Draw the Preview's Toolpath Segments From an Index Buffer

The preview's frame cost is dominated by one call: a single instanced draw of
every visible toolpath segment. On a tall multi-filament print the wipe tower
supplies most of those segments, which is why the preview of a large tower is
slow and why shrinking the layer range speeds it up again.

That draw is not fill bound. Shrinking the model to about a fortieth of its
screen area moved the frame from 419 ms to 401 ms, so the cost is per segment,
not per pixel, and it is paid in the vertex shader: five texelFetch calls plus
several cross/normalize per invocation.

Each segment is a box of eight corners, but it was submitted with
glDrawArraysInstanced over a 24 entry array, so every corner was transformed
once per triangle that touches it and the shader ran 24 times per segment. The
same 24 entries are now an element buffer over the eight distinct corners, which
lets the post-transform cache reuse them and drops the shader to 8 runs per
segment. The triangles, their winding and the vertex_id each corner receives are
unchanged.

Measured over 100 frames on the 636-layer, 351k-vertex three-filament fixture,
the segment draw goes from 381 ms to 322 ms per frame. That is a software
rasterizer, where triangle setup dominates and understates the win; the drop in
shader invocations is the transferable part.

Verified by loading the same project in this build and in a build of the parent
commit and comparing the canvas across three states - the default view, a
rotated camera, and a reduced layer range: pixel identical in all three. The
rotated case matters because the shader picks its corner offsets from the camera
direction. The only pixels that differ anywhere on screen are in the G-code text
panel, which prints a per-process object id that varies between any two runs.

src/libvgcode/src/SegmentTemplate.cpp

@@ -15,7 +15,12 @@ namespace libvgcode {
 //|   2--0-------5--7   |
 //|    \ |       | /    |
 //|      3-------4      | 
-static constexpr const std::array<uint8_t, 24> VERTEX_DATA = {
+// The eight corners the vertex shader knows how to place. Each is sent once and
+// referenced by INDEX_DATA below, so the post-transform cache can reuse it across
+// the triangles that share it: the shader runs 8 times per segment instead of 24.
+static constexpr const std::array<uint8_t, 8> VERTEX_DATA = { 0, 1, 2, 3, 4, 5, 6, 7 };
+
+static constexpr const std::array<uint8_t, 24> INDEX_DATA = {
     0, 1, 2, // front spike
     0, 2, 3, // front spike
     0, 3, 4, // right/bottom body 
@@ -31,7 +36,7 @@ void SegmentTemplate::init()
     if (m_vao_id != 0)
         return;
 
-    m_size_in_bytes_gpu += VERTEX_DATA.size() * sizeof(uint8_t);
+    m_size_in_bytes_gpu += (VERTEX_DATA.size() + INDEX_DATA.size()) * sizeof(uint8_t);
 
     int curr_vertex_array;
     glsafe(glGetIntegerv(GL_VERTEX_ARRAY_BINDING, &curr_vertex_array));
@@ -51,12 +56,22 @@ void SegmentTemplate::init()
     glsafe(glVertexAttribIPointer(0, 1, GL_UNSIGNED_BYTE, 0, (const void*)0));
 #endif // ENABLE_OPENGL_ES
 
+    // The element buffer binding is part of the vao state, so it is left bound here
+    // and restored together with the vao.
+    glsafe(glGenBuffers(1, &m_ibo_id));
+    glsafe(glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, m_ibo_id));
+    glsafe(glBufferData(GL_ELEMENT_ARRAY_BUFFER, INDEX_DATA.size() * sizeof(uint8_t), INDEX_DATA.data(), GL_STATIC_DRAW));
+
     glsafe(glBindBuffer(GL_ARRAY_BUFFER, curr_array_buffer));
     glsafe(glBindVertexArray(curr_vertex_array));
 }
 
 void SegmentTemplate::shutdown()
 {
+    if (m_ibo_id != 0) {
+        glsafe(glDeleteBuffers(1, &m_ibo_id));
+        m_ibo_id = 0;
+    }
     if (m_vbo_id != 0) {
         glsafe(glDeleteBuffers(1, &m_vbo_id));
         m_vbo_id = 0;
@@ -71,14 +86,15 @@ void SegmentTemplate::shutdown()
 
 void SegmentTemplate::render(size_t count)
 {
-    if (m_vao_id == 0 || m_vbo_id == 0 || count == 0)
+    if (m_vao_id == 0 || m_vbo_id == 0 || m_ibo_id == 0 || count == 0)
         return;
 
     int curr_vertex_array;
     glsafe(glGetIntegerv(GL_VERTEX_ARRAY_BINDING, &curr_vertex_array));
 
     glsafe(glBindVertexArray(m_vao_id));
-    glsafe(glDrawArraysInstanced(GL_TRIANGLES, 0, static_cast<GLsizei>(VERTEX_DATA.size()), static_cast<GLsizei>(count)));
+    glsafe(glDrawElementsInstanced(GL_TRIANGLES, static_cast<GLsizei>(INDEX_DATA.size()), GL_UNSIGNED_BYTE,
+                                   nullptr, static_cast<GLsizei>(count)));
     glsafe(glBindVertexArray(curr_vertex_array));
 }
 

src/libvgcode/src/SegmentTemplate.hpp

@@ -40,6 +40,7 @@ private:
     //
     unsigned int m_vao_id{ 0 };
     unsigned int m_vbo_id{ 0 };
+    unsigned int m_ibo_id{ 0 };
     //
     // Size of the data sent to gpu, in bytes.
     //

src/libvgcode/src/ViewerImpl.cpp

@@ -875,6 +875,12 @@ void ViewerImpl::reset()
     m_travels_time = { 0.0f, 0.0f };
     m_vertices.clear();
     m_vertices_colors.clear();
+    // swap rather than clear: these are sized by the print, and a reset means the memory
+    // should go back, not sit reserved until the next load
+    for (std::vector<float>& times : m_layer_start_times)
+        std::vector<float>().swap(times);
+    std::vector<uint32_t>().swap(m_layer_first_vertex);
+    std::vector<float>().swap(m_colors_scratch);
     m_valid_lines_bitset.clear();
 #if VGCODE_ENABLE_COG_AND_TOOL_MARKERS
     m_cog_marker.reset();
@@ -1048,6 +1054,37 @@ void ViewerImpl::load(GCodeInputData&& gcode_data)
         v.layer_duration = m_layers.get_layer_time(m_settings.time_mode, static_cast<size_t>(v.layer_id));
     }
 
+    // Index of the first vertex of each layer, walked back to front so that a layer with no
+    // vertex of its own inherits the next layer's index and the array stays non-decreasing.
+    if (!m_layers.empty()) {
+        const uint32_t vertices_count = static_cast<uint32_t>(m_vertices.size());
+        m_layer_first_vertex.assign(m_layers.count(), vertices_count);
+        for (uint32_t i = vertices_count; i > 0; --i) {
+            const uint32_t layer_id = m_vertices[i - 1].layer_id;
+            if (layer_id < m_layer_first_vertex.size())
+                m_layer_first_vertex[layer_id] = i - 1;
+        }
+        for (size_t i = m_layer_first_vertex.size() - 1; i > 0; --i)
+            m_layer_first_vertex[i - 1] = std::min(m_layer_first_vertex[i - 1], m_layer_first_vertex[i]);
+
+        // the running time at each layer's first vertex, summed in vertex order so that
+        // get_estimated_time_at() matches a full accumulation exactly
+        std::array<float, TIME_MODES_COUNT> running{};
+        for (std::vector<float>& times : m_layer_start_times)
+            times.assign(m_layer_first_vertex.size(), 0.0f);
+        size_t layer = 0;
+        for (size_t i = 0; i <= m_vertices.size(); ++i) {
+            for (; layer < m_layer_first_vertex.size() && m_layer_first_vertex[layer] == i; ++layer) {
+                for (size_t j = 0; j < TIME_MODES_COUNT; ++j)
+                    m_layer_start_times[j][layer] = running[j];
+            }
+            if (i < m_vertices.size()) {
+                for (size_t j = 0; j < TIME_MODES_COUNT; ++j)
+                    running[j] += m_vertices[i].times[j];
+            }
+        }
+    }
+
     if (!m_layers.empty())
         m_layers.set_view_range(0, static_cast<uint32_t>(m_layers.count()) - 1);
 
