Files
Mujoco_WASM/src/engine/engine_island.c
T
Alessio Quaglino ea230a950c Implicit flex elasticity in the CG constraint solver via an effective metric
This CL replaces the post-hoc implicit flex correction (`flexInterp_cgsolve`) with a **linearly-implicit effective metric** `M̃ = M + (h² + h·damping)·K` carried by the CG constraint solver itself. Contact/friction forces and implicit flex elasticity are now computed against one consistent metric, instead of the solver seeing `M` and a post-solve correction changing `qacc` behind its back.

Gate (unchanged semantics): `solver="CG"` + implicit/implicitfast integrator + pyramidal cones + flex stiffness present. Newton and PGS are untouched. `solver="CG"` remains the user-facing contract — the factorization is an implementation detail of the preconditioner.

### What's in the metric

- **mjData `efm_*`** (arena, efc-like lifetime/skip semantics; built in `mj_fwdPosition`, value-refreshed in `mj_fwdVelocity`): the per-step stiffness CSR `efm_B_*`, its reverse-Cholesky factor `efm_dofid` + `efm_L_*` (nested-dissection ordered, separators-first for the reverse factorization), and the smooth-force shift `efm_c = h·K·qvel`.
- **`mjd_flexStiff_assemble`** now assembles stretch (Gauss–Newton), standard dim-2 bending, and — via the cached corotated stiffness `d->flexelem_krot` — interp stiffness (all node bodies on simple sliders: point Jacobian is I₃, `flex_centered` not required; fixed nodes drop like pins) into one dof-level CSR. `mjd_effMulAdd`/`mjd_effSolve` apply the metric, with matrix-free operator fallbacks where assembly does not apply.
- **mjModel `efm0_*`** (`nefm0dof`/`nefm0L`): the constant part of the metric factor — currently the dim-2 bending factor, computed once in `mj_setConst` — so bending-only models pay zero per-step factorization cost. Naming mirrors mjData's `efm_*` with the standard `0`-suffix (reference/constant) idiom, and is deliberately not bending-specific: future constant contributors extend it without renames.
- The solver consumes the metric through pre-shifted `qfrc_smooth` and the metric products `Ma`/`Mv`/`Mgrad`; `qacc_smooth` becomes the unconstrained minimizer of the implicit dynamics, which makes the no-constraint shortcut and the warmstart choice consistent by construction.
- **`mj_inverse` adds `B·qacc − c`**, making inverse dynamics discrete-consistent with the gated forward dynamics — exact, since the gated path has no qDeriv term (new test `ForwardTest.GatedFlexInverseConsistency`).

### Performance

All numbers: ms/step over the same 2000-step window, models as shipped on each side (old code with the old model settings vs this CL with the new ones).

The new solver path activates on exactly two shipped models — the ponchos, the only flex models that need an implicit integrator (poncho on Euler degenerates to >200 ms/step). For them, this CL trades speed for consistency: the implicit bending solve now runs inside every solver iteration, where the contact solve can see the stiffness, instead of once after the solve. Solver iterations drop because the curvature is visible, but each iteration pays for the implicit solve:

| model | before | after | solver iters/step |
|---|---|---|---|
| poncho | 2.47 | 3.30 (1.33×) | 16.8 → 11.8 |
| poncho_edgeequality | 1.96 | 2.72 (1.39×) | 13.2 → 10.0 |

What that price buys: contact forces consistent with the implicit elasticity (previously the post-hoc correction changed `qacc` after the constraint solve), discrete-consistent inverse dynamics, and the removal of the post-hoc special case from the integration path. Raising poncho's timestep from 2 to 5 ms leaves its per-step cost nearly flat, so the consistency price can be recovered by taking fewer steps where accuracy allows.

Every other flex model was measured stable on Euler at its shipped timestep and switches to it (these models predate the post-hoc integrator; implicit was never load-bearing for them). They end up equal or faster than before: bunny_multicell 0.47 → 0.40, trampoline 0.28 → 0.25, plate 1.02 → 0.99, pancake 0.34 → 0.33.

Finally, the per-step factorization makes configurations practical that the old code could only integrate explicitly: implicit stretch elasticity (`elastic2d="stretch"`/`"both"`, dim-3 solids) and factorized interp stiffness. No before/after exists for these — stock has no implicit treatment of stretch at all.

