Chris@10: /* Chris@10: * Copyright (c) 2003, 2007-11 Matteo Frigo Chris@10: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology Chris@10: * Chris@10: * This program is free software; you can redistribute it and/or modify Chris@10: * it under the terms of the GNU General Public License as published by Chris@10: * the Free Software Foundation; either version 2 of the License, or Chris@10: * (at your option) any later version. Chris@10: * Chris@10: * This program is distributed in the hope that it will be useful, Chris@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@10: * GNU General Public License for more details. Chris@10: * Chris@10: * You should have received a copy of the GNU General Public License Chris@10: * along with this program; if not, write to the Free Software Chris@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@10: * Chris@10: */ Chris@10: Chris@10: /* This file was automatically generated --- DO NOT EDIT */ Chris@10: /* Generated on Sun Nov 25 07:42:29 EST 2012 */ Chris@10: Chris@10: #include "codelet-rdft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2cdft_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -trivial-stores -variables 32 -no-generate-bytw -n 16 -dit -name hc2cfdftv_16 -include hc2cfv.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 103 FP additions, 96 FP multiplications, Chris@10: * (or, 53 additions, 46 multiplications, 50 fused multiply/add), Chris@10: * 92 stack variables, 4 constants, and 32 memory accesses Chris@10: */ Chris@10: #include "hc2cfv.h" Chris@10: Chris@10: static void hc2cfdftv_16(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DVK(KP923879532, +0.923879532511286756128183189396788286822416626); Chris@10: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@10: DVK(KP414213562, +0.414213562373095048801688724209698078569671875); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 30)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 30), MAKE_VOLATILE_STRIDE(64, rs)) { Chris@10: V T8, Tc, TQ, TZ, T1J, T1x, T12, TH, T1I, T1q, Tp, TJ, Te, Tf, Td; Chris@10: V TN, Tj, Tk, Ti, TK, Tg, TO, Tl, TL, T1r, Th, TR, T1y, T1s, Tq; Chris@10: V TM, T1z, T1N, T1t, T10, Tr, T13, TS, T1K, T1A, T1E, T1u, T1f, T11, T1c; Chris@10: V Ts, T1d, T14, T1g, TT; Chris@10: { Chris@10: V T3, Tw, TF, TW, Tz, TA, Ty, TX, T7, Tu, T1, T2, Tv, TD, TE; Chris@10: V TC, TV, T5, T6, T4, Tt, TB, TY, T1o, T1v, Tx, Ta, Tb, T9, TP; Chris@10: V T1w, TG, T1p, Tn, To, Tm, TI; Chris@10: T1 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@10: T2 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@10: Tv = LDW(&(W[0])); Chris@10: TD = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@10: TE = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@10: TC = LDW(&(W[TWVL * 8])); Chris@10: TV = LDW(&(W[TWVL * 6])); Chris@10: T5 = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0])); Chris@10: T6 = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0])); Chris@10: T3 = VFMACONJ(T2, T1); Chris@10: Tw = VZMULIJ(Tv, VFNMSCONJ(T2, T1)); Chris@10: T4 = LDW(&(W[TWVL * 14])); Chris@10: Tt = LDW(&(W[TWVL * 16])); Chris@10: TF = VZMULIJ(TC, VFNMSCONJ(TE, TD)); Chris@10: TW = VZMULJ(TV, VFMACONJ(TE, TD)); Chris@10: Tz = LD(&(Rp[WS(rs, 6)]), ms, &(Rp[0])); Chris@10: TA = LD(&(Rm[WS(rs, 6)]), -ms, &(Rm[0])); Chris@10: Ty = LDW(&(W[TWVL * 24])); Chris@10: TX = LDW(&(W[TWVL * 22])); Chris@10: T7 = VZMULJ(T4, VFMACONJ(T6, T5)); Chris@10: Tu = VZMULIJ(Tt, VFNMSCONJ(T6, T5)); Chris@10: Ta = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Tb = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: T9 = LDW(&(W[TWVL * 2])); Chris@10: TP = LDW(&(W[TWVL * 4])); Chris@10: TB = VZMULIJ(Ty, VFNMSCONJ(TA, Tz)); Chris@10: TY = VZMULJ(TX, VFMACONJ(TA, Tz)); Chris@10: T1o = VADD(T3, T7); Chris@10: T8 = VSUB(T3, T7); Chris@10: T1v = VADD(Tw, Tu); Chris@10: Tx = VSUB(Tu, Tw); Chris@10: Tc = VZMULJ(T9, VFMACONJ(Tb, Ta)); Chris@10: TQ = VZMULIJ(TP, VFNMSCONJ(Tb, Ta)); Chris@10: T1w = VADD(TF, TB); Chris@10: TG = VSUB(TB, TF); Chris@10: T1p = VADD(TW, TY); Chris@10: TZ = VSUB(TW, TY); Chris@10: Tn = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); Chris@10: To = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Tm = LDW(&(W[TWVL * 10])); Chris@10: TI = LDW(&(W[TWVL * 12])); Chris@10: T1J = VSUB(T1w, T1v); Chris@10: T1x = VADD(T1v, T1w); Chris@10: T12 = VFMA(LDK(KP414213562), Tx, TG); Chris@10: TH = VFNMS(LDK(KP414213562), TG, Tx); Chris@10: T1I = VSUB(T1o, T1p); Chris@10: T1q = VADD(T1o, T1p); Chris@10: Tp = VZMULJ(Tm, VFMACONJ(To, Tn)); Chris@10: TJ = VZMULIJ(TI, VFNMSCONJ(To, Tn)); Chris@10: Te = LD(&(Rp[WS(rs, 5)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Tf = LD(&(Rm[WS(rs, 5)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Td = LDW(&(W[TWVL * 18])); Chris@10: TN = LDW(&(W[TWVL * 20])); Chris@10: Tj = LD(&(Rp[WS(rs, 7)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Tk = LD(&(Rm[WS(rs, 7)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Ti = LDW(&(W[TWVL * 26])); Chris@10: TK = LDW(&(W[TWVL * 28])); Chris@10: } Chris@10: Tg = VZMULJ(Td, VFMACONJ(Tf, Te)); Chris@10: TO = VZMULIJ(TN, VFNMSCONJ(Tf, Te)); Chris@10: Tl = VZMULJ(Ti, VFMACONJ(Tk, Tj)); Chris@10: TL = VZMULIJ(TK, VFNMSCONJ(Tk, Tj)); Chris@10: T1r = VADD(Tc, Tg); Chris@10: Th = VSUB(Tc, Tg); Chris@10: TR = VSUB(TO, TQ); Chris@10: T1y = VADD(TQ, TO); Chris@10: T1s = VADD(Tl, Tp); Chris@10: Tq = VSUB(Tl, Tp); Chris@10: TM = VSUB(TJ, TL); Chris@10: T1z = VADD(TL, TJ); Chris@10: T1N = VSUB(T1s, T1r); Chris@10: T1t = VADD(T1r, T1s); Chris@10: T10 = VSUB(Tq, Th); Chris@10: Tr = VADD(Th, Tq); Chris@10: T13 = VFNMS(LDK(KP414213562), TM, TR); Chris@10: TS = VFMA(LDK(KP414213562), TR, TM); Chris@10: T1K = VSUB(T1y, T1z); Chris@10: T1A = VADD(T1y, T1z); Chris@10: T1E = VADD(T1q, T1t); Chris@10: T1u = VSUB(T1q, T1t); Chris@10: T1f = VFMA(LDK(KP707106781), T10, TZ); Chris@10: T11 = VFNMS(LDK(KP707106781), T10, TZ); Chris@10: T1c = VFNMS(LDK(KP707106781), Tr, T8); Chris@10: Ts = VFMA(LDK(KP707106781), Tr, T8); Chris@10: T1d = VSUB(T12, T13); Chris@10: T14 = VADD(T12, T13); Chris@10: T1g = VSUB(TS, TH); Chris@10: TT = VADD(TH, TS); Chris@10: { Chris@10: V T1O, T1L, T1F, T1B, T1k, T1e, T19, T15, T1l, T1h, T18, TU, T1T, T1P, T1S; Chris@10: V T1M, T1H, T1G, T1D, T1C, T1m, T1n, T1j, T1i, T1a, T1b, T17, T16, T1U, T1V; Chris@10: V T1R, T1Q; Chris@10: T1O = VSUB(T1K, T1J); Chris@10: T1L = VADD(T1J, T1K); Chris@10: T1F = VADD(T1x, T1A); Chris@10: T1B = VSUB(T1x, T1A); Chris@10: T1k = VFNMS(LDK(KP923879532), T1d, T1c); Chris@10: T1e = VFMA(LDK(KP923879532), T1d, T1c); Chris@10: T19 = VFNMS(LDK(KP923879532), T14, T11); Chris@10: T15 = VFMA(LDK(KP923879532), T14, T11); Chris@10: T1l = VFNMS(LDK(KP923879532), T1g, T1f); Chris@10: T1h = VFMA(LDK(KP923879532), T1g, T1f); Chris@10: T18 = VFNMS(LDK(KP923879532), TT, Ts); Chris@10: TU = VFMA(LDK(KP923879532), TT, Ts); Chris@10: T1T = VFNMS(LDK(KP707106781), T1O, T1N); Chris@10: T1P = VFMA(LDK(KP707106781), T1O, T1N); Chris@10: T1S = VFNMS(LDK(KP707106781), T1L, T1I); Chris@10: T1M = VFMA(LDK(KP707106781), T1L, T1I); Chris@10: T1H = VCONJ(VMUL(LDK(KP500000000), VADD(T1F, T1E))); Chris@10: T1G = VMUL(LDK(KP500000000), VSUB(T1E, T1F)); Chris@10: T1D = VCONJ(VMUL(LDK(KP500000000), VFNMSI(T1B, T1u))); Chris@10: T1C = VMUL(LDK(KP500000000), VFMAI(T1B, T1u)); Chris@10: T1m = VMUL(LDK(KP500000000), VFNMSI(T1l, T1k)); Chris@10: T1n = VCONJ(VMUL(LDK(KP500000000), VFMAI(T1l, T1k))); Chris@10: T1j = VMUL(LDK(KP500000000), VFMAI(T1h, T1e)); Chris@10: T1i = VCONJ(VMUL(LDK(KP500000000), VFNMSI(T1h, T1e))); Chris@10: T1a = VCONJ(VMUL(LDK(KP500000000), VFNMSI(T19, T18))); Chris@10: T1b = VMUL(LDK(KP500000000), VFMAI(T19, T18)); Chris@10: T17 = VCONJ(VMUL(LDK(KP500000000), VFMAI(T15, TU))); Chris@10: T16 = VMUL(LDK(KP500000000), VFNMSI(T15, TU)); Chris@10: T1U = VMUL(LDK(KP500000000), VFNMSI(T1T, T1S)); Chris@10: T1V = VCONJ(VMUL(LDK(KP500000000), VFMAI(T1T, T1S))); Chris@10: T1R = VMUL(LDK(KP500000000), VFMAI(T1P, T1M)); Chris@10: T1Q = VCONJ(VMUL(LDK(KP500000000), VFNMSI(T1P, T1M))); Chris@10: ST(&(Rm[WS(rs, 7)]), T1H, -ms, &(Rm[WS(rs, 1)])); Chris@10: ST(&(Rp[0]), T1G, ms, &(Rp[0])); Chris@10: ST(&(Rm[WS(rs, 3)]), T1D, -ms, &(Rm[WS(rs, 1)])); Chris@10: ST(&(Rp[WS(rs, 4)]), T1C, ms, &(Rp[0])); Chris@10: ST(&(Rp[WS(rs, 5)]), T1m, ms, &(Rp[WS(rs, 1)])); Chris@10: ST(&(Rm[WS(rs, 4)]), T1n, -ms, &(Rm[0])); Chris@10: ST(&(Rp[WS(rs, 3)]), T1j, ms, &(Rp[WS(rs, 1)])); Chris@10: ST(&(Rm[WS(rs, 2)]), T1i, -ms, &(Rm[0])); Chris@10: ST(&(Rm[WS(rs, 6)]), T1a, -ms, &(Rm[0])); Chris@10: ST(&(Rp[WS(rs, 7)]), T1b, ms, &(Rp[WS(rs, 1)])); Chris@10: ST(&(Rm[0]), T17, -ms, &(Rm[0])); Chris@10: ST(&(Rp[WS(rs, 1)]), T16, ms, &(Rp[WS(rs, 1)])); Chris@10: ST(&(Rp[WS(rs, 6)]), T1U, ms, &(Rp[0])); Chris@10: ST(&(Rm[WS(rs, 5)]), T1V, -ms, &(Rm[WS(rs, 1)])); Chris@10: ST(&(Rp[WS(rs, 2)]), T1R, ms, &(Rp[0])); Chris@10: ST(&(Rm[WS(rs, 1)]), T1Q, -ms, &(Rm[WS(rs, 1)])); Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: VTW(1, 1), Chris@10: VTW(1, 2), Chris@10: VTW(1, 3), Chris@10: VTW(1, 4), Chris@10: VTW(1, 5), Chris@10: VTW(1, 6), Chris@10: VTW(1, 7), Chris@10: VTW(1, 8), Chris@10: VTW(1, 9), Chris@10: VTW(1, 10), Chris@10: VTW(1, 11), Chris@10: VTW(1, 12), Chris@10: VTW(1, 13), Chris@10: VTW(1, 14), Chris@10: VTW(1, 15), Chris@10: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 16, XSIMD_STRING("hc2cfdftv_16"), twinstr, &GENUS, {53, 46, 50, 0} }; Chris@10: Chris@10: void XSIMD(codelet_hc2cfdftv_16) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdftv_16, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2cdft_c.native -simd -compact -variables 4 -pipeline-latency 8 -trivial-stores -variables 32 -no-generate-bytw -n 16 -dit -name hc2cfdftv_16 -include hc2cfv.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 103 FP additions, 56 FP multiplications, Chris@10: * (or, 99 additions, 52 multiplications, 4 fused multiply/add), Chris@10: * 101 stack variables, 5 constants, and 32 memory accesses Chris@10: */ Chris@10: #include "hc2cfv.h" Chris@10: Chris@10: static void hc2cfdftv_16(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@10: DVK(KP353553390, +0.353553390593273762200422181052424519642417969); Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DVK(KP382683432, +0.382683432365089771728459984030398866761344562); Chris@10: DVK(KP923879532, +0.923879532511286756128183189396788286822416626); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 