Ports/perfect_dark
1
0 Fork 1
mirror of https://github.com/perfect-dark-pc-port/perfect_dark.git synced 2026-10-23 15:52:41 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'fgs/port' into port-x64

Browse Source
This commit is contained in:
Raf committed 2024-08-08 17:17:20 +08:00
2 parents 8b1e81f32a bdbf0ab789
commit b44dbb16b5
22 files changed
+449 -246
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

port/fast3d/gfx_api.h
+1 -3
View File
@@ -44,7 +44,7 @@ void gfx_start_frame(void);
void gfx_run(Gfx* commands);
void gfx_end_frame(void);
void gfx_set_target_fps(int);
void gfx_set_maximum_frame_latency(int latency);
void gfx_set_texture_filter(enum FilteringMode mode);
void gfx_texture_cache_clear(void);
void gfx_texture_cache_delete(const uint8_t *orig_addr);
int gfx_create_framebuffer(uint32_t width, uint32_t height, int upscale, int autoresize);
@@ -52,7 +52,5 @@ void gfx_resize_framebuffer(int fb, uint32_t width, uint32_t height, int upscale
void gfx_set_framebuffer(int fb, float noise_scale) ;
void gfx_reset_framebuffer(void);
void gfx_copy_framebuffer(int fb_dst, int fb_src, int left, int top, int use_back);
void gfx_get_pixel_depth_prepare(float x, float y);
uint16_t gfx_get_pixel_depth(float x, float y);
#endif
port/fast3d/gfx_opengl.cpp
+195 -160
View File
@@ -32,6 +32,7 @@ struct ShaderProgram {
uint8_t num_attribs;
GLint frame_count_location;
GLint noise_scale_location;
GLint three_point_filter_locations[2];
};
struct Framebuffer {
@@ -54,6 +55,7 @@ static std::vector<Framebuffer> framebuffers;
static size_t current_framebuffer;
static float current_noise_scale;
static FilteringMode current_filter_mode = FILTER_LINEAR;
static bool current_textures_linear_filter[2] = {false, false};
static GLenum gl_mirror_clamp = GL_MIRROR_CLAMP_TO_EDGE;
@@ -94,6 +96,12 @@ static void gfx_opengl_set_uniforms(struct ShaderProgram* prg) {
if (prg->noise_scale_location >= 0) {
glUniform1f(prg->noise_scale_location, current_noise_scale);
}
if (prg->three_point_filter_locations[0] >= 0) {
glUniform1i(prg->three_point_filter_locations[0], current_textures_linear_filter[0]);
}
if (prg->three_point_filter_locations[1] >= 0) {
glUniform1i(prg->three_point_filter_locations[1], current_textures_linear_filter[1]);
}
}
static void gfx_opengl_unload_shader(struct ShaderProgram* old_prg) {
@@ -234,144 +242,124 @@ static struct ShaderProgram* gfx_opengl_create_and_load_new_shader(uint64_t shad
size_t num_floats = 4;
// Vertex shader
#ifdef __APPLE__
append_line(vs_buf, &vs_len, "#version 410 core");
append_line(vs_buf, &vs_len, "in vec4 aVtxPos;");
append_line(vs_buf, &vs_len, "#define INPUT in");
append_line(vs_buf, &vs_len, "#define OUTPUT out");
#else
append_line(vs_buf, &vs_len, "#version 110");
append_line(vs_buf, &vs_len, "attribute vec4 aVtxPos;");
append_line(vs_buf, &vs_len, "#define INPUT attribute");
append_line(vs_buf, &vs_len, "#define OUTPUT varying");
#endif
append_line(vs_buf, &vs_len, "INPUT vec4 aVtxPos;");
for (int i = 0; i < 2; i++) {
if (cc_features.used_textures[i]) {
#ifdef __APPLE__
vs_len += sprintf(vs_buf + vs_len, "in vec2 aTexCoord%d;\n", i);
vs_len += sprintf(vs_buf + vs_len, "out vec2 vTexCoord%d;\n", i);
#else
vs_len += sprintf(vs_buf + vs_len, "attribute vec2 aTexCoord%d;\n", i);
vs_len += sprintf(vs_buf + vs_len, "varying vec2 vTexCoord%d;\n", i);
#endif
vs_len += sprintf(vs_buf + vs_len, "INPUT vec2 aTexCoord%d;\n", i);
vs_len += sprintf(vs_buf + vs_len, "OUTPUT vec2 vTexCoord%d;\n", i);
num_floats += 2;
for (int j = 0; j < 2; j++) {
if (cc_features.clamp[i][j]) {
#ifdef __APPLE__
vs_len += sprintf(vs_buf + vs_len, "in float aTexClamp%s%d;\n", j == 0 ? "S" : "T", i);
vs_len += sprintf(vs_buf + vs_len, "out float vTexClamp%s%d;\n", j == 0 ? "S" : "T", i);
#else
vs_len += sprintf(vs_buf + vs_len, "attribute float aTexClamp%s%d;\n", j == 0 ? "S" : "T", i);
vs_len += sprintf(vs_buf + vs_len, "varying float vTexClamp%s%d;\n", j == 0 ? "S" : "T", i);
#endif
vs_len += sprintf(vs_buf + vs_len, "INPUT float aTexClamp%s%d;\n", j == 0 ? "S" : "T", i);
vs_len += sprintf(vs_buf + vs_len, "OUTPUT float vTexClamp%s%d;\n", j == 0 ? "S" : "T", i);
num_floats += 1;
}
}
}
}
if (cc_features.opt_fog) {
#ifdef __APPLE__
append_line(vs_buf, &vs_len, "in vec4 aFog;");
append_line(vs_buf, &vs_len, "out vec4 vFog;");
#else
append_line(vs_buf, &vs_len, "attribute vec4 aFog;");
append_line(vs_buf, &vs_len, "varying vec4 vFog;");
