cannam@95: /* cannam@95: * Copyright (c) 2003, 2007-11 Matteo Frigo cannam@95: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology cannam@95: * cannam@95: * This program is free software; you can redistribute it and/or modify cannam@95: * it under the terms of the GNU General Public License as published by cannam@95: * the Free Software Foundation; either version 2 of the License, or cannam@95: * (at your option) any later version. cannam@95: * cannam@95: * This program is distributed in the hope that it will be useful, cannam@95: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@95: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@95: * GNU General Public License for more details. cannam@95: * cannam@95: * You should have received a copy of the GNU General Public License cannam@95: * along with this program; if not, write to the Free Software cannam@95: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@95: * cannam@95: */ cannam@95: cannam@95: /* This file was automatically generated --- DO NOT EDIT */ cannam@95: /* Generated on Sun Nov 25 07:42:29 EST 2012 */ cannam@95: cannam@95: #include "codelet-rdft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* 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 12 -dit -name hc2cfdftv_12 -include hc2cfv.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 71 FP additions, 66 FP multiplications, cannam@95: * (or, 41 additions, 36 multiplications, 30 fused multiply/add), cannam@95: * 86 stack variables, 2 constants, and 24 memory accesses cannam@95: */ cannam@95: #include "hc2cfv.h" cannam@95: cannam@95: static void hc2cfdftv_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 22)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 22), MAKE_VOLATILE_STRIDE(48, rs)) { cannam@95: V T3, T7, TH, TE, Th, TC, Tq, T11, TU, Tx, Tb, Tz, Tu, Tw, Tp; cannam@95: V Tl, T9, Ta, T8, Ty, Tn, To, Tm, TG, T1, T2, Tt, T5, T6, T4; cannam@95: V Tv, Tj, Tk, Ti, TD, Tf, Tg, Te, TB, TT, TF, TR, Tr; cannam@95: T1 = LD(&(Rp[0]), ms, &(Rp[0])); cannam@95: T2 = LD(&(Rm[0]), -ms, &(Rm[0])); cannam@95: Tt = LDW(&(W[0])); cannam@95: T5 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); cannam@95: T6 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); cannam@95: T4 = LDW(&(W[TWVL * 6])); cannam@95: Tv = LDW(&(W[TWVL * 8])); cannam@95: Tn = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); cannam@95: To = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: T3 = VFMACONJ(T2, T1); cannam@95: Tu = VZMULIJ(Tt, VFNMSCONJ(T2, T1)); cannam@95: Tm = LDW(&(W[TWVL * 2])); cannam@95: TG = LDW(&(W[TWVL * 4])); cannam@95: T7 = VZMULJ(T4, VFMACONJ(T6, T5)); cannam@95: Tw = VZMULIJ(Tv, VFNMSCONJ(T6, T5)); cannam@95: Tj = LD(&(Rp[WS(rs, 