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