#ifndef PATTERNS_MATH_H #define PATTERNS_MATH_H r32 fractf(r32 v) { r64 int_part = 0; return (r32)modf((r64)v, &int_part); } r32 pm_smoothstep_r32(r32 t) { r32 r = (t * t * (3 - (2 * t))); return r; } r32 pm_smoothstep_range_r32(r32 t, r32 min, r32 max) { r32 tt = pm_smoothstep_r32(t); r32 r = lerp(min, tt, max); return r; } v3 pm_smoothstep_v3(v3 p) { v3 result = { .x = pm_smoothstep_r32(p.x), .y = pm_smoothstep_r32(p.y), .z = pm_smoothstep_r32(p.z), }; return result; } r32 pm_easeinout_cubic_r32(r32 v) { assert(v >= 0 && v <= 1); // Equation: // [0,0.5) = 4 * v^3 // [0.5,1] = 1 - (-2 * v + 2)^3 / 2 if (v < 0.5f) { return 4 * v * v * v; } else { r32 a = -2 * v + 2; r32 a3 = a * a * a; return 1 - a3 / 2; } } ///// vector extensions v2 pm_lerp_v2(v2 a, r32 t, v2 b) { return (v2){ .x = lerp(a.x, t, b.x), .y = lerp(a.y, t, b.y), }; } v3 pm_lerp_v3(v3 a, r32 t, v3 b) { return (v3){ .x = lerp(a.x, t, b.x), .y = lerp(a.y, t, b.y), .z = lerp(a.z, t, b.z), }; } v4 pm_lerp_v4(v4 a, r32 t, v4 b) { return (v4){ .x = lerp(a.x, t, b.x), .y = lerp(a.y, t, b.y), .z = lerp(a.z, t, b.z), .w = lerp(a.w, t, b.w), }; } v2 pm_abs_v2(v2 v) { return (v2){ .x = fabsf(v.x), .y = fabsf(v.y) }; } v3 pm_abs_v3(v3 v) { return (v3){ .x = fabsf(v.x), .y = fabsf(v.y), .z = fabsf(v.z) }; } v4 pm_abs_v4(v4 v) { return (v4){ .x = fabsf(v.x), .y = fabsf(v.y), .z = fabsf(v.z), .w = fabsf(v.w) }; } v2 pm_floor_v2(v2 v) { return (v2){ .x = floorf(v.x), .y = floorf(v.y) }; } v3 pm_floor_v3(v3 v) { return (v3){ .x = floorf(v.x), .y = floorf(v.y), .z = floorf(v.z) }; } v4 pm_floor_v4(v4 v) { return (v4){ .x = floorf(v.x), .y = floorf(v.y), .z = floorf(v.z), .w = floorf(v.w) }; } v2 pm_fract_v2(v2 v) { return (v2){ .x = fractf(v.x), .y = fractf(v.y) }; } v3 pm_fract_v3(v3 v) { return (v3){ .x = fractf(v.x), .y = fractf(v.y), .z = fractf(v.z) }; } v4 pm_fract_v4(v4 v) { return (v4){ .x = fractf(v.x), .y = fractf(v.y), .z = fractf(v.z), .w = fractf(v.w) }; } r32 pm_sinf_01(r32 v) { return 0.5f + (0.5f * sinf(v)); } r32 pm_cosf_01(r32 v) { return 0.5f + (0.5f * cosf(v)); } v2 pm_sin_v2(v2 v) { return (v2){ .x = sinf(v.x), .y = sinf(v.y) }; } v3 pm_sin_v3(v3 v) { return (v3){ .x = sinf(v.x), .y = sinf(v.y), .z = sinf(v.z) }; } v4 pm_sin_v4(v4 v) { return (v4){ .x = sinf(v.x), .y = sinf(v.y), .z = sinf(v.z), .w = sinf(v.w) }; } v2 pm_cos_v2(v2 v) { return (v2){ .x = cosf(v.x), .y = cosf(v.y) }; } v3 pm_cos_v3(v3 v) { return (v3){ .x = cosf(v.x), .y = cosf(v.y), .z = cosf(v.z) }; } v4 pm_cos_v4(v4 v) { return (v4){ .x = cosf(v.x), .y = cosf(v.y), .z = cosf(v.z), .w = cosf(v.w) }; } ////// hash functions r32 pm_hash_v2_to_r32(v2 p) { v2 r = HMM_MultiplyVec2f(p, 0.3183099f); r = pm_fract_v2(r); r = HMM_MultiplyVec2f(r, 50); r32 result = fractf(r.x * r.y * (r.x + r.y)); return result; } r32 pm_hash_r32_to_r32(r32 n) { return fractf(n * 17 * fractf(n * 0.3183099f)); } v2 pm_hash_r32_to_v2(r32 n) { v2 a = pm_sin_v2((v2){ n, n + 1.0f }); v2 b = HMM_MultiplyVec2(a, (v2){ 43758.5453123f, 22578.1459123f }); v2 r = pm_fract_v2(b); return r; } v2 pm_hash_v2_to_v2(v2 p) { v2 k = (v2){ 0.3183099f, 0.3678794f }; v2 kp = (v2){k.y, k.x}; v2 r0 = HMM_MultiplyVec2(p, k); v2 r1 = HMM_AddVec2(r0, kp); r32 f = 16.0f * fractf(p.x * p.y * (p.x + p.y)); v2 r2 = HMM_MultiplyVec2f(k, f); v2 r3 = pm_fract_v2(r2); return r3; } v3 pm_hash_v2_to_v3(v2 p) { v3 q = (v3){ .x = HMM_DotVec2(p, (v2){127.1f, 311.7f}), .y = HMM_DotVec2(p, (v2){267.5f, 183.3f}), .z = HMM_DotVec2(p, (v2){419.2f, 371.9f}) }; v3 r0 = pm_sin_v3(q); v3 r1 = HMM_MultiplyVec3f(r0, 43758.5453f); v3 r2 = pm_fract_v3(r1); return r2; } r32 pm_hash_v3_to_r32(v3 p) { v3 p0 = HMM_MultiplyVec3f(p, 0.3183099f); v3 p1 = HMM_AddVec3(p0, (v3){ 0.1f, 0.1f, 0.1f }); v3 p2 = pm_fract_v3(p1); v3 p3 = HMM_MultiplyVec3f(p2, 17.0f); r32 r0 = fractf(p3.x * p3.y * p3.z * (p3.x + p3.y + p3.z)); return r0; } r32 pm_random_v2_to_r32(v2 n) { v2 v = (v2){ 12.9898f, 4.1414f }; r32 r0 = HMM_DotVec2(n, v); r32 r1 = sinf(r0); r32 r2 = fractf(r1 * 43758.5453); return r2; } internal r32 pm_noise_v3_to_r32(v3 p) { p = pm_abs_v3(p); v3 p_fl = pm_floor_v3(p); v3 p_fr = pm_fract_v3(p); v3 f = pm_smoothstep_v3(p_fr); v3 p_fl_0 = p_fl; v3 p_fl_1 = HMM_AddVec3(p_fl, (v3){1, 0, 0}); v3 p_fl_2 = HMM_AddVec3(p_fl, (v3){0, 1, 0}); v3 p_fl_3 = HMM_AddVec3(p_fl, (v3){1, 1, 0}); v3 p_fl_4 = HMM_AddVec3(p_fl, (v3){0, 0, 1}); v3 p_fl_5 = HMM_AddVec3(p_fl, (v3){1, 0, 1}); v3 p_fl_6 = HMM_AddVec3(p_fl, (v3){0, 1, 1}); v3 p_fl_7 = HMM_AddVec3(p_fl, (v3){1, 1, 1}); r32 h0 = pm_hash_v3_to_r32(p_fl_0); r32 h1 = pm_hash_v3_to_r32(p_fl_1); r32 h2 = pm_hash_v3_to_r32(p_fl_2); r32 h3 = pm_hash_v3_to_r32(p_fl_3); r32 h4 = pm_hash_v3_to_r32(p_fl_4); r32 h5 = pm_hash_v3_to_r32(p_fl_5); r32 h6 = pm_hash_v3_to_r32(p_fl_6); r32 h7 = pm_hash_v3_to_r32(p_fl_7); r32 h0_1 = lerp(h0, f.x, h1); r32 h2_3 = lerp(h2, f.x, h3); r32 h4_5 = lerp(h4, f.x, h5); r32 h6_7 = lerp(h6, f.x, h7); r32 h01_23 = lerp(h0_1, f.y, h2_3); r32 h45_67 = lerp(h4_5, f.y, h6_7); // r32 result = lerp( // lerp( // lerp(h0, f.x, h1), // f.y, // lerp(h2, f.x, h3) // ), // f.z, // lerp( // lerp(h4, f.x, h5), // f.y, // lerp(h6, f.x, h7) // ) // ); r32 result = lerp(h01_23, f.z, h45_67); assert(result >= 0 && result <= 1); return result; } r32 pm_fmb_3d(v3 x, r32 h) { // r32 G = powf(2, -h); // r32 f = 1.0f; // r32 a = 1.0f; // r32 t = 0.0f; // for(s32 i = 0; i < 4; i++) // { // v3 xx = HMM_MultiplyVec3f(x, f); // r32 n = pm_noise_v3_to_r32(xx); // t += a * n; // f *= 2.0f; // a *= G; // } // return (t - .17f) / 1.2f; // float t = 0.0; // for(s32 i = 0; i < 4; i++) // { // r32 f = powf(2.0, (r32)(i)); // r32 a = powf(f, -h); // r32 n = pm_noise_v3_to_r32( // HMM_MultiplyVec3f(x, f) // ); // r32 ns = a * n; // t += ns; // } // return t; v3 ts = (v3){h, h, h}; v3 pp = x; r32 f = 0.0; v3 pp0 = HMM_AddVec3(pp, ts); v3 pp1 = HMM_SubtractVec3(pp, ts); f += 0.500000f * pm_noise_v3_to_r32(pp0); pp = HMM_MultiplyVec3f(pp, 2.02); f += 0.300000f * pm_noise_v3_to_r32(pp1); pp = HMM_MultiplyVec3f(pp, 2.03); f += 0.125000f * pm_noise_v3_to_r32(pp); pp = HMM_MultiplyVec3f(pp, 2.01); f += 0.062500f * pm_noise_v3_to_r32(pp0); pp = HMM_MultiplyVec3f(pp, 2.04); r32 d = 0.9875f; f = f / d; return f; } // internal