/************************************************************************** * * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. * All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the * "Software"), to deal in the Software without restriction, including * without limitation the rights to use, copy, modify, merge, publish, * distribute, sub license, and/or sell copies of the Software, and to * permit persons to whom the Software is furnished to do so, subject to * the following conditions: * * The above copyright notice and this permission notice (including the * next paragraph) shall be included in all copies or substantial portions * of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. * **************************************************************************/ /** * Texture sampling * * Authors: * Brian Paul */ #include "main/macros.h" #include "sp_context.h" #include "sp_surface.h" #include "sp_tex_sample.h" #include "pipe/p_context.h" #include "pipe/p_defines.h" #include "pipe/tgsi/core/tgsi_exec.h" /* * Note, the FRAC macro has to work perfectly. Otherwise you'll sometimes * see 1-pixel bands of improperly weighted linear-filtered textures. * The tests/texwrap.c demo is a good test. * Also note, FRAC(x) doesn't truly return the fractional part of x for x < 0. * Instead, if x < 0 then FRAC(x) = 1 - true_frac(x). */ #define FRAC(f) ((f) - IFLOOR(f)) /** * Linear interpolation macro */ #define LERP(T, A, B) ( (A) + (T) * ((B) - (A)) ) /** * Do 2D/biliner interpolation of float values. * v00, v10, v01 and v11 are typically four texture samples in a square/box. * a and b are the horizontal and vertical interpolants. * It's important that this function is inlined when compiled with * optimization! If we find that's not true on some systems, convert * to a macro. */ static INLINE float lerp_2d(float a, float b, float v00, float v10, float v01, float v11) { const float temp0 = LERP(a, v00, v10); const float temp1 = LERP(a, v01, v11); return LERP(b, temp0, temp1); } /** * Compute the remainder of a divided by b, but be careful with * negative values so that REPEAT mode works right. */ static INLINE GLint repeat_remainder(GLint a, GLint b) { if (a >= 0) return a % b; else return (a + 1) % b + b - 1; } /** * Apply texture coord wrapping mode and return integer texture index. * \param wrapMode PIPE_TEX_WRAP_x * \param s the texcoord * \param size the texture image size * \return integer texture index */ static INLINE GLint nearest_texcoord(GLuint wrapMode, float s, GLuint size) { GLint i; switch (wrapMode) { case PIPE_TEX_WRAP_REPEAT: /* s limited to [0,1) */ /* i limited to [0,size-1] */ i = IFLOOR(s * size); i = repeat_remainder(i, size); return i; case PIPE_TEX_WRAP_CLAMP: /* s limited to [0,1] */ /* i limited to [0,size-1] */ if (s <= 0.0F) i = 0; else if (s >= 1.0F) i = size - 1; else i = IFLOOR(s * size); return i; case PIPE_TEX_WRAP_CLAMP_TO_EDGE: { /* s limited to [min,max] */ /* i limited to [0, size-1] */ const float min = 1.0F / (2.0F * size); const float max = 1.0F - min; if (s < min) i = 0; else if (s > max) i = size - 1; else i = IFLOOR(s * size); } return i; case PIPE_TEX_WRAP_CLAMP_TO_BORDER: { /* s limited to [min,max] */ /* i limited to [-1, size] */ const float min = -1.0F / (2.0F * size); const float max = 1.0F - min; if (s <= min) i = -1; else if (s >= max) i = size; else i = IFLOOR(s * size); } return i; case PIPE_TEX_WRAP_MIRROR_REPEAT: { const float min = 1.0F / (2.0F * size); const float max = 1.0F - min; const GLint flr = IFLOOR(s); float u; if (flr & 1) u = 1.0F - (s - (float) flr); else u = s - (float) flr; if (u < min) i = 0; else if (u > max) i = size - 1; else i = IFLOOR(u * size); } return i; case PIPE_TEX_WRAP_MIRROR_CLAMP: { /* s limited to [0,1] */ /* i limited to [0,size-1] */ const float u = FABSF(s); if (u <= 0.0F) i = 0; else if (u >= 1.0F) i = size - 1; else i = IFLOOR(u * size); } return i; case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: { /* s limited to [min,max] */ /* i limited to [0, size-1] */ const float min = 1.0F / (2.0F * size); const float max = 1.0F - min; const float u = FABSF(s); if (u < min) i = 0; else if (u > max) i = size - 1; else i = IFLOOR(u * size); } return i; case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: { /* s limited to [min,max] */ /* i limited to [0, size-1] */ const float min = -1.0F / (2.0F * size); const float max = 1.0F - min; const float u = FABSF(s); if (u < min) i = -1; else if (u > max) i = size; else i = IFLOOR(u * size); } return i; default: assert(0); return 0; } } /** * Used to compute texel locations for linear sampling. * \param wrapMode PIPE_TEX_WRAP_x * \param s the texcoord * \param size the texture image size * \param i0 returns first texture index * \param i1 returns second texture index (usually *i0 + 1) * \param a returns blend factor/weight between texture indexes */ static INLINE void linear_texcoord(GLuint wrapMode, float s, GLuint size, GLint *i0, GLint *i1, float *a) { float u; switch (wrapMode) { case PIPE_TEX_WRAP_REPEAT: u = s * size - 0.5F; *i0 = repeat_remainder(IFLOOR(u), size); *i1 = repeat_remainder(*i0 + 1, size); break; case PIPE_TEX_WRAP_CLAMP: if (s <= 0.0F) u = 0.0F; else if (s >= 1.0F) u = (float) size; else u = s * size; u -= 0.5F; *i0 = IFLOOR(u); *i1 = *i0 + 1; break; case PIPE_TEX_WRAP_CLAMP_TO_EDGE: if (s <= 0.0F) u = 0.0F; else if (s >= 1.0F) u = (float) size; else u = s * size; u -= 0.5F; *i0 = IFLOOR(u); *i1 = *i0 + 1; if (*i0 < 0) *i0 = 0; if (*i1 >= (GLint) size) *i1 = size - 1; break; case PIPE_TEX_WRAP_CLAMP_TO_BORDER: { const float min = -1.0F / (2.0F * size); const float max = 1.0F - min; if (s <= min) u = min * size; else if (s >= max) u = max * size; else u = s * size; u -= 0.5F; *i0 = IFLOOR(u); *i1 = *i0 + 1; } break; case PIPE_TEX_WRAP_MIRROR_REPEAT: { const GLint flr = IFLOOR(s); if (flr & 1) u = 1.0F - (s - (float) flr); else u = s - (float) flr; u = (u * size) - 0.5F; *i0 = IFLOOR(u); *i1 = *i0 + 1; if (*i0 < 0) *i0 = 0; if (*i1 >= (GLint) size) *i1 = size - 1; } break; case PIPE_TEX_WRAP_MIRROR_CLAMP: u = FABSF(s); if (u >= 1.0F) u = (float) size; else u *= size; u -= 0.5F; *i0 = IFLOOR(u); *i1 = *i0 + 1; break; case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: u = FABSF(s); if (u >= 1.0F) u = (float) size; else u *= size; u -= 0.5F; *i0 = IFLOOR(u); *i1 = *i0 + 1; if (*i0 < 0) *i0 = 0; if (*i1 >= (GLint) size) *i1 = size - 1; break; case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: { const float min = -1.0F / (2.0F * size); const float max = 1.0F - min; u = FABSF(s); if (u <= min) u = min * size; else if (u >= max) u = max * size; else u *= size; u -= 0.5F; *i0 = IFLOOR(u); *i1 = *i0 + 1; } break; default: assert(0); } *a = FRAC(u); } static GLuint choose_cube_face(float rx, float ry, float rz, float newCoord[4]) { /* major axis direction target sc tc ma ---------- ------------------------------- --- --- --- +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz */ const float arx = FABSF(rx), ary = FABSF(ry), arz = FABSF(rz); GLuint face; float sc, tc, ma; if (arx > ary && arx > arz) { if (rx >= 0.0F) { face = PIPE_TEX_FACE_POS_X; sc = -rz; tc = -ry; ma = arx; } else { face = PIPE_TEX_FACE_NEG_X; sc = rz; tc = -ry; ma = arx; } } else if (ary > arx && ary > arz) { if (ry >= 0.0F) { face = PIPE_TEX_FACE_POS_Y; sc = rx; tc = rz; ma = ary; } else { face = PIPE_TEX_FACE_NEG_Y; sc = rx; tc = -rz; ma = ary; } } else { if (rz > 0.0F) { face = PIPE_TEX_FACE_POS_Z; sc = rx; tc = -ry; ma = arz; } else { face = PIPE_TEX_FACE_NEG_Z; sc = -rx; tc = -ry; ma = arz; } } newCoord[0] = ( sc / ma + 1.0F ) * 0.5F; newCoord[1] = ( tc / ma + 1.0F ) * 0.5F; return face; } /** * Examine the quad's texture coordinates to compute the partial * derivatives w.r.t X and Y, then compute lambda (level of detail). * * This is only done for fragment shaders, not vertex shaders. */ static float compute_lambda(struct tgsi_sampler *sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], float lodbias) { float rho, lambda; assert(s); { float dsdx = s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]; float dsdy = s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]; dsdx = FABSF(dsdx); dsdy = FABSF(dsdy); rho = MAX2(dsdx, dsdy) * sampler->texture->width0; } if (t) { float dtdx = t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]; float dtdy = t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]; float max; dtdx = FABSF(dtdx); dtdy = FABSF(dtdy); max = MAX2(dtdx, dtdy) * sampler->texture->height0; rho = MAX2(rho, max); } if (p) { float dpdx = p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]; float dpdy = p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT]; float max; dpdx = FABSF(dpdx); dpdy = FABSF(dpdy); max = MAX2(dpdx, dpdy) * sampler->texture->depth0; rho = MAX2(rho, max); } lambda = LOG2(rho); lambda += lodbias + sampler->state->lod_bias; lambda = CLAMP(lambda, sampler->state->min_lod, sampler->state->max_lod); return lambda; } static void sp_get_samples_1d(struct tgsi_sampler *sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], float lodbias, float rgba[NUM_CHANNELS][QUAD_SIZE]) { struct pipe_context *pipe = (struct pipe_context *) sampler->pipe; struct pipe_surface *ps = pipe->get_tex_surface(pipe, sampler->texture, 0, 0, 0); switch (sampler->state->min_img_filter) { case PIPE_TEX_FILTER_NEAREST: { GLuint j; for (j = 0; j < QUAD_SIZE; j++) { GLint x = nearest_texcoord(sampler->state->wrap_s, s[j], sampler->texture->width0); float texel[4]; ps->get_tile(ps, x, 0, 1, 1, texel); rgba[0][j] = texel[0]; rgba[1][j] = texel[1]; rgba[2][j] = texel[2]; rgba[3][j] = texel[3]; } } break; case PIPE_TEX_FILTER_LINEAR: { GLuint j; for (j = 0; j < QUAD_SIZE; j++) { float t0[4], t1[4], texel[4]; GLint x0, x1; float a; linear_texcoord(sampler->state->wrap_s, s[j], sampler->texture->width0, &x0, &x1, &a); ps->get_tile(ps, x0, 0, 1, 1, t0); ps->get_tile(ps, x1, 0, 1, 1, t1); texel[0] = LERP(a, t0[0], t1[0]); texel[1] = LERP(a, t0[1], t1[1]); texel[2] = LERP(a, t0[2], t1[2]); texel[3] = LERP(a, t0[3], t1[3]); rgba[0][j] = texel[0]; rgba[1][j] = texel[1]; rgba[2][j] = texel[2]; rgba[3][j] = texel[3]; } } break; default: assert(0); } } /** * From lambda and the min_mip_filter setting, choose the mipmap