Optimize GCM

By allowing multiple blocks for clmul, slight speedup there though still
far behind optimum.

Precompute a table of multiples of H, 3-4x faster on systems without clmul
(and still no secret indexes).

Refactor GMAC to not derive from GHASH

1 files changed, 2 insertions, 0 deletions

diff --git a/src/tests/test_mac.cpp b/src/tests/test_mac.cpp
index 471a15fed..2792aeb3e 100644
--- a/src/tests/test_mac.cpp
+++ b/src/tests/test_mac.cpp
@@ -68,6 +68,7 @@ class Message_Auth_Tests final : public Text_Based_Test
 
             // Test to make sure clear() resets what we need it to
             mac->set_key(key);
+            mac->start(iv);
             mac->update("some discarded input");
             mac->clear();
 
@@ -81,6 +82,7 @@ class Message_Auth_Tests final : public Text_Based_Test
             result.confirm("Clone has different pointer", mac.get() != clone.get());
             result.test_eq("Clone has same name", mac->name(), clone->name());
             clone->set_key(key);
+            clone->start(iv);
             clone->update(Test::rng().random_vec(32));
 
             result.test_eq(provider + " correct mac", mac->verify_mac(expected.data(), expected.size()), true);