obs-studio/libobs/graphics/graphics.c

2696 lines
61 KiB
C

/******************************************************************************
Copyright (C) 2013 by Hugh Bailey <obs.jim@gmail.com>
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
******************************************************************************/
#include <assert.h>
#include "../util/base.h"
#include "../util/bmem.h"
#include "../util/platform.h"
#include "graphics-internal.h"
#include "vec2.h"
#include "vec3.h"
#include "quat.h"
#include "axisang.h"
#include "effect-parser.h"
#include "effect.h"
static THREAD_LOCAL graphics_t *thread_graphics = NULL;
static inline bool gs_obj_valid(const void *obj, const char *f,
const char *name)
{
if (!obj) {
blog(LOG_DEBUG, "%s: Null '%s' parameter", f, name);
return false;
}
return true;
}
static inline bool gs_valid(const char *f)
{
if (!thread_graphics) {
blog(LOG_DEBUG, "%s: called while not in a graphics context",
f);
return false;
}
return true;
}
#define ptr_valid(ptr, func) gs_obj_valid(ptr, func, #ptr)
#define gs_valid_p(func, param1) \
(gs_valid(func) && ptr_valid(param1, func))
#define gs_valid_p2(func, param1, param2) \
(gs_valid(func) && ptr_valid(param1, func) \
&& ptr_valid(param2, func))
#define gs_valid_p3(func, param1, param2, param3) \
(gs_valid(func) && ptr_valid(param1, func) \
&& ptr_valid(param2, func) && ptr_valid(param3, func))
#define IMMEDIATE_COUNT 512
void gs_enum_adapters(
bool (*callback)(void *param, const char *name, uint32_t id),
void *param)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_enum_adapters", callback))
return;
if (graphics->exports.device_enum_adapters) {
if (graphics->exports.device_enum_adapters(callback, param)) {
return;
}
}
/* If the subsystem does not currently support device enumeration of
* adapters or fails to enumerate adapters, just set it to one adapter
* named "Default" */
callback(param, "Default", 0);
}
extern void gs_init_image_deps(void);
extern void gs_free_image_deps(void);
bool load_graphics_imports(struct gs_exports *exports, void *module,
const char *module_name);
static bool graphics_init_immediate_vb(struct graphics_subsystem *graphics)
{
struct gs_vb_data *vbd;
vbd = gs_vbdata_create();
vbd->num = IMMEDIATE_COUNT;
vbd->points = bmalloc(sizeof(struct vec3)*IMMEDIATE_COUNT);
vbd->normals = bmalloc(sizeof(struct vec3)*IMMEDIATE_COUNT);
vbd->colors = bmalloc(sizeof(uint32_t) *IMMEDIATE_COUNT);
vbd->num_tex = 1;
vbd->tvarray = bmalloc(sizeof(struct gs_tvertarray));
vbd->tvarray[0].width = 2;
vbd->tvarray[0].array =
bmalloc(sizeof(struct vec2) * IMMEDIATE_COUNT);
graphics->immediate_vertbuffer = graphics->exports.
device_vertexbuffer_create(graphics->device, vbd, GS_DYNAMIC);
if (!graphics->immediate_vertbuffer)
return false;
return true;
}
static bool graphics_init_sprite_vb(struct graphics_subsystem *graphics)
{
struct gs_vb_data *vbd;
vbd = gs_vbdata_create();
vbd->num = 4;
vbd->points = bmalloc(sizeof(struct vec3) * 4);
vbd->num_tex = 1;
vbd->tvarray = bmalloc(sizeof(struct gs_tvertarray));
vbd->tvarray[0].width = 2;
vbd->tvarray[0].array = bmalloc(sizeof(struct vec2) * 4);
memset(vbd->points, 0, sizeof(struct vec3) * 4);
memset(vbd->tvarray[0].array, 0, sizeof(struct vec2) * 4);
graphics->sprite_buffer = graphics->exports.
device_vertexbuffer_create(graphics->device, vbd, GS_DYNAMIC);
if (!graphics->sprite_buffer)
return false;
return true;
}
static bool graphics_init(struct graphics_subsystem *graphics)
{
struct matrix4 top_mat;
matrix4_identity(&top_mat);
da_push_back(graphics->matrix_stack, &top_mat);
graphics->exports.device_enter_context(graphics->device);
if (!graphics_init_immediate_vb(graphics))
return false;
if (!graphics_init_sprite_vb(graphics))
return false;
if (pthread_mutex_init(&graphics->mutex, NULL) != 0)
return false;
if (pthread_mutex_init(&graphics->effect_mutex, NULL) != 0)
return false;
graphics->exports.device_blend_function_separate(graphics->device,
GS_BLEND_SRCALPHA, GS_BLEND_INVSRCALPHA,
GS_BLEND_ONE, GS_BLEND_ONE);
graphics->cur_blend_state.enabled = true;
graphics->cur_blend_state.src_c = GS_BLEND_SRCALPHA;
graphics->cur_blend_state.dest_c = GS_BLEND_INVSRCALPHA;
graphics->cur_blend_state.src_a = GS_BLEND_ONE;
graphics->cur_blend_state.dest_a = GS_BLEND_ONE;
graphics->exports.device_leave_context(graphics->device);
gs_init_image_deps();
return true;
}
int gs_create(graphics_t **pgraphics, const char *module, uint32_t adapter)
{
int errcode = GS_ERROR_FAIL;
graphics_t *graphics = bzalloc(sizeof(struct graphics_subsystem));
pthread_mutex_init_value(&graphics->mutex);
pthread_mutex_init_value(&graphics->effect_mutex);
graphics->module = os_dlopen(module);
if (!graphics->module) {
errcode = GS_ERROR_MODULE_NOT_FOUND;
goto error;
}
if (!load_graphics_imports(&graphics->exports, graphics->module,
module))
goto error;
errcode = graphics->exports.device_create(&graphics->device, adapter);
if (errcode != GS_SUCCESS)
goto error;
if (!graphics_init(graphics)) {
errcode = GS_ERROR_FAIL;
goto error;
}
*pgraphics = graphics;
return errcode;
error:
gs_destroy(graphics);
return errcode;
}
extern void gs_effect_actually_destroy(gs_effect_t *effect);
void gs_destroy(graphics_t *graphics)
{
if (!ptr_valid(graphics, "gs_destroy"))
return;
while (thread_graphics)
gs_leave_context();
if (graphics->device) {
struct gs_effect *effect = graphics->first_effect;
thread_graphics = graphics;
graphics->exports.device_enter_context(graphics->device);
while (effect) {
struct gs_effect *next = effect->next;
gs_effect_actually_destroy(effect);
effect = next;
}
graphics->exports.gs_vertexbuffer_destroy(
graphics->sprite_buffer);
graphics->exports.gs_vertexbuffer_destroy(
graphics->immediate_vertbuffer);
graphics->exports.device_destroy(graphics->device);
thread_graphics = NULL;
}
pthread_mutex_destroy(&graphics->mutex);
pthread_mutex_destroy(&graphics->effect_mutex);
da_free(graphics->matrix_stack);
da_free(graphics->viewport_stack);
da_free(graphics->blend_state_stack);
if (graphics->module)
os_dlclose(graphics->module);
bfree(graphics);
gs_free_image_deps();
}
void gs_enter_context(graphics_t *graphics)
{
if (!ptr_valid(graphics, "gs_enter_context"))
return;
bool is_current = thread_graphics == graphics;
if (thread_graphics && !is_current) {
while (thread_graphics)
gs_leave_context();
}
if (!is_current) {
pthread_mutex_lock(&graphics->mutex);
graphics->exports.device_enter_context(graphics->device);
thread_graphics = graphics;
}
os_atomic_inc_long(&graphics->ref);
}
void gs_leave_context(void)
{
if (gs_valid("gs_leave_context")) {
if (!os_atomic_dec_long(&thread_graphics->ref)) {
graphics_t *graphics = thread_graphics;
graphics->exports.device_leave_context(
graphics->device);
pthread_mutex_unlock(&graphics->mutex);
thread_graphics = NULL;
}
}
}
graphics_t *gs_get_context(void)
{
return thread_graphics;
}
const char *gs_get_device_name(void)
{
return gs_valid("gs_get_device_name") ?
