// +build arm64,!noasm #include "textflag.h" TEXT ·fp503ConditionalSwap(SB), NOSPLIT, $0-17 MOVD x+0(FP), R0 MOVD y+8(FP), R1 MOVB choice+16(FP), R2 // Set flags // If choice is not 0 or 1, this implementation will swap completely CMP $0, R2 LDP 0(R0), (R3, R4) LDP 0(R1), (R5, R6) CSEL EQ, R3, R5, R7 CSEL EQ, R4, R6, R8 STP (R7, R8), 0(R0) CSEL NE, R3, R5, R9 CSEL NE, R4, R6, R10 STP (R9, R10), 0(R1) LDP 16(R0), (R3, R4) LDP 16(R1), (R5, R6) CSEL EQ, R3, R5, R7 CSEL EQ, R4, R6, R8 STP (R7, R8), 16(R0) CSEL NE, R3, R5, R9 CSEL NE, R4, R6, R10 STP (R9, R10), 16(R1) LDP 32(R0), (R3, R4) LDP 32(R1), (R5, R6) CSEL EQ, R3, R5, R7 CSEL EQ, R4, R6, R8 STP (R7, R8), 32(R0) CSEL NE, R3, R5, R9 CSEL NE, R4, R6, R10 STP (R9, R10), 32(R1) LDP 48(R0), (R3, R4) LDP 48(R1), (R5, R6) CSEL EQ, R3, R5, R7 CSEL EQ, R4, R6, R8 STP (R7, R8), 48(R0) CSEL NE, R3, R5, R9 CSEL NE, R4, R6, R10 STP (R9, R10), 48(R1) RET TEXT ·fp503AddReduced(SB), NOSPLIT, $0-24 MOVD z+0(FP), R2 MOVD x+8(FP), R0 MOVD y+16(FP), R1 // Load first summand into R3-R10 // Add first summand and second summand and store result in R3-R10 LDP 0(R0), (R3, R4) LDP 0(R1), (R11, R12) LDP 16(R0), (R5, R6) LDP 16(R1), (R13, R14) ADDS R11, R3 ADCS R12, R4 ADCS R13, R5 ADCS R14, R6 LDP 32(R0), (R7, R8) LDP 32(R1), (R11, R12) LDP 48(R0), (R9, R10) LDP 48(R1), (R13, R14) ADCS R11, R7 ADCS R12, R8 ADCS R13, R9 ADC R14, R10 // Subtract 2 * p503 in R11-R17 from the result in R3-R10 LDP ·p503x2+0(SB), (R11, R12) LDP ·p503x2+24(SB), (R13, R14) SUBS R11, R3 SBCS R12, R4 LDP ·p503x2+40(SB), (R15, R16) SBCS R12, R5 SBCS R13, R6 MOVD ·p503x2+56(SB), R17 SBCS R14, R7 SBCS R15, R8 SBCS R16, R9 SBCS R17, R10 SBC ZR, ZR, R19 // If x + y - 2 * p503 < 0, R19 is 1 and 2 * p503 should be added AND R19, R11 AND R19, R12 AND R19, R13 AND R19, R14 AND R19, R15 AND R19, R16 AND R19, R17 ADDS R11, R3 ADCS R12, R4 STP (R3, R4), 0(R2) ADCS R12, R5 ADCS R13, R6 STP (R5, R6), 16(R2) ADCS R14, R7 ADCS R15, R8 STP (R7, R8), 32(R2) ADCS R16, R9 ADC R17, R10 STP (R9, R10), 48(R2) RET TEXT ·fp503SubReduced(SB), NOSPLIT, $0-24 MOVD z+0(FP), R2 MOVD x+8(FP), R0 MOVD y+16(FP), R1 // Load x into R3-R10 // Subtract y from x and store result in R3-R10 LDP 0(R0), (R3, R4) LDP 0(R1), (R11, R12) LDP 16(R0), (R5, R6) LDP 16(R1), (R13, R14) SUBS R11, R3 SBCS R12, R4 SBCS R13, R5 SBCS R14, R6 LDP 32(R0), (R7, R8) LDP 32(R1), (R11, R12) LDP 48(R0), (R9, R10) LDP 48(R1), (R13, R14) SBCS