Skip to content

Commit bcecdfd

Browse files
jeffzhou2000jeffzhou2000
committed
project: sync with upstream
1 parent b8a47ee commit bcecdfd

File tree

2 files changed

+57
-19
lines changed

2 files changed

+57
-19
lines changed

‎scripts/build-run-android.sh‎

Lines changed: 15 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -534,12 +534,12 @@ function run_test-op()
534534

535535
echo "adb shell cd ${REMOTE_PATH} \
536536
&& export LD_LIBRARY_PATH=${REMOTE_PATH} \
537-
&& ${REMOTE_PATH}/ggmlhexagon-testops test -o ${opname} -a ${mulmat_algotype}"
537+
&& ${REMOTE_PATH}/ggmlhexagon-testops test -o ${opname} -a ${mulmat_algotype} -i ${hexagon_backend}"
538538

539539
echo "\n"
540540
adb shell "cd ${REMOTE_PATH} \
541541
&& export LD_LIBRARY_PATH=${REMOTE_PATH} \
542-
&& ${REMOTE_PATH}/ggmlhexagon-testops test -o ${opname} -a ${mulmat_algotype}"
542+
&& ${REMOTE_PATH}/ggmlhexagon-testops test -o ${opname} -a ${mulmat_algotype} -i ${hexagon_backend}"
543543

544544
}
545545

@@ -673,11 +673,12 @@ function show_usage()
673673
echo " $0 run_llamabench 0(QNN_CPU)/1(QNN_GPU)/2(QNN_NPU)/3(cdsp)/4(ggml)"
674674
echo " $0 run_threadsafety 0(QNN_CPU)/1(QNN_GPU)/2(QNN_NPU)/3(cdsp)/4(ggml)"
675675
echo " $0 run_perfop MUL_MAT 0(QNN_CPU)/1(QNN_GPU)/2(QNN_NPU)/3(cdsp)/4(ggml)"
676+
echo " $0 run_testop MUL_MAT 0(QNN_CPU)/1(QNN_GPU)/2(QNN_NPU)/3(cdsp)/4(ggml)"
676677
echo " $0 run_benchmark ADD/MUL_MAT 0(QNN_CPU)/1(QNN_GPU)/2(QNN_NPU)/3(cdsp)/4(ggml)"
677678
echo " $0 run_benchmark ADD/MUL_MAT 0(QNN_CPU)/1(QNN_GPU)/2(QNN_NPU)/3(cdsp)/4(ggml) 256/512/1024/2048/4096 256/512/1024/2048/4096"
678679
echo " $0 run_benchmark MUL_MAT 3(cdsp) mulmat_algotype(0,1,2,3,4,5,6,32,33) (verify performance of mulmat on cDSP)"
679680
echo " $0 run_perfop MUL_MAT 3(cdsp) mulmat_algotype(0,1,2,3,4,5,6,32,33) (verify performance of mulmat on cDSP)"
680-
echo " $0 run_testop MUL_MAT mulmat_algotype(0,1,2,3,4,5,6,32,33) (verify accuracy of mulmat on cDSP)"
681+
echo " $0 run_testop MUL_MAT 3(cdsp) mulmat_algotype(0,1,2,3,4,5,6,32,33) (verify accuracy of mulmat on cDSP)"
681682
echo " $0 run_testop ADD (verify accuracy of add on cDSP)"
682683

683684
echo -e "\n\n\n"
@@ -769,9 +770,10 @@ elif [ $# == 3 ]; then
769770
run_benchmark
770771
exit 0
771772
elif [ "$1" == "run_testop" ]; then
772-
opname=$2
773-
mulmat_algotype=$3
774-
check_mulmat_algotype
773+
opname=MUL_MAT
774+
mulmat_algotype=32
775+
hexagon_backend=$3
776+
check_hexagon_backend
775777
run_test-op
776778
exit 0
777779
elif [ "$1" == "run_perfop" ]; then
@@ -802,6 +804,13 @@ elif [ $# == 4 ]; then
802804
check_mulmat_algotype
803805
run_perf-op
804806
exit 0
807+
elif [ "$1" == "run_testop" ]; then
808+
opname=MUL_MAT
809+
hexagon_backend=3
810+
mulmat_algotype=$4
811+
check_mulmat_algotype
812+
run_test-op
813+
exit 0
805814
else
806815
show_usage
807816
exit 1

‎tests/ggmlhexagon-testops.cpp‎

Lines changed: 42 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -2088,30 +2088,59 @@ struct test_ssm_scan : public test_case {
20882088
const ggml_type type;
20892089

20902090
const int64_t d_state;
2091-
const int64_t d_inner;
2091+
const int64_t head_dim;
2092+
const int64_t n_head;
2093+
const int64_t n_group;
20922094
const int64_t n_seq_tokens;
20932095
const int64_t n_seqs;
20942096

20952097
std::string vars() override {
2096-
return VARS_TO_STR5(type, d_state, d_inner, n_seq_tokens, n_seqs);
2098+
return VARS_TO_STR7(type, d_state, head_dim, n_head, n_group, n_seq_tokens, n_seqs);
20972099
}
20982100