thread_graphics->exports.device_get_name() : NULL;
}
int gs_get_device_type(void)
{
return gs_valid("gs_get_device_type") ?
thread_graphics->exports.device_get_type() : -1;
}
static inline struct matrix4 *top_matrix(graphics_t *graphics)
{
return graphics->matrix_stack.array + graphics->cur_matrix;
}
void gs_matrix_push(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_matrix_push"))
return;
struct matrix4 mat, *top_mat = top_matrix(graphics);
memcpy(&mat, top_mat, sizeof(struct matrix4));
da_push_back(graphics->matrix_stack, &mat);
graphics->cur_matrix++;
}
void gs_matrix_pop(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_matrix_pop"))
return;
if (graphics->cur_matrix == 0) {
blog(LOG_ERROR, "Tried to pop last matrix on stack");
return;
}
da_erase(graphics->matrix_stack, graphics->cur_matrix);
graphics->cur_matrix--;
}
void gs_matrix_identity(void)
{
struct matrix4 *top_mat;
if (!gs_valid("gs_matrix_identity"))
return;
top_mat = top_matrix(thread_graphics);
if (top_mat)
matrix4_identity(top_mat);
}
void gs_matrix_transpose(void)
{
struct matrix4 *top_mat;
if (!gs_valid("gs_matrix_transpose"))
return;
top_mat = top_matrix(thread_graphics);
if (top_mat)
matrix4_transpose(top_mat, top_mat);
}
void gs_matrix_set(const struct matrix4 *matrix)
{
struct matrix4 *top_mat;
if (!gs_valid("gs_matrix_set"))
return;
top_mat = top_matrix(thread_graphics);
if (top_mat)
matrix4_copy(top_mat, matrix);
}
void gs_matrix_get(struct matrix4 *dst)
{
struct matrix4 *top_mat;
if (!gs_valid("gs_matrix_get"))
return;
top_mat = top_matrix(thread_graphics);
if (top_mat)
matrix4_copy(dst, top_mat);
}
void gs_matrix_mul(const struct matrix4 *matrix)
{
struct matrix4 *top_mat;
if (!gs_valid("gs_matrix_mul"))
return;
top_mat = top_matrix(thread_graphics);
if (top_mat)
matrix4_mul(top_mat, matrix, top_mat);
}
void gs_matrix_rotquat(const struct quat *rot)
{
struct matrix4 *top_mat;
if (!gs_valid("gs_matrix_rotquat"))
return;
top_mat = top_matrix(thread_graphics);
if (top_mat)
matrix4_rotate_i(top_mat, rot, top_mat);
}
void gs_matrix_rotaa(const struct axisang *rot)
{
struct matrix4 *top_mat;
if (!gs_valid("gs_matrix_rotaa"))
return;
top_mat = top_matrix(thread_graphics);
if (top_mat)
matrix4_rotate_aa_i(top_mat, rot, top_mat);
}
void gs_matrix_translate(const struct vec3 *pos)
{
struct matrix4 *top_mat;
if (!gs_valid("gs_matrix_translate"))
return;
top_mat = top_matrix(thread_graphics);
if (top_mat)
matrix4_translate3v_i(top_mat, pos, top_mat);
}
void gs_matrix_scale(const struct vec3 *scale)
{
struct matrix4 *top_mat;
if (!gs_valid("gs_matrix_scale"))
return;
top_mat = top_matrix(thread_graphics);
if (top_mat)
matrix4_scale_i(top_mat, scale, top_mat);
}
void gs_matrix_rotaa4f(float x, float y, float z, float angle)
{
struct matrix4 *top_mat;
struct axisang aa;
if (!gs_valid("gs_matrix_rotaa4f"))
return;
top_mat = top_matrix(thread_graphics);
if (top_mat) {
axisang_set(&aa, x, y, z, angle);
matrix4_rotate_aa_i(top_mat, &aa, top_mat);
}
}
void gs_matrix_translate3f(float x, float y, float z)
{
struct matrix4 *top_mat;
struct vec3 p;
if (!gs_valid("gs_matrix_translate3f"))
return;
top_mat = top_matrix(thread_graphics);
if (top_mat) {
vec3_set(&p, x, y, z);
matrix4_translate3v_i(top_mat, &p, top_mat);
}
}
void gs_matrix_scale3f(float x, float y, float z)
{
struct matrix4 *top_mat = top_matrix(thread_graphics);
struct vec3 p;
if (top_mat) {
vec3_set(&p, x, y, z);
matrix4_scale_i(top_mat, &p, top_mat);
}
}
static inline void reset_immediate_arrays(graphics_t *graphics)
{
da_init(graphics->verts);
da_init(graphics->norms);
da_init(graphics->colors);
for (size_t i = 0; i < 16; i++)
da_init(graphics->texverts[i]);
}
void gs_render_start(bool b_new)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_render_start"))
return;
graphics->using_immediate = !b_new;
reset_immediate_arrays(graphics);
if (b_new) {
graphics->vbd = gs_vbdata_create();
} else {
graphics->vbd = gs_vertexbuffer_get_data(
graphics->immediate_vertbuffer);
memset(graphics->vbd->colors, 0xFF,
sizeof(uint32_t) * IMMEDIATE_COUNT);
graphics->verts.array = graphics->vbd->points;
graphics->norms.array = graphics->vbd->normals;
graphics->colors.array = graphics->vbd->colors;
graphics->texverts[0].array = graphics->vbd->tvarray[0].array;
graphics->verts.capacity = IMMEDIATE_COUNT;
graphics->norms.capacity = IMMEDIATE_COUNT;
graphics->colors.capacity = IMMEDIATE_COUNT;
graphics->texverts[0].capacity = IMMEDIATE_COUNT;
}
}
static inline size_t min_size(const size_t a, const size_t b)
{
return (a < b) ? a : b;
}
void gs_render_stop(enum gs_draw_mode mode)
{
graphics_t *graphics = thread_graphics;
size_t i, num;
if (!gs_valid("gs_render_stop"))
return;
num = graphics->verts.num;
if (!num) {
if (!graphics->using_immediate) {
da_free(graphics->verts);
da_free(graphics->norms);
da_free(graphics->colors);
for (i = 0; i < 16; i++)
da_free(graphics->texverts[i]);
gs_vbdata_destroy(graphics->vbd);
}
return;
}
if (graphics->norms.num &&
(graphics->norms.num != graphics->verts.num)) {
blog(LOG_ERROR, "gs_render_stop: normal count does "
"not match vertex count");
num = min_size(num, graphics->norms.num);
}
if (graphics->colors.num &&
(graphics->colors.num != graphics->verts.num)) {
blog(LOG_ERROR, "gs_render_stop: color count does "
"not match vertex count");
num = min_size(num, graphics->colors.num);
}
if (graphics->texverts[0].num &&
(graphics->texverts[0].num != graphics->verts.num)) {
blog(LOG_ERROR, "gs_render_stop: texture vertex count does "
"not match vertex count");
num = min_size(num, graphics->texverts[0].num);
}
if (graphics->using_immediate) {
gs_vertexbuffer_flush(graphics->immediate_vertbuffer);
gs_load_vertexbuffer(graphics->immediate_vertbuffer);
gs_load_indexbuffer(NULL);
gs_draw(mode, 0, (uint32_t)num);
reset_immediate_arrays(graphics);
} else {
gs_vertbuffer_t *vb = gs_render_save();
gs_load_vertexbuffer(vb);
gs_load_indexbuffer(NULL);
gs_draw(mode, 0, 0);
gs_vertexbuffer_destroy(vb);
}
graphics->vbd = NULL;
}
gs_vertbuffer_t *gs_render_save(void)
{
graphics_t *graphics = thread_graphics;
size_t num_tex, i;
if (!gs_valid("gs_render_save"))
return NULL;
if (graphics->using_immediate)
return NULL;
if (!graphics->verts.num) {
gs_vbdata_destroy(graphics->vbd);
return NULL;
}
for (num_tex = 0; num_tex < 16; num_tex++)
if (!graphics->texverts[num_tex].num)
break;
graphics->vbd->points = graphics->verts.array;
graphics->vbd->normals = graphics->norms.array;
graphics->vbd->colors = graphics->colors.array;
graphics->vbd->num = graphics->verts.num;
graphics->vbd->num_tex = num_tex;
if (graphics->vbd->num_tex) {
graphics->vbd->tvarray =
bmalloc(sizeof(struct gs_tvertarray) * num_tex);
for (i = 0; i < num_tex; i++) {
graphics->vbd->tvarray[i].width = 2;
graphics->vbd->tvarray[i].array =
graphics->texverts[i].array;
}
}
reset_immediate_arrays(graphics);
return gs_vertexbuffer_create(graphics->vbd, 0);
}
void gs_vertex2f(float x, float y)
{
struct vec3 v3;
if (!gs_valid("gs_verte"))
return;
vec3_set(&v3, x, y, 0.0f);
gs_vertex3v(&v3);
}
void gs_vertex3f(float x, float y, float z)
{
struct vec3 v3;
if (!gs_valid("gs_vertex3f"))
return;
vec3_set(&v3, x, y, z);
gs_vertex3v(&v3);
}
void gs_normal3f(float x, float y, float z)
{
struct vec3 v3;
if (!gs_valid("gs_normal3f"))
return;
vec3_set(&v3, x, y, z);
gs_normal3v(&v3);
