From 2539fb7c3c9b8c26f4f942571371c6eb17a97589 Mon Sep 17 00:00:00 2001
From: "Unknown W. Brackets" <checkins@unknownbrackets.org>
Date: Sat, 15 Jan 2022 18:43:44 -0800
Subject: [PATCH 1/2] softgpu: Tune queue push/pop to reduce overhead.

These aren't safetly atomic with concurrent pushers or poppers, but as
long as there's only one of each, they're still safe.

Shaves a decent % off Drain time for heavy scenes.
---
 GPU/Software/BinManager.cpp | 19 ++++++++++++-------
 GPU/Software/BinManager.h   | 28 ++++++++++++++++++++++++++++
 2 files changed, 40 insertions(+), 7 deletions(-)

diff --git a/GPU/Software/BinManager.cpp b/GPU/Software/BinManager.cpp
index d85708c541..993bf8712e 100644
--- a/GPU/Software/BinManager.cpp
+++ b/GPU/Software/BinManager.cpp
@@ -105,8 +105,9 @@ public:
 private:
 	void ProcessItems() {
 		while (!items_.Empty()) {
-			const BinItem item = items_.Pop();
+			const BinItem &item = items_.PeekNext();
 			DrawBinItem(item, states_[item.stateIndex]);
+			items_.SkipNext();
 		}
 	}
 
@@ -261,12 +262,13 @@ void BinManager::Drain() {
 
 	if (taskRanges_.size() <= 1) {
 		while (!queue_.Empty()) {
-			const BinItem item = queue_.Pop();
+			const BinItem &item = queue_.PeekNext();
 			DrawBinItem(item, states_[item.stateIndex]);
+			queue_.SkipNext();
 		}
 	} else {
 		while (!queue_.Empty()) {
-			const BinItem item = queue_.Pop();
+			const BinItem &item = queue_.PeekNext();
 			for (int i = 0; i < (int)taskRanges_.size(); ++i) {
 				const BinCoords range = taskRanges_[i].Intersect(item.range);
 				if (range.Invalid())
@@ -276,10 +278,13 @@ void BinManager::Drain() {
 				if (taskQueues_[i].Full())
 					waitable_->Wait();
 
-				BinItem subitem = item;
-				subitem.range = range;
-				taskQueues_[i].Push(subitem);
+				BinItem &taskItem = taskQueues_[i].PeekPush();
+				taskItem = item;
+				taskItem.range = range;
+				taskQueues_[i].PushPeeked();
+
 			}
+			queue_.SkipNext();
 		}
 
 		for (int i = 0; i < (int)taskRanges_.size(); ++i) {
@@ -302,7 +307,7 @@ void BinManager::Flush() {
 
 	queue_.Reset();
 	while (states_.Size() > 1)
-		states_.Pop();
+		states_.SkipNext();
 
 	queueRange_.x1 = 0x7FFFFFFF;
 	queueRange_.y1 = 0x7FFFFFFF;
diff --git a/GPU/Software/BinManager.h b/GPU/Software/BinManager.h
index b25921cf8f..0848ed8d5c 100644
--- a/GPU/Software/BinManager.h
+++ b/GPU/Software/BinManager.h
@@ -89,6 +89,34 @@ struct BinQueue {
 		return item;
 	}
 
+	// Only safe if you're the only one reading.
+	T &PeekNext() {
+		_dbg_assert_(!Empty());
+		return items_[head_];
+	}
+
+	void SkipNext() {
+		_dbg_assert_(!Empty());
+		size_t i = head_++;
+		if (i + 1 == N)
+			head_ -= N;
+		size_--;
+	}
+
+	// Only safe if you're the only one writing.
+	T &PeekPush() {
+		_dbg_assert_(size_ < N);
+		return items_[tail_];
+	}
+
+	void PushPeeked() {
+		_dbg_assert_(size_ < N);
+		size_t i = tail_++;
+		if (i + 1 == N)
+			tail_ -= N;
+		size_++;
+	}
+
 	size_t Size() const {
 		return size_;
 	}

From b42ebe15d8d3eddd78ab990b22ad27c3496e9d22 Mon Sep 17 00:00:00 2001
From: "Unknown W. Brackets" <checkins@unknownbrackets.org>
Date: Sat, 15 Jan 2022 21:59:23 -0800
Subject: [PATCH 2/2] softgpu: Fix off-by-one size limit on bin queues.

---
 GPU/Software/BinManager.h | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/GPU/Software/BinManager.h b/GPU/Software/BinManager.h
index 0848ed8d5c..4d32ea0335 100644
--- a/GPU/Software/BinManager.h
+++ b/GPU/Software/BinManager.h
@@ -70,7 +70,7 @@ struct BinQueue {
 	}
 
 	size_t Push(const T &item) {
-		_dbg_assert_(size_ < N);
+		_dbg_assert_(size_ < N - 1);
 		size_t i = tail_++;
 		if (i + 1 == N)
 			tail_ -= N;
@@ -105,12 +105,12 @@ struct BinQueue {
 
 	// Only safe if you're the only one writing.
 	T &PeekPush() {
-		_dbg_assert_(size_ < N);
+		_dbg_assert_(size_ < N - 1);
 		return items_[tail_];
 	}
 
 	void PushPeeked() {
-		_dbg_assert_(size_ < N);
+		_dbg_assert_(size_ < N - 1);
 		size_t i = tail_++;
 		if (i + 1 == N)
 			tail_ -= N;
@@ -122,7 +122,7 @@ struct BinQueue {
 	}
 
 	bool Full() const {
-		return size_ == N;
+		return size_ == N - 1;
 	}
 
 	bool Empty() const {