From 846005eeda4b1355a84217943e1886e31859dfd1 Mon Sep 17 00:00:00 2001
From: Sun Yimin <emmansun@users.noreply.github.com>
Date: Tue, 11 Jul 2023 15:21:24 +0800
Subject: [PATCH] =?UTF-8?q?Updated=20SM9=E5=AE=9E=E7=8E=B0=E5=8F=8A?=
 =?UTF-8?q?=E4=BC=98=E5=8C=96=20(markdown)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 SM9实现及优化.md | 31 +++++++++++++++++++++++++++++++
 1 file changed, 31 insertions(+)

diff --git a/SM9实现及优化.md b/SM9实现及优化.md
index fb1bc1f..6091934 100644
--- a/SM9实现及优化.md
+++ b/SM9实现及优化.md
@@ -41,6 +41,37 @@ Go语言相对简单，但是为了简单，编译器做了很多额外的操作
 ## 应用SIMD复制值
 也就是Set操作的汇编实现，同时也尽量减少Set操作（这个“优化”导致了实现的复杂性、影响了代码的可维护性，可能不值得）。
 
+## Neg改用Sub实现
+无意中发现Neg方法不如后来实现的Sub性能好，这个挺奇怪的，单独测试，gfpNeg性能(BenchmarkGfPNeg-6)要比gfpSub()性能好(BenchmarkGfPNeg2-6)：
+```
+goos: windows
+goarch: amd64
+pkg: github.com/emmansun/gmsm/sm9/bn256
+cpu: Intel(R) Core(TM) i5-9500 CPU @ 3.00GHz
+BenchmarkGfPNeg-6   	349538827	         3.399 ns/op	       0 B/op	       0 allocs/op
+BenchmarkGfPNeg2-6   	282038318	         4.208 ns/op	       0 B/op	       0 allocs/op
+```
+
+但是应用到gfP2的MulUNC方法：
+gfpNeg
+```
+goos: windows
+goarch: amd64
+pkg: github.com/emmansun/gmsm/sm9/bn256
+cpu: Intel(R) Core(TM) i5-9500 CPU @ 3.00GHz
+BenchmarkGfP2MulU-6   	 8290990	       141.1 ns/op	      64 B/op	       1 allocs/op
+BenchmarkGfP2SquareU-6   	10009350	       117.0 ns/op	      64 B/op	       1 allocs/op
+```
+gfpSub
+```
+goos: windows
+goarch: amd64
+pkg: github.com/emmansun/gmsm/sm9/bn256
+cpu: Intel(R) Core(TM) i5-9500 CPU @ 3.00GHz
+BenchmarkGfP2MulU-6   	12727611	        92.70 ns/op	       0 B/op	       0 allocs/op
+BenchmarkGfP2SquareU-6   	17728008	        66.35 ns/op	       0 B/op	       0 allocs/op
+```
+
 ## 下一步
 * 参考《New software speed records for cryptographic pairings》使用浮点运算和SIMD实现？
 * [High-Speed Software Implementation of the Optimal Ate Pairing over Barreto–Naehrig Curves](https://eprint.iacr.org/2010/354.pdf)，平方扩域上的运算优化，不过由于他的p选择，有其特殊性。