/* Assignment operator overloading. We signal to
* the references counter that we release the
* previous address and hold the new one */
SmartPointer<T>& operator=(SmartPointer<T> sp) {
unuse();
pointer = sp.pointer;
use();
return *this;
}
Program::~Program()
{
if (m_Id == m_usedProgramId) {
//if this program is used take care of "unusing"
unuse();
}
//detach shaders
detachAllShaders();
//finally delete
glDeleteProgram(m_Id);
}
void Batch3DGeometry::render_batch() {
//Vertices data
glBindBuffer(GL_ARRAY_BUFFER, m_vbo);
glBufferData(GL_ARRAY_BUFFER, vertices.size() * sizeof(Vertex3D), vertices.data(), GL_STATIC_DRAW);
//Indices data
glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, m_index_vbo);
glBufferData(GL_ELEMENT_ARRAY_BUFFER, indices.size() * sizeof(U32), indices.data(), GL_STATIC_DRAW);
Shader* old_shader = nullptr;
U64 buffer_data_offset = 0;
for (const auto& rb : m_render_batches){
auto curr_shader = rb.model->material.shader;
curr_shader->use();
U32 old_shader_size = 0;//keep track of the size of the last used shader
if (old_shader)old_shader_size = old_shader->attributes_struct_size;
//If the new batch has a different shader than the one before,
//enable attributes and send global data
if (old_shader != curr_shader){
old_shader = curr_shader;
enable_attributes(curr_shader);
U8 curr_samp = 0;
//Send global shader uniforms
for (const auto& cnt : curr_shader->global_uniforms_data){
curr_shader->send_global_uniform_to_shader(cnt.first, cnt.second.value, &curr_samp);
}
}
U8 current_sampler = 0;
for (const auto& cnt : rb.model->material.uniform_map)
{
curr_shader->send_material_uniform_to_shader(cnt.first, cnt.second, ¤t_sampler);
}
glBindBuffer(GL_ARRAY_BUFFER, m_i_vbo);
glBufferData(GL_ARRAY_BUFFER, rb.num_instances * curr_shader->attributes_struct_size, (void*)((U64)input_instances + buffer_data_offset ), GL_STATIC_DRAW);
glBindBuffer(GL_ARRAY_BUFFER, 0);
buffer_data_offset += rb.num_instances* curr_shader->attributes_struct_size;
U32 offset = rb.indices_offset * sizeof(U32);
glDrawElementsInstancedBaseVertex(GL_TRIANGLES, rb.num_indices, GL_UNSIGNED_INT, (void*)offset, rb.num_instances, rb.vertex_offset);
curr_shader->unuse();
}
}
void Lightmap::render(const Falltergeist::Point &pos)
{
if (_indexes<=0) return;
GL_CHECK(glBlendFunc(GL_DST_COLOR, GL_SRC_COLOR));
auto shader = ResourceManager::getInstance()->shader("lightmap");
GL_CHECK(shader->use());
GL_CHECK(shader->setUniform(_uniformMVP, Game::getInstance()->renderer()->getMVP()));
// set camera offset
GL_CHECK(shader->setUniform(_uniformOffset, glm::vec2((float)pos.x(), (float)pos.y()) ));
GL_CHECK(shader->setUniform(_uniformFade,Game::getInstance()->renderer()->fadeColor()));
GL_CHECK(glBindVertexArray(_vao));
GL_CHECK(glBindBuffer(GL_ARRAY_BUFFER, _coords));
GL_CHECK(glVertexAttribPointer(_attribPos, 2, GL_FLOAT, GL_FALSE, 0, (void*)0 ));
GL_CHECK(glBindBuffer(GL_ARRAY_BUFFER, _lights));
GL_CHECK(glVertexAttribPointer(_attribLights, 1, GL_FLOAT, GL_FALSE, 0, (void*)0 ));
GL_CHECK(glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, _ebo));
GL_CHECK(glEnableVertexAttribArray(_attribPos));
GL_CHECK(glEnableVertexAttribArray(_attribLights));
GL_CHECK(glDrawElements(GL_TRIANGLES, _indexes, GL_UNSIGNED_INT, 0 ));
GL_CHECK(glDisableVertexAttribArray(_attribPos));
GL_CHECK(glDisableVertexAttribArray(_attribLights));
GL_CHECK(glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0));
GL_CHECK(glBindBuffer(GL_ARRAY_BUFFER, 0));
GL_CHECK(glBindVertexArray(0));
GL_CHECK(shader->unuse());
GL_CHECK(glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA));
}
bool StGLStereoFrameBuffer::StGLStereoProgram::link(StGLContext& theCtx) {
if(!StGLProgram::link(theCtx)) {
return false;
}
StGLVarLocation uniTexLLoc = StGLProgram::getUniformLocation(theCtx, "texL");
StGLVarLocation uniTexRLoc = StGLProgram::getUniformLocation(theCtx, "texR");
atrVVertexLoc = StGLProgram::getAttribLocation(theCtx, "vVertex");
atrVTexCoordLoc = StGLProgram::getAttribLocation(theCtx, "vTexCoord");
if(uniTexLLoc.isValid() && uniTexRLoc.isValid()) {
use(theCtx);
theCtx.core20fwd->glUniform1i(uniTexLLoc, StGLProgram::TEXTURE_SAMPLE_0); // GL_TEXTURE0 in multitexture
theCtx.core20fwd->glUniform1i(uniTexRLoc, StGLProgram::TEXTURE_SAMPLE_1); // GL_TEXTURE1 in multitexture
unuse(theCtx);
}
return uniTexRLoc.isValid() && uniTexLLoc.isValid() && atrVVertexLoc.isValid() && atrVTexCoordLoc.isValid();
}
/* Signal to the references counter that we
* release the current address */
~SmartPointer() {
unuse();
}
/**
* Find instruction src's which are mov's that can be collapsed, replacing
* the mov dst with the mov src
*/
static void
instr_cp(struct ir3_cp_ctx *ctx, struct ir3_instruction *instr)
{
struct ir3_register *reg;
if (instr->regs_count == 0)
return;
if (ir3_instr_check_mark(instr))
return;
/* walk down the graph from each src: */
foreach_src_n(reg, n, instr) {
struct ir3_instruction *src = ssa(reg);
if (!src)
continue;
instr_cp(ctx, src);
/* TODO non-indirect access we could figure out which register
* we actually want and allow cp..
*/
if (reg->flags & IR3_REG_ARRAY)
continue;
/* Don't CP absneg into meta instructions, that won't end well: */
if (is_meta(instr) && (src->opc != OPC_MOV))
continue;
reg_cp(ctx, instr, reg, n);
}
if (instr->regs[0]->flags & IR3_REG_ARRAY) {
struct ir3_instruction *src = ssa(instr->regs[0]);
if (src)
instr_cp(ctx, src);
}
if (instr->address) {
instr_cp(ctx, instr->address);
ir3_instr_set_address(instr, eliminate_output_mov(instr->address));
}
/* we can end up with extra cmps.s from frontend, which uses a
*
* cmps.s p0.x, cond, 0
*
* as a way to mov into the predicate register. But frequently 'cond'
* is itself a cmps.s/cmps.f/cmps.u. So detect this special case and
* just re-write the instruction writing predicate register to get rid
* of the double cmps.
