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 12 -dit -name hc2cfdftv_12 -include hc2cfv.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 71 FP additions, 66 FP multiplications, Chris@10: * (or, 41 additions, 36 multiplications, 30 fused multiply/add), Chris@10: * 86 stack variables, 2 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "hc2cfv.h" Chris@10: Chris@10: static void hc2cfdftv_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: 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)) { Chris@10: V T3, T7, TH, TE, Th, TC, Tq, T11, TU, Tx, Tb, Tz, Tu, Tw, Tp; Chris@10: V Tl, T9, Ta, T8, Ty, Tn, To, Tm, TG, T1, T2, Tt, T5, T6, T4; Chris@10: V Tv, Tj, Tk, Ti, TD, Tf, Tg, Te, TB, TT, TF, TR, Tr; Chris@10: T1 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@10: T2 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@10: Tt = LDW(&(W[0])); Chris@10: T5 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@10: T6 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@10: T4 = LDW(&(W[TWVL * 6])); Chris@10: Tv = LDW(&(W[TWVL * 8])); Chris@10: Tn = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@10: To = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: T3 = VFMACONJ(T2, T1); Chris@10: Tu = VZMULIJ(Tt, VFNMSCONJ(T2, T1)); Chris@10: Tm = LDW(&(W[TWVL * 2])); Chris@10: TG = LDW(&(W[TWVL * 4])); Chris@10: T7 = VZMULJ(T4, VFMACONJ(T6, T5)); Chris@10: Tw = VZMULIJ(Tv, VFNMSCONJ(T6, T5)); Chris@10: Tj = LD(&(Rp[WS(rs, 5)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Tk = LD(&(Rm[WS(rs, 5)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Ti = LDW(&(W[TWVL * 18])); Chris@10: TD = LDW(&(W[TWVL * 20])); Chris@10: Tp = VZMULJ(Tm, VFMACONJ(To, Tn)); Chris@10: TH = VZMULIJ(TG, VFNMSCONJ(To, Tn)); Chris@10: Tf = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Tg = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Te = LDW(&(W[TWVL * 10])); Chris@10: TB = LDW(&(W[TWVL * 12])); Chris@10: Tl = VZMULJ(Ti, VFMACONJ(Tk, Tj)); Chris@10: TE = VZMULIJ(TD, VFNMSCONJ(Tk, Tj)); Chris@10: T9 = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0])); Chris@10: Ta = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0])); Chris@10: T8 = LDW(&(W[TWVL * 14])); Chris@10: Ty = LDW(&(W[TWVL * 16])); Chris@10: Th = VZMULJ(Te, VFMACONJ(Tg, Tf)); Chris@10: TC = VZMULIJ(TB, VFNMSCONJ(Tg, Tf)); Chris@10: Tq = VADD(Tl, Tp); Chris@10: T11 = VSUB(Tp, Tl); Chris@10: TU = VSUB(Tu, Tw); Chris@10: Tx = VADD(Tu, Tw); Chris@10: Tb = VZMULJ(T8, VFMACONJ(Ta, T9)); Chris@10: Tz = VZMULIJ(Ty, VFNMSCONJ(Ta, T9)); Chris@10: TT = VSUB(TC, TE); Chris@10: TF = VADD(TC, TE); Chris@10: TR = VFNMS(LDK(KP500000000), Tq, Th); Chris@10: Tr = VADD(Th, Tq); Chris@10: { Chris@10: V TX, TA, T1d, TV, TY, TI, T1e, T12, TQ, Td, T10, Tc, T1a, TN, TJ; Chris@10: V