#endif
append_line(vs_buf, &vs_len, "INPUT vec4 aFog;");
append_line(vs_buf, &vs_len, "OUTPUT vec4 vFog;");
num_floats += 4;
}
if (cc_features.opt_grayscale) {
#ifdef __APPLE__
append_line(vs_buf, &vs_len, "in vec4 aGrayscaleColor;");
append_line(vs_buf, &vs_len, "out vec4 vGrayscaleColor;");
#else
append_line(vs_buf, &vs_len, "attribute vec4 aGrayscaleColor;");
append_line(vs_buf, &vs_len, "varying vec4 vGrayscaleColor;");
#endif
append_line(vs_buf, &vs_len, "INPUT vec4 aGrayscaleColor;");
append_line(vs_buf, &vs_len, "OUTPUT vec4 vGrayscaleColor;");
num_floats += 4;
}
for (int i = 0; i < cc_features.num_inputs; i++) {
#ifdef __APPLE__
vs_len += sprintf(vs_buf + vs_len, "in vec%d aInput%d;\n", cc_features.opt_alpha ? 4 : 3, i + 1);
vs_len += sprintf(vs_buf + vs_len, "out vec%d vInput%d;\n", cc_features.opt_alpha ? 4 : 3, i + 1);
#else
vs_len += sprintf(vs_buf + vs_len, "attribute vec%d aInput%d;\n", cc_features.opt_alpha ? 4 : 3, i + 1);
vs_len += sprintf(vs_buf + vs_len, "varying vec%d vInput%d;\n", cc_features.opt_alpha ? 4 : 3, i + 1);
#endif
vs_len += sprintf(vs_buf + vs_len, "INPUT vec%d aInput%d;\n", cc_features.opt_alpha ? 4 : 3, i + 1);
vs_len += sprintf(vs_buf + vs_len, "OUTPUT vec%d vInput%d;\n", cc_features.opt_alpha ? 4 : 3, i + 1);
num_floats += cc_features.opt_alpha ? 4 : 3;
}
append_line(vs_buf, &vs_len, "void main() {");
for (int i = 0; i < 2; i++) {
if (cc_features.used_textures[i]) {
vs_len += sprintf(vs_buf + vs_len, "vTexCoord%d = aTexCoord%d;\n", i, i);
vs_len += sprintf(vs_buf + vs_len, " vTexCoord%d = aTexCoord%d;\n", i, i);
for (int j = 0; j < 2; j++) {
if (cc_features.clamp[i][j]) {
vs_len += sprintf(vs_buf + vs_len, "vTexClamp%s%d = aTexClamp%s%d;\n", j == 0 ? "S" : "T", i,
vs_len += sprintf(vs_buf + vs_len, " vTexClamp%s%d = aTexClamp%s%d;\n", j == 0 ? "S" : "T", i,
j == 0 ? "S" : "T", i);
}
}
}
}
if (cc_features.opt_fog) {
append_line(vs_buf, &vs_len, "vFog = aFog;");
append_line(vs_buf, &vs_len, " vFog = aFog;");
}
if (cc_features.opt_grayscale) {
append_line(vs_buf, &vs_len, "vGrayscaleColor = aGrayscaleColor;");
append_line(vs_buf, &vs_len, " vGrayscaleColor = aGrayscaleColor;");
}
for (int i = 0; i < cc_features.num_inputs; i++) {
vs_len += sprintf(vs_buf + vs_len, "vInput%d = aInput%d;\n", i + 1, i + 1);
vs_len += sprintf(vs_buf + vs_len, " vInput%d = aInput%d;\n", i + 1, i + 1);
}
append_line(vs_buf, &vs_len, "gl_Position = aVtxPos;");
append_line(vs_buf, &vs_len, " gl_Position = aVtxPos;");
append_line(vs_buf, &vs_len, "}");
// Fragment shader
#ifdef __APPLE__
append_line(fs_buf, &fs_len, "#version 410 core");
append_line(fs_buf, &fs_len, "#define INPUT in");
append_line(fs_buf, &fs_len, "#define OUTPUT_COLOR outColor");
append_line(fs_buf, &fs_len, "#define SAMPLE_TEX(tex, uv) texture(tex, uv)");
#else
append_line(fs_buf, &fs_len, "#version 130");
append_line(fs_buf, &fs_len, "#define INPUT varying");
append_line(fs_buf, &fs_len, "#define OUTPUT_COLOR gl_FragColor");
append_line(fs_buf, &fs_len, "#define SAMPLE_TEX(tex, uv) texture2D(tex, uv)");
#endif
// Reference approach to color wrapping as per GLideN64
// Return wrapped value of x in interval [low, high)
append_line(fs_buf, &fs_len, "#define WRAP(x, low, high) mod((x)-(low), (high)-(low)) + (low)");
append_line(fs_buf, &fs_len, "#define TEX_OFFSET(tex, uv, texSize, off) SAMPLE_TEX(tex, uv - (off)/texSize)");
// append_line(fs_buf, &fs_len, "precision mediump float;");
for (int i = 0; i < 2; i++) {
if (cc_features.used_textures[i]) {
#ifdef __APPLE__
fs_len += sprintf(fs_buf + fs_len, "in vec2 vTexCoord%d;\n", i);
#else
fs_len += sprintf(fs_buf + fs_len, "varying vec2 vTexCoord%d;\n", i);
#endif
fs_len += sprintf(fs_buf + fs_len, "INPUT vec2 vTexCoord%d;\n", i);
for (int j = 0; j < 2; j++) {
if (cc_features.clamp[i][j]) {
#ifdef __APPLE__
fs_len += sprintf(fs_buf + fs_len, "in float vTexClamp%s%d;\n", j == 0 ? "S" : "T", i);
#else
fs_len += sprintf(fs_buf + fs_len, "varying float vTexClamp%s%d;\n", j == 0 ? "S" : "T", i);
#endif
fs_len += sprintf(fs_buf + fs_len, "INPUT float vTexClamp%s%d;\n", j == 0 ? "S" : "T", i);
}
}
}
}
if (cc_features.opt_fog) {
#ifdef __APPLE__
append_line(fs_buf, &fs_len, "in vec4 vFog;");
#else
append_line(fs_buf, &fs_len, "varying vec4 vFog;");
#endif
append_line(fs_buf, &fs_len, "INPUT vec4 vFog;");
}
if (cc_features.opt_grayscale) {
#ifdef __APPLE__
append_line(fs_buf, &fs_len, "in vec4 vGrayscaleColor;");
#else
append_line(fs_buf, &fs_len, "varying vec4 vGrayscaleColor;");
#endif
append_line(fs_buf, &fs_len, "INPUT vec4 vGrayscaleColor;");
}
for (int i = 0; i < cc_features.num_inputs; i++) {
#ifdef __APPLE__
fs_len += sprintf(fs_buf + fs_len, "in vec%d vInput%d;\n", cc_features.opt_alpha ? 4 : 3, i + 1);
#else
fs_len += sprintf(fs_buf + fs_len, "varying vec%d vInput%d;\n", cc_features.opt_alpha ? 4 : 3, i + 1);
#endif
fs_len += sprintf(fs_buf + fs_len, "INPUT vec%d vInput%d;\n", cc_features.opt_alpha ? 4 : 3, i + 1);
}
if (cc_features.used_textures[0]) {
append_line(fs_buf, &fs_len, "uniform sampler2D uTex0;");
if (current_filter_mode == FILTER_THREE_POINT)
append_line(fs_buf, &fs_len, "uniform int three_point_filter0;");
}
if (cc_features.used_textures[1]) {
append_line(fs_buf, &fs_len, "uniform sampler2D uTex1;");
if (current_filter_mode == FILTER_THREE_POINT)
append_line(fs_buf, &fs_len, "uniform int three_point_filter1;");
}
append_line(fs_buf, &fs_len, "uniform int frame_count;");
@@ -382,46 +370,52 @@ static struct ShaderProgram* gfx_opengl_create_and_load_new_shader(uint64_t shad
append_line(fs_buf, &fs_len, " return fract(sin(random) * 143758.5453);");
append_line(fs_buf, &fs_len, "}");
if (current_filter_mode == FILTER_THREE_POINT) {
append_line(fs_buf, &fs_len, "vec4 filter3point(in sampler2D tex, in vec2 texCoord, in vec2 texSize) {");
append_line(fs_buf, &fs_len, " vec2 offset = fract(texCoord*texSize - vec2(0.5));");
append_line(fs_buf, &fs_len, " offset -= step(1.0, offset.x + offset.y);");
append_line(fs_buf, &fs_len, " vec4 c0 = TEX_OFFSET(tex, texCoord, texSize, offset);");
append_line(fs_buf, &fs_len, " vec4 c1 = TEX_OFFSET(tex, texCoord, texSize, vec2(offset.x - sign(offset.x), offset.y));");
append_line(fs_buf, &fs_len, " vec4 c2 = TEX_OFFSET(tex, texCoord, texSize, vec2(offset.x, offset.y - sign(offset.y)));");
append_line(fs_buf, &fs_len, " return c0 + abs(offset.x)*(c1-c0) + abs(offset.y)*(c2-c0);");
append_line(fs_buf, &fs_len, "}");
}
if (cc_features.opt_blur) {
// blur filter, used for menu backgrounds
// used to be two for loops from 0 to 4, but apparently intel drivers crashed trying to unroll it
// used to have a const weight array, but apparently drivers for the GT620 don't like const array initializers
append_line(fs_buf, &fs_len, R"(
lowp vec4 hookTexture2D(in sampler2D t, in vec2 uv, in vec2 tsize) {
lowp vec4 cw = vec4(0.0);
for (int i = 0; i < 16; ++i) {
vec2 xy = vec2(float(i & 3), float(i >> 2));
lowp float w = 0.009947 - length(xy) * 0.001;
cw += vec4(texture2D(t, uv + (vec2(-1.5) + xy) / tsize).rgb * w, w);
}
return vec4(cw.rgb / cw.a, 1.0);
})"
);
} else if (current_filter_mode == FILTER_THREE_POINT) {
#if __APPLE__
append_line(fs_buf, &fs_len, "#define TEX_OFFSET(off) texture(tex, texCoord - (off)/texSize)");
#else
append_line(fs_buf, &fs_len, "#define TEX_OFFSET(off) texture2D(tex, texCoord - (off)/texSize)");
#endif
append_line(fs_buf, &fs_len, "vec4 filter3point(in sampler2D tex, in vec2 texCoord, in vec2 texSize) {");
append_line(fs_buf, &fs_len, " vec2 offset = fract(texCoord*texSize - vec2(0.5));");
append_line(fs_buf, &fs_len, " offset -= step(1.0, offset.x + offset.y);");
append_line(fs_buf, &fs_len, " vec4 c0 = TEX_OFFSET(offset);");
append_line(fs_buf, &fs_len, " vec4 c1 = TEX_OFFSET(vec2(offset.x - sign(offset.x), offset.y));");
append_line(fs_buf, &fs_len, " vec4 c2 = TEX_OFFSET(vec2(offset.x, offset.y - sign(offset.y)));");
append_line(fs_buf, &fs_len, " return c0 + abs(offset.x)*(c1-c0) + abs(offset.y)*(c2-c0);");
append_line(fs_buf, &fs_len, "}");
append_line(fs_buf, &fs_len, "vec4 hookTexture2D(in sampler2D tex, in vec2 uv, in vec2 texSize) {");
append_line(fs_buf, &fs_len, " return filter3point(tex, uv, texSize);");
if (current_filter_mode == FILTER_THREE_POINT)
append_line(fs_buf, &fs_len, "lowp vec4 hookTexture2D(in sampler2D t, in vec2 uv, in vec2 texSize, in int three_point_filter) {");
else
append_line(fs_buf, &fs_len, "lowp vec4 hookTexture2D(in sampler2D t, in vec2 uv, in vec2 texSize) {");
append_line(fs_buf, &fs_len, " lowp vec4 cw = vec4(0.0);");
append_line(fs_buf, &fs_len, " for (int i = 0; i < 16; ++i) {");
append_line(fs_buf, &fs_len, " vec2 xy = vec2(float(i & 3), float(i >> 2));");
append_line(fs_buf, &fs_len, " lowp float w = 0.009947 - length(xy) * 0.001;");
append_line(fs_buf, &fs_len, " vec2 scaled_uv = uv + (vec2(-1.5) + xy) / texSize;");
