optimize receiving data with TLS 1.2 and aes-128-gcm

[bdaehlie]

We ran the Rustls benchmarking tool with the Rustls v0.23 release and Rustls appears to be a bit slower than OpenSSL 3.2.0 when receiving data with TLS 1.2 and aes-128-gcm. We should optimize until we are faster than OpenSSL in this benchmark.

[SimonTate]

I believe I can reproduce with openssl-bench ( https://github.com/ctz/openssl-bench.git @ ea4f92bb969d151b8cedf10083819b980d19e533) on M3 Pro MacBook with latest master of openssl and rustls.

OpenSSL: dc9bc6c8e1bd329ead703417a2235ab3e97557ec
rustls: 38573b2

rustls:

make -f admin/bench-measure.mk measure
target/release/examples/bench bulk TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256
bulk    TLSv1_2 TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256   max_fragment_size:default       send    6842.16 MB/s
bulk    TLSv1_2 TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256   max_fragment_size:default       recv    4874.97 MB/s

openssl with openssl-bench:

state@BTM-FQV52JYD09:~/Developer/openssl-bench(master)
$ make measure
env LD_LIBRARY_PATH=../openssl ./bench bulk ECDHE-RSA-AES128-GCM-SHA256 1048576
negotiated TLSv1.2 with ECDHE-RSA-AES128-GCM-SHA256
send: 7001.01 MB/s
recv: 7779.95 MB/s

Is this the test you're referring to @bdaehlie ?