R11, R7 SBCS R12, R8 SBCS R13, R9 SBCS R14, R10 SBC ZR, ZR, R19 // If x - y < 0, R19 is 1 and 2 * p503 should be added LDP ·p503x2+0(SB), (R11, R12) LDP ·p503x2+24(SB), (R13, R14) AND R19, R11 AND R19, R12 LDP ·p503x2+40(SB), (R15, R16) AND R19, R13 AND R19, R14 MOVD ·p503x2+56(SB), R17 AND R19, R15 AND R19, R16 AND R19, R17 ADDS R11, R3 ADCS R12, R4 STP (R3, R4), 0(R2) ADCS R12, R5 ADCS R13, R6 STP (R5, R6), 16(R2) ADCS R14, R7 ADCS R15, R8 STP (R7, R8), 32(R2) ADCS R16, R9 ADC R17, R10 STP (R9, R10), 48(R2) RET TEXT ·fp503AddLazy(SB), NOSPLIT, $0-24 MOVD z+0(FP), R2 MOVD x+8(FP), R0 MOVD y+16(FP), R1 // Load first summand into R3-R10 // Add first summand and second summand and store result in R3-R10 LDP 0(R0), (R3, R4) LDP 0(R1), (R11, R12) LDP 16(R0), (R5, R6) LDP 16(R1), (R13, R14) ADDS R11, R3 ADCS R12, R4 STP (R3, R4), 0(R2) ADCS R13, R5 ADCS R14, R6 STP (R5, R6), 16(R2) LDP 32(R0), (R7, R8) LDP 32(R1), (R11, R12) LDP 48(R0), (R9, R10) LDP 48(R1), (R13, R14) ADCS R11, R7 ADCS R12, R8 STP (R7, R8), 32(R2) ADCS R13, R9 ADC R14, R10 STP (R9, R10), 48(R2) RET TEXT ·fp503X2AddLazy(SB), NOSPLIT, $0-24 MOVD z+0(FP), R2 MOVD x+8(FP), R0 MOVD y+16(FP), R1 LDP 0(R0), (R3, R4) LDP 0(R1), (R11, R12) LDP 16(R0), (R5, R6) LDP 16(R1), (R13, R14) ADDS R11, R3 ADCS R12, R4 STP (R3, R4), 0(R2) ADCS R13, R5 ADCS R14, R6 STP (R5, R6), 16(R2) LDP 32(R0), (R7, R8) LDP 32(R1), (R11, R12) LDP 48(R0), (R9, R10) LDP 48(R1), (R13, R14) ADCS R11, R7 ADCS R12, R8 STP (R7, R8), 32(R2) ADCS R13, R9 ADCS R14, R10 STP (R9, R10), 48(R2) LDP 64(R0), (R3, R4) LDP 64(R1), (R11, R12) LDP 80(R0), (R5, R6) LDP 80(R1), (R13, R14) ADCS R11, R3 ADCS R12, R4 STP (R3, R4), 64(R2) ADCS R13, R5 ADCS R14, R6 STP (R5, R6), 80(R2) LDP 96(R0), (R7, R8) LDP 96(R1), (R11, R12) LDP 112(R0), (R9, R10) LDP 112(R1), (R13, R14) ADCS R11, R7 ADCS R12, R8 STP (R7, R8), 96(R2) ADCS R13, R9 ADC R14, R10 STP (R9, R10), 112(R2) RET TEXT ·fp503X2SubLazy(SB), NOSPLIT, $0-24 MOVD z+0(FP), R2 MOVD x+8(FP), R0 MOVD y+16(FP), R1 LDP 0(R0), (R3, R4) LDP 0(R1), (R11, R12) LDP 16(R0), (R5, R6) LDP 16(R1), (R13, R14) SUBS R11, R3 SBCS R12, R4 STP (R3, R4), 0(R2) SBCS R13, R5 SBCS R14, R6 STP (R5, R6), 16(R2) LDP 32(R0), (R7, R8) LDP 32(R1), (R11, R12) LDP 48(R0), (R9, R10) LDP 48(R1), (R13, R14) SBCS R11, R7 SBCS R12, R8 STP (R7, R8), 32(R2) SBCS R13, R9 SBCS R14, R10 STP (R9, R10), 48(R2) LDP 64(R0), (R3, R4) LDP 64(R1), (R11, R12) LDP 