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 16:52:43 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 10 -dif -sign 1 -name hc2cbdftv_10 -include hc2cbv.h */ Chris@42: Chris@42: /* Chris@42: * This function contains 61 FP additions, 50 FP multiplications, Chris@42: * (or, 33 additions, 22 multiplications, 28 fused multiply/add), Chris@42: * 76 stack variables, 4 constants, and 20 memory accesses Chris@42: */ Chris@42: #include "hc2cbv.h" Chris@42: Chris@42: static void hc2cbdftv_10(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@42: { Chris@42: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@42: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@42: DVK(KP618033988, +0.618033988749894848204586834365638117720309180); Chris@42: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@42: { Chris@42: INT m; Chris@42: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 18)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 18), MAKE_VOLATILE_STRIDE(40, rs)) { Chris@42: V Ts, T4, TR, T1, TZ, TD, Ty, Tn, Ti, TT, T11, TJ, T15, Tr, TN; Chris@42: V TE, Tv, To, Tb, T8, Tw, Te, Tx, Th, Tt, T7, T9, T2, T3, Tc; Chris@42: V Td, Tf, Tg, T5, T6, Tu, Ta; Chris@42: T2 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@42: T3 = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0])); Chris@42: Tc = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0])); Chris@42: Td = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@42: Tf = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@42: Tg = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: T5 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@42: T6 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@42: T8 = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); Chris@42: Ts = VFMACONJ(T3, T2); Chris@42: T4 = VFNMSCONJ(T3, T2); Chris@42: Tw = VFMACONJ(Td, Tc); Chris@42: Te = VFNMSCONJ(Td, Tc); Chris@42: Tx = VFMACONJ(Tg, Tf); Chris@42: Th = VFMSCONJ(Tg, Tf); Chris@42: Tt = VFMACONJ(T6, T5); Chris@42: T7 = VFNMSCONJ(T6, T5); Chris@42: T9 = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: TR = LDW(&(W[TWVL * 8])); Chris@42: T1 = LDW(&(W[TWVL * 4])); Chris@42: TZ = LDW(&(W[TWVL * 12])); Chris@42: TD = VSUB(Tw, Tx); Chris@42: Ty = VADD(Tw, Tx); Chris@42: Tn = VSUB(Te, Th); Chris@42: Ti = VADD(Te, Th); Chris@42: Tu = VFMACONJ(T9, T8); Chris@42: Ta = VFMSCONJ(T9, T8); Chris@42: TT = LDW(&(W[TWVL * 6])); Chris@42: T11 = LDW(&(W[TWVL * 10])); Chris@42: TJ = LDW(&(W[TWVL * 16])); Chris@42: T15 = LDW(&(W[0])); Chris@42: Tr = LDW(&(W[TWVL * 2])); Chris@42: TN = LDW(&(W[TWVL * 14])); Chris@42: TE = VSUB(Tt, Tu); Chris@42: Tv = VADD(Tt, Tu); Chris@42: To = VSUB(T7, Ta); Chris@42: Tb = VADD(T7, Ta); Chris@42: { Chris@42: V TV, TF, Tz, TB, TL, Tp, Tj, Tl, T17, TA, TS, Tk, TC, TU, TK; Chris@42: V Tm, TO, TG, T12, TW, T16, TM, T10, Tq, TX, TY, T18, T19, TQ, TP; Chris@42: V T13, T14, TI, TH; Chris@42: TV = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), TD, TE)); Chris@42: TF = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), TE, TD)); Chris@42: Tz = VADD(Tv, Ty); Chris@42: TB = VSUB(Tv, Ty); Chris@42: TL = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), Tn, To)); Chris@42: Tp = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), To, Tn)); Chris@42: Tj = VADD(Tb, Ti); Chris@42: Tl = VSUB(Tb, Ti); Chris@42: T17 = VADD(Ts, Tz); Chris@42: TA = VFNMS(LDK(KP250000000), Tz, Ts); Chris@42: TS = VZMULI(TR, VADD(T4, Tj)); Chris@42: Tk = VFNMS(LDK(KP250000000), Tj, T4); Chris@42: TC = VFNMS(LDK(KP559016994), TB, TA); Chris@42: TU = VFMA(LDK(KP559016994), TB, TA); Chris@42: TK = VFMA(LDK(KP559016994), Tl, Tk); Chris@42: Tm = VFNMS(LDK(KP559016994), Tl, Tk); Chris@42: TO = VZMUL(TN, VFMAI(TF, TC)); Chris@42: TG = VZMUL(Tr, VFNMSI(TF, TC)); Chris@42: T12 = VZMUL(T11, VFMAI(TV, TU)); Chris@42: TW = VZMUL(TT, VFNMSI(TV, TU)); Chris@42: T16 = VZMULI(T15, VFMAI(TL, TK)); Chris@42: TM = VZMULI(TJ, VFNMSI(TL, TK)); Chris@42: T10 = VZMULI(TZ, VFNMSI(Tp, Tm)); Chris@42: Tq = VZMULI(T1, VFMAI(Tp, Tm)); Chris@42: TX = VADD(TS, TW); Chris@42: TY = VCONJ(VSUB(TW, TS)); Chris@42: T18 = VADD(T16, T17); Chris@42: T19 = VCONJ(VSUB(T17, T16)); Chris@42: TQ = VCONJ(VSUB(TO, TM)); Chris@42: TP = VADD(TM, TO); Chris@42: T13 = VADD(T10, T12); Chris@42: T14 = VCONJ(VSUB(T12, T10)); Chris@42: TI = VCONJ(VSUB(TG, Tq)); Chris@42: TH = VADD(Tq, TG); Chris@42: ST(&(Rp[WS(rs, 2)]), TX, ms, &(Rp[0])); Chris@42: ST(&(Rm[WS(rs, 2)]), TY, -ms, &(Rm[0])); Chris@42: ST(&(Rp[0]), T18, ms, &(Rp[0])); Chris@42: ST(&(Rm[0]), T19, -ms, &(Rm[0])); Chris@42: ST(&(Rm[WS(rs, 4)]), TQ, -ms, &(Rm[0])); Chris@42: ST(&(Rp[WS(rs, 4)]), TP, ms, &(Rp[0])); Chris@42: ST(&(Rp[WS(rs, 3)]), T13, ms, &(Rp[WS(rs, 1)])); Chris@42: ST(&(Rm[WS(rs, 3)]), T14, -ms, &(Rm[WS(rs, 1)])); Chris@42: ST(&(Rm[WS(rs, 1)]), TI, -ms, &(Rm[WS(rs, 1)])); Chris@42: ST(&(Rp[WS(rs, 1)]), TH, ms, &(Rp[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: {TW_NEXT, VL, 0} Chris@42: }; Chris@42: Chris@42: static const hc2c_desc desc = { 10, XSIMD_STRING("hc2cbdftv_10"), twinstr, &GENUS, {33, 22, 28, 0} }; Chris@42: Chris@42: void XSIMD(codelet_hc2cbdftv_10) (planner *p) { Chris@42: X(khc2c_register) (p, hc2cbdftv_10, &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 10 -dif -sign 1 -name hc2cbdftv_10 -include hc2cbv.h */ Chris@42: Chris@42: /* Chris@42: * This function contains 61 FP additions, 30 FP multiplications, Chris@42: * (or, 55 additions, 24 multiplications, 6 fused multiply/add), Chris@42: * 81 stack variables, 4 constants, and 20 memory accesses Chris@42: */ Chris@42: #include "hc2cbv.h" Chris@42: Chris@42: static void hc2cbdftv_10(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@42: { Chris@42: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@42: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@42: DVK(KP587785252, +0.587785252292473129168705954639072768597652438); Chris@42: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@42: { Chris@42: INT m; Chris@42: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 18)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 18), MAKE_VOLATILE_STRIDE(40, rs)) { Chris@42: V T5, TE, Ts, Tt, TC, Tz, TH, TJ, To, Tq, T2, T4, T3, T9, Tx; Chris@42: V Tm, TB, Td, Ty, Ti, TA, T6, T8, T7, Tl, Tk, Tj, Tc, Tb, Ta; Chris@42: V Tf, Th, Tg, TF, TG, Te, Tn; Chris@42: T2 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@42: T3 = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0])); Chris@42: T4 = VCONJ(T3); Chris@42: T5 = VSUB(T2, T4); Chris@42: TE = VADD(T2, T4); Chris@42: T6 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@42: T7 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@42: T8 = VCONJ(T7); Chris@42: T9 = VSUB(T6, T8); Chris@42: Tx = VADD(T6, T8); Chris@42: Tl = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@42: Tj = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: Tk = VCONJ(Tj); Chris@42: Tm = VSUB(Tk, Tl); Chris@42: TB = VADD(Tk, Tl); Chris@42: Tc = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); Chris@42: Ta = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: Tb = VCONJ(Ta); Chris@42: Td = VSUB(Tb, Tc); Chris@42: Ty = VADD(Tb, Tc); Chris@42: Tf = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0])); Chris@42: Tg = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@42: Th = VCONJ(Tg); Chris@42: Ti = VSUB(Tf, Th); Chris@42: TA = VADD(Tf, Th); Chris@42: Ts = VSUB(T9, Td); Chris@42: Tt = VSUB(Ti, Tm); Chris@42: TC = VSUB(TA, TB); Chris@42: Tz = VSUB(Tx, Ty); Chris@42: TF = VADD(Tx, Ty); Chris@42: TG = VADD(TA, TB); Chris@42: TH = VADD(TF, TG); Chris@42: TJ = VMUL(LDK(KP559016994), VSUB(TF, TG)); Chris@42: Te = VADD(T9, Td); Chris@42: Tn = VADD(Ti, Tm); Chris@42: To = VADD(Te, Tn); Chris@42: Tq = VMUL(LDK(KP559016994), VSUB(Te, Tn)); Chris@42: { Chris@42: V T1c, TX, Tv, T1b, TR, T15, TL, T17, TT, T11, TW, Tu, TQ, Tr, TP; Chris@42: V Tp, T1, T1a, TO, T14, TD, T10, TK, TZ, TI, Tw, T16, TS, TY, TM; Chris@42: V TU, T1e, TN, T1d, T19, T13, TV, T18, T12; Chris@42: T1c = VADD(TE, TH); Chris@42: TW = LDW(&(W[TWVL * 8])); Chris@42: TX = VZMULI(TW, VADD(T5, To)); Chris@42: Tu = VBYI(VFNMS(LDK(KP951056516), Tt, VMUL(LDK(KP587785252), Ts))); Chris@42: TQ = VBYI(VFMA(LDK(KP951056516), Ts, VMUL(LDK(KP587785252), Tt))); Chris@42: Tp = VFNMS(LDK(KP250000000), To, T5); Chris@42: Tr = VSUB(Tp, Tq); Chris@42: TP = VADD(Tq, Tp); Chris@42: T1 = LDW(&(W[TWVL * 4])); Chris@42: Tv = VZMULI(T1, VSUB(Tr, Tu)); Chris@42: T1a = LDW(&(W[0])); Chris@42: T1b = VZMULI(T1a, VADD(TQ, TP)); Chris@42: TO = LDW(&(W[TWVL * 16])); Chris@42: TR = VZMULI(TO, VSUB(TP, TQ)); Chris@42: T14 = LDW(&(W[TWVL * 12])); Chris@42: T15 = VZMULI(T14, VADD(Tu, Tr)); Chris@42: TD = VBYI(VFNMS(LDK(KP951056516), TC, VMUL(LDK(KP587785252), Tz))); Chris@42: T10 = VBYI(VFMA(LDK(KP951056516), Tz, VMUL(LDK(KP587785252), TC))); Chris@42: TI = VFNMS(LDK(KP250000000), TH, TE); Chris@42: TK = VSUB(TI, TJ); Chris@42: TZ = VADD(TJ, TI); Chris@42: Tw = LDW(&(W[TWVL * 2])); Chris@42: TL = VZMUL(Tw, VADD(TD, TK)); Chris@42: T16 = LDW(&(W[TWVL * 10])); Chris@42: T17 = VZMUL(T16, VADD(T10, TZ)); Chris@42: TS = LDW(&(W[TWVL * 14])); Chris@42: TT = VZMUL(TS, VSUB(TK, TD)); Chris@42: TY = LDW(&(W[TWVL * 6])); Chris@42: T11 = VZMUL(TY, VSUB(TZ, T10)); Chris@42: TM = VADD(Tv, TL); Chris@42: ST(&(Rp[WS(rs, 1)]), TM, ms, &(Rp[WS(rs, 1)])); Chris@42: TU = VADD(TR, TT); Chris@42: ST(&(Rp[WS(rs, 4)]), TU, ms, &(Rp[0])); Chris@42: T1e = VCONJ(VSUB(T1c, T1b)); Chris@42: ST(&(Rm[0]), T1e, -ms, &(Rm[0])); Chris@42: TN = VCONJ(VSUB(TL, Tv)); Chris@42: ST(&(Rm[WS(rs, 1)]), TN, -ms, &(Rm[WS(rs, 1)])); Chris@42: T1d = VADD(T1b, T1c); Chris@42: ST(&(Rp[0]), T1d, ms, &(Rp[0])); Chris@42: T19 = VCONJ(VSUB(T17, T15)); Chris@42: ST(&(Rm[WS(rs, 3)]), T19, -ms, &(Rm[WS(rs, 1)])); Chris@42: T13 = VCONJ(VSUB(T11, TX)); Chris@42: ST(&(Rm[WS(rs, 2)]), T13, -ms, &(Rm[0])); Chris@42: TV = VCONJ(VSUB(TT, TR)); Chris@42: ST(&(Rm[WS(rs, 4)]), TV, -ms, &(Rm[0])); Chris@42: T18 = VADD(T15, T17); Chris@42: ST(&(Rp[WS(rs, 3)]), T18, ms, &(Rp[WS(rs, 1)])); Chris@42: T12 = VADD(TX, T11); Chris@42: ST(&(Rp[WS(rs, 2)]), T12, ms, &(Rp[0])); 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: {TW_NEXT, VL, 0} Chris@42: }; Chris@42: Chris@42: static const hc2c_desc desc = { 10, XSIMD_STRING("hc2cbdftv_10"), twinstr, &GENUS, {55, 24, 6, 0} }; Chris@42: Chris@42: void XSIMD(codelet_hc2cbdftv_10) (planner *p) { Chris@42: X(khc2c_register) (p, hc2cbdftv_10, &desc, HC2C_VIA_DFT); Chris@42: } Chris@42: #endif /* HAVE_FMA */