diff options
| author | Andy Polyakov <appro@openssl.org> | 2015-12-21 14:26:12 +0100 |
|---|---|---|
| committer | Andy Polyakov <appro@openssl.org> | 2015-12-22 10:30:03 +0100 |
| commit | b859d70d4a04db2e3b39b5c86cb6750d3c5b9593 (patch) | |
| tree | 8e739dc3aeb492fc986b3e70ae7e29dded9011a2 /crypto/bn/asm | |
| parent | cfe670732b63b875054aabd965a7bcecc6508657 (diff) | |
| download | openssl-b859d70d4a04db2e3b39b5c86cb6750d3c5b9593.tar.gz | |
bn/asm/bn-c64xplus.asm: update commentary.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Diffstat (limited to 'crypto/bn/asm')
| -rw-r--r-- | crypto/bn/asm/bn-c64xplus.asm | 5 |
1 files 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 |