if (current_filter_mode == FILTER_THREE_POINT)
append_line(fs_buf, &fs_len, " lowp vec4 tex = mix(SAMPLE_TEX(t, scaled_uv), filter3point(t, scaled_uv, texSize), three_point_filter);");
else
append_line(fs_buf, &fs_len, " lowp vec4 tex = SAMPLE_TEX(t, scaled_uv);");
append_line(fs_buf, &fs_len, " cw += vec4(tex.rgb * w, w);");
append_line(fs_buf, &fs_len, " }");
append_line(fs_buf, &fs_len, " return vec4(cw.rgb / cw.a, 1.0);");
append_line(fs_buf, &fs_len, "}");
} else {
append_line(fs_buf, &fs_len, "vec4 hookTexture2D(in sampler2D tex, in vec2 uv, in vec2 texSize) {");
#if __APPLE__
append_line(fs_buf, &fs_len, " return texture(tex, uv);");
#else
append_line(fs_buf, &fs_len, " return texture2D(tex, uv);");
#endif
append_line(fs_buf, &fs_len, "}");
if (current_filter_mode == FILTER_THREE_POINT) {
append_line(fs_buf, &fs_len, "vec4 hookTexture2D(in sampler2D tex, in vec2 uv, in vec2 texSize, in int three_point_filter) {");
append_line(fs_buf, &fs_len, " return mix(SAMPLE_TEX(tex, uv), filter3point(tex, uv, texSize), three_point_filter);");
append_line(fs_buf, &fs_len, "}");
} else {
append_line(fs_buf, &fs_len, "vec4 hookTexture2D(in sampler2D tex, in vec2 uv, in vec2 texSize) {");
append_line(fs_buf, &fs_len, " return SAMPLE_TEX(tex, uv);");
append_line(fs_buf, &fs_len, "}");
}
}
#if __APPLE__
@@ -430,44 +424,43 @@ static struct ShaderProgram* gfx_opengl_create_and_load_new_shader(uint64_t shad
append_line(fs_buf, &fs_len, "void main() {");
// Reference approach to color wrapping as per GLideN64
// Return wrapped value of x in interval [low, high)
append_line(fs_buf, &fs_len, "#define WRAP(x, low, high) mod((x)-(low), (high)-(low)) + (low)");
for (int i = 0; i < 2; i++) {
if (cc_features.used_textures[i]) {
bool s = cc_features.clamp[i][0], t = cc_features.clamp[i][1];
fs_len += sprintf(fs_buf + fs_len, "vec2 texSize%d = textureSize(uTex%d, 0);\n", i, i);
fs_len += sprintf(fs_buf + fs_len, " vec2 texSize%d = textureSize(uTex%d, 0);\n", i, i);
if (!s && !t) {
fs_len += sprintf(fs_buf + fs_len, "vec2 vTexCoordAdj%d = vTexCoord%d;\n", i, i);
fs_len += sprintf(fs_buf + fs_len, " vec2 vTexCoordAdj%d = vTexCoord%d;\n", i, i);
} else {
if (s && t) {
fs_len += sprintf(fs_buf + fs_len,
"vec2 vTexCoordAdj%d = clamp(vTexCoord%d, 0.5 / texSize%d, "
" vec2 vTexCoordAdj%d = clamp(vTexCoord%d, 0.5 / texSize%d, "
"vec2(vTexClampS%d, vTexClampT%d));\n",
i, i, i, i, i);
} else if (s) {
fs_len += sprintf(fs_buf + fs_len,
"vec2 vTexCoordAdj%d = vec2(clamp(vTexCoord%d.s, 0.5 / "
" vec2 vTexCoordAdj%d = vec2(clamp(vTexCoord%d.s, 0.5 / "
"texSize%d.s, vTexClampS%d), vTexCoord%d.t);\n",
i, i, i, i, i);
} else {
fs_len += sprintf(fs_buf + fs_len,
"vec2 vTexCoordAdj%d = vec2(vTexCoord%d.s, clamp(vTexCoord%d.t, "
" vec2 vTexCoordAdj%d = vec2(vTexCoord%d.s, clamp(vTexCoord%d.t, "
"0.5 / texSize%d.t, vTexClampT%d));\n",
i, i, i, i, i);
}
}
fs_len += sprintf(fs_buf + fs_len, "vec4 texVal%d = hookTexture2D(uTex%d, vTexCoordAdj%d, texSize%d);\n", i, i, i, i);
if (current_filter_mode == FILTER_THREE_POINT)
fs_len += sprintf(fs_buf + fs_len, " vec4 texVal%d = hookTexture2D(uTex%d, vTexCoordAdj%d, texSize%d, three_point_filter%d);\n", i, i, i, i, i);
else
fs_len += sprintf(fs_buf + fs_len, " vec4 texVal%d = hookTexture2D(uTex%d, vTexCoordAdj%d, texSize%d);\n", i, i, i, i);
}
}
append_line(fs_buf, &fs_len, cc_features.opt_alpha ? "vec4 texel;" : "vec3 texel;");
append_line(fs_buf, &fs_len, cc_features.opt_alpha ? " vec4 texel;" : " vec3 texel;");
for (int c = 0; c < (cc_features.opt_2cyc ? 2 : 1); c++) {
append_str(fs_buf, &fs_len, "texel = ");
append_str(fs_buf, &fs_len, " texel = ");
if (!cc_features.color_alpha_same[c] && cc_features.opt_alpha) {
append_str(fs_buf, &fs_len, "vec4(");
append_formula(fs_buf, &fs_len, cc_features.c[c], cc_features.do_single[c][0],
@@ -484,55 +477,48 @@ static struct ShaderProgram* gfx_opengl_create_and_load_new_shader(uint64_t shad
append_line(fs_buf, &fs_len, ";");
if (c == 0) {
append_str(fs_buf, &fs_len, "texel = WRAP(texel, -1.01, 1.01);");
append_line(fs_buf, &fs_len, " texel = WRAP(texel, -1.01, 1.01);");
}
}
append_str(fs_buf, &fs_len, "texel = WRAP(texel, -0.51, 1.51);");
append_str(fs_buf, &fs_len, "texel = clamp(texel, 0.0, 1.0);");
append_line(fs_buf, &fs_len, " texel = WRAP(texel, -0.51, 1.51);");
append_line(fs_buf, &fs_len, " texel = clamp(texel, 0.0, 1.0);");
// TODO discard if alpha is 0?