@@ -1261,7 +1298,10 @@ void ViewerImpl::update_colors_texture()
 
     // Based on current settings and slider position, we might want to render some
     // vertices as dark grey (or darkened, see above). Use either that or the normal color (from the cache).
-    std::vector<float> colors(m_vertices_colors.size());
+    // Reused across calls: this runs on every slider tick, and the allocation alone is
+    // 4 bytes per vertex of the whole print each time.
+    std::vector<float>& colors = m_colors_scratch;
+    colors.resize(m_vertices_colors.size());
     assert(colors.size() == m_vertices.size() && m_vertices_colors.size() == m_vertices.size());
     for (size_t i=0; i<m_vertices.size(); ++i) {
         const PathVertex& v = m_vertices[i];
@@ -1516,8 +1556,19 @@ void ViewerImpl::set_view_visible_range(Interval::value_type min, Interval::valu
 
 float ViewerImpl::get_estimated_time_at(size_t id) const
 {
-    return std::accumulate(m_vertices.begin(), m_vertices.begin() + id + 1, 0.0f, 
-        [this](float a, const PathVertex& v) { return a + v.times[static_cast<size_t>(m_settings.time_mode)]; });
+    const size_t mode = static_cast<size_t>(m_settings.time_mode);
+    if (mode >= TIME_MODES_COUNT || id >= m_vertices.size())
+        return 0.0f;
+    size_t first = 0;
+    float time = 0.0f;
+    const size_t layer = static_cast<size_t>(m_vertices[id].layer_id);
+    if (layer < m_layer_first_vertex.size() && m_layer_first_vertex[layer] <= id) {
+        first = m_layer_first_vertex[layer];
+        time = m_layer_start_times[mode][layer];
+    }
+    for (size_t i = first; i <= id; ++i)
+        time += m_vertices[i].times[mode];
+    return time;
 }
 
 Color ViewerImpl::get_vertex_color(const PathVertex& v) const
@@ -1722,6 +1773,10 @@ size_t ViewerImpl::get_used_cpu_memory() const
     ret += sizeof(m_extrusion_roles_colors);
     ret += sizeof(m_options_colors);
     ret += STDVEC_MEMSIZE(m_vertices, PathVertex);
+    for (const std::vector<float>& times : m_layer_start_times)
+        ret += STDVEC_MEMSIZE(times, float);
+    ret += STDVEC_MEMSIZE(m_layer_first_vertex, uint32_t);
+    ret += STDVEC_MEMSIZE(m_colors_scratch, float);
     ret += m_valid_lines_bitset.size_in_bytes_cpu();
     ret += m_height_range.size_in_bytes_cpu();
     ret += m_width_range.size_in_bytes_cpu();
@@ -1787,7 +1842,11 @@ void ViewerImpl::update_view_full_range()
     const bool travels_visible = m_settings.options_visibility[size_t(EOptionType::Travels)];
     const bool wipes_visible   = m_settings.options_visibility[size_t(EOptionType::Wipes)];
 
+    // every vertex before m_layer_first_vertex[layers_range[0]] has a smaller layer_id, so the loop
+    // below would skip all of them anyway
     auto first_it = m_vertices.begin();
+    if (layers_range[0] < m_layer_first_vertex.size())
+        first_it += m_layer_first_vertex[layers_range[0]];
     while (first_it != m_vertices.end() &&
            (first_it->layer_id < layers_range[0] || !is_visible(*first_it, m_settings))) {
         ++first_it;

src/libvgcode/src/ViewerImpl.hpp

@@ -234,6 +234,20 @@ private:
     //
     std::array<float, TIME_MODES_COUNT> m_total_time{ 0.0f, 0.0f };
     //
+    // Running sum of the vertex estimated times at each layer's first vertex, for each time mode,
+    // so that get_estimated_time_at() only accumulates the vertices of one layer.
+    //
+    std::array<std::vector<float>, TIME_MODES_COUNT> m_layer_start_times;
+    //
+    // For each layer L, the index of the first vertex whose layer_id is >= L (m_vertices.size()
+    // if there is none). Derived from the vertices, so it stays exact whatever order they arrive in.
+    //
+    std::vector<uint32_t> m_layer_first_vertex;
+    //
+    // Scratch buffer for update_colors_texture(), kept alive across slider steps
+    //
+    std::vector<float> m_colors_scratch;
+    //
     // Detected travel moves times
     //
     std::array<float, TIME_MODES_COUNT> m_travels_time{ 0.0f, 0.0f };

src/slic3r/GUI/GCodeViewer.cpp

@@ -1627,7 +1627,7 @@ void GCodeViewer::render_scene(int canvas_width, int canvas_height)
     glsafe(::glEnable(GL_DEPTH_TEST));
     render_shells(canvas_width, canvas_height);
 
-    if (m_viewer.get_extrusion_roles().empty())
+    if (m_viewer.get_extrusion_roles_count() == 0)
         return;
 
     render_toolpaths();
@@ -3426,6 +3426,12 @@ void GCodeViewer::render_legend(float &legend_height, int canvas_width, int canv
         std::vector<std::pair<ColorRGBA, std::pair<double, double>>> ret;
         ret.reserve(custom_gcode_per_print_z.size());
 
+        // Loop invariant, but built lazily: this lambda runs once per extruder on every frame
+        // and most prints reach neither colour change below, so fetching it up front would cost
+        // more than the per-item fetch it replaces.
+        std::vector<float> zs;
+        bool zs_built = false;
+
         for (const auto& item : custom_gcode_per_print_z) {
             if (extruder_id + 1 != static_cast<unsigned char>(item.extruder))
                 continue;
@@ -3433,7 +3439,10 @@ void GCodeViewer::render_legend(float &legend_height, int canvas_width, int canv
             if (item.type != ColorChange)
                 continue;
 
-            const std::vector<float> zs = m_viewer.get_layers_zs();
+            if (!zs_built) {
+                zs = m_viewer.get_layers_zs();
+                zs_built = true;
+            }
             auto lower_b = std::lower_bound(zs.begin(), zs.end(),
                 static_cast<float>(item.print_z - epsilon()));
             if (lower_b == zs.end())
@@ -4582,6 +4591,8 @@ void GCodeViewer::render_legend(float &legend_height, int canvas_width, int canv
 
         // ORCA: Get layer Zs as doubles
         std::vector<double> layer_zs = get_layers_zs();
+        // loop invariant, same reason as the layer Zs above
+        const std::vector<float> layer_times = m_viewer.get_layers_estimated_times();
 
         for (Slic3r::CustomGCode::Item custom_gcode : custom_gcode_per_print_z) {
             ImGui::Dummy({window_padding, window_padding});
@@ -4601,7 +4612,6 @@ void GCodeViewer::render_legend(float &legend_height, int canvas_width, int canv
             imgui.text(buf);
             ImGui::SameLine(max_len * 1.5);
 
-            std::vector<float> layer_times = m_viewer.get_layers_estimated_times();
             float custom_gcode_time = 0;
             if (layer > 0)
             {
@@ -4650,7 +4660,7 @@ void GCodeViewer::render_legend(float &legend_height, int canvas_width, int canv
     std::string print_str = _u8L("Model printing time");
     std::string total_str = _u8L("Total time");
     float max_len = window_padding + 2 * ImGui::GetStyle().ItemSpacing.x;
-    if (m_viewer.get_layers_estimated_times().empty())
+    if (m_viewer.get_layers_count() == 0)
         max_len += ImGui::CalcTextSize(total_str.c_str()).x;
     else {
         if (m_viewer.get_view_type() == libvgcode::EViewType::FeatureType)