20992101
test_ssm_scan(ggml_type type = GGML_TYPE_F32,
2100-
int64_t d_state = 32, int64_t d_inner = 32, int64_t n_seq_tokens = 32, int64_t n_seqs = 32)
2101-
: type(type), d_state(d_state), d_inner(d_inner), n_seq_tokens(n_seq_tokens), n_seqs(n_seqs) {}
2102+
int64_t d_state = 32,
2103+
int64_t head_dim = 1, // non-zero for Mamba-2
2104+
int64_t n_head = 32,
2105+
int64_t n_group = 1,
2106+
int64_t n_seq_tokens = 32,
2107+
int64_t n_seqs = 32)
2108+
: type(type), d_state(d_state), head_dim(head_dim), n_head(n_head), n_group(n_group), n_seq_tokens(n_seq_tokens), n_seqs(n_seqs) {}
21022109

21032110
ggml_tensor * build_graph(ggml_context * ctx) override {
2104-
ggml_tensor * s = ggml_new_tensor(ctx, type, 4, std::vector<int64_t>{ d_state, d_inner, n_seqs, 1 }.data());
2105-
ggml_tensor * x = ggml_new_tensor(ctx, type, 4, std::vector<int64_t>{ d_inner, n_seq_tokens, n_seqs, 1 }.data());
2106-
ggml_tensor * dt = ggml_new_tensor(ctx, type, 4, std::vector<int64_t>{ d_inner, n_seq_tokens, n_seqs, 1 }.data());
2107-
ggml_tensor * A = ggml_new_tensor(ctx, type, 4, std::vector<int64_t>{ d_state, d_inner, 1 , 1 }.data());
2108-
ggml_tensor * B = ggml_new_tensor(ctx, type, 4, std::vector<int64_t>{ d_state, n_seq_tokens, n_seqs, 1 }.data());
2109-
ggml_tensor * C = ggml_new_tensor(ctx, type, 4, std::vector<int64_t>{ d_state, n_seq_tokens, n_seqs, 1 }.data());
2110-
ggml_tensor * out = ggml_ssm_scan(ctx, s, x, dt, A, B, C);
2111+
ggml_tensor * s = ggml_new_tensor_4d(ctx, type, d_state, head_dim, n_head, n_seqs);
2112+
ggml_tensor * x = ggml_new_tensor_4d(ctx, type, head_dim, n_head, n_seq_tokens, n_seqs);
2113+
ggml_tensor * dt = ggml_new_tensor_3d(ctx, type, n_head, n_seq_tokens, n_seqs);
2114+
ggml_tensor * A = ggml_new_tensor_2d(ctx, type, (head_dim > 1) ? 1 : d_state, n_head);
2115+
ggml_tensor * B = ggml_new_tensor_4d(ctx, type, d_state, n_group, n_seq_tokens, n_seqs);
2116+
ggml_tensor * C = ggml_new_tensor_4d(ctx, type, d_state, n_group, n_seq_tokens, n_seqs);
2117+
ggml_tensor * ids = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, n_seqs);
2118+
ggml_tensor * out = ggml_ssm_scan(ctx, s, x, dt, A, B, C, ids);
21112119
return out;
21122120
}
2113-
};
21142121

2122+
// similar to test_mul_mat_id
2123+
void initialize_tensors(ggml_context * ctx) override {
2124+
std::random_device rd;
2125+
std::default_random_engine rng(rd());
2126+
for (ggml_tensor * t = ggml_get_first_tensor(ctx); t != NULL; t = ggml_get_next_tensor(ctx, t)) {
2127+
if (t->type == GGML_TYPE_I32) {
2128+
if (ggml_is_view_op(t->op)) { continue; }
2129+
// ids
2130+
for (int64_t r = 0; r < ggml_nrows(t); r++) {
2131+
std::vector<int32_t> data(t->ne[0]);
2132+
for (int i = 0; i < t->ne[0]; i++) {
2133+
data[i] = i;
2134+
}
2135+
std::shuffle(data.begin(), data.end(), rng);
2136+
ggml_backend_tensor_set(t, data.data(), r * t->nb[1], t->ne[0] * sizeof(int32_t));
2137+
}
2138+
} else {
2139+
init_tensor_uniform(t);
2140+
}
2141+
}
2142+
}
2143+
};
21152144
// GGML_OP_RWKV_WKV6
21162145
struct test_rwkv_wkv6 : public test_case {
21172146
const ggml_type type;
@@ -3321,7 +3350,7 @@ struct test_upscale_ext : public test_case {
33213350
ggml_tensor * a = ggml_new_tensor(ctx, type, 4, ne.data());
33223351
ggml_set_name(a, "a");
33233352

3324-
ggml_tensor * out = ggml_upscale_ext(ctx, a, ne_tgt[0], ne_tgt[1],ne_tgt[2], ne_tgt[3], mode);
3353+
ggml_tensor * out = ggml_interpolate(ctx, a, ne_tgt[0], ne_tgt[1],ne_tgt[2], ne_tgt[3], mode);
33253354
ggml_set_name(out, "out");
33263355

33273356
return out;

0 commit comments

Comments
 (0)