#ifndef SMALL3DLIB_H #define SMALL3DLIB_H /* Simple realtime 3D software rasterization renderer. It is fast, focused on resource-limited computers, located in a single C header file, with no dependencies, using only 32bit integer arithmetics. author: Miloslav Ciz license: CC0 1.0 (public domain) found at https://creativecommons.org/publicdomain/zero/1.0/ + additional waiver of all IP version: 0.905d Before including the library, define S3L_PIXEL_FUNCTION to the name of the function you'll be using to draw single pixels (this function will be called by the library to render the frames). Also either init S3L_resolutionX and S3L_resolutionY or define S3L_RESOLUTION_X and S3L_RESOLUTION_Y. You'll also need to decide what rendering strategy and other settings you want to use, depending on your specific usecase. You may want to use a z-buffer (full or reduced, S3L_Z_BUFFER), sorted-drawing (S3L_SORT), or even none of these. See the description of the options in this file. The rendering itself is done with S3L_drawScene, usually preceded by S3L_newFrame (for clearing zBuffer etc.). The library is meant to be used in not so huge programs that use single translation unit and so includes both declarations and implementation at once. If you for some reason use multiple translation units (which include the library), you'll have to handle this yourself (e.g. create a wrapper, manually split the library into .c and .h etc.). -------------------- This work's goal is to never be encumbered by any exclusive intellectual property rights. The work is therefore provided under CC0 1.0 + additional WAIVER OF ALL INTELLECTUAL PROPERTY RIGHTS that waives the rest of intellectual property rights not already waived by CC0 1.0. The WAIVER OF ALL INTELLECTUAL PROPERTY RGHTS is as follows: Each contributor to this work agrees that they waive any exclusive rights, including but not limited to copyright, patents, trademark, trade dress, industrial design, plant varieties and trade secrets, to any and all ideas, concepts, processes, discoveries, improvements and inventions conceived, discovered, made, designed, researched or developed by the contributor either solely or jointly with others, which relate to this work or result from this work. Should any waiver of such right be judged legally invalid or ineffective under applicable law, the contributor hereby grants to each affected person a royalty-free, non transferable, non sublicensable, non exclusive, irrevocable and unconditional license to this right. -------------------- CONVENTIONS: This library should never draw pixels outside the specified screen boundaries, so you don't have to check this (that would cost CPU time)! You can safely assume that triangles are rasterized one by one and from top down, left to right (so you can utilize e.g. various caches), and if sorting is disabled the order of rasterization will be that specified in the scene structure and model arrays (of course, some triangles and models may be skipped due to culling etc.). Angles are in S3L_Units, a full angle (2 pi) is S3L_FRACTIONS_PER_UNITs. We use row vectors. In 3D space, a left-handed coord. system is used. One spatial unit is split into S3L_FRACTIONS_PER_UNITs fractions (fixed point arithmetic). y ^ | _ | /| z | / | / [0,0,0]-------> x Untransformed camera is placed at [0,0,0], looking forward along +z axis. The projection plane is centered at [0,0,0], stretrinch from -S3L_FRACTIONS_PER_UNIT to S3L_FRACTIONS_PER_UNIT horizontally (x), vertical size (y) depends on the aspect ratio (S3L_RESOLUTION_X and S3L_RESOLUTION_Y). Camera FOV is defined by focal length in S3L_Units. Rotations use Euler angles and are generally in the extrinsic Euler angles in ZXY order (by Z, then by X, then by Y). Positive rotation about an axis rotates CW (clock-wise) when looking in the direction of the axis. Coordinates of pixels on the screen start at the top left, from [0,0]. There is NO subpixel accuracy (screen coordinates are only integer). Triangle rasterization rules are these (mostly same as OpenGL, D3D etc.): - Let's define: - left side: - not exactly horizontal, and on the left side of triangle - exactly horizontal and above the topmost (in other words: its normal points at least a little to the left or completely up) - right side: not left side - Pixel centers are at integer coordinates and triangle for drawing are specified with integer coordinates of pixel centers. - A pixel is rasterized: - if its center is inside the triangle OR - if its center is exactly on the triangle side which is left and at the same time is not on the side that's right (case of a triangle that's on a single line) OR - if its center is exactly on the triangle corner of sides neither of which is right. These rules imply among others: - Adjacent triangles don't have any overlapping pixels, nor gaps between. - Triangles of points that lie on a single line are NOT rasterized. - A single "long" triangle CAN be rasterized as isolated islands of pixels. - Transforming (e.g. mirroring, rotating by 90 degrees etc.) a result of rasterizing triangle A is NOT generally equal to applying the same transformation to triangle A first and then rasterizing it. Even the number of rasterized pixels is usually different. - If specifying a triangle with integer coordinates (which we are), then: - The bottom-most corner (or side) of a triangle is never rasterized (because it is connected to a right side). - The top-most corner can only be rasterized on completely horizontal side (otherwise it is connected to a right side). - Vertically middle corner is rasterized if and only if it is on the left of the triangle and at the same time is also not the bottom-most corner. */ #include #ifdef S3L_RESOLUTION_X #ifdef S3L_RESOLUTION_Y #define S3L_MAX_PIXELS (S3L_RESOLUTION_X * S3L_RESOLUTION_Y) #endif #endif #ifndef S3L_RESOLUTION_X #ifndef S3L_MAX_PIXELS #error Dynamic resolution set (S3L_RESOLUTION_X not defined), but\ S3L_MAX_PIXELS not defined! #endif uint16_t S3L_resolutionX = 512; /**< If a static resolution is not set with S3L_RESOLUTION_X, this variable can be used to change X resolution at runtime, in which case S3L_MAX_PIXELS has to be defined (to allocate zBuffer etc.)! */ #define S3L_RESOLUTION_X S3L_resolutionX #endif #ifndef S3L_RESOLUTION_Y #ifndef S3L_MAX_PIXELS #error Dynamic resolution set (S3L_RESOLUTION_Y not defined), but\ S3L_MAX_PIXELS not defined! #endif uint16_t S3L_resolutionY = 512; /**< Same as S3L_resolutionX, but for Y resolution. */ #define S3L_RESOLUTION_Y S3L_resolutionY #endif #ifndef S3L_USE_WIDER_TYPES /** If true, the library will use wider data types which will largely supress many rendering bugs and imprecisions happening due to overflows, but this will also consumer more RAM and may potentially be slower on computers with smaller native integer. */ #define S3L_USE_WIDER_TYPES 0 #endif #ifndef S3L_SIN_METHOD /** Says which method should be used for computing sin/cos functions, possible values: 0 (lookup table, takes more program memory), 1 (Bhaskara's approximation, slower). This may cause the trigonometric functions give slightly different results. */ #define S3L_SIN_METHOD 0 #endif /** Units of measurement in 3D space. There is S3L_FRACTIONS_PER_UNIT in one spatial unit. By dividing the unit into fractions we effectively achieve a fixed point arithmetic. The number of fractions is a constant that serves as 1.0 in floating point arithmetic (normalization etc.). */ typedef #if S3L_USE_WIDER_TYPES int64_t #else int32_t #endif S3L_Unit; /** How many fractions a spatial unit is split into, i.e. this is the fixed point scaling. This is NOT SUPPOSED TO BE REDEFINED, so rather don't do it (otherwise things may overflow etc.). */ #define S3L_FRACTIONS_PER_UNIT 512 #define S3L_F S3L_FRACTIONS_PER_UNIT typedef #if S3L_USE_WIDER_TYPES int32_t #else int16_t #endif S3L_ScreenCoord; typedef #if S3L_USE_WIDER_TYPES uint32_t #else uint16_t #endif S3L_Index; #ifndef S3L_NEAR_CROSS_STRATEGY /** Specifies how the library will handle triangles that partially cross the near plane. These are problematic and require special handling. Possible values: 0: Strictly cull any triangle crossing the near plane. This will make such triangles disappear. This is good for performance or models viewed only from at least small distance. 1: Forcefully push the vertices crossing near plane in front of it. This is a cheap technique that can be good enough for displaying simple environments on slow devices, but texturing and geometric artifacts/warps will appear. 2: Geometrically correct the triangles crossing the near plane. This may result in some triangles being subdivided into two and is a little more expensive, but the results will be geometrically correct, even though barycentric correction is not performed so texturing artifacts will appear. Can be ideal with S3L_FLAT. 3: Perform both geometrical and barycentric correction of triangle crossing the near plane. This is significantly more expensive but results in correct rendering. */ #define S3L_NEAR_CROSS_STRATEGY 0 #endif #ifndef S3L_FLAT /** If on, disables computation of per-pixel values such as barycentric coordinates and depth -- these will still be available but will be the same for the whole triangle. This can be used to create flat-shaded renders and will be a lot faster. With this option on you will probably want to use sorting instead of z-buffer. */ #define S3L_FLAT 0 #endif #if S3L_FLAT #define S3L_COMPUTE_DEPTH 0 #define S3L_PERSPECTIVE_CORRECTION 0 // don't disable z-buffer, it makes sense to use it with no sorting #endif #ifndef S3L_PERSPECTIVE_CORRECTION /** Specifies what type of perspective correction (PC) to use. Remember this is an expensive operation! Possible values: 0: No perspective correction. Fastest, inaccurate from most angles. 1: Per-pixel perspective correction, accurate but very expensive. 2: Approximation (computing only at every S3L_PC_APPROX_LENGTHth pixel). Quake-style approximation is used, which only computes the PC after S3L_PC_APPROX_LENGTH pixels. This is reasonably accurate and fast. */ #define S3L_PERSPECTIVE_CORRECTION 0 #endif #ifndef S3L_PC_APPROX_LENGTH /** For S3L_PERSPECTIVE_CORRECTION == 2, this specifies after how many pixels PC is recomputed. Should be a power of two to keep up the performance. Smaller is nicer but slower. */ #define S3L_PC_APPROX_LENGTH 32 #endif #if S3L_PERSPECTIVE_CORRECTION #define S3L_COMPUTE_DEPTH 1 // PC inevitably computes depth, so enable it #endif #ifndef S3L_COMPUTE_DEPTH /** Whether to compute depth for each pixel (fragment). Some other options may turn this on automatically. If you don't need depth information, turning this off can save performance. Depth will still be accessible in S3L_PixelInfo, but will be constant -- equal to center point depth -- over the whole triangle. */ #define S3L_COMPUTE_DEPTH 1 #endif #ifndef S3L_Z_BUFFER /** What type of z-buffer (depth buffer) to use for visibility determination. Possible values: 0: Don't use z-buffer. This saves a lot of memory, but visibility checking won't be pixel-accurate and has to mostly be done by other means (typically sorting). 1: Use full z-buffer (of S3L_Units) for visibiltiy determination. This is the most accurate option (and also a fast one), but requires a big amount of memory. 2: Use reduced-size z-buffer (of bytes). This is fast and somewhat accurate, but inaccuracies can occur and a considerable amount of memory is needed. */ #define S3L_Z_BUFFER 0 #endif #ifndef S3L_REDUCED_Z_BUFFER_GRANULARITY /** For S3L_Z_BUFFER == 2 this sets the reduced z-buffer granularity. */ #define S3L_REDUCED_Z_BUFFER_GRANULARITY 5 #endif #ifndef S3L_STENCIL_BUFFER /** Whether to use stencil buffer for drawing -- with this a pixel that would be resterized over an already rasterized pixel (within a frame) will be discarded. This is mostly for front-to-back sorted drawing. */ #define S3L_STENCIL_BUFFER 0 #endif #ifndef S3L_SORT /** Defines how to sort triangles before drawing a frame. This can be used to solve visibility in case z-buffer is not used, to prevent overwriting already rasterized pixels, implement transparency etc. Note that for simplicity and performance a relatively simple sorting is used which doesn't work completely correctly, so mistakes can occur (even the best sorting wouldn't be able to solve e.g. intersecting triangles). Note that sorting requires a bit of extra memory -- an array of the triangles to sort -- the size of this array limits the maximum number of triangles that can be drawn in a single frame (S3L_MAX_TRIANGES_DRAWN). Possible values: 0: Don't sort triangles. This is fastest and doesn't use extra memory. 1: Sort triangles from back to front. This can in most cases solve visibility without requiring almost any extra memory compared to z-buffer. 2: Sort triangles from front to back. This can be faster than back to front because we prevent computing pixels that will be overwritten by nearer ones, but we need a 1b stencil buffer for this (enable S3L_STENCIL_BUFFER), so a bit more memory is needed. */ #define S3L_SORT 0 #endif #ifndef S3L_MAX_TRIANGES_DRAWN /** Maximum number of triangles that can be drawn in sorted modes. This affects the size of the cache used for triangle sorting. */ #define S3L_MAX_TRIANGES_DRAWN 128 #endif #ifndef S3L_NEAR /** Distance of the near clipping plane. Points in front or EXATLY ON this plane are considered outside the frustum. This must be >= 0. */ #define S3L_NEAR (S3L_F / 4) #endif #if S3L_NEAR <= 0 #define S3L_NEAR 1 // Can't be <= 0. #endif #ifndef S3L_NORMAL_COMPUTE_MAXIMUM_AVERAGE /** Affects the S3L_computeModelNormals function. See its description for details. */ #define S3L_NORMAL_COMPUTE_MAXIMUM_AVERAGE 6 #endif #ifndef S3L_FAST_LERP_QUALITY /** Quality (scaling) of SOME (stepped) linear interpolations. 0 will most likely be a tiny bit faster, but artifacts can occur for bigger tris, while higher values can fix this -- in theory all higher values will have the same speed (it is a shift value), but it mustn't be too high to prevent overflow. */ #define S3L_FAST_LERP_QUALITY 11 #endif /** Vector that consists of four scalars and can represent homogenous coordinates, but is generally also used as Vec3 and Vec2 for various purposes. */ typedef struct { S3L_Unit x; S3L_Unit y; S3L_Unit z; S3L_Unit w; } S3L_Vec4; #define S3L_logVec4(v)\ printf("Vec4: %d %d %d %d\n",((v).x),((v).y),((v).z),((v).w)) static inline void S3L_vec4Init(S3L_Vec4 *v); static inline void S3L_vec4Set(S3L_Vec4 *v, S3L_Unit x, S3L_Unit y, S3L_Unit z, S3L_Unit w); static inline void S3L_vec3Add(S3L_Vec4 *result, S3L_Vec4 added); static inline void S3L_vec3Sub(S3L_Vec4 *result, S3L_Vec4 substracted); S3L_Unit S3L_vec3Length(S3L_Vec4 v); /** Normalizes Vec3. Note that this function tries to normalize correctly rather than quickly! If you need to normalize quickly, do it yourself in a way that best fits your case. */ void S3L_vec3Normalize(S3L_Vec4 *v); /** Like S3L_vec3Normalize, but doesn't perform any checks on the input vector, which is faster, but can be very innacurate or overflowing. You are supposed to provide a "nice" vector (not too big or small). */ static inline void S3L_vec3NormalizeFast(S3L_Vec4 *v); S3L_Unit S3L_vec2Length(S3L_Vec4 v); void S3L_vec3Cross(S3L_Vec4 a, S3L_Vec4 b, S3L_Vec4 *result); static inline S3L_Unit S3L_vec3Dot(S3L_Vec4 a, S3L_Vec4 b); /** Computes a reflection direction (typically used e.g. for specular component in Phong illumination). The input vectors must be normalized. The result will be normalized as well. */ void S3L_reflect(S3L_Vec4 toLight, S3L_Vec4 normal, S3L_Vec4 *result); /** Determines the winding of a triangle, returns 1 (CW, clockwise), -1 (CCW, counterclockwise) or 0 (points lie on a single line). */ static inline int8_t S3L_triangleWinding( S3L_ScreenCoord x0, S3L_ScreenCoord y0, S3L_ScreenCoord x1, S3L_ScreenCoord y1, S3L_ScreenCoord x2, S3L_ScreenCoord y2); typedef struct { S3L_Vec4 translation; S3L_Vec4 rotation; /**< Euler angles. Rortation is applied in this order: 1. z = by z (roll) CW looking along z+ 2. x = by x (pitch) CW looking along x+ 3. y = by y (yaw) CW looking along y+ */ S3L_Vec4 scale; } S3L_Transform3D; #define S3L_logTransform3D(t)\ printf("Transform3D: T = [%d %d %d], R = [%d %d %d], S = [%d %d %d]\n",\ (t).translation.x,(t).translation.y,(t).translation.z,\ (t).rotation.x,(t).rotation.y,(t).rotation.z,\ (t).scale.x,(t).scale.y,(t).scale.z) static inline void S3L_transform3DInit(S3L_Transform3D *t); void S3L_lookAt(S3L_Vec4 pointTo, S3L_Transform3D *t); void S3L_transform3DSet( S3L_Unit tx, S3L_Unit ty, S3L_Unit tz, S3L_Unit rx, S3L_Unit ry, S3L_Unit rz, S3L_Unit sx, S3L_Unit sy, S3L_Unit sz, S3L_Transform3D *t); /** Converts rotation transformation to three direction vectors of given length (any one can be NULL, in which case it won't be computed). */ void S3L_rotationToDirections( S3L_Vec4 rotation, S3L_Unit length, S3L_Vec4 *forw, S3L_Vec4 *right, S3L_Vec4 *up); /** 4x4 matrix, used mostly for 3D transforms. The indexing is this: matrix[column][row]. */ typedef S3L_Unit S3L_Mat4[4][4]; #define S3L_logMat4(m)\ printf("Mat4:\n %d %d %d %d\n %d %d %d %d\n %d %d %d %d\n %d %d %d %d\n"\ ,(m)[0][0],(m)[1][0],(m)[2][0],(m)[3][0],\ (m)[0][1],(m)[1][1],(m)[2][1],(m)[3][1],\ (m)[0][2],(m)[1][2],(m)[2][2],(m)[3][2],\ (m)[0][3],(m)[1][3],(m)[2][3],(m)[3][3]) /** Initializes a 4x4 matrix to identity. */ static inline void S3L_mat4Init(S3L_Mat4 m); void S3L_mat4Copy(S3L_Mat4 src, S3L_Mat4 dst); void S3L_mat4Transpose(S3L_Mat4 m); void S3L_makeTranslationMat( S3L_Unit offsetX, S3L_Unit offsetY, S3L_Unit offsetZ, S3L_Mat4 m); /** Makes a scaling matrix. DON'T FORGET: scale of 1.0 is set with S3L_FRACTIONS_PER_UNIT! */ void S3L_makeScaleMatrix( S3L_Unit scaleX, S3L_Unit scaleY, S3L_Unit scaleZ, S3L_Mat4 m); /** Makes a matrix for rotation in the ZXY order. */ void S3L_makeRotationMatrixZXY( S3L_Unit byX, S3L_Unit byY, S3L_Unit byZ, S3L_Mat4 m); void S3L_makeWorldMatrix(S3L_Transform3D worldTransform, S3L_Mat4 m); void S3L_makeCameraMatrix(S3L_Transform3D cameraTransform, S3L_Mat4 m); /** Multiplies a vector by a matrix with normalization by S3L_FRACTIONS_PER_UNIT. Result is stored in the input vector. */ void S3L_vec4Xmat4(S3L_Vec4 *v, S3L_Mat4 m); /** Same as S3L_vec4Xmat4 but faster, because this version doesn't compute the W component of the result, which is usually not needed. */ void S3L_vec3Xmat4(S3L_Vec4 *v, S3L_Mat4 m); /** Multiplies two matrices with normalization by S3L_FRACTIONS_PER_UNIT. Result is stored in the first matrix. The result represents a transformation that has the same effect as applying the transformation represented by m1 and then m2 (in that order). */ void S3L_mat4Xmat4(S3L_Mat4 m1, S3L_Mat4 m2); typedef struct { S3L_Unit focalLength; /**< Defines the field of view (FOV). 