compute shaders!

demos/99-compute.c

@@ -0,0 +1,19 @@
+#include "v4k.h"
+
+int main() {
+    // 75% sized, MSAAx2
+    window_create(50, WINDOW_SQUARE);
+    window_title(__FILE__);
+
+    unsigned comp = compute(vfs_read("shaders/compute-test.glsl"));
+    texture_t tex = texture_create(512, 512, 4, 0, TEXTURE_LINEAR|TEXTURE_FLOAT);
+
+    while ( window_swap() && !input_down(KEY_ESC) ){
+        shader_bind(comp);
+        shader_float("t", (float)window_time());
+        shader_image(&tex, 0, 0, -1, READ);
+        dispatch(512, 512, 1);
+        imageWriteBarrier();
+        fullscreen_quad_rgb(tex, 2.2);
+    }
+}

demos/art/shaders/compute-test.glsl

@@ -0,0 +1,18 @@
+layout (local_size_x = 10, local_size_y = 10, local_size_z = 1) in;
+
+layout(rgba32f, binding = 0) uniform image2D imgOutput;
+
+layout (location = 0) uniform float t;                 /** Time */
+
+void main() {
+    vec4 value = vec4(0.0, 0.0, 0.0, 1.0);
+    ivec2 texelCoord = ivec2(gl_GlobalInvocationID.xy);
+    
+    float speed = 100;
+    // the width of the texture
+    float width = 1000;
+
+    value.x = mod(float(texelCoord.x) + t * speed, width) / (gl_NumWorkGroups.x * gl_WorkGroupSize.x);
+    value.y = float(texelCoord.y)/(gl_NumWorkGroups.y*gl_WorkGroupSize.y);
+    imageStore(imgOutput, texelCoord, value);
+}

engine/bind/v4k.lua

@@ -2230,6 +2230,7 @@ union { unsigned y, h; };
 union { unsigned z, d; };
 union { unsigned n, bpp; };
 handle id, unit;
+unsigned texel_type;
 unsigned flags;
 char* filename;
 bool transparent;
@@ -2358,6 +2359,16 @@ int texture_width;
      void shader_colormap(const char *name, colormap_t cm);
  unsigned shader_get_active();
  void     shader_destroy(unsigned shader);
+enum ACCESS_MODE {
+READ,
+WRITE,
+READ_WRITE
+};
+ unsigned compute(const char *cs);
+ void dispatch(unsigned wx, unsigned wy, unsigned wz);
+ void shader_image(texture_t *t, unsigned unit, unsigned level, int layer , unsigned access);
+ void shader_image_unit(unsigned texture, unsigned unit, unsigned level, int layer , unsigned texel_type, unsigned access);
+ void imageWriteBarrier();
 enum MESH_FLAGS {
 MESH_STATIC = 0,
 MESH_STREAM = 1,

engine/joint/v4k.h

@@ -16280,6 +16280,7 @@ typedef struct texture_t {
     union { unsigned z, d; };
     union { unsigned n, bpp; };
     handle id, unit;
+    unsigned texel_type;
     unsigned flags;
     char* filename;
     bool transparent;
@@ -16481,6 +16482,19 @@ API     void shader_colormap(const char *name, colormap_t cm);
 API unsigned shader_get_active();
 API void     shader_destroy(unsigned shader);
 
+// compute shaders
+enum ACCESS_MODE {
+    READ,
+    WRITE,
+    READ_WRITE
+};
+
+API unsigned compute(const char *cs);
+API void dispatch(unsigned wx, unsigned wy, unsigned wz);
+API void shader_image(texture_t *t, unsigned unit, unsigned level, int layer /* -1 to disable layered access */, unsigned access);
+API void shader_image_unit(unsigned texture, unsigned unit, unsigned level, int layer /* -1 to disable layered access */, unsigned texel_type, unsigned access);
+API void imageWriteBarrier();
+
 // -----------------------------------------------------------------------------
 // meshes (@fixme: deprecate?)
 
