From Wang Rui, "In the attached files I've added the Compute Shader support for OSG, as well as serializer updates and a new osgcomputeshaders example. My submission also include a setComputeGroups() function in Program for setting compute-shader work groups, and a bindToImageUnit() function in Texture for binding textures as image variables in shaders.
All code are tested on Windows 7 + NVIDIA GFX 570 with the latest GeForce 310.70 Driver (BETA), which could support OpenGL 4.3. Compute shader information can be found at "http://www.opengl.org/registry/specs/ARB/compute_shader.txt" "
This commit is contained in:
Robert Osfield
@@ -442,6 +442,28 @@ typedef char GLchar;
|
||||
#define GL_UNSIGNED_INT_ATOMIC_COUNTER 0x92DB
|
||||
#endif
|
||||
|
||||
// ARB_compute_shader
|
||||
#ifndef GL_ARB_compute_shader
|
||||
#define GL_COMPUTE_SHADER 0x91B9
|
||||
#define GL_MAX_COMPUTE_UNIFORM_BLOCKS 0x91BB
|
||||
#define GL_MAX_COMPUTE_TEXTURE_IMAGE_UNITS 0x91BC
|
||||
#define GL_MAX_COMPUTE_IMAGE_UNIFORMS 0x91BD
|
||||
#define GL_MAX_COMPUTE_SHARED_MEMORY_SIZE 0x8262
|
||||
#define GL_MAX_COMPUTE_UNIFORM_COMPONENTS 0x8263
|
||||
#define GL_MAX_COMPUTE_ATOMIC_COUNTER_BUFFERS 0x8264
|
||||
#define GL_MAX_COMPUTE_ATOMIC_COUNTERS 0x8265
|
||||
#define GL_MAX_COMBINED_COMPUTE_UNIFORM_COMPONENTS 0x8266
|
||||
#define GL_MAX_COMPUTE_LOCAL_INVOCATIONS 0x90EB
|
||||
#define GL_MAX_COMPUTE_WORK_GROUP_COUNT 0x91BE
|
||||
#define GL_MAX_COMPUTE_WORK_GROUP_SIZE 0x91BF
|
||||
#define GL_COMPUTE_LOCAL_WORK_SIZE 0x8267
|
||||
#define GL_UNIFORM_BLOCK_REFERENCED_BY_COMPUTE_SHADER 0x90EC
|
||||
#define GL_ATOMIC_COUNTER_BUFFER_REFERENCED_BY_COMPUTE_SHADER 0x90ED
|
||||
#define GL_DISPATCH_INDIRECT_BUFFER 0x90EE
|
||||
#define GL_DISPATCH_INDIRECT_BUFFER_BINDING 0x90EF
|
||||
#define GL_COMPUTE_SHADER_BIT 0x00000020
|
||||
#endif
|
||||
|
||||
#ifndef GL_ARB_depth_clamp
|
||||
#define GL_DEPTH_CLAMP 0x864F
|
||||
#endif
|
||||
@@ -689,6 +711,8 @@ class OSG_EXPORT GL2Extensions : public osg::Referenced
|
||||
// ARB_shader_atomic_counters
|
||||
void glGetActiveAtomicCounterBufferiv( GLuint program, GLuint bufferIndex, GLenum pname, GLint* params ) const;
|
||||
|
||||
// ARB_compute_shader
|
||||
void glDispatchCompute( GLuint numGroupsX, GLuint numGroupsY, GLuint numGroupsZ ) const;
|
||||
|
||||
protected:
|
||||
~GL2Extensions() {}
|
||||
@@ -855,6 +879,7 @@ class OSG_EXPORT GL2Extensions : public osg::Referenced
|
||||
typedef void (GL_APIENTRY * UniformMatrix3x4dvProc)( GLint location, GLsizei count, GLboolean transpose, const GLdouble* value );
|
||||
typedef void (GL_APIENTRY * UniformMatrix4x3dvProc)( GLint location, GLsizei count, GLboolean transpose, const GLdouble* value );
|
||||
typedef void (GL_APIENTRY * GetActiveAtomicCounterBufferivProc)( GLuint program, GLuint bufferIndex, GLenum pname, GLint* params );
|
||||
typedef void (GL_APIENTRY * DispatchComputeProc)( GLuint numGroupsX, GLuint numGroupsY, GLuint numGroupsZ );
|
||||
|
||||
BlendEquationSeparateProc _glBlendEquationSeparate;
|
||||
DrawBuffersProc _glDrawBuffers;
|
||||
@@ -1020,6 +1045,9 @@ class OSG_EXPORT GL2Extensions : public osg::Referenced
