From 6fb700bc6f5f6afd1a16f775c8a81f1770f2f1f7 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Pavel=20Poto=C4=8Dek?= Date: Thu, 26 Nov 2020 17:43:43 +0100 Subject: [PATCH] ImDrawList: wip reimplement thick line strip rendering. (merged commits from PR 2964, rebased, minor styling tweaks) --- imgui_draw.cpp | 259 ++++++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 258 insertions(+), 1 deletion(-) diff --git a/imgui_draw.cpp b/imgui_draw.cpp index 57486a6e2..a9763ab6c 100644 --- a/imgui_draw.cpp +++ b/imgui_draw.cpp @@ -716,7 +716,7 @@ void ImDrawList::PrimQuadUV(const ImVec2& a, const ImVec2& b, const ImVec2& c, c #define IM_FIXNORMAL2F_MAX_INVLEN2 100.0f // 500.0f (see #4053, #3366) #define IM_FIXNORMAL2F(VX,VY) { float d2 = VX*VX + VY*VY; if (d2 > 0.000001f) { float inv_len2 = 1.0f / d2; if (inv_len2 > IM_FIXNORMAL2F_MAX_INVLEN2) inv_len2 = IM_FIXNORMAL2F_MAX_INVLEN2; VX *= inv_len2; VY *= inv_len2; } } (void)0 -// TODO: Thickness anti-aliased lines cap are missing their AA fringe. +// FIXME: Thickness anti-aliased lines cap are missing their AA fringe. // We avoid using the ImVec2 math operators here to reduce cost to a minimum for debug/non-inlined builds. void ImDrawList::AddPolyline(const ImVec2* points, const int points_count, ImU32 col, ImDrawFlags flags, float thickness) { @@ -728,6 +728,262 @@ void ImDrawList::AddPolyline(const ImVec2* points, const int points_count, ImU32 const int count = closed ? points_count : points_count - 1; // The number of line segments we need to draw const bool thick_line = (thickness > _FringeScale); +// (Intentionally incorrect indentation to avoid altering) +bool USE_NEW_CODE = true; +//bool USE_NEW_CODE = ImGui::GetIO().KeyShift; +if (USE_NEW_CODE) +{ + const float AA_SIZE = 1.0f; + const ImU32 col_trans = col & ~IM_COL32_A_MASK; + const bool antialias = (Flags & ImDrawListFlags_AntiAliasedLines) != 0; + + if (antialias && !thick_line) + { + // Anti-aliased stroke approximation + const int idx_count = count * 12; + const int vtx_count = count * 6; // FIXME-OPT: Not sharing edges + PrimReserve(idx_count, vtx_count); + ImU32 col_faded = col; + if (thickness < 1.0f) + col_faded = col_trans | ((int)(((col >> IM_COL32_A_SHIFT) & 0xFF) * thickness) << IM_COL32_A_SHIFT); + + for (int i1 = 0; i1 < count; i1++) + { + const int i2 = (i1 + 1) == points_count ? 0 : i1 + 1; + const ImVec2& p1 = points[i1]; + const ImVec2& p2 = points[i2]; + + float dx = p2.x - p1.x; + float dy = p2.y - p1.y; + IM_NORMALIZE2F_OVER_ZERO(dx, dy); + dx *= AA_SIZE; + dy *= AA_SIZE; + + _VtxWritePtr[0].pos.x = p1.x + dy; _VtxWritePtr[0].pos.y = p1.y - dx; _VtxWritePtr[0].uv = opaque_uv; _VtxWritePtr[0].col = col_trans; + _VtxWritePtr[1].pos.x = p1.x; _VtxWritePtr[1].pos.y = p1.y; _VtxWritePtr[1].uv = opaque_uv; _VtxWritePtr[1].col = col_faded; + _VtxWritePtr[2].pos.x = p1.x - dy; _VtxWritePtr[2].pos.y = p1.y + dx; _VtxWritePtr[2].uv = opaque_uv; _VtxWritePtr[2].col = col_trans; + + _VtxWritePtr[3].pos.x = p2.x + dy; _VtxWritePtr[3].pos.y = p2.y - dx; _VtxWritePtr[3].uv = opaque_uv; _VtxWritePtr[3].col = col_trans; + _VtxWritePtr[4].pos.x = p2.x; _VtxWritePtr[4].pos.y = p2.y; _VtxWritePtr[4].uv = opaque_uv; _VtxWritePtr[4].col = col_faded; + _VtxWritePtr[5].pos.x = p2.x - dy; _VtxWritePtr[5].pos.y = p2.y + dx; _VtxWritePtr[5].uv = opaque_uv; _VtxWritePtr[5].col = col_trans; + _VtxWritePtr += 6; + + _IdxWritePtr[0] = (ImDrawIdx)(_VtxCurrentIdx); _IdxWritePtr[1] = (ImDrawIdx)(_VtxCurrentIdx + 1); _IdxWritePtr[2] = (ImDrawIdx)(_VtxCurrentIdx + 4); + _IdxWritePtr[3] = (ImDrawIdx)(_VtxCurrentIdx); _IdxWritePtr[4] = (ImDrawIdx)(_VtxCurrentIdx + 4); _IdxWritePtr[5] = (ImDrawIdx)(_VtxCurrentIdx + 3); + _IdxWritePtr[6] = (ImDrawIdx)(_VtxCurrentIdx + 1); _IdxWritePtr[7] = (ImDrawIdx)(_VtxCurrentIdx + 2); _IdxWritePtr[8] = (ImDrawIdx)(_VtxCurrentIdx + 5); + _IdxWritePtr[9] = (ImDrawIdx)(_VtxCurrentIdx + 1); _IdxWritePtr[10]= (ImDrawIdx)(_VtxCurrentIdx + 5); _IdxWritePtr[11]= (ImDrawIdx)(_VtxCurrentIdx + 4); + _IdxWritePtr += 12; + _VtxCurrentIdx += 6; + } + } + else + { + // Precise line with bevels on acute angles + const int max_n_vtx = antialias ? 6 : 3; + const int max_n_idx = 3 * (antialias ? 9 : 3); + const int vtx_count = points_count * max_n_vtx; + const int idx_count = count * max_n_idx; + PrimReserve(idx_count, vtx_count); + + const float half_thickness = (antialias ? thickness - AA_SIZE : thickness) * 0.5f; + const float half_thickness_aa = half_thickness + AA_SIZE; + const unsigned int first_vtx_idx = _VtxCurrentIdx; + + float sqlen1 = 0.0f; + float dx1, dy1; + if (closed) + { + dx1 = points[0].x - points[points_count - 1].x; + dy1 = points[0].y - points[points_count - 1].y; + sqlen1 = dx1 * dx1 + dy1 * dy1; + IM_NORMALIZE2F_OVER_ZERO(dx1, dy1); + } + + for (int i1 = 0; i1 < points_count; i1++) + { + const int i2 = (i1 + 1 == points_count) ? 0 : i1 + 1; + const ImVec2& p1 = points[i1]; + const ImVec2& p2 = points[i2]; + float dx2 = p1.x - p2.x; + float dy2 = p1.y - p2.y; + float sqlen2 = dx2 * dx2 + dy2 * dy2; + IM_NORMALIZE2F_OVER_ZERO(dx2, dy2); + + if (!closed && i1 == 0) + { + dx1 = -dx2; + dy1 = -dy2; + sqlen1 = sqlen2; + } + else if (!closed && i1 == points_count - 1) + { + dx2 = -dx1; + dy2 = -dy1; + sqlen2 = sqlen1; + } + + float miter_l_recip = dx1 * dy2 - dy1 * dx2; + float mlx, mly, mrx, mry; // Left and right miters + float mlax, mlay, mrax, mray; // Left and right miters including anti-aliasing + const bool bevel = (dx1 * dx2 + dy1 * dy2) > 1e-5f; + if (ImFabs(miter_l_recip) > 1e-5f) + { + float miter_l = half_thickness / miter_l_recip; + // Limit (inner) miter so it doesn't shoot away when miter is longer than adjacent line segments on acute angles + if (bevel) + { + // This is too aggressive (not exactly precise) + float min_sqlen = sqlen1 > sqlen2 ? sqlen2 : sqlen1; + float miter_sqlen = ((dx1 + dx2) * (dx1 + dx2) + (dy1 + dy2) * (dy1 + dy2)) * miter_l * miter_l; + if (miter_sqlen > min_sqlen) + miter_l *= ImSqrt(min_sqlen / miter_sqlen); + } + mlx = p1.x - (dx1 + dx2) * miter_l; + mly = p1.y - (dy1 + dy2) * miter_l; + mrx = p1.x + (dx1 + dx2) * miter_l; + mry = p1.y + (dy1 + dy2) * miter_l; + if (antialias) + { + float miter_al = half_thickness_aa / miter_l_recip; + mlax = p1.x - (dx1 + dx2) * miter_al; + mlay = p1.y - (dy1 + dy2) * miter_al; + mrax = p1.x + (dx1 + dx2) * miter_al; + mray = p1.y + (dy1 + dy2) * miter_al; + } + } + else + { + // Avoid degeneracy for (nearly) straight lines + mlx = p1.x + dy1 * half_thickness; + mly = p1.y - dx1 * half_thickness; + mrx = p1.x - dy1 * half_thickness; + mry = p1.y + dx1 * half_thickness; + if (antialias) + { + mlax = p1.x + dy1 * half_thickness_aa; + mlay = p1.y - dx1 * half_thickness_aa; + mrax = p1.x - dy1 * half_thickness_aa; + mray = p1.y + dx1 * half_thickness_aa; + } + } + // The two bevel vertices if the angle is right or obtuse + // miter_sign == 1, iff the outer (maybe bevelled) edge is on the right, -1 iff it is on the left + int miter_sign = (miter_l_recip >= 0) - (miter_l_recip < 0); + float b1x, b1y, b2x, b2y; // First and second bevel point + float b1ax, b1ay, b2ax, b2ay; // First and second bevel point including anti-aliasing + if (bevel) + { + // FIXME-OPT: benchmark if doing these computations only once in AA case saves cycles + b1x = p1.x + (dx1 - dy1 * miter_sign) * half_thickness; + b1y = p1.y + (dy1 + dx1 * miter_sign) * half_thickness; + b2x = p1.x + (dx2 + dy2 * miter_sign) * half_thickness; + b2y = p1.y + (dy2 - dx2 * miter_sign) * half_thickness; + if (antialias) + { + b1ax = p1.x + (dx1 - dy1 * miter_sign) * half_thickness_aa; + b1ay = p1.y + (dy1 + dx1 * miter_sign) * half_thickness_aa; + b2ax = p1.x + (dx2 + dy2 * miter_sign) * half_thickness_aa; + b2ay = p1.y + (dy2 - dx2 * miter_sign) * half_thickness_aa; + } + } + + // Set the previous line direction so it doesn't need to be recomputed + dx1 = -dx2; + dy1 = -dy2; + sqlen1 = sqlen2; + + // Now that we have all the point coordinates, put them into buffers + + // Vertices for each point are ordered in vertex buffer like this (looking in the direction of the polyline): + // - left vertex* + // - right vertex* + // - left vertex AA fringe* (if antialias) + // - right vertex AA fringe* (if antialias) + // - the remaining vertex (if bevel) + // - the remaining vertex AA fringe (if bevel and antialias) + // (*) if there is bevel, these vertices are the ones on the incoming edge. + // Having all the vertices of the incoming edge in predictable positions is important - we reference them + // even if we don't know relevant line properties yet + + const int vertex_count = antialias ? (bevel ? 6 : 4) : (bevel ? 3 : 2); // FIXME: shorten the expression + const unsigned int bi = antialias ? 4 : 2; // Outgoing edge bevel vertex index + const bool bevel_l = bevel && miter_sign < 0; + const bool bevel_r = bevel && miter_sign > 0; + + _VtxWritePtr[0].pos.x = bevel_l ? b1x : mlx; _VtxWritePtr[0].pos.y = bevel_l ? b1y : mly; _VtxWritePtr[0].uv = opaque_uv; _VtxWritePtr[0].col = col; + _VtxWritePtr[1].pos.x = bevel_r ? b1x : mrx; _VtxWritePtr[1].pos.y = bevel_r ? b1y : mry; _VtxWritePtr[1].uv = opaque_uv; _VtxWritePtr[1].col = col; + if (bevel) + { + _VtxWritePtr[bi].pos.x = b2x; _VtxWritePtr[bi].pos.y = b2y; _VtxWritePtr[bi].uv = opaque_uv; _VtxWritePtr[bi].col = col; + } + + if (antialias) + { + _VtxWritePtr[2].pos.x = bevel_l ? b1ax : mlax; _VtxWritePtr[2].pos.y = bevel_l ? b1ay : mlay; _VtxWritePtr[2].uv = opaque_uv; _VtxWritePtr[2].col = col_trans; + _VtxWritePtr[3].pos.x = bevel_r ? b1ax : mrax; _VtxWritePtr[3].pos.y = bevel_r ? b1ay : mray; _VtxWritePtr[3].uv = opaque_uv; _VtxWritePtr[3].col = col_trans; + if (bevel) + { + _VtxWritePtr[5].pos.x = b2ax; _VtxWritePtr[5].pos.y = b2ay; _VtxWritePtr[5].uv = opaque_uv; _VtxWritePtr[5].col = col_trans; + } + } + _VtxWritePtr += vertex_count; + + if (i1 < count) + { + const int vtx_next_id = i1 < points_count - 1 ? _VtxCurrentIdx + vertex_count : first_vtx_idx; + unsigned int l1i = _VtxCurrentIdx + (bevel_l ? bi : 0); + unsigned int r1i = _VtxCurrentIdx + (bevel_r ? bi : 1); + unsigned int l2i = vtx_next_id; + unsigned int r2i = vtx_next_id + 1; + unsigned int ebi = _VtxCurrentIdx + (bevel_l ? 0 : 1); // incoming edge bevel vertex index + + _IdxWritePtr[0] = (ImDrawIdx)l1i; _IdxWritePtr[1] = (ImDrawIdx)r1i; _IdxWritePtr[2] = (ImDrawIdx)r2i; + _IdxWritePtr[3] = (ImDrawIdx)l1i; _IdxWritePtr[4] = (ImDrawIdx)r2i; _IdxWritePtr[5] = (ImDrawIdx)l2i; + _IdxWritePtr += 6; + + if (bevel) + { + _IdxWritePtr[0] = (ImDrawIdx)l1i; _IdxWritePtr[1] = (ImDrawIdx)r1i; _IdxWritePtr[2] = (ImDrawIdx)ebi; + _IdxWritePtr += 3; + } + + if (antialias) + { + unsigned int l1ai = _VtxCurrentIdx + (bevel_l ? 5 : 2); + unsigned int r1ai = _VtxCurrentIdx + (bevel_r ? 5 : 3); + unsigned int l2ai = vtx_next_id + 2; + unsigned int r2ai = vtx_next_id + 3; + + _IdxWritePtr[0] = (ImDrawIdx)l1ai; _IdxWritePtr[1] = (ImDrawIdx)l1i; _IdxWritePtr[2] = (ImDrawIdx)l2i; + _IdxWritePtr[3] = (ImDrawIdx)l1ai; _IdxWritePtr[4] = (ImDrawIdx)l2i; _IdxWritePtr[5] = (ImDrawIdx)l2ai; + _IdxWritePtr[6] = (ImDrawIdx)r1ai; _IdxWritePtr[7] = (ImDrawIdx)r1i; _IdxWritePtr[8] = (ImDrawIdx)r2i; + _IdxWritePtr[9] = (ImDrawIdx)r1ai; _IdxWritePtr[10] = (ImDrawIdx)r2i; _IdxWritePtr[11] = (ImDrawIdx)r2ai; + _IdxWritePtr += 12; + + if (bevel) + { + _IdxWritePtr[0] = (ImDrawIdx)(_VtxCurrentIdx + (bevel_r ? 1 : 2)); + _IdxWritePtr[1] = (ImDrawIdx)(_VtxCurrentIdx + (bevel_r ? 3 : 0)); + _IdxWritePtr[2] = (ImDrawIdx)(_VtxCurrentIdx + (bevel_r ? 5 : 4)); + + _IdxWritePtr[3] = (ImDrawIdx)(_VtxCurrentIdx + (bevel_r ? 1 : 2)); + _IdxWritePtr[4] = (ImDrawIdx)(_VtxCurrentIdx + (bevel_r ? 5 : 4)); + _IdxWritePtr[5] = (ImDrawIdx)(_VtxCurrentIdx + (bevel_r ? 4 : 5)); + _IdxWritePtr += 6; + } + } + } + _VtxCurrentIdx += vertex_count; + } + + const int unused_indices = (int)(IdxBuffer.Data + IdxBuffer.Size - _IdxWritePtr); + const int unused_vertices = (int)(VtxBuffer.Size - _VtxCurrentIdx - _CmdHeader.VtxOffset); + if (unused_indices > 0 || unused_vertices > 0) + PrimUnreserve(unused_indices, unused_vertices); + } +} +else +{ if (Flags & ImDrawListFlags_AntiAliasedLines) { // Anti-aliased stroke @@ -973,6 +1229,7 @@ void ImDrawList::AddPolyline(const ImVec2* points, const int points_count, ImU32 } } } +} // - We intentionally avoid using ImVec2 and its math operators here to reduce cost to a minimum for debug/non-inlined builds. // - Filled shapes must always use clockwise winding order. The anti-aliasing fringe depends on it. Counter-clockwise shapes will have "inward" anti-aliasing.