0 sets an orthographics projection (scale is controlled with camera's scale in its transform). */ S3L_Transform3D transform; } S3L_Camera; void S3L_cameraInit(S3L_Camera *camera); typedef struct { uint8_t backfaceCulling; /**< What backface culling to use. Possible values: - 0 none - 1 clock-wise - 2 counter clock-wise */ int8_t visible; /**< Can be used to easily hide the model. */ } S3L_DrawConfig; void S3L_drawConfigInit(S3L_DrawConfig *config); typedef struct { const S3L_Unit *vertices; S3L_Index vertexCount; const S3L_Index *triangles; S3L_Index triangleCount; S3L_Transform3D transform; S3L_Mat4 *customTransformMatrix; /**< This can be used to override the transform (if != 0) with a custom transform matrix, which is more general. */ S3L_DrawConfig config; } S3L_Model3D; ///< Represents a 3D model. void S3L_model3DInit( const S3L_Unit *vertices, S3L_Index vertexCount, const S3L_Index *triangles, S3L_Index triangleCount, S3L_Model3D *model); typedef struct { S3L_Model3D *models; S3L_Index modelCount; S3L_Camera camera; } S3L_Scene; ///< Represent the 3D scene to be rendered. void S3L_sceneInit( S3L_Model3D *models, S3L_Index modelCount, S3L_Scene *scene); typedef struct { S3L_ScreenCoord x; ///< Screen X coordinate. S3L_ScreenCoord y; ///< Screen Y coordinate. S3L_Unit barycentric[3]; /**< Barycentric coords correspond to the three vertices. These serve to locate the pixel on a triangle and interpolate values between its three points. Each one goes from 0 to S3L_FRACTIONS_PER_UNIT (including), but due to rounding error may fall outside this range (you can use S3L_correctBarycentricCoords to fix this for the price of some performance). The sum of the three coordinates will always be exactly S3L_FRACTIONS_PER_UNIT. */ S3L_Index modelIndex; ///< Model index within the scene. S3L_Index triangleIndex; ///< Triangle index within the model. uint32_t triangleID; /**< Unique ID of the triangle withing the whole scene. This can be used e.g. by a cache to quickly find out if a triangle has changed. */ S3L_Unit depth; ///< Depth (only if depth is turned on). S3L_Unit previousZ; /**< Z-buffer value (not necessarily world depth in S3L_Units!) that was in the z-buffer on the pixels position before this pixel was rasterized. This can be used to set the value back, e.g. for transparency. */ S3L_ScreenCoord triangleSize[2]; /**< Rasterized triangle width and height, can be used e.g. for MIP mapping. */ } S3L_PixelInfo; /**< Used to pass the info about a rasterized pixel (fragment) to the user-defined drawing func. */ static inline void S3L_pixelInfoInit(S3L_PixelInfo *p); /** Corrects barycentric coordinates so that they exactly meet the defined conditions (each fall into <0,S3L_FRACTIONS_PER_UNIT>, sum = S3L_FRACTIONS_PER_UNIT). Note that doing this per-pixel can slow the program down significantly. */ static inline void S3L_correctBarycentricCoords(S3L_Unit barycentric[3]); // general helper functions static inline S3L_Unit S3L_abs(S3L_Unit value); static inline S3L_Unit S3L_min(S3L_Unit v1, S3L_Unit v2); static inline S3L_Unit S3L_max(S3L_Unit v1, S3L_Unit v2); static inline S3L_Unit S3L_clamp(S3L_Unit v, S3L_Unit v1, S3L_Unit v2); static inline S3L_Unit S3L_wrap(S3L_Unit value, S3L_Unit mod); static inline S3L_Unit S3L_nonZero(S3L_Unit value); static inline S3L_Unit S3L_zeroClamp(S3L_Unit value); S3L_Unit S3L_sin(S3L_Unit x); S3L_Unit S3L_asin(S3L_Unit x); static inline S3L_Unit S3L_cos(S3L_Unit x); S3L_Unit S3L_vec3Length(S3L_Vec4 v); S3L_Unit S3L_sqrt(S3L_Unit value); /** Projects a single point from 3D space to the screen space (pixels), which can be useful e.g. for drawing sprites. The w component of input and result holds the point size. If this size is 0 in the result, the sprite is outside the view. */ void S3L_project3DPointToScreen( S3L_Vec4 point, S3L_Camera camera, S3L_Vec4 *result); /** Computes a normalized normal of given triangle. */ void S3L_triangleNormal(S3L_Vec4 t0, S3L_Vec4 t1, S3L_Vec4 t2, S3L_Vec4 *n); /** Helper function for retrieving per-vertex indexed values from an array, e.g. texturing (UV) coordinates. The 'indices' array contains three indices for each triangle, each index pointing into 'values' array, which contains the values, each one consisting of 'numComponents' components (e.g. 2 for UV coordinates). The three values are retrieved into 'v0', 'v1' and 'v2' vectors (into x, y, z and w, depending on 'numComponents'). This function is meant to be used per-triangle (typically from a cache), NOT per-pixel, as it is not as fast as possible! */ void S3L_getIndexedTriangleValues( S3L_Index triangleIndex, const S3L_Index *indices, const S3L_Unit *values, uint8_t numComponents, S3L_Vec4 *v0, S3L_Vec4 *v1, S3L_Vec4 *v2); /** Computes a normalized normal for every vertex of given model (this is relatively slow and SHOUDN'T be done each frame). The dst array must have a sufficient size preallocated! The size is: number of model vertices * 3 * sizeof(S3L_Unit). Note that for advanced allowing sharp edges it is not sufficient to have per-vertex normals, but must be per-triangle. This function doesn't support this. The function computes a normal for each vertex by averaging normals of the triangles containing the vertex. The maximum number of these triangle normals that will be averaged is set with S3L_NORMAL_COMPUTE_MAXIMUM_AVERAGE. */ void S3L_computeModelNormals(S3L_Model3D model, S3L_Unit *dst, int8_t transformNormals); /** Interpolated between two values, v1 and v2, in the same ratio as t is to tMax. Does NOT prevent zero division. */ static inline S3L_Unit S3L_interpolate( S3L_Unit v1, S3L_Unit v2, S3L_Unit t, S3L_Unit tMax); /** Same as S3L_interpolate but with v1 == 0. Should be faster. */ static inline S3L_Unit S3L_interpolateFrom0( S3L_Unit v2, S3L_Unit t, S3L_Unit tMax); /** Like S3L_interpolate, but uses a parameter that goes from 0 to S3L_FRACTIONS_PER_UNIT - 1, which can be faster. */ static inline S3L_Unit S3L_interpolateByUnit( S3L_Unit v1, S3L_Unit v2, S3L_Unit t); /** Same as S3L_interpolateByUnit but with v1 == 0. Should be faster. */ static inline S3L_Unit S3L_interpolateByUnitFrom0( S3L_Unit v2, S3L_Unit t); static inline S3L_Unit S3L_distanceManhattan(S3L_Vec4 a, S3L_Vec4 b); /** Returns a value interpolated between the three triangle vertices based on barycentric coordinates. */ static inline S3L_Unit S3L_interpolateBarycentric( S3L_Unit value0, S3L_Unit value1, S3L_Unit value2, S3L_Unit barycentric[3]); static inline void S3L_mapProjectionPlaneToScreen( S3L_Vec4 point, S3L_ScreenCoord *screenX, S3L_ScreenCoord *screenY); /** Draws a triangle according to given config. The vertices are specified in Screen Space space (pixels). If perspective correction is enabled, each vertex has to have a depth (Z position in camera space) specified in the Z component. */ void S3L_drawTriangle( S3L_Vec4 point0, S3L_Vec4 point1, S3L_Vec4 point2, S3L_Index modelIndex, S3L_Index triangleIndex); /** This should be called before rendering each frame. The function clears buffers and does potentially other things needed for the frame. */ void S3L_newFrame(void); void S3L_zBufferClear(void); void S3L_stencilBufferClear(void); /** Writes a value (not necessarily depth! depends on the format of z-buffer) to z-buffer (if enabled). Does NOT check boundaries! */ void S3L_zBufferWrite(S3L_ScreenCoord x, S3L_ScreenCoord y, S3L_Unit value); /** Reads a value (not necessarily depth! depends on the format of z-buffer) from z-buffer (if enabled). Does NOT check boundaries! */ S3L_Unit S3L_zBufferRead(S3L_ScreenCoord x, S3L_ScreenCoord y); static inline void S3L_rotate2DPoint(S3L_Unit *x, S3L_Unit *y, S3L_Unit angle); /** Predefined vertices of a cube to simply insert in an array. These come with S3L_CUBE_TRIANGLES and S3L_CUBE_TEXCOORDS. */ #define S3L_CUBE_VERTICES(m)\ /* 0 front, bottom, right */\ m/2, -m/2, -m/2,\ /* 1 front, bottom, left */\ -m/2, -m/2, -m/2,\ /* 2 front, top, right */\ m/2, m/2, -m/2,\ /* 3 front, top, left */\ -m/2, m/2, -m/2,\ /* 4 back, bottom, right */\ m/2, -m/2, m/2,\ /* 5 back, bottom, left */\ -m/2, -m/2, m/2,\ /* 6 back, top, right */\ m/2, m/2, m/2,\ /* 7 back, top, left */\ -m/2, m/2, m/2 #define S3L_CUBE_VERTEX_COUNT 8 /** Predefined triangle indices of a cube, to be used with S3L_CUBE_VERTICES and S3L_CUBE_TEXCOORDS. */ #define S3L_CUBE_TRIANGLES\ 3, 0, 2, /* front */\ 1, 0, 3,\ 0, 4, 2, /* right */\ 2, 4, 6,\ 4, 5, 6, /* back */\ 7, 6, 5,\ 3, 7, 1, /* left */\ 1, 7, 5,\ 6, 3, 2, /* top */\ 7, 3, 6,\ 1, 4, 0, /* bottom */\ 5, 4, 1 #define S3L_CUBE_TRIANGLE_COUNT 12 /** Predefined texture coordinates of a cube, corresponding to triangles (NOT vertices), to be used with S3L_CUBE_VERTICES and S3L_CUBE_TRIANGLES. */ #define S3L_CUBE_TEXCOORDS(m)\ 0,0, m,m, m,0,\ 0,m, m,m, 0,0,\ m,m, m,0, 0,m,\ 0,m, m,0, 0,0,\ m,0, 0,0, m,m,\ 0,m, m,m, 0,0,\ 0,0, 0,m, m,0,\ m,0, 0,m, m,m,\ 0,0, m,m, m,0,\ 0,m, m,m, 0,0,\ m,0, 0,m, m,m,\ 0,0, 0,m, m,0 //============================================================================= // privates #define S3L_UNUSED(what) (void)(what) ///< helper macro for unused vars #define S3L_HALF_RESOLUTION_X (S3L_RESOLUTION_X >> 1) #define S3L_HALF_RESOLUTION_Y (S3L_RESOLUTION_Y >> 1) #define S3L_PROJECTION_PLANE_HEIGHT\ ((S3L_RESOLUTION_Y * S3L_F * 2) / S3L_RESOLUTION_X) #if S3L_Z_BUFFER == 1 #define