30)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 30), MAKE_VOLATILE_STRIDE(64, rs)) { Chris@10: V T1D, T1E, T1R, TP, T1b, Ta, T1w, T18, T1x, T1z, T1A, T1G, T1H, T1S, Tx; Chris@10: V T13, T10, T1a, T1, T3, TA, TM, TL, TN, T6, T8, TC, TH, TG, TI; Chris@10: V T2, Tz, TK, TJ, T7, TB, TF, TE, TD, TO, T4, T9, T5, T15, T17; Chris@10: V T14, T16; Chris@10: T1 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@10: T2 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@10: T3 = VCONJ(T2); Chris@10: Tz = LDW(&(W[0])); Chris@10: TA = VZMULIJ(Tz, VSUB(T3, T1)); Chris@10: TM = LD(&(Rp[WS(rs, 6)]), ms, &(Rp[0])); Chris@10: TK = LD(&(Rm[WS(rs, 6)]), -ms, &(Rm[0])); Chris@10: TL = VCONJ(TK); Chris@10: TJ = LDW(&(W[TWVL * 24])); Chris@10: TN = VZMULIJ(TJ, VSUB(TL, TM)); Chris@10: T6 = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0])); Chris@10: T7 = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0])); Chris@10: T8 = VCONJ(T7); Chris@10: TB = LDW(&(W[TWVL * 16])); Chris@10: TC = VZMULIJ(TB, VSUB(T8, T6)); Chris@10: TH = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@10: TF = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@10: TG = VCONJ(TF); Chris@10: TE = LDW(&(W[TWVL * 8])); Chris@10: TI = VZMULIJ(TE, VSUB(TG, TH)); Chris@10: T1D = VADD(TA, TC); Chris@10: T1E = VADD(TI, TN); Chris@10: T1R = VSUB(T1D, T1E); Chris@10: TD = VSUB(TA, TC); Chris@10: TO = VSUB(TI, TN); Chris@10: TP = VFNMS(LDK(KP382683432), TO, VMUL(LDK(KP923879532), TD)); Chris@10: T1b = VFMA(LDK(KP382683432), TD, VMUL(LDK(KP923879532), TO)); Chris@10: T4 = VADD(T1, T3); Chris@10: T5 = LDW(&(W[TWVL * 14])); Chris@10: T9 = VZMULJ(T5, VADD(T6, T8)); Chris@10: Ta = VMUL(LDK(KP500000000), VSUB(T4, T9)); Chris@10: T1w = VADD(T4, T9); Chris@10: T14 = LDW(&(W[TWVL * 6])); Chris@10: T15 = VZMULJ(T14, VADD(TH, TG)); Chris@10: T16 = LDW(&(W[TWVL * 22])); Chris@10: T17 = VZMULJ(T16, VADD(TM, TL)); Chris@10: T18 = VSUB(T15, T17); Chris@10: T1x = VADD(T15, T17); Chris@10: { Chris@10: V Tf, TR, Tv, TY, Tk, TT, Tq, TW, Tc, Te, Td, Tb, TQ, Ts, Tu; Chris@10: V Tt, Tr, TX, Th, Tj, Ti, Tg, TS, Tn, Tp, To, Tm, TV, Tl, Tw; Chris@10: V TU, TZ; Chris@10: Tc = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Td = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Te = VCONJ(Td); Chris@10: Tb = LDW(&(W[TWVL * 2])); Chris@10: Tf = VZMULJ(Tb, VADD(Tc, Te)); Chris@10: TQ = LDW(&(W[TWVL * 4])); Chris@10: TR = VZMULIJ(TQ, VSUB(Te, Tc)); Chris@10: Ts = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Tt = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Tu = VCONJ(Tt); Chris@10: Tr = LDW(&(W[TWVL * 10])); Chris@10: Tv = VZMULJ(Tr, VADD(Ts, Tu)); Chris@10: TX = LDW(&(W[TWVL * 12])); Chris@10: TY = VZMULIJ(TX, VSUB(Tu, Ts)); Chris@10: Th = LD(&(Rp[WS(rs, 5)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Ti = LD(&(Rm[WS(rs, 5)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Tj = VCONJ(Ti); Chris@10: Tg = LDW(&(W[TWVL * 18])); Chris@10: Tk = VZMULJ(Tg, VADD(Th, Tj)); Chris@10: TS = LDW(&(W[TWVL * 20])); Chris@10: TT = VZMULIJ(TS, VSUB(Tj, Th)); Chris@10: Tn = LD(&(Rp[WS(rs, 7)]), ms, &(Rp[WS(rs, 1)])); Chris@10: To = LD(&(Rm[WS(rs, 7)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Tp = VCONJ(To); Chris@10: Tm = LDW(&(W[TWVL * 26])); Chris@10: Tq = VZMULJ(Tm, VADD(Tn, Tp)); Chris@10: TV = LDW(&(W[TWVL * 28])); Chris@10: TW = VZMULIJ(TV, VSUB(Tp, Tn)); Chris@10: T1z = VADD(Tf, Tk); Chris@10: T1A = VADD(Tq, Tv); Chris@10: T1G = VADD(TR, TT); Chris@10: T1H = VADD(TW, TY); Chris@10: T1S = VSUB(T1H, T1G); Chris@10: Tl = VSUB(Tf, Tk); Chris@10: Tw = VSUB(Tq, Tv); Chris@10: Tx = VMUL(LDK(KP353553390), VADD(Tl, Tw)); Chris@10: T13 = VMUL(LDK(KP707106781), VSUB(Tw, Tl)); Chris@10: TU = VSUB(TR, TT); Chris@10: TZ = VSUB(TW, TY); Chris@10: T10 = VFMA(LDK(KP382683432), TU, VMUL(LDK(KP923879532), TZ)); Chris@10: T1a = VFNMS(LDK(KP923879532), TU, VMUL(LDK(KP382683432), TZ)); Chris@10: } Chris@10: { Chris@10: V T1U, T20, T1X, T21, T1Q, T1T, T1V, T1W, T1Y, T23, T1Z, T22, T1C, T1M, T1J; Chris@10: V T1N, T1y, T1B, T1F, T1I, T1K, T1P, T1L, T1O, T12, T1g, T1d, T1h, Ty, T11; Chris@10: V T19, T1c, T1e, T1j, T1f, T1i, T1m, T1s, T1p, T1t, T1k, T1l, T1n, T1o, T1q; Chris@10: V T1v, T1r, T1u; Chris@10: T1Q = VMUL(LDK(KP500000000), VSUB(T1w, T1x)); Chris@10: T1T = VMUL(LDK(KP353553390), VADD(T1R, T1S)); Chris@10: T1U = VADD(T1Q, T1T); Chris@10: T20 = VSUB(T1Q, T1T); Chris@10: T1V = VSUB(T1A, T1z); Chris@10: T1W = VMUL(LDK(KP707106781), VSUB(T1S, T1R)); Chris@10: T1X = VMUL(LDK(KP500000000), VBYI(VADD(T1V, T1W))); Chris@10: T21 = VMUL(LDK(KP500000000), VBYI(VSUB(T1W, T1V))); Chris@10: T1Y = VCONJ(VSUB(T1U, T1X)); Chris@10: ST(&(Rm[WS(rs, 1)]), T1Y, -ms, &(Rm[WS(rs, 1)])); Chris@10: T23 = VADD(T20, T21); Chris@10: ST(&(Rp[WS(rs, 6)]), T23, ms, &(Rp[0])); Chris@10: T1Z = VADD(T1U, T1X); Chris@10: ST(&(Rp[WS(rs, 2)]), T1Z, ms, &(Rp[0])); Chris@10: T22 = VCONJ(VSUB(T20, T21)); Chris@10: ST(&(Rm[WS(rs, 5)]), T22, -ms, &(Rm[WS(rs, 1)])); Chris@10: T1y = VADD(T1w, T1x); Chris@10: T1B = VADD(T1z, T1A); Chris@10: T1C = VADD(T1y, T1B); Chris@10: T1M = VSUB(T1y, T1B); Chris@10: T1F = VADD(T1D, T1E); Chris@10: T1I = VADD(T1G, T1H); Chris@10: T1J = VADD(T1F, T1I); Chris@10: T1N = VBYI(VSUB(T1I, T1F)); Chris@10: T1K = VCONJ(VMUL(LDK(KP500000000), VSUB(T1C, T1J))); Chris@10: ST(&(Rm[WS(rs, 7)]), T1K, -ms, &(Rm[WS(rs, 1)])); Chris@10: T1P = VMUL(LDK(KP500000000), VADD(T1M, T1N)); Chris@10: ST(&(Rp[WS(rs, 4)]), T1P, ms, &(Rp[0])); Chris@10: T1L = VMUL(LDK(KP500000000), VADD(T1C, T1J)); Chris@10: ST(&(Rp[0]), T1L, ms, &(Rp[0])); Chris@10: T1O = VCONJ(VMUL(LDK(KP500000000), VSUB(T1M, T1N))); Chris@10: ST(&(Rm[WS(rs, 3)]), T1O, -ms, &(Rm[WS(rs, 1)])); Chris@10: Ty = VADD(Ta, Tx); Chris@10: T11 = VMUL(LDK(KP500000000), VADD(TP, T10)); Chris@10: T12 = VADD(Ty, T11); Chris@10: T1g = VSUB(Ty, T11); Chris@10: T19 = VSUB(T13, T18); Chris@10: T1c = VSUB(T1a, T1b); Chris@10: T1d = VMUL(LDK(KP500000000), VBYI(VADD(T19, T1c))); Chris@10: T1h = VMUL(LDK(KP500000000), VBYI(VSUB(T1c, T19))); Chris@10: T1e = VCONJ(VSUB(T12, T1d)); Chris@10: ST(&(Rm[0]), T1e, -ms, &(Rm[0])); Chris@10: T1j = VADD(T1g, T1h); Chris@10: ST(&(Rp[WS(rs, 7)]), T1j, ms, &(Rp[WS(rs, 1)])); Chris@10: T1f = VADD(T12, T1d); Chris@10: ST(&(Rp[WS(rs, 1)]), T1f, ms, &(Rp[WS(rs, 1)])); Chris@10: T1i = VCONJ(VSUB(T1g, T1h)); Chris@10: ST(&(Rm[WS(rs, 6)]), T1i, -ms, &(Rm[0])); Chris@10: T1k = VSUB(T10, TP); Chris@10: T1l = VADD(T18, T13); Chris@10: T1m = VMUL(LDK(KP500000000), VBYI(VSUB(T1k, T1l))); Chris@10: T1s = VMUL(LDK(KP500000000), VBYI(VADD(T1l, T1k))); Chris@10: T1n = VSUB(Ta, Tx); Chris@10: T1o = VMUL(LDK(KP500000000), VADD(T1b, T1a)); Chris@10: T1p = VSUB(T1n, T1o); Chris@10: T1t = VADD(T1n, T1o); Chris@10: T1q = VADD(T1m, T1p); Chris@10: ST(&(Rp[WS(rs, 5)]), T1q, ms, &(Rp[WS(rs, 1)])); Chris@10: T1v = VCONJ(VSUB(T1t, T1s)); Chris@10: ST(&(Rm[WS(rs, 2)]), T1v, -ms, &(Rm[0])); Chris@10: T1r = VCONJ(VSUB(T1p, T1m)); Chris@10: ST(&(Rm[WS(rs, 4)]), T1r, -ms, &(Rm[0])); Chris@10: T1u = VADD(T1s, T1t); Chris@10: ST(&(Rp[WS(rs, 3)]), T1u, ms, &(Rp[WS(rs, 1)])); Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: VTW(1, 1), Chris@10: VTW(1, 2), Chris@10: VTW(1, 3), Chris@10: VTW(1, 4), Chris@10: VTW(1, 5), Chris@10: VTW(1, 6), Chris@10: VTW(1, 7), Chris@10: VTW(1, 8), Chris@10: VTW(1, 9), Chris@10: VTW(1, 10), Chris@10: VTW(1, 11), Chris@10: VTW(1, 12), Chris@10: VTW(1, 13), Chris@10: VTW(1, 14), Chris@10: VTW(1, 15), Chris@10: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 16, XSIMD_STRING("hc2cfdftv_16"), twinstr, &GENUS, {99, 52, 4, 0} }; Chris@10: Chris@10: void XSIMD(codelet_hc2cfdftv_16) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdftv_16, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */