Skip to content

Commit

Permalink
renderer: micro optimizations
Browse files Browse the repository at this point in the history
  • Loading branch information
@JaCzekanski
JaCzekanski committed May 1, 2019
1 parent 974c366 commit e0ce3af
Show file tree
Hide file tree
Showing 3 changed files with 77 additions and 67 deletions.
4 changes: 2 additions & 2 deletions src/device/gpu/psx_color.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -60,7 +60,7 @@ union PSXColor {
}
};

inline uint32_t to15bit(uint8_t r, uint8_t g, uint8_t b) {
constexpr uint32_t to15bit(uint8_t r, uint8_t g, uint8_t b) {
uint32_t newColor = 0;
newColor |= (r & 0xf8) >> 3;
newColor |= (g & 0xf8) << 2;
Expand All @@ -80,7 +80,7 @@ inline uint32_t to15bit(uint32_t color) {
return c.raw;
}

inline uint32_t to24bit(uint16_t color) {
constexpr uint32_t to24bit(uint16_t color) {
uint32_t newColor = 0;
newColor |= (color & 0x7c00) << 1;
newColor |= (color & 0x3e0) >> 2;
Expand Down
126 changes: 69 additions & 57 deletions src/device/gpu/render/render_polygon.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5,42 +5,48 @@
#include "texture_utils.h"
#include "utils/macros.h"

using glm::ivec2;
using glm::ivec3;
using glm::uvec2;
using glm::vec3;
using gpu::GPU;
using gpu::Vertex;

#undef VRAM
#define VRAM ((uint16_t(*)[gpu::VRAM_WIDTH])gpu->vram.data())

int ditherTable[4][4] = {
const int8_t ditherTable[4][4] = {
{-4, +0, -3, +1}, //
{+2, -2, +3, -1}, //
{-3, +1, -4, +0}, //
{+3, -1, +2, -2} //
};

int orient2d(const glm::ivec2& a, const glm::ivec2& b, const glm::ivec2& c) {
int orient2d(const ivec2& a, const ivec2& b, const ivec2& c) { //
return (b.x - a.x) * (c.y - a.y) - (b.y - a.y) * (c.x - a.x);
}

bool isCw(const glm::ivec2 v[3]) {
glm::vec3 a = glm::vec3(v[0].x, v[0].y, 1);
glm::vec3 b = glm::vec3(v[1].x, v[1].y, 1);
glm::vec3 c = glm::vec3(v[2].x, v[2].y, 1);
bool isCw(const ivec2 v[3]) {
vec3 a = vec3(v[0].x, v[0].y, 1);
vec3 b = vec3(v[1].x, v[1].y, 1);
vec3 c = vec3(v[2].x, v[2].y, 1);

glm::vec3 ab = b - a;
glm::vec3 ac = c - a;
glm::vec3 n = glm::cross(ab, ac);
vec3 ab = b - a;
vec3 ac = c - a;
vec3 n = glm::cross(ab, ac);

return n.z < 0;
}

INLINE glm::ivec2 calculateTexel(glm::ivec3 s, int area, glm::ivec2 tex[3], gpu::GP0_E2 textureWindow) {
glm::ivec2 texel
= glm::ivec2((s.x * tex[0].x + s.y * tex[1].x + s.z * tex[2].x) / area, (s.x * tex[0].y + s.y * tex[1].y + s.z * tex[2].y) / area);
INLINE uvec2 calculateTexel(const ivec3 s, const int area, const ivec2 tex[3], const gpu::GP0_E2 textureWindow) {
uvec2 texel( //
(s.x * tex[0].x + s.y * tex[1].x + s.z * tex[2].x) / area, //
(s.x * tex[0].y + s.y * tex[1].y + s.z * tex[2].y) / area //
);

// Texture is repeated outside of 256x256 window
texel.x %= 256;
texel.y %= 256;
texel.x %= 256u;
texel.y %= 256u;

// Texture masking
// texel = (texel AND(NOT(Mask * 8))) OR((Offset AND Mask) * 8)
Expand All @@ -50,24 +56,26 @@ INLINE glm::ivec2 calculateTexel(glm::ivec3 s, int area, glm::ivec2 tex[3], gpu:
return texel;
}

INLINE PSXColor doShading(glm::ivec3 s, int area, glm::ivec2 p, glm::vec3 color[3], int flags) {
glm::ivec3 outColor((s.x * color[0].r * 255 + s.y * color[1].r * 255 + s.z * color[2].r * 255) / area, //
(s.x * color[0].g * 255 + s.y * color[1].g * 255 + s.z * color[2].g * 255) / area, //
(s.x * color[0].b * 255 + s.y * color[1].b * 255 + s.z * color[2].b * 255) / area);
INLINE PSXColor doShading(const ivec3 s, const int area, const ivec2 p, const ivec3 color[3], const int flags) {
ivec3 outColor( //
(s.x * color[0].r + s.y * color[1].r + s.z * color[2].r) / area, //
(s.x * color[0].g + s.y * color[1].g + s.z * color[2].g) / area, //
(s.x * color[0].b + s.y * color[1].b + s.z * color[2].b) / area //
);

