gpu: optimize resourceCache

By keeping all the information in a single map, we avoid multiple
lookups and can switch between frames more easily.

Before ~35us, after ~20us for adding 50 new+old and switching
the frame.

Signed-off-by: Egon Elbre <egonelbre@gmail.com>

gpu/caches.go

@@ -9,8 +9,12 @@ import (
 )
 
 type resourceCache struct {
-	res    map[interface{}]resource
-	newRes map[interface{}]resource
+	res map[interface{}]resourceCacheValue
+}
+
+type resourceCacheValue struct {
+	used     bool
+	resource resource
 }
 
 // opCache is like a resourceCache but using concrete types and a
@@ -35,46 +39,48 @@ type opCacheValue struct {
 
 func newResourceCache() *resourceCache {
 	return &resourceCache{
-		res:    make(map[interface{}]resource),
-		newRes: make(map[interface{}]resource),
+		res: make(map[interface{}]resourceCacheValue),
 	}
 }
 
 func (r *resourceCache) get(key interface{}) (resource, bool) {
 	v, exists := r.res[key]
-	if exists {
-		r.newRes[key] = v
+	if !exists {
+		return nil, false
 	}
-	return v, exists
+	if !v.used {
+		v.used = true
+		r.res[key] = v
+	}
+	return v.resource, exists
 }
 
 func (r *resourceCache) put(key interface{}, val resource) {
-	if _, exists := r.newRes[key]; exists {
+	v, exists := r.res[key]
+	if exists && v.used {
 		panic(fmt.Errorf("key exists, %p", key))
 	}
-	r.res[key] = val
-	r.newRes[key] = val
+	v.used = true
+	v.resource = val
+	r.res[key] = v
 }
 
 func (r *resourceCache) frame() {
 	for k, v := range r.res {
-		if _, exists := r.newRes[k]; !exists {
+		if v.used {
+			v.used = false
+			r.res[k] = v
+		} else {
 			delete(r.res, k)
-			v.release()
+			v.resource.release()
 		}
 	}
-	for k, v := range r.newRes {
-		delete(r.newRes, k)
-		r.res[k] = v
-	}
 }
 
 func (r *resourceCache) release() {
-	r.frame()
 	for _, v := range r.res {
-		v.release()
+		v.resource.release()
 	}
-	r.newRes = nil
 	r.res = nil
 }
 

gpu/caches_test.go

@@ -0,0 +1,24 @@
+// SPDX-License-Identifier: Unlicense OR MIT
+
+package gpu
+
+import "testing"
+
+func BenchmarkResourceCache(b *testing.B) {
+	offset := 0
+	const N = 100
+
+	cache := newResourceCache()
+	for i := 0; i < b.N; i++ {
+		// half are the same and half updated
+		for k := 0; k < N; k++ {
+			cache.put(offset+k, nullResource{})
+		}
+		cache.frame()
+		offset += N / 2
+	}
+}
+
+type nullResource struct{}
+
+func (nullResource) release() {}