*/
if ((instr->opc == OPC_CMPS_S) &&
(instr->regs[0]->num == regid(REG_P0, 0)) &&
ssa(instr->regs[1]) &&
(instr->regs[2]->flags & IR3_REG_IMMED) &&
(instr->regs[2]->iim_val == 0)) {
struct ir3_instruction *cond = ssa(instr->regs[1]);
switch (cond->opc) {
case OPC_CMPS_S:
case OPC_CMPS_F:
case OPC_CMPS_U:
instr->opc = cond->opc;
instr->flags = cond->flags;
instr->cat2 = cond->cat2;
instr->address = cond->address;
instr->regs[1] = cond->regs[1];
instr->regs[2] = cond->regs[2];
instr->barrier_class |= cond->barrier_class;
instr->barrier_conflict |= cond->barrier_conflict;
unuse(cond);
break;
default:
break;
}
}
/* Handle converting a sam.s2en (taking samp/tex idx params via
* register) into a normal sam (encoding immediate samp/tex idx)
* if they are immediate. This saves some instructions and regs
* in the common case where we know samp/tex at compile time:
*/
if (is_tex(instr) && (instr->flags & IR3_INSTR_S2EN) &&
!(ir3_shader_debug & IR3_DBG_FORCES2EN)) {
/* The first src will be a fan-in (collect), if both of it's
* two sources are mov from imm, then we can
*/
struct ir3_instruction *samp_tex = ssa(instr->regs[1]);
debug_assert(samp_tex->opc == OPC_META_FI);
struct ir3_instruction *samp = ssa(samp_tex->regs[1]);
struct ir3_instruction *tex = ssa(samp_tex->regs[2]);
if ((samp->opc == OPC_MOV) &&
(samp->regs[1]->flags & IR3_REG_IMMED) &&
(tex->opc == OPC_MOV) &&
(tex->regs[1]->flags & IR3_REG_IMMED)) {
instr->flags &= ~IR3_INSTR_S2EN;
instr->cat5.samp = samp->regs[1]->iim_val;
instr->cat5.tex = tex->regs[1]->iim_val;
instr->regs[1]->instr = NULL;
}
}
}
/**
* Handle cp for a given src register. This additionally handles
* the cases of collapsing immedate/const (which replace the src
* register with a non-ssa src) or collapsing mov's from relative
* src (which needs to also fixup the address src reference by the
* instruction).
*/
static void
reg_cp(struct ir3_cp_ctx *ctx, struct ir3_instruction *instr,
struct ir3_register *reg, unsigned n)
{
struct ir3_instruction *src = ssa(reg);
if (is_eligible_mov(src, true)) {
/* simple case, no immed/const/relativ, only mov's w/ ssa src: */
struct ir3_register *src_reg = src->regs[1];
unsigned new_flags = reg->flags;
combine_flags(&new_flags, src);
if (valid_flags(instr, n, new_flags)) {
if (new_flags & IR3_REG_ARRAY) {
debug_assert(!(reg->flags & IR3_REG_ARRAY));
reg->array = src_reg->array;
}
reg->flags = new_flags;
reg->instr = ssa(src_reg);
instr->barrier_class |= src->barrier_class;
instr->barrier_conflict |= src->barrier_conflict;
unuse(src);
reg->instr->use_count++;
}
} else if (is_same_type_mov(src) &&
/* cannot collapse const/immed/etc into meta instrs: */
!is_meta(instr)) {
/* immed/const/etc cases, which require some special handling: */
struct ir3_register *src_reg = src->regs[1];
unsigned new_flags = reg->flags;
combine_flags(&new_flags, src);
if (!valid_flags(instr, n, new_flags)) {
/* See if lowering an immediate to const would help. */
if (valid_flags(instr, n, (new_flags & ~IR3_REG_IMMED) | IR3_REG_CONST)) {
debug_assert(new_flags & IR3_REG_IMMED);
instr->regs[n + 1] = lower_immed(ctx, src_reg, new_flags);
return;
}
/* special case for "normal" mad instructions, we can
* try swapping the first two args if that fits better.
*
* the "plain" MAD's (ie. the ones that don't shift first
* src prior to multiply) can swap their first two srcs if
* src[0] is !CONST and src[1] is CONST:
*/
if ((n == 1) && is_mad(instr->opc) &&
!(instr->regs[0 + 1]->flags & (IR3_REG_CONST | IR3_REG_RELATIV)) &&
valid_flags(instr, 0, new_flags & ~IR3_REG_IMMED)) {
/* swap src[0] and src[1]: */
struct ir3_register *tmp;
tmp = instr->regs[0 + 1];
instr->regs[0 + 1] = instr->regs[1 + 1];
instr->regs[1 + 1] = tmp;
n = 0;
} else {
return;
}
}
/* Here we handle the special case of mov from
* CONST and/or RELATIV. These need to be handled
* specially, because in the case of move from CONST
* there is no src ir3_instruction so we need to
* replace the ir3_register. And in the case of
* RELATIV we need to handle the address register
* dependency.