if (cc_features.opt_fog) {
if (cc_features.opt_alpha) {
append_line(fs_buf, &fs_len, "texel = vec4(mix(texel.rgb, vFog.rgb, vFog.a), texel.a);");
append_line(fs_buf, &fs_len, " texel = vec4(mix(texel.rgb, vFog.rgb, vFog.a), texel.a);");
} else {
append_line(fs_buf, &fs_len, "texel = mix(texel, vFog.rgb, vFog.a);");
append_line(fs_buf, &fs_len, " texel = mix(texel, vFog.rgb, vFog.a);");
}
}
if (cc_features.opt_texture_edge && cc_features.opt_alpha) {
append_line(fs_buf, &fs_len, "if (texel.a > 0.19) texel.a = 1.0; else discard;");
append_line(fs_buf, &fs_len, " if (texel.a > 0.19) texel.a = 1.0; else discard;");
}
if (cc_features.opt_alpha && cc_features.opt_noise) {
append_line(fs_buf, &fs_len,
"texel.a *= floor(clamp(random(vec3(floor(gl_FragCoord.xy * noise_scale), float(frame_count))) + "
" texel.a *= floor(clamp(random(vec3(floor(gl_FragCoord.xy * noise_scale), float(frame_count))) + "
"texel.a, 0.0, 1.0));");
}
if (cc_features.opt_grayscale) {
append_line(fs_buf, &fs_len, "float intensity = (texel.r + texel.g + texel.b) / 3.0;");
append_line(fs_buf, &fs_len, "vec3 new_texel = vGrayscaleColor.rgb * intensity;");
append_line(fs_buf, &fs_len, "texel.rgb = mix(texel.rgb, new_texel, vGrayscaleColor.a);");
append_line(fs_buf, &fs_len, " float intensity = (texel.r + texel.g + texel.b) / 3.0;");
append_line(fs_buf, &fs_len, " vec3 new_texel = vGrayscaleColor.rgb * intensity;");
append_line(fs_buf, &fs_len, " texel.rgb = mix(texel.rgb, new_texel, vGrayscaleColor.a);");
}
if (cc_features.opt_alpha) {
if (cc_features.opt_alpha_threshold) {
append_line(fs_buf, &fs_len, "if (texel.a < 8.0 / 256.0) discard;");
append_line(fs_buf, &fs_len, " if (texel.a < 8.0 / 256.0) discard;");
}
if (cc_features.opt_invisible) {
append_line(fs_buf, &fs_len, "texel.a = 0.0;");
append_line(fs_buf, &fs_len, " texel.a = 0.0;");
}
#if __APPLE__
append_line(fs_buf, &fs_len, "outColor = texel;");
#else
append_line(fs_buf, &fs_len, "gl_FragColor = texel;");
#endif
append_line(fs_buf, &fs_len, " OUTPUT_COLOR = texel;");
} else {
#if __APPLE__
append_line(fs_buf, &fs_len, "outColor = vec4(texel, 1.0);");
#else
append_line(fs_buf, &fs_len, "gl_FragColor = vec4(texel, 1.0);");
#endif
append_line(fs_buf, &fs_len, " OUTPUT_COLOR = vec4(texel, 1.0);");
}
append_line(fs_buf, &fs_len, "}");
@@ -574,6 +560,11 @@ static struct ShaderProgram* gfx_opengl_create_and_load_new_shader(uint64_t shad
glAttachShader(shader_program, fragment_shader);
glLinkProgram(shader_program);
glDetachShader(shader_program, vertex_shader);
glDetachShader(shader_program, fragment_shader);
glDeleteShader(vertex_shader);
glDeleteShader(fragment_shader);
size_t cnt = 0;
struct ShaderProgram* prg = &shader_program_pool[make_pair(shader_id0, shader_id1)];
@@ -640,6 +631,8 @@ static struct ShaderProgram* gfx_opengl_create_and_load_new_shader(uint64_t shad
prg->frame_count_location = glGetUniformLocation(shader_program, "frame_count");
prg->noise_scale_location = glGetUniformLocation(shader_program, "noise_scale");
prg->three_point_filter_locations[0] = glGetUniformLocation(shader_program, "three_point_filter0");
prg->three_point_filter_locations[1] = glGetUniformLocation(shader_program, "three_point_filter1");
gfx_opengl_load_shader(prg);
@@ -657,6 +650,14 @@ static void gfx_opengl_shader_get_info(struct ShaderProgram* prg, uint8_t* num_i
used_textures[1] = prg->used_textures[1];
}
static void gfx_opengl_clear_shaders(void) {
glUseProgram(0);
for (auto& pair : shader_program_pool) {
glDeleteProgram(pair.second.opengl_program_id);
}
shader_program_pool.clear();
}
static GLuint gfx_opengl_new_texture(void) {
GLuint ret;
glGenTextures(1, &ret);
@@ -667,9 +668,11 @@ static void gfx_opengl_delete_texture(uint32_t texID) {
glDeleteTextures(1, &texID);
}
static void gfx_opengl_select_texture(int tile, GLuint texture_id) {
static void gfx_opengl_select_texture(int tile, GLuint texture_id, bool linear_filter) {
glActiveTexture(GL_TEXTURE0 + tile);
glBindTexture(GL_TEXTURE_2D, texture_id);
current_textures_linear_filter[tile] = linear_filter;
}
static void gfx_opengl_upload_texture(const uint8_t* rgba32_buf, uint32_t width, uint32_t height) {
@@ -699,12 +702,42 @@ static void gfx_opengl_set_sampler_parameters(int tile, bool linear_filter, uint
