working compute shaders!

include/high_perf/gl_util.h

@@ -13,6 +13,25 @@
 #include <string>
 #include <unordered_map>
 
+/**
+ * This part was made for this assignment and will likely be used in future projects
+ */
+class compute_shader {
+    private:
+        GLuint shaderID = 0;
+        GLuint programID = 0;
+    public:
+        explicit compute_shader(const std::string& shader_source, bool loadAsString = true);
+        inline void bind() const {
+            glUseProgram(programID);
+        }
+        inline void execute(int x, int y, int z) const {
+            bind();
+            glDispatchCompute(x, y, z);
+        }
+        ~compute_shader();
+};
+
 /**
  * Note: This is taken from my final project,
  * https://github.com/Tri11Paragon/COSC-3P98-Final-Project/blob/main/include/render/gl.h
@@ -24,16 +43,16 @@ class shader {
             GLint i = -1;
         };
         // we can have shaders of many types in OpenGL
-        unsigned int programID = 0;
+        GLuint programID = 0;
         // but we will only make use of these two for now
-        unsigned int vertexShaderID = 0;
+        GLuint vertexShaderID = 0;
-        unsigned int fragmentShaderID = 0;
+        GLuint fragmentShaderID = 0;
         // while these will remain unused. (Webgl2 apparently doesn't support them despite being based on GL4.3? that's a TODO!)
-        unsigned int geometryShaderID = 0;
+        GLuint geometryShaderID = 0;
         // this would be very useful however it is highly unlikely webgl will support it
         // im leaving some of this stuff in here because I might expand the native application to use some of it.
         // im trying to keep the web and native versions the same though
-        unsigned int tessellationShaderID = 0;
+        GLuint tessellationShaderID = 0;
         std::unordered_map<std::string, IntDefaultedToMinusOne> uniformVars;
         
         static unsigned int createShader(const std::string& source, int type);
@@ -47,18 +66,6 @@ class shader {
             uniformVars[name].i = loc;
             return loc;
         }
-        
-        static inline std::string removeEmptyFirstLines(const std::string& string){
-            auto lines = blt::string::split(string, "\n");
-            std::string new_source_string;
-            for (const auto& line : lines) {
-                if (!line.empty() && !blt::string::contains(line, "\"")) {
-                    new_source_string += line;
-                    new_source_string += "\n";
-                }
-            }
-            return new_source_string;
-        }
     
     public:
         /**
@@ -120,12 +127,6 @@ class shader {
             glUseProgram(programID);
         }
         
-        static void updateProjectionMatrix(const blt::mat4x4& projectionMatrix);
-        static void updateOrthographicMatrix(const blt::mat4x4& orthoMatrix);
-        static void updateViewMatrix(const blt::mat4x4& viewMatrix);
-        // returns the perspective view matrix which is calculated per frame. (This is for optimization)
-        static const blt::mat4x4& getPVM();
-        
         ~shader();
 };
 

include/modes/high_perf.h

@@ -29,6 +29,10 @@ void init();
 
 void cleanup();
 
+void runPhysicsShader();
+
+void beginExecution();
+
 #endif
 
 #endif //ASSIGN3_HIGH_PERF_H

include/shaders/fragment.frag

@@ -2,13 +2,21 @@
 #include <string>
 std::string shader_frag = R"("
 #version 460
+precision mediump float;
 
 in vec2 uv_;
+in float index;
 
 out vec4 out_color;
 
+uniform mediump sampler2DArray texture_array;
+
 void main() {
-    out_color = vec4(uv_, 0.0, 1.0);
+    //out_color = vec4(uv_, 0.0, 1.0);
+    out_color = texture(texture_array, vec3(uv_, index));
+
+    if (out_color.a < 0.1)
+        discard;
 }
 