### Behavior changes

- With the post-hoc correction deleted, interp/bending models running `solver="Newton"` (or elliptic cones, or islands) now integrate flex elasticity **explicitly** (previously: post-hoc implicit). Affects e.g. `gripper_trilinear` (stable, and faster, but different semantics). Follow-up options: Newton-side metric support, or a documented fallback.
- With the gate on, `mj_forward` outputs are timestep-dependent for gated models (they answer the linearly-implicit discrete problem); `qacc_smooth` and `mj_inverse` change accordingly. Non-gated models are bit-identical (full suite green throughout).

### Validation

- 1737/1737 tests, including new: `FlexStretchDerivatives` (FD-validated GN operator), `FlexStiffAssemble`/`FlexStiffAssembleInterp` (CSR ≡ operators), `GatedFlexInverseConsistency` (fails pre-change), equivalence tests vs the old post-hoc treatment (bending matches to 2e-11).
- Fingerprint discipline throughout: bending-only models bit-exact across every refactor; permutation/kernel changes verified iteration-identical.

### Known follow-ups (not in this CL)

3×3-block sparse Cholesky kernel (the numeric factorization is index-bound; projected ~3× on the factor); mjModel persistence of the factor's symbolic pattern (rest-pose ND makes sizes compile-time); the general effective-metric mode (all solvers, all PSD-safe force classes, behind an enable flag).

PiperOrigin-RevId: 948561856
Change-Id: I8b8e32ebd0428042af71647d0470d10773bf6daf
2026-07-15 14:57:42 -07:00

622 lines
18 KiB
C

// Copyright 2023 DeepMind Technologies Limited
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "engine/engine_island.h"
#include <stdio.h>
#include <stddef.h>
#include <string.h>
#include <mujoco/mjdata.h>
#include <mujoco/mjmodel.h>
#include <mujoco/mjsan.h> // IWYU pragma: keep
#include <mujoco/mjxmacro.h>
#include "engine/engine_core_util.h"
#include "engine/engine_memory.h"
#include "engine/engine_util_errmem.h"
#include "engine/engine_util_misc.h"
#include "engine/engine_util_sparse.h"
#ifdef MEMORY_SANITIZER
#include <sanitizer/msan_interface.h>
#endif
//-------------------------- local utilities -------------------------------------------------------
// clear island-related arena pointers in mjData
static void clearIsland(mjData* d, size_t parena) {
#define X(type, name, nr, nc) d->name = NULL;
MJDATA_ARENA_POINTERS_ISLAND
#undef X
d->nefc = 0;
d->nisland = 0;
d->nidof = 0;
d->parena = parena;
// poison remaining memory
#ifdef ADDRESS_SANITIZER
ASAN_POISON_MEMORY_REGION(
(char*)d->arena + d->parena, d->narena - d->pstack - d->parena);
#endif
}
// allocate island arrays on arena, return 1 on success, 0 on failure
static int arenaAllocIsland(const mjModel* m, mjData* d) {
#undef MJ_M
#define MJ_M(n) m->n
#undef MJ_D
#define MJ_D(n) d->n
size_t parena_old = d->parena;
#define X(type, name, nr, nc) \
d->name = mj_arenaAllocByte(d, sizeof(type) * (nr) * (nc), _Alignof(type)); \
if (!d->name) { \
mj_warning(d, mjWARN_CNSTRFULL, d->narena); \
clearIsland(d, parena_old); \
return 0; \
}
MJDATA_ARENA_POINTERS_ISLAND
#undef X
#undef MJ_M
#define MJ_M(n) n
#undef MJ_D
#define MJ_D(n) n
return 1;
}
//-------------------------- flood-fill and graph construction ------------------------------------
// find disjoint subgraphs ("islands") given sparse symmetric adjacency matrix
// arguments:
// island (nr) - island index assigned to vertex, -1 if vertex has no edges
// nr - number of rows/columns of adjacency matrix
// rownnz (nr) - matrix row nonzeros
// rowadr (nr) - matrix row addresses
// colind (nnz) - matrix column indices
// stack (nnz) - stack space
// returns number of islands
// note: column indices are not required to be unique or sorted
int mj_floodFill(int* island, int nr, const int* rownnz, const int* rowadr, const int* colind,
int* stack) {
// initialize island count, set ids to -1
int nisland = 0;
mju_fillInt(island, -1, nr);
// iterate over vertices, discover islands
for (int i=0; i < nr; i++) {
// vertex already in island or singleton with no edges: skip
if (island[i] != -1 || !rownnz[i]) {
continue;
}
// push i onto stack
int nstack = 0;
stack[nstack++] = i;
// DFS traversal of island
while (nstack) {
// pop v from stack
int v = stack[--nstack];
// if v is already assigned, continue
if (island[v] != -1) {
continue;
}
// assign v to current island
island[v] = nisland;
// push adjacent vertices onto stack
mju_copyInt(stack + nstack, colind + rowadr[v], rownnz[v]);
nstack += rownnz[v];
}
// island is filled: increment nisland
nisland++;
}
return nisland;
}
// state of iterator for finding trees involved in a constraint
typedef struct {
int trees[2]; // pre-calculated trees (special-cased constraints); -2: empty/sentinel
int jac_idx; // generic scan: current lookup index in Jacobian row; -1: scan disabled
int tree_prev; // generic scan: previous tree in ongoing scan
} mjTreeIter;
// return next tree of constraint i from iterator; -2: no more trees
static int treeNext(const mjModel* m, const mjData* d, int i, mjTreeIter* iter) {
// handle special cases
if (iter->trees[0] != -2) {
// get first tree, queue up second tree, return first tree
int tree = iter->trees[0];
iter->trees[0] = iter->trees[1];
iter->trees[1] = -2;
return tree;
}
// special case mode complete
if (iter->jac_idx == -1) {
return -2;
}
// generic scan mode
int j;
int tree_next = -2;
// sparse
if (mj_isSparse(m)) {
int rownnz = d->efc_J_rownnz[i];
const int* colind = d->efc_J_colind + d->efc_J_rowadr[i];
for (j = iter->jac_idx; j < rownnz; j++) {
int tree_j = m->dof_treeid[colind[j]];
if (tree_j != iter->tree_prev) {
// found new tree
tree_next = tree_j;
break;
}
}
}
// dense
else {
int nv = m->nv;
const mjtNum* J = d->efc_J + nv * i;
for (j = iter->jac_idx; j < nv; j++) {
if (J[j]) {
int tree_j = m->dof_treeid[j];
if (tree_j != iter->tree_prev) {
// found new tree
tree_next = tree_j;
break;
}
// skip to end of tree's dof block
j = m->tree_dofadr[tree_j] + m->tree_dofnum[tree_j] - 1;
}
}
}
// update iterator state
iter->jac_idx = j;
if (tree_next != -2) {
iter->tree_prev = tree_next;
}
return tree_next;
}
// initialize tree iterator, handle special cases
static void treeIterInit(const mjModel* m, const mjData* d, int i, mjTreeIter* iter) {
iter->trees[0] = -2;
iter->trees[1] = -2;
iter->jac_idx = -1;
iter->tree_prev = -1;
int efc_type = d->efc_type[i];
int efc_id = d->efc_id[i];
// ==== special cases: fill iter->trees where possible
// joint friction
if (efc_type == mjCNSTR_FRICTION_DOF) {
iter->trees[0] = m->dof_treeid[efc_id];
}
// joint limit
else if (efc_type == mjCNSTR_LIMIT_JOINT) {
iter->trees[0] = m->dof_treeid[m->jnt_dofadr[efc_id]];
}
// contact
else if (efc_type == mjCNSTR_CONTACT_FRICTIONLESS ||
efc_type == mjCNSTR_CONTACT_PYRAMIDAL ||
efc_type == mjCNSTR_CONTACT_ELLIPTIC) {
int g1 = d->contact[efc_id].geom[0];
int g2 = d->contact[efc_id].geom[1];
// geom-geom contact
if (g1 >= 0 && g2 >= 0) {
iter->trees[0] = m->body_treeid[m->geom_bodyid[g1]];
iter->trees[1] = m->body_treeid[m->geom_bodyid[g2]];
if (iter->trees[0] < 0 && iter->trees[1] < 0) {
mjERROR("contact %d is between two static bodies", efc_id); // SHOULD NOT OCCUR
}
}
// no shortcut for flex contacts: enable generic scan
else {
iter->jac_idx = 0;
}
}
// connect or weld constraints
else if (efc_type == mjCNSTR_EQUALITY &&
(m->eq_type[efc_id] == mjEQ_CONNECT ||
m->eq_type[efc_id] == mjEQ_WELD)) {
int b1 = m->eq_obj1id[efc_id];
int b2 = m->eq_obj2id[efc_id];
// get body ids if using site semantics
if (m->eq_objtype[efc_id] == mjOBJ_SITE) {
b1 = m->site_bodyid[b1];
b2 = m->site_bodyid[b2];
}
// get trees
iter->trees[0] = m->body_treeid[b1];
iter->trees[1] = m->body_treeid[b2];
if (iter->trees[0] < 0 && iter->trees[1] < 0) {
mjERROR("equality %d is between two static bodies", efc_id); // SHOULD NOT OCCUR
}
}
// otherwise enable generic scan
else {
iter->jac_idx = 0;
}
}
// add 0, 1 or 2 edges to uncompressed CSR adjacency matrix
// increment rownnz using tree_tree to de-dupe; return number of edges added
static int addEdge(int* rownnz, int* colind, mjtByte* tree_tree, int ntree, int tree1, int tree2) {
if (tree1 == -1 && tree2 == -1) {
mjERROR("self-edge of the static tree"); // SHOULD NOT OCCUR
return 0;
}
// handle static trees (treat as self-edge)
if (tree1 == -1) tree1 = tree2;
if (tree2 == -1) tree2 = tree1;
// skip if edge already present
if (tree_tree[tree1*ntree + tree2]) {
return 0;
}
// add edge
tree_tree[tree1*ntree + tree2] = 1;
colind[tree1*ntree + rownnz[tree1]++] = tree2; // uncompressed format, rowadr is known
// add flipped edge (off-diagonal)
if (tree1 != tree2) {
tree_tree[tree2*ntree + tree1] = 1;
colind[tree2*ntree + rownnz[tree2]++] = tree1; // uncompressed format, rowadr is known
return 2;
}
return 1;
}
// find tree-tree edges (column indices), return total number of edges
// efc_tree: first nonegative tree index of each constraint
static int findEdges(const mjModel* m, const mjData* d,
int* rownnz, int* colind, mjtByte* tree_tree, int* efc_tree, int ntree) {
int nefc = d->nefc;
int nnz = 0;
int efc_type = -1;
int efc_id = -1;
// clear row nonzeros
mju_zeroInt(rownnz, ntree);
// iterate over constraints, compute tree-tree edges, assign efc_tree
for (int i=0; i < nefc; i++) {
// row i is still in the same constraint: skip it,
if (efc_type == d->efc_type[i] && efc_id == d->efc_id[i]) {
// unless it is a flex equality, where the tree pattern changes per dof
if (!(efc_type == mjCNSTR_EQUALITY &&
(m->eq_type[efc_id] == mjEQ_FLEX ||
m->eq_type[efc_id] == mjEQ_FLEXVERT ||
m->eq_type[efc_id] == mjEQ_FLEXSTRAIN))) {
// copy tree assignment from previous constraint and continue
efc_tree[i] = efc_tree[i-1];
continue;
}
}
efc_type = d->efc_type[i];
efc_id = d->efc_id[i];
