maxwell_3d: Slow implementation of passed samples (query 21)

Implements GL_SAMPLES_PASSED by waiting immediately for queries.

2 files changed, 60 insertions, 17 deletions

diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 0b3e8749b..fe91ff6a0 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -400,6 +400,10 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
         ProcessQueryCondition();
         break;
     }
+    case MAXWELL3D_REG_INDEX(counter_reset): {
+        ProcessCounterReset();
+        break;
+    }
     case MAXWELL3D_REG_INDEX(sync_info): {
         ProcessSyncPoint();
         break;
@@ -544,23 +548,23 @@ void Maxwell3D::ProcessQueryGet() {
                "Units other than CROP are unimplemented");
 
     switch (regs.query.query_get.operation) {
-    case Regs::QueryOperation::Release: {
-        const u64 result = regs.query.query_sequence;
-        StampQueryResult(result, regs.query.query_get.short_query == 0);
+    case Regs::QueryOperation::Release:
+        StampQueryResult(regs.query.query_sequence, regs.query.query_get.short_query == 0);
         break;
-    }
-    case Regs::QueryOperation::Acquire: {
-        // Todo(Blinkhawk): Under this operation, the GPU waits for the CPU
-        // to write a value that matches the current payload.
+    case Regs::QueryOperation::Acquire:
+        // TODO(Blinkhawk): Under this operation, the GPU waits for the CPU to write a value that
+        // matches the current payload.
         UNIMPLEMENTED_MSG("Unimplemented query operation ACQUIRE");
         break;
-    }
     case Regs::QueryOperation::Counter: {
-        u64 result{};
+        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 {}",
@@ -569,15 +573,13 @@ void Maxwell3D::ProcessQueryGet() {
         StampQueryResult(result, regs.query.query_get.short_query == 0);
         break;
     }
-    case Regs::QueryOperation::Trap: {
+    case Regs::QueryOperation::Trap:
         UNIMPLEMENTED_MSG("Unimplemented query operation TRAP");
         break;
-    }
-    default: {
+    default:
         UNIMPLEMENTED_MSG("Unknown query operation");
         break;
     }
-    }
 }
 
 void Maxwell3D::ProcessQueryCondition() {
@@ -619,6 +621,17 @@ void Maxwell3D::ProcessQueryCondition() {
     }
 }
 
+void Maxwell3D::ProcessCounterReset() {
+    switch (regs.counter_reset) {
+    case Regs::CounterReset::SampleCnt:
+        rasterizer.ResetCounter(VideoCore::QueryType::SamplesPassed);
+        break;
+    default:
+        UNIMPLEMENTED_MSG("counter_reset={}", static_cast<u32>(regs.counter_reset));
+        break;
+    }
+}
+
 void Maxwell3D::ProcessSyncPoint() {
     const u32 sync_point = regs.sync_info.sync_point.Value();
     const u32 increment = regs.sync_info.increment.Value();
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index 0a2af54e5..d21f678ed 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -409,6 +409,27 @@ public:
             Linear = 1,
         };
 
+        enum class CounterReset : u32 {
+            SampleCnt = 0x01,
+            Unk02 = 0x02,
+            Unk03 = 0x03,
+            Unk04 = 0x04,
+            EmittedPrimitives = 0x10, // Not tested
+            Unk11 = 0x11,
+            Unk12 = 0x12,
+            Unk13 = 0x13,
+            Unk15 = 0x15,
+            Unk16 = 0x16,
+            Unk17 = 0x17,
+            Unk18 = 0x18,
+            Unk1A = 0x1A,
+            Unk1B = 0x1B,
+            Unk1C = 0x1C,
+            Unk1D = 0x1D,
+            Unk1E = 0x1E,
+            GeneratedPrimitives = 0x1F,
+        };
+
         struct Cull {
             enum class FrontFace : u32 {
                 ClockWise = 0x0900,
@@ -857,7 +878,7 @@ public:
                     BitField<7, 1, u32> c7;
                 } clip_distance_enabled;
 
-                INSERT_UNION_PADDING_WORDS(0x1);
+                u32 samplecnt_enable;
 
                 float point_size;
 
@@ -865,7 +886,11 @@ public:
 
                 u32 point_sprite_enable;
 
-                INSERT_UNION_PADDING_WORDS(0x5);
+                INSERT_UNION_PADDING_WORDS(0x3);
+
+                CounterReset counter_reset;
+
+                INSERT_UNION_PADDING_WORDS(0x1);
 
                 u32 zeta_enable;
 
@@ -1412,12 +1437,15 @@ private:
     /// Handles a write to the QUERY_GET register.
     void ProcessQueryGet();
 
-    // Writes the query result accordingly
+    /// Writes the query result accordingly.
     void StampQueryResult(u64 payload, bool long_query);
 
-    // Handles Conditional Rendering
+    /// Handles conditional rendering.
     void ProcessQueryCondition();
 
+    /// Handles counter resets.
+    void ProcessCounterReset();
+
     /// Handles writes to syncing register.
     void ProcessSyncPoint();
 
@@ -1499,8 +1527,10 @@ ASSERT_REG_POSITION(screen_y_control, 0x4EB);
 ASSERT_REG_POSITION(vb_element_base, 0x50D);
 ASSERT_REG_POSITION(vb_base_instance, 0x50E);
 ASSERT_REG_POSITION(clip_distance_enabled, 0x544);
+ASSERT_REG_POSITION(samplecnt_enable, 0x545);
 ASSERT_REG_POSITION(point_size, 0x546);
 ASSERT_REG_POSITION(point_sprite_enable, 0x548);
+ASSERT_REG_POSITION(counter_reset, 0x54C);
 ASSERT_REG_POSITION(zeta_enable, 0x54E);
 ASSERT_REG_POSITION(multisample_control, 0x54F);
 ASSERT_REG_POSITION(condition, 0x554);