 ")";

include/shaders/physics.comp

@@ -0,0 +1,77 @@
+#ifdef __cplusplus
+#include <string>
+std::string shader_physics = R"("
+#version 460
+
+// execute 1 shader per particle. 128 executions per group
+layout (local_size_x = 128, local_size_y = 1, local_size_z = 1) in;
+
+// due to byte alignment, storing pos and dir as vec4 was required anyways.
+struct particle_t {
+    vec4 pos;
+    vec4 dir;
+};
+
+const int offset_size = 8192;
+const float SPEED = 1.0f;
+const float SPEED_FACTOR = 25.0f;
+const float BOUNCE_FACTOR = 0.75f;
+const float SPREAD = 4.5f;
+const float particle_lifetime = 10.0f;
+const vec2 p_min = vec2(-50, -50);
+const vec2 p_max = vec2(50, 50);
+const vec3 inital_pos = vec3(0.0f, 1.0f, 0.0f);
+const vec4 inital_dir = vec4(0.0f, 1.0f, 0.0f, 0.0f);
+const vec4 GRAVITY = vec4(0.0, -9.8, 0.0, 0.0);
+
+
+layout (std430, binding=0) buffer particle_buffer {
+    particle_t particles[];
+};
+
+layout (std430, binding=1) buffer offset_buffer {
+    vec4 offsets[offset_size];
+};
+
+layout(location=0) uniform float deltaSeconds;
+
+void resetParticle(uint i, float w) {
+    particles[i].dir = inital_dir * SPEED_FACTOR + offsets[i % offset_size] * SPREAD;
+    particles[i].pos = vec4(inital_pos, w);
+}
+
+bool checkBounds(vec2 pos) {
+    return pos.x > p_min.x && pos.y > p_min.y && pos.x < p_max.x && pos.y < p_max.y;
+}
+
+void main() {
+    uint i = gl_GlobalInvocationID.x;
+
+    vec4 pos = particles[i].pos;
+    vec4 dir = particles[i].dir;
+
+    dir.w += deltaSeconds;
+
+    if (dir.w > particle_lifetime) {
+        resetParticle(i, pos.w);
+        return;
+    }
+
+    pos += vec4(dir.xyz * SPEED * deltaSeconds, 0.0);
+    dir += vec4(GRAVITY.xyz * deltaSeconds, 0.0);
+
+    if (pos.y < 0 && checkBounds(pos.xy)) {
+        dir.y = -dir.y * BOUNCE_FACTOR;
+        pos.y = 0;
+    }
+
+    particles[i].dir = dir;
+    particles[i].pos = pos;
+
+    if (pos.y < -50)
+    resetParticle(i, pos.w);
+
+}
+
+")";
+#endif

include/shaders/vertex.vert

@@ -9,12 +9,14 @@ layout (location = 2) in vec4 pos;
 layout (location = 3) in vec4 dir;
 
 out vec2 uv_;
+out float index;
 
 uniform mat4 pvm;
 
 void main() {
     // passthough the UV (OpenGL interpolates this per fragment)
     uv_ = uv;
+    index = pos.w;
     // offset the vertex by the particle's position
     gl_Position = pvm * vec4(vertex + pos.xyz, 1.0);
 }

libs/BLT

@@ -1 +1 @@
-Subproject commit 289af1317141c53e04998b32760efced0956db57
+Subproject commit 08c542658240b60d1221e3580f3c6b4a2f483919

src/high_perf/gl_util.cpp

@@ -12,6 +12,18 @@
  * https://github.com/Tri11Paragon/COSC-3P98-Final-Project/blob/main/include/render/gl.h
  */
 
+static inline std::string removeEmptyFirstLines(const std::string& string){
+    auto lines = blt::string::split(string, "\n");
+    std::string new_source_string;
+    for (const auto& line : lines) {
+        if (!line.empty() && !blt::string::contains(line, "\"")) {
+            new_source_string += line;
+            new_source_string += "\n";
+        }
+    }
+    return new_source_string;
+}
+
 unsigned int shader::createShader(const std::string& source, int type) {
     const char* shader_code = source.c_str();
     // creates a Shader
@@ -153,4 +165,54 @@ shader::shader(shader&& move) noexcept {
         uniformVars.insert(pair);
     // by setting the program ID to -1 we tell the shader it has been moved.
     move.programID = -1;
-}
+}
+
+/**
+ * This part was made for this assignment and will likely be used in future projects
+ */
+
+compute_shader::compute_shader(const std::string& shader_source, bool loadAsString) {
+    int status;
+    std::string source;
+    const char* c_source;
+    
+    if (!loadAsString)
+        source = blt::fs::loadShaderFile(shader_source);
+    else
+        source = removeEmptyFirstLines(shader_source);
+    
+    c_source = source.c_str();
+    
+    shaderID = glCreateShader(GL_COMPUTE_SHADER);
+    
+    glShaderSource(shaderID, 1, &c_source, NULL);
+    glCompileShader(shaderID);
+    
+    glGetShaderiv(shaderID, GL_COMPILE_STATUS, &status);
+    if (!status){
+        int log_length = 0;
+        glGetShaderiv(shaderID, GL_INFO_LOG_LENGTH, &log_length);
+        blt::scoped_buffer<GLchar> infoLog{static_cast<unsigned long>(log_length + 1)};
+        glGetShaderInfoLog(shaderID, log_length + 1, nullptr, infoLog.buffer);
+        BLT_ERROR("Unable to compile compute shader! (%d)", log_length);
+        BLT_ERROR(infoLog.buffer);
+    }
+    
+    programID = glCreateProgram();
+    glAttachShader(programID, shaderID);
+    glLinkProgram(programID);
+    
+    glGetProgramiv(shaderID, GL_LINK_STATUS, &status);
+    if (!status){
+        int log_length = 0;
+        glGetProgramiv(programID, GL_INFO_LOG_LENGTH, &log_length);
+        blt::scoped_buffer<GLchar> infoLog{static_cast<unsigned long>(log_length + 1)};
+        glGetProgramInfoLog(programID, log_length + 1, nullptr, infoLog.buffer);
+        BLT_ERROR("Unable to link compute shader!");
+        BLT_ERROR(infoLog.buffer);
+    }
+}
+
+compute_shader::~compute_shader() {
+    glDeleteShader(shaderID);
+}

src/high_perf/high_perf.cpp

@@ -3,6 +3,7 @@
  * Licensed under GNU General Public License V3.0
  * See LICENSE file for license detail
  */
+#include <locale.h>
 #include <high_perf/gl_util.h>
 #include <modes/high_perf.h>
 #include <util.h>
@@ -10,6 +11,9 @@
 #include "blt/std/memory.h"
 #include <shaders/vertex.vert>
 #include <shaders/fragment.frag>
+#include <shaders/physics.comp>
+#include <stb_image.h>
+#include <stb_image_resize.h>
 
 //static inline float degreesToRadian(float deg) {
 //    return deg * (float)M_PI / 180.0f;
@@ -32,13 +36,26 @@ void window_resize(int width, int height) {
 
 }
 
-GLuint particleTranslationsVBO;
+// -------{VBO}-------
+GLuint particleTranslationsBuffer;
+GLuint particleOffsetsBuffer;
 GLuint verticesVBO;
 GLuint uvsVBO;
 GLuint indicesEBO;
+
+// -------{VAO}-------
 GLuint particleVAO;
 
-const unsigned int particle_count = 25000000;
+// -------{Textures}-------
+GLuint textureArrayID;
+
+const unsigned int TEXTURE_COUNT = 10;
+const unsigned int TEXTURE_WIDTH = 512;
+const unsigned int TEXTURE_HEIGHT = 512;
+
+// -------{Particles}-------
+const unsigned int particle_count = 128 * 10000;
+const unsigned int offset_count = 8192;
 
