Merge pull request #16 from nspcc-dev/feat/cpuid

Move cpu id to a separate package

go.mod

@@ -1,3 +1,5 @@
 module github.com/nspcc-dev/tzhash
 
+go 1.12
+
 require github.com/stretchr/testify v1.3.0

internal/cpuid/cpuid_x86.go

@@ -28,7 +28,7 @@
 
 // +build 386 amd64 amd64p32
 
-package tz
+package cpuid
 
 const (
 	bitOSXSAVE = 1 << 27
@@ -36,14 +36,19 @@ const (
 	bitAVX2    = 1 << 5
 )
 
-func setFeatures() {
+var (
+	hasAVX  bool
+	hasAVX2 bool
+)
+
+func init() {
 	maxID, _, _, _ := cpuid(0, 0)
 	if maxID < 1 {
 		return
 	}
 
 	_, _, ecx1, _ := cpuid(1, 0)
-	hasOSXSAVE = isSet(ecx1, bitOSXSAVE)
+	hasOSXSAVE := isSet(ecx1, bitOSXSAVE)
 
 	osSupportsAVX := false
 	if hasOSXSAVE {
@@ -61,6 +66,9 @@ func setFeatures() {
 	hasAVX2 = isSet(ebx7, bitAVX2) && osSupportsAVX
 }
 
+func HasAVX() bool  { return hasAVX }
+func HasAVX2() bool { return hasAVX2 }
+
 func isSet(hwc uint32, value uint32) bool {
 	return hwc&value != 0
 }

tz/avx2_inline_amd64.s

@@ -1,5 +1,28 @@
 #include "textflag.h"
 
+#define mask(bit, src, tmp, to1, to2) \
+    MOVQ src, tmp \
+    SHRQ bit, tmp \
+    ANDQ $1, tmp  \
+    NEGQ tmp      \
+    MOVQ tmp, to1 \
+    VPBROADCASTB to1, to2
+
+#define mulBit(bit) \
+    VPSLLQ $1, Y0, Y1 \
+    VPALIGNR $8, Y1, Y0, Y2 \
+    VPSRLQ $63, Y2, Y2 \
+    VPXOR Y1, Y2, Y2 \
+    VPSRLQ $63, Y1, Y3 \
+    VPSLLQ $63, Y3, Y3 \
+    VPUNPCKHQDQ Y3, Y3, Y3 \
+    VPXOR Y2, Y3, Y3 \
+    mask(bit, CX, DX, X1, Y2) \
+    VPXOR Y3, Y8, Y3 \
+    VPAND Y3, Y2, Y4 \
+    VPXOR Y4, Y0, Y8 \
+    VMOVDQA Y3, Y0
+
 // func mulByteRightx2(c00c10, c01c11 *[4]uint64, b byte)
 TEXT ·mulByteRightx2(SB),NOSPLIT,$0
     MOVQ c00c10+0(FP), AX
@@ -8,179 +31,16 @@ TEXT ·mulByteRightx2(SB),NOSPLIT,$0
     VMOVDQA (BX), Y8
     MOVB b+16(FP), CX
 
-    // 1 bit
+    mulBit($7)
-    VPSLLQ $1, Y0, Y1
+    mulBit($6)
-    VPALIGNR $8, Y1, Y0, Y2
+    mulBit($5)
-    VPSRLQ $63, Y2, Y2
+    mulBit($4)
-    VPXOR Y1, Y2, Y2
+    mulBit($3)
-    VPSRLQ $63, Y1, Y3
+    mulBit($2)
-    VPSLLQ $63, Y3, Y3
+    mulBit($1)
-    VPUNPCKHQDQ Y3, Y3, Y3
+    mulBit($0)
-    VPXOR Y2, Y3, Y3
 