}
static inline bool validvertsize(graphics_t *graphics, size_t num,
const char *name)
{
if (graphics->using_immediate && num == IMMEDIATE_COUNT) {
blog(LOG_ERROR, "%s: tried to use over %u "
"for immediate rendering",
name, IMMEDIATE_COUNT);
return false;
}
return true;
}
void gs_color(uint32_t color)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_color"))
return;
if (!validvertsize(graphics, graphics->colors.num, "gs_color"))
return;
da_push_back(graphics->colors, &color);
}
void gs_texcoord(float x, float y, int unit)
{
struct vec2 v2;
if (!gs_valid("gs_texcoord"))
return;
vec2_set(&v2, x, y);
gs_texcoord2v(&v2, unit);
}
void gs_vertex2v(const struct vec2 *v)
{
struct vec3 v3;
if (!gs_valid("gs_vertex2v"))
return;
vec3_set(&v3, v->x, v->y, 0.0f);
gs_vertex3v(&v3);
}
void gs_vertex3v(const struct vec3 *v)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_vertex3v"))
return;
if (!validvertsize(graphics, graphics->verts.num, "gs_vertex"))
return;
da_push_back(graphics->verts, v);
}
void gs_normal3v(const struct vec3 *v)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_normal3v"))
return;
if (!validvertsize(graphics, graphics->norms.num, "gs_normal"))
return;
da_push_back(graphics->norms, v);
}
void gs_color4v(const struct vec4 *v)
{
/* TODO */
UNUSED_PARAMETER(v);
}
void gs_texcoord2v(const struct vec2 *v, int unit)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_texcoord2v"))
return;
if (!validvertsize(graphics, graphics->texverts[unit].num,
"gs_texcoord"))
return;
da_push_back(graphics->texverts[unit], v);
}
input_t *gs_get_input(void)
{
/* TODO */
return NULL;
}
gs_effect_t *gs_get_effect(void)
{
if (!gs_valid("gs_get_effect"))
return NULL;
return thread_graphics ? thread_graphics->cur_effect : NULL;
}
static inline struct gs_effect *find_cached_effect(const char *filename)
{
struct gs_effect *effect = thread_graphics->first_effect;
while (effect) {
if (strcmp(effect->effect_path, filename) == 0)
break;
effect = effect->next;
}
return effect;
}
gs_effect_t *gs_effect_create_from_file(const char *file, char **error_string)
{
char *file_string;
gs_effect_t *effect = NULL;
if (!gs_valid_p("gs_effect_create_from_file", file))
return NULL;
effect = find_cached_effect(file);
if (effect)
return effect;
file_string = os_quick_read_utf8_file(file);
if (!file_string) {
blog(LOG_ERROR, "Could not load effect file '%s'", file);
return NULL;
}
effect = gs_effect_create(file_string, file, error_string);
bfree(file_string);
return effect;
}
gs_effect_t *gs_effect_create(const char *effect_string, const char *filename,
char **error_string)
{
if (!gs_valid_p("gs_effect_create", effect_string))
return NULL;
struct gs_effect *effect = bzalloc(sizeof(struct gs_effect));
struct effect_parser parser;
bool success;
effect->graphics = thread_graphics;
effect->effect_path = bstrdup(filename);
ep_init(&parser);
success = ep_parse(&parser, effect, effect_string, filename);
if (!success) {
if (error_string)
*error_string = error_data_buildstring(
&parser.cfp.error_list);
gs_effect_destroy(effect);
effect = NULL;
}
if (effect) {
pthread_mutex_lock(&thread_graphics->effect_mutex);
if (effect->effect_path) {
effect->cached = true;
effect->next = thread_graphics->first_effect;
thread_graphics->first_effect = effect;
}
pthread_mutex_unlock(&thread_graphics->effect_mutex);
}
ep_free(&parser);
return effect;
}
gs_shader_t *gs_vertexshader_create_from_file(const char *file,
char **error_string)
{
if (!gs_valid_p("gs_vertexshader_create_from_file", file))
return NULL;
char *file_string;
gs_shader_t *shader = NULL;
file_string = os_quick_read_utf8_file(file);
if (!file_string) {
blog(LOG_ERROR, "Could not load vertex shader file '%s'",
file);
return NULL;
}
shader = gs_vertexshader_create(file_string, file, error_string);
bfree(file_string);
return shader;
}
gs_shader_t *gs_pixelshader_create_from_file(const char *file,
char **error_string)
{
char *file_string;
gs_shader_t *shader = NULL;
if (!gs_valid_p("gs_pixelshader_create_from_file", file))
return NULL;
file_string = os_quick_read_utf8_file(file);
if (!file_string) {
blog(LOG_ERROR, "Could not load pixel shader file '%s'",
file);
return NULL;
}
shader = gs_pixelshader_create(file_string, file, error_string);
bfree(file_string);
return shader;
}
gs_texture_t *gs_texture_create_from_file(const char *file)
{
enum gs_color_format format;
uint32_t cx;
uint32_t cy;
uint8_t *data = gs_create_texture_file_data(file, &format, &cx, &cy);
gs_texture_t *tex = NULL;
if (data) {
tex = gs_texture_create(cx, cy, format, 1,
(const uint8_t**)&data, 0);
bfree(data);
}
return tex;
}
static inline void assign_sprite_rect(float *start, float *end, float size,
bool flip)
{
if (!flip) {
*start = 0.0f;
*end = size;
} else {
*start = size;
*end = 0.0f;
}
}
static inline void assign_sprite_uv(float *start, float *end, bool flip)
{
if (!flip) {
*start = 0.0f;
*end = 1.0f;
} else {
*start = 1.0f;
*end = 0.0f;
}
}
static void build_sprite(struct gs_vb_data *data, float fcx, float fcy,
float start_u, float end_u, float start_v, float end_v)
{
struct vec2 *tvarray = data->tvarray[0].array;
vec3_zero(data->points);
vec3_set(data->points+1, fcx, 0.0f, 0.0f);
vec3_set(data->points+2, 0.0f, fcy, 0.0f);
vec3_set(data->points+3, fcx, fcy, 0.0f);
vec2_set(tvarray, start_u, start_v);
vec2_set(tvarray+1, end_u, start_v);
vec2_set(tvarray+2, start_u, end_v);
vec2_set(tvarray+3, end_u, end_v);
}
static inline void build_sprite_norm(struct gs_vb_data *data, float fcx,
float fcy, uint32_t flip)
{
float start_u, end_u;
float start_v, end_v;
assign_sprite_uv(&start_u, &end_u, (flip & GS_FLIP_U) != 0);
assign_sprite_uv(&start_v, &end_v, (flip & GS_FLIP_V) != 0);
build_sprite(data, fcx, fcy, start_u, end_u, start_v, end_v);
}
static inline void build_subsprite_norm(struct gs_vb_data *data,
float fsub_x, float fsub_y, float fsub_cx, float fsub_cy,
float fcx, float fcy, uint32_t flip)
{
float start_u, end_u;
float start_v, end_v;
if ((flip & GS_FLIP_U) == 0) {
start_u = fsub_x / fcx;
end_u = (fsub_x + fsub_cx) / fcx;
} else {
start_u = (fsub_x + fsub_cx) / fcx;
end_u = fsub_x / fcx;
}
if ((flip & GS_FLIP_V) == 0) {
start_v = fsub_y / fcy;
end_v = (fsub_y + fsub_cy) / fcy;
} else {
start_v = (fsub_y + fsub_cy) / fcy;
end_v = fsub_y / fcy;
}
build_sprite(data, fsub_cx, fsub_cy, start_u, end_u, start_v, end_v);
}
static inline void build_sprite_rect(struct gs_vb_data *data, gs_texture_t *tex,
float fcx, float fcy, uint32_t flip)
{
float start_u, end_u;
float start_v, end_v;
float width = (float)gs_texture_get_width(tex);
float height = (float)gs_texture_get_height(tex);
assign_sprite_rect(&start_u, &end_u, width, (flip & GS_FLIP_U) != 0);
assign_sprite_rect(&start_v, &end_v, height, (flip & GS_FLIP_V) != 0);
build_sprite(data, fcx, fcy, start_u, end_u, start_v, end_v);
}
void gs_draw_sprite(gs_texture_t *tex, uint32_t flip, uint32_t width,
uint32_t height)
{
graphics_t *graphics = thread_graphics;
float fcx, fcy;
struct gs_vb_data *data;
if (tex) {
if (gs_get_texture_type(tex) != GS_TEXTURE_2D) {
blog(LOG_ERROR, "A sprite must be a 2D texture");
return;
}
} else {
if (!width || !height) {
blog(LOG_ERROR, "A sprite cannot be drawn without "
"a width/height");
return;
}
}
fcx = width ? (float)width : (float)gs_texture_get_width(tex);
