void
TraceAnalyzer::recordSideEffects(trace::Call *call)
{

    const char *name = call->name();

    /* FIXME: If we encode the list of commands that are executed
     * immediately (as opposed to those that are compiled into a
     * display list) then we could generate a "display-list-X"
     * resource just as we do for "texture-X" resources and only
     * emit it in the trace if a glCallList(X) is emitted. For
     * now, simply punt and include anything within glNewList and
     * glEndList in the trim output. This guarantees that display
     * lists will work, but does not trim out unused display
     * lists. */
    if (insideNewEndList != 0) {
        provide("state", call->no);

        /* Also, any texture bound inside a display list is
         * conservatively considered required. */
        if (strcmp(name, "glBindTexture") == 0) {
            GLuint texture = call->arg(1).toUInt();

            linkf("state", "texture-", texture);
        }

        return;
    }

    if (trimFlags & TRIM_FLAG_NO_SIDE_EFFECTS) {

        if (callHasNoSideEffects(call, name)) {
            return;
        }
    }

    if (trimFlags & TRIM_FLAG_TEXTURES) {
        
        if (recordTextureSideEffects(call, name)) {
            return;
        }
    }

    if (trimFlags & TRIM_FLAG_SHADERS) {

        if (recordShaderSideEffects(call, name)) {
            return;
        }
    }

    if (trimFlags & TRIM_FLAG_DRAWING) {

        if (recordDrawingSideEffects(call, name)) {
            return;
        }
    }

    /* By default, assume this call affects the state somehow. */
    resources["state"].insert(call->no);
}
Ejemplo n.º 2
0
int do_link(u8 * file , u8 * link){
	s8 ret;
	if (!check_for_args(LINK,2)) return 1;
	ret = linkf(file,link);
	if(ret == -1){
		vd_puts("CANNOT CREATE LINK\n");
				errormessage(geterror());
		}
		fl_clean();
	return 0;
}
bool
TraceAnalyzer::recordShaderSideEffects(trace::Call *call, const char *name)
{
    if (strcmp(name, "glCreateShader") == 0 ||
        strcmp(name, "glCreateShaderObjectARB") == 0) {

        GLuint shader = call->ret->toUInt();
        providef("shader-", shader, call->no);
        return true;
    }

    if (strcmp(name, "glShaderSource") == 0 ||
        strcmp(name, "glShaderSourceARB") == 0 ||
        strcmp(name, "glCompileShader") == 0 ||
        strcmp(name, "glCompileShaderARB") == 0 ||
        strcmp(name, "glGetShaderiv") == 0 ||
        strcmp(name, "glGetShaderInfoLog") == 0) {

        GLuint shader = call->arg(0).toUInt();
        providef("shader-", shader, call->no);
        return true;
    }

    if (strcmp(name, "glCreateProgram") == 0 ||
        strcmp(name, "glCreateProgramObjectARB") == 0) {

        GLuint program = call->ret->toUInt();
        providef("program-", program, call->no);
        return true;
    }

    if (strcmp(name, "glAttachShader") == 0 ||
        strcmp(name, "glAttachObjectARB") == 0) {

        GLuint program, shader;
        std::stringstream ss_program, ss_shader;

        program = call->arg(0).toUInt();
        shader = call->arg(1).toUInt();

        ss_program << "program-" << program;
        ss_shader << "shader-" << shader;

        link(ss_program.str(), ss_shader.str());
        provide(ss_program.str(), call->no);

        return true;
    }

    if (strcmp(name, "glDetachShader") == 0 ||
        strcmp(name, "glDetachObjectARB") == 0) {

        GLuint program, shader;
        std::stringstream ss_program, ss_shader;

        program = call->arg(0).toUInt();
        shader = call->arg(1).toUInt();

        ss_program << "program-" << program;
        ss_shader << "shader-" << shader;

        unlink(ss_program.str(), ss_shader.str());

        return true;
    }

    if (strcmp(name, "glUseProgram") == 0 ||
        strcmp(name, "glUseProgramObjectARB") == 0) {