-    MOVQ CX, DX
-    SHRQ $7, DX
-    ANDQ $1, DX
-    NEGQ DX
-    MOVQ DX, X1
-    VPBROADCASTB X1, Y2
-
-    VPXOR Y3, Y8, Y3
-    VPAND Y3, Y2, Y4
-    VPXOR Y4, Y0, Y8
-    VMOVDQA Y3, Y0
-
-    // 2 bit
-    VPSLLQ $1, Y0, Y1
-    VPALIGNR $8, Y1, Y0, Y2
-    VPSRLQ $63, Y2, Y2
-    VPXOR Y1, Y2, Y2
-    VPSRLQ $63, Y1, Y3
-    VPSLLQ $63, Y3, Y3
-    VPUNPCKHQDQ Y3, Y3, Y3
-    VPXOR Y2, Y3, Y3
-
-    MOVQ CX, DX
-    SHRQ $6, DX
-    ANDQ $1, DX
-    NEGQ DX
-    MOVQ DX, X1
-    VPBROADCASTB X1, Y2
-
-    VPXOR Y3, Y8, Y3
-    VPAND Y3, Y2, Y4
-    VPXOR Y4, Y0, Y8
-    VMOVDQA Y3, Y0
-
-    // 3 bit
-    VPSLLQ $1, Y0, Y1
-    VPALIGNR $8, Y1, Y0, Y2
-    VPSRLQ $63, Y2, Y2
-    VPXOR Y1, Y2, Y2
-    VPSRLQ $63, Y1, Y3
-    VPSLLQ $63, Y3, Y3
-    VPUNPCKHQDQ Y3, Y3, Y3
-    VPXOR Y2, Y3, Y3
-
-    MOVQ CX, DX
-    SHRQ $5, DX
-    ANDQ $1, DX
-    NEGQ DX
-    MOVQ DX, X1
-    VPBROADCASTB X1, Y2
-
-    VPXOR Y3, Y8, Y3
-    VPAND Y3, Y2, Y4
-    VPXOR Y4, Y0, Y8
-    VMOVDQA Y3, Y0
-
-    // 4 bit
-    VPSLLQ $1, Y0, Y1
-    VPALIGNR $8, Y1, Y0, Y2
-    VPSRLQ $63, Y2, Y2
-    VPXOR Y1, Y2, Y2
-    VPSRLQ $63, Y1, Y3
-    VPSLLQ $63, Y3, Y3
-    VPUNPCKHQDQ Y3, Y3, Y3
-    VPXOR Y2, Y3, Y3
-
-    MOVQ CX, DX
-    SHRQ $4, DX
-    ANDQ $1, DX
-    NEGQ DX
-    MOVQ DX, X1
-    VPBROADCASTB X1, Y2
-
-    VPXOR Y3, Y8, Y3
-    VPAND Y3, Y2, Y4
-    VPXOR Y4, Y0, Y8
-    VMOVDQA Y3, Y0
-
-    // 5 bit
-    VPSLLQ $1, Y0, Y1
-    VPALIGNR $8, Y1, Y0, Y2
-    VPSRLQ $63, Y2, Y2
-    VPXOR Y1, Y2, Y2
-    VPSRLQ $63, Y1, Y3
-    VPSLLQ $63, Y3, Y3
-    VPUNPCKHQDQ Y3, Y3, Y3
-    VPXOR Y2, Y3, Y3
-
-    MOVQ CX, DX
-    SHRQ $3, DX
-    ANDQ $1, DX
-    NEGQ DX
-    MOVQ DX, X1
-    VPBROADCASTB X1, Y2
-
-    VPXOR Y3, Y8, Y3
-    VPAND Y3, Y2, Y4
-    VPXOR Y4, Y0, Y8
-    VMOVDQA Y3, Y0
-
-    // 6 bit
-    VPSLLQ $1, Y0, Y1
-    VPALIGNR $8, Y1, Y0, Y2
-    VPSRLQ $63, Y2, Y2
-    VPXOR Y1, Y2, Y2
-    VPSRLQ $63, Y1, Y3
-    VPSLLQ $63, Y3, Y3
-    VPUNPCKHQDQ Y3, Y3, Y3
-    VPXOR Y2, Y3, Y3
-
-    MOVQ CX, DX
-    SHRQ $2, DX
-    ANDQ $1, DX
-    NEGQ DX
-    MOVQ DX, X1
-    VPBROADCASTB X1, Y2
-
-    VPXOR Y3, Y8, Y3
-    VPAND Y3, Y2, Y4
-    VPXOR Y4, Y0, Y8
-    VMOVDQA Y3, Y0
-
-    // 7 bit
-    VPSLLQ $1, Y0, Y1
-    VPALIGNR $8, Y1, Y0, Y2
-    VPSRLQ $63, Y2, Y2
-    VPXOR Y1, Y2, Y2
-    VPSRLQ $63, Y1, Y3
-    VPSLLQ $63, Y3, Y3
-    VPUNPCKHQDQ Y3, Y3, Y3
-    VPXOR Y2, Y3, Y3
-
-    MOVQ CX, DX
-    SHRQ $1, DX
-    ANDQ $1, DX
-    NEGQ DX
-    MOVQ DX, X1
-    VPBROADCASTB X1, Y2
-
-    VPXOR Y3, Y8, Y3
-    VPAND Y3, Y2, Y4
-    VPXOR Y4, Y0, Y8
-    VMOVDQA Y3, Y0
-
-    // 8 bit
-    VPSLLQ $1, Y0, Y1
-    VPALIGNR $8, Y1, Y0, Y2
-    VPSRLQ $63, Y2, Y2
-    VPXOR Y1, Y2, Y2
-    VPSRLQ $63, Y1, Y3
-    VPSLLQ $63, Y3, Y3
-    VPUNPCKHQDQ Y3, Y3, Y3
-    VPXOR Y2, Y3, Y3
-
-    MOVQ CX, DX
-    ANDQ $1, DX
-    NEGQ DX
-    MOVQ DX, X1
-    VPBROADCASTB X1, Y2
-
-    VPXOR Y3, Y8, Y3
-    VPAND Y3, Y2, Y4
-    VPXOR Y4, Y0, Y8
     VMOVDQA Y8, (BX)
-    VMOVDQA Y3, (AX)
+    VMOVDQA Y0, (AX)
+
     RET

tz/hash.go

@@ -6,6 +6,8 @@ package tz
 import (
 	"errors"
 	"hash"
+
+	"github.com/nspcc-dev/tzhash/internal/cpuid"
 )
 
 type Implementation int
@@ -22,9 +24,10 @@ const (
 )
 
 var (
-	hasAVX     bool
+	hasAVX = cpuid.HasAVX()
-	hasAVX2    bool
+	// Having AVX2 does not guarantee
-	hasOSXSAVE bool
+	// that AVX is also present.
+	hasAVX2 = cpuid.HasAVX2() && hasAVX
 )
 
 func (impl Implementation) String() string {

tz/sl2.go

@@ -17,8 +17,6 @@ var id = sl2{
 var mul func(a, b, c *sl2, x *[4]gf127.GF127)
 
 func init() {
-	setFeatures()
-
 	if hasAVX {
 		mul = mulSL2AVX
 	} else {