Modified triangle rendering code so that it finally works on real hardware.

Not sure why the original method doesn't work, probably some weird limitation on the memory/GTE register subsystems.
This will get rewritten anyway so it doesn't matter much now.

display.c

@@ -155,6 +155,7 @@ void display_start()
 
 	gte_SetBackColor(48, 48, 48);		// Ambient light color
 	gte_SetColorMatrix(&light_cols);	// Light color (up to three different lights)
+	gte_SetLightMatrix(&identity);
 
 	MATRIX proj_matrix = quake_swizzle;			// Swizzle coordinates so that everything is in Quake coordinate system (Z up)
 	ScaleMatrixL(&proj_matrix, &aspect_scale);	// Apply aspect ratio correction for the current resolution

main.c

@@ -46,11 +46,12 @@ int main(int argc, const char *argv[])
 
 		display_start();	// Works
 
+		FntPrint(-1, "Time: %d, ticks = %d, delta time = %d, fps = %d\n", (time_getRealTime() * 1000) >> 12, time_getFrameNumber(), time_getDeltaTime(), time_getFrameRate() >> 8);
+		FntPrint(-1, "Camera pos = (%d, %d, %d) rot = (%d, %d, %d)\n", cam_pos.vx, cam_pos.vy, cam_pos.vz, cam_rot.vx, cam_rot.vy, cam_rot.vz);
+
 		// Draw stuff
 		world_draw(&world);	// Doesn't work >:(
 	
-		FntPrint(-1, "Time: %d, ticks = %d, delta time = %d, fps = %d\n", (time_getRealTime() * 1000) >> 12, time_getFrameNumber(), time_getDeltaTime(), time_getFrameRate() >> 8);
-		FntPrint(-1, "Camera pos = (%d, %d, %d) rot = (%d, %d, %d)\n", cam_pos.vx, cam_pos.vy, cam_pos.vz, cam_rot.vx, cam_rot.vy, cam_rot.vz);
 		FntFlush(-1);
 		
 		display_finish();	// Works

world.c

@@ -40,6 +40,7 @@ void world_load(const u_long *data, world_t *world)
 
 void world_draw(const world_t *world)
 {
+    SVECTOR vecs[3];
     int p;
 
     // The world doesn't move, so we just set the camera view-projection matrix
@@ -51,16 +52,27 @@ void world_draw(const world_t *world)
         const ps1bsp_face_t *face = &world->faces[faceIdx];
         const CVECTOR *col = &colors[faceIdx % numColors];
 
-        for (int triangleIdx = 0; triangleIdx < face->numTriangles; ++triangleIdx)
+        const ps1bsp_triangle_t *tri = &world->triangles[face->firstTriangleId];
+        for (int triangleIdx = 0; triangleIdx < face->numTriangles; ++triangleIdx, ++tri)
         {
-            const ps1bsp_triangle_t *tri = &world->triangles[face->firstTriangleId + triangleIdx];
-
             // Naively draw the triangle with GTE, nothing special or optimized about this
-            SVECTOR *v0 = (SVECTOR*)&world->vertices[tri->vertex0];
-            SVECTOR *v1 = (SVECTOR*)&world->vertices[tri->vertex1];
-            SVECTOR *v2 = (SVECTOR*)&world->vertices[tri->vertex2];
+            const ps1bsp_vertex_t *v0 = &world->vertices[tri->vertex0];
+            const ps1bsp_vertex_t *v1 = &world->vertices[tri->vertex1];
+            const ps1bsp_vertex_t *v2 = &world->vertices[tri->vertex2];
+
+            vecs[0].vx = v0->x; vecs[0].vy = v0->y; vecs[0].vz = v0->z;
+            vecs[1].vx = v1->x; vecs[1].vy = v1->y; vecs[1].vz = v1->z;
+            vecs[2].vx = v2->x; vecs[2].vy = v2->y; vecs[2].vz = v2->z;
+            gte_ldv3c(vecs);
+
+            // This method *should* work but it doesn't on real hardware, for some reason.
+            // Perhaps the PS1 doesn't like loading data from random far-off memory locations directly into GTE registers?
+            // We probably want to use the scratchpad memory to prepare the vertex data for the GTE anyway.
+            // SVECTOR *v0 = (SVECTOR*)&world->vertices[tri->vertex0];
+            // SVECTOR *v1 = (SVECTOR*)&world->vertices[tri->vertex1];
+            // SVECTOR *v2 = (SVECTOR*)&world->vertices[tri->vertex2];
+            // gte_ldv3(v0, v1, v2);
 
-            gte_ldv3(v0, v1, v2);
             gte_rtpt();     // Rotation, translation, perspective projection
 
             // Normal clipping for backface culling