@@ -340094,6 +340108,56 @@ unsigned shader(const char *vs, const char *fs, const char *attribs, const char
     return program;
 }
 
+unsigned compute(const char *cs){
+    #if is(ems)
+    return 0;
+    #else
+    PRINTF(/*"!"*/"Compiling compute shader\n");
+
+    cs = cs[0] == '#' && cs[1] == 'c' ? cs : va("#version 430 core\n%s", cs ? cs : "");
+
+    GLuint comp = shader_compile(GL_COMPUTE_SHADER, cs);
+    GLuint program = 0;
+
+    if( comp ) {
+        program = glCreateProgram();
+
+        glAttachShader(program, comp);
+
+        glLinkProgram(program);
+
+        GLint status = GL_FALSE, length;
+        glGetProgramiv(program, GL_LINK_STATUS, &status);
+#ifdef DEBUG_SHADER
+        if (status != GL_FALSE && program == DEBUG_SHADER) {
+#else
+        if (status == GL_FALSE) {
+#endif
+            glGetProgramiv(program, GL_INFO_LOG_LENGTH, &length);
+            char *buf = stack(length+1);
+            glGetProgramInfoLog(program, length, NULL, buf);
+            puts("--- cs:");
+            shader_print(cs);
+        }
+        if (status == GL_FALSE) {
+            PANIC("ERROR: shader(): Shader/program link: %s\n", buf);
+            return 0;
+        }
+
+        glDeleteShader(comp);
+    }
+    return program;
+    #endif
+}
+
+void dispatch(unsigned wx, unsigned wy, unsigned wz){
+    glDispatchCompute(wx, wy, wz);
+}
+
+void imageWriteBarrier(){
+    glMemoryBarrier(GL_SHADER_IMAGE_ACCESS_BARRIER_BIT);
+}
+
 void shader_destroy(unsigned program){
     if( program == ~0u ) return;
     glDeleteProgram(program);
@@ -340126,6 +340190,13 @@ void shader_texture_unit(const char *sampler, unsigned id, unsigned unit) {
     glActiveTexture(GL_TEXTURE0 + unit);
     glBindTexture(GL_TEXTURE_2D, id);
 }
+void shader_image(texture_t *t, unsigned unit, unsigned level, int layer /* -1 to disable layered access */, unsigned access){
+    shader_image_unit(t->id, unit, level, layer, t->texel_type, access);
+}
+void shader_image_unit(unsigned texture, unsigned unit, unsigned level, int layer, unsigned texel_type, unsigned access){
+    GLenum gl_access[] = {GL_READ_ONLY, GL_WRITE_ONLY, GL_READ_WRITE};
+    glBindImageTexture(unit, texture, level, layer!=-1, layer!=-1?layer:0, gl_access[access], texel_type);
+}
 
 void shader_colormap(const char *name, colormap_t c ) {
     // assumes shader uses `struct { vec4 color; bool has_tex } name + sampler2D name_tex;`
@@ -340251,7 +340322,7 @@ unsigned texture_update(texture_t *t, unsigned w, unsigned h, unsigned n, const
     GLuint pixel_types[] = { GL_RED, GL_RED, GL_RG, GL_RGB, GL_RGBA, GL_R32F, GL_R32F, GL_RG32F, GL_RGB32F, GL_RGBA32F };
     GLenum pixel_storage = flags & TEXTURE_FLOAT ? GL_FLOAT : GL_UNSIGNED_BYTE;
     GLuint pixel_type = pixel_types[ n ];
-    GLuint texel_type = pixel_types[ n + 5 * !!(flags & TEXTURE_FLOAT) ];
+    GLuint texel_type = t->texel_type = pixel_types[ n + 5 * !!(flags & TEXTURE_FLOAT) ];
     GLenum wrap = GL_CLAMP_TO_EDGE;
     GLenum min_filter = GL_NEAREST, mag_filter = GL_NEAREST;
 //    GLfloat color = (flags&7)/7.f, border_color[4] = { color, color, color, 1.f };
@@ -349463,9 +349534,14 @@ void window_hints(unsigned flags) {
     //glfwWindowHint( GLFW_COCOA_MENUBAR, GLFW_FALSE );
     #endif
 