T1j, T1f, T1b, TS, TM, Ts, T17, T13, TZ, T1i, T1c, T16, TW, TP, TO; Chris@10: V TL, TK, T1k, T1l, T1h, T1g, T18, T19, T15, T14; Chris@10: T10 = VSUB(Tb, T7); Chris@10: Tc = VADD(T7, Tb); Chris@10: TX = VFNMS(LDK(KP500000000), Tx, Tz); Chris@10: TA = VADD(Tx, Tz); Chris@10: T1d = VADD(TU, TT); Chris@10: TV = VSUB(TT, TU); Chris@10: TY = VFNMS(LDK(KP500000000), TF, TH); Chris@10: TI = VADD(TF, TH); Chris@10: T1e = VADD(T10, T11); Chris@10: T12 = VSUB(T10, T11); Chris@10: TQ = VFNMS(LDK(KP500000000), Tc, T3); Chris@10: Td = VADD(T3, Tc); Chris@10: T1a = VADD(TX, TY); Chris@10: TZ = VSUB(TX, TY); Chris@10: TN = VADD(TA, TI); Chris@10: TJ = VSUB(TA, TI); Chris@10: T1j = VMUL(LDK(KP866025403), VADD(T1d, T1e)); Chris@10: T1f = VMUL(LDK(KP866025403), VSUB(T1d, T1e)); Chris@10: T1b = VADD(TQ, TR); Chris@10: TS = VSUB(TQ, TR); Chris@10: TM = VADD(Td, Tr); Chris@10: Ts = VSUB(Td, Tr); Chris@10: T17 = VFMA(LDK(KP866025403), T12, TZ); Chris@10: T13 = VFNMS(LDK(KP866025403), T12, TZ); Chris@10: T1i = VSUB(T1b, T1a); Chris@10: T1c = VADD(T1a, T1b); Chris@10: T16 = VFNMS(LDK(KP866025403), TV, TS); Chris@10: TW = VFMA(LDK(KP866025403), TV, TS); Chris@10: TP = VCONJ(VMUL(LDK(KP500000000), VADD(TN, TM))); Chris@10: TO = VMUL(LDK(KP500000000), VSUB(TM, TN)); Chris@10: TL = VCONJ(VMUL(LDK(KP500000000), VFNMSI(TJ, Ts))); Chris@10: TK = VMUL(LDK(KP500000000), VFMAI(TJ, Ts)); Chris@10: T1k = VCONJ(VMUL(LDK(KP500000000), VFNMSI(T1j, T1i))); Chris@10: T1l = VMUL(LDK(KP500000000), VFMAI(T1j, T1i)); Chris@10: T1h = VMUL(LDK(KP500000000), VFMAI(T1f, T1c)); Chris@10: T1g = VCONJ(VMUL(LDK(KP500000000), VFNMSI(T1f, T1c))); Chris@10: T18 = VMUL(LDK(KP500000000), VFNMSI(T17, T16)); Chris@10: T19 = VCONJ(VMUL(LDK(KP500000000), VFMAI(T17, T16))); Chris@10: T15 = VCONJ(VMUL(LDK(KP500000000), VFMAI(T13, TW))); Chris@10: T14 = VMUL(LDK(KP500000000), VFNMSI(T13, TW)); Chris@10: ST(&(Rm[WS(rs, 5)]), TP, -ms, &(Rm[WS(rs, 1)])); Chris@10: ST(&(Rp[0]), TO, ms, &(Rp[0])); Chris@10: ST(&(Rm[WS(rs, 2)]), TL, -ms, &(Rm[0])); Chris@10: ST(&(Rp[WS(rs, 3)]), TK, ms, &(Rp[WS(rs, 1)])); Chris@10: ST(&(Rm[WS(rs, 3)]), T1k, -ms, &(Rm[WS(rs, 1)])); Chris@10: ST(&(Rp[WS(rs, 4)]), T1l, ms, &(Rp[0])); Chris@10: ST(&(Rp[WS(rs, 2)]), T1h, ms, &(Rp[0])); Chris@10: ST(&(Rm[WS(rs, 1)]), T1g, -ms, &(Rm[WS(rs, 1)])); Chris@10: ST(&(Rp[WS(rs, 5)]), T18, ms, &(Rp[WS(rs, 1)])); Chris@10: ST(&(Rm[WS(rs, 4)]), T19, -ms, &(Rm[0])); Chris@10: ST(&(Rm[0]), T15, -ms, &(Rm[0])); Chris@10: ST(&(Rp[WS(rs, 1)]), T14, 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: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 12, XSIMD_STRING("hc2cfdftv_12"), twinstr, &GENUS, {41, 36, 30, 0} }; Chris@10: Chris@10: void XSIMD(codelet_hc2cfdftv_12) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdftv_12, &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 12 -dit -name hc2cfdftv_12 -include hc2cfv.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 71 FP additions, 41 FP multiplications, Chris@10: * (or, 67 additions, 37 multiplications, 4 fused multiply/add), Chris@10: * 58 stack variables, 4 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "hc2cfv.h" Chris@10: Chris@10: static void hc2cfdftv_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DVK(KP433012701, +0.433012701892219323381861585376468091735701313); Chris@10: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: 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)) { Chris@10: V TX, T13, T4, Tf, TZ, TD, TF, T17, TW, T14, Tw, Tl, T10, TL, TN; Chris@10: V T16; Chris@10: { Chris@10: V T1, T3, TA, Tb, Td, Te, T9, TC, T2, Tz, Tc, Ta, T6, T8, T7; Chris@10: V T5, TB, TE, Ti, Tk, TI, Ts, Tu, Tv, Tq, TK, Tj, TH, Tt, Tr; Chris@10: V Tn, Tp, To, Tm, TJ, Th, TM; 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: Tb = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0])); Chris@10: Tc = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0])); Chris@10: Td = VCONJ(Tc); Chris@10: Ta = LDW(&(W[TWVL * 14])); Chris@10: Te = VZMULJ(Ta, VADD(Tb, Td)); Chris@10: T6 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@10: T7 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@10: T8 = VCONJ(T7); Chris@10: T5 = LDW(&(W[TWVL * 6])); Chris@10: T9 = VZMULJ(T5, VADD(T6, T8)); Chris@10: TB = LDW(&(W[TWVL * 8])); Chris@10: TC = VZMULIJ(TB, VSUB(T8, T6)); Chris@10: TX = VSUB(TC, TA); Chris@10: T13 = VSUB(Te, T9); Chris@10: T4 = VADD(T1, T3); Chris@10: Tf = VADD(T9, Te); Chris@10: TZ = VFNMS(LDK(KP250000000), Tf, VMUL(LDK(KP500000000), T4)); Chris@10: TD = VADD(TA, TC); Chris@10: TE = LDW(&(W[TWVL * 16])); Chris@10: TF = VZMULIJ(TE, VSUB(Td, Tb)); Chris@10: T17 = VFNMS(LDK(KP500000000), TD, TF); Chris@10: Ti = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Tj = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Tk = VCONJ(Tj); Chris@10: TH = LDW(&(W[TWVL * 12])); Chris@10: TI = VZMULIJ(TH, VSUB(Tk, Ti)); Chris@10: Ts = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Tt = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Tu = VCONJ(Tt); Chris@10: Tr = LDW(&(W[TWVL * 2])); Chris@10: Tv = VZMULJ(Tr, VADD(Ts, Tu)); Chris@10: Tn = LD(&(Rp[WS(rs, 5)]), ms, &(Rp[WS(rs, 1)])); Chris@10: To = LD(&(Rm[WS(rs, 5)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Tp = VCONJ(To); Chris@10: Tm = LDW(&(W[TWVL * 18])); Chris@10: Tq = VZMULJ(Tm, VADD(Tn, Tp)); Chris@10: TJ = LDW(&(W[TWVL * 20])); Chris@10: TK = VZMULIJ(TJ, VSUB(Tp, Tn)); Chris@10: TW = VSUB(TK, TI); Chris@10: T14 = VSUB(Tv, Tq); Chris@10: Tw = VADD(Tq, Tv); Chris@10: Th = LDW(&(W[TWVL * 10])); Chris@10: Tl = VZMULJ(Th, VADD(Ti, Tk)); Chris@10: T10 = VFNMS(LDK(KP250000000), Tw, VMUL(LDK(KP500000000), Tl)); Chris@10: TL = VADD(TI, TK); Chris@10: TM = LDW(&(W[TWVL * 4])); Chris@10: TN = VZMULIJ(TM, VSUB(Tu, Ts)); Chris@10: T16 = VFNMS(LDK(KP500000000), TL, TN); Chris@10: } Chris@10: { Chris@10: V Ty, TS, TP, TT, Tg, Tx, TG, TO, TQ, TV, TR, TU, T1i, T1o, T1l; Chris@10: V T1p, T1g, T1h, T1j, T1k, T1m, T1r, T1n, T1q, T12, T1c, T19, T1d, TY, T11; Chris@10: V T15, T18, T1a, T1f, T1b, T1e; Chris@10: Tg = VADD(T4, Tf); Chris@10: Tx = VADD(Tl, Tw); Chris@10: Ty = VADD(Tg, Tx); Chris@10: TS = VSUB(Tg, Tx); Chris@10: TG = VADD(TD, TF); Chris@10: TO = VADD(TL, TN); Chris@10: TP = VADD(TG, TO); Chris@10: TT = VBYI(VSUB(TO, TG)); Chris@10: TQ = VCONJ(VMUL(LDK(KP500000000), VSUB(Ty, TP))); Chris@10: ST(&(Rm[WS(rs, 5)]), TQ, -ms, &(Rm[WS(rs, 1)])); Chris@10: TV = VMUL(LDK(KP500000000), VADD(TS, TT)); Chris@10: ST(&(Rp[WS(rs, 3)]), TV, ms, &(Rp[WS(rs, 1)])); Chris@10: TR = VMUL(LDK(KP500000000), VADD(Ty, TP)); Chris@10: ST(&(Rp[0]), TR, ms, &(Rp[0])); Chris@10: TU = VCONJ(VMUL(LDK(KP500000000), VSUB(TS, TT))); Chris@10: ST(&(Rm[WS(rs, 2)]), TU, -ms, &(Rm[0])); Chris@10: T1g = VADD(TX, TW); Chris@10: T1h = VADD(T13, T14); Chris@10: T1i = VMUL(LDK(KP500000000), VBYI(VMUL(LDK(KP866025403), VSUB(T1g, T1h)))); Chris@10: T1o = VMUL(LDK(KP500000000), VBYI(VMUL(LDK(KP866025403), VADD(T1g, T1h)))); Chris@10: T1j = VADD(TZ, T10); Chris@10: T1k = VMUL(LDK(KP500000000), VADD(T17, T16)); Chris@10: T1l = VSUB(T1j, T1k); Chris@10: T1p = VADD(T1j, T1k); Chris@10: T1m = VADD(T1i, T1l); Chris@10: ST(&(Rp[WS(rs, 2)]), T1m, ms, &(Rp[0])); Chris@10: T1r = VCONJ(VSUB(T1p, T1o)); Chris@10: ST(&(Rm[WS(rs, 3)]), T1r, -ms, &(Rm[WS(rs, 1)])); Chris@10: T1n = VCONJ(VSUB(T1l, T1i)); Chris@10: ST(&(Rm[WS(rs, 1)]), T1n, -ms, &(Rm[WS(rs, 1)])); Chris@10: T1q = VADD(T1o, T1p); Chris@10: ST(&(Rp[WS(rs, 4)]), T1q, ms, &(Rp[0])); Chris@10: TY = VMUL(LDK(KP433012701), VSUB(TW, TX)); Chris@10: T11 = VSUB(TZ, T10); Chris@10: T12 = VADD(TY, T11); Chris@10: T1c = VSUB(T11, TY); Chris@10: T15 = VMUL(LDK(KP866025403), VSUB(T13, T14)); Chris@10: T18 = VSUB(T16, T17); Chris@10: T19 = VMUL(LDK(KP500000000), VBYI(VSUB(T15, T18))); Chris@10: T1d = VMUL(LDK(KP500000000), VBYI(VADD(T15, T18))); Chris@10: T1a = VCONJ(VSUB(T12, T19)); Chris@10: ST(&(Rm[0]), T1a, -ms, &(Rm[0])); Chris@10: T1f = VCONJ(VADD(T1c, T1d)); Chris@10: ST(&(Rm[WS(rs, 4)]), T1f, -ms, &(Rm[0])); Chris@10: T1b = VADD(T12, T19); Chris@10: ST(&(Rp[WS(rs, 1)]), T1b, ms, &(Rp[WS(rs, 1)])); Chris@10: T1e = VSUB(T1c, T1d); Chris@10: ST(&(Rp[WS(rs, 5)]), T1e, 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: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 12, XSIMD_STRING("hc2cfdftv_12"), twinstr, &GENUS, {67, 37, 4, 0} }; Chris@10: Chris@10: void XSIMD(codelet_hc2cfdftv_12) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdftv_12, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */