Make the software implementation use the same clipping code

This commit is contained in:
Luke Benstead 2022-06-13 20:06:04 +01:00
parent 6fb15ee4c7
commit 99ae70a72b

View File

@ -3,11 +3,14 @@
#include <stdlib.h> #include <stdlib.h>
#include <string.h> #include <string.h>
#include "../private.h"
#include "../platform.h" #include "../platform.h"
#include "software.h" #include "software.h"
#include "software/edge_equation.h" #include "software/edge_equation.h"
#include "software/parameter_equation.h" #include "software/parameter_equation.h"
#define CLIP_DEBUG 0
static size_t AVAILABLE_VRAM = 16 * 1024 * 1024; static size_t AVAILABLE_VRAM = 16 * 1024 * 1024;
static Matrix4x4 MATRIX; static Matrix4x4 MATRIX;
@ -23,28 +26,16 @@ static VideoMode vid_mode = {
640, 480 640, 480
}; };
typedef struct GPUVertex {
uint32_t flags;
float x;
float y;
float z;
float u;
float v;
uint8_t bgra[4];
uint8_t obgra[4];
} GPUVertex;
#define MIN(x, y) ((x) < (y) ? (x) : (y)) #define MIN(x, y) ((x) < (y) ? (x) : (y))
#define MAX(x, y) ((x) > (y) ? (x) : (y)) #define MAX(x, y) ((x) > (y) ? (x) : (y))
static void DrawTriangle(GPUVertex* v0, GPUVertex* v1, GPUVertex* v2) { static void DrawTriangle(Vertex* v0, Vertex* v1, Vertex* v2) {
// Compute triangle bounding box. // Compute triangle bounding box.
int minX = MIN(MIN(v0->x, v1->x), v2->x); int minX = MIN(MIN(v0->xyz[0], v1->xyz[0]), v2->xyz[0]);
int maxX = MAX(MAX(v0->x, v1->x), v2->x); int maxX = MAX(MAX(v0->xyz[0], v1->xyz[0]), v2->xyz[0]);
int minY = MIN(MIN(v0->y, v1->y), v2->y); int minY = MIN(MIN(v0->xyz[1], v1->xyz[1]), v2->xyz[1]);
int maxY = MAX(MAX(v0->y, v1->y), v2->y); int maxY = MAX(MAX(v0->xyz[1], v1->xyz[1]), v2->xyz[1]);
// Clip to scissor rect. // Clip to scissor rect.
@ -56,9 +47,9 @@ static void DrawTriangle(GPUVertex* v0, GPUVertex* v1, GPUVertex* v2) {
// Compute edge equations. // Compute edge equations.
EdgeEquation e0, e1, e2; EdgeEquation e0, e1, e2;
EdgeEquationInit(&e0, &v0->x, &v1->x); EdgeEquationInit(&e0, &v0->xyz[0], &v1->xyz[0]);
EdgeEquationInit(&e1, &v1->x, &v2->x); EdgeEquationInit(&e1, &v1->xyz[0], &v2->xyz[0]);
EdgeEquationInit(&e2, &v2->x, &v0->x); EdgeEquationInit(&e2, &v2->xyz[0], &v0->xyz[0]);
float area = 0.5 * (e0.c + e1.c + e2.c); float area = 0.5 * (e0.c + e1.c + e2.c);
@ -66,12 +57,12 @@ static void DrawTriangle(GPUVertex* v0, GPUVertex* v1, GPUVertex* v2) {
* so I just swap the vertex order if something is back-facing * so I just swap the vertex order if something is back-facing
* and we want to render it. Patches welcome! */ * and we want to render it. Patches welcome! */
#define REVERSE_WINDING() \ #define REVERSE_WINDING() \
GPUVertex* tv = v0; \ Vertex* tv = v0; \
v0 = v1; \ v0 = v1; \
v1 = tv; \ v1 = tv; \
EdgeEquationInit(&e0, &v0->x, &v1->x); \ EdgeEquationInit(&e0, &v0->xyz[0], &v1->xyz[0]); \
EdgeEquationInit(&e1, &v1->x, &v2->x); \ EdgeEquationInit(&e1, &v1->xyz[0], &v2->xyz[0]); \
EdgeEquationInit(&e2, &v2->x, &v0->x); \ EdgeEquationInit(&e2, &v2->xyz[0], &v0->xyz[0]); \
area = 0.5f * (e0.c + e1.c + e2.c) \ area = 0.5f * (e0.c + e1.c + e2.c) \
// Check if triangle is backfacing. // Check if triangle is backfacing.