fcy = height ? (float)height : (float)gs_texture_get_height(tex);
data = gs_vertexbuffer_get_data(graphics->sprite_buffer);
if (tex && gs_texture_is_rect(tex))
build_sprite_rect(data, tex, fcx, fcy, flip);
else
build_sprite_norm(data, fcx, fcy, flip);
gs_vertexbuffer_flush(graphics->sprite_buffer);
gs_load_vertexbuffer(graphics->sprite_buffer);
gs_load_indexbuffer(NULL);
gs_draw(GS_TRISTRIP, 0, 0);
}
void gs_draw_sprite_subregion(gs_texture_t *tex, uint32_t flip,
uint32_t sub_x, uint32_t sub_y,
uint32_t sub_cx, uint32_t sub_cy)
{
graphics_t *graphics = thread_graphics;
float fcx, fcy;
struct gs_vb_data *data;
if (tex) {
if (gs_get_texture_type(tex) != GS_TEXTURE_2D) {
blog(LOG_ERROR, "A sprite must be a 2D texture");
return;
}
}
fcx = (float)gs_texture_get_width(tex);
fcy = (float)gs_texture_get_height(tex);
data = gs_vertexbuffer_get_data(graphics->sprite_buffer);
build_subsprite_norm(data,
(float)sub_x, (float)sub_y,
(float)sub_cx, (float)sub_cy,
fcx, fcy, flip);
gs_vertexbuffer_flush(graphics->sprite_buffer);
gs_load_vertexbuffer(graphics->sprite_buffer);
gs_load_indexbuffer(NULL);
gs_draw(GS_TRISTRIP, 0, 0);
}
void gs_draw_cube_backdrop(gs_texture_t *cubetex, const struct quat *rot,
float left, float right, float top, float bottom, float znear)
{
/* TODO */
UNUSED_PARAMETER(cubetex);
UNUSED_PARAMETER(rot);
UNUSED_PARAMETER(left);
UNUSED_PARAMETER(right);
UNUSED_PARAMETER(top);
UNUSED_PARAMETER(bottom);
UNUSED_PARAMETER(znear);
}
void gs_reset_viewport(void)
{
uint32_t cx, cy;
if (!gs_valid("gs_reset_viewport"))
return;
gs_get_size(&cx, &cy);
gs_set_viewport(0, 0, (int)cx, (int)cy);
}
void gs_set_2d_mode(void)
{
uint32_t cx, cy;
if (!gs_valid("gs_set_2d_mode"))
return;
gs_get_size(&cx, &cy);
gs_ortho(0.0f, (float)cx, 0.0f, (float)cy, -1.0, -1024.0f);
}
void gs_set_3d_mode(double fovy, double znear, double zvar)
{
/* TODO */
UNUSED_PARAMETER(fovy);
UNUSED_PARAMETER(znear);
UNUSED_PARAMETER(zvar);
}
void gs_viewport_push(void)
{
if (!gs_valid("gs_viewport_push"))
return;
struct gs_rect *rect = da_push_back_new(
thread_graphics->viewport_stack);
gs_get_viewport(rect);
}
void gs_viewport_pop(void)
{
struct gs_rect *rect;
if (!gs_valid("gs_viewport_pop"))
return;
if (!thread_graphics->viewport_stack.num)
return;
rect = da_end(thread_graphics->viewport_stack);
gs_set_viewport(rect->x, rect->y, rect->cx, rect->cy);
da_pop_back(thread_graphics->viewport_stack);
}
void gs_texture_set_image(gs_texture_t *tex, const uint8_t *data,
uint32_t linesize, bool flip)
{
uint8_t *ptr;
uint32_t linesize_out;
uint32_t row_copy;
int32_t height;
int32_t y;
if (!gs_valid_p2("gs_texture_set_image", tex, data))
return;
height = (int32_t)gs_texture_get_height(tex);
if (!gs_texture_map(tex, &ptr, &linesize_out))
return;
row_copy = (linesize < linesize_out) ? linesize : linesize_out;
if (flip) {
for (y = height-1; y >= 0; y--)
memcpy(ptr + (uint32_t)y * linesize_out,
data + (uint32_t)(height - y - 1) * linesize,
row_copy);
} else if (linesize == linesize_out) {
memcpy(ptr, data, row_copy * height);
} else {
for (y = 0; y < height; y++)
memcpy(ptr + (uint32_t)y * linesize_out,
data + (uint32_t)y * linesize,
row_copy);
}
gs_texture_unmap(tex);
}
void gs_cubetexture_set_image(gs_texture_t *cubetex, uint32_t side,
const void *data, uint32_t linesize, bool invert)
{
/* TODO */
UNUSED_PARAMETER(cubetex);
UNUSED_PARAMETER(side);
UNUSED_PARAMETER(data);
UNUSED_PARAMETER(linesize);
UNUSED_PARAMETER(invert);
}
void gs_perspective(float angle, float aspect, float near, float far)
{
graphics_t *graphics = thread_graphics;
float xmin, xmax, ymin, ymax;
if (!gs_valid("gs_perspective"))
return;
ymax = near * tanf(RAD(angle)*0.5f);
ymin = -ymax;
xmin = ymin * aspect;
xmax = ymax * aspect;
graphics->exports.device_frustum(graphics->device, xmin, xmax,
ymin, ymax, near, far);
}
void gs_blend_state_push(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_blend_state_push"))
return;
da_push_back(graphics->blend_state_stack, &graphics->cur_blend_state);
}
void gs_blend_state_pop(void)
{
graphics_t *graphics = thread_graphics;
struct blend_state *state;
if (!gs_valid("gs_blend_state_pop"))
return;
state = da_end(graphics->blend_state_stack);
if (!state)
return;
gs_enable_blending(state->enabled);
gs_blend_function_separate(state->src_c, state->dest_c,
state->src_a, state->dest_a);
da_pop_back(graphics->blend_state_stack);
}
void gs_reset_blend_state(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_preprocessor_name"))
return;
if (!graphics->cur_blend_state.enabled)
gs_enable_blending(true);
if (graphics->cur_blend_state.src_c != GS_BLEND_SRCALPHA ||
graphics->cur_blend_state.dest_c != GS_BLEND_INVSRCALPHA ||
graphics->cur_blend_state.src_a != GS_BLEND_ONE ||
graphics->cur_blend_state.dest_a != GS_BLEND_ONE)
gs_blend_function_separate(
GS_BLEND_SRCALPHA, GS_BLEND_INVSRCALPHA,
GS_BLEND_ONE, GS_BLEND_ONE);
}
/* ------------------------------------------------------------------------- */
const char *gs_preprocessor_name(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_preprocessor_name"))
return NULL;
return graphics->exports.device_preprocessor_name();
}
gs_swapchain_t *gs_swapchain_create(const struct gs_init_data *data)
{
struct gs_init_data new_data = *data;
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_swapchain_create", data))
return NULL;
if (new_data.num_backbuffers == 0)
new_data.num_backbuffers = 1;
return graphics->exports.device_swapchain_create(graphics->device,
&new_data);
}
void gs_resize(uint32_t x, uint32_t y)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_resize"))
return;
graphics->exports.device_resize(graphics->device, x, y);
}
void gs_get_size(uint32_t *x, uint32_t *y)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_get_size"))
return;
graphics->exports.device_get_size(graphics->device, x, y);
}
uint32_t gs_get_width(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_get_width"))
return 0;
return graphics->exports.device_get_width(graphics->device);
}
uint32_t gs_get_height(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_get_height"))
return 0;
return graphics->exports.device_get_height(graphics->device);
}
static inline bool is_pow2(uint32_t size)
{
return size >= 2 && (size & (size-1)) == 0;
}
gs_texture_t *gs_texture_create(uint32_t width, uint32_t height,
enum gs_color_format color_format, uint32_t levels,
const uint8_t **data, uint32_t flags)
{
graphics_t *graphics = thread_graphics;
bool pow2tex = is_pow2(width) && is_pow2(height);
bool uses_mipmaps = (flags & GS_BUILD_MIPMAPS || levels != 1);
if (!gs_valid("gs_texture_create"))
return NULL;
if (uses_mipmaps && !pow2tex) {
blog(LOG_WARNING, "Cannot use mipmaps with a "
"non-power-of-two texture. Disabling "
"mipmaps for this texture.");
uses_mipmaps = false;
flags &= ~GS_BUILD_MIPMAPS;
levels = 1;
}
if (uses_mipmaps && flags & GS_RENDER_TARGET) {
blog(LOG_WARNING, "Cannot use mipmaps with render targets. "
"Disabling mipmaps for this texture.");
flags &= ~GS_BUILD_MIPMAPS;
levels = 1;
}
return graphics->exports.device_texture_create(graphics->device,
width, height, color_format, levels, data, flags);
}
gs_texture_t *gs_cubetexture_create(uint32_t size,
enum gs_color_format color_format, uint32_t levels,
const uint8_t **data, uint32_t flags)
{
graphics_t *graphics = thread_graphics;
bool pow2tex = is_pow2(size);
bool uses_mipmaps = (flags & GS_BUILD_MIPMAPS || levels != 1);
if (!gs_valid("gs_cubetexture_create"))
return NULL;
if (uses_mipmaps && !pow2tex) {
blog(LOG_WARNING, "Cannot use mipmaps with a "
"non-power-of-two texture. Disabling "
"mipmaps for this texture.");
uses_mipmaps = false;
flags &= ~GS_BUILD_MIPMAPS;
levels = 1;
}
if (uses_mipmaps && flags & GS_RENDER_TARGET) {
blog(LOG_WARNING, "Cannot use mipmaps with render targets. "
"Disabling mipmaps for this texture.");
flags &= ~GS_BUILD_MIPMAPS;
levels = 1;
data = NULL;
}
return graphics->exports.device_cubetexture_create(graphics->device,
size, color_format, levels, data, flags);
}
gs_texture_t *gs_voltexture_create(uint32_t width, uint32_t height,
uint32_t depth, enum gs_color_format color_format,
uint32_t levels, const uint8_t **data, uint32_t flags)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_voltexture_create"))
return NULL;
return graphics->exports.device_voltexture_create(graphics->device,
width, height, depth, color_format, levels, data,
flags);
}
gs_zstencil_t *gs_zstencil_create(uint32_t width, uint32_t height,
enum gs_zstencil_format format)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_zstencil_create"))
return NULL;
return graphics->exports.device_zstencil_create(graphics->device,
width, height, format);
}
gs_stagesurf_t *gs_stagesurface_create(uint32_t width, uint32_t height,
enum gs_color_format color_format)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_stagesurface_create"))
return NULL;
return graphics->exports.device_stagesurface_create(graphics->device,
width, height, color_format);
}
gs_samplerstate_t *gs_samplerstate_create(const struct gs_sampler_info *info)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_samplerstate_create", info))
return NULL;
return graphics->exports.device_samplerstate_create(graphics->device,
info);
}
gs_shader_t *gs_vertexshader_create(const char *shader, const char *file,
char **error_string)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_vertexshader_create", shader))
return NULL;
return graphics->exports.device_vertexshader_create(graphics->device,
shader, file, error_string);
}
gs_shader_t *gs_pixelshader_create(const char *shader,
const char *file, char **error_string)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_pixelshader_create", shader))
return NULL;
return graphics->exports.device_pixelshader_create(graphics->device,
shader, file, error_string);
}
gs_vertbuffer_t *gs_vertexbuffer_create(struct gs_vb_data *data,
uint32_t flags)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_vertexbuffer_create"))
return NULL;
if (data && data->num && (flags & GS_DUP_BUFFER) != 0) {
struct gs_vb_data *new_data = gs_vbdata_create();
new_data->num = data->num;
#define DUP_VAL(val) \
do { \
if (data->val) \
new_data->val = bmemdup(data->val, \
sizeof(*data->val) * \
data->num); \
} while (false)
DUP_VAL(points);
DUP_VAL(normals);
DUP_VAL(tangents);
DUP_VAL(colors);
#undef DUP_VAL
if (data->tvarray && data->num_tex) {
new_data->num_tex = data->num_tex;
new_data->tvarray = bzalloc(
sizeof(struct gs_tvertarray) *
data->num_tex);
for (size_t i = 0; i < data->num_tex; i++) {
struct gs_tvertarray *tv = &data->tvarray[i];
struct gs_tvertarray *new_tv =
&new_data->tvarray[i];
size_t size = tv->width * sizeof(float);
new_tv->width = tv->width;
new_tv->array = bmemdup(tv->array,
size * data->num);
}
}
data = new_data;
}
return graphics->exports.device_vertexbuffer_create(graphics->device,
data, flags);
}
gs_indexbuffer_t *gs_indexbuffer_create(enum gs_index_type type,
void *indices, size_t num, uint32_t flags)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_indexbuffer_create"))
return NULL;
if (indices && num && (flags & GS_DUP_BUFFER) != 0) {
size_t size = type == GS_UNSIGNED_SHORT
? sizeof(unsigned short)
: sizeof(unsigned long);
indices = bmemdup(indices, size * num);
}
return graphics->exports.device_indexbuffer_create(graphics->device,
type, indices, num, flags);
}
enum gs_texture_type gs_get_texture_type(const gs_texture_t *texture)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_get_texture_type", texture))
return GS_TEXTURE_2D;
return graphics->exports.device_get_texture_type(texture);
}
void gs_load_vertexbuffer(gs_vertbuffer_t *vertbuffer)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_load_vertexbuffer"))
return;
graphics->exports.device_load_vertexbuffer(graphics->device,
vertbuffer);
}
void gs_load_indexbuffer(gs_indexbuffer_t *indexbuffer)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_load_indexbuffer"))
return;
graphics->exports.device_load_indexbuffer(graphics->device,
indexbuffer);
}
void gs_load_texture(gs_texture_t *tex, int unit)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_load_texture"))
return;
graphics->exports.device_load_texture(graphics->device, tex, unit);
}
void gs_load_samplerstate(gs_samplerstate_t *samplerstate, int unit)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_load_samplerstate"))
return;
graphics->exports.device_load_samplerstate(graphics->device,
samplerstate, unit);
}
void gs_load_vertexshader(gs_shader_t *vertshader)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_load_vertexshader"))
return;
graphics->exports.device_load_vertexshader(graphics->device,
vertshader);
}
void gs_load_pixelshader(gs_shader_t *pixelshader)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_load_pixelshader"))
return;
graphics->exports.device_load_pixelshader(graphics->device,
pixelshader);
}
void gs_load_default_samplerstate(bool b_3d, int unit)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_load_default_samplerstate"))
return;
graphics->exports.device_load_default_samplerstate(graphics->device,
b_3d, unit);
}
gs_shader_t *gs_get_vertex_shader(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_get_vertex_shader"))
return NULL;
return graphics->exports.device_get_vertex_shader(graphics->device);
}
gs_shader_t *gs_get_pixel_shader(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_get_pixel_shader"))
return NULL;
return graphics->exports.device_get_pixel_shader(graphics->device);
}
gs_texture_t *gs_get_render_target(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_get_render_target"))
return NULL;
return graphics->exports.device_get_render_target(graphics->device);
}
gs_zstencil_t *gs_get_zstencil_target(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_get_zstencil_target"))
return NULL;
return graphics->exports.device_get_zstencil_target(graphics->device);
}
void gs_set_render_target(gs_texture_t *tex, gs_zstencil_t *zstencil)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_set_render_target"))
return;
graphics->exports.device_set_render_target(graphics->device, tex,
zstencil);
}
void gs_set_cube_render_target(gs_texture_t *cubetex, int side,
gs_zstencil_t *zstencil)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_set_cube_render_target"))
return;
graphics->exports.device_set_cube_render_target(graphics->device,
cubetex, side, zstencil);
}
void gs_copy_texture(gs_texture_t *dst, gs_texture_t *src)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p2("gs_copy_texture", dst, src))
return;
graphics->exports.device_copy_texture(graphics->device, dst, src);
}
void gs_copy_texture_region(gs_texture_t *dst, uint32_t dst_x, uint32_t dst_y,
gs_texture_t *src, uint32_t src_x, uint32_t src_y,
uint32_t src_w, uint32_t src_h)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_copy_texture_region", dst))
return;
graphics->exports.device_copy_texture_region(graphics->device,
dst, dst_x, dst_y,
src, src_x, src_y, src_w, src_h);
}
void gs_stage_texture(gs_stagesurf_t *dst, gs_texture_t *src)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_stage_texture"))
return;
graphics->exports.device_stage_texture(graphics->device, dst, src);
}
void gs_begin_scene(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_begin_scene"))
return;
graphics->exports.device_begin_scene(graphics->device);
}
void gs_draw(enum gs_draw_mode draw_mode, uint32_t start_vert,
uint32_t num_verts)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_draw"))
return;
graphics->exports.device_draw(graphics->device, draw_mode,
start_vert, num_verts);
}
void gs_end_scene(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_end_scene"))
return;
graphics->exports.device_end_scene(graphics->device);
}
void gs_load_swapchain(gs_swapchain_t *swapchain)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_load_swapchain"))
return;
graphics->exports.device_load_swapchain(graphics->device, swapchain);
}
void gs_clear(uint32_t clear_flags, const struct vec4 *color, float depth,
uint8_t stencil)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_clear"))
return;
graphics->exports.device_clear(graphics->device, clear_flags, color,
depth, stencil);
}
void gs_present(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_present"))
return;
graphics->exports.device_present(graphics->device);
}
void gs_flush(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_flush"))
return;
graphics->exports.device_flush(graphics->device);
}
void gs_set_cull_mode(enum gs_cull_mode mode)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_set_cull_mode"))
return;
graphics->exports.device_set_cull_mode(graphics->device, mode);
}
enum gs_cull_mode gs_get_cull_mode(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_get_cull_mode"))
return GS_NEITHER;
return graphics->exports.device_get_cull_mode(graphics->device);
}
void gs_enable_blending(bool enable)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_enable_blending"))
return;
graphics->cur_blend_state.enabled = enable;
graphics->exports.device_enable_blending(graphics->device, enable);
}
void gs_enable_depth_test(bool enable)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_enable_depth_test"))
return;
graphics->exports.device_enable_depth_test(graphics->device, enable);
}
void gs_enable_stencil_test(bool enable)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_enable_stencil_test"))
return;
graphics->exports.device_enable_stencil_test(graphics->device, enable);
}
void gs_enable_stencil_write(bool enable)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_enable_stencil_write"))
return;
graphics->exports.device_enable_stencil_write(graphics->device, enable);
}
void gs_enable_color(bool red, bool green, bool blue, bool alpha)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_enable_color"))
return;
graphics->exports.device_enable_color(graphics->device, red, green,
blue, alpha);
}
void gs_blend_function(enum gs_blend_type src, enum gs_blend_type dest)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_blend_function"))
return;
graphics->cur_blend_state.src_c = src;
graphics->cur_blend_state.dest_c = dest;
graphics->cur_blend_state.src_a = src;
graphics->cur_blend_state.dest_a = dest;
graphics->exports.device_blend_function(graphics->device, src, dest);
}
void gs_blend_function_separate(
enum gs_blend_type src_c, enum gs_blend_type dest_c,
enum gs_blend_type src_a, enum gs_blend_type dest_a)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_blend_function_separate"))
return;
graphics->cur_blend_state.src_c = src_c;
graphics->cur_blend_state.dest_c = dest_c;
graphics->cur_blend_state.src_a = src_a;
graphics->cur_blend_state.dest_a = dest_a;
graphics->exports.device_blend_function_separate(graphics->device,
src_c, dest_c, src_a, dest_a);
}
void gs_depth_function(enum gs_depth_test test)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_depth_function"))
return;
graphics->exports.device_depth_function(graphics->device, test);
}
void gs_stencil_function(enum gs_stencil_side side, enum gs_depth_test test)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_stencil_function"))
return;
graphics->exports.device_stencil_function(graphics->device, side, test);
}
void gs_stencil_op(enum gs_stencil_side side, enum gs_stencil_op_type fail,
enum gs_stencil_op_type zfail, enum gs_stencil_op_type zpass)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_stencil_op"))
return;
graphics->exports.device_stencil_op(graphics->device, side, fail, zfail,
zpass);
}
void gs_set_viewport(int x, int y, int width, int height)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_set_viewport"))
return;
graphics->exports.device_set_viewport(graphics->device, x, y, width,
height);
}
void gs_get_viewport(struct gs_rect *rect)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_get_viewport", rect))
return;
graphics->exports.device_get_viewport(graphics->device, rect);
}
void gs_set_scissor_rect(const struct gs_rect *rect)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_set_scissor_rect"))
return;
graphics->exports.device_set_scissor_rect(graphics->device, rect);
}
void gs_ortho(float left, float right, float top, float bottom, float znear,
float zfar)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_ortho"))
return;
graphics->exports.device_ortho(graphics->device, left, right, top,
bottom, znear, zfar);
}
void gs_frustum(float left, float right, float top, float bottom, float znear,
float zfar)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_frustum"))
return;
graphics->exports.device_frustum(graphics->device, left, right, top,
bottom, znear, zfar);
}
void gs_projection_push(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_projection_push"))
return;
graphics->exports.device_projection_push(graphics->device);
}
void gs_projection_pop(void)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_projection_pop"))
return;
graphics->exports.device_projection_pop(graphics->device);
}
void gs_swapchain_destroy(gs_swapchain_t *swapchain)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_swapchain_destroy"))
return;
if (!swapchain)
return;
graphics->exports.gs_swapchain_destroy(swapchain);
}
void gs_shader_destroy(gs_shader_t *shader)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_shader_destroy"))
return;
if (!shader)
return;
graphics->exports.gs_shader_destroy(shader);
}
int gs_shader_get_num_params(const gs_shader_t *shader)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_shader_get_num_params", shader))
return 0;
return graphics->exports.gs_shader_get_num_params(shader);
}
gs_sparam_t *gs_shader_get_param_by_idx(gs_shader_t *shader, uint32_t param)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_shader_get_param_by_idx", shader))
return NULL;
return graphics->exports.gs_shader_get_param_by_idx(shader, param);
}
gs_sparam_t *gs_shader_get_param_by_name(gs_shader_t *shader, const char *name)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p2("gs_shader_get_param_by_name", shader, name))
