From 9fc872439c811ab0a05f2172d33be6287bcaf7c1 Mon Sep 17 00:00:00 2001
From: Kris Kwiatkowski <kris@cloudflare.com>
Date: Wed, 3 Oct 2018 11:54:17 +0100
Subject: [PATCH] PERF

---
 p503/arith_decl.go | 37 ++++++++++++++++++-------------------
 p751/arith_decl.go | 24 +++++++++---------------
 2 files changed, 27 insertions(+), 34 deletions(-)

diff --git a/p503/arith_decl.go b/p503/arith_decl.go
index 0129559..a3e3f59 100644
--- a/p503/arith_decl.go
+++ b/p503/arith_decl.go
@@ -41,7 +41,15 @@ func fp503StrongReduce(x *FpElement)
 // Concrete implementation depends on capabilities of the CPU which
 // are resolved at runtime. CPUs with ADCX, ADOX and MULX support
 // run most optimized implementation
-var fp503Mul func(z *FpElementX2, x, y *FpElement)
+func fp503Mul(z *FpElementX2, x, y *FpElement) {
+	if cpu.X86.HasBMI2 && cpu.X86.HasADX {
+		mulWithMULXADX(z,x,y)
+	} else if cpu.X86.HasBMI2 {
+		mulWithMULX(z,x,y)
+	} else {
+		mul(z,x,y)
+	}
+}
 
 // Mul implementattion for legacy CPUs
 //go:noescape
@@ -58,7 +66,15 @@ func mulWithMULXADX(z *FpElementX2, x, y *FpElement)
 
 // Computes the Montgomery reduction z = x R^{-1} (mod 2*p). On return value
 // of x may be changed. z=x not allowed.
-var fp503MontgomeryReduce func(z *FpElement, x *FpElementX2)
+func fp503MontgomeryReduce(z *FpElement, x *FpElementX2) {
+	if cpu.X86.HasBMI2 && cpu.X86.HasADX {
+		redcWithMULXADX(z,x)
+	} else if cpu.X86.HasBMI2 {
+		redcWithMULX(z,x)
+	} else {
+		redc(z,x)
+	}
+}
 
 func redc(z *FpElement, x *FpElementX2)
 
@@ -70,20 +86,3 @@ func redcWithMULX(z *FpElement, x *FpElementX2)
 // (ADOX/ADCX) instructions and carry-less MULX multiplier
 //go:noescape
 func redcWithMULXADX(z *FpElement, x *FpElementX2)
-
-// On initialization, set the fp503Mul function pointer to the
-// fastest implementation depending on CPU capabilities.
-func init() {
-	if cpu.X86.HasBMI2 {
-		if cpu.X86.HasADX {
-			fp503Mul = mulWithMULXADX
-			fp503MontgomeryReduce = redcWithMULXADX
-		} else {
-			fp503Mul = mulWithMULX
-			fp503MontgomeryReduce = redcWithMULX
-		}
-	} else {
-		fp503Mul = mul
-		fp503MontgomeryReduce = redc
-	}
-}
diff --git a/p751/arith_decl.go b/p751/arith_decl.go
index d56f73b..35e46e2 100644
--- a/p751/arith_decl.go
+++ b/p751/arith_decl.go
@@ -41,7 +41,15 @@ func fp751Mul(z *FpElementX2, x, y *FpElement)
 // fp751MontgomeryReduce implementations below.
 // When set, it performs Montgomery reduction: set z = x R^{-1} (mod 2*p).
 // It may destroy the input value.
-var fp751MontgomeryReduce func(z *FpElement, x *FpElementX2)
+func fp751MontgomeryReduce(z *FpElement, x *FpElementX2) {
+	if cpu.X86.HasBMI2 && cpu.X86.HasADX {
+		fp751MontgomeryReduceBMI2ADX(z,x)
+	} else if cpu.X86.HasBMI2 {
+		fp751MontgomeryReduceBMI2(z,x)
+	} else {
+		fp751MontgomeryReduceFallback(z,x)
+	}
+}
 
 //go:noescape
 func fp751MontgomeryReduceBMI2ADX(z *FpElement, x *FpElementX2)
@@ -55,17 +63,3 @@ func fp751MontgomeryReduceFallback(z *FpElement, x *FpElementX2)
 // Reduce a field element in [0, 2*p) to one in [0,p).
 //go:noescape
 func fp751StrongReduce(x *FpElement)
-
-// On initialization, set the fp751MontgomeryReduce function pointer to the
-// fastest implementation depending on CPU capabilities.
-func init() {
-	if cpu.X86.HasBMI2 {
-		if cpu.X86.HasADX {
-			fp751MontgomeryReduce = fp751MontgomeryReduceBMI2ADX
-		} else {
-			fp751MontgomeryReduce = fp751MontgomeryReduceBMI2
-		}
-	} else {
-		fp751MontgomeryReduce = fp751MontgomeryReduceFallback
-	}
-}