// TODO: THPS2 fading screen doesn't look as it should
if (flags & Vertex::Dithering && !(flags & Vertex::RawTexture)) {
outColor += ditherTable[p.y % 4][p.x % 4];
outColor += ditherTable[p.y % 4u][p.x % 4u];
outColor = glm::clamp(outColor, 0, 255);
}

return to15bit(outColor.r, outColor.g, outColor.b);
}

template <ColorDepth bits>
INLINE void plotPixel(GPU* gpu, glm::ivec2 p, glm::ivec3 s, int area, glm::vec3* color, glm::ivec2 tex[3], glm::ivec2 texPage,
glm::ivec2 clut, int flags, gpu::GP0_E2 textureWindow) {
glm::ivec2 texel = calculateTexel(s, area, tex, textureWindow);
INLINE void plotPixel(GPU* gpu, const ivec2 p, const ivec3 s, const int area, const ivec3 color[3], const ivec2 tex[3], const ivec2 texPage,
const ivec2 clut, const int flags, const gpu::GP0_E2 textureWindow) {
uvec2 texel = calculateTexel(s, area, tex, textureWindow);

PSXColor c;
switch (bits) {
Expand All @@ -81,14 +89,20 @@ INLINE void plotPixel(GPU* gpu, glm::ivec2 p, glm::ivec3 s, int area, glm::vec3*

// If texture blending is enabled
if (bits != ColorDepth::NONE && !(flags & Vertex::RawTexture)) {
glm::vec3 brightness;
vec3 brightness;

if (flags & Vertex::GouroudShading) {
brightness = glm::vec3((s.x * color[0].r + s.y * color[1].r + s.z * color[2].r) / area, //
(s.x * color[0].g + s.y * color[1].g + s.z * color[2].g) / area, //
(s.x * color[0].b + s.y * color[1].b + s.z * color[2].b) / area);
// TODO: Get rid of float colors
vec3 fcolor[3];
fcolor[0] = vec3(color[0]) / 255.f;
fcolor[1] = vec3(color[1]) / 255.f;
fcolor[2] = vec3(color[2]) / 255.f;
brightness = vec3( //
(s.x * fcolor[0].r + s.y * fcolor[1].r + s.z * fcolor[2].r) / area, //
(s.x * fcolor[0].g + s.y * fcolor[1].g + s.z * fcolor[2].g) / area, //
(s.x * fcolor[0].b + s.y * fcolor[1].b + s.z * fcolor[2].b) / area);
} else { // Flat shading
brightness = glm::vec3(color[0]);
brightness = vec3(color[0]) / 255.f;
}

c = c * (brightness * 2.f);
Expand All @@ -113,41 +127,39 @@ INLINE void plotPixel(GPU* gpu, glm::ivec2 p, glm::ivec3 s, int area, glm::vec3*
}

template <ColorDepth bits>
void triangle(GPU* gpu, glm::ivec2 pos[3], glm::vec3 color[3], glm::ivec2 tex[3], glm::ivec2 texPage, glm::ivec2 clut, int flags,
gpu::GP0_E2 textureWindow) {
// clang-format off
glm::ivec2 min = glm::ivec2(
gpu->minDrawingX(std::min({ pos[0].x, pos[1].x, pos[2].x })),
gpu->minDrawingY(std::min({ pos[0].y, pos[1].y, pos[2].y }))
INLINE void triangle(GPU* gpu, const ivec2 pos[3], const ivec3 color[3], const ivec2 tex[3], const ivec2 texPage, const ivec2 clut,
const int flags, const gpu::GP0_E2 textureWindow) {
const ivec2 min = ivec2( //
gpu->minDrawingX(std::min({pos[0].x, pos[1].x, pos[2].x})), //
gpu->minDrawingY(std::min({pos[0].y, pos[1].y, pos[2].y})) //
);
glm::ivec2 max = glm::ivec2(
gpu->maxDrawingX(std::max({ pos[0].x, pos[1].x, pos[2].x })),
gpu->maxDrawingY(std::max({ pos[0].y, pos[1].y, pos[2].y }))
const ivec2 max = ivec2( //
gpu->maxDrawingX(std::max({pos[0].x, pos[1].x, pos[2].x})), //
gpu->maxDrawingY(std::max({pos[0].y, pos[1].y, pos[2].y})) //
);
// clang-format on