|
||||
|
||||
// ARB_shader_atomic_counters
|
||||
GetActiveAtomicCounterBufferivProc _glGetActiveAtomicCounterBufferiv;
|
||||
|
||||
// ARB_compute_shader
|
||||
DispatchComputeProc _glDispatchCompute;
|
||||
};
|
||||
|
||||
}
|
||||
|
||||
@@ -101,6 +101,10 @@ class OSG_EXPORT Program : public osg::StateAttribute
|
||||
|
||||
void setParameterfv( GLenum pname, const GLfloat* value );
|
||||
const GLfloat* getParameterfv( GLenum pname ) const;
|
||||
|
||||
/** Set/get compute shader work groups */
|
||||
void setComputeGroups( GLint numGroupsX, GLint numGroupsY, GLint numGroupsZ );
|
||||
void getComputeGroups( GLint& numGroupsX, GLint& numGroupsY, GLint& numGroupsZ ) const;
|
||||
|
||||
/** Add an attribute location binding. */
|
||||
void addBindAttribLocation( const std::string& name, GLuint index );
|
||||
@@ -388,6 +392,11 @@ class OSG_EXPORT Program : public osg::StateAttribute
|
||||
// todo add tessellation default level
|
||||
//GLfloat _patchDefaultInnerLevel[2];
|
||||
//GLfloat _patchDefaultOuterLevel[4];
|
||||
|
||||
/** Parameter maintained with glDispatchCompute */
|
||||
GLint _numGroupsX;
|
||||
GLint _numGroupsY;
|
||||
GLint _numGroupsZ;
|
||||
|
||||
private:
|
||||
Program& operator=(const Program&); // disallowed
|
||||
|
||||
@@ -92,6 +92,7 @@ class OSG_EXPORT Shader : public osg::Object
|
||||
TESSEVALUATION = GL_TESS_EVALUATION_SHADER,
|
||||
GEOMETRY = GL_GEOMETRY_SHADER_EXT,
|
||||
FRAGMENT = GL_FRAGMENT_SHADER,
|
||||
COMPUTE = GL_COMPUTE_SHADER,
|
||||
UNDEFINED = -1
|
||||
};
|
||||
|
||||
|
||||
@@ -302,7 +302,6 @@
|
||||
#define GL_RGBA_INTEGER_MODE_EXT 0x8D9E
|
||||
#endif
|
||||
|
||||
|
||||
#ifndef GL_ARB_texture_rg
|
||||
#define GL_RG 0x8227
|
||||
#define GL_RG_INTEGER 0x8228
|
||||
@@ -328,12 +327,77 @@
|
||||
#define GL_RG32UI 0x823C
|
||||
#endif
|
||||
|
||||
#ifndef GL_ARB_shader_image_load_store
|
||||
#define GL_VERTEX_ATTRIB_ARRAY_BARRIER_BIT 0x00000001
|
||||
#define GL_ELEMENT_ARRAY_BARRIER_BIT 0x00000002
|
||||
#define GL_UNIFORM_BARRIER_BIT 0x00000004
|
||||
#define GL_TEXTURE_FETCH_BARRIER_BIT 0x00000008
|
||||
#define GL_SHADER_IMAGE_ACCESS_BARRIER_BIT 0x00000020
|
||||
#define GL_COMMAND_BARRIER_BIT 0x00000040
|
||||
#define GL_PIXEL_BUFFER_BARRIER_BIT 0x00000080
|
||||
#define GL_TEXTURE_UPDATE_BARRIER_BIT 0x00000100
|
||||
#define GL_BUFFER_UPDATE_BARRIER_BIT 0x00000200
|
||||
#define GL_FRAMEBUFFER_BARRIER_BIT 0x00000400
|
||||
#define GL_TRANSFORM_FEEDBACK_BARRIER_BIT 0x00000800
|
||||
#define GL_ATOMIC_COUNTER_BARRIER_BIT 0x00001000
|
||||
#define GL_ALL_BARRIER_BITS 0xFFFFFFFF
|
||||
#define GL_MAX_IMAGE_UNITS 0x8F38
|
||||
#define GL_MAX_COMBINED_IMAGE_UNITS_AND_FRAGMENT_OUTPUTS 0x8F39
|
||||
#define GL_IMAGE_BINDING_NAME 0x8F3A
|
||||
#define GL_IMAGE_BINDING_LEVEL 0x8F3B
|
||||
#define GL_IMAGE_BINDING_LAYERED 0x8F3C
|
||||
#define GL_IMAGE_BINDING_LAYER 0x8F3D
|
||||
#define GL_IMAGE_BINDING_ACCESS 0x8F3E
|
||||
#define GL_IMAGE_1D 0x904C
|
||||
#define GL_IMAGE_2D 0x904D
|
||||
#define GL_IMAGE_3D 0x904E
|
||||
#define GL_IMAGE_2D_RECT 0x904F
|
||||
#define GL_IMAGE_CUBE 0x9050
|