80(R0), (R5, R6) LDP 80(R1), (R13, R14) SBCS R11, R3 SBCS R12, R4 SBCS R13, R5 SBCS R14, R6 LDP 96(R0), (R7, R8) LDP 96(R1), (R11, R12) LDP 112(R0), (R9, R10) LDP 112(R1), (R13, R14) SBCS R11, R7 SBCS R12, R8 SBCS R13, R9 SBCS R14, R10 SBC ZR, ZR, R15 // If x - y < 0, R15 is 1 and p503 should be added LDP ·p503+16(SB), (R16, R17) LDP ·p503+32(SB), (R19, R20) AND R15, R16 AND R15, R17 LDP ·p503+48(SB), (R21, R22) AND R15, R19 AND R15, R20 AND R15, R21 AND R15, R22 ADDS R16, R3 ADCS R16, R4 STP (R3, R4), 64(R2) ADCS R16, R5 ADCS R17, R6 STP (R5, R6), 80(R2) ADCS R19, R7 ADCS R20, R8 STP (R7, R8), 96(R2) ADCS R21, R9 ADC R22, R10 STP (R9, R10), 112(R2) RET // Expects that X0*Y0 is already in Z0(low),Z3(high) and X0*Y1 in Z1(low),Z2(high) // Z0 is not actually touched // Result of (X0-X1) * (Y0-Y1) will be in Z0-Z3 // Inputs get overwritten, except for X1 #define mul128x128comba(X0, X1, Y0, Y1, Z0, Z1, Z2, Z3, T0) \ MUL X1, Y0, X0 \ UMULH X1, Y0, Y0 \ ADDS Z3, Z1 \ ADC ZR, Z2 \ \ MUL Y1, X1, T0 \ UMULH Y1, X1, Y1 \ ADDS X0, Z1 \ ADCS Y0, Z2 \ ADC ZR, ZR, Z3 \ \ ADDS T0, Z2 \ ADC Y1, Z3 // Expects that X points to (X0-X1) // Result of (X0-X3) * (Y0-Y3) will be in Z0-Z7 // Inputs get overwritten, except X2-X3 and Y2-Y3 #define mul256x256karatsuba(X, X0, X1, X2, X3, Y0, Y1, Y2, Y3, Z0, Z1, Z2, Z3, Z4, Z5, Z6, Z7, T0, T1)\ ADDS X2, X0 \ // xH + xL, destroys xL ADCS X3, X1 \ ADCS ZR, ZR, T0 \ \ ADDS Y2, Y0, Z6 \ // yH + yL ADCS Y3, Y1, T1 \ ADC ZR, ZR, Z7 \ \ SUB T0, ZR, Z2 \ SUB Z7, ZR, Z3 \ AND Z7, T0 \ // combined carry \ AND Z2, Z6, Z0 \ // masked(yH + yL) AND Z2, T1, Z1 \ \ AND Z3, X0, Z4 \ // masked(xH + xL) AND Z3, X1, Z5 \ \ MUL Z6, X0, Z2 \ MUL T1, X0, Z3 \ \ ADDS Z4, Z0 \ UMULH T1, X0, Z4 \ ADCS Z5, Z1 \ UMULH Z6, X0, Z5 \ ADC ZR, T0 \ \ // (xH + xL) * (yH + yL) mul128x128comba(X0, X1, Z6, T1, Z2, Z3, Z4, Z5, Z7)\ \ LDP 0+X, (X0, X1) \ \ ADDS Z0, Z4 \ UMULH Y0, X0, Z7 \ UMULH Y1, X0, T1 \ ADCS Z1, Z5 \ MUL Y0, X0, Z0 \ MUL Y1, X0, Z1 \ ADC ZR, T0 \ \ // xL * yL mul128x128comba(X0, X1, Y0, Y1, Z0, Z1, T1, Z7, Z6)\ \ MUL Y2, X2, X0 \ UMULH Y2, X2, Y0 \ SUBS Z0, Z2 \ // (xH + xL) * (yH + yL) - xL * yL SBCS Z1, Z3 \ SBCS T1, Z4 \ MUL Y3, X2, X1 \ UMULH Y3, X2, Z6 \ SBCS Z7, Z5 \ SBCS ZR, T0 \ \ // xH * yH mul128x128comba(X2, X3, Y2, Y3, X0, X1, Z6, Y0, Y1)\ \ SUBS X0, Z2 \ // (xH + xL) * (yH + yL) - xL * yL - xH * yH SBCS X1, Z3 \ SBCS Z6, Z4 \ SBCS Y0, Z5 \ SBCS ZR, T0 \ \ ADDS T1, Z2 \ // (xH * yH) * 2^256 + ((xH + xL) * (yH + yL) - xL * yL - xH * yH) * 2^128 + xL * yL ADCS Z7, Z3 \ ADCS X0, Z4 \ ADCS X1, Z5 \ ADCS T0, Z6 \ ADC Y0, ZR, Z7 // This implements two-level Karatsuba with a 128x128 Comba multiplier // at the bottom TEXT ·fp503Mul(SB), NOSPLIT, $0-24 MOVD z+0(FP), R2 MOVD x+8(FP), R0 MOVD y+16(FP), R1 // Load xL in R3-R6, xH in R7-R10 // (xH + xL) in R25-R29 LDP 0(R0), (R3, R4) LDP 32(R0), (R7, R8) ADDS R3, R7, R25 ADCS R4, R8, R26 LDP 16(R0), (R5, R6) LDP 48(R0), (R9, R10) ADCS R5, R9, R27 ADCS R6, R10, R29 ADC ZR, ZR, R7 // Load yL in R11-R14, yH in R15-19 // (yH + yL) in R11-R14, destroys yL LDP 0(R1), (R11, R12) LDP 32(R1), (R15, R16) ADDS R15, R11 ADCS R16, R12 LDP 16(R1), (R13, R14) LDP 48(R1), (R17, R19) ADCS R17, R13 ADCS R19, R14 ADC ZR, ZR, R8 // Compute maskes and combined carry SUB R7, ZR, R9 SUB R8, ZR, R10 AND R8, R7 // masked(yH + yL) AND R9, R11, R15 AND R9, R12, R16 AND R9, R13, R17 AND R9, R14, R19 // masked(xH + xL) AND R10, R25, R20 AND R10, R26, R21 AND R10, R27, R22 AND R10, R29, R23 // masked(xH + xL) + masked(yH + yL) in R15-R19 ADDS R20, R15 ADCS R21, R16 ADCS R22, R17 ADCS R23, R19 ADC ZR, R7 // Use z as temporary storage STP (R25, R26), 0(R2) // (xH + xL) * (yH + yL) mul256x256karatsuba(0(R2), R25, R26, R27, R29, R11, R12, R13, R14, R8, R9, R10, R20, R21, R22, R23, R24, R0, R1) MOVD x+8(FP), R0 MOVD y+16(FP), R1 ADDS R21, R15 ADCS R22, R16 ADCS R23, R17 ADCS R24, R19 ADC ZR, R7 // Load yL in R11-R14 LDP 0(R1), (R11, R12) LDP 16(R1), (R13, R14) // xL * yL mul256x256karatsuba(0(R0), R3, R4, R5, R6, R11, R12, R13, R14, R21, R22, R23, R24, R25, R26, R27, R29, R1, R2) MOVD z+0(FP), R2 MOVD y+16(FP), R1 // (xH + xL) * (yH + yL) - xL * yL SUBS R21, R8 SBCS R22, R9 STP (R21, R22), 0(R2) SBCS R23, R10 SBCS R24, R20 STP (R23, R24), 16(R2) SBCS R25, R15 SBCS R26, R16 SBCS R27, R17 SBCS R29, R19 SBC ZR, R7 // Load xH in R3-R6, yH in R11-R14 LDP 32(R0), (R3, R4) LDP 48(R0), (R5, R6) LDP 32(R1), (R11, R12) LDP 48(R1), (R13, R14) ADDS R25, R8 ADCS R26, R9 ADCS R27, R10 ADCS R29, R20 ADC ZR, ZR, R1 MOVD R20, 32(R2) // xH * yH mul256x256karatsuba(32(R0), R3, R4, R5, R6, R11, R12, R13, R14, R21, R22, R23, R24, R25, R26, R27, R29, R2, R20) NEG R1, R1 MOVD z+0(FP), R2 MOVD 32(R2), R20 // (xH + xL) * (yH + yL) - xL * yL - xH * yH in R8-R10,R20,R15-R19 // Store lower half in z, that's done SUBS R21, R8 SBCS R22, R9 STP (R8, R9), 32(R2) SBCS R23, R10 SBCS R24, R20 STP (R10, R20), 48(R2) SBCS R25, R15 SBCS R26, R16 SBCS R27, R17 SBCS R29, R19 SBC ZR, R7 // (xH * yH) * 2^512 + ((xH + xL) * (yH + yL) - xL * yL - xH * yH) * 2^256 + xL * yL // Store remaining limbs in z ADDS $1, R1 ADCS R21, R15 ADCS R22, R16 STP (R15, R16), 64(R2) ADCS R23, R17 ADCS R24, R19 STP (R17, R19), 80(R2) ADCS R7, R25 ADCS ZR, R26 STP (R25, R26), 96(R2) ADCS ZR, R27 ADC ZR, R29 STP (R27, R29), 112(R2) RET // Expects that X0*Y0 is already in Z0(low),Z3(high) and X0*Y1 in Z1(low),Z2(high) // Z0 is not actually touched // Result of (X0-X1) * (Y0-Y3) will be in Z0-Z5 // Inputs remain intact #define mul128x256comba(X0, X1, Y0, Y1, Y2, Y3, Z0, Z1, Z2, Z3, Z4, Z5, T0, T1, T2, T3)\ MUL X1, Y0, T0 \ UMULH X1, Y0, T1 \ ADDS Z3, Z1 \ ADC ZR, Z2 \ \ MUL X0, Y2, T2 \ UMULH X0, Y2, T3 \ ADDS T0, Z1 \ ADCS T1, Z2 \ ADC ZR, ZR, Z3 \ \ MUL X1, Y1, T0 \ UMULH X1, Y1, T1 \ ADDS T2, Z2 \ ADCS T3, Z3 \ ADC ZR, ZR, Z4 \ \ MUL X0, Y3, T2 \ UMULH X0, Y3, T3 \ ADDS T0, Z2 \ ADCS T1, Z3 \ ADC ZR, Z4 \ \ MUL X1, Y2, T0 \ UMULH X1, Y2, T1 \ ADDS T2, Z3 \ ADCS T3, Z4 \ ADC ZR, ZR, Z5 \ \ MUL X1, Y3, T2 \ UMULH X1, Y3, T3 \ ADDS T0, Z3 \ ADCS T1, Z4 \ ADC ZR, Z5 \ ADDS T2, Z4 \ ADC T3, Z5 // This implements the shifted 2^(B*w) Montgomery reduction from // https://eprint.iacr.org/2016/986.pdf, section Section 3.2, with // B = 4, w = 64. Performance results were reported in // https://eprint.iacr.org/2018/700.pdf Section 6. TEXT ·fp503MontgomeryReduce(SB), NOSPLIT, $0-16 MOVD x+8(FP), R0 // Load x0-x1 LDP 0(R0), (R2, R3) // Load the prime constant in R25-R29 LDP ·p503p1s8+32(SB), (R25, R26) LDP ·p503p1s8+48(SB), (R27, R29) // [x0,x1] * p503p1s8 to R4-R9 MUL R2, R25, R4 // x0 * p503p1s8[0] UMULH R2, R25, R7 MUL R2, R26, R5 // x0 * p503p1s8[1] UMULH R2, R26, R6 mul128x256comba(R2, R3, R25, R26, R27, R29, R4, R5, R6, R7, R8, R9, R10, R11, R12, R13) LDP 16(R0), (R3, R11) // x2 LDP 32(R0), (R12, R13) LDP 48(R0), (R14, R15) // Left-shift result in R4-R9 by 56 to R4-R10 ORR R9>>8, ZR, R10 LSL $56, R9 ORR R8>>8, R9 LSL $56, R8 ORR R7>>8, R8 LSL $56, R7 ORR R6>>8, R7 LSL $56, R6 ORR R5>>8, R6 LSL $56, R5 ORR R4>>8, R5 LSL $56, R4 ADDS R4, R11 // x3 ADCS R5, R12 // x4 ADCS R6, R13 ADCS R7, R14 ADCS R8, R15 LDP 64(R0), (R16, R17) LDP 80(R0), (R19, R20) MUL R3, R25, R4 // x2 * p503p1s8[0] UMULH R3, R25, R7 ADCS R9, R16 ADCS R10, R17 ADCS ZR, R19 ADCS ZR, R20 LDP 