return NULL;
return graphics->exports.gs_shader_get_param_by_name(shader, name);
}
gs_sparam_t *gs_shader_get_viewproj_matrix(const gs_shader_t *shader)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_shader_get_viewproj_matrix", shader))
return NULL;
return graphics->exports.gs_shader_get_viewproj_matrix(shader);
}
gs_sparam_t *gs_shader_get_world_matrix(const gs_shader_t *shader)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_shader_get_world_matrix", shader))
return NULL;
return graphics->exports.gs_shader_get_world_matrix(shader);
}
void gs_shader_get_param_info(const gs_sparam_t *param,
struct gs_shader_param_info *info)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p2("gs_shader_get_param_info", param, info))
return;
graphics->exports.gs_shader_get_param_info(param, info);
}
void gs_shader_set_bool(gs_sparam_t *param, bool val)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_shader_set_bool", param))
return;
graphics->exports.gs_shader_set_bool(param, val);
}
void gs_shader_set_float(gs_sparam_t *param, float val)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_shader_set_float", param))
return;
graphics->exports.gs_shader_set_float(param, val);
}
void gs_shader_set_int(gs_sparam_t *param, int val)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_shader_set_int", param))
return;
graphics->exports.gs_shader_set_int(param, val);
}
void gs_shader_set_matrix3(gs_sparam_t *param, const struct matrix3 *val)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p2("gs_shader_set_matrix3", param, val))
return;
graphics->exports.gs_shader_set_matrix3(param, val);
}
void gs_shader_set_matrix4(gs_sparam_t *param, const struct matrix4 *val)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p2("gs_shader_set_matrix4", param, val))
return;
graphics->exports.gs_shader_set_matrix4(param, val);
}
void gs_shader_set_vec2(gs_sparam_t *param, const struct vec2 *val)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p2("gs_shader_set_vec2", param, val))
return;
graphics->exports.gs_shader_set_vec2(param, val);
}
void gs_shader_set_vec3(gs_sparam_t *param, const struct vec3 *val)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p2("gs_shader_set_vec3", param, val))
return;
graphics->exports.gs_shader_set_vec3(param, val);
}
void gs_shader_set_vec4(gs_sparam_t *param, const struct vec4 *val)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p2("gs_shader_set_vec4", param, val))
return;
graphics->exports.gs_shader_set_vec4(param, val);
}
void gs_shader_set_texture(gs_sparam_t *param, gs_texture_t *val)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_shader_set_texture", param))
return;
graphics->exports.gs_shader_set_texture(param, val);
}
void gs_shader_set_val(gs_sparam_t *param, const void *val, size_t size)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p2("gs_shader_set_val", param, val))
return;
graphics->exports.gs_shader_set_val(param, val, size);
}
void gs_shader_set_default(gs_sparam_t *param)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_shader_set_default", param))
return;
graphics->exports.gs_shader_set_default(param);
}
void gs_shader_set_next_sampler(gs_sparam_t *param, gs_samplerstate_t *sampler)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_shader_set_next_sampler", param))
return;
graphics->exports.gs_shader_set_next_sampler(param, sampler);
}
void gs_texture_destroy(gs_texture_t *tex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_texture_destroy"))
return;
if (!tex)
return;
graphics->exports.gs_texture_destroy(tex);
}
uint32_t gs_texture_get_width(const gs_texture_t *tex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_texture_get_width", tex))
return 0;
return graphics->exports.gs_texture_get_width(tex);
}
uint32_t gs_texture_get_height(const gs_texture_t *tex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_texture_get_height", tex))
return 0;
return graphics->exports.gs_texture_get_height(tex);
}
enum gs_color_format gs_texture_get_color_format(const gs_texture_t *tex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_texture_get_color_format", tex))
return GS_UNKNOWN;
return graphics->exports.gs_texture_get_color_format(tex);
}
bool gs_texture_map(gs_texture_t *tex, uint8_t **ptr, uint32_t *linesize)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p3("gs_texture_map", tex, ptr, linesize))
return false;
return graphics->exports.gs_texture_map(tex, ptr, linesize);
}
void gs_texture_unmap(gs_texture_t *tex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_texture_unmap", tex))
return;
graphics->exports.gs_texture_unmap(tex);
}
bool gs_texture_is_rect(const gs_texture_t *tex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_texture_is_rect", tex))
return false;
if (graphics->exports.gs_texture_is_rect)
return graphics->exports.gs_texture_is_rect(tex);
else
return false;
}
void *gs_texture_get_obj(gs_texture_t *tex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_texture_get_obj", tex))
return NULL;
return graphics->exports.gs_texture_get_obj(tex);
}
void gs_cubetexture_destroy(gs_texture_t *cubetex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_cubetexture_destroy"))
return;
if (!cubetex)
return;
graphics->exports.gs_cubetexture_destroy(cubetex);
}
uint32_t gs_cubetexture_get_size(const gs_texture_t *cubetex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_cubetexture_get_size", cubetex))
return 0;
return graphics->exports.gs_cubetexture_get_size(cubetex);
}
enum gs_color_format gs_cubetexture_get_color_format(
const gs_texture_t *cubetex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_cubetexture_get_color_format", cubetex))
return GS_UNKNOWN;
return graphics->exports.gs_cubetexture_get_color_format(cubetex);
}
void gs_voltexture_destroy(gs_texture_t *voltex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_voltexture_destroy"))
return;
if (!voltex)
return;
graphics->exports.gs_voltexture_destroy(voltex);
}
uint32_t gs_voltexture_get_width(const gs_texture_t *voltex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_voltexture_get_width", voltex))
return 0;
return graphics->exports.gs_voltexture_get_width(voltex);
}
uint32_t gs_voltexture_get_height(const gs_texture_t *voltex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_voltexture_get_height", voltex))
return 0;
return graphics->exports.gs_voltexture_get_height(voltex);
}
uint32_t gs_voltexture_get_depth(const gs_texture_t *voltex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_voltexture_get_depth", voltex))
return 0;
return graphics->exports.gs_voltexture_get_depth(voltex);
}
enum gs_color_format gs_voltexture_get_color_format(const gs_texture_t *voltex)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_voltexture_get_color_format", voltex))
return GS_UNKNOWN;
return graphics->exports.gs_voltexture_get_color_format(voltex);
}
void gs_stagesurface_destroy(gs_stagesurf_t *stagesurf)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_stagesurface_destroy"))
return;
if (!stagesurf)
return;
graphics->exports.gs_stagesurface_destroy(stagesurf);
}
uint32_t gs_stagesurface_get_width(const gs_stagesurf_t *stagesurf)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_stagesurface_get_width", stagesurf))
return 0;
