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 6 -dif -sign 1 -name hc2cbdftv_6 -include hc2cbv.h */ Chris@42: Chris@42: /* Chris@42: * This function contains 29 FP additions, 24 FP multiplications, Chris@42: * (or, 17 additions, 12 multiplications, 12 fused multiply/add), Chris@42: * 38 stack variables, 2 constants, and 12 memory accesses Chris@42: */ Chris@42: #include "hc2cbv.h" Chris@42: Chris@42: static void hc2cbdftv_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@42: { Chris@42: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@42: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@42: { Chris@42: INT m; Chris@42: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 10)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 10), MAKE_VOLATILE_STRIDE(24, rs)) { Chris@42: V Tv, Tn, Tr, Te, T4, Tg, Ta, Tf, T7, T1, Td, T2, T3, T8, T9; Chris@42: V T5, T6, Th, Tj, Tb, Tp, Tx, Ti, Tc, To, Tk, Ts, Tq, Tw, Tm; Chris@42: V Tl, Tu, Tt, Tz, Ty; Chris@42: T2 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@42: T3 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@42: T8 = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@42: T9 = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: T5 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@42: T6 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@42: Tv = LDW(&(W[0])); Chris@42: Tn = LDW(&(W[TWVL * 8])); Chris@42: Tr = LDW(&(W[TWVL * 6])); Chris@42: Te = VFMACONJ(T3, T2); Chris@42: T4 = VFNMSCONJ(T3, T2); Chris@42: Tg = VFMACONJ(T9, T8); Chris@42: Ta = VFMSCONJ(T9, T8); Chris@42: Tf = VFMACONJ(T6, T5); Chris@42: T7 = VFNMSCONJ(T6, T5); Chris@42: T1 = LDW(&(W[TWVL * 4])); Chris@42: Td = LDW(&(W[TWVL * 2])); Chris@42: Th = VADD(Tf, Tg); Chris@42: Tj = VMUL(LDK(KP866025403), VSUB(Tf, Tg)); Chris@42: Tb = VADD(T7, Ta); Chris@42: Tp = VMUL(LDK(KP866025403), VSUB(T7, Ta)); Chris@42: Tx = VADD(Te, Th); Chris@42: Ti = VFNMS(LDK(KP500000000), Th, Te); Chris@42: Tc = VZMULI(T1, VADD(T4, Tb)); Chris@42: To = VFNMS(LDK(KP500000000), Tb, T4); Chris@42: Tk = VZMUL(Td, VFNMSI(Tj, Ti)); Chris@42: Ts = VZMUL(Tr, VFMAI(Tj, Ti)); Chris@42: Tq = VZMULI(Tn, VFNMSI(Tp, To)); Chris@42: Tw = VZMULI(Tv, VFMAI(Tp, To)); Chris@42: Tm = VCONJ(VSUB(Tk, Tc)); Chris@42: Tl = VADD(Tc, Tk); Chris@42: Tu = VCONJ(VSUB(Ts, Tq)); Chris@42: Tt = VADD(Tq, Ts); Chris@42: Tz = VCONJ(VSUB(Tx, Tw)); Chris@42: Ty = VADD(Tw, Tx); Chris@42: ST(&(Rm[WS(rs, 1)]), Tm, -ms, &(Rm[WS(rs, 1)])); Chris@42: ST(&(Rp[WS(rs, 1)]), Tl, ms, &(Rp[WS(rs, 1)])); Chris@42: ST(&(Rm[WS(rs, 2)]), Tu, -ms, &(Rm[0])); Chris@42: ST(&(Rp[WS(rs, 2)]), Tt, ms, &(Rp[0])); Chris@42: ST(&(Rm[0]), Tz, -ms, &(Rm[0])); Chris@42: ST(&(Rp[0]), Ty, 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: {TW_NEXT, VL, 0} Chris@42: }; Chris@42: Chris@42: static const hc2c_desc desc = { 6, XSIMD_STRING("hc2cbdftv_6"), twinstr, &GENUS, {17, 12, 12, 0} }; Chris@42: Chris@42: void XSIMD(codelet_hc2cbdftv_6) (planner *p) { Chris@42: X(khc2c_register) (p, hc2cbdftv_6, &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 6 -dif -sign 