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, gfx_cm_to_opengl(cmt));
}
static void gfx_opengl_set_depth_test_and_mask(bool depth_test, bool z_upd) {
if (depth_test || z_upd) {
static void gfx_opengl_set_depth_mode(bool depth_test, bool depth_update, bool depth_compare, bool depth_source_prim, uint16_t zmode) {
if (depth_test) {
glEnable(GL_DEPTH_TEST);
glDepthMask(z_upd ? GL_TRUE : GL_FALSE);
glDepthFunc(depth_test ? GL_LEQUAL : GL_ALWAYS);
current_depth_mask = z_upd;
glDepthMask(depth_update ? GL_TRUE : GL_FALSE);
current_depth_mask = depth_update;
if (depth_compare) {
switch (zmode) {
case ZMODE_INTER:
glDepthFunc(GL_LEQUAL);
glDisable(GL_POLYGON_OFFSET_FILL);
glPolygonOffset(0, 0);
break;
case ZMODE_OPA:
case ZMODE_XLU:
if (depth_source_prim) {
glDepthFunc(GL_LEQUAL);
} else {
glDepthFunc(GL_LESS);
}
glDisable(GL_POLYGON_OFFSET_FILL);
glPolygonOffset(0, 0);
break;
case ZMODE_DEC:
glDepthFunc(GL_LEQUAL);
glEnable(GL_POLYGON_OFFSET_FILL);
glPolygonOffset(-2, -2);
break;
}
} else {
glDepthFunc(GL_ALWAYS);
glDisable(GL_POLYGON_OFFSET_FILL);
glPolygonOffset(0, 0);
}
} else {
glDisable(GL_DEPTH_TEST);
}
@@ -714,16 +747,6 @@ static void gfx_opengl_set_depth_range(float znear, float zfar) {
glDepthRange(znear, zfar);
}
static void gfx_opengl_set_zmode_decal(bool zmode_decal) {
if (zmode_decal) {
glPolygonOffset(-2, -2);
glEnable(GL_POLYGON_OFFSET_FILL);
} else {
glPolygonOffset(0, 0);
glDisable(GL_POLYGON_OFFSET_FILL);
}
}
static void gfx_opengl_set_viewport(int x, int y, int width, int height) {
glViewport(x, y, width, height);
}
@@ -1045,12 +1068,23 @@ bool gfx_opengl_start_draw_to_framebuffer(int fb_id, float noise_scale) {
}
}
void gfx_opengl_clear_framebuffer() {
void gfx_opengl_clear_framebuffer(bool clear_color, bool clear_depth) {
glDisable(GL_SCISSOR_TEST);
glDepthMask(GL_TRUE);
glClearColor(0.0f, 0.0f, 0.0f, 1.0f);
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
glDepthMask(current_depth_mask ? GL_TRUE : GL_FALSE);
GLbitfield mask = 0;
if (clear_color) {
glClearColor(0.0f, 0.0f, 0.0f, 1.0f);
mask |= GL_COLOR_BUFFER_BIT;
}
if (clear_depth) {
glDepthMask(GL_TRUE);
mask |= GL_DEPTH_BUFFER_BIT;
}
glClear(mask);
if (clear_depth) {
glDepthMask(current_depth_mask ? GL_TRUE : GL_FALSE);
}
glEnable(GL_SCISSOR_TEST);
}
@@ -1078,6 +1112,8 @@ void gfx_opengl_select_texture_fb(int fb_id) {
// glDisable(GL_DEPTH_TEST);
glActiveTexture(GL_TEXTURE0 + 0);
glBindTexture(GL_TEXTURE_2D, framebuffers[fb_id].clrbuf);
current_textures_linear_filter[0] = true;
}
void gfx_opengl_copy_framebuffer(int fb_dst, int fb_src, int left, int top, bool flip_y, bool use_back) {
@@ -1136,7 +1172,6 @@ void gfx_opengl_copy_framebuffer(int fb_dst, int fb_src, int left, int top, bool
void gfx_opengl_set_texture_filter(FilteringMode mode) {
current_filter_mode = mode;
gfx_texture_cache_clear();
}
FilteringMode gfx_opengl_get_texture_filter(void) {
@@ -1152,13 +1187,13 @@ struct GfxRenderingAPI gfx_opengl_api = {
gfx_opengl_create_and_load_new_shader,
gfx_opengl_lookup_shader,
gfx_opengl_shader_get_info,
gfx_opengl_clear_shaders,
gfx_opengl_new_texture,
gfx_opengl_select_texture,
gfx_opengl_upload_texture,
gfx_opengl_set_sampler_parameters,
gfx_opengl_set_depth_test_and_mask,
gfx_opengl_set_depth_mode,
gfx_opengl_set_depth_range,
gfx_opengl_set_zmode_decal,
gfx_opengl_set_viewport,
gfx_opengl_set_scissor,
gfx_opengl_set_use_alpha,
port/fast3d/gfx_pc.cpp
+32 -41
View File
@@ -125,8 +125,6 @@ static struct RSP {
float aspect_ofs;
float aspect_scale;
float depth_zfar;
struct {
// U0.16
uint16_t s, t;
@@ -197,9 +195,7 @@ static struct RDP {
} rdp;
static struct RenderingState {
uint8_t depth_test_and_mask; // 1: depth test, 2: depth mask
float depth_zfar;
bool decal_mode;
uint8_t depth_mode;
bool alpha_blend;
bool modulate;
struct XYWidthHeight viewport, scissor;
@@ -523,7 +519,7 @@ static bool gfx_texture_cache_lookup(int i, const TextureCacheKey& key) {
TextureCacheNode** n = &rendering_state.textures[i];
if (it != gfx_texture_cache.map.end()) {
gfx_rapi->select_texture(i, it->second.texture_id);
gfx_rapi->select_texture(i, it->second.texture_id, it->second.linear_filter);
*n = &*it;
gfx_texture_cache.lru.splice(gfx_texture_cache.lru.end(), gfx_texture_cache.lru,
