Closed jalil-salame closed 10 months ago
Looks to be an instruction perf win over the board:
iai_templates::templates::render teams:teams_setup() Instructions: 34396|39255 (-12.3780%) [-1.14127x] L1 Hits: 48685|56277 (-13.4904%) [-1.15594x] L2 Hits: 29|38 (-23.6842%) [-1.31034x] RAM Hits: 344|381 (-9.71129%) [-1.10756x] Total read+write: 49058|56696 (-13.4718%) [-1.15569x] Estimated Cycles: 60870|69802 (-12.7962%) [-1.14674x] iai_templates::templates::render big_table:big_table_setup() Instructions: 6669990|12080468 (-44.7870%) [-1.81117x] L1 Hits: 10067825|18575012 (-45.7991%) [-1.84499x] L2 Hits: 12260|12282 (-0.17912%) [-1.00179x] RAM Hits: 1979|2008 (-1.44422%) [-1.01465x] Total read+write: 10082064|18589302 (-45.7642%) [-1.84380x] Estimated Cycles: 10198390|18706702 (-45.4827%) [-1.83428x] iai_templates::templates::render realistic:realistic_setup() Instructions: 187747|206193 (-8.94599%) [-1.09825x] L1 Hits: 260913|289793 (-9.96573%) [-1.11069x] L2 Hits: 510|513 (-0.58480%) [-1.00588x] RAM Hits: 684|706 (-3.11615%) [-1.03216x] Total read+write: 262107|291012 (-9.93258%) [-1.11028x] Estimated Cycles: 287403|317068 (-9.35604%) [-1.10322x] iai_templates::templates::context big_context:big_context_setup() Instructions: 490359|491400 (-0.21184%) [-1.00212x] L1 Hits: 684141|685723 (-0.23071%) [-1.00231x] L2 Hits: 5288|5333 (-0.84380%) [-1.00851x] RAM Hits: 181|167 (+8.38323%) [+1.08383x] Total read+write: 689610|691223 (-0.23335%) [-1.00234x] Estimated Cycles: 716916|718233 (-0.18337%) [-1.00184x] Running benches/iai-value.rs (target/release/deps/iai_value-452022e514e9078a) iai_value::serialize::serialize_value page:& Page :: default() Instructions: 7859|7876 (-0.21585%) [-1.00216x] L1 Hits: 10762|10781 (-0.17624%) [-1.00177x] L2 Hits: 4|3 (+33.3333%) [+1.33333x] RAM Hits: 136|138 (-1.44928%) [-1.01471x] Total read+write: 10902|10922 (-0.18312%) [-1.00183x] Estimated Cycles: 15542|15626 (-0.53757%) [-1.00540x]
Thanks. I'll probably change a bit the features later but that's some solid improvement!
Looks to be an instruction perf win over the board: