Introduces support for 3D and 4D inputs in ACL acl_lowp and acl_lowp_sq matmul.

[renato-arantes]

Description

This PR introduces 3D and 4D support for acl_lowp_matmul_t and acl_lowp_sq_matmul_t in ACL. Solves issue.

General

• Do all unit and benchdnn tests (make test and make test_benchdnn_*) pass locally for each commit?
• Have you formatted the code using clang-format?
• Have you submitted performance data that demonstrates performance improvements?

Performance improvements

Below are small test case logs that demonstrate performance numbers before and after.

• 3D lowp:

OMP_NUM_THREADS=16 ./tests/benchdnn/benchdnn --mode=pc --matmul --dt=s8:s8:f32 4x1024x1024:1x1024x1024

ACL: total perf: min(ms):1.53076 avg(ms):1.55271
JIT: total perf: min(ms):1029.39 avg(ms):1029.92

• 3D lowp_sq:

OMP_NUM_THREADS=16 ./tests/benchdnn/benchdnn --mode=pc --matmul --dt=s8:s8:s8 4x1024x1024:1x1024x1024

ACL: total perf: min(ms):1.38965 avg(ms):1.4086
JIT: total perf: min(ms):1032.73 avg(ms):1033.18

• 4D lowp:

OMP_NUM_THREADS=16 ./tests/benchdnn/benchdnn --mode=pc --matmul --dt=s8:s8:f32 4x4x1024x1024:1x1x1024x1024

ACL: total perf: min(ms):5.95874 avg(ms):5.9957
JIT: total perf: min(ms):4003.9 avg(ms):4005.17

• 4D lowp_sq:

OMP_NUM_THREADS=16 ./tests/benchdnn/benchdnn --mode=pc --matmul --dt=s8:s8:s8 4x4x1024x1024:1x1x1024x1024

ACL: total perf: min(ms):5.31323 avg(ms):5.34218
JIT: total perf: min(ms):4022.92 avg(ms):4023.37