migraphx/env.hpp

@@ -19,7 +19,7 @@ bool enabled(const char* name);
 bool disabled(const char* name);
 std::vector<std::string> env(const char* name);
 
-std::size_t value_of(const char* name);
+std::size_t value_of(const char* name, std::size_t fallback = 0);
 
 template <class T>
 bool enabled(T)
@@ -36,9 +36,9 @@ bool disabled(T)
 }
 
 template <class T>
-std::size_t value_of(T)
+std::size_t value_of(T, std::size_t fallback = 0)
 {
-    static const std::size_t result = value_of(T::value());
+    static const std::size_t result = value_of(T::value(), fallback);
     return result;
 }
 

src/program.cpp

@@ -384,25 +384,26 @@ argument generic_eval(const program& p,
     values.reserve(16);
     for(auto ins : iterator_for(p))
     {
-        if(ins->name() == "@literal")
+        const auto& name = ins->name();
+        if(name == "@literal")
         {
             results.emplace(ins, trace(ins, [&] { return ins->get_literal().get_argument(); }));
         }
-        else if(ins->name() == "@param")
+        else if(name == "@param")
         {
             results.emplace(
                 ins, trace(ins, [&] {
                     auto param_name = any_cast<builtin::param>(ins->get_operator()).parameter;
                     if(not contains(params, param_name))
                         MIGRAPHX_THROW("Parameter not found: " + param_name);
-                    auto param = params.at(param_name);
+                    auto param = params[param_name];
                     if(param.get_shape() != ins->get_shape())
                         MIGRAPHX_THROW("Incorrect shape {" + to_string(param.get_shape()) +
                                        "} for parameter: " + param_name);
                     return param;
                 }));
         }
-        else if(ins->name() == "@outline")
+        else if(name == "@outline")
         {
             results.emplace(ins, trace(ins, [&] { return argument{ins->get_shape(), nullptr}; }));
         }

src/schedule.cpp

@@ -9,6 +9,7 @@
 #include <migraphx/ranges.hpp>
 #include <unordered_map>
 #include <unordered_set>
+#include <queue>
 #include <thread>
 #include <mutex>
 #include <set>
@@ -18,6 +19,8 @@
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 
+MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_TRACE_SCHEDULE)
+
 auto get_inputs()
 {
     return [](auto i) { return i->inputs(); };
@@ -54,6 +57,17 @@ struct stream_info
         })(last);
     }
 
+    template <class Compare>
+    void sort_args_by_weight(std::vector<instruction_ref>& args, Compare compare) const
+    {
+        if(args.size() < 2)
+            return;
+        std::sort(args.begin(), args.end(), by(compare, [this](auto x) {
+                      return std::make_tuple(
+                          this->weights.at(x), x->inputs().size(), std::addressof(*x));
+                  }));
+    }
+
     std::vector<instruction_ref>::iterator sort_args(std::vector<instruction_ref>& args)
     {
         if(args.size() < 2)
@@ -61,11 +75,8 @@ struct stream_info
             return args.end();
         }
 
-        const std::size_t min_partition_threshold = 2;
-        auto compare                              = by(std::greater<>{}, [&](auto x) {
-            return std::make_tuple(this->weights[x], x->inputs().size());
-        });
-        std::sort(args.begin(), args.end(), compare);
+        const std::size_t min_partition_threshold = 1;
+        sort_args_by_weight(args, std::greater<>{});
 
         auto it = std::lower_bound(std::next(args.begin()),
                                    args.end(),
@@ -89,8 +100,9 @@ struct stream_info
         }
     };
 
-    void assign_streams(program& p, std::size_t n)
+    std::size_t assign_streams(program& p, std::size_t n)
     {
+        assert(n > 0);
         partition critical;
         std::unordered_map<instruction_ref, std::deque<partition>> partitions;
         partitions.reserve(weights.size());
@@ -126,19 +138,77 @@ struct stream_info
 