// https://fgiesen.wordpress.com/2013/02/10/optimizing-the-basic-rasterizer/
int A01 = pos[0].y - pos[1].y;
int B01 = pos[1].x - pos[0].x;
const int A01 = pos[0].y - pos[1].y;
const int B01 = pos[1].x - pos[0].x;

int A12 = pos[1].y - pos[2].y;
int B12 = pos[2].x - pos[1].x;
const int A12 = pos[1].y - pos[2].y;
const int B12 = pos[2].x - pos[1].x;

int A20 = pos[2].y - pos[0].y;
int B20 = pos[0].x - pos[2].x;
const int A20 = pos[2].y - pos[0].y;
const int B20 = pos[0].x - pos[2].x;

glm::ivec2 minp = glm::ivec2(min.x, min.y);
const ivec2 minp = ivec2(min.x, min.y);
int w0_row = orient2d(pos[1], pos[2], minp);
int w1_row = orient2d(pos[2], pos[0], minp);
int w2_row = orient2d(pos[0], pos[1], minp);

int area = orient2d(pos[0], pos[1], pos[2]);
const int area = orient2d(pos[0], pos[1], pos[2]);
if (area == 0) return;

glm::ivec2 p;
ivec2 p;

for (p.y = min.y; p.y < max.y; p.y++) {
glm::ivec3 is = glm::ivec3(w0_row, w1_row, w2_row);
ivec3 is = ivec3(w0_row, w1_row, w2_row);
for (p.x = min.x; p.x < max.x; p.x++) {
if ((is.x | is.y | is.z) >= 0) {
plotPixel<bits>(gpu, p, is, area, color, tex, texPage, clut, flags, textureWindow);
Expand All @@ -165,18 +177,18 @@ void triangle(GPU* gpu, glm::ivec2 pos[3], glm::vec3 color[3], glm::ivec2 tex[3]

// TODO: Render in batches
void Render::drawTriangle(GPU* gpu, Vertex v[3]) {
glm::ivec2 pos[3];
glm::vec3 color[3];
glm::ivec2 texcoord[3];
glm::ivec2 texpage;
glm::ivec2 clut;
ivec2 pos[3];
ivec3 color[3];
ivec2 texcoord[3];
ivec2 texpage;
ivec2 clut;
int bits;
int flags;
gpu::GP0_E2 textureWindow;
for (int j = 0; j < 3; j++) {
pos[j] = glm::ivec2(v[j].position[0], v[j].position[1]);
color[j] = glm::vec3(v[j].color[0] / 255.f, v[j].color[1] / 255.f, v[j].color[2] / 255.f);
texcoord[j] = glm::ivec2(v[j].texcoord[0], v[j].texcoord[1]);
pos[j] = ivec2(v[j].position[0], v[j].position[1]);
color[j] = ivec3(v[j].color[0], v[j].color[1], v[j].color[2]);
texcoord[j] = ivec2(v[j].texcoord[0], v[j].texcoord[1]);
}

// TODO: Remove this hack
Expand All @@ -185,8 +197,8 @@ void Render::drawTriangle(GPU* gpu, Vertex v[3]) {
std::swap(color[1], color[2]);
std::swap(texcoord[1], texcoord[2]);
}
texpage = glm::ivec2(v[0].texpage[0], v[0].texpage[1]);
clut = glm::ivec2(v[0].clut[0], v[0].clut[1]);
texpage = ivec2(v[0].texpage[0], v[0].texpage[1]);
clut = ivec2(v[0].clut[0], v[0].clut[1]);
bits = v[0].bitcount;
flags = v[0].flags;
textureWindow = v[0].textureWindow;
Expand Down
14 changes: 6 additions & 8 deletions src/device/gpu/render/texture_utils.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -6,21 +6,19 @@

enum class ColorDepth { NONE, BIT_4, BIT_8, BIT_16 };

INLINE uint16_t tex4bit(gpu::GPU* gpu, glm::ivec2 tex, glm::ivec2 texPage, glm::ivec2 clut) {
// Using unsigned vectors allows compiler to generate slightly faster division code
INLINE uint16_t tex4bit(gpu::GPU* gpu, glm::uvec2 tex, glm::uvec2 texPage, glm::uvec2 clut) {
uint16_t index = gpuVRAM[texPage.y + tex.y][texPage.x + tex.x / 4];
uint16_t entry = (index >> ((tex.x & 3) * 4)) & 0xf;
uint8_t entry = (index >> ((tex.x & 3) * 4)) & 0xf;
return gpuVRAM[clut.y][clut.x + entry];
}

INLINE uint16_t tex8bit(gpu::GPU* gpu, glm::ivec2 tex, glm::ivec2 texPage, glm::ivec2 clut) {
INLINE uint16_t tex8bit(gpu::GPU* gpu, glm::uvec2 tex, glm::uvec2 texPage, glm::uvec2 clut) {
uint16_t index = gpuVRAM[texPage.y + tex.y][texPage.x + tex.x / 2];
uint16_t entry = (index >> ((tex.x & 1) * 8)) & 0xff;
uint8_t entry = (index >> ((tex.x & 1) * 8)) & 0xff;
return gpuVRAM[clut.y][clut.x + entry];
}

INLINE uint16_t tex16bit(gpu::GPU* gpu, glm::ivec2 tex, glm::ivec2 texPage) {
return gpuVRAM[texPage.y + tex.y][texPage.x + tex.x];
// TODO: In PSOne BIOS colors are swapped (r == b, g == g, b == r, k == k)
}
INLINE uint16_t tex16bit(gpu::GPU* gpu, glm::uvec2 tex, glm::uvec2 texPage) { return gpuVRAM[texPage.y + tex.y][texPage.x + tex.x]; }

#undef gpuVRAM

0 comments on commit e0ce3af

Please sign in to comment.