chenlin/Mujoco_WASM
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add mj_stackAllocInt to get correct size for allocating ints on mjData stack. Reducing stack memory usage by 10% - 15%.

Browse Source 
PiperOrigin-RevId: 517672195
Change-Id: Ie14e6c9c99e8b9e1c5d3686b77631f2df0651c67
This commit is contained in:
Kyle Bayes
2023-03-18 13:03:11 -07:00
committed by Copybara-Service
parent fae5e822c0
commit 8696247f9d
17 changed files with 103 additions and 48 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/benchmark/engine_util_sparse_benchmark_test.cc
+10 -10
View File
@@ -47,7 +47,7 @@ void ABSL_ATTRIBUTE_NOINLINE mju_sqrMatTDSparse_baseline(
const int* rowsuper, const int* rownnzT, const int* rowadrT,
const int* colindT, const int* rowsuperT, mjData* d) {
mjMARKSTACK;
int* chain = (int*)mj_stackAlloc(d, 2 * nc);
int* chain = mj_stackAllocInt(d, 2 * nc);
mjtNum* buffer = mj_stackAlloc(d, nc);
for (int r = 0; r < nc; r++) {
@@ -433,9 +433,9 @@ static void BM_combineSparse(benchmark::State& state, CombineFuncPtr func) {
// allocate
mjMARKSTACK;
mjtNum* H = mj_stackAlloc(d, m->nv*m->nv);
int* rownnz = (int*)mj_stackAlloc(d, m->nv);
int* rowadr = (int*)mj_stackAlloc(d, m->nv);
int* colind = (int*)mj_stackAlloc(d, m->nv*m->nv);
int* rownnz = mj_stackAllocInt(d, m->nv);
int* rowadr = mj_stackAllocInt(d, m->nv);
int* colind = mj_stackAllocInt(d, m->nv*m->nv);
// compute D corresponding to quad states
mjtNum* D = mj_stackAlloc(d, d->nefc);
@@ -511,9 +511,9 @@ static void BM_transposeSparse(benchmark::State& state, TransposeFuncPtr func) {
// need uncompressed layout
mjtNum* res = mj_stackAlloc(d, m->nv * d->nefc);
int* res_rownnz = (int*)mj_stackAlloc(d, m->nv);
int* res_rowadr = (int*)mj_stackAlloc(d, m->nv);
int* res_colind = (int*)mj_stackAlloc(d, m->nv * d->nefc);
int* res_rownnz = mj_stackAllocInt(d, m->nv);
int* res_rowadr = mj_stackAllocInt(d, m->nv);
int* res_colind = mj_stackAllocInt(d, m->nv * d->nefc);
// time benchmark
for (auto s : state) {
@@ -555,9 +555,9 @@ static void BM_sqrMatTDSparse(benchmark::State& state, SqrMatTDFuncPtr func) {
// allocate
mjMARKSTACK;
mjtNum* H = mj_stackAlloc(d, m->nv * m->nv);
int* rownnz = (int*)mj_stackAlloc(d, m->nv);
int* rowadr = (int*)mj_stackAlloc(d, m->nv);
int* colind = (int*)mj_stackAlloc(d, m->nv * m->nv);
int* rownnz = mj_stackAllocInt(d, m->nv);
int* rowadr = mj_stackAllocInt(d, m->nv);
int* colind = mj_stackAllocInt(d, m->nv * m->nv);
// compute D corresponding to quad states
mjtNum* D = mj_stackAlloc(d, d->nefc);