1 -name hc2cbdftv_6 -include hc2cbv.h */ Chris@42: Chris@42: /* Chris@42: * This function contains 29 FP additions, 14 FP multiplications, Chris@42: * (or, 27 additions, 12 multiplications, 2 fused multiply/add), Chris@42: * 41 stack variables, 2 constants, and 12 memory accesses Chris@42: */ Chris@42: #include "hc2cbv.h" Chris@42: Chris@42: static void hc2cbdftv_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@42: { Chris@42: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@42: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@42: { Chris@42: INT m; Chris@42: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 10)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 10), MAKE_VOLATILE_STRIDE(24, rs)) { Chris@42: V T5, Th, Te, Ts, Tk, Tm, T2, T4, T3, T6, Tc, T8, Tb, T7, Ta; Chris@42: V T9, Td, Ti, Tj, TA, Tf, Tn, Tv, Tt, Tz, T1, Tl, Tg, Tu, Tr; Chris@42: V Tq, Ty, To, Tp, TC, TB, Tx, Tw; Chris@42: T2 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@42: T3 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@42: T4 = VCONJ(T3); Chris@42: T5 = VSUB(T2, T4); Chris@42: Th = VADD(T2, T4); Chris@42: T6 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@42: Tc = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@42: T7 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@42: T8 = VCONJ(T7); Chris@42: Ta = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@42: Tb = VCONJ(Ta); Chris@42: T9 = VSUB(T6, T8); Chris@42: Td = VSUB(Tb, Tc); Chris@42: Te = VADD(T9, Td); Chris@42: Ts = VBYI(VMUL(LDK(KP866025403), VSUB(T9, Td))); Chris@42: Ti = VADD(T6, T8); Chris@42: Tj = VADD(Tb, Tc); Chris@42: Tk = VADD(Ti, Tj); Chris@42: Tm = VBYI(VMUL(LDK(KP866025403), VSUB(Ti, Tj))); Chris@42: TA = VADD(Th, Tk); Chris@42: T1 = LDW(&(W[TWVL * 4])); Chris@42: Tf = VZMULI(T1, VADD(T5, Te)); Chris@42: Tl = VFNMS(LDK(KP500000000), Tk, Th); Chris@42: Tg = LDW(&(W[TWVL * 2])); Chris@42: Tn = VZMUL(Tg, VSUB(Tl, Tm)); Chris@42: Tu = LDW(&(W[TWVL * 6])); Chris@42: Tv = VZMUL(Tu, VADD(Tm, Tl)); Chris@42: Tr = VFNMS(LDK(KP500000000), Te, T5); Chris@42: Tq = LDW(&(W[TWVL * 8])); Chris@42: Tt = VZMULI(Tq, VSUB(Tr, Ts)); Chris@42: Ty = LDW(&(W[0])); Chris@42: Tz = VZMULI(Ty, VADD(Ts, Tr)); Chris@42: To = VADD(Tf, Tn); Chris@42: ST(&(Rp[WS(rs, 1)]), To, ms, &(Rp[WS(rs, 1)])); Chris@42: Tp = VCONJ(VSUB(Tn, Tf)); Chris@42: ST(&(Rm[WS(rs, 1)]), Tp, -ms, &(Rm[WS(rs, 1)])); Chris@42: TC = VCONJ(VSUB(TA, Tz)); Chris@42: ST(&(Rm[0]), TC, -ms, &(Rm[0])); Chris@42: TB = VADD(Tz, TA); Chris@42: ST(&(Rp[0]), TB, ms, &(Rp[0])); Chris@42: Tx = VCONJ(VSUB(Tv, Tt)); Chris@42: ST(&(Rm[WS(rs, 2)]), Tx, -ms, &(Rm[0])); Chris@42: Tw = VADD(Tt, Tv); Chris@42: ST(&(Rp[WS(rs, 2)]), Tw, 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: {TW_NEXT, VL, 0} Chris@42: }; Chris@42: Chris@42: static const hc2c_desc desc = { 6, XSIMD_STRING("hc2cbdftv_6"), twinstr, &GENUS, {27, 12, 2, 0} }; Chris@42: Chris@42: void XSIMD(codelet_hc2cbdftv_6) (planner *p) { Chris@42: X(khc2c_register) (p, hc2cbdftv_6, &desc, HC2C_VIA_DFT); Chris@42: } Chris@42: #endif /* HAVE_FMA */