        GLuint program;

        program = call->arg(0).toUInt();

        unlinkAll("render-program-state");

        if (program == 0) {
            unlink("render-state", "render-program-state");
            provide("state", call->no);
        } else {
            std::stringstream ss;

            ss << "program-" << program;

            link("render-state", "render-program-state");
            link("render-program-state", ss.str());

            provide(ss.str(), call->no);
        }

        return true;
    }

    if (strcmp(name, "glGetUniformLocation") == 0 ||
        strcmp(name, "glGetUniformLocationARB") == 0 ||
        strcmp(name, "glGetFragDataLocation") == 0 ||
        strcmp(name, "glGetFragDataLocationEXT") == 0 ||
        strcmp(name, "glGetSubroutineUniformLocation") == 0 ||
        strcmp(name, "glGetProgramResourceLocation") == 0 ||
        strcmp(name, "glGetProgramResourceLocationIndex") == 0 ||
        strcmp(name, "glGetVaryingLocationNV") == 0) {

        GLuint program = call->arg(0).toUInt();

        providef("program-", program, call->no);

        return true;
    }

    /* For any call that accepts 'location' as its first argument,
     * perform a lookup in our location->program map and add a
     * dependence on the program we find there. */
    if (call->sig->num_args > 0 &&
        strcmp(call->sig->arg_names[0], "location") == 0) {

        providef("program-", activeProgram, call->no);

        /* We can't easily tell if this uniform is being used to
         * associate a sampler in the shader with a texture
         * unit. The conservative option is to assume that it is
         * and create a link from the active program to any bound
         * textures for the given unit number.
         *
         * FIXME: We should be doing the same thing for calls to
         * glUniform1iv. */
        if (strcmp(name, "glUniform1i") == 0 ||
            strcmp(name, "glUniform1iARB") == 0) {

            GLint max_unit = MAX(GL_MAX_TEXTURE_COORDS, GL_MAX_COMBINED_TEXTURE_IMAGE_UNITS);

            GLint unit = call->arg(1).toSInt();
            std::stringstream ss_program;
            std::stringstream ss_texture;

            if (unit < max_unit) {

                ss_program << "program-" << activeProgram;

                ss_texture << "texture-unit-" << GL_TEXTURE0 + unit << "-target-";

                /* We don't know what target(s) might get bound to
                 * this texture unit, so conservatively link to
                 * all. Only bound textures will actually get inserted
                 * into the output call stream. */
                linkf(ss_program.str(), ss_texture.str(), GL_TEXTURE_1D);
                linkf(ss_program.str(), ss_texture.str(), GL_TEXTURE_2D);
                linkf(ss_program.str(), ss_texture.str(), GL_TEXTURE_3D);
                linkf(ss_program.str(), ss_texture.str(), GL_TEXTURE_CUBE_MAP);
            }
        }

        return true;
    }

    /* FIXME: We cut a huge swath by assuming that any unhandled
     * call that has a first argument named "program" should not
     * be included in the trimmed output unless the program of
     * that number is also included.
     *
     * This heuristic is correct for many cases, but we should
     * actually carefully verify if this includes some calls
     * inappropriately, or if it misses some.
     */
    if (strcmp(name, "glLinkProgram") == 0 ||
        strcmp(name, "glLinkProgramARB") == 0 ||
        (call->sig->num_args > 0 &&
         (strcmp(call->sig->arg_names[0], "program") == 0 ||
          strcmp(call->sig->arg_names[0], "programObj") == 0))) {

        GLuint program = call->arg(0).toUInt();
        providef("program-", program, call->no);
        return true;
    }

    /* No known shader-related side effects. Return false for more analysis. */
    return false;
}
bool
TraceAnalyzer::recordTextureSideEffects(trace::Call *call, const char *name)
{
    if (strcmp(name, "glGenTextures") == 0) {
        const trace::Array *textures = call->arg(1).toArray();
        size_t i;
        GLuint texture;

        if (textures) {
            for (i = 0; i < textures->size(); i++) {
                texture = textures->values[i]->toUInt();
                providef("texture-", texture, call->no);
            }
        }
        return true;
    }

