PR #8634 Optimize quantized LSTM

optimize some lstm gate computation. Remove no need string constructions.

zhanghuanrong committed 4 years ago

change gcc optimization flags for computation bound logics in rnn_helpers

zhanghuanrong committed 4 years ago

better qgemm for M=1

zhanghuanrong committed 4 years ago

Some improve on avx512

zhanghuanrong committed 4 years ago

add condition to limit GCC related marcros

zhanghuanrong committed 4 years ago

Correct QGemm assembly for M=1 AVX2 optimization to pass mlas_test.

zhanghuanrong committed 4 years ago

Fix rnn_helper build issue for wasm.

zhanghuanrong committed 4 years ago

better asm code here according to feedbacks.

zhanghuanrong committed 4 years ago

Remove customized vectorize and unroll option for GCC.

zhanghuanrong committed 4 years ago

Better restrict semantic for merge_lstm_gates_to_memory() by adding in_place().

zhanghuanrong committed 4 years ago

Force CI restart as it stucked by the onnxruntime-python-checks-ci-pipeline which can not restart.

zhanghuanrong committed 4 years ago

onnxruntime Optimize quantized LSTM #8634 Merged