@ -135,18 +126,382 @@ void SceneBegin() {
SDL_RenderClear(RENDERER); SDL_RenderClear(RENDERER);
} }
void SceneListBegin(GPUList list) { static Vertex BUFFER[1024 * 32];
static uint32_t vertex_counter = 0;
GL_FORCE_INLINE bool glIsVertex(const float flags) {
return flags == GPU_CMD_VERTEX_EOL || flags == GPU_CMD_VERTEX;
}
GL_FORCE_INLINE bool glIsLastVertex(const float flags) {
return flags == GPU_CMD_VERTEX_EOL;
}
void SceneListBegin(GPUList list) {
vertex_counter = 0;
}
GL_FORCE_INLINE void _glPerspectiveDivideVertex(Vertex* vertex, const float h) {
const float f = 1.0f / (vertex->w);
/* Convert to NDC and apply viewport */
vertex->xyz[0] = __builtin_fmaf(
VIEWPORT.hwidth, vertex->xyz[0] * f, VIEWPORT.x_plus_hwidth
);
vertex->xyz[1] = h - __builtin_fmaf(
VIEWPORT.hheight, vertex->xyz[1] * f, VIEWPORT.y_plus_hheight
);
if(vertex->w == 1.0f) {
vertex->xyz[2] = 1.0f / (1.0001f + vertex->xyz[2]);
} else {
vertex->xyz[2] = f;
}
}
GL_FORCE_INLINE void _glSubmitHeaderOrVertex(const Vertex* v) {
#ifndef NDEBUG
if(glIsVertex(v->flags)) {
assert(!isnan(v->xyz[2]));
assert(!isnan(v->w));
}
#endif
#if CLIP_DEBUG
printf("Submitting: %x (%x)\n", v, v->flags);
#endif
BUFFER[vertex_counter++] = *v;
}
static struct {
Vertex* v;
int visible;
} triangle[3];
static int tri_count = 0;
static int strip_count = 0;
GL_FORCE_INLINE void interpolateColour(const uint8_t* v1, const uint8_t* v2, const float t, uint8_t* out) {
const int MASK1 = 0x00FF00FF;
const int MASK2 = 0xFF00FF00;
const int f2 = 256 * t;
const int f1 = 256 - f2;
const uint32_t a = *(uint32_t*) v1;
const uint32_t b = *(uint32_t*) v2;
*((uint32_t*) out) = (((((a & MASK1) * f1) + ((b & MASK1) * f2)) >> 8) & MASK1) |
(((((a & MASK2) * f1) + ((b & MASK2) * f2)) >> 8) & MASK2);
}
GL_FORCE_INLINE void _glClipEdge(const Vertex* v1, const Vertex* v2, Vertex* vout) {
/* Clipping time! */
const float d0 = v1->w + v1->xyz[2];
const float d1 = v2->w + v2->xyz[2];
const float epsilon = (d0 < d1) ? -0.00001f : 0.00001f;
float t = (d0 / (d0 - d1)) + epsilon;
t = (t > 1.0f) ? 1.0f : t;
t = (t < 0.0f) ? 0.0f : t;
vout->xyz[0] = __builtin_fmaf(v2->xyz[0] - v1->xyz[0], t, v1->xyz[0]);
vout->xyz[1] = __builtin_fmaf(v2->xyz[1] - v1->xyz[1], t, v1->xyz[1]);
vout->xyz[2] = __builtin_fmaf(v2->xyz[2] - v1->xyz[2], t, v1->xyz[2]);
vout->w = __builtin_fmaf(v2->w - v1->w, t, v1->w);
vout->uv[0] = __builtin_fmaf(v2->uv[0] - v1->uv[0], t, v1->uv[0]);
vout->uv[1] = __builtin_fmaf(v2->uv[1] - v1->uv[1], t, v1->uv[1]);
interpolateColour(v1->bgra, v2->bgra, t, vout->bgra);
}
GL_FORCE_INLINE void ClearTriangle() {
tri_count = 0;
}
GL_FORCE_INLINE void ShiftTriangle() {
if(!tri_count) {
return;
}
tri_count--;
triangle[0] = triangle[1];
triangle[1] = triangle[2];
#ifndef NDEBUG
triangle[2].v = NULL;
triangle[2].visible = false;
#endif
}
GL_FORCE_INLINE void ShiftRotateTriangle() {
if(!tri_count) {
return;
}
if(triangle[0].v < triangle[1].v) {
triangle[0] = triangle[2];
} else {
triangle[1] = triangle[2];
}
tri_count--;
} }
void SceneListSubmit(void* src, int n) { void SceneListSubmit(void* src, int n) {
uint32_t vertex_counter = 0; /* Perform perspective divide on each vertex */
const uint32_t* flags = (const uint32_t*) src; Vertex* vertex = (Vertex*) src;
uint32_t step = sizeof(GPUVertex) / sizeof(uint32_t);
for(int i = 0; i < n; ++i, flags += step) { const float h = GetVideoMode()->height;
tri_count = 0;
strip_count = 0;
#if CLIP_DEBUG
printf("----\n");
#endif
for(int i = 0; i < n; ++i, ++vertex) {
PREFETCH(vertex + 1);
bool is_last_in_strip = glIsLastVertex(vertex->flags);
/* Wait until we fill the triangle */
if(tri_count < 3) {
if(glIsVertex(vertex->flags)) {
triangle[tri_count].v = vertex;
triangle[tri_count].visible = vertex->xyz[2] >= -vertex->w;
tri_count++;
strip_count++;
} else {
/* We hit a header */
tri_count = 0;
strip_count = 0;
_glSubmitHeaderOrVertex(vertex);
}
if(tri_count < 3) {
continue;
}
}
#if CLIP_DEBUG
printf("SC: %d\n", strip_count);
#endif
/* If we got here, then triangle contains 3 vertices */
int visible_mask = triangle[0].visible | (triangle[1].visible << 1) | (triangle[2].visible << 2);
if(visible_mask == 7) {
#if CLIP_DEBUG
printf("Visible\n");
#endif
/* All the vertices are visible! We divide and submit v0, then shift */
_glPerspectiveDivideVertex(vertex - 2, h);
_glSubmitHeaderOrVertex(vertex - 2);
if(is_last_in_strip) {
_glPerspectiveDivideVertex(vertex - 1, h);
_glSubmitHeaderOrVertex(vertex - 1);
_glPerspectiveDivideVertex(vertex, h);
_glSubmitHeaderOrVertex(vertex);
tri_count = 0;
strip_count = 0;
}
ShiftRotateTriangle();
} else if(visible_mask) {
/* Clipping time!
There are 6 distinct possibilities when clipping a triangle. 3 of them result
in another triangle, 3 of them result in a quadrilateral.
Assuming you iterate the edges of the triangle in order, and create a new *visible*
vertex when you cross the plane, and discard vertices behind the plane, then the only
difference between the two cases is that the final two vertices that need submitting have
to be reversed.
Unfortunately we have to copy vertices here, because if we persp-divide a vertex it may
be used in a subsequent triangle in the strip and would end up being double divided.
*/
#if CLIP_DEBUG
printf("Clip: %d, SC: %d\n", visible_mask, strip_count);
printf("%d, %d, %d\n", triangle[0].v - (Vertex*) src - 1, triangle[1].v - (Vertex*) src - 1, triangle[2].v - (Vertex*) src - 1);
#endif
Vertex tmp;
if(strip_count > 3) {
#if CLIP_DEBUG
printf("Flush\n");
#endif
tmp = *(vertex - 2);
/* If we had triangles ahead of this one, submit and finalize */
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
tmp = *(vertex - 1);
tmp.flags = GPU_CMD_VERTEX_EOL;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
}
switch(visible_mask) {
case 1: {
/* 0, 0a, 2a */
tmp = *triangle[0].v;
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
_glClipEdge(triangle[0].v, triangle[1].v, &tmp);
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
_glClipEdge(triangle[2].v, triangle[0].v, &tmp);
tmp.flags = GPU_CMD_VERTEX_EOL;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
} break;
case 2: {
/* 0a, 1, 1a */
_glClipEdge(triangle[0].v, triangle[1].v, &tmp);
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
tmp = *triangle[1].v;
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
_glClipEdge(triangle[1].v, triangle[2].v, &tmp);
tmp.flags = GPU_CMD_VERTEX_EOL;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
} break;
case 3: {
/* 0, 1, 2a, 1a */
tmp = *triangle[0].v;
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
tmp = *triangle[1].v;
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
_glClipEdge(triangle[2].v, triangle[0].v, &tmp);
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
_glClipEdge(triangle[1].v, triangle[2].v, &tmp);
tmp.flags = GPU_CMD_VERTEX_EOL;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
} break;
case 4: {
/* 1a, 2, 2a */
_glClipEdge(triangle[1].v, triangle[2].v, &tmp);
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
tmp = *triangle[2].v;
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
_glClipEdge(triangle[2].v, triangle[0].v, &tmp);
tmp.flags = GPU_CMD_VERTEX_EOL;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
} break;
case 5: {
/* 0, 0a, 2, 1a */
tmp = *triangle[0].v;
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
_glClipEdge(triangle[0].v, triangle[1].v, &tmp);
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
tmp = *triangle[2].v;
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
_glClipEdge(triangle[1].v, triangle[2].v, &tmp);
tmp.flags = GPU_CMD_VERTEX_EOL;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
} break;
case 6: {
/* 0a, 1, 2a, 2 */
_glClipEdge(triangle[0].v, triangle[1].v, &tmp);
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
tmp = *triangle[1].v;
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
_glClipEdge(triangle[2].v, triangle[0].v, &tmp);
tmp.flags = GPU_CMD_VERTEX;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
tmp = *triangle[2].v;
tmp.flags = GPU_CMD_VERTEX_EOL;
_glPerspectiveDivideVertex(&tmp, h);
_glSubmitHeaderOrVertex(&tmp);
} break;
default:
break;
}
/* If this was the last in the strip, we don't need to
submit anything else, we just wipe the tri_count */
if(is_last_in_strip) {
tri_count = 0;
strip_count = 0;
} else {
ShiftRotateTriangle();
strip_count = 2;
}
} else {
/* Invisible? Move to the next in the strip */
if(is_last_in_strip) {
tri_count = 0;
strip_count = 0;
}
strip_count = 2;
ShiftRotateTriangle();
}
}
}
void SceneListFinish() {
uint32_t vidx = 0;
const uint32_t* flags = (const uint32_t*) BUFFER;
uint32_t step = sizeof(Vertex) / sizeof(uint32_t);
for(int i = 0; i < vertex_counter; ++i, flags += step) {
if((*flags & GPU_CMD_POLYHDR) == GPU_CMD_POLYHDR) { if((*flags & GPU_CMD_POLYHDR) == GPU_CMD_POLYHDR) {
vertex_counter = 0; vidx = 0;
uint32_t mode1 = *(flags + 1); uint32_t mode1 = *(flags + 1);
// Extract culling mode // Extract culling mode
@ -157,33 +512,29 @@ void SceneListSubmit(void* src, int n) {
switch(*flags) { switch(*flags) {
case GPU_CMD_VERTEX_EOL: case GPU_CMD_VERTEX_EOL:
case GPU_CMD_VERTEX: // Fallthrough case GPU_CMD_VERTEX: // Fallthrough
vertex_counter++; vidx++;
break; break;
default: default:
break; break;
} }
} }
if(vertex_counter > 2) { if(vidx > 2) {
GPUVertex* v0 = (GPUVertex*) (flags - step - step); Vertex* v0 = (Vertex*) (flags - step - step);
GPUVertex* v1 = (GPUVertex*) (flags - step); Vertex* v1 = (Vertex*) (flags - step);
GPUVertex* v2 = (GPUVertex*) (flags); Vertex* v2 = (Vertex*) (flags);
(vertex_counter % 2 == 0) ? DrawTriangle(v0, v1, v2) : DrawTriangle(v1, v0, v2); (vidx % 2 == 0) ? DrawTriangle(v0, v1, v2) : DrawTriangle(v1, v0, v2);
} }
if((*flags) == GPU_CMD_VERTEX_EOL) { if((*flags) == GPU_CMD_VERTEX_EOL) {
vertex_counter = 0; vidx = 0;
} }
} }
} }
void SceneListFinish() {
}
void SceneFinish() { void SceneFinish() {
SDL_RenderPresent(RENDERER); SDL_RenderPresent(RENDERER);
return;
/* Only sensible place to hook the quit signal */ /* Only sensible place to hook the quit signal */
SDL_Event e; SDL_Event e;
while (SDL_PollEvent(&e)) { while (SDL_PollEvent(&e)) {