// initialize tree iterator
mjTreeIter iter;
treeIterInit(m, d, i, &iter);
// iterate over trees involved in constraint i
int tree1 = treeNext(m, d, i, &iter);
if (tree1 != -2) {
int tree2 = treeNext(m, d, i, &iter);
// assign tree to constraint, one of (tree1, tree2) must be non-negative
efc_tree[i] = tree1 >= 0 ? tree1 : tree2;
if (efc_tree[i] < 0) {
mjERROR("constraint %d is between two static bodies", i); // SHOULD NOT OCCUR
}
// add one edge or continue to search for more edges
if (tree2 == -2) {
nnz += addEdge(rownnz, colind, tree_tree, ntree, tree1, -1);
} else {
while (tree2 != -2) {
nnz += addEdge(rownnz, colind, tree_tree, ntree, tree1, tree2);
tree1 = tree2;
tree2 = treeNext(m, d, i, &iter);
}
}
} else {
mjERROR("no tree found for constraint %d", i); // SHOULD NOT OCCUR
}
}
// flex stiffness couples all vertices (nodes for interpolated flexes) of a flex without any
// constraint row representing the coupling: union the trees of every stiffness-active flex
// (star around the first dynamic tree). This keeps the partition valid when the implicit
// effective metric (mj_flexCG) carries the stiffness inside the constraint solve. Awake
// trees only: sleeping trees must stay out of islands (mj_sleep invariant, matching the
// constraint filter); waking a flex as a unit remains the wake machinery's job.
for (int f=0; f < m->nflex; f++) {
// mirror the stiffness-activity conditions of engine_derivative's flexStiff_active /
// flexInterp_processed: deformable dim>=2 flex with bending or nonzero stiffness
if (m->flex_rigid[f] || m->flex_dim[f] < 2) {
continue;
}
int sadr = m->flex_stiffnessadr[f];
if (m->flex_bendingadr[f] < 0 && (sadr < 0 || m->flex_stiffness[sadr] == 0)) {
continue;
}
int num, adr;
const int* bodyid;
if (m->flex_interp[f]) {
num = m->flex_nodenum[f];
adr = m->flex_nodeadr[f];
bodyid = m->flex_nodebodyid;
} else {
num = m->flex_vertnum[f];
adr = m->flex_vertadr[f];
bodyid = m->flex_vertbodyid;
}
int tree1 = -1;
for (int j=0; j < num; j++) {
int treeid = m->body_treeid[bodyid[adr+j]];
if (treeid < 0 || treeid == tree1 || !d->tree_awake[treeid]) {
continue;
}
if (tree1 < 0) {
tree1 = treeid;
} else {
nnz += addEdge(rownnz, colind, tree_tree, ntree, tree1, treeid);
}
}
}
return nnz;
}
//-------------------------- main entry-point -----------------------------------------------------
// discover islands:
// nisland, island_idofadr, dof_island, dof_islandnext, island_efcadr, efc_island, efc_islandnext
void mj_island(const mjModel* m, mjData* d) {
int nv = m->nv, nefc = d->nefc, ntree = m->ntree;
// no constraints or islands disabled: quick return
if (mjDISABLED(mjDSBL_ISLAND) || !nefc) {
d->nisland = d->nidof = 0;
return;
}
mj_markStack(d);
// dense tree-tree adjacency matrix
int ntree2 = ntree * ntree;
mjtByte* tree_tree = mjSTACKALLOC(d, ntree2, mjtByte);
memset(tree_tree, 0, ntree2);
// CSR representation of tree-tree adjacency matrix (uncompressed)
int* colind = mjSTACKALLOC(d, ntree2, int);
int* rownnz = mjSTACKALLOC(d, ntree, int);
int* rowadr = mjSTACKALLOC(d, ntree, int);
for (int r=0; r < ntree; r++) {
rowadr[r] = r * ntree;
}
// first non-negative tree index of each constraint, used later for computing efc_island
int* efc_tree = mjSTACKALLOC(d, nefc, int);
// compute tree-tree adjacency matrix: fill rownnz and colind
int nnz = findEdges(m, d, rownnz, colind, tree_tree, efc_tree, ntree);
// discover islands
int* tree_island = mjSTACKALLOC(d, ntree, int);
int* stack = mjSTACKALLOC(d, nnz, int);
d->nisland = mj_floodFill(tree_island, ntree, rownnz, rowadr, colind, stack);
// no islands found: quick return
if (!d->nisland) {
d->nidof = 0;
mj_freeStack(d);
return;
}
// count nidof: total number of dofs in islands
int nidof = 0;
for (int i=0; i < ntree; i++) {
if (tree_island[i] >= 0) {
nidof += m->tree_dofnum[i];
}
}
d->nidof = nidof;
// allocate island arrays on arena
if (!arenaAllocIsland(m, d)) {
mj_freeStack(d);
return;
}
// local copy
int nisland = d->nisland;
// ------------------------------------- trees ---------------------------------------------------
// copy tree_island from stack to arena