+    #ifdef __APPLE__
     /* We need to explicitly ask for a 3.2 context on OS X */
     glfwWindowHint(GLFW_CONTEXT_VERSION_MAJOR, 3); // osx
     glfwWindowHint(GLFW_CONTEXT_VERSION_MINOR, 2); // osx, 2:#version150,3:330
+    #else
+    glfwWindowHint(GLFW_CONTEXT_VERSION_MAJOR, 4);
+    glfwWindowHint(GLFW_CONTEXT_VERSION_MINOR, 3);
+    #endif
     #ifdef __APPLE__
         glfwWindowHint(GLFW_OPENGL_FORWARD_COMPAT, GL_TRUE); //osx
     #endif

engine/split/v4k_render.c

@@ -194,6 +194,56 @@ unsigned shader(const char *vs, const char *fs, const char *attribs, const char
     return program;
 }
 
+unsigned compute(const char *cs){
+    #if is(ems)
+    return 0;
+    #else
+    PRINTF(/*"!"*/"Compiling compute shader\n");
+
+    cs = cs[0] == '#' && cs[1] == 'c' ? cs : va("#version 430 core\n%s", cs ? cs : "");
+
+    GLuint comp = shader_compile(GL_COMPUTE_SHADER, cs);
+    GLuint program = 0;
+
+    if( comp ) {
+        program = glCreateProgram();
+
+        glAttachShader(program, comp);
+
+        glLinkProgram(program);
+
+        GLint status = GL_FALSE, length;
+        glGetProgramiv(program, GL_LINK_STATUS, &status);
+#ifdef DEBUG_SHADER
+        if (status != GL_FALSE && program == DEBUG_SHADER) {
+#else
+        if (status == GL_FALSE) {
+#endif
+            glGetProgramiv(program, GL_INFO_LOG_LENGTH, &length);
+            char *buf = stack(length+1);
+            glGetProgramInfoLog(program, length, NULL, buf);
+            puts("--- cs:");
+            shader_print(cs);
+        }
+        if (status == GL_FALSE) {
+            PANIC("ERROR: shader(): Shader/program link: %s\n", buf);
+            return 0;
+        }
+
+        glDeleteShader(comp);
+    }
+    return program;
+    #endif
+}
+
+void dispatch(unsigned wx, unsigned wy, unsigned wz){
+    glDispatchCompute(wx, wy, wz);
+}
+
+void imageWriteBarrier(){
+    glMemoryBarrier(GL_SHADER_IMAGE_ACCESS_BARRIER_BIT);
+}
+
 void shader_destroy(unsigned program){
     if( program == ~0u ) return;
     glDeleteProgram(program);
@@ -226,6 +276,13 @@ void shader_texture_unit(const char *sampler, unsigned id, unsigned unit) {
     glActiveTexture(GL_TEXTURE0 + unit);
     glBindTexture(GL_TEXTURE_2D, id);
 }
+void shader_image(texture_t *t, unsigned unit, unsigned level, int layer /* -1 to disable layered access */, unsigned access){
+    shader_image_unit(t->id, unit, level, layer, t->texel_type, access);
+}
+void shader_image_unit(unsigned texture, unsigned unit, unsigned level, int layer, unsigned texel_type, unsigned access){
+    GLenum gl_access[] = {GL_READ_ONLY, GL_WRITE_ONLY, GL_READ_WRITE};
+    glBindImageTexture(unit, texture, level, layer!=-1, layer!=-1?layer:0, gl_access[access], texel_type);
+}
 
 void shader_colormap(const char *name, colormap_t c ) {
     // assumes shader uses `struct { vec4 color; bool has_tex } name + sampler2D name_tex;`
@@ -351,7 +408,7 @@ unsigned texture_update(texture_t *t, unsigned w, unsigned h, unsigned n, const
     GLuint pixel_types[] = { GL_RED, GL_RED, GL_RG, GL_RGB, GL_RGBA, GL_R32F, GL_R32F, GL_RG32F, GL_RGB32F, GL_RGBA32F };
     GLenum pixel_storage = flags & TEXTURE_FLOAT ? GL_FLOAT : GL_UNSIGNED_BYTE;
     GLuint pixel_type = pixel_types[ n ];
-    GLuint texel_type = pixel_types[ n + 5 * !!(flags & TEXTURE_FLOAT) ];
+    GLuint texel_type = t->texel_type = pixel_types[ n + 5 * !!(flags & TEXTURE_FLOAT) ];
     GLenum wrap = GL_CLAMP_TO_EDGE;
     GLenum min_filter = GL_NEAREST, mag_filter = GL_NEAREST;
 //    GLfloat color = (flags&7)/7.f, border_color[4] = { color, color, color, 1.f };

