fixed sizing issues for radius cache

GrigoryGraborenko · Dec 22, 2024 · f853595 · f853595
1 parent 26a3294
commit f853595
Show file tree

Hide file tree

Showing 4 changed files with 33 additions and 10 deletions.
diff --git a/Src/Shaders/GridCache3D.wgsl b/Src/Shaders/GridCache3D.wgsl
@@ -9,6 +9,7 @@ fn ItemMain(item_index: i32, pos: vec3f) {
     let index: i32 = (grid_pos.x + (grid_pos.y + grid_pos.z * Uniform.GridSize.y) * Uniform.GridSize.x) * 3;
 
 #ifdef PHASE_INDEX
+    atomicAdd(&TotalSize, 1);
     atomicAdd(&b_Index[index], 1);
     atomicCompareExchangeWeak(&b_Index[index + 1], 0, -1);
 #elifdef PHASE_ALLOCATE
@@ -48,6 +49,7 @@ fn ItemRadiusMain(item_index: i32, pos: vec3f, radius: f32) {
                 let grid_pos = vec3i(x, y, z);
                 let index : i32 = (grid_pos.x + (grid_pos.y + grid_pos.z * Uniform.GridSize.y) * Uniform.GridSize.x) * 3;
 #ifdef PHASE_INDEX
+                atomicAdd(&TotalSize, 1);
                 atomicAdd(&b_Index[index], 1);
                 atomicCompareExchangeWeak(&b_Index[index + 1], 0, -1);
 #elifdef PHASE_ALLOCATE

diff --git a/Src/UnitTests/EntityTest.cpp b/Src/UnitTests/EntityTest.cpp
@@ -541,7 +541,7 @@ namespace Test {
 	////////////////////////////////////////////////////////////////////////////////
 	void GPUEntityCache3D(bool use_cursor, bool use_radius = false) {
 
-		const int prey_count = 50;
+		const int prey_count = 2000;
 		const int hunter_count = 20;
 
 		const float map_radius = 50;
@@ -560,7 +560,7 @@ namespace Test {
 		Neshny::GPUEntity hunter_entities("Hunter", &GPUOther::p_Id, "Id");
 #endif
 
-		prey_entities.Init(1000);
+		prey_entities.Init(prey_count + 1);
 		hunter_entities.Init(1000);
 
 		Neshny::Grid3DCache cache(prey_entities, "ThreeDim", use_radius ? std::optional<std::string_view>("Float") : std::nullopt);

diff --git a/Src/WebGPU/PipelineWebGPU.cpp b/Src/WebGPU/PipelineWebGPU.cpp
@@ -836,8 +836,9 @@ void Grid3DCache::GenerateCache(iVec3 grid_size, Vec3 grid_min, Vec3 grid_max) {
 	m_GridMin = grid_min;
 	m_GridMax = grid_max;
 
+	m_RequiredItemCacheSize = std::max(m_RequiredItemCacheSize, m_Entity.GetMaxCount() * (m_RadiusName.has_value() ? 4 : 1)); // quadruple entity count to begin with for inefficient use of space
 	m_GridIndices.EnsureSizeBytes(grid_size.x * grid_size.y * grid_size.z * 3 * sizeof(int), true);
-	m_GridItems.EnsureSizeBytes(m_Entity.GetMaxCount() * sizeof(int), false);
+	m_GridItems.EnsureSizeBytes(m_RequiredItemCacheSize * sizeof(int), false);
 
 	std::string func_call = m_RadiusName.has_value() ? std::format("ItemRadiusMain(item_index, item.{}, item.{})", m_PosName, *m_RadiusName) : std::format("ItemMain(item_index, item.{})", m_PosName);
 	std::string main_func = std::format("fn {0}Main(item_index: i32, item: {0}) {{ {1}; }}", m_Entity.GetName(), func_call);
@@ -849,12 +850,31 @@ void Grid3DCache::GenerateCache(iVec3 grid_size, Vec3 grid_min, Vec3 grid_max) {
 		fVec4(grid_max.x, grid_max.y, grid_max.z, 0)
 	};
 
-	Neshny::EntityPipeline::IterateEntity(std::format("{}:INDEX", base_id), m_Entity, "GridCache3D", true)
-		.AddBuffer("b_Index", m_GridIndices, MemberSpec::T_INT, EntityPipeline::BufferAccess::READ_WRITE_ATOMIC)
-		.AddCode("#define PHASE_INDEX")
-		.AddCode(main_func)
-		.SetUniform(uniform)
-		.Run();
+	while (true) {
+		auto async = Neshny::EntityPipeline::IterateEntity(std::format("{}:INDEX", base_id), m_Entity, "GridCache3D", true)
+			.AddBuffer("b_Index", m_GridIndices, MemberSpec::T_INT, EntityPipeline::BufferAccess::READ_WRITE_ATOMIC)
+			.AddCode("#define PHASE_INDEX")
+			.AddCode(main_func)
+			.AddInputOutputVar("TotalSize", 0)
+			.SetUniform(uniform)
+			.Run([this] (const Neshny::EntityPipeline::OutputResults& results) {
+				int total_size;
+				results.GetValue("TotalSize", total_size);
+				if (m_RadiusName.has_value()) {
+					total_size *= 2; // twice the required size to allow flexibility in future frames
+				}
+				m_RequiredItemCacheSize = std::max(total_size, m_RequiredItemCacheSize);
+			});
+		if (m_InitialRun) {
+			m_InitialRun = false;
+			if (m_RadiusName.has_value()) {
+				async.Wait();
+				m_GridItems.EnsureSizeBytes(m_RequiredItemCacheSize * sizeof(int), false);
+			}
+			continue;
+		}
+		break;
+	}
 
 	for (int i = 0; i < 2; i++) {
 		Neshny::EntityPipeline::IterateEntity(std::format("{}:ALLOC", base_id), m_Entity, "GridCache3D", true)
@@ -887,7 +907,6 @@ void Grid3DCache::Bind(EntityPipeline& target_stage, bool initial_creation) {
 	target_stage.AddBuffer(std::format("b_{0}GridIndices", name), m_GridIndices, MemberSpec::Type::T_INT, EntityPipeline::BufferAccess::READ_ONLY);
 	target_stage.AddBuffer(std::format("b_{0}GridItems", name), m_GridItems, MemberSpec::Type::T_INT, EntityPipeline::BufferAccess::READ_ONLY);
 
-	target_stage.AddStructBuffer<Grid3DCacheUniform>(std::format("b_{0}GridUniform", name), std::format("{0}GridUniformStruct", name), m_Uniform, EntityPipeline::BufferAccess::READ_ONLY, false);
 	target_stage.AddStructBuffer<GridCacheUniform>(std::format("b_{0}GridUniform", name), std::format("{0}GridUniformStruct", name), m_Uniform, EntityPipeline::BufferAccess::READ_ONLY, false);
 
 	if (!initial_creation) {

diff --git a/Src/WebGPU/PipelineWebGPU.h b/Src/WebGPU/PipelineWebGPU.h
@@ -300,6 +300,8 @@ class Grid3DCache : public BaseCache {
 	iVec3						m_GridSize;
 	Vec3						m_GridMin;
 	Vec3						m_GridMax;
+	int							m_RequiredItemCacheSize = 0;
+	bool						m_InitialRun = true;
 };
 
 } // namespace Neshny