mju_copyInt(d->tree_island, tree_island, ntree);
// compute island_ntree, number of trees per island
mju_zeroInt(d->island_ntree, nisland);
for (int i=0; i < ntree; i++) {
int island = tree_island[i];
if (island >= 0) {
d->island_ntree[island]++;
}
}
// compute island_itreeadr (cumsum of island_ntree)
d->island_itreeadr[0] = 0;
for (int i=1; i < nisland; i++) {
d->island_itreeadr[i] = d->island_itreeadr[i-1] + d->island_ntree[i-1];
}
int last_tree = d->island_itreeadr[nisland-1] + d->island_ntree[nisland-1];
// compute map_itree2tree
int* island_ntree2 = mjSTACKALLOC(d, nisland + 1, int); // last elem counts unconstrained trees
mju_zeroInt(island_ntree2, nisland + 1);
for (int i=0; i < ntree; i++) {
int island = tree_island[i];
if (island >= 0) {
d->map_itree2tree[d->island_itreeadr[island] + island_ntree2[island]++] = i;
} else {
d->map_itree2tree[last_tree + island_ntree2[nisland]++] = i;
}
}
// SHOULD NOT OCCUR
if (!mju_compare(island_ntree2, d->island_ntree, nisland)) mjERROR("island_ntree miscount");
if (last_tree + island_ntree2[nisland] != ntree) mjERROR("miscount of unconstrained trees");
// ------------------------------------- degrees of freedom --------------------------------------
// compute dof_island, island_nv
mju_zeroInt(d->island_nv, nisland);
for (int i=0; i < nv; i++) {
// assign DOFs to islands
int island = tree_island[m->dof_treeid[i]]; // -1 if unconstrained
d->dof_island[i] = island;
// increment island_nv
if (island >= 0) {
d->island_nv[island]++;
}
}
// compute island_idofadr (cumsum of island_nv)
d->island_idofadr[0] = 0;
for (int i=1; i < nisland; i++) {
d->island_idofadr[i] = d->island_idofadr[i-1] + d->island_nv[i-1];
}
// compute dof <-> idof maps
int* island_nv2 = mjSTACKALLOC(d, nisland + 1, int); // last element counts unconstrained DOFs
mju_zeroInt(island_nv2, nisland + 1);
for (int dof=0; dof < nv; dof++) {
int island = d->dof_island[dof];
int idof;
if (island >= 0) {
// constrained dof
idof = d->island_idofadr[island] + island_nv2[island]++;
} else {
// unconstrained dof
idof = nidof + island_nv2[nisland]++;
}
d->map_dof2idof[dof] = idof;
d->map_idof2dof[idof] = dof; // only the first nidof elements of map_idof2dof are in some island
}
// SHOULD NOT OCCUR
if (!mju_compare(island_nv2, d->island_nv, nisland)) mjERROR("island_nv miscount");
if (nidof + island_nv2[nisland] != nv) mjERROR("miscount of unconstrained dofs");
// compute island_dofadr (used for visualization)
for (int i=0; i < nisland; i++) {
d->island_dofadr[i] = d->map_idof2dof[d->island_idofadr[i]];
}
// ------------------------------------- constraints ---------------------------------------------
// compute efc_island, island_{ne,nf,nefc}
mju_zeroInt(d->island_ne, nisland);
mju_zeroInt(d->island_nf, nisland);
mju_zeroInt(d->island_nefc, nisland);
for (int i=0; i < nefc; i++) {
int island = tree_island[efc_tree[i]];
d->efc_island[i] = island;
d->island_nefc[island]++;
switch (d->efc_type[i]) {
case mjCNSTR_EQUALITY:
d->island_ne[island]++;
break;
case mjCNSTR_FRICTION_DOF:
case mjCNSTR_FRICTION_TENDON:
d->island_nf[island]++;
break;
default:
break;
}
}
// compute island_iefcadr (cumsum of island_nefc)
d->island_iefcadr[0] = 0;
for (int i=1; i < nisland; i++) {
d->island_iefcadr[i] = d->island_iefcadr[i-1] + d->island_nefc[i-1];
}
// compute efc <-> iefc maps
int* island_nefc2 = island_nv2; // reuse island_nv2
mju_zeroInt(island_nefc2, nisland);
for (int c=0; c < nefc; c++) {
int island = d->efc_island[c];
int ic = d->island_iefcadr[island] + island_nefc2[island]++;
d->map_efc2iefc[c] = ic;
d->map_iefc2efc[ic] = c;
}
// SHOULD NOT OCCUR
if (!mju_compare(island_nefc2, d->island_nefc, nisland)) mjERROR("island_nefc miscount");
// copy position-dependent efc vectors required by solver
mju_gatherInt(d->iefc_type, d->efc_type, d->map_iefc2efc, nefc);
mju_gatherInt(d->iefc_id, d->efc_id, d->map_iefc2efc, nefc);
mju_gather(d->iefc_frictionloss, d->efc_frictionloss, d->map_iefc2efc, nefc);
mju_gather(d->iefc_D, d->efc_D, d->map_iefc2efc, nefc);
mju_gather(d->iefc_R, d->efc_R, d->map_iefc2efc, nefc);
mj_freeStack(d);
}