*/
if (src_reg->flags & IR3_REG_CONST) {
/* an instruction cannot reference two different
* address registers:
*/
if ((src_reg->flags & IR3_REG_RELATIV) &&
conflicts(instr->address, reg->instr->address))
return;
/* This seems to be a hw bug, or something where the timings
* just somehow don't work out. This restriction may only
* apply if the first src is also CONST.
*/
if ((opc_cat(instr->opc) == 3) && (n == 2) &&
(src_reg->flags & IR3_REG_RELATIV) &&
(src_reg->array.offset == 0))
return;
src_reg = ir3_reg_clone(instr->block->shader, src_reg);
src_reg->flags = new_flags;
instr->regs[n+1] = src_reg;
if (src_reg->flags & IR3_REG_RELATIV)
ir3_instr_set_address(instr, reg->instr->address);
return;
}
if ((src_reg->flags & IR3_REG_RELATIV) &&
!conflicts(instr->address, reg->instr->address)) {
src_reg = ir3_reg_clone(instr->block->shader, src_reg);
src_reg->flags = new_flags;
instr->regs[n+1] = src_reg;
ir3_instr_set_address(instr, reg->instr->address);
return;
}
/* NOTE: seems we can only do immed integers, so don't
* need to care about float. But we do need to handle
* abs/neg *before* checking that the immediate requires
* few enough bits to encode:
*
* TODO: do we need to do something to avoid accidentally
* catching a float immed?
*/
if (src_reg->flags & IR3_REG_IMMED) {
int32_t iim_val = src_reg->iim_val;
debug_assert((opc_cat(instr->opc) == 1) ||
(opc_cat(instr->opc) == 6) ||
ir3_cat2_int(instr->opc) ||
(is_mad(instr->opc) && (n == 0)));
if (new_flags & IR3_REG_SABS)
iim_val = abs(iim_val);
if (new_flags & IR3_REG_SNEG)
iim_val = -iim_val;
if (new_flags & IR3_REG_BNOT)
iim_val = ~iim_val;
/* other than category 1 (mov) we can only encode up to 10 bits: */
if ((instr->opc == OPC_MOV) ||
!((iim_val & ~0x3ff) && (-iim_val & ~0x3ff))) {
new_flags &= ~(IR3_REG_SABS | IR3_REG_SNEG | IR3_REG_BNOT);
src_reg = ir3_reg_clone(instr->block->shader, src_reg);
src_reg->flags = new_flags;
src_reg->iim_val = iim_val;
instr->regs[n+1] = src_reg;
} else if (valid_flags(instr, n, (new_flags & ~IR3_REG_IMMED) | IR3_REG_CONST)) {
/* See if lowering an immediate to const would help. */
instr->regs[n+1] = lower_immed(ctx, src_reg, new_flags);
}
return;
}
}
}
std::shared_ptr<Texture> RenderPass::
get_buffer(std::string const& name, CameraMode mode, bool draw_fps) {
// check for existance of desired buffer
if ((mode == CENTER
&& center_eye_buffers_.find(name) == center_eye_buffers_.end())
|| (mode == LEFT
&& left_eye_buffers_.find(name) == left_eye_buffers_.end())
|| (mode == RIGHT
&& right_eye_buffers_.find(name) == right_eye_buffers_.end())) {
WARNING("Failed to get buffer \"%s\" from pass \"%s\": "
"A buffer with this name does not exist!",
name.c_str(), get_name().c_str());
return NULL;
}
// return appropriate buffer if it has been rendered already
if (mode == CENTER && rendererd_center_eye_)
return center_eye_buffers_[name];
if (mode == LEFT && rendererd_left_eye_)
return left_eye_buffers_[name];
if (mode == RIGHT && rendererd_right_eye_)
return right_eye_buffers_[name];
// serialize the scenegraph
Optimizer optimizer;
optimizer.check(pipeline_->get_current_graph(), render_mask_);
// if there are dynamic texture inputs for this render pass, get the
// according buffers recursively
for (auto& node: optimizer.get_data().nodes_) {
auto material(inputs_.find(node.material_));
if (material != inputs_.end()) {
for (auto& uniform: material->second) {
overwrite_uniform_texture(material->first, uniform.first,
pipeline_->get_render_pass(uniform.second.first)->
get_buffer(uniform.second.second, mode));
}
}
}
// we'll need these two very often now...
OptimizedScene const& scene(optimizer.get_data());
RenderContext const& ctx(pipeline_->get_context());
// get the fbo which should be rendered to
FrameBufferObject* fbo(NULL);
switch (mode) {
case CENTER:
fbo = ¢er_eye_fbo_;
break;
case LEFT:
fbo = &left_eye_fbo_;
break;
case RIGHT:
fbo = &right_eye_fbo_;
break;
}
fbo->bind(ctx);
fbo->clear_color_buffers(ctx);
fbo->clear_depth_stencil_buffer(ctx);
ctx.render_context->set_viewport(scm::gl::viewport(math::vec2(0,0),
math::vec2(fbo->width(),
fbo->height())));
auto camera_it(scene.cameras_.find(camera_));
auto screen_it(scene.screens_.find(screen_));
if (camera_it != scene.cameras_.end()
&& screen_it != scene.screens_.end()) {
auto camera(camera_it->second);
auto screen(screen_it->second);
math::mat4 camera_transform(camera.transform_);
if (mode == LEFT) {
scm::math::translate(camera_transform,
-camera.stereo_width_*0.5f, 0.f, 0.f);
} else if (mode == RIGHT) {
scm::math::translate(camera_transform,
camera.stereo_width_*0.5f, 0.f, 0.f);
}
auto projection(math::compute_frustum(camera_transform.column(3),
screen.transform_,
0.1, 100000.f));
math::mat4 view_transform(screen.transform_);
view_transform[12] = 0.f;
view_transform[13] = 0.f;
view_transform[14] = 0.f;
view_transform[15] = 1.f;
math::vec3 camera_position(camera_transform.column(3)[0],
camera_transform.column(3)[1],
camera_transform.column(3)[2]);
view_transform = scm::math::make_translation(camera_position)
* view_transform;
math::mat4 view_matrix(scm::math::inverse(view_transform));
// update light data uniform block
if (scene.lights_.size() > 0) {
if (!light_information_) {
light_information_ =
new scm::gl::uniform_block<LightInformation>(ctx.render_device);
}
light_information_->begin_manipulation(ctx.render_context);
LightInformation light;
light.light_count = math::vec4i(scene.lights_.size(),
scene.lights_.size(),
scene.lights_.size(),
scene.lights_.size());
for (unsigned i(0); i < scene.lights_.size(); ++i) {
math::mat4 transform(scene.lights_[i].transform_);
// calc light radius and position
light.position[i] = math::vec4(transform[12], transform[13],
transform[14], transform[15]);
float radius = scm::math::length(light.position[i] - transform
* math::vec4(0.f, 0.f, 1.f, 1.f));
light.color_radius[i] = math::vec4(scene.lights_[i].color_.r(),
scene.lights_[i].color_.g(),
scene.lights_[i].color_.b(),
radius);
}
**light_information_ = light;
light_information_->end_manipulation();
ctx.render_context->bind_uniform_buffer(
light_information_->block_buffer(), 0);
}
for (auto& core: scene.nodes_) {
auto geometry = GeometryBase::instance()->get(core.geometry_);
auto material = MaterialBase::instance()->get(core.material_);
if (material && geometry) {
material->use(ctx);
if (float_uniforms_.find(core.material_)
!= float_uniforms_.end()) {
for (auto val : float_uniforms_[core.material_])
material->get_shader()->set_float(ctx, val.first,
val.second);
}
if (texture_uniforms_.find(core.material_)
!= texture_uniforms_.end()) {
for (auto val : texture_uniforms_[core.material_])
material->get_shader()->set_sampler2D(ctx, val.first,
*val.second);
}
material->get_shader()->set_mat4(ctx, "projection_matrix",
projection);
material->get_shader()->set_mat4(ctx, "view_matrix",
view_matrix);
material->get_shader()->set_mat4(ctx, "model_matrix",
core.transform_);
material->get_shader()->set_mat4(ctx, "normal_matrix",
scm::math::transpose(
scm::math::inverse(core.transform_)));
geometry->draw(ctx);
material->unuse(ctx);
} else if (material) {
WARNING("Cannot render geometry \"%s\": Undefined geometry "
"name!", core.geometry_.c_str());
} else if (geometry) {
WARNING("Cannot render geometry \"%s\": Undefined material "
"name: \"%s\"!", core.geometry_.c_str(),
core.material_.c_str());
} else {
WARNING("Cannot render geometry \"%s\": Undefined geometry "
"and material name: \"%s\"!", core.geometry_.c_str(),
core.material_.c_str());
}
}
if (scene.lights_.size() > 0) {
ctx.render_context->reset_uniform_buffers();
}
}
fbo->unbind(ctx);
// draw fps on the screen
if (draw_fps) {
if (!text_renderer_)
text_renderer_ = new TextRenderer(pipeline_->get_context());
if (mode == CENTER) {
text_renderer_->render_fps(pipeline_->get_context(), center_eye_fbo_,
pipeline_->get_application_fps(),
pipeline_->get_rendering_fps());
} else if (mode == LEFT) {
text_renderer_->render_fps(pipeline_->get_context(), left_eye_fbo_,
pipeline_->get_application_fps(),
pipeline_->get_rendering_fps());
} else {
text_renderer_->render_fps(pipeline_->get_context(), right_eye_fbo_,
pipeline_->get_application_fps(),
pipeline_->get_rendering_fps());
}
}
// return the buffer and set the already-rendered-flag
if (mode == CENTER) {
rendererd_center_eye_ = true;
return center_eye_buffers_[name];
} else if (mode == LEFT) {
rendererd_left_eye_ = true;
return left_eye_buffers_[name];
} else {
rendererd_right_eye_ = true;
return right_eye_buffers_[name];
}
}
void Shader::disable(void) { unuse(); }
int main_w2(int base_in) {
// use the input number to set the global variable
_base_ = base_in;
// see http://stackoverflow.com/questions/6792235/declaring-global-variable-array-inside-a-function-in-c
// http://forum.codecall.net/topic/51010-dynamic-arrays-using-malloc-and-realloc/
used_number = (int *)malloc(sizeof(int*) * _base_); // seems need (int *); should not be under c99
if (used_number ==NULL) {
printf("Error allocating memory!\n"); //print an error message
return 1; //return with failure
}
// used_number[0] = 90; <-- remember to set it back to 0 if used
// used_number[_base_ - 1] = 99;
// test_used_number(0);
// test_used_number(_base_ - 1);
// used_number[0] = 0;
// used_number[_base_ - 1] = 0;
clock_t tick_l;
int cnt_l = 0;
int ab, cd, ef, gh;
tick_l = clock();
reset_w2();
for (ab = START_WITH_2_ALWAYS * _base_; ab < (_base_ * _base_); ab++)
// a cannot be 0 and 1, start from 20
if (use(ab)) { //try to use ab
for (cd = START_WITH_2_ALWAYS * _base_; cd < ab; cd++)
// ab > cd to avoid ef < 0 and cd run into ab
if (use(cd)) {
ef = ab - cd; // you do not need to loop over ef and gh
gh = (1 * (_base_ * _base_ + _base_ + 1)) - ef;
// a - b = c + d = p = 111
// fixed a and b then c and d is fixed by logic
if (use(ef)) {
if (use(gh)) {
if (cnt_l < MAX_SOL_TO_PRINT) {
printf("%d: %d - %d = %d, %d + %d = %d\n", cnt_l + 1, ab, cd, ef, ef, gh, (1 * (_base_ * _base_ + _base_ + 1)) );
}
cnt_l++; // need that for those time you cannot print all
unuse(gh);
}
unuse(ef);
}
unuse(cd);
}
unuse(ab); // if used(ab) unuse it
}
tick_l = clock()-tick_l;
printf("\n_base_ is: %d\n", _base_);
printf("time used: %f\n", ((double)tick_l)/CLOCKS_PER_SEC);
// printf("MAX_SOL_TO_PRINT: %d\n\n", MAX_SOL_TO_PRINT);
printf("number of solutions: %d\n\n", cnt_l);
free(used_number);
return 0;
}