engine/split/v4k_render.h

@@ -125,6 +125,7 @@ typedef struct texture_t {
     union { unsigned z, d; };
     union { unsigned n, bpp; };
     handle id, unit;
+    unsigned texel_type;
     unsigned flags;
     char* filename;
     bool transparent;
@@ -326,6 +327,19 @@ API     void shader_colormap(const char *name, colormap_t cm);
 API unsigned shader_get_active();
 API void     shader_destroy(unsigned shader);
 
+// compute shaders
+enum ACCESS_MODE {
+    READ,
+    WRITE,
+    READ_WRITE
+};
+
+API unsigned compute(const char *cs);
+API void dispatch(unsigned wx, unsigned wy, unsigned wz);
+API void shader_image(texture_t *t, unsigned unit, unsigned level, int layer /* -1 to disable layered access */, unsigned access);
+API void shader_image_unit(unsigned texture, unsigned unit, unsigned level, int layer /* -1 to disable layered access */, unsigned texel_type, unsigned access);
+API void imageWriteBarrier();
+
 // -----------------------------------------------------------------------------
 // meshes (@fixme: deprecate?)
 

engine/split/v4k_window.c

@@ -169,9 +169,14 @@ void window_hints(unsigned flags) {
     //glfwWindowHint( GLFW_COCOA_MENUBAR, GLFW_FALSE );
     #endif
 
+    #ifdef __APPLE__
     /* We need to explicitly ask for a 3.2 context on OS X */
     glfwWindowHint(GLFW_CONTEXT_VERSION_MAJOR, 3); // osx
     glfwWindowHint(GLFW_CONTEXT_VERSION_MINOR, 2); // osx, 2:#version150,3:330
+    #else
+    glfwWindowHint(GLFW_CONTEXT_VERSION_MAJOR, 4);
+    glfwWindowHint(GLFW_CONTEXT_VERSION_MINOR, 3);
+    #endif
     #ifdef __APPLE__
         glfwWindowHint(GLFW_OPENGL_FORWARD_COMPAT, GL_TRUE); //osx
     #endif

engine/v4k.c

@@ -11108,6 +11108,56 @@ unsigned shader(const char *vs, const char *fs, const char *attribs, const char
     return program;
 }
 
+unsigned compute(const char *cs){
+    #if is(ems)
+    return 0;
+    #else
+    PRINTF(/*"!"*/"Compiling compute shader\n");
+
+    cs = cs[0] == '#' && cs[1] == 'c' ? cs : va("#version 430 core\n%s", cs ? cs : "");
+
+    GLuint comp = shader_compile(GL_COMPUTE_SHADER, cs);
+    GLuint program = 0;
+
+    if( comp ) {
+        program = glCreateProgram();
+
+        glAttachShader(program, comp);
+
+        glLinkProgram(program);
+
+        GLint status = GL_FALSE, length;
+        glGetProgramiv(program, GL_LINK_STATUS, &status);
+#ifdef DEBUG_SHADER
+        if (status != GL_FALSE && program == DEBUG_SHADER) {
+#else
+        if (status == GL_FALSE) {
+#endif
+            glGetProgramiv(program, GL_INFO_LOG_LENGTH, &length);
+            char *buf = stack(length+1);
+            glGetProgramInfoLog(program, length, NULL, buf);
+            puts("--- cs:");
+            shader_print(cs);
+        }
+        if (status == GL_FALSE) {
+            PANIC("ERROR: shader(): Shader/program link: %s\n", buf);
+            return 0;
+        }
+
+        glDeleteShader(comp);
+    }
+    return program;
+    #endif
+}
+
+void dispatch(unsigned wx, unsigned wy, unsigned wz){
+    glDispatchCompute(wx, wy, wz);
+}
+
+void imageWriteBarrier(){
+    glMemoryBarrier(GL_SHADER_IMAGE_ACCESS_BARRIER_BIT);
+}
+
 void shader_destroy(unsigned program){
     if( program == ~0u ) return;
     glDeleteProgram(program);
@@ -11140,6 +11190,13 @@ void shader_texture_unit(const char *sampler, unsigned id, unsigned unit) {
     glActiveTexture(GL_TEXTURE0 + unit);
     glBindTexture(GL_TEXTURE_2D, id);
 }
+void shader_image(texture_t *t, unsigned unit, unsigned level, int layer /* -1 to disable layered access */, unsigned access){
+    shader_image_unit(t->id, unit, level, layer, t->texel_type, access);
+}
+void shader_image_unit(unsigned texture, unsigned unit, unsigned level, int layer, unsigned texel_type, unsigned access){
+    GLenum gl_access[] = {GL_READ_ONLY, GL_WRITE_ONLY, GL_READ_WRITE};
+    glBindImageTexture(unit, texture, level, layer!=-1, layer!=-1?layer:0, gl_access[access], texel_type);
+}
 