||||
#define GL_IMAGE_BUFFER 0x9051
|
||||
#define GL_IMAGE_1D_ARRAY 0x9052
|
||||
#define GL_IMAGE_2D_ARRAY 0x9053
|
||||
#define GL_IMAGE_CUBE_MAP_ARRAY 0x9054
|
||||
#define GL_IMAGE_2D_MULTISAMPLE 0x9055
|
||||
#define GL_IMAGE_2D_MULTISAMPLE_ARRAY 0x9056
|
||||
#define GL_INT_IMAGE_1D 0x9057
|
||||
#define GL_INT_IMAGE_2D 0x9058
|
||||
#define GL_INT_IMAGE_3D 0x9059
|
||||
#define GL_INT_IMAGE_2D_RECT 0x905A
|
||||
#define GL_INT_IMAGE_CUBE 0x905B
|
||||
#define GL_INT_IMAGE_BUFFER 0x905C
|
||||
#define GL_INT_IMAGE_1D_ARRAY 0x905D
|
||||
#define GL_INT_IMAGE_2D_ARRAY 0x905E
|
||||
#define GL_INT_IMAGE_CUBE_MAP_ARRAY 0x905F
|
||||
#define GL_INT_IMAGE_2D_MULTISAMPLE 0x9060
|
||||
#define GL_INT_IMAGE_2D_MULTISAMPLE_ARRAY 0x9061
|
||||
#define GL_UNSIGNED_INT_IMAGE_1D 0x9062
|
||||
#define GL_UNSIGNED_INT_IMAGE_2D 0x9063
|
||||
#define GL_UNSIGNED_INT_IMAGE_3D 0x9064
|
||||
#define GL_UNSIGNED_INT_IMAGE_2D_RECT 0x9065
|
||||
#define GL_UNSIGNED_INT_IMAGE_CUBE 0x9066
|
||||
#define GL_UNSIGNED_INT_IMAGE_BUFFER 0x9067
|
||||
#define GL_UNSIGNED_INT_IMAGE_1D_ARRAY 0x9068
|
||||
#define GL_UNSIGNED_INT_IMAGE_2D_ARRAY 0x9069
|
||||
#define GL_UNSIGNED_INT_IMAGE_CUBE_MAP_ARRAY 0x906A
|
||||
#define GL_UNSIGNED_INT_IMAGE_2D_MULTISAMPLE 0x906B
|
||||
#define GL_UNSIGNED_INT_IMAGE_2D_MULTISAMPLE_ARRAY 0x906C
|
||||
#define GL_MAX_IMAGE_SAMPLES 0x906D
|
||||
#define GL_IMAGE_BINDING_FORMAT 0x906E
|
||||
#define GL_IMAGE_FORMAT_COMPATIBILITY_TYPE 0x90C7
|
||||
#define GL_IMAGE_FORMAT_COMPATIBILITY_BY_SIZE 0x90C8
|
||||
#define GL_IMAGE_FORMAT_COMPATIBILITY_BY_CLASS 0x90C9
|
||||
#define GL_MAX_VERTEX_IMAGE_UNIFORMS 0x90CA
|
||||
#define GL_MAX_TESS_CONTROL_IMAGE_UNIFORMS 0x90CB
|
||||
#define GL_MAX_TESS_EVALUATION_IMAGE_UNIFORMS 0x90CC
|
||||
#define GL_MAX_GEOMETRY_IMAGE_UNIFORMS 0x90CD
|
||||
#define GL_MAX_FRAGMENT_IMAGE_UNIFORMS 0x90CE
|
||||
#define GL_MAX_COMBINED_IMAGE_UNIFORMS 0x90CF
|
||||
#endif
|
||||
|
||||
#ifndef GL_ARB_half_float_pixel
|
||||
#define GL_HALF_FLOAT 0x140B
|
||||
#endif
|
||||
|
||||
|
||||
|
||||
namespace osg {
|
||||
|
||||
|
||||
@@ -590,6 +654,37 @@ class OSG_EXPORT Texture : public osg::StateAttribute
|
||||
* min filter is used. */
|
||||
void allocateMipmapLevels();
|
||||
|
||||
/** Encapsulates texture image load/store attributes */
|
||||
struct ImageAttachment
|
||||
{
|
||||
GLuint unit;
|
||||
GLint level;
|
||||
GLboolean layered;
|
||||
GLint layer;
|
||||
GLenum access;
|
||||
GLenum format;
|
||||
|
||||
ImageAttachment()
|
||||
: unit(0), level(0), layered(GL_FALSE), layer(0), access(0), format(0) {}
|
||||
};
|
||||
|
||||
/** Type of access that will be performed on the texture image. */
|
||||
enum ImageAccess
|
||||
{
|
||||
NOT_USED = 0,
|
||||
READ_ONLY = GL_READ_ONLY_ARB,
|
||||
WRITE_ONLY = GL_WRITE_ONLY_ARB,
|
||||
READ_WRITE = GL_READ_WRITE_ARB
|
||||
};
|
||||
|
||||
/** Bind texture to an image unit (available only if GL version is 4.2 or greater)
|
||||
* The format parameter for the image unit need not exactly match the texture internal format,
|
||||
* but if it is set to 0, the texture internal format will be used.
