forked from btcsuite/btcd
-
Notifications
You must be signed in to change notification settings - Fork 31
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
Showing
1 changed file
with
120 additions
and
106 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,119 +1,133 @@ | ||
| #include "textflag.h" | ||
|
|
||
| #define EORP(n, Ra1, Ra2, Rb1, Rb2, Rc1, Rc2) \ | ||
| LDP n(Ra1), (Rb1, Rb2) \ | ||
| LDP n(Ra2), (Rc1, Rc2) \ | ||
| EOR Rb1, Rc1, Rc1 \ | ||
| EOR Rb2, Rc2, Rc2 \ | ||
| STP (Rc1, Rc2), n(Ra1) | ||
|
|
||
| #define ADDPW(n, Ra, _, Rb1, Rb2, Rc1, Rc2) \ | ||
| LDPW n(Ra), (Rb1, Rb2) \ | ||
| ADDW Rb1, Rc1, Rc1 \ | ||
| ADDW Rb2, Rc2, Rc2 \ | ||
| STPW (Rc1, Rc2), n(Ra) | ||
|
|
||
| #define BLK(OP, w, n, Ra, Rb, Rc, Rd, Re, Rf, Rg, Rh) \ | ||
| OP(0*w+n, R16, R17, R19, R20, Ra, Rb) \ | ||
| OP(1*w+n, R16, R17, R21, R22, Rc, Rd) \ | ||
| OP(2*w+n, R16, R17, R23, R24, Re, Rf) \ | ||
| OP(3*w+n, R16, R17, R25, R26, Rg, Rh) | ||
|
|
||
| #define ADDEORW(Ra, Rb, Rc, n, Rd) \ | ||
| ADDW Ra, Rb, Rc \ | ||
| EORW Rc@>n, Rd, Rd | ||
|
|
||
| #define QRTRND(Ra, Rb, Rc, Rd, Re, Rf, Rg, Rh, Ri, Rj, Rk, Rl, n) \ | ||
| ADDEORW(Ra, Rb, R19, n, Rc) \ | ||
| ADDEORW(Rd, Re, R20, n, Rf) \ | ||
| ADDEORW(Rg, Rh, R21, n, Ri) \ | ||
| ADDEORW(Rj, Rk, R22, n, Rl) | ||
| #define BLK_COPY \ | ||
| FLDPQ 0(R16), (F0, F1) \ | ||
| FSTPQ (F0, F1), 0(R17) \ | ||
| FLDPQ 32(R16), (F2, F3) \ | ||
| FSTPQ (F2, F3), 32(R17) \ | ||
| FLDPQ 64(R16), (F4, F5) \ | ||
| FSTPQ (F4, F5), 64(R17) \ | ||
| FLDPQ 96(R16), (F6, F7) \ | ||
| FSTPQ (F6, F7), 96(R17) | ||
|
|
||
| #define BLK_MIX \ | ||
| ADD $64, R16, R17 \ | ||
| CALL ·eorSalsa8(SB) \ | ||
| MOVD X+0(FP), R17 \ | ||
| ADD $64, R17, R16 \ | ||
| CALL ·eorSalsa8(SB) | ||
|
|
||
| #define MIX_LOOP1 \ | ||
| MOVD X+0(FP), R16 \ | ||
| MOVD R17, p-8(SP) \ | ||
| BLK_COPY \ | ||
| BLK_MIX \ | ||
| MOVD p-8(SP), R17 \ | ||
| ADD $128, R17, R17 | ||
|
|
||
| #define EORP(n, Ra, Rb) \ | ||
| LDP n(R16), (R19, R20) \ | ||
| LDP n(R17), (Ra, Rb) \ | ||
| EOR R19, Ra, Ra \ | ||
| EOR R20, Rb, Rb | ||
|
|
||
| #define EORP_STP(n) \ | ||
| EORP (n, R0, R1) \ | ||
| STP (R0, R1), n(R16) | ||
|
|
||
| #define MIX_LOOP2 \ | ||
| MOVD X+0(FP), R16 \ | ||
| MOVD V+8(FP), R17 \ | ||
| MOVWU 64(R16), R0 \ | ||
| AND $1023, R0, R0 \ | ||
| ADD R0<<7, R17, R17 \ | ||
| EORP_STP(0); EORP_STP(16) \ | ||
| EORP_STP(32); EORP_STP(48) \ | ||
| EORP_STP(64); EORP_STP(80) \ | ||
| EORP_STP(96); EORP_STP(112) \ | ||
| BLK_MIX | ||
|
|
||
| TEXT ·scrypt(SB), NOSPLIT, $8-16 | ||
| MOVD V+8(FP), R1 | ||
|
|
||
| L1: MOVD X+0(FP), R0 | ||
|
|
||
| FLDPQ 0(R0), (F0, F1) | ||
| FSTPQ (F0, F1), 0(R1) | ||
| FLDPQ 32(R0), (F2, F3) | ||
| FSTPQ (F2, F3), 32(R1) | ||
| FLDPQ 64(R0), (F4, F5) | ||
| FSTPQ (F4, F5), 64(R1) | ||
| FLDPQ 96(R0), (F6, F7) | ||
| FSTPQ (F6, F7), 96(R1) | ||
|
|
||
| MOVD R1, 8(RSP) | ||
| ADD $64, R0, R1 | ||
| CALL eor_salsa8(SB) | ||
| MOVD X+0(FP), R1 | ||
| ADD $64, R1, R0 | ||
| CALL eor_salsa8(SB) | ||
|
|
||
| MOVD 8(RSP), R1 | ||
| ADD $128, R1, R1 | ||
| MOVD V+8(FP), R2 | ||
| ADD $0x20000, R2, R2 | ||
| CMP R1, R2 | ||
| MOVD V+8(FP), R17 | ||
|
|
||
| L1: MIX_LOOP1 | ||
| MIX_LOOP1 | ||
| MIX_LOOP1 | ||
| MIX_LOOP1 | ||
|
|
||
| MOVD V+8(FP), R0 | ||
| ADD $0x20000, R0, R0 | ||
| CMP R17, R0 | ||
| BNE L1 | ||
|
|
||
| MOVD $0, R1 | ||
| MOVD $1024, R0 | ||
| MOVD R0, i-8(SP) | ||
|
|
||
| L2: MOVD R1, 8(RSP) | ||
| MOVD X+0(FP), R16 | ||
| MOVD V+8(FP), R17 | ||
| MOVWU 64(R16), R0 | ||
| AND $1023, R0, R0 | ||
| ADD R0<<7, R17, R17 | ||
|
|
||
| BLK(EORP, 16, 0, R0, R1, R2, R3, R4, R5, R6, R7) | ||
| BLK(EORP, 16, 64, R0, R1, R2, R3, R4, R5, R6, R7) | ||
|
|
||
| MOVD R16, R0 | ||
| ADD $64, R0, R1 | ||
| CALL eor_salsa8(SB) | ||
| MOVD X+0(FP), R1 | ||
| ADD $64, R1, R0 | ||
| CALL eor_salsa8(SB) | ||
|
|
||
| MOVD 8(RSP), R1 | ||
| ADD $1, R1, R1 | ||
| CMP $1024, R1 | ||
| L2: MIX_LOOP2 | ||
| MIX_LOOP2 | ||
| MIX_LOOP2 | ||
| MIX_LOOP2 | ||
|
|
||
| MOVD i-8(SP), R0 | ||
| SUBS $4, R0, R0 | ||
| MOVD R0, i-8(SP) | ||
| BNE L2 | ||
|
|
||
| RET | ||
|
|
||
| TEXT eor_salsa8(SB), NOSPLIT, $0 | ||
| MOVD R0, R16 | ||
| MOVD R1, R17 | ||
|
|
||
| BLK(EORP, 16, 0, R0, R2, R4, R6, R8, R10, R12, R14) | ||
|
|
||
| LSR $32, R0, R1 | ||
| LSR $32, R2, R3 | ||
| LSR $32, R4, R5 | ||
| LSR $32, R6, R7 | ||
| LSR $32, R8, R9 | ||
| LSR $32, R10, R11 | ||
| LSR $32, R12, R13 | ||
| LSR $32, R14, R15 | ||
|
|
||
| MOVD $0, R17 | ||
|
|
||
| L: QRTRND(R0, R12, R4, R5, R1, R9, R10, R6, R14, R15, R11, R3, 25) | ||
| QRTRND(R4, R0, R8, R9, R5, R13, R14, R10, R2, R3, R15, R7, 23) | ||
| QRTRND(R8, R4, R12, R13, R9, R1, R2, R14, R6, R7, R3, R11, 19) | ||
| QRTRND(R12, R8, R0, R1, R13, R5, R6, R2, R10, R11, R7, R15, 14) | ||
| QRTRND(R0, R3, R1, R5, R4, R6, R10, R9, R11, R15, R14, R12, 25) | ||
| QRTRND(R1, R0, R2, R6, R5, R7, R11, R10, R8, R12, R15, R13, 23) | ||
| QRTRND(R2, R1, R3, R7, R6, R4, R8, R11, R9, R13, R12, R14, 19) | ||
| QRTRND(R3, R2, R0, R4, R7, R5, R9, R8, R10, R14, R13, R15, 14) | ||
|
|
||
| ADD $1, R17, R17 | ||
| CMP $4, R17 | ||
| BNE L | ||
|
|
||
| BLK(ADDPW, 8, 0, R0, R1, R2, R3, R4, R5, R6, R7) | ||
| BLK(ADDPW, 8, 32, R8, R9, R10, R11, R12, R13, R14, R15) | ||
| #define EORP_LSR(n, Ra, Rb, Rc, Rd) \ | ||
| EORP (n, Ra, Rc) \ | ||
| LSR $32, Ra, Rb \ | ||
| LSR $32, Rc, Rd | ||
|
|
||
| #define EORP_LSR_MOV(n, Ra, Rb, Rc, Rd, Re, Rf) \ | ||
| EORP_LSR(n, Ra, Rb, Rc, Rd) \ | ||
| MOVD Ra, Re \ | ||
| MOVD Rc, Rf | ||
|
|
||
| #define ADD_ROR_EOR(Ra, Rb, n, Rc) \ | ||
| ADDW Ra, Rb, R17 \ | ||
| EORW R17@>n, Rc, Rc | ||
|
|
||
| #define QR(Ra, Rb, Rc, Rd, Re, Rf, Rg, Rh, Ri, Rj, Rk, Rl, n) \ | ||
| ADD_ROR_EOR(Ra, Rb, n, Rc) \ | ||
| ADD_ROR_EOR(Rd, Re, n, Rf) \ | ||
| ADD_ROR_EOR(Rg, Rh, n, Ri) \ | ||
| ADD_ROR_EOR(Rj, Rk, n, Rl) | ||
|
|
||
| #define DBL_ROUND \ | ||
| QR(R0, R12, R4, R5, R1, R9, R10, R6, R14, R15, R11, R3, 25) \ | ||
| QR(R4, R0, R8, R9, R5, R13, R14, R10, R2, R3, R15, R7, 23) \ | ||
| QR(R8, R4, R12, R13, R9, R1, R2, R14, R6, R7, R3, R11, 19) \ | ||
| QR(R12, R8, R0, R1, R13, R5, R6, R2, R10, R11, R7, R15, 14) \ | ||
| QR(R0, R3, R1, R5, R4, R6, R10, R9, R11, R15, R14, R12, 25) \ | ||
| QR(R1, R0, R2, R6, R5, R7, R11, R10, R8, R12, R15, R13, 23) \ | ||
| QR(R2, R1, R3, R7, R6, R4, R8, R11, R9, R13, R12, R14, 19) \ | ||
| QR(R3, R2, R0, R4, R7, R5, R9, R8, R10, R14, R13, R15, 14) | ||
|
|
||
| #define ADD_STP(n, Ra, Rb, Rc) \ | ||
| ADDW Ra, Rb, Rb \ | ||
| ADD Ra>>32, Rc, Rc \ | ||
| STPW (Rb, Rc), n(R16) | ||
|
|
||
| TEXT ·eorSalsa8(SB), NOSPLIT, $0 | ||
| EORP_LSR_MOV(0, R0, R1, R2, R3, R21, R22) | ||
| EORP_LSR_MOV(16, R4, R5, R6, R7, R23, R24) | ||
| EORP_LSR_MOV(32, R8, R9, R10, R11, R25, R26) | ||
| EORP_LSR_MOV(48, R12, R13, R14, R15, R19, R20) | ||
|
|
||
| DBL_ROUND | ||
| DBL_ROUND | ||
| DBL_ROUND | ||
| DBL_ROUND | ||
|
|
||
| ADD_STP(0, R21, R0, R1) | ||
| ADD_STP(8, R22, R2, R3) | ||
| ADD_STP(16, R23, R4, R5) | ||
| ADD_STP(24, R24, R6, R7) | ||
| ADD_STP(32, R25, R8, R9) | ||
| ADD_STP(40, R26, R10, R11) | ||
| ADD_STP(48, R19, R12, R13) | ||
| ADD_STP(56, R20, R14, R15) | ||
|
|
||
| RET |