    /* FIXME: When we start tracking framebuffer objects as their own
     * resources, we will want to link the FBO to the given texture
     * resource, (and to this call). For now, just link render state
     * to the texture, and force this call to be required. */
    if (strcmp(name, "glFramebufferTexture2D") == 0) {
        GLuint texture;

        texture = call->arg(3).toUInt();

        linkf("render-state", "texture-", texture);

        provide("state", call->no);
    }

    if (strcmp(name, "glBindTexture") == 0) {
        GLenum target;
        GLuint texture;

        std::stringstream ss_target, ss_texture;

        target = static_cast<GLenum>(call->arg(0).toSInt());
        texture = call->arg(1).toUInt();

        ss_target << "texture-unit-" << activeTextureUnit << "-target-" << target;
        ss_texture << "texture-" << texture;

        resources.erase(ss_target.str());
        provide(ss_target.str(), call->no);

        unlinkAll(ss_target.str());
        link(ss_target.str(), ss_texture.str());

        /* FIXME: This really shouldn't be necessary. The effect
         * this provide() has is that all glBindTexture calls will
         * be preserved in the output trace (never trimmed). Carl
         * has a trace ("btr") where a glBindTexture call should
         * not be necessary at all, (it's immediately followed
         * with a glBindTexture to a different texture and no
         * intervening texture-related calls), yet this 'provide'
         * makes the difference between a trim_stress test failing
         * and passing.
         *
         * More investigation is necessary, but for now, be
         * conservative and don't trim. */
        provide("state", call->no);

        return true;
    }

    /* FIXME: Need to handle glMultiTexImage and friends. */
    if (STRNCMP_LITERAL(name, "glTexImage") == 0 ||
        STRNCMP_LITERAL(name, "glTexSubImage") == 0 ||
        STRNCMP_LITERAL(name, "glCopyTexImage") == 0 ||
        STRNCMP_LITERAL(name, "glCopyTexSubImage") == 0 ||
        STRNCMP_LITERAL(name, "glCompressedTexImage") == 0 ||
        STRNCMP_LITERAL(name, "glCompressedTexSubImage") == 0 ||
        strcmp(name, "glInvalidateTexImage") == 0 ||
        strcmp(name, "glInvalidateTexSubImage") == 0) {

        std::set<unsigned> *calls;
        std::set<unsigned>::iterator c;
        std::stringstream ss_target, ss_texture;

        GLenum target = static_cast<GLenum>(call->arg(0).toSInt());

        ss_target << "texture-unit-" << activeTextureUnit << "-target-" << target;
        ss_texture << "texture-" << texture_map[target];

        /* The texture resource depends on this call and any calls
         * providing the given texture target. */
        provide(ss_texture.str(), call->no);

        if (resources.count(ss_target.str())) {
            calls = &resources[ss_target.str()];
            for (c = calls->begin(); c != calls->end(); c++) {
                provide(ss_texture.str(), *c);
            }
        }

        return true;
    }

    if (strcmp(name, "glEnable") == 0) {
        GLenum cap;

        cap = static_cast<GLenum>(call->arg(0).toSInt());

        if (cap == GL_TEXTURE_1D ||
            cap == GL_TEXTURE_2D ||
            cap == GL_TEXTURE_3D ||
            cap == GL_TEXTURE_CUBE_MAP)
        {
            std::stringstream ss;

            ss << "texture-unit-" << activeTextureUnit << "-target-" << cap;

            link("render-state", ss.str());
        }

        provide("state", call->no);
        return true;
    }

    if (strcmp(name, "glDisable") == 0) {
        GLenum cap;

        cap = static_cast<GLenum>(call->arg(0).toSInt());

        if (cap == GL_TEXTURE_1D ||
            cap == GL_TEXTURE_2D ||
            cap == GL_TEXTURE_3D ||
            cap == GL_TEXTURE_CUBE_MAP)
        {
            std::stringstream ss;

            ss << "texture-unit-" << activeTextureUnit << "-target-" << cap;

            unlink("render-state", ss.str());
        }

        provide("state", call->no);
        return true;
    }

    /* No known texture-related side effects. Return false for more analysis. */
    return false;
}
Ejemplo n.º 5
0
void
TraceAnalyzer::recordSideEffects(trace::Call *call)
{

    const char *name = call->name();

    /* Handle display lists before any other processing. */

    /* FIXME: If we encode the list of commands that are executed
     * immediately (as opposed to those that are compiled into a
     * display list) then we could generate a "display-list-X"
     * resource just as we do for "texture-X" resources and only
     * emit it in the trace if a glCallList(X) is emitted. For
     * now, simply punt and include anything within glNewList and
     * glEndList in the trim output. This guarantees that display
     * lists will work, but does not trim out unused display
     * lists. */
    if (insideNewEndList != 0) {
        provide("state", call->no);

        /* Also, any texture bound inside a display list is
         * conservatively considered required. */
        if (strcmp(name, "glBindTexture") == 0) {
            GLuint texture = call->arg(1).toUInt();

            linkf("state", "texture-", texture);
        }

        return;
    }

    /* If call is flagged as no side effects, then we are done here. */
    if (call->flags & trace::CALL_FLAG_NO_SIDE_EFFECTS) {
        return;
    }

    /* Similarly, swap-buffers calls don't have interesting side effects. */
    if (call->flags & trace::CALL_FLAG_SWAP_RENDERTARGET &&
            call->flags & trace::CALL_FLAG_END_FRAME) {
        return;
    }

    if (strcmp(name, "glGenTextures") == 0) {
        const trace::Array *textures = dynamic_cast<const trace::Array *>(&call->arg(1));
        size_t i;
        GLuint texture;

        if (textures) {
            for (i = 0; i < textures->size(); i++) {
                texture = textures->values[i]->toUInt();
                providef("texture-", texture, call->no);
            }
        }
        return;
    }

    /* FIXME: When we start tracking framebuffer objects as their own
     * resources, we will want to link the FBO to the given texture
     * resource, (and to this call). For now, just link render state
     * to the texture, and force this call to be required. */
    if (strcmp(name, "glFramebufferTexture2D") == 0) {
        GLuint texture;

        texture = call->arg(3).toUInt();

        linkf("render-state", "texture-", texture);

        required.insert(call->no);
    }

    if (strcmp(name, "glBindTexture") == 0) {
        GLenum target;
        GLuint texture;

        std::stringstream ss_target, ss_texture;

        target = static_cast<GLenum>(call->arg(0).toSInt());
        texture = call->arg(1).toUInt();

        ss_target << "texture-unit-" << activeTextureUnit << "-target-" << target;
        ss_texture << "texture-" << texture;

        resources.erase(ss_target.str());
        provide(ss_target.str(), call->no);

        unlinkAll(ss_target.str());
        link(ss_target.str(), ss_texture.str());

        /* FIXME: This really shouldn't be necessary. The effect
         * this provide() has is that all glBindTexture calls will
         * be preserved in the output trace (never trimmed). Carl
         * has a trace ("btr") where a glBindTexture call should
         * not be necessary at all, (it's immediately followed
         * with a glBindTexture to a different texture and no
         * intervening texture-related calls), yet this 'provide'
         * makes the difference between a trim_stress test failing
         * and passing.
         *
         * More investigation is necessary, but for now, be
         * conservative and don't trim. */
        provide("state", call->no);

        return;
    }

    /* FIXME: Need to handle glMultTetImage and friends. */
    if (STRNCMP_LITERAL(name, "glTexImage") == 0 ||
            STRNCMP_LITERAL(name, "glTexSubImage") == 0 ||
            STRNCMP_LITERAL(name, "glCopyTexImage") == 0 ||
            STRNCMP_LITERAL(name, "glCopyTexSubImage") == 0 ||
            STRNCMP_LITERAL(name, "glCompressedTexImage") == 0 ||
            STRNCMP_LITERAL(name, "glCompressedTexSubImage") == 0 ||
            strcmp(name, "glInvalidateTexImage") == 0 ||
            strcmp(name, "glInvalidateTexSubImage") == 0) {

        std::set<unsigned> *calls;
        std::set<unsigned>::iterator c;
        std::stringstream ss_target, ss_texture;