5)]), ms, &(Rp[WS(rs, 1)])); cannam@95: Tk = LD(&(Rm[WS(rs, 5)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: Ti = LDW(&(W[TWVL * 18])); cannam@95: TD = LDW(&(W[TWVL * 20])); cannam@95: Tp = VZMULJ(Tm, VFMACONJ(To, Tn)); cannam@95: TH = VZMULIJ(TG, VFNMSCONJ(To, Tn)); cannam@95: Tf = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); cannam@95: Tg = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: Te = LDW(&(W[TWVL * 10])); cannam@95: TB = LDW(&(W[TWVL * 12])); cannam@95: Tl = VZMULJ(Ti, VFMACONJ(Tk, Tj)); cannam@95: TE = VZMULIJ(TD, VFNMSCONJ(Tk, Tj)); cannam@95: T9 = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0])); cannam@95: Ta = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0])); cannam@95: T8 = LDW(&(W[TWVL * 14])); cannam@95: Ty = LDW(&(W[TWVL * 16])); cannam@95: Th = VZMULJ(Te, VFMACONJ(Tg, Tf)); cannam@95: TC = VZMULIJ(TB, VFNMSCONJ(Tg, Tf)); cannam@95: Tq = VADD(Tl, Tp); cannam@95: T11 = VSUB(Tp, Tl); cannam@95: TU = VSUB(Tu, Tw); cannam@95: Tx = VADD(Tu, Tw); cannam@95: Tb = VZMULJ(T8, VFMACONJ(Ta, T9)); cannam@95: Tz = VZMULIJ(Ty, VFNMSCONJ(Ta, T9)); cannam@95: TT = VSUB(TC, TE); cannam@95: TF = VADD(TC, TE); cannam@95: TR = VFNMS(LDK(KP500000000), Tq, Th); cannam@95: Tr = VADD(Th, Tq); cannam@95: { cannam@95: V TX, TA, T1d, TV, TY, TI, T1e, T12, TQ, Td, T10, Tc, T1a, TN, TJ; cannam@95: V T1j, T1f, T1b, TS, TM, Ts, T17, T13, TZ, T1i, T1c, T16, TW, TP, TO; cannam@95: V TL, TK, T1k, T1l, T1h, T1g, T18, T19, T15, T14; cannam@95: T10 = VSUB(Tb, T7); cannam@95: Tc = VADD(T7, Tb); cannam@95: TX = VFNMS(LDK(KP500000000), Tx, Tz); cannam@95: TA = VADD(Tx, Tz); cannam@95: T1d = VADD(TU, TT); cannam@95: TV = VSUB(TT, TU); cannam@95: TY = VFNMS(LDK(KP500000000), TF, TH); cannam@95: TI = VADD(TF, TH); cannam@95: T1e = VADD(T10, T11); cannam@95: T12 = VSUB(T10, T11); cannam@95: TQ = VFNMS(LDK(KP500000000), Tc, T3); cannam@95: Td = VADD(T3, Tc); cannam@95: T1a = VADD(TX, TY); cannam@95: TZ = VSUB(TX, TY); cannam@95: TN = VADD(TA, TI); cannam@95: TJ = VSUB(TA, TI); cannam@95: T1j = VMUL(LDK(KP866025403), VADD(T1d, T1e)); cannam@95: T1f = VMUL(LDK(KP866025403), VSUB(T1d, T1e)); cannam@95: T1b = VADD(TQ, TR); cannam@95: TS = VSUB(TQ, TR); cannam@95: TM = VADD(Td, Tr); cannam@95: Ts = VSUB(Td, Tr); cannam@95: T17 = VFMA(LDK(KP866025403), T12, TZ); cannam@95: T13 = VFNMS(LDK(KP866025403), T12, TZ); cannam@95: T1i = VSUB(T1b, T1a); cannam@95: T1c = VADD(T1a, T1b); cannam@95: T16 = VFNMS(LDK(KP866025403), TV, TS); cannam@95: TW = VFMA(LDK(KP866025403), TV, TS); cannam@95: TP = VCONJ(VMUL(LDK(KP500000000), VADD(TN, TM))); cannam@95: TO = VMUL(LDK(KP500000000), VSUB(TM, TN)); cannam@95: TL = VCONJ(VMUL(LDK(KP500000000), VFNMSI(TJ, Ts))); cannam@95: TK = VMUL(LDK(KP500000000), VFMAI(TJ, Ts)); cannam@95: T1k = VCONJ(VMUL(LDK(KP500000000), VFNMSI(T1j, T1i))); cannam@95: T1l = VMUL(LDK(KP500000000), VFMAI(T1j, T1i)); cannam@95: T1h = VMUL(LDK(KP500000000), VFMAI(T1f, T1c)); cannam@95: T1g = VCONJ(VMUL(LDK(KP500000000), VFNMSI(T1f, T1c))); cannam@95: T18 = VMUL(LDK(KP500000000), VFNMSI(T17, T16)); cannam@95: T19 = VCONJ(VMUL(LDK(KP500000000), VFMAI(T17, T16))); cannam@95: T15 = VCONJ(VMUL(LDK(KP500000000), VFMAI(T13, TW))); cannam@95: T14 = VMUL(LDK(KP500000000), VFNMSI(T13, TW)); cannam@95: ST(&(Rm[WS(rs, 5)]), TP, -ms, &(Rm[WS(rs, 1)])); cannam@95: ST(&(Rp[0]), TO, ms, &(Rp[0])); cannam@95: ST(&(Rm[WS(rs, 2)]), TL, -ms, &(Rm[0])); cannam@95: ST(&(Rp[WS(rs, 3)]), TK, ms, &(Rp[WS(rs, 1)])); cannam@95: ST(&(Rm[WS(rs, 3)]), T1k, -ms, &(Rm[WS(rs, 1)])); cannam@95: ST(&(Rp[WS(rs, 4)]), T1l, ms, &(Rp[0])); cannam@95: ST(&(Rp[WS(rs, 2)]), T1h, ms, &(Rp[0])); cannam@95: ST(&(Rm[WS(rs, 1)]), T1g, -ms, &(Rm[WS(rs, 1)])); cannam@95: ST(&(Rp[WS(rs, 5)]), T18, ms, &(Rp[WS(rs, 1)])); cannam@95: ST(&(Rm[WS(rs, 4)]), T19, -ms, &(Rm[0])); cannam@95: ST(&(Rm[0]), T15, -ms, &(Rm[0])); cannam@95: ST(&(Rp[WS(rs, 1)]), T14, ms, &(Rp[WS(rs, 1)])); cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(1, 1), cannam@95: VTW(1, 2), cannam@95: VTW(1, 3), cannam@95: VTW(1, 4), cannam@95: VTW(1, 5), cannam@95: VTW(1, 6), cannam@95: VTW(1, 7), cannam@95: VTW(1, 8), cannam@95: VTW(1, 9), cannam@95: VTW(1, 10), cannam@95: VTW(1, 11), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 12, XSIMD_STRING("hc2cfdftv_12"), twinstr, &GENUS, {41, 36, 30, 0} }; cannam@95: cannam@95: void XSIMD(codelet_hc2cfdftv_12) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cfdftv_12, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2cdft_c.native -simd -compact -variables 4 -pipeline-latency 8 -trivial-stores -variables 32 -no-generate-bytw -n 12 -dit -name hc2cfdftv_12 -include hc2cfv.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 71 FP additions, 41 FP multiplications, cannam@95: * (or, 67 additions, 37 multiplications, 4 fused multiply/add), cannam@95: * 58 stack variables, 4 constants, and 24 memory accesses cannam@95: */ cannam@95: #include "hc2cfv.h" cannam@95: cannam@95: static void hc2cfdftv_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP433012701, +0.433012701892219323381861585376468091735701313); cannam@95: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 22)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 22), MAKE_VOLATILE_STRIDE(48, rs)) { cannam@95: V TX, T13, T4, Tf, TZ, TD, TF, T17, TW, T14, Tw, Tl, T10, TL, TN; cannam@95: V T16; cannam@95: { cannam@95: V T1, T3, TA, Tb, Td, Te, T9, TC, T2, Tz, Tc, Ta, T6, T8, T7; cannam@95: V T5, TB, TE, Ti, Tk, TI, Ts, Tu, Tv, Tq, TK, Tj, TH, Tt, Tr; cannam@95: V Tn, Tp, To, Tm, TJ, Th, TM; cannam@95: T1 = LD(&(Rp[0]), ms, &(Rp[0])); cannam@95: T2 = LD(&(Rm[0]), -ms, &(Rm[0])); cannam@95: T3 = VCONJ(T2); cannam@95: Tz = LDW(&(W[0])); cannam@95: TA = VZMULIJ(Tz, VSUB(T3, T1)); cannam@95: Tb = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0])); cannam@95: Tc = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0])); cannam@95: Td = VCONJ(Tc); cannam@95: Ta = LDW(&(W[TWVL * 14])); cannam@95: Te = VZMULJ(Ta, VADD(Tb, Td)); cannam@95: T6 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); cannam@95: T7 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); cannam@95: T8 = VCONJ(T7); cannam@95: T5 = LDW(&(W[TWVL * 6])); cannam@95: T9 = VZMULJ(T5, VADD(T6, T8)); cannam@95: TB = LDW(&(W[TWVL * 8])); cannam@95: TC = VZMULIJ(TB, VSUB(T8, T6)); cannam@95: TX = VSUB(TC, TA); cannam@95: T13 = VSUB(Te, T9); cannam@95: T4 = VADD(T1, T3); cannam@95: Tf = VADD(T9, Te); cannam@95: TZ = VFNMS(LDK(KP250000000), Tf, VMUL(LDK(KP500000000), T4)); cannam@95: TD = VADD(TA, TC); cannam@95: TE = LDW(&(W[TWVL * 16])); cannam@95: TF = VZMULIJ(TE, VSUB(Td, Tb)); cannam@95: T17 = VFNMS(LDK(KP500000000), TD, TF); cannam@95: Ti = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); cannam@95: Tj = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: Tk = VCONJ(Tj); cannam@95: TH = LDW(&(W[TWVL * 12])); cannam@95: TI = VZMULIJ(TH, VSUB(Tk, Ti)); cannam@95: Ts = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); cannam@95: Tt = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: Tu = VCONJ(Tt); cannam@95: Tr = LDW(&(W[TWVL * 2])); cannam@95: Tv = VZMULJ(Tr, VADD(Ts, Tu)); cannam@95: Tn = LD(&(Rp[WS(rs, 5)]), ms, &(Rp[WS(rs, 1)])); cannam@95: To = LD(&(Rm[WS(rs, 5)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: Tp = VCONJ(To); cannam@95: Tm = LDW(&(W[TWVL * 18])); cannam@95: Tq = VZMULJ(Tm, VADD(Tn, Tp)); cannam@95: TJ = LDW(&(W[TWVL * 20])); cannam@95: TK = VZMULIJ(TJ, VSUB(Tp, Tn)); cannam@95: TW = VSUB(TK, TI); cannam@95: T14 = VSUB(Tv, Tq); cannam@95: Tw = VADD(Tq, Tv); cannam@95: Th = LDW(&(W[TWVL * 10])); cannam@95: Tl = VZMULJ(Th, VADD(Ti, Tk)); cannam@95: T10 = VFNMS(LDK(KP250000000), Tw, VMUL(LDK(KP500000000), Tl)); cannam@95: TL = VADD(TI, TK); cannam@95: TM = LDW(&(W[TWVL * 4])); cannam@95: TN = VZMULIJ(TM, VSUB(Tu, Ts)); cannam@95: T16 = VFNMS(LDK(KP500000000), TL, TN); cannam@95: } cannam@95: { cannam@95: V Ty, TS, TP, TT, Tg, Tx, TG, TO, TQ, TV, TR, TU, T1i, T1o, T1l; cannam@95: V T1p, T1g, T1h, T1j, T1k, T1m, T1r, T1n, T1q, T12, T1c, T19, T1d, TY, T11; cannam@95: V T15, T18, T1a, T1f, T1b, T1e; cannam@95: Tg = VADD(T4, Tf); cannam@95: Tx = VADD(Tl, Tw); cannam@95: Ty = VADD(Tg, Tx); cannam@95: TS = VSUB(Tg, Tx); cannam@95: TG = VADD(TD, TF); cannam@95: TO = VADD(TL, TN); cannam@95: TP = VADD(TG, TO); cannam@95: TT = VBYI(VSUB(TO, TG)); cannam@95: TQ = VCONJ(VMUL(LDK(KP500000000), VSUB(Ty, TP))); cannam@95: ST(&(Rm[WS(rs, 5)]), TQ, -ms, &(Rm[WS(rs, 1)])); cannam@95: TV = VMUL(LDK(KP500000000), VADD(TS, TT)); cannam@95: ST(&(Rp[WS(rs, 3)]), TV, ms, &(Rp[WS(rs, 1)])); cannam@95: TR = VMUL(LDK(KP500000000), VADD(Ty, TP)); cannam@95: ST(&(Rp[0]), TR, ms, &(Rp[0])); cannam@95: TU = VCONJ(VMUL(LDK(KP500000000), VSUB(TS, TT))); cannam@95: ST(&(Rm[WS(rs, 2)]), TU, -ms, &(Rm[0])); cannam@95: T1g = VADD(TX, TW); cannam@95: T1h = VADD(T13, T14); cannam@95: T1i = VMUL(LDK(KP500000000), VBYI(VMUL(LDK(KP866025403), VSUB(T1g, T1h)))); cannam@95: T1o = VMUL(LDK(KP500000000), VBYI(VMUL(LDK(KP866025403), VADD(T1g, T1h)))); cannam@95: T1j = VADD(TZ, T10); cannam@95: T1k = VMUL(LDK(KP500000000), VADD(T17, T16)); cannam@95: T1l = VSUB(T1j, T1k); cannam@95: T1p = VADD(T1j, T1k); cannam@95: T1m = VADD(T1i, T1l); cannam@95: ST(&(Rp[WS(rs, 2)]), T1m, ms, &(Rp[0])); cannam@95: T1r = VCONJ(VSUB(T1p, T1o)); cannam@95: ST(&(Rm[WS(rs, 3)]), T1r, -ms, &(Rm[WS(rs, 1)])); cannam@95: T1n = VCONJ(VSUB(T1l, T1i)); cannam@95: ST(&(Rm[WS(rs, 1)]), T1n, -ms, &(Rm[WS(rs, 1)])); cannam@95: T1q = VADD(T1o, T1p); cannam@95: ST(&(Rp[WS(rs, 4)]), T1q, ms, &(Rp[0])); cannam@95: TY = VMUL(LDK(KP433012701), VSUB(TW, TX)); cannam@95: T11 = VSUB(TZ, T10); cannam@95: T12 = VADD(TY, T11); cannam@95: T1c = VSUB(T11, TY); cannam@95: T15 = VMUL(LDK(KP866025403), VSUB(T13, T14)); cannam@95: T18 = VSUB(T16, T17); cannam@95: T19 = VMUL(LDK(KP500000000), VBYI(VSUB(T15, T18))); cannam@95: T1d = VMUL(LDK(KP500000000), VBYI(VADD(T15, T18))); cannam@95: T1a = VCONJ(VSUB(T12, T19)); cannam@95: ST(&(Rm[0]), T1a, -ms, &(Rm[0])); cannam@95: T1f = VCONJ(VADD(T1c, T1d)); cannam@95: ST(&(Rm[WS(rs, 4)]), T1f, -ms, &(Rm[0])); cannam@95: T1b = VADD(T12, T19); cannam@95: ST(&(Rp[WS(rs, 1)]), T1b, ms, &(Rp[WS(rs, 1)])); cannam@95: T1e = VSUB(T1c, T1d); cannam@95: ST(&(Rp[WS(rs, 5)]), T1e, ms, &(Rp[WS(rs, 1)])); cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(1, 1), cannam@95: VTW(1, 2), cannam@95: VTW(1, 3), cannam@95: VTW(1, 4), cannam@95: VTW(1, 5), cannam@95: VTW(1, 6), cannam@95: VTW(1, 7), cannam@95: VTW(1, 8), cannam@95: VTW(1, 9), cannam@95: VTW(1, 10), cannam@95: VTW(1, 11), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 12, XSIMD_STRING("hc2cfdftv_12"), twinstr, &GENUS, {67, 37, 4, 0} }; cannam@95: cannam@95: void XSIMD(codelet_hc2cfdftv_12) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cfdftv_12, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */