mirror/gmsm
1
0
Fork 0
mirror of https://github.com/emmansun/gmsm.git synced 2025-04-26 12:16:20 +08:00
Code Issues Packages Projects Releases Wiki Activity

sm9/bn256: use ADCX ADOX together with MULX #132

Browse Source
This commit is contained in:
Sun Yimin 2023-06-28 17:38:05 +08:00 committed by GitHub
parent ce489e2b4b
commit 0afaeb49eb
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 228 additions and 202 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

391
sm9/bn256/gfp_amd64.s
View File

@ -136,198 +136,192 @@ TEXT ·gfpSub(SB),0,$0-24
RET
TEXT ·gfpMul(SB),0,$0-24
MOVQ res+0(FP), res_ptr
MOVQ in1+8(FP), x_ptr
MOVQ in2+16(FP), y_ptr
CMPB ·hasBMI2(SB), $0
JE nobmi2Mul
CMPB ·supportADX(SB), $0
JE noAdxMul
XORQ acc5, acc5
XORQ res_ptr, res_ptr
// x * y[0]
MOVQ (8*0)(y_ptr), DX
MULXQ (8*0)(x_ptr), acc0, acc1
MULXQ (8*1)(x_ptr), AX, acc2
ADDQ AX, acc1
ADCQ $0, acc2
ADCXQ AX, acc1
MULXQ (8*2)(x_ptr), AX, acc3
ADDQ AX, acc2
ADCQ $0, acc3
ADCXQ AX, acc2
MULXQ (8*3)(x_ptr), AX, acc4
ADDQ AX, acc3
ADCQ $0, acc4
XORQ acc5, acc5
ADCXQ AX, acc3
ADCXQ acc5, acc4
// First reduction step
MOVQ acc0, DX
MULXQ ·np+0x00(SB), DX, AX
MULXQ ·p2+0x00(SB), AX, t1
ADDQ AX, acc0
ADCQ t1, acc1
MULXQ ·p2+0x00(SB), AX, t0
ADOXQ AX, acc0
MULXQ ·p2+0x08(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc1
ADCQ t1, acc2
ADCXQ t0, AX
ADOXQ AX, acc1
MULXQ ·p2+0x10(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc2
ADCQ t1, acc3
MULXQ ·p2+0x10(SB), AX, t0
ADCXQ t1, AX
ADOXQ AX, acc2
MULXQ ·p2+0x18(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc3
ADCQ t1, acc4
ADCQ $0, acc5
ADCXQ t0, AX
ADOXQ AX, acc3
ADCXQ res_ptr, t1
ADOXQ t1, acc4
ADOXQ res_ptr, acc5
XORQ acc0, acc0
// x * y[1]
MOVQ (8*1)(y_ptr), DX
MULXQ (8*0)(x_ptr), AX, t1
ADDQ AX, acc1
ADCQ t1, acc2
MULXQ (8*0)(x_ptr), AX, t0
ADOXQ AX, acc1
MULXQ (8*1)(x_ptr), AX, t1
ADCQ $0, t1
ADDQ AX, acc2
ADCQ t1, acc3
ADCXQ t0, AX
ADOXQ AX, acc2
MULXQ (8*2)(x_ptr), AX, t1
ADCQ $0, t1
ADDQ AX, acc3
ADCQ t1, acc4
MULXQ (8*2)(x_ptr), AX, t0
ADCXQ t1, AX
ADOXQ AX, acc3
MULXQ (8*3)(x_ptr), AX, t1
ADCQ $0, t1
ADDQ AX, acc4
ADCQ t1, acc5
ADCQ $0, acc0
MULXQ (8*3)(x_ptr), AX, t1
ADCXQ t0, AX
ADOXQ AX, acc4
ADCXQ acc0, t1
ADOXQ t1, acc5
ADOXQ res_ptr, acc0
// Second reduction step
MOVQ acc1, DX
MULXQ ·np+0x00(SB), DX, AX
MULXQ ·p2+0x00(SB), AX, t1
ADDQ AX, acc1
ADCQ t1, acc2
MULXQ ·p2+0x00(SB), AX, t0
ADOXQ AX, acc1
MULXQ ·p2+0x08(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc2
ADCQ t1, acc3
ADCXQ t0, AX
ADOXQ AX, acc2
MULXQ ·p2+0x10(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc3
ADCQ t1, acc4
MULXQ ·p2+0x10(SB), AX, t0
ADCXQ t1, AX
ADOXQ AX, acc3
MULXQ ·p2+0x18(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc4
ADCQ t1, acc5
ADCQ $0, acc0
ADCXQ t0, AX
ADOXQ AX, acc4
ADCXQ res_ptr, t1
ADOXQ t1, acc5
ADOXQ res_ptr, acc0
XORQ acc1, acc1
// x * y[2]
MOVQ (8*2)(y_ptr), DX
MULXQ (8*0)(x_ptr), AX, t1
ADDQ AX, acc2
ADCQ t1, acc3
MULXQ (8*0)(x_ptr), AX, t0
ADOXQ AX, acc2
MULXQ (8*1)(x_ptr), AX, t1
ADCQ $0, t1
ADDQ AX, acc3
ADCQ t1, acc4
ADCXQ t0, AX
ADOXQ AX, acc3
MULXQ (8*2)(x_ptr), AX, t1
ADCQ $0, t1
ADDQ AX, acc4
ADCQ t1, acc5
MULXQ (8*2)(x_ptr), AX, t0
ADCXQ t1, AX
ADOXQ AX, acc4
MULXQ (8*3)(x_ptr), AX, t1
ADCQ $0, t1
ADDQ AX, acc5
ADCQ t1, acc0
ADCQ $0, acc1
MULXQ (8*3)(x_ptr), AX, t1
ADCXQ t0, AX
ADOXQ AX, acc5
ADCXQ res_ptr, t1
ADOXQ t1, acc0
ADOXQ res_ptr, acc1
// Third reduction step
MOVQ acc2, DX
MULXQ ·np+0x00(SB), DX, AX
MULXQ ·p2+0x00(SB), AX, t1
ADDQ AX, acc2
ADCQ t1, acc3
MULXQ ·p2+0x00(SB), AX, t0
ADOXQ AX, acc2
MULXQ ·p2+0x08(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc3
ADCQ t1, acc4
ADCXQ t0, AX
ADOXQ AX, acc3
MULXQ ·p2+0x10(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc4
ADCQ t1, acc5
MULXQ ·p2+0x10(SB), AX, t0
ADCXQ t1, AX
ADOXQ AX, acc4
MULXQ ·p2+0x18(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc5
ADCQ t1, acc0
ADCQ $0, acc1
ADCXQ t0, AX
ADOXQ AX, acc5
ADCXQ res_ptr, t1
ADOXQ t1, acc0
ADOXQ res_ptr, acc1
XORQ acc2, acc2
// x * y[3]
MOVQ (8*3)(y_ptr), DX
MULXQ (8*0)(x_ptr), AX, t1
ADDQ AX, acc3
ADCQ t1, acc4
MULXQ (8*0)(x_ptr), AX, t0
ADOXQ AX, acc3
MULXQ (8*1)(x_ptr), AX, t1
ADCQ $0, t1
ADDQ AX, acc4
ADCQ t1, acc5
ADCXQ t0, AX
ADOXQ AX, acc4
MULXQ (8*2)(x_ptr), AX, t1
ADCQ $0, t1
ADDQ AX, acc5
ADCQ t1, acc0
MULXQ (8*2)(x_ptr), AX, t0
ADCXQ t1, AX
ADOXQ AX, acc5
MULXQ (8*3)(x_ptr), AX, t1
ADCQ $0, t1
ADDQ AX, acc0
ADCQ t1, acc1
ADCQ $0, acc2
MULXQ (8*3)(x_ptr), AX, t1
ADCXQ t0, AX
ADOXQ AX, acc0
ADCXQ res_ptr, t1
ADOXQ t1, acc1
ADOXQ res_ptr, acc2
// Last reduction step
MOVQ acc3, DX
MULXQ ·np+0x00(SB), DX, AX
MULXQ ·p2+0x00(SB), AX, t1
ADDQ AX, acc3
ADCQ t1, acc4
MULXQ ·p2+0x00(SB), AX, t0
ADOXQ AX, acc3
MULXQ ·p2+0x08(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc4
ADCQ t1, acc5
ADCXQ t0, AX
ADOXQ AX, acc4
MULXQ ·p2+0x10(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc5
ADCQ t1, acc0
MULXQ ·p2+0x10(SB), AX, t0
ADCXQ t1, AX
ADOXQ AX, acc5
MULXQ ·p2+0x18(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc0
ADCQ t1, acc1
ADCQ $0, acc2
ADCXQ t0, AX
ADOXQ AX, acc0
ADCXQ res_ptr, t1
ADOXQ t1, acc1
ADOXQ res_ptr, acc2
// Copy result [255:0]
gfpCarry(acc4,acc5,acc0,acc1, x_ptr,acc3,t0,t1,acc2)
MOVQ res+0(FP), res_ptr
storeBlock(acc4,acc5,acc0,acc1, 0(res_ptr))
RET
nobmi2Mul:
noAdxMul:
// x * y[0]
MOVQ (8*0)(y_ptr), t0
@ -588,6 +582,7 @@ nobmi2Mul:
ADCQ $0, acc2
// Copy result [255:0]
gfpCarry(acc4,acc5,acc0,acc1, x_ptr,acc3,t0,t1,acc2)
MOVQ res+0(FP), res_ptr
storeBlock(acc4,acc5,acc0,acc1, 0(res_ptr))
RET
@ -598,175 +593,171 @@ TEXT ·gfpSqr(SB),NOSPLIT,$0
MOVQ in+8(FP), x_ptr
MOVQ n+16(FP), BX
CMPB ·hasBMI2(SB), $0
CMPB ·supportADX(SB), $0
JE gfpSqrLoop
gfpSqrLoopBMI2:
gfpSqrLoopAdx:
XORQ acc0, acc0
XORQ y_ptr, y_ptr
// y[1:] * y[0]
MOVQ (8*0)(x_ptr), DX
MULXQ (8*1)(x_ptr), acc1, acc2
MULXQ (8*2)(x_ptr), AX, acc3
ADDQ AX, acc2
ADCQ $0, acc3
ADOXQ AX, acc2
MULXQ (8*3)(x_ptr), AX, acc4
ADDQ AX, acc3
ADCQ $0, acc4
ADOXQ AX, acc3
ADOXQ y_ptr, acc4
// y[2:] * y[1]
MOVQ (8*1)(x_ptr), DX
MULXQ (8*2)(x_ptr), AX, t1
ADDQ AX, acc3
ADCQ t1, acc4
ADOXQ AX, acc3
MULXQ (8*3)(x_ptr), AX, acc5
ADCQ $0, acc5
ADDQ AX, acc4
ADCQ $0, acc5
ADCXQ t1, AX
ADOXQ AX, acc4
ADCXQ y_ptr, acc5
// y[3] * y[2]
MOVQ (8*2)(x_ptr), DX
MULXQ (8*3)(x_ptr), AX, y_ptr
ADDQ AX, acc5
ADCQ $0, y_ptr
ADOXQ AX, acc5
ADOXQ acc0, y_ptr
XORQ t1, t1
// *2
ADDQ acc1, acc1
ADCQ acc2, acc2
ADCQ acc3, acc3
ADCQ acc4, acc4
ADCQ acc5, acc5
ADCQ y_ptr, y_ptr
ADCQ $0, t1
ADOXQ acc1, acc1
ADOXQ acc2, acc2
ADOXQ acc3, acc3
ADOXQ acc4, acc4
ADOXQ acc5, acc5
ADOXQ y_ptr, y_ptr
ADOXQ acc0, t1
// Missing products
MOVQ (8*0)(x_ptr), DX
MULXQ DX, acc0, t0
ADDQ t0, acc1
ADCXQ t0, acc1
MOVQ (8*1)(x_ptr), DX
MULXQ DX, AX, t0
ADCQ AX, acc2
ADCQ t0, acc3
ADCXQ AX, acc2
ADCXQ t0, acc3
MOVQ (8*2)(x_ptr), DX
MULXQ DX, AX, t0
ADCQ AX, acc4
ADCQ t0, acc5
ADCXQ AX, acc4
ADCXQ t0, acc5
MOVQ (8*3)(x_ptr), DX
MULXQ DX, AX, x_ptr
ADCQ AX, y_ptr
ADCQ t1, x_ptr
ADCXQ AX, y_ptr
ADCXQ t1, x_ptr
// First reduction step
MOVQ acc0, DX
MULXQ ·np+0x00(SB), DX, AX
MULXQ ·p2+0x00(SB), AX, t1
ADDQ AX, acc0 // (carry1, acc0) = acc0 + t0 * ord0
ADCQ t1, acc1
MULXQ ·p2+0x00(SB), AX, t0
ADOXQ AX, acc0 // (carry1, acc0) = acc0 + t0 * ord0
MULXQ ·p2+0x08(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc1
ADCQ t1, acc2
ADCXQ t0, AX
ADOXQ AX, acc1
MULXQ ·p2+0x10(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc2
ADCQ t1, acc3
MULXQ ·p2+0x10(SB), AX, t0
ADCXQ t1, AX
ADOXQ AX, acc2
MULXQ ·p2+0x18(SB), AX, acc0
ADCQ $0, acc0
ADDQ AX, acc3
ADCQ $0, acc0
ADCXQ t0, AX
ADOXQ AX, acc3
MOVQ $0, t0
ADCXQ t0, acc0
ADOXQ t0, acc0
// Second reduction step
MOVQ acc1, DX
MULXQ ·np+0x00(SB), DX, AX
MULXQ ·p2+0x00(SB), AX, t1
ADDQ AX, acc1
ADCQ t1, acc2
MULXQ ·p2+0x00(SB), AX, t0
ADOXQ AX, acc1
MULXQ ·p2+0x08(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc2
ADCQ t1, acc3
ADCXQ t0, AX
ADOXQ AX, acc2
MULXQ ·p2+0x10(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc3
ADCQ t1, acc0
MULXQ ·p2+0x10(SB), AX, t0
ADCXQ t1, AX
ADOXQ AX, acc3
MULXQ ·p2+0x18(SB), AX, acc1
ADCQ $0, acc1
ADDQ AX, acc0
ADCQ $0, acc1
ADCXQ t0, AX
ADOXQ AX, acc4
MOVQ $0, t0
ADCXQ t0, acc1
ADOXQ t0, acc1
// Third reduction step
MOVQ acc2, DX
MULXQ ·np+0x00(SB), DX, AX
MULXQ ·p2+0x00(SB), AX, t1
ADDQ AX, acc2
ADCQ t1, acc3
MULXQ ·p2+0x00(SB), AX, t0
ADOXQ AX, acc2
MULXQ ·p2+0x08(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc3
ADCQ t1, acc0
ADCXQ t0, AX
ADOXQ AX, acc3
MULXQ ·p2+0x10(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc0
ADCQ t1, acc1
MULXQ ·p2+0x10(SB), AX, t0
ADCXQ t1, AX
ADOXQ AX, acc0
MULXQ ·p2+0x18(SB), AX, acc2
ADCQ $0, acc2
ADDQ AX, acc1
ADCQ $0, acc2
ADCXQ t0, AX
ADOXQ AX, acc1
MOVQ $0, t0
ADCXQ t0, acc2
ADOXQ t0, acc2
// Last reduction step
MOVQ acc3, DX
MULXQ ·np+0x00(SB), DX, AX
MULXQ ·p2+0x00(SB), AX, t1
ADDQ AX, acc3
ADCQ t1, acc0
MULXQ ·p2+0x00(SB), AX, t0
ADOXQ AX, acc3
MULXQ ·p2+0x08(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc0
ADCQ t1, acc1
ADCXQ t0, AX
ADOXQ AX, acc0
MULXQ ·p2+0x10(SB), AX, t1
ADCQ $0, t1
ADDQ AX, acc1
ADCQ t1, acc2
MULXQ ·p2+0x10(SB), AX, t0
ADCXQ t1, AX
ADOXQ AX, acc1
MULXQ ·p2+0x18(SB), AX, acc3
ADCQ $0, acc3
ADDQ AX, acc2
ADCQ $0, acc3
ADCXQ t0, AX
ADOXQ AX, acc2
MOVQ $0, t0
ADCXQ t0, acc3
ADOXQ t0, acc3
XORQ t0, t0
XORQ t1, t1
// Add bits [511:256] of the sqr result
ADDQ acc4, acc0
ADCQ acc5, acc1
ADCQ y_ptr, acc2
ADCQ x_ptr, acc3
ADCQ $0, t0
ADCXQ acc4, acc0
ADCXQ acc5, acc1
ADCXQ y_ptr, acc2
ADCXQ x_ptr, acc3
ADCXQ t1, t0
gfpCarry(acc0,acc1,acc2,acc3, acc4,acc5,y_ptr,t1,t0)
storeBlock(acc0,acc1,acc2,acc3, 0(res_ptr))
MOVQ res_ptr, x_ptr
DECQ BX
JNE gfpSqrLoopBMI2
JNE gfpSqrLoopAdx
RET

5
sm9/bn256/gfp_decl.go
View File

@ -10,7 +10,10 @@ import (
"golang.org/x/sys/cpu"
)
var hasBMI2 = cpu.X86.HasBMI2
// amd64 assembly uses ADCX/ADOX/MULX if ADX is available to run two carry
// chains in the flags in parallel across the whole operation, and aggressively
// unrolls loops. arm64 processes four words at a time.
var supportADX = cpu.X86.HasADX && cpu.X86.HasBMI2
// Set c = p - a, if c == p, then c = 0
//

12
sm9/bn256/gfp_generic.go
View File

@ -134,5 +134,15 @@ func gfpSqr(res, in *gfP, n int) {
}
func gfpFromMont(res, in *gfP) {
gfpMul(res, in, &gfP{1})
var T [8]uint64
var carry uint64
copy(T[:], in[:])
for i := 0; i < 4; i++ {
Y := T[i] * np[0]
c2 := addMulVVW(T[i:4+i], p2[:], Y)
T[4+i], carry = bits.Add64(uint64(0), c2, carry)
}
*res = gfP{T[4], T[5], T[6], T[7]}
gfpCarry(res, carry)
}

22
sm9/bn256/gfp_test.go
View File

@ -48,6 +48,28 @@ func Test_gfpBasicOperations(t *testing.T) {
}
}
func Test_gfpSqr(t *testing.T) {
// p - 1
pMinusOne := new(big.Int).Sub(p, big.NewInt(1))
x := fromBigInt(pMinusOne)
ret := &gfP{}
gfpSqr(ret, x, 1)
pMinusOne.Mul(pMinusOne, pMinusOne)
pMinusOne.Mod(pMinusOne, p)
if *ret != *fromBigInt(pMinusOne) {
t.Errorf("bad sqr")
}
// p + 1
pPlusOne := new(big.Int).Add(p, big.NewInt(1))
x = fromBigInt(pPlusOne)
gfpSqr(ret, x, 1)
pPlusOne.Mul(pPlusOne, pPlusOne)
pPlusOne.Mod(pPlusOne, p)
if *ret != *fromBigInt(pPlusOne) {
t.Errorf("bad sqr")
}
}
func TestFromMont(t *testing.T) {
x := fromBigInt(bigFromHex("85AEF3D078640C98597B6027B441A01FF1DD2C190F5E93C454806C11D8806141"))
ret1, ret2 := &gfP{}, &gfP{}