r32 // pm_voronoise(v2 p, r32 u, r32 v) // { // r32 k = 1.0f + 63.0f + powf(1.0f - v, 6.0f); // v2 i = pm_floor_v2(p); // v2 f = pm_fract_v2(p); // v2 a = (v2){0, 0}; // for (s32 y = -2; y <= 2; y++) // { // for (s32 x = -2; x <= 2; x++) // { // v2 g = (v2){(r32)x, (r32)y}; // v2 hi = HMM_AddVec2(g, i); // v3 h = pm_hash_v2_to_v3(hi); // v3 o = HMM_MultiplyVec3(h, (v3){ u, u, 1.0f }); // v2 d0 = HMM_SubtractVec2(g, f); // v2 d1 = HMM_AddVec2(d0, o.XY); // r32 d1m = HMM_LengthVec2(d1); // r32 w = powf(1.0f - pm_smoothstep_range_r32(d1m, 0.0f, 1.414f), k); // a = HMM_AddVec2(a, (v2){o.z * w, w}); // } // } // return a.x / a.y; // } // Color ramps typedef struct { r32 pct; v3 color; } Color_Ramp_Anchor; typedef struct { Color_Ramp_Anchor anchors[8]; u32 anchors_count; } Color_Ramp; Color_Ramp color_ramp_reverse(Color_Ramp ramp) { Color_Ramp result = { .anchors_count = ramp.anchors_count }; for (u32 i = 0; i < ramp.anchors_count; i++) { u32 new_i = (ramp.anchors_count - 1) - i; assert(new_i < ramp.anchors_count); r32 rev_pct = 1.f - ramp.anchors[i].pct; result.anchors[new_i] = (Color_Ramp_Anchor){ .color = ramp.anchors[i].color, .pct = rev_pct, }; } return result; } v3 color_ramp_eval(Color_Ramp ramp, r32 pct) { // find nearest two anchors // TODO: do a binary search and we just have to assume that the anchors // are in order from least to greatest Color_Ramp_Anchor nearest_below = { .pct = 0, .color = BLACK_V4.xyz }; Color_Ramp_Anchor nearest_above = { .pct = 1, .color = BLACK_V4.xyz }; r32 dist_below = 1; r32 dist_above = -1; for (u32 i = 0; i < ramp.anchors_count; i++) { Color_Ramp_Anchor anchor = ramp.anchors[i]; r32 dist = pct - anchor.pct; if (dist >= 0 && dist_below > dist) { nearest_below = anchor; dist_below = dist; } if (dist <= 0 && dist_above < dist) { nearest_above = anchor; dist_above = dist; } } // interpolate between them r32 anchor_range = nearest_above.pct - nearest_below.pct; if (anchor_range == 0) anchor_range = 1; r32 pct_remapped = (pct - nearest_below.pct) / anchor_range; v3 result = pm_lerp_v3(nearest_below.color, pct_remapped, nearest_above.color); return result; } // Common SDFs // all assume the shape is centered at 0, 0, 0 internal r32 sdf_sphere2_d(r32 radius_squared, r32 dist_squared) { r32 d = radius_squared - dist_squared; r32 sdf = (d / radius_squared); return max(0, sdf); } internal r32 sdf_sphere2(r32 radius_squared, v3 p) { r32 d2 = HMM_LengthSquaredVec3(p); return sdf_sphere2_d(radius_squared, d2); } internal r32 sdf_sphere(r32 radius, v3 p) { return sdf_sphere2(radius * radius, p); } internal r32 sdf_sphere_hull2_d(r32 radius_squared, r32 falloff, r32 dist_squared) { r32 d = fabsf(falloff * (radius_squared - dist_squared)); r32 sdf = 1.0f - (d / radius_squared); r32 result = max(0, sdf); return result; } internal r32 sdf_sphere_hull2(r32 radius_squared, r32 falloff, v3 p) { r32 d2 = HMM_LengthSquaredVec3(p); return sdf_sphere_hull2_d(radius_squared, falloff, d2); } internal r32 sdf_sphere_hull(r32 radius, r32 falloff, v3 p) { return sdf_sphere_hull2(radius * radius, falloff, p); } #endif // PATTERNS_MATH_H