        GLenum target = static_cast<GLenum>(call->arg(0).toSInt());

        ss_target << "texture-unit-" << activeTextureUnit << "-target-" << target;
        ss_texture << "texture-" << texture_map[target];

        /* The texture resource depends on this call and any calls
         * providing the given texture target. */
        provide(ss_texture.str(), call->no);

        if (resources.count(ss_target.str())) {
            calls = &resources[ss_target.str()];
            for (c = calls->begin(); c != calls->end(); c++) {
                provide(ss_texture.str(), *c);
            }
        }

        return;
    }

    if (strcmp(name, "glEnable") == 0) {
        GLenum cap;

        cap = static_cast<GLenum>(call->arg(0).toSInt());

        if (cap == GL_TEXTURE_1D ||
                cap == GL_TEXTURE_2D ||
                cap == GL_TEXTURE_3D ||
                cap == GL_TEXTURE_CUBE_MAP)
        {
            std::stringstream ss;

            ss << "texture-unit-" << activeTextureUnit << "-target-" << cap;

            link("render-state", ss.str());
        }

        provide("state", call->no);
        return;
    }

    if (strcmp(name, "glDisable") == 0) {
        GLenum cap;

        cap = static_cast<GLenum>(call->arg(0).toSInt());

        if (cap == GL_TEXTURE_1D ||
                cap == GL_TEXTURE_2D ||
                cap == GL_TEXTURE_3D ||
                cap == GL_TEXTURE_CUBE_MAP)
        {
            std::stringstream ss;

            ss << "texture-unit-" << activeTextureUnit << "-target-" << cap;

            unlink("render-state", ss.str());
        }

        provide("state", call->no);
        return;
    }

    if (strcmp(name, "glCreateShader") == 0 ||
            strcmp(name, "glCreateShaderObjectARB") == 0) {

        GLuint shader = call->ret->toUInt();
        providef("shader-", shader, call->no);
        return;
    }

    if (strcmp(name, "glShaderSource") == 0 ||
            strcmp(name, "glShaderSourceARB") == 0 ||
            strcmp(name, "glCompileShader") == 0 ||
            strcmp(name, "glCompileShaderARB") == 0 ||
            strcmp(name, "glGetShaderiv") == 0 ||
            strcmp(name, "glGetShaderInfoLog") == 0) {

        GLuint shader = call->arg(0).toUInt();
        providef("shader-", shader, call->no);
        return;
    }

    if (strcmp(name, "glCreateProgram") == 0 ||
            strcmp(name, "glCreateProgramObjectARB") == 0) {

        GLuint program = call->ret->toUInt();
        providef("program-", program, call->no);
        return;
    }

    if (strcmp(name, "glAttachShader") == 0 ||
            strcmp(name, "glAttachObjectARB") == 0) {

        GLuint program, shader;
        std::stringstream ss_program, ss_shader;

        program = call->arg(0).toUInt();
        shader = call->arg(1).toUInt();

        ss_program << "program-" << program;
        ss_shader << "shader-" << shader;

        link(ss_program.str(), ss_shader.str());
        provide(ss_program.str(), call->no);

        return;
    }

    if (strcmp(name, "glDetachShader") == 0 ||
            strcmp(name, "glDetachObjectARB") == 0) {

        GLuint program, shader;
        std::stringstream ss_program, ss_shader;

        program = call->arg(0).toUInt();
        shader = call->arg(1).toUInt();

        ss_program << "program-" << program;
        ss_shader << "shader-" << shader;

        unlink(ss_program.str(), ss_shader.str());

        return;
    }

    if (strcmp(name, "glUseProgram") == 0 ||
            strcmp(name, "glUseProgramObjectARB") == 0) {