S3L_MAX_DEPTH 2147483647 S3L_Unit S3L_zBuffer[S3L_MAX_PIXELS]; #define S3L_zBufferFormat(depth) (depth) #elif S3L_Z_BUFFER == 2 #define S3L_MAX_DEPTH 255 uint8_t S3L_zBuffer[S3L_MAX_PIXELS]; #define S3L_zBufferFormat(depth)\ S3L_min(255,(depth) >> S3L_REDUCED_Z_BUFFER_GRANULARITY) #endif #if S3L_Z_BUFFER static inline int8_t S3L_zTest( S3L_ScreenCoord x, S3L_ScreenCoord y, S3L_Unit depth) { uint32_t index = y * S3L_RESOLUTION_X + x; depth = S3L_zBufferFormat(depth); #if S3L_Z_BUFFER == 2 #define cmp <= /* For reduced z-buffer we need equality test, because otherwise pixels at the maximum depth (255) would never be drawn over the background (which also has the depth of 255). */ #else #define cmp < /* For normal z-buffer we leave out equality test to not waste time by drawing over already drawn pixls. */ #endif if (depth cmp S3L_zBuffer[index]) { S3L_zBuffer[index] = depth; return 1; } #undef cmp return 0; } #endif S3L_Unit S3L_zBufferRead(S3L_ScreenCoord x, S3L_ScreenCoord y) { #if S3L_Z_BUFFER return S3L_zBuffer[y * S3L_RESOLUTION_X + x]; #else S3L_UNUSED(x); S3L_UNUSED(y); return 0; #endif } void S3L_zBufferWrite(S3L_ScreenCoord x, S3L_ScreenCoord y, S3L_Unit value) { #if S3L_Z_BUFFER S3L_zBuffer[y * S3L_RESOLUTION_X + x] = value; #else S3L_UNUSED(x); S3L_UNUSED(y); S3L_UNUSED(value); #endif } #if S3L_STENCIL_BUFFER #define S3L_STENCIL_BUFFER_SIZE\ ((S3L_RESOLUTION_X * S3L_RESOLUTION_Y - 1) / 8 + 1) uint8_t S3L_stencilBuffer[S3L_STENCIL_BUFFER_SIZE]; static inline int8_t S3L_stencilTest( S3L_ScreenCoord x, S3L_ScreenCoord y) { uint32_t index = y * S3L_RESOLUTION_X + x; uint32_t bit = (index & 0x00000007); index = index >> 3; uint8_t val = S3L_stencilBuffer[index]; if ((val >> bit) & 0x1) return 0; S3L_stencilBuffer[index] = val | (0x1 << bit); return 1; } #endif #define S3L_COMPUTE_LERP_DEPTH\ (S3L_COMPUTE_DEPTH && (S3L_PERSPECTIVE_CORRECTION == 0)) #define S3L_SIN_TABLE_LENGTH 128 #if S3L_SIN_METHOD == 0 static const S3L_Unit S3L_sinTable[S3L_SIN_TABLE_LENGTH] = { /* 511 was chosen here as a highest number that doesn't overflow during compilation for S3L_F == 1024 */ (0*S3L_F)/511, (6*S3L_F)/511, (12*S3L_F)/511, (18*S3L_F)/511, (25*S3L_F)/511, (31*S3L_F)/511, (37*S3L_F)/511, (43*S3L_F)/511, (50*S3L_F)/511, (56*S3L_F)/511, (62*S3L_F)/511, (68*S3L_F)/511, (74*S3L_F)/511, (81*S3L_F)/511, (87*S3L_F)/511, (93*S3L_F)/511, (99*S3L_F)/511, (105*S3L_F)/511, (111*S3L_F)/511, (118*S3L_F)/511, (124*S3L_F)/511, (130*S3L_F)/511, (136*S3L_F)/511, (142*S3L_F)/511, (148*S3L_F)/511, (154*S3L_F)/511, (160*S3L_F)/511, (166*S3L_F)/511, (172*S3L_F)/511, (178*S3L_F)/511, (183*S3L_F)/511, (189*S3L_F)/511, (195*S3L_F)/511, (201*S3L_F)/511, (207*S3L_F)/511, (212*S3L_F)/511, (218*S3L_F)/511, (224*S3L_F)/511, (229*S3L_F)/511, (235*S3L_F)/511, (240*S3L_F)/511, (246*S3L_F)/511, (251*S3L_F)/511, (257*S3L_F)/511, (262*S3L_F)/511, (268*S3L_F)/511, (273*S3L_F)/511, (278*S3L_F)/511, (283*S3L_F)/511, (289*S3L_F)/511, (294*S3L_F)/511, (299*S3L_F)/511, (304*S3L_F)/511, (309*S3L_F)/511, (314*S3L_F)/511, (319*S3L_F)/511, (324*S3L_F)/511, (328*S3L_F)/511, (333*S3L_F)/511, (338*S3L_F)/511, (343*S3L_F)/511, (347*S3L_F)/511, (352*S3L_F)/511, (356*S3L_F)/511, (361*S3L_F)/511, (365*S3L_F)/511, (370*S3L_F)/511, (374*S3L_F)/511, (378*S3L_F)/511, (382*S3L_F)/511, (386*S3L_F)/511, (391*S3L_F)/511, (395*S3L_F)/511, (398*S3L_F)/511, (402*S3L_F)/511, (406*S3L_F)/511, (410*S3L_F)/511, (414*S3L_F)/511, (417*S3L_F)/511, (421*S3L_F)/511, (424*S3L_F)/511, (428*S3L_F)/511, (431*S3L_F)/511, (435*S3L_F)/511, (438*S3L_F)/511, (441*S3L_F)/511, (444*S3L_F)/511, (447*S3L_F)/511, (450*S3L_F)/511, (453*S3L_F)/511, (456*S3L_F)/511, (459*S3L_F)/511, (461*S3L_F)/511, (464*S3L_F)/511, (467*S3L_F)/511, (469*S3L_F)/511, (472*S3L_F)/511, (474*S3L_F)/511, (476*S3L_F)/511, (478*S3L_F)/511, (481*S3L_F)/511, (483*S3L_F)/511, (485*S3L_F)/511, (487*S3L_F)/511, (488*S3L_F)/511, (490*S3L_F)/511, (492*S3L_F)/511, (494*S3L_F)/511, (495*S3L_F)/511, (497*S3L_F)/511, (498*S3L_F)/511, (499*S3L_F)/511, (501*S3L_F)/511, (502*S3L_F)/511, (503*S3L_F)/511, (504*S3L_F)/511, (505*S3L_F)/511, (506*S3L_F)/511, (507*S3L_F)/511, (507*S3L_F)/511, (508*S3L_F)/511, (509*S3L_F)/511, (509*S3L_F)/511, (510*S3L_F)/511, (510*S3L_F)/511, (510*S3L_F)/511, (510*S3L_F)/511, (510*S3L_F)/511 }; #endif #define S3L_SIN_TABLE_UNIT_STEP\ (S3L_F / (S3L_SIN_TABLE_LENGTH * 4)) void S3L_vec4Init(S3L_Vec4 *v) { v->x = 0; v->y = 0; v->z = 0; v->w = S3L_F; } void S3L_vec4Set(S3L_Vec4 *v, S3L_Unit x, S3L_Unit y, S3L_Unit z, S3L_Unit w) { v->x = x; v->y = y; v->z = z; v->w = w; } void S3L_vec3Add(S3L_Vec4 *result, S3L_Vec4 added) { result->x += added.x; result->y += added.y; result->z += added.z; } void S3L_vec3Sub(S3L_Vec4 *result, S3L_Vec4 substracted) { result->x -= substracted.x; result->y -= substracted.y; result->z -= substracted.z; } void S3L_mat4Init(S3L_Mat4 m) { #define M(x,y) m[x][y] #define S S3L_F M(0,0) = S; M(1,0) = 0; M(2,0) = 0; M(3,0) = 0; M(0,1) = 0; M(1,1) = S; M(2,1) = 0; M(3,1) = 0; M(0,2) = 0; M(1,2) = 0; M(2,2) = S; M(3,2) = 0; M(0,3) = 0; M(1,3) = 0; M(2,3) = 0; M(3,3) = S; #undef M #undef S } void S3L_mat4Copy(S3L_Mat4 src, S3L_Mat4 dst) { for (uint8_t j = 0; j < 4; ++j) for (uint8_t i = 0; i < 4; ++i) dst[i][j] = src[i][j]; } S3L_Unit S3L_vec3Dot(S3L_Vec4 a, S3L_Vec4 b) { return (a.x * b.x + a.y * b.y + a.z * b.z) / S3L_F; } void S3L_reflect(S3L_Vec4 toLight, S3L_Vec4 normal, S3L_Vec4 *result) { S3L_Unit d = 2 * S3L_vec3Dot(toLight,normal); result->x = (normal.x * d) / S3L_F - toLight.x; result->y = (normal.y * d) / S3L_F - toLight.y; result->z = (normal.z * d) / S3L_F - toLight.z; } void S3L_vec3Cross(S3L_Vec4 a, S3L_Vec4 b, S3L_Vec4 *result) { result->x = a.y * b.z - a.z * b.y; result->y = a.z * b.x - a.x * b.z; result->z = a.x * b.y - a.y * b.x; } void S3L_triangleNormal(S3L_Vec4 t0, S3L_Vec4 t1, S3L_Vec4 t2, S3L_Vec4 *n) { #define ANTI_OVERFLOW 32 t1.x = (t1.x - t0.x) / ANTI_OVERFLOW; t1.y = (t1.y - t0.y) / ANTI_OVERFLOW; t1.z = (t1.z - t0.z) / ANTI_OVERFLOW; t2.x = (t2.x - t0.x) / ANTI_OVERFLOW; t2.y = (t2.y - t0.y) / ANTI_OVERFLOW; t2.z = (t2.z - t0.z) / ANTI_OVERFLOW; #undef ANTI_OVERFLOW S3L_vec3Cross(t1,t2,n); S3L_vec3Normalize(n); } void S3L_getIndexedTriangleValues( S3L_Index triangleIndex, const S3L_Index *indices, const S3L_Unit *values, uint8_t numComponents, S3L_Vec4 *v0, S3L_Vec4 *v1, S3L_Vec4 *v2) { uint32_t i0, i1; S3L_Unit *value; i0 = triangleIndex * 3; i1 = indices[i0] * numComponents; value = (S3L_Unit *) v0; if (numComponents > 4) numComponents = 4; for (uint8_t j = 0; j < numComponents; ++j) { *value = values[i1]; i1++; value++; } i0++; i1 = indices[i0] * numComponents; value = (S3L_Unit *) v1; for (uint8_t j = 0; j < numComponents; ++j) { *value = values[i1]; i1++; value++; } i0++; i1 = indices[i0] * numComponents; value = (S3L_Unit *) v2; for (uint8_t j = 0; j < numComponents; ++j) { *value = values[i1]; i1++; value++; } } void S3L_computeModelNormals(S3L_Model3D model, S3L_Unit *dst, int8_t transformNormals) { S3L_Index vPos = 0; S3L_Vec4 n; n.w = 0; S3L_Vec4 ns[S3L_NORMAL_COMPUTE_MAXIMUM_AVERAGE]; S3L_Index normalCount; for (uint32_t i = 0; i < model.vertexCount; ++i) { normalCount = 0; for (uint32_t j = 0; j < model.triangleCount * 3; j += 3) { if ( (model.triangles[j] == i) || (model.triangles[j + 1] == i) || (model.triangles[j + 2] == i)) { S3L_Vec4 t0, t1, t2; uint32_t vIndex; #define getVertex(n)\ vIndex = model.triangles[j + n] * 3;\ t##n.x = model.vertices[vIndex];\ vIndex++;\ t##n.y = model.vertices[vIndex];\ vIndex++;\ t##n.z = model.vertices[vIndex]; getVertex(0) getVertex(1) getVertex(2) #undef getVertex S3L_triangleNormal(t0,t1,t2,&(ns[normalCount])); normalCount++; if (normalCount >= S3L_NORMAL_COMPUTE_MAXIMUM_AVERAGE) break; } } n.x = S3L_F; n.y = 0; n.z = 0; if (normalCount != 0) { // compute average n.x = 0; for (uint8_t i = 0; i < normalCount; ++i) { n.x += ns[i].x; n.y += ns[i].y; n.z += ns[i].z; } n.x /= normalCount; n.y /= normalCount; n.z /= normalCount; S3L_vec3Normalize(&n); } dst[vPos] = n.x; vPos++; dst[vPos] = n.y; vPos++; dst[vPos] = n.z; vPos++; } S3L_Mat4 m; S3L_makeWorldMatrix(model.transform,m); if (transformNormals) for (S3L_Index i = 0; i < model.vertexCount * 3; i += 3) { n.x = dst[i]; n.y = dst[i + 1]; n.z = dst[i + 2]; S3L_vec4Xmat4(&n,m); dst[i] = n.x; dst[i + 1] = n.y; dst[i + 2] = n.z; } } void S3L_vec4Xmat4(S3L_Vec4 *v, S3L_Mat4 m) { S3L_Vec4 vBackup; vBackup.x = v->x; vBackup.y = v->y; vBackup.z = v->z; vBackup.w = v->w; #define dotCol(col)\ ((vBackup.x * m[col][0]) +\ (vBackup.y * m[col][1]) +\ (vBackup.z * m[col][2]) +\ (vBackup.w * m[col][3])) / S3L_F v->x = dotCol(0); v->y = dotCol(1); v->z = dotCol(2); v->w = dotCol(3); } void S3L_vec3Xmat4(S3L_Vec4 *v, S3L_Mat4 m) { S3L_Vec4 vBackup; #undef dotCol #define dotCol(col)\ (vBackup.x * m[col][0]) / S3L_F +\ (vBackup.y * m[col][1]) / S3L_F +\ (vBackup.z * m[col][2]) / S3L_F +\ m[col][3] vBackup.x = v->x; vBackup.y = v->y; vBackup.z = v->z; vBackup.w = v->w; v->x = dotCol(0); v->y = dotCol(1); v->z = dotCol(2); v->w = S3L_F; } #undef dotCol S3L_Unit S3L_abs(S3L_Unit value) { return value * (((value >= 0) << 1) - 1); } S3L_Unit S3L_min(S3L_Unit v1, S3L_Unit v2) { return v1 >= v2 ? v2 : v1; } S3L_Unit S3L_max(S3L_Unit v1, S3L_Unit v2) { return v1 >= v2 ? v1 : v2; } S3L_Unit S3L_clamp(S3L_Unit v, S3L_Unit v1, S3L_Unit v2) { return v >= v1 ? (v <= v2 ? v : v2) : v1; } S3L_Unit S3L_zeroClamp(S3L_Unit