 void shader_colormap(const char *name, colormap_t c ) {
     // assumes shader uses `struct { vec4 color; bool has_tex } name + sampler2D name_tex;`
@@ -11265,7 +11322,7 @@ unsigned texture_update(texture_t *t, unsigned w, unsigned h, unsigned n, const
     GLuint pixel_types[] = { GL_RED, GL_RED, GL_RG, GL_RGB, GL_RGBA, GL_R32F, GL_R32F, GL_RG32F, GL_RGB32F, GL_RGBA32F };
     GLenum pixel_storage = flags & TEXTURE_FLOAT ? GL_FLOAT : GL_UNSIGNED_BYTE;
     GLuint pixel_type = pixel_types[ n ];
-    GLuint texel_type = pixel_types[ n + 5 * !!(flags & TEXTURE_FLOAT) ];
+    GLuint texel_type = t->texel_type = pixel_types[ n + 5 * !!(flags & TEXTURE_FLOAT) ];
     GLenum wrap = GL_CLAMP_TO_EDGE;
     GLenum min_filter = GL_NEAREST, mag_filter = GL_NEAREST;
 //    GLfloat color = (flags&7)/7.f, border_color[4] = { color, color, color, 1.f };
@@ -20477,9 +20534,14 @@ void window_hints(unsigned flags) {
     //glfwWindowHint( GLFW_COCOA_MENUBAR, GLFW_FALSE );
     #endif
 
+    #ifdef __APPLE__
     /* We need to explicitly ask for a 3.2 context on OS X */
     glfwWindowHint(GLFW_CONTEXT_VERSION_MAJOR, 3); // osx
     glfwWindowHint(GLFW_CONTEXT_VERSION_MINOR, 2); // osx, 2:#version150,3:330
+    #else
+    glfwWindowHint(GLFW_CONTEXT_VERSION_MAJOR, 4);
+    glfwWindowHint(GLFW_CONTEXT_VERSION_MINOR, 3);
+    #endif
     #ifdef __APPLE__
         glfwWindowHint(GLFW_OPENGL_FORWARD_COMPAT, GL_TRUE); //osx
     #endif

engine/v4k.h

@@ -2363,6 +2363,7 @@ typedef struct texture_t {
     union { unsigned z, d; };
     union { unsigned n, bpp; };
     handle id, unit;
+    unsigned texel_type;
     unsigned flags;
     char* filename;
     bool transparent;
@@ -2564,6 +2565,19 @@ API     void shader_colormap(const char *name, colormap_t cm);
 API unsigned shader_get_active();
 API void     shader_destroy(unsigned shader);
 
+// compute shaders
+enum ACCESS_MODE {
+    READ,
+    WRITE,
+    READ_WRITE
+};
+
+API unsigned compute(const char *cs);
+API void dispatch(unsigned wx, unsigned wy, unsigned wz);
+API void shader_image(texture_t *t, unsigned unit, unsigned level, int layer /* -1 to disable layered access */, unsigned access);
+API void shader_image_unit(unsigned texture, unsigned unit, unsigned level, int layer /* -1 to disable layered access */, unsigned texel_type, unsigned access);
+API void imageWriteBarrier();
+
 // -----------------------------------------------------------------------------
 // meshes (@fixme: deprecate?)