 // generally alignment to multiples of 4 floats helps performance, plus we can use that extra space for info we need.
 typedef struct {
@@ -69,19 +86,43 @@ blt::mat4x4 perspectiveMatrix;
 blt::mat4x4 viewMatrix;
 
 shader* instance_shader;
+compute_shader* physics_shader;
 
 void updateView() {
     viewMatrix = createViewMatrix();
 }
 
+bool execute = false;
+
+void beginExecution() {
+    execute = true;
+}
+
+void runPhysicsShader(){
+    if (!execute)
+        return;
+    
+    physics_shader->bind();
+    glUniform1f(0, (float)((double) getDelta() / 1000000000.0));
+    glBindBuffer(GL_SHADER_STORAGE_BUFFER, particleTranslationsBuffer);
+    physics_shader->execute(particle_count / 128, 1, 1);
+    
+    glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
+}
+
 void render() {
     updateView();
     perspectiveMatrix = blt::perspective(FOV, (float)WINDOW_WIDTH / (float)WINDOW_HEIGHT, 0.1f, 1000.0f);
     auto pvm = perspectiveMatrix * viewMatrix;
     
+    runPhysicsShader();
+    
     instance_shader->bind();
     instance_shader->setMatrix("pvm", pvm);
     
+    glActiveTexture(GL_TEXTURE0);
+    glBindTexture(GL_TEXTURE_2D_ARRAY, textureArrayID);
+    
     glBindVertexArray(particleVAO);
     glDrawElementsInstanced(GL_TRIANGLES, 6, GL_UNSIGNED_INT, 0, particle_count);
     glBindVertexArray(0);
@@ -89,23 +130,65 @@ void render() {
 }
 
 void init() {
-    blt::scoped_buffer<particle_record> translations{particle_count};
+    BLT_DEBUG("High performance subsystem init");
-    blt::random<float> pos{-50.0, 50.0};
+    setlocale(LC_NUMERIC, "");
+    BLT_INFO("Using %'d particle count", particle_count);
+    BLT_INFO("Loading %d texture(s) of size (%d, %d)", TEXTURE_COUNT, TEXTURE_WIDTH, TEXTURE_HEIGHT);
+    BLT_TRACE("Checking system constants");
     
+    // number of work groups allowed (min: 65535)
+    GLint workGroupCountX;
+    GLint workGroupCountY;
+    GLint workGroupCountZ;
+    // max local size of the work groups (min: 1024, 1024, 64)
+    GLint workGroupSize;
+    // max number of work group invocations (min: 1024)
+    GLint workGroupInvocations;
+    
+    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_COUNT, 0, &workGroupCountX);
+    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_COUNT, 1, &workGroupCountY);
+    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_COUNT, 2, &workGroupCountZ);
+    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_SIZE, 0, &workGroupSize);
+    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_INVOCATIONS, 0, &workGroupInvocations);
+    
+    BLT_INFO("This system's OpenGL supports (%d, %d, %d) work groups", workGroupCountX, workGroupCountY, workGroupCountZ);
+    BLT_INFO("\tLocal group max total size: %d", workGroupSize);
+    BLT_INFO("\tMax work group invocations: %d", workGroupInvocations);
+    blt::scoped_buffer<particle_record> translations{particle_count};
+    blt::scoped_buffer<vec4> offsets{offset_count};
+    blt::random<float> dir{-1, 1};
+    blt::random<float> lifetime{0, 25};
+    
+    BLT_TRACE("Creating particles");
     for (int i = 0; i < particle_count; i++)
-        translations[i] = particle_record{vec4{pos.get(), pos.get() / 2, pos.get(), (float)(i % 10)}, vec4{0, 0, 0, 0}};
+        translations[i] = particle_record{vec4{0, 1, 0, (float)(i % 10)}, vec4{0, 1, 0, lifetime.get()}};
+    
+    for (int i = 0; i < offset_count; i++) {
+        blt::vec2 v {dir.get(), dir.get()};
+        v = v.normalize();
+        offsets[i] = vec4{v[0], 0, v[1], 0};
+    }
     
