Make PCF filtering configurable, improve shader performance

data/graphics_presets.cfg

@@ -1,11 +1,14 @@
 [Low]
 shadowResolution=128
 dofDownscale=8
+filterPCF=0
 
 [Medium]
 shadowResolution=256
 dofDownscale=4
+filterPCF=1
 
 [High]
 shadowResolution=512
 dofDownscale=2
+filterPCF=1

include/renderer.h

@@ -70,7 +70,7 @@ struct RenderTarget {
 
 struct GraphicsConfig {
     int shadowResolution, dofDownscale;
-    bool vsync = true;
+    bool vsync = true, filterPCF;
 };
 
 class World;

shaders/mesh.frag

@@ -8,7 +8,7 @@ layout(location = 3) in vec2 inUV;
 layout(location = 0) out vec4 outColor;
 
 struct Light {
-    vec4 position, direction;
+    vec4 position;
     vec3 color;
 };
 
@@ -19,6 +19,8 @@ layout(set = 0, binding = 0) uniform Lights {
 layout(set = 0, binding = 1) uniform sampler2D shadowSampler;
 layout(set = 1, binding = 0) uniform sampler2D albedoSampler;
 
+layout(constant_id = 0) const int shadowFilterPCF = 1;
+
 float textureProj(vec4 P, vec2 off) {
     const vec4 shadowCoord = P / P.w; // perspective divide
 
@@ -54,16 +56,19 @@ float filterPCF(vec4 sc) {
 void main() {
     vec3 diffuse = vec3(0);
     for(int i = 0; i < 32; i++) {
-        const vec3 norm = normalize(inNormal);
-
         vec3 lightDir = vec3(0);
         if(lights[i].position.w == 0)
+            continue;
+        else if(lights[i].position.w == 1)
             lightDir = normalize(lights[i].position.xyz - inFragPos);
-        else
-            lightDir = normalize(-lights[i].direction.xyz);
+        else if(lights[i].position.w == 2)
+            lightDir = normalize(-lights[i].position.xyz);
+
+        const vec3 norm = normalize(inNormal);
 
         const float diff = max(dot(norm, lightDir), 0.0);
-        diffuse += vec3(diff) * lights[i].color * filterPCF(inShadowPos / inShadowPos.w);
+        const float shadow = shadowFilterPCF == 1 ? filterPCF(inShadowPos / inShadowPos.w) : textureProj(inShadowPos / inShadowPos.w, vec2(0, 0));
+        diffuse += vec3(diff) * lights[i].color * shadow;
     }
 
     outColor = vec4(vec3(0.1) + diffuse * texture(albedoSampler, inUV).rgb, 1.0);

src/main.cpp

@@ -61,6 +61,7 @@ void loadGraphicsConfig() {
 
     graphicsConfig.shadowResolution = toInt(config.get(currentGraphicsPreset.c_str(), "shadowResolution"));
     graphicsConfig.dofDownscale = toInt(config.get(currentGraphicsPreset.c_str(), "dofDownscale"));
+    graphicsConfig.filterPCF = toInt(config.get(currentGraphicsPreset.c_str(), "filterPCF"));
 }
 
 void readConfig() {

src/worldpass.cpp

@@ -32,16 +32,21 @@ WorldPass::~WorldPass() {
 
 void WorldPass::render(VkCommandBuffer commandBuffer, World& world, Camera& camera, RenderTarget* target) {
     struct ShaderLight {
-        glm::vec4 position, direction;
+        glm::vec4 position;
         glm::vec3 color;
     };
 
     ShaderLight* data;
-    vkMapMemory(renderer_.getDevice(), lightMemory_, 0, sizeof(float) * (4 + 4 + 3) * 32, 0, reinterpret_cast<void**>(&data));
+    vkMapMemory(renderer_.getDevice(), lightMemory_, 0, sizeof(float) * (4 + 3) * 32, 0, reinterpret_cast<void**>(&data));
 
     for(const auto& light : world.lights) {
-        data->position = glm::vec4(light->position, (int)light->type);
-        data->direction = glm::vec4(glm::normalize(glm::vec3(0) - light->position), 0.0);
+        glm::vec3 position;
+        if(light->type == LightType::Point)
+            position = light->position;
+        else
+            position = glm::normalize(glm::vec3(0) - light->position);
+
+        data->position = glm::vec4(position, ((int)light->type) + 1);
         data->color = light->color;
 
         data++;
@@ -174,11 +179,23 @@ void WorldPass::createPipeline() {
     vertShaderStageInfo.module = vertShaderModule;
     vertShaderStageInfo.pName = "main";
 
+    VkSpecializationMapEntry mapEntry = {};
+    mapEntry.size = sizeof(int);
+
+    VkSpecializationInfo specializationInfo = {};
+    specializationInfo.mapEntryCount = 1;
+    specializationInfo.pMapEntries = &mapEntry;
+    specializationInfo.dataSize = sizeof(int);
+
+    int shadowFilterPCF = renderer_.getConfig().filterPCF;
+    specializationInfo.pData = &shadowFilterPCF;
+
     VkPipelineShaderStageCreateInfo fragShaderStageInfo = {};
     fragShaderStageInfo.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
     fragShaderStageInfo.stage = VK_SHADER_STAGE_FRAGMENT_BIT;
     fragShaderStageInfo.module = fragShaderModule;
     fragShaderStageInfo.pName = "main";
+    fragShaderStageInfo.pSpecializationInfo = &specializationInfo;
 
     const std::array<VkPipelineShaderStageCreateInfo, 2> shaderStages = {vertShaderStageInfo, fragShaderStageInfo};
 
@@ -298,7 +315,7 @@ void WorldPass::createPipeline() {
 void WorldPass::createUniformBuffer() {
     VkBufferCreateInfo bufferInfo = {};
     bufferInfo.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
-    bufferInfo.size = sizeof(float) * (4 + 4 + 3) * 32;
+    bufferInfo.size = sizeof(float) * (4 + 3) * 32;
     bufferInfo.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT;
     bufferInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
 
@@ -327,7 +344,7 @@ void WorldPass::createDescriptorSet() {
 
     VkDescriptorBufferInfo bufferInfo = {};
     bufferInfo.buffer = lightBuffer_;
-    bufferInfo.range = sizeof(float) * (4 + 4 + 3) * 32;
+    bufferInfo.range = sizeof(float) * (4 + 3) * 32;
 
     VkDescriptorImageInfo imageInfo = {};
     imageInfo.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
@@ -357,9 +374,9 @@ void WorldPass::createDescriptorSet() {
     vkUpdateDescriptorSets(renderer_.getDevice(), descriptorWrites.size(), descriptorWrites.data(), 0, nullptr);
 
     float* data;
-    vkMapMemory(renderer_.getDevice(), lightMemory_, 0, sizeof(float) * (4 + 4 + 3) * 32, 0, reinterpret_cast<void**>(&data));
+    vkMapMemory(renderer_.getDevice(), lightMemory_, 0, sizeof(float) * (4 + 3) * 32, 0, reinterpret_cast<void**>(&data));
 
-    for(uint32_t i = 0; i < (4 + 4 + 3) * 32; i++)
+    for(uint32_t i = 0; i < (4 + 3) * 32; i++)
         data[i] = 0.0f;
 
     vkUnmapMemory(renderer_.getDevice(), lightMemory_);