        GLuint program;

        program = call->arg(0).toUInt();

        unlinkAll("render-program-state");

        if (program == 0) {
            unlink("render-state", "render-program-state");
            provide("state", call->no);
        } else {
            std::stringstream ss;

            ss << "program-" << program;

            link("render-state", "render-program-state");
            link("render-program-state", ss.str());

            provide(ss.str(), call->no);
        }

        return;
    }

    if (strcmp(name, "glGetUniformLocation") == 0 ||
            strcmp(name, "glGetUniformLocationARB") == 0 ||
            strcmp(name, "glGetFragDataLocation") == 0 ||
            strcmp(name, "glGetFragDataLocationEXT") == 0 ||
            strcmp(name, "glGetSubroutineUniformLocation") == 0 ||
            strcmp(name, "glGetProgramResourceLocation") == 0 ||
            strcmp(name, "glGetProgramResourceLocationIndex") == 0 ||
            strcmp(name, "glGetVaryingLocationNV") == 0) {

        GLuint program = call->arg(0).toUInt();

        providef("program-", program, call->no);

        return;
    }

    /* For any call that accepts 'location' as its first argument,
     * perform a lookup in our location->program map and add a
     * dependence on the program we find there. */
    if (call->sig->num_args > 0 &&
            strcmp(call->sig->arg_names[0], "location") == 0) {

        providef("program-", activeProgram, call->no);

        /* We can't easily tell if this uniform is being used to
         * associate a sampler in the shader with a texture
         * unit. The conservative option is to assume that it is
         * and create a link from the active program to any bound
         * textures for the given unit number.
         *
         * FIXME: We should be doing the same thing for calls to
         * glUniform1iv. */
        if (strcmp(name, "glUniform1i") == 0 ||
                strcmp(name, "glUniform1iARB") == 0) {

            GLint max_unit = MAX(GL_MAX_TEXTURE_COORDS, GL_MAX_COMBINED_TEXTURE_IMAGE_UNITS);

            GLint unit = call->arg(1).toSInt();
            std::stringstream ss_program;
            std::stringstream ss_texture;

            if (unit < max_unit) {

                ss_program << "program-" << activeProgram;

                ss_texture << "texture-unit-" << GL_TEXTURE0 + unit << "-target-";

                /* We don't know what target(s) might get bound to
                 * this texture unit, so conservatively link to
                 * all. Only bound textures will actually get inserted
                 * into the output call stream. */
                linkf(ss_program.str(), ss_texture.str(), GL_TEXTURE_1D);
                linkf(ss_program.str(), ss_texture.str(), GL_TEXTURE_2D);
                linkf(ss_program.str(), ss_texture.str(), GL_TEXTURE_3D);
                linkf(ss_program.str(), ss_texture.str(), GL_TEXTURE_CUBE_MAP);
            }
        }

        return;
    }

    /* FIXME: We cut a huge swath by assuming that any unhandled
     * call that has a first argument named "program" should not
     * be included in the trimmed output unless the program of
     * that number is also included.
     *
     * This heuristic is correct for many cases, but we should
     * actually carefully verify if this includes some calls
     * inappropriately, or if it misses some.
     */
    if (strcmp(name, "glLinkProgram") == 0 ||
            strcmp(name, "glLinkProgramARB") == 0 ||
            (call->sig->num_args > 0 &&
             (strcmp(call->sig->arg_names[0], "program") == 0 ||
              strcmp(call->sig->arg_names[0], "programObj") == 0))) {

        GLuint program = call->arg(0).toUInt();
        providef("program-", program, call->no);
        return;
    }

    /* Handle all rendering operations, (even though only glEnd is
     * flagged as a rendering operation we treat everything from
     * glBegin through glEnd as a rendering operation). */
    if (call->flags & trace::CALL_FLAG_RENDER ||
            insideBeginEnd) {

        std::set<unsigned> calls;
        std::set<unsigned>::iterator c;

        provide("framebuffer", call->no);

        calls = resolve("render-state");

        for (c = calls.begin(); c != calls.end(); c++) {
            provide("framebuffer", *c);
        }

        /* In some cases, rendering has side effects beyond the
         * framebuffer update. */
        if (renderingHasSideEffect()) {
            provide("state", call->no);
            for (c = calls.begin(); c != calls.end(); c++) {
                provide("state", *c);
            }
        }

        return;
    }

    /* By default, assume this call affects the state somehow. */
    resources["state"].insert(call->no);
}