     // ----------------------------------
     //       Create OpenGL Objects
     // ----------------------------------
+    BLT_TRACE("Creating OpenGL objects");
     // create our VAO
     glGenVertexArrays(1, &particleVAO);
     // create our VBOs
-    glGenBuffers(1, &particleTranslationsVBO);
+    glGenBuffers(1, &particleTranslationsBuffer);
+    glGenBuffers(1, &particleOffsetsBuffer);
     glGenBuffers(1, &verticesVBO);
     glGenBuffers(1, &uvsVBO);
     glGenBuffers(1, &indicesEBO);
+    // create our texture
+    glGenTextures(1, &textureArrayID);
     
+    // ----------------------------------
+    //    Upload/Assign OpenGL Objects
+    // ----------------------------------
+    
+    BLT_TRACE("Uploading VBO data and assigning to VAO");
     glBindVertexArray(particleVAO);
     
     // bind and upload vertices data to the GPU
@@ -124,7 +207,7 @@ void init() {
     glEnableVertexAttribArray(1);
     
     int translations_size = sizeof(particle_record) * particle_count;
-    glBindBuffer(GL_ARRAY_BUFFER, particleTranslationsVBO);
+    glBindBuffer(GL_ARRAY_BUFFER, particleTranslationsBuffer);
     glBufferData(GL_ARRAY_BUFFER, translations_size, translations.buffer, GL_DYNAMIC_DRAW); // allocate some memory on the GPU
     glVertexAttribPointer(2, 4, GL_FLOAT, GL_FALSE, sizeof(particle_record), (void*) 0);
     glVertexAttribPointer(3, 4, GL_FLOAT, GL_FALSE, sizeof(particle_record), (void*) offsetof(particle_record, dir));
@@ -134,18 +217,108 @@ void init() {
     glEnableVertexAttribArray(2);
     glEnableVertexAttribArray(3);
     
+    // allow the particle buffer to be used in the computer shader!
+    glBindBuffer(GL_SHADER_STORAGE_BUFFER, particleTranslationsBuffer);
+    glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, particleTranslationsBuffer);
+    
+    // generating random numbers on the GPU is hard, we can use enough precomputed random offsets to simulate real time randomness
+    glBindBuffer(GL_SHADER_STORAGE_BUFFER, particleOffsetsBuffer);
+    glBufferData(GL_SHADER_STORAGE_BUFFER, offset_count * sizeof(vec4), offsets.buffer, GL_STATIC_DRAW);
+    glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 1, particleOffsetsBuffer);
     
     glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, indicesEBO);
     glBufferData(GL_ELEMENT_ARRAY_BUFFER, sizeof(float) * 6, indices, GL_STATIC_DRAW);
     