it->second.lru_location); // move to back
@@ -551,7 +547,7 @@ static bool gfx_texture_cache_lookup(int i, const TextureCacheKey& key) {
node->second.texture_id = texture_id;
node->second.lru_location = gfx_texture_cache.lru.insert(gfx_texture_cache.lru.end(), { it });
gfx_rapi->select_texture(i, texture_id);
gfx_rapi->select_texture(i, texture_id, false);
gfx_rapi->set_sampler_parameters(i, false, 0, 0);
*n = node;
return false;
@@ -1240,25 +1236,18 @@ static void gfx_sp_tri1(uint8_t vtx1_idx, uint8_t vtx2_idx, uint8_t vtx3_idx, bo
}
}
bool depth_test = (rsp.geometry_mode & G_ZBUFFER) == G_ZBUFFER;
bool depth_mask = (rdp.other_mode_l & Z_UPD) == Z_UPD;
uint8_t depth_test_and_mask = (depth_test ? 1 : 0) | (depth_mask ? 2 : 0);
if (depth_test_and_mask != rendering_state.depth_test_and_mask) {
gfx_flush();
gfx_rapi->set_depth_test_and_mask(depth_test, depth_mask);
rendering_state.depth_test_and_mask = depth_test_and_mask;
}
if (rsp.depth_zfar != rendering_state.depth_zfar) {
gfx_flush();
gfx_rapi->set_depth_range(0.0f, rsp.depth_zfar);
rendering_state.depth_zfar = rsp.depth_zfar;
}
bool depth_test = ((rsp.geometry_mode & G_ZBUFFER) == G_ZBUFFER || (rdp.other_mode_l & G_ZS_PRIM) == G_ZS_PRIM) &&
((rdp.other_mode_h & G_CYC_1CYCLE) == G_CYC_1CYCLE || (rdp.other_mode_h & G_CYC_2CYCLE) == G_CYC_2CYCLE);
bool depth_update = (rdp.other_mode_l & Z_UPD) == Z_UPD;
bool depth_compare = (rdp.other_mode_l & Z_CMP) == Z_CMP;
bool depth_source_prim = (rdp.other_mode_l & G_ZS_PRIM) == G_ZS_PRIM /* && gDP.primDepth.z == 1.0f */;
uint16_t zmode = rdp.other_mode_l & ZMODE_DEC;
uint8_t depth_mode = (depth_test ? 1 : 0) | (depth_update ? 2 : 0) | (depth_compare ? 4 : 0) | (depth_source_prim ? 8 : 0) | (zmode >> 6);
bool zmode_decal = (rdp.other_mode_l & ZMODE_DEC) == ZMODE_DEC;
if (zmode_decal != rendering_state.decal_mode) {
if (depth_mode != rendering_state.depth_mode) {
gfx_flush();
gfx_rapi->set_zmode_decal(zmode_decal);
rendering_state.decal_mode = zmode_decal;
gfx_rapi->set_depth_mode(depth_test, depth_update, depth_compare, depth_source_prim, zmode);
rendering_state.depth_mode = depth_mode;
}
if (rdp.viewport_or_scissor_changed) {
@@ -1733,18 +1722,6 @@ static void gfx_sp_moveword(uint8_t index, uint16_t offset, uintptr_t data) {
case G_MW_SEGMENT:
segmentPointers[(offset >> 2) & 0xff] = data;
break;
case G_MW_PERSPNORM:
// the default z range is around [100, 10000]
// data is 2 / (znear + zfar) represented as a 0.16 fixed point
// => (znear + zfar) = (2 / (data / 65536)) = 131072 / data
constexpr float full_range_mul = 1.f / 11000.f; // that's around the biggest value I got when testing
if (data == 0) {
rsp.depth_zfar = 1.f;
} else {
// sometimes this will overshoot 1 but GL can handle that
rsp.depth_zfar =((131072.f * full_range_mul) / (float)data);
}
break;
}
}
@@ -2504,6 +2481,10 @@ static void gfx_run_dl(Gfx* cmd) {
case G_RDPFLUSH_EXT:
gfx_flush();
break;
case G_CLEAR_DEPTH_EXT:
gfx_flush();
gfx_rapi->clear_framebuffer(false, true);
break;
case G_RDPPIPESYNC:
case G_RDPFULLSYNC:
case G_RDPLOADSYNC:
@@ -2554,8 +2535,6 @@ extern "C" void gfx_init(const GfxInitSettings *settings) {
tex_upload_buffer = (uint8_t*)malloc(max_tex_size * max_tex_size * 4);
}
rsp.depth_zfar = 1.0f;
rsp.lookat[0].dir[0] = rsp.lookat[1].dir[1] = 0x7F;
rsp.current_lookat_coeffs[0][0] = rsp.current_lookat_coeffs[1][1] = 1.f;
rsp.lookat_enabled = true;
@@ -2661,7 +2640,7 @@ extern "C" void gfx_run(Gfx* commands) {
gfx_rapi->start_frame();
gfx_rapi->start_draw_to_framebuffer(game_renders_to_framebuffer ? game_framebuffer : 0,
(float)gfx_current_dimensions.height / SCREEN_HEIGHT);
gfx_rapi->clear_framebuffer();
gfx_rapi->clear_framebuffer(true, false);
rdp.viewport_or_scissor_changed = true;
rendering_state.viewport = {};
rendering_state.scissor = {};
@@ -2671,7 +2650,7 @@ extern "C" void gfx_run(Gfx* commands) {
if (game_renders_to_framebuffer) {
gfx_rapi->start_draw_to_framebuffer(0, 1);
gfx_rapi->clear_framebuffer();
gfx_rapi->clear_framebuffer(true, true);
if (gfx_msaa_level > 1) {
bool different_size = gfx_current_dimensions.width != gfx_current_game_window_viewport.width ||
@@ -2703,6 +2682,18 @@ extern "C" void gfx_set_target_fps(int fps) {
gfx_wapi->set_target_fps(fps);
}
