gl_query_cache: Implement host queries using a deferred cache

Instead of waiting immediately for executed commands, defer the query
until the guest CPU reads it. This way we get closer to what the guest
program is doing.

To archive this we have to build a dependency queue, because host APIs
(like OpenGL and Vulkan) use ranged queries instead of counters like
NVN.

Waiting for queries implicitly uses fences and this requires a command
being queued, otherwise the driver will lock waiting until a timeout. To
fix this when there are no commands queued, we explicitly call glFlush.

src/video_core/engines/maxwell_3d.cpp

@@ -556,23 +556,13 @@ void Maxwell3D::ProcessQueryGet() {
         // matches the current payload.
         UNIMPLEMENTED_MSG("Unimplemented query operation ACQUIRE");
         break;
-    case Regs::QueryOperation::Counter: {
-        u64 result;
-        switch (regs.query.query_get.select) {
-        case Regs::QuerySelect::Zero:
-            result = 0;
-            break;
-        case Regs::QuerySelect::SamplesPassed:
-            result = rasterizer.Query(VideoCore::QueryType::SamplesPassed);
-            break;
-        default:
-            result = 1;
-            UNIMPLEMENTED_MSG("Unimplemented query select type {}",
-                              static_cast<u32>(regs.query.query_get.select.Value()));
+    case Regs::QueryOperation::Counter:
+        if (const std::optional<u64> result = GetQueryResult()) {
+            // If the query returns an empty optional it means it's cached and deferred.
+            // In this case we have a non-empty result, so we stamp it immediately.
+            StampQueryResult(*result, regs.query.query_get.short_query == 0);
         }
-        StampQueryResult(result, regs.query.query_get.short_query == 0);
         break;
-    }
     case Regs::QueryOperation::Trap:
         UNIMPLEMENTED_MSG("Unimplemented query operation TRAP");
         break;
@@ -595,20 +585,20 @@ void Maxwell3D::ProcessQueryCondition() {
     }
     case Regs::ConditionMode::ResNonZero: {
         Regs::QueryCompare cmp;
-        memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
+        memory_manager.ReadBlock(condition_address, &cmp, sizeof(cmp));
         execute_on = cmp.initial_sequence != 0U && cmp.initial_mode != 0U;
         break;
     }
     case Regs::ConditionMode::Equal: {
         Regs::QueryCompare cmp;
-        memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
+        memory_manager.ReadBlock(condition_address, &cmp, sizeof(cmp));
         execute_on =
             cmp.initial_sequence == cmp.current_sequence && cmp.initial_mode == cmp.current_mode;
         break;
     }
     case Regs::ConditionMode::NotEqual: {
         Regs::QueryCompare cmp;
-        memory_manager.ReadBlockUnsafe(condition_address, &cmp, sizeof(cmp));
+        memory_manager.ReadBlock(condition_address, &cmp, sizeof(cmp));
         execute_on =
             cmp.initial_sequence != cmp.current_sequence || cmp.initial_mode != cmp.current_mode;
         break;
@@ -674,6 +664,21 @@ void Maxwell3D::DrawArrays() {
     }
 }
 
+std::optional<u64> Maxwell3D::GetQueryResult() {
+    switch (regs.query.query_get.select) {
+    case Regs::QuerySelect::Zero:
+        return 0;
+    case Regs::QuerySelect::SamplesPassed:
+        // Deferred.
+        rasterizer.Query(regs.query.QueryAddress(), VideoCore::QueryType::SamplesPassed);
+        return {};
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented query select type {}",
+                          static_cast<u32>(regs.query.query_get.select.Value()));
+        return 1;
+    }
+}
+
 void Maxwell3D::ProcessCBBind(std::size_t stage_index) {
     // Bind the buffer currently in CB_ADDRESS to the specified index in the desired shader stage.
     auto& shader = state.shader_stages[stage_index];

src/video_core/engines/maxwell_3d.h

@@ -6,6 +6,7 @@
 
 #include <array>
 #include <bitset>
+#include <optional>
 #include <type_traits>
 #include <unordered_map>
 #include <vector>
@@ -1462,6 +1463,9 @@ private:
 
     // Handles a instance drawcall from MME
     void StepInstance(MMEDrawMode expected_mode, u32 count);
+
+    /// Returns a query's value or an empty object if the value will be deferred through a cache.
+    std::optional<u64> GetQueryResult();
 };
 
 #define ASSERT_REG_POSITION(field_name, position)                                                  \