     
+    // ----------------------------------
+    //             Texturing
+    // ----------------------------------
+    BLT_TRACE("Creating array texture");
+    // based on my final project's texture array implementation
+    glBindTexture(GL_TEXTURE_2D_ARRAY, textureArrayID);
+    // allocate immutable storage for our textures
+    // we can change what is stored inside the texture, but we cannot change its size
+    // which is why we need to be specific here about the type of data we will be storing.
+    glTexStorage3D(GL_TEXTURE_2D_ARRAY, 4, GL_RGBA8, TEXTURE_WIDTH, TEXTURE_HEIGHT, TEXTURE_COUNT);
+    
+    std::string texture_locations[TEXTURE_COUNT] = {
+            "wayland.png",
+            "SPONGEBOB_YOUTUBE.jpg",
+            "1618325873904.png",
+            "1665624414712991.jpg",
+            "stonks.png",
+            "yak.png",
+            "penguin.jpg",
+            "fFTkb.png",
+            "depression.png",
+            "1665624414712991.jpg"
+    };
+    
+    constexpr int channel_count = 4;
+    
+    int level = 0;
+    stbi_set_flip_vertically_on_load(true);
+    for (const std::string& texture_loc : texture_locations){
+        // load the texture
+        int width, height, channels;
+        auto* data = stbi_load(
+                (std::string("resources/") += texture_loc).c_str(), &width, &height,
+                &channels, channel_count
+        );
+        auto* resized_data = data;
+        
+        // resize if necessary
+        if (width != TEXTURE_WIDTH || height != TEXTURE_HEIGHT){
+            // needs to be malloc since stbi_image_free is just free()
+            auto output_data = (unsigned char*) malloc(
+                    TEXTURE_WIDTH * TEXTURE_HEIGHT * channel_count
+            );
+            if (stbir_resize_uint8(
+                    // input
+                    data, width, height, 0,
+                    // output
+                    output_data, TEXTURE_WIDTH, TEXTURE_HEIGHT, 0,
+                    // channels
+                    channel_count
+            )) {
+                BLT_WARN("Error resizing block texture image!");
+            }
+            stbi_image_free(data);
+            resized_data = output_data;
+        }
+        
+        // upload image to the gpu
+        glTexSubImage3D(
+                GL_TEXTURE_2D_ARRAY, 0, 0, 0, level++, TEXTURE_WIDTH, TEXTURE_HEIGHT, 1,
+                GL_RGBA, GL_UNSIGNED_BYTE, resized_data
+        );
+        
+        stbi_image_free(resized_data);
+    }
+    
+    glTexParameteri(GL_TEXTURE_2D_ARRAY, GL_TEXTURE_WRAP_S, GL_REPEAT);
+    glTexParameteri(GL_TEXTURE_2D_ARRAY, GL_TEXTURE_WRAP_T, GL_REPEAT);
+    glTexParameteri(GL_TEXTURE_2D_ARRAY, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR);
+    glTexParameteri(GL_TEXTURE_2D_ARRAY, GL_TEXTURE_MAG_FILTER, GL_LINEAR_MIPMAP_LINEAR);
+    // Anisotropy helps preserve textures at oblique angles
+    float a = 0;
+    glGetFloatv(GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT, &a);
+    glTexParameterf(GL_TEXTURE_2D_ARRAY, GL_TEXTURE_MAX_ANISOTROPY_EXT, a);
+    // mipmaps reduce resolution of textures as the distance to them increases
+    glGenerateMipmap(GL_TEXTURE_2D_ARRAY);
+    
+    BLT_TRACE("Loading shaders");
+    
     instance_shader = new shader(shader_vert, shader_frag, "", true);
+    physics_shader = new compute_shader(shader_physics);
+    
+    BLT_DEBUG("High performance subsystem init complete!");
 }
 
 void cleanup() {
     // cleanup opengl resources
     glDeleteVertexArrays(1, &particleVAO);
-    glDeleteBuffers(1, &particleTranslationsVBO);
+    glDeleteBuffers(1, &particleTranslationsBuffer);
     glDeleteBuffers(1, &verticesVBO);
     glDeleteBuffers(1, &uvsVBO);
     glDeleteBuffers(1, &indicesEBO);

src/main.cpp

@@ -140,6 +140,8 @@ int main(int argc, char** argv) {
             fountain->toggleSpray();
         if (key == 't')
             fountain->toggleTexRandomizer();
+        if (key == 'p')
+            beginExecution();
     });
     glutSpecialFunc([](int k, int x, int y) -> void {
         cam.specialPress(k);

src/texture.cpp

@@ -5,6 +5,8 @@
 #include <texture.h>
 #define STB_IMAGE_IMPLEMENTATION
 #include <stb_image.h>
+#define STB_IMAGE_RESIZE_IMPLEMENTATION
+#include <stb_image_resize.h>
 
 
 texture* loadTexture(const std::string& path) {