WIP

Makefile

@@ -12,7 +12,7 @@ OPTS         ?=
 OPTS_TAGS    ?= -tags=noasm
 NOASM        ?=
 # -run="NonExistent" is set to make sure tests are not run before benchmarking
-BENCH_OPTS   ?= -bench=. -run="NonExistent"
+BENCH_OPTS   ?= -bench=. -run="NonExistent" -benchmem
 # whether to be verbose
 V            ?= 1
 
@@ -22,7 +22,7 @@ endif
 
 ifeq ($(V),1)
 	OPTS += -v              # Be verbose
-	BENCH_OPTS += -gcflags=-m     # Show results from inlining
+	BENCH_OPTS += -gcflags="-m -m"    # Show results from inlining
 endif
 
 all: test

p503/arith_decl.go

@@ -41,7 +41,13 @@ 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.HasBMI2 {
+		mulWithMULXADX(z, x, y)
+	} else {
+		mul(z, x, y)
+	}
+}
 
 // Mul implementattion for legacy CPUs
 //go:noescape
@@ -58,7 +64,17 @@ 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.HasBMI2 {
+		if cpu.HasADX {
+			redcWithMULXADX(z, x)
+		} else {
+			redcWithMULX(z, x)
+		}
+	} else {
+		redc(z, x)
+	}
+}
 
 func redc(z *FpElement, x *FpElementX2)
 
@@ -76,14 +92,14 @@ func redcWithMULXADX(z *FpElement, x *FpElementX2)
 func init() {
 	if cpu.HasBMI2 {
 		if cpu.HasADX {
-			fp503Mul = mulWithMULXADX
-			fp503MontgomeryReduce = redcWithMULXADX
+			//fp503Mul = mulWithMULXADX
+			//fp503MontgomeryReduce = redcWithMULXADX
 		} else {
-			fp503Mul = mulWithMULX
-			fp503MontgomeryReduce = redcWithMULX
+			//fp503Mul = mulWithMULX
+			//fp503MontgomeryReduce = redcWithMULX
 		}
 	} else {
-		fp503Mul = mul
-		fp503MontgomeryReduce = redc
+		//fp503Mul = mul
+		//fp503MontgomeryReduce = redc
 	}
 }