P2Pool/edwards25519
P2Pool/edwards25519
1
0
Fork 0
Code Issues Pull requests Releases Activity

asm: upgrade avo and regenerate fe_amd64.s

Browse source 
Fixes #16
This commit is contained in:
Filippo Valsorda 2021-04-19 00:45:35 -04:00
parent c6be47d677
commit 8e7780424d
3 changed files with 219 additions and 216 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
asm/go.mod
View file

@ -4,7 +4,7 @@ go 1.16
require (
filippo.io/edwards25519 v0.0.0
github.com/mmcloughlin/avo v0.0.0-20210104032911-599bdd1269f4
github.com/mmcloughlin/avo v0.2.0
)
replace filippo.io/edwards25519 => ../

13
asm/go.sum
View file

@ -1,7 +1,7 @@
github.com/mmcloughlin/avo v0.0.0-20210104032911-599bdd1269f4 h1:ExoghBBFY7A3RzgkAOq0XbHs9zaT/bHq7xysgyp3z3Q=
github.com/mmcloughlin/avo v0.0.0-20210104032911-599bdd1269f4/go.mod h1:6aKT4zZIrpGqB3RpFU14ByCSSyKY6LfJz4J/JJChHfI=
github.com/mmcloughlin/avo v0.2.0 h1:6vhoSaKtxb6f4RiH+LK2qL6GSMpFzhEwJYTTSZNy09w=
github.com/mmcloughlin/avo v0.2.0/go.mod h1:5tidO2Z9Z7N6X7UMcGg+1KTj51O8OxYDCMHxCZTVpEA=
github.com/yuin/goldmark v1.2.1/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74=
golang.org/x/arch v0.0.0-20201008161808-52c3e6f60cff/go.mod h1:flIaEI6LNU6xOCD5PaJvn9wGP0agmIOqjrtsKGRguv4=
golang.org/x/arch v0.0.0-20210405154355-08b684f594a5/go.mod h1:flIaEI6LNU6xOCD5PaJvn9wGP0agmIOqjrtsKGRguv4=
golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
golang.org/x/crypto v0.0.0-20191011191535-87dc89f01550/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
golang.org/x/crypto v0.0.0-20200622213623-75b288015ac9/go.mod h1:LzIPMQfyMNhhGPhUkYOs5KpL4U8rLKemX1yGLhDgUto=
@ -15,12 +15,15 @@ golang.org/x/sync v0.0.0-20201020160332-67f06af15bc9/go.mod h1:RxMgew5VJxzue5/jJ
golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
golang.org/x/sys v0.0.0-20190412213103-97732733099d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
golang.org/x/sys v0.0.0-20200930185726-fdedc70b468f/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
golang.org/x/sys v0.0.0-20210119212857-b64e53b001e4/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
golang.org/x/sys v0.0.0-20210403161142-5e06dd20ab57 h1:F5Gozwx4I1xtr/sr/8CFbb57iKi3297KFs0QDbGN60A=
golang.org/x/sys v0.0.0-20210403161142-5e06dd20ab57/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
golang.org/x/tools v0.0.0-20201105001634-bc3cf281b174 h1:0rx0F4EjJNbxTuzWe0KjKcIzs+3VEb/Mrs/d1ciNz1c=
golang.org/x/tools v0.0.0-20201105001634-bc3cf281b174/go.mod h1:emZCQorbCU4vsT4fOWvOPXz4eW1wZW4PmDk9uLelYpA=
golang.org/x/tools v0.1.0 h1:po9/4sTYwZU9lPhi1tOrb4hCv3qrhiQ77LZfGa2OjwY=
golang.org/x/tools v0.1.0/go.mod h1:xkSsbof2nBLbhDlRMhhhyNLN/zl3eTqcnHD5viDpcZ0=
golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
golang.org/x/xerrors v0.0.0-20200804184101-5ec99f83aff1 h1:go1bK/D/BFZV2I8cIQd1NKEZ+0owSTG1fDTci4IqFcE=

420
fe_amd64.s
View file

@ -12,166 +12,323 @@ TEXT ·feMul(SB), NOSPLIT, $0-24
// r0 = a0×b0
MOVQ (CX), AX
MULQ (BX)
MOVQ AX, SI
MOVQ DX, BP
MOVQ AX, DI
MOVQ DX, SI
// r0 += 19×a1×b4
MOVQ 8(CX), AX
IMUL3Q $0x13, AX, AX
MULQ 32(BX)
ADDQ AX, SI
ADCQ DX, BP
ADDQ AX, DI
ADCQ DX, SI
// r0 += 19×a2×b3
MOVQ 16(CX), AX
IMUL3Q $0x13, AX, AX
MULQ 24(BX)
ADDQ AX, SI
ADCQ DX, BP
ADDQ AX, DI
ADCQ DX, SI
// r0 += 19×a3×b2
MOVQ 24(CX), AX
IMUL3Q $0x13, AX, AX
MULQ 16(BX)
ADDQ AX, SI
ADCQ DX, BP
ADDQ AX, DI
ADCQ DX, SI
// r0 += 19×a4×b1
MOVQ 32(CX), AX
IMUL3Q $0x13, AX, AX
MULQ 8(BX)
ADDQ AX, SI
ADCQ DX, BP
ADDQ AX, DI
ADCQ DX, SI
// r1 = a0×b1
MOVQ (CX), AX
MULQ 8(BX)
MOVQ AX, R8
MOVQ DX, DI
MOVQ AX, R9
MOVQ DX, R8
// r1 += a1×b0
MOVQ 8(CX), AX
MULQ (BX)
ADDQ AX, R8
ADCQ DX, DI
ADDQ AX, R9
ADCQ DX, R8
// r1 += 19×a2×b4
MOVQ 16(CX), AX
IMUL3Q $0x13, AX, AX
MULQ 32(BX)
ADDQ AX, R8
ADCQ DX, DI
ADDQ AX, R9
ADCQ DX, R8
// r1 += 19×a3×b3
MOVQ 24(CX), AX
IMUL3Q $0x13, AX, AX
MULQ 24(BX)
ADDQ AX, R8
ADCQ DX, DI
ADDQ AX, R9
ADCQ DX, R8
// r1 += 19×a4×b2
MOVQ 32(CX), AX
IMUL3Q $0x13, AX, AX
MULQ 16(BX)
ADDQ AX, R8
ADCQ DX, DI
ADDQ AX, R9
ADCQ DX, R8
// r2 = a0×b2
MOVQ (CX), AX
MULQ 16(BX)
MOVQ AX, R10
MOVQ DX, R9
MOVQ AX, R11
MOVQ DX, R10
// r2 += a1×b1
MOVQ 8(CX), AX
MULQ 8(BX)
ADDQ AX, R10
ADCQ DX, R9
ADDQ AX, R11
ADCQ DX, R10
// r2 += a2×b0
MOVQ 16(CX), AX
MULQ (BX)
ADDQ AX, R10
ADCQ DX, R9
ADDQ AX, R11
ADCQ DX, R10
// r2 += 19×a3×b4
MOVQ 24(CX), AX
IMUL3Q $0x13, AX, AX
MULQ 32(BX)
ADDQ AX, R10
ADCQ DX, R9
ADDQ AX, R11
ADCQ DX, R10
// r2 += 19×a4×b3
MOVQ 32(CX), AX
IMUL3Q $0x13, AX, AX
MULQ 24(BX)
ADDQ AX, R10
ADCQ DX, R9
ADDQ AX, R11
ADCQ DX, R10
// r3 = a0×b3
MOVQ (CX), AX
MULQ 24(BX)
MOVQ AX, R12
MOVQ DX, R11
MOVQ AX, R13
MOVQ DX, R12
// r3 += a1×b2
MOVQ 8(CX), AX
MULQ 16(BX)
ADDQ AX, R12
ADCQ DX, R11
ADDQ AX, R13
ADCQ DX, R12
// r3 += a2×b1
MOVQ 16(CX), AX
MULQ 8(BX)
ADDQ AX, R12
ADCQ DX, R11
ADDQ AX, R13
ADCQ DX, R12
// r3 += a3×b0
MOVQ 24(CX), AX
MULQ (BX)
ADDQ AX, R12
ADCQ DX, R11
ADDQ AX, R13
ADCQ DX, R12
// r3 += 19×a4×b4
MOVQ 32(CX), AX
IMUL3Q $0x13, AX, AX
MULQ 32(BX)
ADDQ AX, R12
ADCQ DX, R11
ADDQ AX, R13
ADCQ DX, R12
// r4 = a0×b4
MOVQ (CX), AX
MULQ 32(BX)
MOVQ AX, R14
MOVQ DX, R13
MOVQ AX, R15
MOVQ DX, R14
// r4 += a1×b3
MOVQ 8(CX), AX
MULQ 24(BX)
ADDQ AX, R14
ADCQ DX, R13
ADDQ AX, R15
ADCQ DX, R14
// r4 += a2×b2
MOVQ 16(CX), AX
MULQ 16(BX)
ADDQ AX, R14
ADCQ DX, R13
ADDQ AX, R15
ADCQ DX, R14
// r4 += a3×b1
MOVQ 24(CX), AX
MULQ 8(BX)
ADDQ AX, R14
ADCQ DX, R13
ADDQ AX, R15
ADCQ DX, R14
// r4 += a4×b0
MOVQ 32(CX), AX
MULQ (BX)
ADDQ AX, R15
ADCQ DX, R14
// First reduction chain
MOVQ $0x0007ffffffffffff, AX
SHLQ $0x0d, DI, SI
SHLQ $0x0d, R9, R8
SHLQ $0x0d, R11, R10
SHLQ $0x0d, R13, R12
SHLQ $0x0d, R15, R14
ANDQ AX, DI
IMUL3Q $0x13, R14, R14
ADDQ R14, DI
ANDQ AX, R9
ADDQ SI, R9
ANDQ AX, R11
ADDQ R8, R11
ANDQ AX, R13
ADDQ R10, R13
ANDQ AX, R15
ADDQ R12, R15
// Second reduction chain (carryPropagate)
MOVQ DI, SI
SHRQ $0x33, SI
MOVQ R9, R8
SHRQ $0x33, R8
MOVQ R11, R10
SHRQ $0x33, R10
MOVQ R13, R12
SHRQ $0x33, R12
MOVQ R15, R14
SHRQ $0x33, R14
ANDQ AX, DI
IMUL3Q $0x13, R14, R14
ADDQ R14, DI
ANDQ AX, R9
ADDQ SI, R9
ANDQ AX, R11
ADDQ R8, R11
ANDQ AX, R13
ADDQ R10, R13
ANDQ AX, R15
ADDQ R12, R15
// Store output
MOVQ out+0(FP), AX
MOVQ DI, (AX)
MOVQ R9, 8(AX)
MOVQ R11, 16(AX)
MOVQ R13, 24(AX)
MOVQ R15, 32(AX)
RET
// func feSquare(out *fieldElement, a *fieldElement)
TEXT ·feSquare(SB), NOSPLIT, $0-16
MOVQ a+8(FP), CX
// r0 = l0×l0
MOVQ (CX), AX
MULQ (CX)
MOVQ AX, SI
MOVQ DX, BX
// r0 += 38×l1×l4
MOVQ 8(CX), AX
IMUL3Q $0x26, AX, AX
MULQ 32(CX)
ADDQ AX, SI
ADCQ DX, BX
// r0 += 38×l2×l3
MOVQ 16(CX), AX
IMUL3Q $0x26, AX, AX
MULQ 24(CX)
ADDQ AX, SI
ADCQ DX, BX
// r1 = 2×l0×l1
MOVQ (CX), AX
SHLQ $0x01, AX
MULQ 8(CX)
MOVQ AX, R8
MOVQ DX, DI
// r1 += 38×l2×l4
MOVQ 16(CX), AX
IMUL3Q $0x26, AX, AX
MULQ 32(CX)
ADDQ AX, R8
ADCQ DX, DI
// r1 += 19×l3×l3
MOVQ 24(CX), AX
IMUL3Q $0x13, AX, AX
MULQ 24(CX)
ADDQ AX, R8
ADCQ DX, DI
// r2 = 2×l0×l2
MOVQ (CX), AX
SHLQ $0x01, AX
MULQ 16(CX)
MOVQ AX, R10
MOVQ DX, R9
// r2 += l1×l1
MOVQ 8(CX), AX
MULQ 8(CX)
ADDQ AX, R10
ADCQ DX, R9
// r2 += 38×l3×l4
MOVQ 24(CX), AX
IMUL3Q $0x26, AX, AX
MULQ 32(CX)
ADDQ AX, R10
ADCQ DX, R9
// r3 = 2×l0×l3
MOVQ (CX), AX
SHLQ $0x01, AX
MULQ 24(CX)
MOVQ AX, R12
MOVQ DX, R11
// r3 += 2×l1×l2
MOVQ 8(CX), AX
IMUL3Q $0x02, AX, AX
MULQ 16(CX)
ADDQ AX, R12
ADCQ DX, R11
// r3 += 19×l4×l4
MOVQ 32(CX), AX
IMUL3Q $0x13, AX, AX
MULQ 32(CX)
ADDQ AX, R12
ADCQ DX, R11
// r4 = 2×l0×l4
MOVQ (CX), AX
SHLQ $0x01, AX
MULQ 32(CX)
MOVQ AX, R14
MOVQ DX, R13
// r4 += 2×l1×l3
MOVQ 8(CX), AX
IMUL3Q $0x02, AX, AX
MULQ 24(CX)
ADDQ AX, R14
ADCQ DX, R13
// r4 += l2×l2
MOVQ 16(CX), AX
MULQ 16(CX)
ADDQ AX, R14
ADCQ DX, R13
// First reduction chain
MOVQ $0x0007ffffffffffff, AX
SHLQ $0x0d, SI, BP
SHLQ $0x0d, SI, BX
SHLQ $0x0d, R8, DI
SHLQ $0x0d, R10, R9
SHLQ $0x0d, R12, R11
@ -180,7 +337,7 @@ TEXT ·feMul(SB), NOSPLIT, $0-24
IMUL3Q $0x13, R13, R13
ADDQ R13, SI
ANDQ AX, R8
ADDQ BP, R8
ADDQ BX, R8
ANDQ AX, R10
ADDQ DI, R10
ANDQ AX, R12
@ -189,8 +346,8 @@ TEXT ·feMul(SB), NOSPLIT, $0-24
ADDQ R11, R14
// Second reduction chain (carryPropagate)
MOVQ SI, BP
SHRQ $0x33, BP
MOVQ SI, BX
SHRQ $0x33, BX
MOVQ R8, DI
SHRQ $0x33, DI
MOVQ R10, R9
@ -203,7 +360,7 @@ TEXT ·feMul(SB), NOSPLIT, $0-24
IMUL3Q $0x13, R13, R13
ADDQ R13, SI
ANDQ AX, R8
ADDQ BP, R8
ADDQ BX, R8
ANDQ AX, R10
ADDQ DI, R10
ANDQ AX, R12
@ -219,160 +376,3 @@ TEXT ·feMul(SB), NOSPLIT, $0-24
MOVQ R12, 24(AX)
MOVQ R14, 32(AX)
RET
// func feSquare(out *fieldElement, a *fieldElement)
TEXT ·feSquare(SB), NOSPLIT, $0-16
MOVQ a+8(FP), CX
// r0 = l0×l0
MOVQ (CX), AX
MULQ (CX)
MOVQ AX, BP
MOVQ DX, BX
// r0 += 38×l1×l4
MOVQ 8(CX), AX
IMUL3Q $0x26, AX, AX
MULQ 32(CX)
ADDQ AX, BP
ADCQ DX, BX
// r0 += 38×l2×l3
MOVQ 16(CX), AX
IMUL3Q $0x26, AX, AX
MULQ 24(CX)
ADDQ AX, BP
ADCQ DX, BX
// r1 = 2×l0×l1
MOVQ (CX), AX
SHLQ $0x01, AX
MULQ 8(CX)
MOVQ AX, DI
MOVQ DX, SI
// r1 += 38×l2×l4
MOVQ 16(CX), AX
IMUL3Q $0x26, AX, AX
MULQ 32(CX)
ADDQ AX, DI
ADCQ DX, SI
// r1 += 19×l3×l3
MOVQ 24(CX), AX
IMUL3Q $0x13, AX, AX
MULQ 24(CX)
ADDQ AX, DI
ADCQ DX, SI
// r2 = 2×l0×l2
MOVQ (CX), AX
SHLQ $0x01, AX
MULQ 16(CX)
MOVQ AX, R9
MOVQ DX, R8
// r2 += l1×l1
MOVQ 8(CX), AX
MULQ 8(CX)
ADDQ AX, R9
ADCQ DX, R8
// r2 += 38×l3×l4
MOVQ 24(CX), AX
IMUL3Q $0x26, AX, AX
MULQ 32(CX)
ADDQ AX, R9
ADCQ DX, R8
// r3 = 2×l0×l3
MOVQ (CX), AX
SHLQ $0x01, AX
MULQ 24(CX)
MOVQ AX, R11
MOVQ DX, R10
// r3 += 2×l1×l2
MOVQ 8(CX), AX
IMUL3Q $0x02, AX, AX
MULQ 16(CX)
ADDQ AX, R11
ADCQ DX, R10
// r3 += 19×l4×l4
MOVQ 32(CX), AX
IMUL3Q $0x13, AX, AX
MULQ 32(CX)
ADDQ AX, R11
ADCQ DX, R10
// r4 = 2×l0×l4
MOVQ (CX), AX
SHLQ $0x01, AX
MULQ 32(CX)
MOVQ AX, R13
MOVQ DX, R12
// r4 += 2×l1×l3
MOVQ 8(CX), AX
IMUL3Q $0x02, AX, AX
MULQ 24(CX)
ADDQ AX, R13
ADCQ DX, R12
// r4 += l2×l2
MOVQ 16(CX), AX
MULQ 16(CX)
ADDQ AX, R13
ADCQ DX, R12
// First reduction chain
MOVQ $0x0007ffffffffffff, AX
SHLQ $0x0d, BP, BX
SHLQ $0x0d, DI, SI
SHLQ $0x0d, R9, R8
SHLQ $0x0d, R11, R10
SHLQ $0x0d, R13, R12
ANDQ AX, BP
IMUL3Q $0x13, R12, R12
ADDQ R12, BP
ANDQ AX, DI
ADDQ BX, DI
ANDQ AX, R9
ADDQ SI, R9
ANDQ AX, R11
ADDQ R8, R11
ANDQ AX, R13
ADDQ R10, R13
// Second reduction chain (carryPropagate)
MOVQ BP, BX
SHRQ $0x33, BX
MOVQ DI, SI
SHRQ $0x33, SI
MOVQ R9, R8
SHRQ $0x33, R8
MOVQ R11, R10
SHRQ $0x33, R10
MOVQ R13, R12
SHRQ $0x33, R12
ANDQ AX, BP
IMUL3Q $0x13, R12, R12
ADDQ R12, BP
ANDQ AX, DI
ADDQ BX, DI
ANDQ AX, R9
ADDQ SI, R9
ANDQ AX, R11
ADDQ R8, R11
ANDQ AX, R13
ADDQ R10, R13
// Store output
MOVQ out+0(FP), AX
MOVQ BP, (AX)
MOVQ DI, 8(AX)
MOVQ R9, 16(AX)
MOVQ R11, 24(AX)
MOVQ R13, 32(AX)
RET