return graphics->exports.gs_stagesurface_get_width(stagesurf);
}
uint32_t gs_stagesurface_get_height(const gs_stagesurf_t *stagesurf)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_stagesurface_get_height", stagesurf))
return 0;
return graphics->exports.gs_stagesurface_get_height(stagesurf);
}
enum gs_color_format gs_stagesurface_get_color_format(
const gs_stagesurf_t *stagesurf)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_stagesurface_get_color_format", stagesurf))
return GS_UNKNOWN;
return graphics->exports.gs_stagesurface_get_color_format(stagesurf);
}
bool gs_stagesurface_map(gs_stagesurf_t *stagesurf, uint8_t **data,
uint32_t *linesize)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p3("gs_stagesurface_map", stagesurf, data, linesize))
return 0;
return graphics->exports.gs_stagesurface_map(stagesurf, data, linesize);
}
void gs_stagesurface_unmap(gs_stagesurf_t *stagesurf)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_stagesurface_unmap", stagesurf))
return;
graphics->exports.gs_stagesurface_unmap(stagesurf);
}
void gs_zstencil_destroy(gs_zstencil_t *zstencil)
{
if (!gs_valid("gs_zstencil_destroy"))
return;
if (!zstencil)
return;
thread_graphics->exports.gs_zstencil_destroy(zstencil);
}
void gs_samplerstate_destroy(gs_samplerstate_t *samplerstate)
{
if (!gs_valid("gs_samplerstate_destroy"))
return;
if (!samplerstate)
return;
thread_graphics->exports.gs_samplerstate_destroy(samplerstate);
}
void gs_vertexbuffer_destroy(gs_vertbuffer_t *vertbuffer)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_vertexbuffer_destroy"))
return;
if (!vertbuffer)
return;
graphics->exports.gs_vertexbuffer_destroy(vertbuffer);
}
void gs_vertexbuffer_flush(gs_vertbuffer_t *vertbuffer)
{
if (!gs_valid_p("gs_vertexbuffer_flush", vertbuffer))
return;
thread_graphics->exports.gs_vertexbuffer_flush(vertbuffer);
}
void gs_vertexbuffer_flush_direct(gs_vertbuffer_t *vertbuffer,
const struct gs_vb_data *data)
{
if (!gs_valid_p2("gs_vertexbuffer_flush_direct", vertbuffer, data))
return;
thread_graphics->exports.gs_vertexbuffer_flush_direct(vertbuffer,
data);
}
struct gs_vb_data *gs_vertexbuffer_get_data(const gs_vertbuffer_t *vertbuffer)
{
if (!gs_valid_p("gs_vertexbuffer_get_data", vertbuffer))
return NULL;
return thread_graphics->exports.gs_vertexbuffer_get_data(vertbuffer);
}
void gs_indexbuffer_destroy(gs_indexbuffer_t *indexbuffer)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_indexbuffer_destroy"))
return;
if (!indexbuffer)
return;
graphics->exports.gs_indexbuffer_destroy(indexbuffer);
}
void gs_indexbuffer_flush(gs_indexbuffer_t *indexbuffer)
{
if (!gs_valid_p("gs_indexbuffer_flush", indexbuffer))
return;
thread_graphics->exports.gs_indexbuffer_flush(indexbuffer);
}
void gs_indexbuffer_flush_direct(gs_indexbuffer_t *indexbuffer,
const void *data)
{
if (!gs_valid_p2("gs_indexbuffer_flush_direct", indexbuffer, data))
return;
thread_graphics->exports.gs_indexbuffer_flush_direct(indexbuffer, data);
}
void *gs_indexbuffer_get_data(const gs_indexbuffer_t *indexbuffer)
{
if (!gs_valid_p("gs_indexbuffer_get_data", indexbuffer))
return NULL;
return thread_graphics->exports.gs_indexbuffer_get_data(indexbuffer);
}
size_t gs_indexbuffer_get_num_indices(const gs_indexbuffer_t *indexbuffer)
{
if (!gs_valid_p("gs_indexbuffer_get_num_indices", indexbuffer))
return 0;
return thread_graphics->exports.gs_indexbuffer_get_num_indices(
indexbuffer);
}
enum gs_index_type gs_indexbuffer_get_type(const gs_indexbuffer_t *indexbuffer)
{
if (!gs_valid_p("gs_indexbuffer_get_type", indexbuffer))
return (enum gs_index_type)0;
return thread_graphics->exports.gs_indexbuffer_get_type(indexbuffer);
}
#ifdef __APPLE__
/** Platform specific functions */
gs_texture_t *gs_texture_create_from_iosurface(void *iosurf)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_texture_create_from_iosurface", iosurf))
return NULL;
if (!graphics->exports.device_texture_create_from_iosurface)
return NULL;
return graphics->exports.device_texture_create_from_iosurface(
graphics->device, iosurf);
}
bool gs_texture_rebind_iosurface(gs_texture_t *texture, void *iosurf)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid_p("gs_texture_rebind_iosurface", texture))
return false;
if (!graphics->exports.gs_texture_rebind_iosurface)
return false;
return graphics->exports.gs_texture_rebind_iosurface(texture, iosurf);
}
#elif _WIN32
bool gs_gdi_texture_available(void)
{
if (!gs_valid("gs_gdi_texture_available"))
return false;
return thread_graphics->exports.device_gdi_texture_available();
}
bool gs_shared_texture_available(void)
{
if (!gs_valid("gs_shared_texture_available"))
return false;
return thread_graphics->exports.device_shared_texture_available();
}
bool gs_get_duplicator_monitor_info(int monitor_idx,
struct gs_monitor_info *monitor_info)
{
if (!gs_valid_p("gs_get_duplicator_monitor_info", monitor_info))
return false;
if (!thread_graphics->exports.device_get_duplicator_monitor_info)
return false;
return thread_graphics->exports.device_get_duplicator_monitor_info(
thread_graphics->device, monitor_idx,
monitor_info);
}
gs_duplicator_t *gs_duplicator_create(int monitor_idx)
{
if (!gs_valid("gs_duplicator_create"))
return NULL;
if (!thread_graphics->exports.device_duplicator_create)
return NULL;
return thread_graphics->exports.device_duplicator_create(
thread_graphics->device, monitor_idx);
}
void gs_duplicator_destroy(gs_duplicator_t *duplicator)
{
if (!gs_valid("gs_duplicator_destroy"))
return;
if (!duplicator)
return;
if (!thread_graphics->exports.gs_duplicator_destroy)
return;
thread_graphics->exports.gs_duplicator_destroy(duplicator);
}
bool gs_duplicator_update_frame(gs_duplicator_t *duplicator)
{
if (!gs_valid_p("gs_duplicator_update_frame", duplicator))
return false;
if (!thread_graphics->exports.gs_duplicator_get_texture)
return false;
return thread_graphics->exports.gs_duplicator_update_frame(duplicator);
}
gs_texture_t *gs_duplicator_get_texture(gs_duplicator_t *duplicator)
{
if (!gs_valid_p("gs_duplicator_get_texture", duplicator))
return NULL;
if (!thread_graphics->exports.gs_duplicator_get_texture)
return NULL;
return thread_graphics->exports.gs_duplicator_get_texture(duplicator);
}
/** creates a windows GDI-lockable texture */
gs_texture_t *gs_texture_create_gdi(uint32_t width, uint32_t height)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_texture_create_gdi"))
return NULL;
if (graphics->exports.device_texture_create_gdi)
return graphics->exports.device_texture_create_gdi(
graphics->device, width, height);
return NULL;
}
void *gs_texture_get_dc(gs_texture_t *gdi_tex)
{
if (!gs_valid_p("gs_texture_release_dc", gdi_tex))
return NULL;
if (thread_graphics->exports.gs_texture_get_dc)
return thread_graphics->exports.gs_texture_get_dc(gdi_tex);
return NULL;
}
void gs_texture_release_dc(gs_texture_t *gdi_tex)
{
if (!gs_valid_p("gs_texture_release_dc", gdi_tex))
return;
if (thread_graphics->exports.gs_texture_release_dc)
thread_graphics->exports.gs_texture_release_dc(gdi_tex);
}
gs_texture_t *gs_texture_open_shared(uint32_t handle)
{
graphics_t *graphics = thread_graphics;
if (!gs_valid("gs_texture_open_shared"))
return NULL;
if (graphics->exports.device_texture_open_shared)
return graphics->exports.device_texture_open_shared(
graphics->device, handle);
return NULL;
}
#endif