value) { return (value * (value >= 0)); } S3L_Unit S3L_wrap(S3L_Unit value, S3L_Unit mod) { return value >= 0 ? (value % mod) : (mod + (value % mod) - 1); } S3L_Unit S3L_nonZero(S3L_Unit value) { return (value + (value == 0)); } S3L_Unit S3L_interpolate(S3L_Unit v1, S3L_Unit v2, S3L_Unit t, S3L_Unit tMax) { return v1 + ((v2 - v1) * t) / tMax; } S3L_Unit S3L_interpolateByUnit(S3L_Unit v1, S3L_Unit v2, S3L_Unit t) { return v1 + ((v2 - v1) * t) / S3L_F; } S3L_Unit S3L_interpolateByUnitFrom0(S3L_Unit v2, S3L_Unit t) { return (v2 * t) / S3L_F; } S3L_Unit S3L_interpolateFrom0(S3L_Unit v2, S3L_Unit t, S3L_Unit tMax) { return (v2 * t) / tMax; } S3L_Unit S3L_distanceManhattan(S3L_Vec4 a, S3L_Vec4 b) { return S3L_abs(a.x - b.x) + S3L_abs(a.y - b.y) + S3L_abs(a.z - b.z); } void S3L_mat4Xmat4(S3L_Mat4 m1, S3L_Mat4 m2) { S3L_Mat4 mat1; for (uint16_t row = 0; row < 4; ++row) for (uint16_t col = 0; col < 4; ++col) mat1[col][row] = m1[col][row]; for (uint16_t row = 0; row < 4; ++row) for (uint16_t col = 0; col < 4; ++col) { m1[col][row] = 0; for (uint16_t i = 0; i < 4; ++i) m1[col][row] += (mat1[i][row] * m2[col][i]) / S3L_F; } } S3L_Unit S3L_sin(S3L_Unit x) { #if S3L_SIN_METHOD == 0 x = S3L_wrap(x / S3L_SIN_TABLE_UNIT_STEP,S3L_SIN_TABLE_LENGTH * 4); int8_t positive = 1; if (x < S3L_SIN_TABLE_LENGTH) { } else if (x < S3L_SIN_TABLE_LENGTH * 2) { x = S3L_SIN_TABLE_LENGTH * 2 - x - 1; } else if (x < S3L_SIN_TABLE_LENGTH * 3) { x = x - S3L_SIN_TABLE_LENGTH * 2; positive = 0; } else { x = S3L_SIN_TABLE_LENGTH - (x - S3L_SIN_TABLE_LENGTH * 3) - 1; positive = 0; } return positive ? S3L_sinTable[x] : -1 * S3L_sinTable[x]; #else int8_t sign = 1; if (x < 0) // odd function { x *= -1; sign = -1; } x %= S3L_F; if (x > S3L_F / 2) { x -= S3L_F / 2; sign *= -1; } S3L_Unit tmp = S3L_F - 2 * x; #define _PI2 ((S3L_Unit) (9.8696044 * S3L_F)) return sign * // Bhaskara's approximation (((32 * x * _PI2) / S3L_F) * tmp) / ((_PI2 * (5 * S3L_F - (8 * x * tmp) / S3L_F)) / S3L_F); #undef _PI2 #endif } S3L_Unit S3L_asin(S3L_Unit x) { #if S3L_SIN_METHOD == 0 x = S3L_clamp(x,-S3L_F,S3L_F); int8_t sign = 1; if (x < 0) { sign = -1; x *= -1; } int16_t low = 0, high = S3L_SIN_TABLE_LENGTH -1, middle; while (low <= high) // binary search { middle = (low + high) / 2; S3L_Unit v = S3L_sinTable[middle]; if (v > x) high = middle - 1; else if (v < x) low = middle + 1; else break; } middle *= S3L_SIN_TABLE_UNIT_STEP; return sign * middle; #else S3L_Unit low = -1 * S3L_F / 4, high = S3L_F / 4, middle; while (low <= high) // binary search { middle = (low + high) / 2; S3L_Unit v = S3L_sin(middle); if (v > x) high = middle - 1; else if (v < x) low = middle + 1; else break; } return middle; #endif } S3L_Unit S3L_cos(S3L_Unit x) { return S3L_sin(x + S3L_F / 4); } void S3L_correctBarycentricCoords(S3L_Unit barycentric[3]) { barycentric[0] = S3L_clamp(barycentric[0],0,S3L_F); barycentric[1] = S3L_clamp(barycentric[1],0,S3L_F); S3L_Unit d = S3L_F - barycentric[0] - barycentric[1]; if (d < 0) { barycentric[0] += d; barycentric[2] = 0; } else barycentric[2] = d; } void S3L_makeTranslationMat( S3L_Unit offsetX, S3L_Unit offsetY, S3L_Unit offsetZ, S3L_Mat4 m) { #define M(x,y) m[x][y] #define S S3L_F M(0,0) = S; M(1,0) = 0; M(2,0) = 0; M(3,0) = 0; M(0,1) = 0; M(1,1) = S; M(2,1) = 0; M(3,1) = 0; M(0,2) = 0; M(1,2) = 0; M(2,2) = S; M(3,2) = 0; M(0,3) = offsetX; M(1,3) = offsetY; M(2,3) = offsetZ; M(3,3) = S; #undef M #undef S } void S3L_makeScaleMatrix( S3L_Unit scaleX, S3L_Unit scaleY, S3L_Unit scaleZ, S3L_Mat4 m) { #define M(x,y) m[x][y] M(0,0) = scaleX; M(1,0) = 0; M(2,0) = 0; M(3,0) = 0; M(0,1) = 0; M(1,1) = scaleY; M(2,1) = 0; M(3,1) = 0; M(0,2) = 0; M(1,2) = 0; M(2,2) = scaleZ; M(3,2) = 0; M(0,3) = 0; M(1,3) = 0; M(2,3) = 0; M(3,3) = S3L_F; #undef M } void S3L_makeRotationMatrixZXY( S3L_Unit byX, S3L_Unit byY, S3L_Unit byZ, S3L_Mat4 m) { byX *= -1; byY *= -1; byZ *= -1; S3L_Unit sx = S3L_sin(byX); S3L_Unit sy = S3L_sin(byY); S3L_Unit sz = S3L_sin(byZ); S3L_Unit cx = S3L_cos(byX); S3L_Unit cy = S3L_cos(byY); S3L_Unit cz = S3L_cos(byZ); #define M(x,y) m[x][y] #define S S3L_F M(0,0) = (cy * cz) / S + (sy * sx * sz) / (S * S); M(1,0) = (cx * sz) / S; M(2,0) = (cy * sx * sz) / (S * S) - (cz * sy) / S; M(3,0) = 0; M(0,1) = (cz * sy * sx) / (S * S) - (cy * sz) / S; M(1,1) = (cx * cz) / S; M(2,1) = (cy * cz * sx) / (S * S) + (sy * sz) / S; M(3,1) = 0; M(0,2) = (cx * sy) / S; M(1,2) = -1 * sx; M(2,2) = (cy * cx) / S; M(3,2) = 0; M(0,3) = 0; M(1,3) = 0; M(2,3) = 0; M(3,3) = S3L_F; #undef M #undef S } S3L_Unit S3L_sqrt(S3L_Unit value) { int8_t sign = 1; if (value < 0) { sign = -1; value *= -1; } uint32_t result = 0; uint32_t a = value; uint32_t b = 1u << 30; while (b > a) b >>= 2; while (b != 0) { if (a >= result + b) { a -= result + b; result = result + 2 * b; } b >>= 2; result >>= 1; } return result * sign; } S3L_Unit S3L_vec3Length(S3L_Vec4 v) { return S3L_sqrt(v.x * v.x + v.y * v.y + v.z * v.z); } S3L_Unit S3L_vec2Length(S3L_Vec4 v) { return S3L_sqrt(v.x * v.x + v.y * v.y); } void S3L_vec3Normalize(S3L_Vec4 *v) { #define SCALE 16 #define BOTTOM_LIMIT 16 #define UPPER_LIMIT 900 /* Here we try to decide if the vector is too small and would cause inaccurate result due to very its inaccurate length. If so, we scale it up. We can't scale up everything as big vectors overflow in length calculations. */ if ( S3L_abs(v->x) <= BOTTOM_LIMIT && S3L_abs(v->y) <= BOTTOM_LIMIT && S3L_abs(v->z) <= BOTTOM_LIMIT) { v->x *= SCALE; v->y *= SCALE; v->z *= SCALE; } else if ( S3L_abs(v->x) > UPPER_LIMIT || S3L_abs(v->y) > UPPER_LIMIT || S3L_abs(v->z) > UPPER_LIMIT) { v->x /= SCALE; v->y /= SCALE; v->z /= SCALE; } #undef SCALE #undef BOTTOM_LIMIT #undef UPPER_LIMIT S3L_Unit l = S3L_vec3Length(*v); if (l == 0) return; v->x = (v->x * S3L_F) / l; v->y = (v->y * S3L_F) / l; v->z = (v->z * S3L_F) / l; } void S3L_vec3NormalizeFast(S3L_Vec4 *v) { S3L_Unit l = S3L_vec3Length(*v); if (l == 0) return; v->x = (v->x * S3L_F) / l; v->y = (v->y * S3L_F) / l; v->z = (v->z * S3L_F) / l; } void S3L_transform3DInit(S3L_Transform3D *t) { S3L_vec4Init(&(t->translation)); S3L_vec4Init(&(t->rotation)); t->scale.x = S3L_F; t->scale.y = S3L_F; t->scale.z = S3L_F; t->scale.w = 0; } /** Performs perspecive division (z-divide). Does NOT check for division by zero. */ static inline void S3L_perspectiveDivide(S3L_Vec4 *vector, S3L_Unit focalLength) { if (focalLength == 0) return; vector->x = (vector->x * focalLength) / vector->z; vector->y = (vector->y * focalLength) / vector->z; } void S3L_project3DPointToScreen( S3L_Vec4 point, S3L_Camera camera, S3L_Vec4 *result) { // TODO: hotfix to prevent a mapping bug probably to overlfows S3L_Vec4 toPoint = point, camForw; S3L_vec3Sub(&toPoint,camera.transform.translation); S3L_vec3Normalize(&toPoint); S3L_rotationToDirections(camera.transform.rotation,S3L_FRACTIONS_PER_UNIT, &camForw,0,0); if (S3L_vec3Dot(toPoint,camForw) < S3L_FRACTIONS_PER_UNIT / 6) { result->z = -1; result->w = 0; return; } // end of hotfix S3L_Mat4 m; S3L_makeCameraMatrix(camera.transform,m); S3L_Unit s = point.w; point.w = S3L_F; S3L_vec3Xmat4(&point,m); point.z = S3L_nonZero(point.z); S3L_perspectiveDivide(&point,camera.focalLength); S3L_ScreenCoord x, y; S3L_mapProjectionPlaneToScreen(point,&x,&y); result->x = x; result->y = y; result->z = point.z; result->w = (point.z <= 0) ? 0 : ( camera.focalLength > 0 ?( (s * camera.focalLength * S3L_RESOLUTION_X) / (point.z * S3L_F)) : ((camera.transform.scale.x * S3L_RESOLUTION_X) / S3L_F) ); } void S3L_lookAt(S3L_Vec4 pointTo, S3L_Transform3D *t) { S3L_Vec4 v; v.x = pointTo.x - t->translation.x; v.y = pointTo.z - t->translation.z; S3L_Unit dx = v.x; S3L_Unit l = S3L_vec2Length(v); dx = (v.x * S3L_F) / S3L_nonZero(l); // normalize t->rotation.y = -1 * S3L_asin(dx); if (v.y < 0) t->rotation.y = S3L_F / 2 - t->rotation.y; v.x = pointTo.y - t->translation.y; v.y = l; l = S3L_vec2Length(v); dx = (v.x * S3L_F) / S3L_nonZero(l); t->rotation.x = S3L_asin(dx); } void S3L_transform3DSet( S3L_Unit tx, S3L_Unit ty, S3L_Unit tz, S3L_Unit rx, S3L_Unit ry, S3L_Unit rz, S3L_Unit sx, S3L_Unit sy, S3L_Unit sz, S3L_Transform3D *t) { t->translation.x = tx; t->translation.y = ty; t->translation.z = tz; t->rotation.x = rx; t->rotation.y = ry; t->rotation.z = rz; t->scale.x = sx; t->scale.y = sy; t->scale.z = sz; } void S3L_cameraInit(S3L_Camera *camera) { camera->focalLength = S3L_F; S3L_transform3DInit(&(camera->transform)); } void S3L_rotationToDirections( S3L_Vec4 rotation, S3L_Unit length, S3L_Vec4 *forw, S3L_Vec4 *right, S3L_Vec4 *up) { S3L_Mat4 m; S3L_makeRotationMatrixZXY(rotation.x,rotation.y,rotation.z,m); if (forw != 0) { forw->x = 0; forw->y = 0; forw->z = length; S3L_vec3Xmat4(forw,m); } if (right != 0) { right->x = length; right->y = 0; right->z = 0; S3L_vec3Xmat4(right,m); } if (up != 0) { up->x = 0; up->y = length; up->z = 0; S3L_vec3Xmat4(up,m); } } void S3L_pixelInfoInit(S3L_PixelInfo *p) { p->x = 0; p->y = 0; p->barycentric[0] = S3L_F; p->barycentric[1] = 0; p->barycentric[2] = 0; p->modelIndex = 0; p->triangleIndex = 0; p->triangleID = 0; p->depth = 0; p->previousZ = 0; } void S3L_model3DInit( const S3L_Unit *vertices, S3L_Index vertexCount, const S3L_Index *triangles, S3L_Index triangleCount, S3L_Model3D *model) { model->vertices = vertices; model->vertexCount = vertexCount; model->triangles = triangles; model->triangleCount = triangleCount; model->customTransformMatrix = 