         // Set the critical partition to stream 0
         set_stream(critical, 0);
-        std::vector<std::size_t> streams(n - 1);
-        // Assign streams for the other partitions
-        for(auto&& ins_part : partitions)
+        if(n == 1)
+        {
+            // Assign streams for the other partitions
+            for(auto&& ins_part : partitions)
+                for(auto&& part : ins_part.second)
+                    set_stream(part, 0);
+            return 1;
+        }
+        else
         {
-            std::sort(
-                ins_part.second.begin(), ins_part.second.end(), by(std::greater<>{}, [](auto&& x) {
-                    return std::make_tuple(x.weight, x.instructions.size());
-                }));
-            for(auto&& part : ins_part.second)
+            std::vector<std::size_t> streams(n - 1);
+            // Assign streams for the other partitions
+            for(auto&& ins_part : partitions)
             {
-                auto stream = std::min_element(streams.begin(), streams.end()) - streams.begin();
-                set_stream(part, stream + 1);
-                streams[stream] += part.weight;
+                std::sort(ins_part.second.begin(),
+                          ins_part.second.end(),
+                          by(std::greater<>{}, [](auto&& x) {
+                              return std::make_tuple(x.weight, x.instructions.size());
+                          }));
+                for(auto&& part : ins_part.second)
+                {
+                    auto stream =
+                        std::min_element(streams.begin(), streams.end()) - streams.begin();
+                    set_stream(part, stream + 1);
+                    streams[stream] += part.weight;
+                }
+            }
+            return 1 + std::count_if(streams.begin(), streams.end(), [](auto x) { return x > 0; });
+        }
+    }
+
+    using weight_ins = std::pair<std::size_t, instruction_ref>;
+    struct compare_weight_ins
+    {
+        bool operator()(const weight_ins& x, const weight_ins& y) const
+        {
+            return std::make_pair(x.first, std::addressof(*x.second)) <
+                   std::make_pair(y.first, std::addressof(*y.second));
+        }
+    };
+
+    void sort(program& p, std::size_t) const
+    {
+        std::set<weight_ins, compare_weight_ins> children;
+        std::unordered_map<instruction_ref, std::size_t> visited;
+        auto last      = std::prev(p.end());
+        auto mw        = this->weights.at(last);
+        auto nw        = mw / (p.size() + 1);
+        auto add_child = [&](auto ins) {
+            auto x  = 1 + (mw - this->weights.at(ins)) / (nw + 1);
+            auto w  = x * this->iweights.at(ins);
+            auto& v = visited[ins];
+            auto it = children.find(std::make_pair(v * w, ins));
+            if(it == children.end())
+            {
+                v++;
+                children.insert(std::make_pair(v * w, ins));
+            }
+        };
+        add_child(last);
+
+        while(not children.empty())
+        {
+            // Pop the first element
+            auto top = children.begin()->second;
+            children.erase(children.begin());
+
+            p.move_instruction(top, p.begin());
+            for(auto ins : top->inputs())
+            {
+                add_child(ins);
             }
         }
     }
@@ -398,9 +468,10 @@ void schedule::apply(program& p) const
     stream_info si;
     auto last = std::prev(p.end());
     si.accumulate_weights(last, model);
-    si.assign_streams(p, model.concurrency());
+    auto nstreams = si.assign_streams(p, model.concurrency());
+    si.sort(p, model.concurrency());
 
-    if(enabled(MIGRAPHX_TRACE_COMPILE{}))
+    if(enabled(MIGRAPHX_TRACE_COMPILE{}) or enabled(MIGRAPHX_TRACE_SCHEDULE{}))
     {
         p.annotate(std::cout, [&](auto ins) {
             std::cout << ":";
@@ -417,6 +488,10 @@ void schedule::apply(program& p) const
         std::cout << std::endl;
     }
 
+    // No concurrency
+    if(nstreams < 2)
+        return;
+
     // Schedule instructions
     std::size_t wait_id = 0;
     std::unordered_map<instruction_ref, std::size_t> ins2wait;

src/targets/gpu/include/migraphx/gpu/context.hpp

@@ -12,6 +12,7 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 
 MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_ENABLE_NULL_STREAM)
+MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_NSTREAMS)
 
 using hip_event_ptr = MIGRAPHX_MANAGE_PTR(hipEvent_t, hipEventDestroy);
 
@@ -126,7 +127,7 @@ struct hip_device
 
 struct context
 {
-    context(std::size_t device_id = 0, std::size_t n = 4)
+    context(std::size_t device_id = 0, std::size_t n = value_of(MIGRAPHX_NSTREAMS{}, 1))
         : current_device(std::make_shared<hip_device>(device_id, n))
     {
     }

src/targets/gpu/schedule_model.cpp

@@ -97,11 +97,10 @@ static std::unordered_map<std::string, std::size_t> create_weight_map()
 {
     return {{"hip::load_literal", 0},
             {"hip::allocate", 0},
-            {"gpu::convolution", 4},
-            {"gpu::conv_bias_relu", 4},
-            {"gpu::pooling", 2},
-            {"gpu::gemm", 2},
-            {"gpu::concat", 1}};
+            {"gpu::convolution", 8},
+            {"gpu::conv_bias_relu", 8},
+            {"gpu::pooling", 4},
+            {"gpu::gemm", 4}};
 }
 
 static const std::unordered_map<std::string, std::size_t>& weight_map()
@@ -114,7 +113,7 @@ std::size_t schedule_model::weight(const operation& op) const
 {
     if(weight_map().count(op.name()) == 0)
     {
-        return 1;
+        return 2;
     }
     return weight_map().at(op.name());
 }

src/targets/gpu/target.cpp

@@ -30,7 +30,7 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 
-MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_ENABLE_SCHEDULE_PASS)
+MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_DISABLE_SCHEDULE_PASS)
 
 std::vector<pass> target::get_passes(migraphx::context& gctx) const
 {
@@ -70,7 +70,7 @@ std::vector<pass> target::get_passes(migraphx::context& gctx) const
         fuse_ops{&ctx},
         dead_code_elimination{},
         write_literals{&ctx},
-        schedule{gpu::schedule_model{ctx.get_current_device().nstreams()}, enabled(MIGRAPHX_ENABLE_SCHEDULE_PASS{})},
+        schedule{gpu::schedule_model{ctx.get_current_device().nstreams()}, not enabled(MIGRAPHX_DISABLE_SCHEDULE_PASS{})},
         memory_coloring{"hip::allocate"},
         dead_code_elimination{},
         eliminate_workspace{},

test/include/test.hpp

@@ -116,6 +116,7 @@ struct lhs_expression
     TEST_LHS_REOPERATOR(%)
     TEST_LHS_REOPERATOR(&)
     TEST_LHS_REOPERATOR(|)
+    TEST_LHS_REOPERATOR (^)
     TEST_LHS_REOPERATOR(&&)
     TEST_LHS_REOPERATOR(||)
 };

test/schedule_test.cpp

@@ -280,7 +280,7 @@ TEST_CASE(stream_free)
     EXPECT(not t.has_stream(one));
     EXPECT(not t.has_stream(onep1));
     EXPECT(not t.has_stream(onep2));
-    EXPECT(t.get_stream(binary) == 0);
+    EXPECT(not t.has_stream(binary));
 }
 
 TEST_CASE(zero_record)
@@ -616,8 +616,9 @@ TEST_CASE(inner_par_merge)
                                                  t.get_stream(outer1),
                                                  t.get_stream(outer2)}));
 
-    EXPECT(t.get_stream(outer1) == 1);
-    EXPECT(t.get_stream(outer2) == 2);
+    EXPECT(t.get_stream(outer1) != t.get_stream(outer2));
+    EXPECT(migraphx::contains({1, 2}, t.get_stream(outer1)));
+    EXPECT(migraphx::contains({1, 2}, t.get_stream(outer2)));
 
     EXPECT(t.get_stream(i1) != t.get_stream(i2));
     for(auto ins : c1)
@@ -704,9 +705,8 @@ TEST_CASE(inner_split1)
     EXPECT(
         get_wait_for(output) ==
         get_wait_for(t.get_stream(output), {t.get_stream(i1), t.get_stream(s1), t.get_stream(s2)}));
-    EXPECT(get_wait_for(s1).empty());
-    // TODO: Remove the extra wait here
-    // EXPECT(get_wait_for(s2).empty());
+    // Either s1 or s2 has a wait depending on the sort order but not both
+    EXPECT(get_wait_for(s1).empty() xor get_wait_for(s2).empty());
     t.check_conflicts(p, {c1, {i1}, {s1}, {s2}});
 }