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 16:52:42 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 8 -dif -sign 1 -name hc2cbdftv_8 -include hc2cbv.h */ Chris@42: Chris@42: /* Chris@42: * This function contains 41 FP additions, 32 FP multiplications, Chris@42: * (or, 23 additions, 14 multiplications, 18 fused multiply/add), Chris@42: * 51 stack variables, 1 constants, and 16 memory accesses Chris@42: */ Chris@42: #include "hc2cbv.h" Chris@42: Chris@42: static void hc2cbdftv_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@42: { Chris@42: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@42: { Chris@42: INT m; Chris@42: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 14)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 14), MAKE_VOLATILE_STRIDE(32, rs)) { Chris@42: V TJ, T4, Tf, TB, TD, TE, Tm, T1, Tj, TF, Tp, Tb, Tg, Tt, Tx; Chris@42: V T2, T3, Td, Te, T5, T6, T8, T9, Tn, T7, To, Ta, Tk, Tl, TG; Chris@42: V TL, Tq, Tc, Tu, Th, Tv, Ty, Tw, TC, Ti, TK, TA, Tz, TI, TH; Chris@42: V Ts, Tr, TN, TM; Chris@42: T2 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@42: T3 = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: Td = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@42: Te = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: T5 = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); 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: T9 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@42: TJ = LDW(&(W[0])); Chris@42: Tk = VFMACONJ(T3, T2); Chris@42: T4 = VFNMSCONJ(T3, T2); Chris@42: Tl = VFMACONJ(Te, Td); Chris@42: Tf = VFNMSCONJ(Te, Td); Chris@42: Tn = VFMACONJ(T6, T5); Chris@42: T7 = VFNMSCONJ(T6, T5); Chris@42: To = VFMACONJ(T9, T8); Chris@42: Ta = VFMSCONJ(T9, T8); Chris@42: TB = LDW(&(W[TWVL * 8])); Chris@42: TD = LDW(&(W[TWVL * 6])); Chris@42: TE = VADD(Tk, Tl); Chris@42: Tm = VSUB(Tk, Tl); Chris@42: T1 = LDW(&(W[TWVL * 12])); Chris@42: Tj = LDW(&(W[TWVL * 10])); Chris@42: TF = VADD(Tn, To); Chris@42: Tp = VSUB(Tn, To); Chris@42: Tb = VADD(T7, Ta); Chris@42: Tg = VSUB(T7, Ta); Chris@42: Tt = LDW(&(W[TWVL * 4])); Chris@42: Tx = LDW(&(W[TWVL * 2])); Chris@42: TG = VZMUL(TD, VSUB(TE, TF)); Chris@42: TL = VADD(TE, TF); Chris@42: Tq = VZMUL(Tj, VFNMSI(Tp, Tm)); Chris@42: Tc = VFMA(LDK(KP707106781), Tb, T4); Chris@42: Tu = VFNMS(LDK(KP707106781), Tb, T4); Chris@42: Th = VFMA(LDK(KP707106781), Tg, Tf); Chris@42: Tv = VFNMS(LDK(KP707106781), Tg, Tf); Chris@42: Ty = VZMUL(Tx, VFMAI(Tp, Tm)); Chris@42: Tw = VZMULI(Tt, VFNMSI(Tv, Tu)); Chris@42: TC = VZMULI(TB, VFMAI(Tv, Tu)); Chris@42: Ti = VZMULI(T1, VFNMSI(Th, Tc)); Chris@42: TK = VZMULI(TJ, VFMAI(Th, Tc)); Chris@42: TA = VCONJ(VSUB(Ty, Tw)); Chris@42: Tz = VADD(Tw, Ty); Chris@42: TI = VCONJ(VSUB(TG, TC)); Chris@42: TH = VADD(TC, TG); Chris@42: Ts = VCONJ(VSUB(Tq, Ti)); Chris@42: Tr = VADD(Ti, Tq); Chris@42: TN = VCONJ(VSUB(TL, TK)); Chris@42: TM = VADD(TK, TL); Chris@42: ST(&(Rm[WS(rs, 1)]), TA, -ms, &(Rm[WS(rs, 1)])); Chris@42: ST(&(Rp[WS(rs, 1)]), Tz, ms, &(Rp[WS(rs, 1)])); Chris@42: ST(&(Rm[WS(rs, 2)]), TI, -ms, &(Rm[0])); Chris@42: ST(&(Rp[WS(rs, 2)]), TH, ms, &(Rp[0])); Chris@42: ST(&(Rm[WS(rs, 3)]), Ts, -ms, &(Rm[WS(rs, 1)])); Chris@42: ST(&(Rp[WS(rs, 3)]), Tr, ms, &(Rp[WS(rs, 1)])); Chris@42: ST(&(Rm[0]), TN, -ms, &(Rm[0])); Chris@42: ST(&(Rp[0]), TM, ms, &(Rp[0])); 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: {TW_NEXT, VL, 0} Chris@42: }; Chris@42: Chris@42: static const hc2c_desc desc = { 8, XSIMD_STRING("hc2cbdftv_8"), twinstr, &GENUS, {23, 14, 18, 0} }; Chris@42: Chris@42: void XSIMD(codelet_hc2cbdftv_8) (planner *p) { Chris@42: X(khc2c_register) (p, hc2cbdftv_8, &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 8 -dif -sign 1 -name hc2cbdftv_8 -include hc2cbv.h */ Chris@42: Chris@42: /* Chris@42: * This function contains 41 FP additions, 16 FP multiplications, Chris@42: * (or, 41 additions, 16 multiplications, 0 fused multiply/add), Chris@42: * 55 stack variables, 1 constants, and 16 memory accesses Chris@42: */ Chris@42: #include "hc2cbv.h" Chris@42: Chris@42: static void hc2cbdftv_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@42: { Chris@42: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@42: { Chris@42: INT m; Chris@42: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 14)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 14), MAKE_VOLATILE_STRIDE(32, rs)) { Chris@42: V T5, Tj, Tq, TI, Te, Tk, Tt, TJ, T2, Tg, T4, Ti, T3, Th, To; Chris@42: V Tp, T6, Tc, T8, Tb, T7, Ta, T9, Td, Tr, Ts, TP, Tu, Tm, TO; Chris@42: V Tn, Tf, Tl, T1, TN, Tv, TR, Tw, TQ, TC, TK, TA, TG, TB, TH; Chris@42: V Ty, Tz, Tx, TF, TD, TM, TE, TL; Chris@42: T2 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@42: Tg = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@42: T3 = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: T4 = VCONJ(T3); Chris@42: Th = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: Ti = VCONJ(Th); Chris@42: T5 = VSUB(T2, T4); Chris@42: Tj = VSUB(Tg, Ti); Chris@42: To = VADD(T2, T4); Chris@42: Tp = VADD(Tg, Ti); Chris@42: Tq = VSUB(To, Tp); Chris@42: TI = VADD(To, Tp); Chris@42: T6 = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@42: Tc = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); Chris@42: T7 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@42: T8 = VCONJ(T7); Chris@42: Ta = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@42: Tb = VCONJ(Ta); Chris@42: T9 = VSUB(T6, T8); Chris@42: Td = VSUB(Tb, Tc); Chris@42: Te = VMUL(LDK(KP707106781), VADD(T9, Td)); Chris@42: Tk = VMUL(LDK(KP707106781), VSUB(T9, Td)); Chris@42: Tr = VADD(T6, T8); Chris@42: Ts = VADD(Tb, Tc); Chris@42: Tt = VBYI(VSUB(Tr, Ts)); Chris@42: TJ = VADD(Tr, Ts); Chris@42: TP = VADD(TI, TJ); Chris@42: Tn = LDW(&(W[TWVL * 10])); Chris@42: Tu = VZMUL(Tn, VSUB(Tq, Tt)); Chris@42: Tf = VADD(T5, Te); Chris@42: Tl = VBYI(VADD(Tj, Tk)); Chris@42: T1 = LDW(&(W[TWVL * 12])); Chris@42: Tm = VZMULI(T1, VSUB(Tf, Tl)); Chris@42: TN = LDW(&(W[0])); Chris@42: TO = VZMULI(TN, VADD(Tl, Tf)); Chris@42: Tv = VADD(Tm, Tu); Chris@42: ST(&(Rp[WS(rs, 3)]), Tv, ms, &(Rp[WS(rs, 1)])); Chris@42: TR = VCONJ(VSUB(TP, TO)); Chris@42: ST(&(Rm[0]), TR, -ms, &(Rm[0])); Chris@42: Tw = VCONJ(VSUB(Tu, Tm)); Chris@42: ST(&(Rm[WS(rs, 3)]), Tw, -ms, &(Rm[WS(rs, 1)])); Chris@42: TQ = VADD(TO, TP); Chris@42: ST(&(Rp[0]), TQ, ms, &(Rp[0])); Chris@42: TB = LDW(&(W[TWVL * 2])); Chris@42: TC = VZMUL(TB, VADD(Tq, Tt)); Chris@42: TH = LDW(&(W[TWVL * 6])); Chris@42: TK = VZMUL(TH, VSUB(TI, TJ)); Chris@42: Ty = VBYI(VSUB(Tk, Tj)); Chris@42: Tz = VSUB(T5, Te); Chris@42: Tx = LDW(&(W[TWVL * 4])); Chris@42: TA = VZMULI(Tx, VADD(Ty, Tz)); Chris@42: TF = LDW(&(W[TWVL * 8])); Chris@42: TG = VZMULI(TF, VSUB(Tz, Ty)); Chris@42: TD = VADD(TA, TC); Chris@42: ST(&(Rp[WS(rs, 1)]), TD, ms, &(Rp[WS(rs, 1)])); Chris@42: TM = VCONJ(VSUB(TK, TG)); Chris@42: ST(&(Rm[WS(rs, 2)]), TM, -ms, &(Rm[0])); Chris@42: TE = VCONJ(VSUB(TC, TA)); Chris@42: ST(&(Rm[WS(rs, 1)]), TE, -ms, &(Rm[WS(rs, 1)])); Chris@42: TL = VADD(TG, TK); Chris@42: ST(&(Rp[WS(rs, 2)]), TL, ms, &(Rp[0])); 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: {TW_NEXT, VL, 0} Chris@42: }; Chris@42: Chris@42: static const hc2c_desc desc = { 8, XSIMD_STRING("hc2cbdftv_8"), twinstr, &GENUS, {41, 16, 0, 0} }; Chris@42: Chris@42: void XSIMD(codelet_hc2cbdftv_8) (planner *p) { Chris@42: X(khc2c_register) (p, hc2cbdftv_8, &desc, HC2C_VIA_DFT); Chris@42: } Chris@42: #endif /* HAVE_FMA */