extern "C" void gfx_set_texture_filter(enum FilteringMode mode) {
gfx_texture_cache_clear();
if (rendering_state.shader_program) {
gfx_rapi->unload_shader(rendering_state.shader_program);
rendering_state.shader_program = nullptr;
}
gfx_rapi->clear_shaders();
color_combiner_pool.clear();
prev_combiner = color_combiner_pool.end();
gfx_rapi->set_texture_filter(mode);
}
extern "C" int gfx_create_framebuffer(uint32_t width, uint32_t height, int upscale, int autoresize) {
int fb = gfx_rapi->create_framebuffer();
gfx_resize_framebuffer(fb, width, height, upscale, autoresize);
@@ -2734,7 +2725,7 @@ extern "C" void gfx_resize_framebuffer(int fb, uint32_t width, uint32_t height,
extern "C" void gfx_set_framebuffer(int fb, float noise_scale) {
gfx_rapi->start_draw_to_framebuffer(fb, noise_scale);
gfx_rapi->clear_framebuffer();
gfx_rapi->clear_framebuffer(true, true);
}
extern "C" void gfx_copy_framebuffer(int fb_dst, int fb_src, int left, int top, int use_back) {
port/fast3d/gfx_rendering_api.h
+4 -4
View File
@@ -23,13 +23,13 @@ struct GfxRenderingAPI {
struct ShaderProgram* (*create_and_load_new_shader)(uint64_t shader_id0, uint32_t shader_id1);
struct ShaderProgram* (*lookup_shader)(uint64_t shader_id0, uint32_t shader_id1);
void (*shader_get_info)(struct ShaderProgram* prg, uint8_t* num_inputs, bool used_textures[2]);
void (*clear_shaders)(void);
uint32_t (*new_texture)(void);
void (*select_texture)(int tile, uint32_t texture_id);
void (*select_texture)(int tile, uint32_t texture_id, bool linear_filter);
void (*upload_texture)(const uint8_t* rgba32_buf, uint32_t width, uint32_t height);
void (*set_sampler_parameters)(int sampler, bool linear_filter, uint32_t cms, uint32_t cmt);
void (*set_depth_test_and_mask)(bool depth_test, bool z_upd);
void (*set_depth_mode)(bool depth_test, bool depth_update, bool depth_compare, bool depth_source_prim, uint16_t zmode);
void (*set_depth_range)(float znear, float zfar);
void (*set_zmode_decal)(bool zmode_decal);
void (*set_viewport)(int x, int y, int width, int height);
void (*set_scissor)(int x, int y, int width, int height);
void (*set_use_alpha)(bool use_alpha, bool modulate);
@@ -45,7 +45,7 @@ struct GfxRenderingAPI {
bool can_extract_depth);
bool (*start_draw_to_framebuffer)(int fb_id, float noise_scale);
void (*copy_framebuffer)(int fb_dst, int fb_src, int left, int top, bool flip_y, bool use_back);
void (*clear_framebuffer)(void);
void (*clear_framebuffer)(bool clear_color, bool clear_depth);
void (*resolve_msaa_color_buffer)(int fb_id_target, int fb_id_source);
void* (*get_framebuffer_texture_id)(int fb_id);
void (*select_texture_fb)(int fb_id);
port/src/optionsmenu.c
+16 -5
View File
@@ -702,12 +702,23 @@ static MenuItemHandlerResult menuhandlerMaximizeWindow(s32 operation, struct men
static MenuItemHandlerResult menuhandlerTexFilter(s32 operation, struct menuitem *item, union handlerdata *data)
{
static const char *opts[] = {
"Nearest",
"Bilinear",
"Three Point"
};
switch (operation) {
case MENUOP_GET:
return (videoGetTextureFilter() != 0);
case MENUOP_SET:
videoSetTextureFilter(data->checkbox.value);
case MENUOP_GETOPTIONCOUNT:
data->dropdown.value = ARRAYCOUNT(opts);
break;
case MENUOP_GETOPTIONTEXT:
return (intptr_t)opts[data->dropdown.value];
case MENUOP_SET:
videoSetTextureFilter(data->dropdown.value);
break;
case MENUOP_GETSELECTEDINDEX:
data->dropdown.value = videoGetTextureFilter();
}
return 0;
@@ -826,7 +837,7 @@ struct menuitem g_ExtendedVideoMenuItems[] = {
menuhandlerTexDetail,
},
{
MENUITEMTYPE_CHECKBOX,
MENUITEMTYPE_DROPDOWN,
0,
MENUITEMFLAG_LITERAL_TEXT,
(uintptr_t)"Texture Filtering",
port/src/video.c
+3 -2
View File
@@ -77,7 +77,7 @@ s32 videoInit(void)
}
wmAPI->set_target_fps(vidFramerateLimit); // disabled because vsync is on
renderingAPI->set_texture_filter((enum FilteringMode)texFilter);
gfx_set_texture_filter((enum FilteringMode)texFilter);
initDone = true;
return 0;
@@ -218,8 +218,9 @@ void videoSetMaximizeWindow(s32 fs)
void videoSetTextureFilter(u32 filter)
{
if (filter > FILTER_THREE_POINT) filter = FILTER_THREE_POINT;
if (texFilter == filter) return;
texFilter = filter;
renderingAPI->set_texture_filter((enum FilteringMode)filter);
gfx_set_texture_filter((enum FilteringMode)filter);
}
void videoSetTextureFilter2D(s32 filter)