|
||||
* See http://www.opengl.org/registry/specs/ARB/shader_image_load_store.txt */
|
||||
void bindToImageUnit(unsigned int unit, GLenum access, GLenum format=0, int level=0, bool layered=false, int layer=0);
|
||||
|
||||
ImageAttachment& getImageAttachment() { return _imageAttachment; }
|
||||
const ImageAttachment& getImageAttachment() const { return _imageAttachment; }
|
||||
|
||||
/** Sets GL_TEXTURE_COMPARE_MODE_ARB to GL_COMPARE_R_TO_TEXTURE_ARB
|
||||
* See http://oss.sgi.com/projects/ogl-sample/registry/ARB/shadow.txt. */
|
||||
@@ -746,6 +841,8 @@ class OSG_EXPORT Texture : public osg::StateAttribute
|
||||
void setTextureIntegerSupported(bool flag) { _isTextureIntegerEXTSupported=flag; }
|
||||
bool isTextureIntegerSupported() const { return _isTextureIntegerEXTSupported; }
|
||||
|
||||
bool isBindImageTextureSupported() const { return _glBindImageTexture!=0; }
|
||||
|
||||
void glCompressedTexImage2D(GLenum target, GLint level, GLenum internalformat, GLsizei width, GLsizei height, GLint border, GLsizei imageSize, const GLvoid *data) const
|
||||
{
|
||||
_glCompressedTexImage2D(target, level, internalformat, width, height, border, imageSize, data);
|
||||
@@ -776,6 +873,12 @@ class OSG_EXPORT Texture : public osg::StateAttribute
|
||||
_glTexParameterIuiv(target, pname, data);
|
||||
}
|
||||
|
||||
// ARB_shader_image_load_store
|
||||
void glBindImageTexture(GLuint unit, GLuint texture, GLint level, GLboolean layered, GLint layer, GLenum access, GLenum format) const
|
||||
{
|
||||
_glBindImageTexture(unit, texture, level, layered, layer, access, format);
|
||||
}
|
||||
|
||||
protected:
|
||||
|
||||
~Extensions() {}
|
||||
@@ -786,6 +889,7 @@ class OSG_EXPORT Texture : public osg::StateAttribute
|
||||
typedef void (GL_APIENTRY * TexImage2DMultisample)(GLenum target, GLsizei samples, GLint internalformat, GLsizei width, GLsizei height, GLboolean fixedsamplelocations);
|
||||
typedef void (GL_APIENTRY * TexParameterIivProc)(GLenum target, GLenum pname, const GLint* data);
|
||||
typedef void (GL_APIENTRY * TexParameterIuivProc)(GLenum target, GLenum pname, const GLuint* data);
|
||||
typedef void (GL_APIENTRY * BindImageTextureProc)( GLuint unit, GLuint texture, GLint level, GLboolean layered, GLint layer, GLenum access, GLenum format );
|
||||
|
||||
CompressedTexImage2DArbProc _glCompressedTexImage2D;
|
||||
CompressedTexSubImage2DArbProc _glCompressedTexSubImage2D;
|
||||
@@ -794,6 +898,8 @@ class OSG_EXPORT Texture : public osg::StateAttribute
|
||||
TexParameterIivProc _glTexParameterIiv;
|
||||
TexParameterIuivProc _glTexParameterIuiv;
|
||||
|
||||
// ARB_shader_image_load_store
|
||||
BindImageTextureProc _glBindImageTexture;
|
||||
|
||||
bool _isMultiTexturingSupported;
|
||||
bool _isTextureFilterAnisotropicSupported;
|
||||
@@ -938,6 +1044,8 @@ class OSG_EXPORT Texture : public osg::StateAttribute
|
||||
ShadowTextureMode _shadow_texture_mode;
|
||||
float _shadow_ambient;
|
||||
|
||||
ImageAttachment _imageAttachment;
|
||||
|
||||
public:
|
||||
|
||||
struct OSG_EXPORT TextureProfile
|
||||
|
||||
Reference in New Issue
Block a user