level(s) * that are to be sampled from. */ static void choose_mipmap_levels(struct tgsi_sampler *sampler, float lambda, unsigned *level0, unsigned *level1) { switch (sampler->state->min_mip_filter) { case PIPE_TEX_MIPFILTER_NONE: *level0 = *level1 = 0; return; case PIPE_TEX_MIPFILTER_NEAREST: { const int lvl = (int) (lambda + 0.5); *level0 = *level1 = CLAMP(lvl, (int) sampler->texture->first_level, (int) sampler->texture->last_level); } return; case PIPE_TEX_MIPFILTER_LINEAR: { const int lvl = (int) lambda; *level0 = CLAMP(lvl, (int) sampler->texture->first_level, (int) sampler->texture->last_level); *level1 = CLAMP(lvl + 1, (int) sampler->texture->first_level, (int) sampler->texture->last_level); assert(*level0 < 100); assert(*level1 < 100); } return; default: assert(0); } } /** * Load the texture cache with a new texture tile. */ static void cache_tex_tile(struct tgsi_sampler *sampler, unsigned face, unsigned level, unsigned zslice, int cx, int cy) { struct pipe_context *pipe = (struct pipe_context *) sampler->pipe; struct pipe_surface *ps = pipe->get_tex_surface(pipe, sampler->texture, face, level, zslice); assert(ps->width == sampler->texture->level[level].width); assert(ps->height == sampler->texture->level[level].height); sampler->cache_level = level; sampler->cache_x = cx; sampler->cache_y = cy; ps->get_tile(ps, cx * SAMPLER_CACHE_SIZE, cy * SAMPLER_CACHE_SIZE, SAMPLER_CACHE_SIZE, SAMPLER_CACHE_SIZE, (float *) sampler->cache); } /** * Get a texel from a texture. * \param face the cube face in 0..5 * \param level the mipmap level * \param zslize which slice of a 3D texture * \param x the x coord of texel within 2D image * \param y the y coord of texel within 2D image * \param rgba the quad to put the texel/color into * \param j which element of the rgba quad to write to */ static void get_texel(struct tgsi_sampler *sampler, unsigned face, unsigned level, unsigned zslice, int x, int y, float rgba[NUM_CHANNELS][QUAD_SIZE], GLuint j) { int cx = x / SAMPLER_CACHE_SIZE; int cy = y / SAMPLER_CACHE_SIZE; if (cx != sampler->cache_x || cy != sampler->cache_y || level != sampler->cache_level) { cache_tex_tile(sampler, face, level, zslice, cx, cy); /* printf("cache miss (%d, %d)\n", x, y); */ } else { /* printf("cache hit (%d, %d)\n", x, y); */ } /* get texel from cache */ cx = x % SAMPLER_CACHE_SIZE; cy = y % SAMPLER_CACHE_SIZE; rgba[0][j] = sampler->cache[cy][cx][0]; rgba[1][j] = sampler->cache[cy][cx][1]; rgba[2][j] = sampler->cache[cy][cx][2]; rgba[3][j] = sampler->cache[cy][cx][3]; } /** * Called via tgsi_sampler::get_samples() * Use the sampler's state setting to get a filtered RGBA value * from the sampler's texture (mipmap tree). * * XXX we can implement many versions of this function, each * tightly coded for a specific combination of sampler state * (nearest + repeat), (bilinear mipmap + clamp), etc. * * The update_samplers() function in st_atom_sampler.c could create * a new tgsi_sampler object for each state combo it finds.... */ static void sp_get_samples_2d(struct tgsi_sampler *sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], float lodbias, float rgba[NUM_CHANNELS][QUAD_SIZE]) { unsigned level0, level1, j, imgFilter; int width, height; float mipBlend; /* compute level0 and imgFilter */ switch (sampler->state->min_mip_filter) { case PIPE_TEX_MIPFILTER_NONE: imgFilter = sampler->state->mag_img_filter; level0 = level1 = 0; assert(sampler->state->min_img_filter == sampler->state->mag_img_filter); break; default: { float lambda; int fragment = 1; if (fragment) lambda = compute_lambda(sampler, s, t, p, lodbias); else lambda = lodbias; /* not really a bias, but absolute LOD */ if (lambda < 0.0) { /* magnifying */ imgFilter = sampler->state->mag_img_filter; level0 = level1 = 0; } else { /* minifying */ imgFilter = sampler->state->min_img_filter; choose_mipmap_levels(sampler, lambda, &level0, &level1); mipBlend = FRAC(lambda); /* blending weight between levels */ } } } width = sampler->texture->level[level0].width; height = sampler->texture->level[level0].height; assert(width > 0); switch (imgFilter) { case PIPE_TEX_FILTER_NEAREST: for (j = 0; j < QUAD_SIZE; j++) { int x = nearest_texcoord(sampler->state->wrap_s, s[j], width); int y = nearest_texcoord(sampler->state->wrap_t, t[j], height); get_texel(sampler, 0, level0, 0, x, y, rgba, j); if (level0 != level1) { /* get texels from second mipmap level and blend */ float rgba2[4][4]; unsigned c; x = x / 2; y = y / 2; get_texel(sampler, 0, level1, 0, x, y, rgba2, j); for (c = 0; c < NUM_CHANNELS; c++) { rgba[c][j] = LERP(mipBlend, rgba2[c][j], rgba[c][j]); } } } break; case PIPE_TEX_FILTER_LINEAR: for (j = 0; j < QUAD_SIZE; j++) { float tx[4][4], a, b; int x0, y0, x1, y1, c; linear_texcoord(sampler->state->wrap_s, s[j], width, &x0, &x1, &a); linear_texcoord(sampler->state->wrap_t, t[j], height, &y0, &y1, &b); get_texel(sampler, 0, level0, 0, x0, y0, tx, 0); get_texel(sampler, 0, level0, 0, x1, y0, tx, 1); get_texel(sampler, 0, level0, 0, x0, y1, tx, 2); get_texel(sampler, 0, level0, 0, x1, y1, tx, 3); for (c = 0; c < 4; c++) { rgba[c][j] = lerp_2d(a, b, tx[c][0], tx[c][1], tx[c][2], tx[c][3]); } if (level0 != level1) { /* get texels from second mipmap level and blend */ float rgba2[4][4]; unsigned c; x0 = x0 / 2; y0 = y0 / 2; x1 = x1 / 2; y1 = y1 / 2; get_texel(sampler, 0, level1, 0, x0, y0, tx, 0); get_texel(sampler, 0, level1, 0, x1, y0, tx, 1); get_texel(sampler, 0, level1, 0, x0, y1, tx, 2); get_texel(sampler, 0, level1, 0, x1, y1, tx, 3); for (c = 0; c < 4; c++) { rgba2[c][j] = lerp_2d(a, b, tx[c][0], tx[c][1], tx[c][2], tx[c][3]); } for (c = 0; c < NUM_CHANNELS; c++) { rgba[c][j] = LERP(mipBlend, rgba[c][j], rgba2[c][j]); } } } break; default: assert(0); } } static void sp_get_samples_3d(struct tgsi_sampler *sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], float lodbias, float rgba[NUM_CHANNELS][QUAD_SIZE]) { /* get/map pipe_surfaces corresponding to 3D tex slices */ } static void sp_get_samples_cube(struct tgsi_sampler *sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], float lodbias, float rgba[NUM_CHANNELS][QUAD_SIZE]) { GLuint j; for (j = 0; j < QUAD_SIZE; j++) { float st[4]; GLuint face = choose_cube_face(s[j], t[j], p[j], st); (void) face; /* get/map surface corresponding to the face */ } } void sp_get_samples(struct tgsi_sampler *sampler, const float s[QUAD_SIZE], const float t[QUAD_SIZE], const float p[QUAD_SIZE], float lodbias, float rgba[NUM_CHANNELS][QUAD_SIZE]) { switch (sampler->texture->target) { case GL_TEXTURE_1D: sp_get_samples_1d(sampler, s, t, p, lodbias, rgba); break; case GL_TEXTURE_2D: sp_get_samples_2d(sampler, s, t, p, lodbias, rgba); break; case GL_TEXTURE_3D: sp_get_samples_3d(sampler, s, t, p, lodbias, rgba); break; case GL_TEXTURE_CUBE_MAP: sp_get_samples_cube(sampler, s, t, p, lodbias, rgba); break; default: assert(0); } }