0; S3L_transform3DInit(&(model->transform)); S3L_drawConfigInit(&(model->config)); } void S3L_sceneInit( S3L_Model3D *models, S3L_Index modelCount, S3L_Scene *scene) { scene->models = models; scene->modelCount = modelCount; S3L_cameraInit(&(scene->camera)); } void S3L_drawConfigInit(S3L_DrawConfig *config) { config->backfaceCulling = 2; config->visible = 1; } #ifndef S3L_PIXEL_FUNCTION #error Pixel rendering function (S3L_PIXEL_FUNCTION) not specified! #endif static inline void S3L_PIXEL_FUNCTION(S3L_PixelInfo *pixel); // forward decl /** Serves to accelerate linear interpolation for performance-critical code. Functions such as S3L_interpolate require division to compute each interpolated value, while S3L_FastLerpState only requires a division for the initiation and a shift for retrieving each interpolated value. S3L_FastLerpState stores a value and a step, both scaled (shifted by S3L_FAST_LERP_QUALITY) to increase precision. The step is being added to the value, which achieves the interpolation. This will only be useful for interpolations in which we need to get the interpolated value in every step. BEWARE! Shifting a negative value is undefined, so handling shifting of negative values has to be done cleverly. */ typedef struct { S3L_Unit valueScaled; S3L_Unit stepScaled; } S3L_FastLerpState; #define S3L_getFastLerpValue(state)\ (state.valueScaled >> S3L_FAST_LERP_QUALITY) #define S3L_stepFastLerp(state)\ state.valueScaled += state.stepScaled static inline S3L_Unit S3L_interpolateBarycentric( S3L_Unit value0, S3L_Unit value1, S3L_Unit value2, S3L_Unit barycentric[3]) { return ( (value0 * barycentric[0]) + (value1 * barycentric[1]) + (value2 * barycentric[2]) ) / S3L_F; } void S3L_mapProjectionPlaneToScreen( S3L_Vec4 point, S3L_ScreenCoord *screenX, S3L_ScreenCoord *screenY) { *screenX = S3L_HALF_RESOLUTION_X + (point.x * S3L_HALF_RESOLUTION_X) / S3L_F; *screenY = S3L_HALF_RESOLUTION_Y - (point.y * S3L_HALF_RESOLUTION_X) / S3L_F; } void S3L_zBufferClear(void) { #if S3L_Z_BUFFER for (uint32_t i = 0; i < S3L_RESOLUTION_X * S3L_RESOLUTION_Y; ++i) S3L_zBuffer[i] = S3L_MAX_DEPTH; #endif } void S3L_stencilBufferClear(void) { #if S3L_STENCIL_BUFFER for (uint32_t i = 0; i < S3L_STENCIL_BUFFER_SIZE; ++i) S3L_stencilBuffer[i] = 0; #endif } void S3L_newFrame(void) { S3L_zBufferClear(); S3L_stencilBufferClear(); } /* the following serves to communicate info about if the triangle has been split and how the barycentrics should be remapped. */ uint8_t _S3L_projectedTriangleState = 0; // 0 = normal, 1 = cut, 2 = split #if S3L_NEAR_CROSS_STRATEGY == 3 S3L_Vec4 _S3L_triangleRemapBarycentrics[6]; #endif void S3L_drawTriangle( S3L_Vec4 point0, S3L_Vec4 point1, S3L_Vec4 point2, S3L_Index modelIndex, S3L_Index triangleIndex) { S3L_PixelInfo p; S3L_pixelInfoInit(&p); p.modelIndex = modelIndex; p.triangleIndex = triangleIndex; p.triangleID = (modelIndex << 16) | triangleIndex; S3L_Vec4 *tPointSS, *lPointSS, *rPointSS; /* points in Screen Space (in S3L_Units, normalized by S3L_F) */ S3L_Unit *barycentric0; // bar. coord that gets higher from L to R S3L_Unit *barycentric1; // bar. coord that gets higher from R to L S3L_Unit *barycentric2; // bar. coord that gets higher from bottom up // sort the vertices: #define assignPoints(t,a,b)\ {\ tPointSS = &point##t;\ barycentric2 = &(p.barycentric[t]);\ if (S3L_triangleWinding(point##t.x,point##t.y,point##a.x,point##a.y,\ point##b.x,point##b.y) >= 0)\ {\ lPointSS = &point##a; rPointSS = &point##b;\ barycentric0 = &(p.barycentric[b]);\ barycentric1 = &(p.barycentric[a]);\ }\ else\ {\ lPointSS = &point##b; rPointSS = &point##a;\ barycentric0 = &(p.barycentric[a]);\ barycentric1 = &(p.barycentric[b]);\ }\ } if (point0.y <= point1.y) { if (point0.y <= point2.y) assignPoints(0,1,2) else assignPoints(2,0,1) } else { if (point1.y <= point2.y) assignPoints(1,0,2) else assignPoints(2,0,1) } #undef assignPoints #if S3L_FLAT *barycentric0 = S3L_F / 3; *barycentric1 = S3L_F / 3; *barycentric2 = S3L_F - 2 * (S3L_F / 3); #endif p.triangleSize[0] = rPointSS->x - lPointSS->x; p.triangleSize[1] = (rPointSS->y > lPointSS->y ? rPointSS->y : lPointSS->y) - tPointSS->y; // now draw the triangle line by line: S3L_ScreenCoord splitY; // Y of the vertically middle point of the triangle S3L_ScreenCoord endY; // bottom Y of the whole triangle int splitOnLeft; /* whether splitY is the y coord. of left or right point */ if (rPointSS->y <= lPointSS->y) { splitY = rPointSS->y; splitOnLeft = 0; endY = lPointSS->y; } else { splitY = lPointSS->y; splitOnLeft = 1; endY = rPointSS->y; } S3L_ScreenCoord currentY = tPointSS->y; /* We'll be using an algorithm similar to Bresenham line algorithm. The specifics of this algorithm are among others: - drawing possibly NON-CONTINUOUS line - NOT tracing the line exactly, but rather rasterizing one the right side of it, according to the pixel CENTERS, INCLUDING the pixel centers The principle is this: - Move vertically by pixels and accumulate the error (abs(dx/dy)). - If the error is greater than one (crossed the next pixel center), keep moving horizontally and substracting 1 from the error until it is less than 1 again. - To make this INTEGER ONLY, scale the case so that distance between pixels is equal to dy (instead of 1). This way the error becomes dx/dy * dy == dx, and we're comparing the error to (and potentially substracting) 1 * dy == dy. */ int16_t /* triangle side: left right */ lX, rX, // current x position on the screen lDx, rDx, // dx (end point - start point) lDy, rDy, // dy (end point - start point) lInc, rInc, // direction in which to increment (1 or -1) lErr, rErr, // current error (Bresenham) lErrCmp, rErrCmp, // helper for deciding comparison (> vs >=) lErrAdd, rErrAdd, // error value to add in each Bresenham cycle lErrSub, rErrSub; // error value to substract when moving in x direction S3L_FastLerpState lSideFLS, rSideFLS; #if S3L_COMPUTE_LERP_DEPTH S3L_FastLerpState lDepthFLS, rDepthFLS; #define initDepthFLS(s,p1,p2)\ s##DepthFLS.valueScaled = p1##PointSS->z << S3L_FAST_LERP_QUALITY;\ s##DepthFLS.stepScaled = ((p2##PointSS->z << S3L_FAST_LERP_QUALITY) -\ s##DepthFLS.valueScaled) / (s##Dy != 0 ? s##Dy : 1); #else #define initDepthFLS(s,p1,p2) ; #endif /* init side for the algorithm, params: s - which side (l or r) p1 - point from (t, l or r) p2 - point to (t, l or r) down - whether the side coordinate goes top-down or vice versa */ #define initSide(s,p1,p2,down)\ s##X = p1##PointSS->x;\ s##Dx = p2##PointSS->x - p1##PointSS->x;\ s##Dy = p2##PointSS->y - p1##PointSS->y;\ initDepthFLS(s,p1,p2)\ s##SideFLS.stepScaled = (S3L_F << S3L_FAST_LERP_QUALITY)\ / (s##Dy != 0 ? s##Dy : 1);\ s##SideFLS.valueScaled = 0;\ if (!down)\ {\ s##SideFLS.valueScaled =\ S3L_F << S3L_FAST_LERP_QUALITY;\ s##SideFLS.stepScaled *= -1;\ }\ s##Inc = s##Dx >= 0 ? 1 : -1;\ if (s##Dx < 0)\ {s##Err = 0; s##ErrCmp = 0;}\ else\ {s##Err = s##Dy; s##ErrCmp = 1;}\ s##ErrAdd = S3L_abs(s##Dx);\ s##ErrSub = s##Dy != 0 ? s##Dy : 1; /* don't allow 0, could lead to an infinite substracting loop */ #define stepSide(s)\ while (s##Err - s##Dy >= s##ErrCmp)\ {\ s##X += s##Inc;\ s##Err -= s##ErrSub;\ }\ s##Err += s##ErrAdd; initSide(r,t,r,1) initSide(l,t,l,1) #if S3L_PERSPECTIVE_CORRECTION /* PC is done by linearly interpolating reciprocals from which the corrected velues can be computed. See http://www.lysator.liu.se/~mikaelk/doc/perspectivetexture/ */ #if S3L_PERSPECTIVE_CORRECTION == 1 #define Z_RECIP_NUMERATOR\ (S3L_F * S3L_F * S3L_F) #elif S3L_PERSPECTIVE_CORRECTION == 2 #define Z_RECIP_NUMERATOR\ (S3L_F * S3L_F) #endif /* ^ This numerator is a number by which we divide values for the reciprocals. For PC == 2 it has to be lower because linear interpolation scaling would make it overflow -- this results in lower depth precision in bigger distance for PC == 2. */ S3L_Unit tPointRecipZ, lPointRecipZ, rPointRecipZ, /* Reciprocals of the depth of each triangle point. */ lRecip0, lRecip1, rRecip0, rRecip1; /* Helper variables for swapping the above after split. */ tPointRecipZ = Z_RECIP_NUMERATOR / S3L_nonZero(tPointSS->z); lPointRecipZ = Z_RECIP_NUMERATOR / S3L_nonZero(lPointSS->z); rPointRecipZ = Z_RECIP_NUMERATOR / S3L_nonZero(rPointSS->z); lRecip0 = tPointRecipZ; lRecip1 = lPointRecipZ; rRecip0 = tPointRecipZ; rRecip1 = rPointRecipZ; #define manageSplitPerspective(b0,b1)\ b1##Recip0 = b0##PointRecipZ;\ b1##Recip1 = b1##PointRecipZ;\ b0##Recip0 = b0##PointRecipZ;\ b0##Recip1 = tPointRecipZ; #else #define manageSplitPerspective(b0,b1) ; #endif // clip to the screen in y dimension: endY = S3L_min(endY,S3L_RESOLUTION_Y); /* Clipping above the screen (y < 0) can't be easily done here, will be handled inside the loop. */ while (currentY < endY) /* draw the triangle from top to bottom -- the bottom-most row is left out because, following from the rasterization rules (see start of the file), it is to never be rasterized. */ { if (currentY == splitY) // reached a vertical split of the triangle? { #define manageSplit(b0,b1,s0,s1)\ S3L_Unit *tmp = barycentric##b0;\ barycentric##b0 = barycentric##b1;\ barycentric##b1 = tmp;\ s0##SideFLS.valueScaled = (S3L_F\ << S3L_FAST_LERP_QUALITY) - s0##SideFLS.valueScaled;\ s0##SideFLS.stepScaled *= -1;\ manageSplitPerspective(s0,s1) if (splitOnLeft) { initSide(l,l,r,0); manageSplit(0,2,r,l) } else { initSide(r,r,l,0); manageSplit(1,2,l,r) } } stepSide(r) stepSide(l) if (currentY >= 0) /* clipping of pixels whose y < 0 (can't be easily done outside the loop because of the Bresenham-like algorithm steps) */ { p.y = currentY; // draw the horizontal line #if !S3L_FLAT S3L_Unit rowLength = S3L_nonZero(rX - lX - 1); // prevent zero div #if S3L_PERSPECTIVE_CORRECTION S3L_Unit lOverZ, lRecipZ, rOverZ, rRecipZ, lT, rT; lT = S3L_getFastLerpValue(lSideFLS); rT = S3L_getFastLerpValue(rSideFLS); lOverZ = S3L_interpolateByUnitFrom0(lRecip1,lT); lRecipZ = S3L_interpolateByUnit(lRecip0,lRecip1,lT); rOverZ = S3L_interpolateByUnitFrom0(rRecip1,rT); rRecipZ = S3L_interpolateByUnit(rRecip0,rRecip1,rT); #else S3L_FastLerpState b0FLS, b1FLS; #if S3L_COMPUTE_LERP_DEPTH S3L_FastLerpState depthFLS; depthFLS.valueScaled = lDepthFLS.valueScaled; depthFLS.stepScaled = (rDepthFLS.valueScaled - lDepthFLS.valueScaled) / rowLength; #endif b0FLS.valueScaled = 0; b1FLS.valueScaled = lSideFLS.valueScaled; b0FLS.stepScaled = rSideFLS.valueScaled / rowLength; b1FLS.stepScaled = -1 * lSideFLS.valueScaled / rowLength; #endif #endif // clip to the screen in x dimension: S3L_ScreenCoord rXClipped = S3L_min(rX,S3L_RESOLUTION_X), lXClipped = lX; if (lXClipped < 0) { lXClipped = 0; #if !S3L_PERSPECTIVE_CORRECTION && !S3L_FLAT b0FLS.valueScaled -= lX * b0FLS.stepScaled; b1FLS.valueScaled -= lX * b1FLS.stepScaled; #if S3L_COMPUTE_LERP_DEPTH depthFLS.valueScaled -= lX * depthFLS.stepScaled; #endif #endif } #if S3L_PERSPECTIVE_CORRECTION S3L_ScreenCoord i = lXClipped - lX; /* helper var to save one substraction in the inner loop */ #endif #if S3L_PERSPECTIVE_CORRECTION == 2 S3L_FastLerpState depthPC, // interpolates depth between row segments b0PC, // interpolates barycentric0 between row segments b1PC; // interpolates barycentric1 between row segments /* ^ These interpolate values between row segments (lines of pixels of S3L_PC_APPROX_LENGTH length). After each row segment perspective correction is recomputed. */ depthPC.valueScaled = (Z_RECIP_NUMERATOR / S3L_nonZero(S3L_interpolate(lRecipZ,rRecipZ,i,rowLength))) << S3L_FAST_LERP_QUALITY; b0PC.valueScaled = ( S3L_interpolateFrom0(rOverZ,i,rowLength) * depthPC.valueScaled ) / (Z_RECIP_NUMERATOR / S3L_F); b1PC.valueScaled = ( (lOverZ - S3L_interpolateFrom0(lOverZ,i,rowLength)) * depthPC.valueScaled ) / (Z_RECIP_NUMERATOR / S3L_F); int8_t rowCount = S3L_PC_APPROX_LENGTH; #endif #if S3L_Z_BUFFER uint32_t zBufferIndex = p.y * S3L_RESOLUTION_X + lXClipped; #endif // draw the row -- inner loop: for (S3L_ScreenCoord x = lXClipped; x < rXClipped; ++x) { int8_t testsPassed = 1; #if S3L_STENCIL_BUFFER if (!S3L_stencilTest(x,p.y)) testsPassed = 0; #endif p.x = x; #if S3L_COMPUTE_DEPTH #if S3L_PERSPECTIVE_CORRECTION == 1 p.depth = Z_RECIP_NUMERATOR / S3L_nonZero(S3L_interpolate(lRecipZ,rRecipZ,i,rowLength)); #elif S3L_PERSPECTIVE_CORRECTION == 2 if (rowCount >= S3L_PC_APPROX_LENGTH) { // init the linear interpolation to the next PC correct value rowCount = 0; S3L_Unit nextI = i + S3L_PC_APPROX_LENGTH; if (nextI < rowLength) { S3L_Unit nextDepthScaled = ( Z_RECIP_NUMERATOR / S3L_nonZero(S3L_interpolate(lRecipZ,rRecipZ,nextI,rowLength)) ) << S3L_FAST_LERP_QUALITY; depthPC.stepScaled = (nextDepthScaled - depthPC.valueScaled) / S3L_PC_APPROX_LENGTH; S3L_Unit nextValue = ( S3L_interpolateFrom0(rOverZ,nextI,rowLength) * nextDepthScaled ) / (Z_RECIP_NUMERATOR / S3L_F); b0PC.stepScaled = (nextValue - b0PC.valueScaled) / S3L_PC_APPROX_LENGTH; nextValue = ( (lOverZ - S3L_interpolateFrom0(lOverZ,nextI,rowLength)) * nextDepthScaled ) / (Z_RECIP_NUMERATOR / S3L_F); b1PC.stepScaled = (nextValue - b1PC.valueScaled) / S3L_PC_APPROX_LENGTH; } else { /* A special case where we'd be interpolating outside the triangle. It seems like a valid approach at first, but it creates a bug in a case when the rasaterized triangle is near screen 0 and can actually never reach the extrapolated screen position. So we have to clamp to the actual end of the triangle here. */ S3L_Unit maxI = S3L_nonZero(rowLength - i); S3L_Unit nextDepthScaled = ( Z_RECIP_NUMERATOR / S3L_nonZero(rRecipZ) ) << S3L_FAST_LERP_QUALITY; depthPC.stepScaled = (nextDepthScaled - depthPC.valueScaled) / maxI; S3L_Unit nextValue = ( rOverZ * nextDepthScaled ) / (Z_RECIP_NUMERATOR / S3L_F); b0PC.stepScaled = (nextValue - b0PC.valueScaled) / maxI; b1PC.stepScaled = -1 * b1PC.valueScaled / maxI; } } p.depth = S3L_getFastLerpValue(depthPC); #else p.depth = S3L_getFastLerpValue(depthFLS); S3L_stepFastLerp(depthFLS); #endif #else // !S3L_COMPUTE_DEPTH p.depth = (tPointSS->z + lPointSS->z + rPointSS->z) / 3; #endif #if S3L_Z_BUFFER p.previousZ = S3L_zBuffer[zBufferIndex]; zBufferIndex++; if (!S3L_zTest(p.x,p.y,p.depth)) testsPassed = 0; #endif if (testsPassed) { #if !S3L_FLAT #if S3L_PERSPECTIVE_CORRECTION == 0 *barycentric0 = S3L_getFastLerpValue(b0FLS); *barycentric1 = S3L_getFastLerpValue(b1FLS); #elif S3L_PERSPECTIVE_CORRECTION == 1 *barycentric0 = ( S3L_interpolateFrom0(rOverZ,i,rowLength) * p.depth ) / (Z_RECIP_NUMERATOR / S3L_F); *barycentric1 = ( (lOverZ - S3L_interpolateFrom0(lOverZ,i,rowLength)) * p.depth ) / (Z_RECIP_NUMERATOR / S3L_F); #elif S3L_PERSPECTIVE_CORRECTION == 2 *barycentric0 = S3L_getFastLerpValue(b0PC); *barycentric1 = S3L_getFastLerpValue(b1PC); #endif *barycentric2 = S3L_F - *barycentric0 - *barycentric1; #endif #if S3L_NEAR_CROSS_STRATEGY == 3 if (_S3L_projectedTriangleState != 0) { S3L_Unit newBarycentric[3]; newBarycentric[0] = S3L_interpolateBarycentric( _S3L_triangleRemapBarycentrics[0].x, _S3L_triangleRemapBarycentrics[1].x, _S3L_triangleRemapBarycentrics[2].x, p.barycentric); newBarycentric[1] = S3L_interpolateBarycentric( _S3L_triangleRemapBarycentrics[0].y, _S3L_triangleRemapBarycentrics[1].y, _S3L_triangleRemapBarycentrics[2].y, p.barycentric); newBarycentric[2] = S3L_interpolateBarycentric( _S3L_triangleRemapBarycentrics[0].z, _S3L_triangleRemapBarycentrics[1].z, _S3L_triangleRemapBarycentrics[2].z, p.barycentric); p.barycentric[0] = newBarycentric[0]; p.barycentric[1] = newBarycentric[1]; p.barycentric[2] = newBarycentric[2]; } #endif S3L_PIXEL_FUNCTION(&p); } // tests passed #if !S3L_FLAT #if S3L_PERSPECTIVE_CORRECTION i++; #if S3L_PERSPECTIVE_CORRECTION == 2 rowCount++; S3L_stepFastLerp(depthPC); S3L_stepFastLerp(b0PC); S3L_stepFastLerp(b1PC); #endif #else S3L_stepFastLerp(b0FLS); S3L_stepFastLerp(b1FLS); #endif #endif } // inner loop } // y clipping #if !S3L_FLAT S3L_stepFastLerp(lSideFLS); S3L_stepFastLerp(rSideFLS); #if S3L_COMPUTE_LERP_DEPTH S3L_stepFastLerp(lDepthFLS); S3L_stepFastLerp(rDepthFLS); #endif #endif ++currentY; } // row drawing #undef manageSplit #undef initPC #undef initSide #undef stepSide #undef Z_RECIP_NUMERATOR } void S3L_rotate2DPoint(S3L_Unit *x, S3L_Unit *y, S3L_Unit angle) { if (angle < S3L_SIN_TABLE_UNIT_STEP) return; // no visible rotation S3L_Unit angleSin = S3L_sin(angle); S3L_Unit angleCos = S3L_cos(angle); S3L_Unit xBackup = *x; *x = (angleCos * (*x)) / S3L_F - (angleSin * (*y)) / S3L_F; *y = (angleSin * xBackup) / S3L_F + (angleCos * (*y)) / S3L_F; } void S3L_makeWorldMatrix(S3L_Transform3D worldTransform, S3L_Mat4 m) { S3L_makeScaleMatrix( worldTransform.scale.x, worldTransform.scale.y, worldTransform.scale.z, m); S3L_Mat4 t; S3L_makeRotationMatrixZXY( worldTransform.rotation.x, worldTransform.rotation.y, worldTransform.rotation.z, t); S3L_mat4Xmat4(m,t); S3L_makeTranslationMat( worldTransform.translation.x, worldTransform.translation.y, worldTransform.translation.z, t); S3L_mat4Xmat4(m,t); } void S3L_mat4Transpose(S3L_Mat4 m) { S3L_Unit tmp; for (uint8_t y = 0; y < 3; ++y) for (uint8_t x = 1 + y; x < 4; ++x) { tmp = m[x][y]; m[x][y] = m[y][x]; m[y][x] = tmp; } } void S3L_makeCameraMatrix(S3L_Transform3D cameraTransform, S3L_Mat4 m) { S3L_makeTranslationMat( -1 * cameraTransform.translation.x, -1 * cameraTransform.translation.y, -1 * cameraTransform.translation.z, m); S3L_Mat4 r; S3L_makeRotationMatrixZXY( cameraTransform.rotation.x, cameraTransform.rotation.y, cameraTransform.rotation.z, r); S3L_mat4Transpose(r); // transposing creates an inverse transform S3L_Mat4 s; S3L_makeScaleMatrix( cameraTransform.scale.x, cameraTransform.scale.y, cameraTransform.scale.z,s); S3L_mat4Xmat4(m,r); S3L_mat4Xmat4(m,s); } int8_t S3L_triangleWinding( S3L_ScreenCoord x0, S3L_ScreenCoord y0, S3L_ScreenCoord x1, S3L_ScreenCoord y1, S3L_ScreenCoord x2, S3L_ScreenCoord y2) { int32_t winding = (y1 - y0) * (x2 - x1) - (x1 - x0) * (y2 - y1); // ^ cross product for points with z == 0 return winding > 0 ? 