From b859d70d4a04db2e3b39b5c86cb6750d3c5b9593 Mon Sep 17 00:00:00 2001 From: Andy Polyakov Date: Mon, 21 Dec 2015 14:26:12 +0100 Subject: [PATCH] bn/asm/bn-c64xplus.asm: update commentary. Reviewed-by: Rich Salz --- crypto/bn/asm/bn-c64xplus.asm | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/crypto/bn/asm/bn-c64xplus.asm b/crypto/bn/asm/bn-c64xplus.asm index 7cd58817e7..7b72bff4ac 100644 --- a/crypto/bn/asm/bn-c64xplus.asm +++ b/crypto/bn/asm/bn-c64xplus.asm @@ -284,8 +284,9 @@ _bn_mul_comba4: .if 0 BNOP sploopNxM?,3 ;; Above mentioned m*2*(n+1)+10 does not apply in n=m=4 case, - ;; because of read-after-write penalties, it's rather - ;; n*2*(n+3)+10, or 66 cycles [plus various overheads]... + ;; because of low-counter effect, when prologue phase finishes + ;; before SPKERNEL instruction is reached. As result it's 25% + ;; slower than expected... MVK 4,B0 ; N, RILC || MVK 4,A0 ; M, outer loop counter || MV ARG1,A5 ; copy ap -- 2.40.0