Chris@42: /* Chris@42: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@42: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@42: * Chris@42: * This program is free software; you can redistribute it and/or modify Chris@42: * it under the terms of the GNU General Public License as published by Chris@42: * the Free Software Foundation; either version 2 of the License, or Chris@42: * (at your option) any later version. Chris@42: * Chris@42: * This program is distributed in the hope that it will be useful, Chris@42: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@42: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@42: * GNU General Public License for more details. Chris@42: * Chris@42: * You should have received a copy of the GNU General Public License Chris@42: * along with this program; if not, write to the Free Software Chris@42: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@42: * Chris@42: */ Chris@42: Chris@42: /* This file was automatically generated --- DO NOT EDIT */ Chris@42: /* Generated on Sat Jul 30 16:52:44 EDT 2016 */ Chris@42: Chris@42: #include "codelet-rdft.h" Chris@42: Chris@42: #ifdef HAVE_FMA Chris@42: Chris@42: /* 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 -dif -sign 1 -name hc2cbdftv_12 -include hc2cbv.h */ Chris@42: Chris@42: /* Chris@42: * This function contains 71 FP additions, 51 FP multiplications, Chris@42: * (or, 45 additions, 25 multiplications, 26 fused multiply/add), Chris@42: * 88 stack variables, 2 constants, and 24 memory accesses Chris@42: */ Chris@42: #include "hc2cbv.h" Chris@42: Chris@42: static void hc2cbdftv_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@42: { Chris@42: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@42: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@42: { Chris@42: INT m; Chris@42: 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@42: V Tz, TT, T1, T1j, TN, TF, TP, TL, Tx, T15, TJ, T1b, T1g, T1l, T18; Chris@42: V T12, TO, TC, TK, Tl, T16, TQ, TU, TG, T1c, TM, T1k, Ty, T19, T1a; Chris@42: V T13, T14, T1h, T1i, TS, TR, T1m, T1n, TI, TH; Chris@42: { Chris@42: V T2, Tm, T7, Tp, T8, Tq, T9, Tu, T5, Tr, Tg, Tn, Tj, Ta, T3; Chris@42: V T4, Te, Tf, Th, Ti, TV, T6, TW, Tk, TD, Tt, TB, T11, T1f, Tw; Chris@42: V TE, TX, Tc, Ts, T10, TZ, To, Tb, Tv, T17, T1d, T1e, TY, TA, Td; Chris@42: T2 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@42: Tm = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); Chris@42: T7 = LD(&(Rm[WS(rs, 5)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: Tp = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@42: T3 = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0])); Chris@42: T4 = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: Te = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@42: Tf = LD(&(Rp[WS(rs, 5)]), ms, &(Rp[WS(rs, 1)])); Chris@42: Th = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@42: Ti = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0])); Chris@42: T8 = VCONJ(T7); Chris@42: Tq = VCONJ(Tp); Chris@42: T9 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@42: Tu = VFNMSCONJ(T4, T3); Chris@42: T5 = VFMACONJ(T4, T3); Chris@42: Tr = VADD(Te, Tf); Chris@42: Tg = VSUB(Te, Tf); Chris@42: Tn = VADD(Ti, Th); Chris@42: Tj = VSUB(Th, Ti); Chris@42: Ta = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: TV = LDW(&(W[TWVL * 4])); Chris@42: Tz = LDW(&(W[TWVL * 18])); Chris@42: T6 = VFNMS(LDK(KP500000000), T5, T2); Chris@42: TW = VADD(T2, T5); Chris@42: Ts = VFNMS(LDK(KP500000000), Tr, Tq); Chris@42: T10 = VFMACONJ(Tp, Tr); Chris@42: TZ = VFMACONJ(Tn, Tm); Chris@42: To = VFNMS(LDK(KP500000000), VCONJ(Tn), Tm); Chris@42: Tk = VFMACONJ(Tj, Tg); Chris@42: TD = VFNMSCONJ(Tj, Tg); Chris@42: Tb = VFMACONJ(Ta, T9); Chris@42: Tv = VFMSCONJ(Ta, T9); Chris@42: TT = LDW(&(W[TWVL * 2])); Chris@42: T1 = LDW(&(W[TWVL * 20])); Chris@42: Tt = VSUB(To, Ts); Chris@42: TB = VADD(To, Ts); Chris@42: T11 = VSUB(TZ, T10); Chris@42: T1f = VADD(TZ, T10); Chris@42: Tw = VSUB(Tu, Tv); Chris@42: TE = VADD(Tu, Tv); Chris@42: TX = VFMACONJ(T7, Tb); Chris@42: Tc = VFNMS(LDK(KP500000000), Tb, T8); Chris@42: T1j = LDW(&(W[0])); Chris@42: T17 = LDW(&(W[TWVL * 16])); Chris@42: T1d = LDW(&(W[TWVL * 10])); Chris@42: TN = LDW(&(W[TWVL * 6])); Chris@42: TF = VMUL(LDK(KP866025403), VSUB(TD, TE)); Chris@42: TP = VMUL(LDK(KP866025403), VADD(TE, TD)); Chris@42: TL = VFNMS(LDK(KP866025403), Tw, Tt); Chris@42: Tx = VFMA(LDK(KP866025403), Tw, Tt); Chris@42: T1e = VADD(TW, TX); Chris@42: TY = VSUB(TW, TX); Chris@42: TA = VADD(T6, Tc); Chris@42: Td = VSUB(T6, Tc); Chris@42: T15 = LDW(&(W[TWVL * 14])); Chris@42: TJ = LDW(&(W[TWVL * 8])); Chris@42: T1b = LDW(&(W[TWVL * 12])); Chris@42: T1g = VZMUL(T1d, VSUB(T1e, T1f)); Chris@42: T1l = VADD(T1e, T1f); Chris@42: T18 = VZMULI(T17, VFMAI(T11, TY)); Chris@42: T12 = VZMULI(TV, VFNMSI(T11, TY)); Chris@42: TO = VADD(TA, TB); Chris@42: TC = VSUB(TA, TB); Chris@42: TK = VFNMS(LDK(KP866025403), Tk, Td); Chris@42: Tl = VFMA(LDK(KP866025403), Tk, Td); Chris@42: } Chris@42: T16 = VZMUL(T15, VFNMSI(TP, TO)); Chris@42: TQ = VZMUL(TN, VFMAI(TP, TO)); Chris@42: TU = VZMUL(TT, VFMAI(TF, TC)); Chris@42: TG = VZMUL(Tz, VFNMSI(TF, TC)); Chris@42: T1c = VZMULI(T1b, VFNMSI(TL, TK)); Chris@42: TM = VZMULI(TJ, VFMAI(TL, TK)); Chris@42: T1k = VZMULI(T1j, VFMAI(Tx, Tl)); Chris@42: Ty = VZMULI(T1, VFNMSI(Tx, Tl)); Chris@42: T19 = VCONJ(VSUB(T16, T18)); Chris@42: T1a = VADD(T16, T18); Chris@42: T13 = VCONJ(VSUB(TU, T12)); Chris@42: T14 = VADD(TU, T12); Chris@42: T1h = VADD(T1c, T1g); Chris@42: T1i = VCONJ(VSUB(T1g, T1c)); Chris@42: TS = VCONJ(VSUB(TQ, TM)); Chris@42: TR = VADD(TM, TQ); Chris@42: T1m = VADD(T1k, T1l); Chris@42: T1n = VCONJ(VSUB(T1l, T1k)); Chris@42: TI = VCONJ(VSUB(TG, Ty)); Chris@42: TH = VADD(Ty, TG); Chris@42: ST(&(Rm[WS(rs, 4)]), T19, -ms, &(Rm[0])); Chris@42: ST(&(Rp[WS(rs, 4)]), T1a, ms, &(Rp[0])); Chris@42: ST(&(Rm[WS(rs, 1)]), T13, -ms, &(Rm[WS(rs, 1)])); Chris@42: ST(&(Rp[WS(rs, 1)]), T14, ms, &(Rp[WS(rs, 1)])); Chris@42: ST(&(Rp[WS(rs, 3)]), T1h, ms, &(Rp[WS(rs, 1)])); Chris@42: ST(&(Rm[WS(rs, 3)]), T1i, -ms, &(Rm[WS(rs, 1)])); Chris@42: ST(&(Rm[WS(rs, 2)]), TS, -ms, &(Rm[0])); Chris@42: ST(&(Rp[WS(rs, 2)]), TR, ms, &(Rp[0])); Chris@42: ST(&(Rp[0]), T1m, ms, &(Rp[0])); Chris@42: ST(&(Rm[0]), T1n, -ms, &(Rm[0])); Chris@42: ST(&(Rm[WS(rs, 5)]), TI, -ms, &(Rm[WS(rs, 1)])); Chris@42: ST(&(Rp[WS(rs, 5)]), TH, ms, &(Rp[WS(rs, 1)])); Chris@42: } Chris@42: } Chris@42: VLEAVE(); Chris@42: } Chris@42: Chris@42: static const tw_instr twinstr[] = { Chris@42: VTW(1, 1), Chris@42: VTW(1, 2), Chris@42: VTW(1, 3), Chris@42: VTW(1, 4), Chris@42: VTW(1, 5), Chris@42: VTW(1, 6), Chris@42: VTW(1, 7), Chris@42: VTW(1, 8), Chris@42: VTW(1, 9), Chris@42: VTW(1, 10), Chris@42: VTW(1, 11), Chris@42: {TW_NEXT, VL, 0} Chris@42: }; Chris@42: Chris@42: static const hc2c_desc desc = { 12, XSIMD_STRING("hc2cbdftv_12"), twinstr, &GENUS, {45, 25, 26, 0} }; Chris@42: Chris@42: void XSIMD(codelet_hc2cbdftv_12) (planner *p) { Chris@42: X(khc2c_register) (p, hc2cbdftv_12, &desc, HC2C_VIA_DFT); Chris@42: } Chris@42: #else /* HAVE_FMA */ Chris@42: Chris@42: /* Generated by: ../../../genfft/gen_hc2cdft_c.native -simd -compact -variables 4 -pipeline-latency 8 -trivial-stores -variables 32 -no-generate-bytw -n 12 -dif -sign 1 -name hc2cbdftv_12 -include hc2cbv.h */ Chris@42: Chris@42: /* Chris@42: * This function contains 71 FP additions, 30 FP multiplications, Chris@42: * (or, 67 additions, 26 multiplications, 4 fused multiply/add), Chris@42: * 90 stack variables, 2 constants, and 24 memory accesses Chris@42: */ Chris@42: #include "hc2cbv.h" Chris@42: Chris@42: static void hc2cbdftv_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@42: { Chris@42: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@42: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@42: { Chris@42: INT m; Chris@42: 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@42: V TY, TZ, Tf, TC, Tq, TG, Tm, TF, Ty, TD, T13, T1h, T2, T9, T3; Chris@42: V T5, T6, Tc, Tb, Td, T8, T4, Ta, T7, Te, To, Tp, Tr, Tv, Ti; Chris@42: V Ts, Tl, Tw, Tu, Tg, Th, Tj, Tk, Tt, Tx, T11, T12; Chris@42: T2 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@42: T8 = LD(&(Rm[WS(rs, 5)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: T9 = VCONJ(T8); Chris@42: T3 = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0])); Chris@42: T4 = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: T5 = VCONJ(T4); Chris@42: T6 = VADD(T3, T5); Chris@42: Tc = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@42: Ta = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: Tb = VCONJ(Ta); Chris@42: Td = VADD(Tb, Tc); Chris@42: TY = VADD(T2, T6); Chris@42: TZ = VADD(T9, Td); Chris@42: T7 = VFNMS(LDK(KP500000000), T6, T2); Chris@42: Te = VFNMS(LDK(KP500000000), Td, T9); Chris@42: Tf = VSUB(T7, Te); Chris@42: TC = VADD(T7, Te); Chris@42: To = VSUB(T3, T5); Chris@42: Tp = VSUB(Tb, Tc); Chris@42: Tq = VMUL(LDK(KP866025403), VSUB(To, Tp)); Chris@42: TG = VADD(To, Tp); Chris@42: Tr = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); Chris@42: Tu = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@42: Tv = VCONJ(Tu); Chris@42: Tg = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0])); Chris@42: Th = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@42: Ti = VCONJ(VSUB(Tg, Th)); Chris@42: Ts = VCONJ(VADD(Tg, Th)); Chris@42: Tj = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@42: Tk = LD(&(Rp[WS(rs, 5)]), ms, &(Rp[WS(rs, 1)])); Chris@42: Tl = VSUB(Tj, Tk); Chris@42: Tw = VADD(Tj, Tk); Chris@42: Tm = VMUL(LDK(KP866025403), VSUB(Ti, Tl)); Chris@42: TF = VADD(Ti, Tl); Chris@42: Tt = VFNMS(LDK(KP500000000), Ts, Tr); Chris@42: Tx = VFNMS(LDK(KP500000000), Tw, Tv); Chris@42: Ty = VSUB(Tt, Tx); Chris@42: TD = VADD(Tt, Tx); Chris@42: T11 = VADD(Tr, Ts); Chris@42: T12 = VADD(Tv, Tw); Chris@42: T13 = VBYI(VSUB(T11, T12)); Chris@42: T1h = VADD(T11, T12); Chris@42: { Chris@42: V T1n, T1i, T14, T1a, TA, T1m, TS, T18, TO, T1e, TI, TW, T1g, T1f, T10; Chris@42: V TX, T19, Tn, Tz, T1, T1l, TQ, TR, TP, T17, TM, TN, TL, T1d, TE; Chris@42: V TH, TB, TV, TJ, T1p, T1k, TT, T1o, TK, TU, T1j, T1b, T16, T1c, T15; Chris@42: T1g = VADD(TY, TZ); Chris@42: T1n = VADD(T1g, T1h); Chris@42: T1f = LDW(&(W[TWVL * 10])); Chris@42: T1i = VZMUL(T1f, VSUB(T1g, T1h)); Chris@42: T10 = VSUB(TY, TZ); Chris@42: TX = LDW(&(W[TWVL * 4])); Chris@42: T14 = VZMULI(TX, VSUB(T10, T13)); Chris@42: T19 = LDW(&(W[TWVL * 16])); Chris@42: T1a = VZMULI(T19, VADD(T10, T13)); Chris@42: Tn = VSUB(Tf, Tm); Chris@42: Tz = VBYI(VADD(Tq, Ty)); Chris@42: T1 = LDW(&(W[TWVL * 20])); Chris@42: TA = VZMULI(T1, VSUB(Tn, Tz)); Chris@42: T1l = LDW(&(W[0])); Chris@42: T1m = VZMULI(T1l, VADD(Tn, Tz)); Chris@42: TQ = VBYI(VMUL(LDK(KP866025403), VADD(TG, TF))); Chris@42: TR = VADD(TC, TD); Chris@42: TP = LDW(&(W[TWVL * 6])); Chris@42: TS = VZMUL(TP, VADD(TQ, TR)); Chris@42: T17 = LDW(&(W[TWVL * 14])); Chris@42: T18 = VZMUL(T17, VSUB(TR, TQ)); Chris@42: TM = VADD(Tf, Tm); Chris@42: TN = VBYI(VSUB(Ty, Tq)); Chris@42: TL = LDW(&(W[TWVL * 8])); Chris@42: TO = VZMULI(TL, VADD(TM, TN)); Chris@42: T1d = LDW(&(W[TWVL * 12])); Chris@42: T1e = VZMULI(T1d, VSUB(TM, TN)); Chris@42: TE = VSUB(TC, TD); Chris@42: TH = VBYI(VMUL(LDK(KP866025403), VSUB(TF, TG))); Chris@42: TB = LDW(&(W[TWVL * 18])); Chris@42: TI = VZMUL(TB, VSUB(TE, TH)); Chris@42: TV = LDW(&(W[TWVL * 2])); Chris@42: TW = VZMUL(TV, VADD(TH, TE)); Chris@42: TJ = VADD(TA, TI); Chris@42: ST(&(Rp[WS(rs, 5)]), TJ, ms, &(Rp[WS(rs, 1)])); Chris@42: T1p = VCONJ(VSUB(T1n, T1m)); Chris@42: ST(&(Rm[0]), T1p, -ms, &(Rm[0])); Chris@42: T1k = VCONJ(VSUB(T1i, T1e)); Chris@42: ST(&(Rm[WS(rs, 3)]), T1k, -ms, &(Rm[WS(rs, 1)])); Chris@42: TT = VADD(TO, TS); Chris@42: ST(&(Rp[WS(rs, 2)]), TT, ms, &(Rp[0])); Chris@42: T1o = VADD(T1m, T1n); Chris@42: ST(&(Rp[0]), T1o, ms, &(Rp[0])); Chris@42: TK = VCONJ(VSUB(TI, TA)); Chris@42: ST(&(Rm[WS(rs, 5)]), TK, -ms, &(Rm[WS(rs, 1)])); Chris@42: TU = VCONJ(VSUB(TS, TO)); Chris@42: ST(&(Rm[WS(rs, 2)]), TU, -ms, &(Rm[0])); Chris@42: T1j = VADD(T1e, T1i); Chris@42: ST(&(Rp[WS(rs, 3)]), T1j, ms, &(Rp[WS(rs, 1)])); Chris@42: T1b = VCONJ(VSUB(T18, T1a)); Chris@42: ST(&(Rm[WS(rs, 4)]), T1b, -ms, &(Rm[0])); Chris@42: T16 = VADD(TW, T14); Chris@42: ST(&(Rp[WS(rs, 1)]), T16, ms, &(Rp[WS(rs, 1)])); Chris@42: T1c = VADD(T18, T1a); Chris@42: ST(&(Rp[WS(rs, 4)]), T1c, ms, &(Rp[0])); Chris@42: T15 = VCONJ(VSUB(TW, T14)); Chris@42: ST(&(Rm[WS(rs, 1)]), T15, -ms, &(Rm[WS(rs, 1)])); Chris@42: } Chris@42: } Chris@42: } Chris@42: VLEAVE(); Chris@42: } Chris@42: Chris@42: static const tw_instr twinstr[] = { Chris@42: VTW(1, 1), Chris@42: VTW(1, 2), Chris@42: VTW(1, 3), Chris@42: VTW(1, 4), Chris@42: VTW(1, 5), Chris@42: VTW(1, 6), Chris@42: VTW(1, 7), Chris@42: VTW(1, 8), Chris@42: VTW(1, 9), Chris@42: VTW(1, 10), Chris@42: VTW(1, 11), Chris@42: {TW_NEXT, VL, 0} Chris@42: }; Chris@42: Chris@42: static const hc2c_desc desc = { 12, XSIMD_STRING("hc2cbdftv_12"), twinstr, &GENUS, {67, 26, 4, 0} }; Chris@42: Chris@42: void XSIMD(codelet_hc2cbdftv_12) (planner *p) { Chris@42: X(khc2c_register) (p, hc2cbdftv_12, &desc, HC2C_VIA_DFT); Chris@42: } Chris@42: #endif /* HAVE_FMA */