96(R0), (R21, R22) LDP 112(R0), (R23, R24) MUL R3, R26, R5 // x2 * p503p1s8[1] UMULH R3, R26, R6 ADCS ZR, R21 ADCS ZR, R22 ADCS ZR, R23 ADC ZR, R24 // [x2,x3] * p503p1s8 to R4-R9 mul128x256comba(R3, R11, R25, R26, R27, R29, R4, R5, R6, R7, R8, R9, R10, R0, R1, R2) ORR R9>>8, ZR, R10 LSL $56, R9 ORR R8>>8, R9 LSL $56, R8 ORR R7>>8, R8 LSL $56, R7 ORR R6>>8, R7 LSL $56, R6 ORR R5>>8, R6 LSL $56, R5 ORR R4>>8, R5 LSL $56, R4 ADDS R4, R13 // x5 ADCS R5, R14 // x6 ADCS R6, R15 ADCS R7, R16 MUL R12, R25, R4 // x4 * p503p1s8[0] UMULH R12, R25, R7 ADCS R8, R17 ADCS R9, R19 ADCS R10, R20 ADCS ZR, R21 MUL R12, R26, R5 // x4 * p503p1s8[1] UMULH R12, R26, R6 ADCS ZR, R22 ADCS ZR, R23 ADC ZR, R24 // [x4,x5] * p503p1s8 to R4-R9 mul128x256comba(R12, R13, R25, R26, R27, R29, R4, R5, R6, R7, R8, R9, R10, R0, R1, R2) ORR R9>>8, ZR, R10 LSL $56, R9 ORR R8>>8, R9 LSL $56, R8 ORR R7>>8, R8 LSL $56, R7 ORR R6>>8, R7 LSL $56, R6 ORR R5>>8, R6 LSL $56, R5 ORR R4>>8, R5 LSL $56, R4 ADDS R4, R15 // x7 ADCS R5, R16 // x8 ADCS R6, R17 ADCS R7, R19 MUL R14, R25, R4 // x6 * p503p1s8[0] UMULH R14, R25, R7 ADCS R8, R20 ADCS R9, R21 ADCS R10, R22 MUL R14, R26, R5 // x6 * p503p1s8[1] UMULH R14, R26, R6 ADCS ZR, R23 ADC ZR, R24 // [x6,x7] * p503p1s8 to R4-R9 mul128x256comba(R14, R15, R25, R26, R27, R29, R4, R5, R6, R7, R8, R9, R10, R0, R1, R2) ORR R9>>8, ZR, R10 LSL $56, R9 ORR R8>>8, R9 LSL $56, R8 ORR R7>>8, R8 LSL $56, R7 ORR R6>>8, R7 LSL $56, R6 ORR R5>>8, R6 LSL $56, R5 ORR R4>>8, R5 LSL $56, R4 MOVD z+0(FP), R0 ADDS R4, R17 ADCS R5, R19 STP (R16, R17), 0(R0) // Store final result to z ADCS R6, R20 ADCS R7, R21 STP (R19, R20), 16(R0) ADCS R8, R22 ADCS R9, R23 STP (R21, R22), 32(R0) ADC R10, R24 STP (R23, R24), 48(R0) RET TEXT ·fp503StrongReduce(SB), NOSPLIT, $0-8 MOVD x+0(FP), R0 // Keep x in R1-R8, p503 in R9-R14, subtract to R1-R8 LDP ·p503+16(SB), (R9, R10) LDP 0(R0), (R1, R2) LDP 16(R0), (R3, R4) SUBS R9, R1 SBCS R9, R2 LDP 32(R0), (R5, R6) LDP ·p503+32(SB), (R11, R12) SBCS R9, R3 SBCS R10, R4 LDP 48(R0), (R7, R8) LDP ·p503+48(SB), (R13, R14) SBCS R11, R5 SBCS R12, R6 SBCS R13, R7 SBCS R14, R8 SBC ZR, ZR, R15 // Mask with the borrow and add p503 AND R15, R9 AND R15, R10 AND R15, R11 AND R15, R12 AND R15, R13 AND R15, R14 ADDS R9, R1 ADCS R9, R2 STP (R1, R2), 0(R0) ADCS R9, R3 ADCS R10, R4 STP (R3, R4), 16(R0) ADCS R11, R5 ADCS R12, R6 STP (R5, R6), 32(R0) ADCS R13, R7 ADCS R14, R8 STP (R7, R8), 48(R0) RET