1 : (winding < 0 ? -1 : 0); } /** Checks if given triangle (in Screen Space) is at least partially visible, i.e. returns false if the triangle is either completely outside the frustum (left, right, top, bottom, near) or is invisible due to backface culling. */ static inline int8_t S3L_triangleIsVisible( S3L_Vec4 p0, S3L_Vec4 p1, S3L_Vec4 p2, uint8_t backfaceCulling) { #define clipTest(c,cmp,v)\ (p0.c cmp (v) && p1.c cmp (v) && p2.c cmp (v)) if ( // outside frustum? #if S3L_NEAR_CROSS_STRATEGY == 0 p0.z <= S3L_NEAR || p1.z <= S3L_NEAR || p2.z <= S3L_NEAR || // ^ partially in front of NEAR? #else clipTest(z,<=,S3L_NEAR) || // completely in front of NEAR? #endif clipTest(x,<,0) || clipTest(x,>=,S3L_RESOLUTION_X) || clipTest(y,<,0) || clipTest(y,>,S3L_RESOLUTION_Y) ) return 0; #undef clipTest if (backfaceCulling != 0) { int8_t winding = S3L_triangleWinding(p0.x,p0.y,p1.x,p1.y,p2.x,p2.y); if ((backfaceCulling == 1 && winding > 0) || (backfaceCulling == 2 && winding < 0)) return 0; } return 1; } #if S3L_SORT != 0 typedef struct { uint8_t modelIndex; S3L_Index triangleIndex; uint16_t sortValue; } _S3L_TriangleToSort; _S3L_TriangleToSort S3L_sortArray[S3L_MAX_TRIANGES_DRAWN]; uint16_t S3L_sortArrayLength; #endif void _S3L_projectVertex(const S3L_Model3D *model, S3L_Index triangleIndex, uint8_t vertex, S3L_Mat4 projectionMatrix, S3L_Vec4 *result) { uint32_t vertexIndex = model->triangles[triangleIndex * 3 + vertex] * 3; result->x = model->vertices[vertexIndex]; result->y = model->vertices[vertexIndex + 1]; result->z = model->vertices[vertexIndex + 2]; result->w = S3L_F; // needed for translation S3L_vec3Xmat4(result,projectionMatrix); result->w = result->z; /* We'll keep the non-clamped z in w for sorting. */ } void _S3L_mapProjectedVertexToScreen(S3L_Vec4 *vertex, S3L_Unit focalLength) { vertex->z = vertex->z >= S3L_NEAR ? vertex->z : S3L_NEAR; /* ^ This firstly prevents zero division in the follwoing z-divide and secondly "pushes" vertices that are in front of near a little bit forward, which makes them behave a bit better. If all three vertices end up exactly on NEAR, the triangle will be culled. */ S3L_perspectiveDivide(vertex,focalLength); S3L_ScreenCoord sX, sY; S3L_mapProjectionPlaneToScreen(*vertex,&sX,&sY); vertex->x = sX; vertex->y = sY; } /** Projects a triangle to the screen. If enabled, a triangle can be potentially subdivided into two if it crosses the near plane, in which case two projected triangles are returned (the info about splitting or cutting the triangle is passed in global variables, see above). */ void _S3L_projectTriangle( const S3L_Model3D *model, S3L_Index triangleIndex, S3L_Mat4 matrix, uint32_t focalLength, S3L_Vec4 transformed[6]) { _S3L_projectVertex(model,triangleIndex,0,matrix,&(transformed[0])); _S3L_projectVertex(model,triangleIndex,1,matrix,&(transformed[1])); _S3L_projectVertex(model,triangleIndex,2,matrix,&(transformed[2])); _S3L_projectedTriangleState = 0; #if S3L_NEAR_CROSS_STRATEGY == 2 || S3L_NEAR_CROSS_STRATEGY == 3 uint8_t infront = 0; uint8_t behind = 0; uint8_t infrontI[3]; uint8_t behindI[3]; for (uint8_t i = 0; i < 3; ++i) if (transformed[i].z < S3L_NEAR) { infrontI[infront] = i; infront++; } else { behindI[behind] = i; behind++; } #if S3L_NEAR_CROSS_STRATEGY == 3 for (int i = 0; i < 3; ++i) S3L_vec4Init(&(_S3L_triangleRemapBarycentrics[i])); _S3L_triangleRemapBarycentrics[0].x = S3L_F; _S3L_triangleRemapBarycentrics[1].y = S3L_F; _S3L_triangleRemapBarycentrics[2].z = S3L_F; #endif #define interpolateVertex \ S3L_Unit ratio =\ ((transformed[be].z - S3L_NEAR) * S3L_F) /\ (transformed[be].z - transformed[in].z);\ transformed[in].x = transformed[be].x - \ ((transformed[be].x - transformed[in].x) * ratio) /\ S3L_F;\ transformed[in].y = transformed[be].y -\ ((transformed[be].y - transformed[in].y) * ratio) /\ S3L_F;\ transformed[in].z = S3L_NEAR;\ if (beI != 0) {\ beI->x = (beI->x * ratio) / S3L_F;\ beI->y = (beI->y * ratio) / S3L_F;\ beI->z = (beI->z * ratio) / S3L_F;\ ratio = S3L_F - ratio;\ beI->x += (beB->x * ratio) / S3L_F;\ beI->y += (beB->y * ratio) / S3L_F;\ beI->z += (beB->z * ratio) / S3L_F; } if (infront == 2) { // shift the two vertices forward along the edge for (uint8_t i = 0; i < 2; ++i) { uint8_t be = behindI[0], in = infrontI[i]; #if S3L_NEAR_CROSS_STRATEGY == 3 S3L_Vec4 *beI = &(_S3L_triangleRemapBarycentrics[in]), *beB = &(_S3L_triangleRemapBarycentrics[be]); #else S3L_Vec4 *beI = 0, *beB = 0; #endif interpolateVertex _S3L_projectedTriangleState = 1; } } else if (infront == 1) { // create another triangle and do the shifts transformed[3] = transformed[behindI[1]]; transformed[4] = transformed[infrontI[0]]; transformed[5] = transformed[infrontI[0]]; #if S3L_NEAR_CROSS_STRATEGY == 3 _S3L_triangleRemapBarycentrics[3] = _S3L_triangleRemapBarycentrics[behindI[1]]; _S3L_triangleRemapBarycentrics[4] = _S3L_triangleRemapBarycentrics[infrontI[0]]; _S3L_triangleRemapBarycentrics[5] = _S3L_triangleRemapBarycentrics[infrontI[0]]; #endif for (uint8_t i = 0; i < 2; ++i) { uint8_t be = behindI[i], in = i + 4; #if S3L_NEAR_CROSS_STRATEGY == 3 S3L_Vec4 *beI = &(_S3L_triangleRemapBarycentrics[in]), *beB = &(_S3L_triangleRemapBarycentrics[be]); #else S3L_Vec4 *beI = 0, *beB = 0; #endif interpolateVertex } #if S3L_NEAR_CROSS_STRATEGY == 3 _S3L_triangleRemapBarycentrics[infrontI[0]] = _S3L_triangleRemapBarycentrics[4]; #endif transformed[infrontI[0]] = transformed[4]; _S3L_mapProjectedVertexToScreen(&transformed[3],focalLength); _S3L_mapProjectedVertexToScreen(&transformed[4],focalLength); _S3L_mapProjectedVertexToScreen(&transformed[5],focalLength); _S3L_projectedTriangleState = 2; } #undef interpolateVertex #endif // S3L_NEAR_CROSS_STRATEGY == 2 _S3L_mapProjectedVertexToScreen(&transformed[0],focalLength); _S3L_mapProjectedVertexToScreen(&transformed[1],focalLength); _S3L_mapProjectedVertexToScreen(&transformed[2],focalLength); } void S3L_drawScene(S3L_Scene scene) { S3L_Mat4 matFinal, matCamera; S3L_Vec4 transformed[6]; // transformed triangle coords, for 2 triangles const S3L_Model3D *model; S3L_Index modelIndex, triangleIndex; S3L_makeCameraMatrix(scene.camera.transform,matCamera); #if S3L_SORT != 0 uint16_t previousModel = 0; S3L_sortArrayLength = 0; #endif for (modelIndex = 0; modelIndex < scene.modelCount; ++modelIndex) { if (!scene.models[modelIndex].config.visible) continue; #if S3L_SORT != 0 if (S3L_sortArrayLength >= S3L_MAX_TRIANGES_DRAWN) break; previousModel = modelIndex; #endif if (scene.models[modelIndex].customTransformMatrix == 0) S3L_makeWorldMatrix(scene.models[modelIndex].transform,matFinal); else { S3L_Mat4 *m = scene.models[modelIndex].customTransformMatrix; for (int8_t j = 0; j < 4; ++j) for (int8_t i = 0; i < 4; ++i) matFinal[i][j] = (*m)[i][j]; } S3L_mat4Xmat4(matFinal,matCamera); S3L_Index triangleCount = scene.models[modelIndex].triangleCount; triangleIndex = 0; model = &(scene.models[modelIndex]); while (triangleIndex < triangleCount) { /* Some kind of cache could be used in theory to not project perviously already projected vertices, but after some testing this was abandoned, no gain was seen. */ _S3L_projectTriangle(model,triangleIndex,matFinal, scene.camera.focalLength,transformed); if (S3L_triangleIsVisible(transformed[0],transformed[1],transformed[2], model->config.backfaceCulling)) { #if S3L_SORT == 0 // without sorting draw right away S3L_drawTriangle(transformed[0],transformed[1],transformed[2],modelIndex, triangleIndex); if (_S3L_projectedTriangleState == 2) // draw potential subtriangle { #if S3L_NEAR_CROSS_STRATEGY == 3 _S3L_triangleRemapBarycentrics[0] = _S3L_triangleRemapBarycentrics[3]; _S3L_triangleRemapBarycentrics[1] = _S3L_triangleRemapBarycentrics[4]; _S3L_triangleRemapBarycentrics[2] = _S3L_triangleRemapBarycentrics[5]; #endif S3L_drawTriangle(transformed[3],transformed[4],transformed[5], modelIndex, triangleIndex); } #else if (S3L_sortArrayLength >= S3L_MAX_TRIANGES_DRAWN) break; // with sorting add to a sort list S3L_sortArray[S3L_sortArrayLength].modelIndex = modelIndex; S3L_sortArray[S3L_sortArrayLength].triangleIndex = triangleIndex; S3L_sortArray[S3L_sortArrayLength].sortValue = S3L_zeroClamp( transformed[0].w + transformed[1].w + transformed[2].w) >> 2; /* ^ The w component here stores non-clamped z. As a simple approximation we sort by the triangle center point, which is a mean coordinate -- we don't actually have to divide by 3 (or anything), that is unnecessary for sorting! We shift by 2 just as a fast operation to prevent overflow of the sum over uint_16t. */ S3L_sortArrayLength++; #endif } triangleIndex++; } } #if S3L_SORT != 0 #if S3L_SORT == 1 #define cmp < #else #define cmp > #endif /* Sort the triangles. We use insertion sort, because it has many advantages, especially for smaller arrays (better than bubble sort, in-place, stable, simple, ...). */ for (int16_t i = 1; i < S3L_sortArrayLength; ++i) { _S3L_TriangleToSort tmp = S3L_sortArray[i]; int16_t j = i - 1; while (j >= 0 && S3L_sortArray[j].sortValue cmp tmp.sortValue) { S3L_sortArray[j + 1] = S3L_sortArray[j]; j--; } S3L_sortArray[j + 1] = tmp; } #undef cmp for (S3L_Index i = 0; i < S3L_sortArrayLength; ++i) // draw sorted triangles { modelIndex = S3L_sortArray[i].modelIndex; triangleIndex = S3L_sortArray[i].triangleIndex; model = &(scene.models[modelIndex]); if (modelIndex != previousModel) { // only recompute the matrix when the model has changed S3L_makeWorldMatrix(model->transform,matFinal); S3L_mat4Xmat4(matFinal,matCamera); previousModel = modelIndex; } /* Here we project the points again, which is redundant and slow as they've already been projected above, but saving the projected points would require a lot of memory, which for small resolutions could be even worse than z-bufer. So this seems to be the best way memory-wise. */ _S3L_projectTriangle(model,triangleIndex,matFinal,scene.camera.focalLength, transformed); S3L_drawTriangle(transformed[0],transformed[1],transformed[2],modelIndex, triangleIndex); if (_S3L_projectedTriangleState == 2) { #if S3L_NEAR_CROSS_STRATEGY == 3 _S3L_triangleRemapBarycentrics[0] = _S3L_triangleRemapBarycentrics[3]; _S3L_triangleRemapBarycentrics[1] = _S3L_triangleRemapBarycentrics[4]; _S3L_triangleRemapBarycentrics[2] = _S3L_triangleRemapBarycentrics[5]; #endif S3L_drawTriangle(transformed[3],transformed[4],transformed[5], modelIndex, triangleIndex); } } #endif } #endif // guard