Switch mj_solveM2 to use CSR representation.

PiperOrigin-RevId: 717639495
Change-Id: I59eee0f23606481480ec5acb8a3bba2c14b2ea2f

src/engine/engine_core_smooth.c

@@ -1668,9 +1668,10 @@ void mj_solveLDs(mjtNum* restrict x, const mjtNum* qLDs, const mjtNum* qLDiagInv
       continue;
     }
 
-    int adr = rowadr[i];
+    int d;
-    int d = rownnz[i] - 1;
+    if ((d = rownnz[i] - 1) > 0) {
-    if (d > 0) {
+      int adr = rowadr[i];
+
       // one vector
       if (n == 1) {
         x[i] -= mju_dotSparse(qLDs+adr, x, d, colind+adr, /*flg_unc1=*/0);
@@ -1769,41 +1770,53 @@ void mj_solveM_island(const mjModel* m, const mjData* d, mjtNum* restrict x, int
 void mj_solveM2(const mjModel* m, mjData* d, mjtNum* x, const mjtNum* y,
                 const mjtNum* sqrtInvD, int n) {
   // local copies of key variables
-  mjtNum* qLD = d->qLD;
+  int nv = m->nv, nC = m->nC;
-  int* dof_Madr = m->dof_Madr;
+  const int* rownnz = d->C_rownnz;
-  int* dof_parentid = m->dof_parentid;
+  const int* rowadr = d->C_rowadr;
-  int nv = m->nv;
+  const int* colind = d->C_colind;
+  const int* diagnum = m->dof_simplenum;
 
   // x = y
   mju_copy(x, y, n * nv);
 
-  // loop over the n input vectors
+  // temporary: make local CSR version of qLD
-  for (int ivec=0; ivec < n; ivec++) {
+  mj_markStack(d);
-    int offset = ivec*nv;
+  mjtNum* qLD = mjSTACKALLOC(d, nC, mjtNum);
+  for (int i=0; i < nC; i++) {
+    qLD[i] = d->qLD[d->mapM2C[i]];
+  }
 
-    // x <- inv(L') * x; skip simple, exploit sparsity of input vector
+  // x <- L^-T x
-    for (int i=nv-1; i >= 0; i--) {
+  for (int i=nv-1; i > 0; i--) {
-      mjtNum tmp;
+    // skip diagonal rows
-      if (!m->dof_simplenum[i] && (tmp = x[i+offset])) {
+    if (diagnum[i]) {
-        // init
+      continue;
-        int Madr_ij = dof_Madr[i]+1;
+    }
-        int j = dof_parentid[i];
 
-        // traverse ancestors backwards
+    // prepare row i column address range
-        while (j >= 0) {
+    int start = rowadr[i];
-          x[j+offset] -= qLD[Madr_ij++] * tmp;        // x(j) -= L(i,j) * x(i)
+    int end = start + rownnz[i] - 1;
 
-          // advance to parent
+    // process all vectors
-          j = dof_parentid[j];
+    for (int offset=0; offset < n*nv; offset+=nv) {
+      mjtNum x_i;
+      if ((x_i = x[i+offset])) {
+        for (int adr=start; adr < end; adr++) {
+          x[offset + colind[adr]] -= qLD[adr] * x_i;
         }
       }
     }
+  }
 
-    // x <- sqrt(inv(D)) * x
+  // x <- D^-1/2 x
-    for (int i=0; i < nv; i++) {
+  for (int i=0; i < nv; i++) {
-      x[i+offset] *= sqrtInvD[i];  // x(i) /= sqrt(L(i,i))
+    mjtNum invD_i = sqrtInvD[i];
+    for (int offset=0; offset < n*nv; offset+=nv) {
+      x[i+offset] *= invD_i;
     }
   }
+
+  mj_freeStack(d);
 }