Chris@82: /* Chris@82: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@82: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@82: * Chris@82: * This program is free software; you can redistribute it and/or modify Chris@82: * it under the terms of the GNU General Public License as published by Chris@82: * the Free Software Foundation; either version 2 of the License, or Chris@82: * (at your option) any later version. Chris@82: * Chris@82: * This program is distributed in the hope that it will be useful, Chris@82: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@82: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@82: * GNU General Public License for more details. Chris@82: * Chris@82: * You should have received a copy of the GNU General Public License Chris@82: * along with this program; if not, write to the Free Software Chris@82: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@82: * Chris@82: */ Chris@82: Chris@82: /* This file was automatically generated --- DO NOT EDIT */ Chris@82: /* Generated on Thu May 24 08:05:07 EDT 2018 */ Chris@82: Chris@82: #include "dft/codelet-dft.h" Chris@82: Chris@82: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_notw_c.native -fma -simd -compact -variables 4 -pipeline-latency 8 -n 8 -name n2fv_8 -with-ostride 2 -include dft/simd/n2f.h -store-multiple 2 */ Chris@82: Chris@82: /* Chris@82: * This function contains 26 FP additions, 10 FP multiplications, Chris@82: * (or, 16 additions, 0 multiplications, 10 fused multiply/add), Chris@82: * 24 stack variables, 1 constants, and 20 memory accesses Chris@82: */ Chris@82: #include "dft/simd/n2f.h" Chris@82: Chris@82: static void n2fv_8(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@82: { Chris@82: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@82: { Chris@82: INT i; Chris@82: const R *xi; Chris@82: R *xo; Chris@82: xi = ri; Chris@82: xo = ro; Chris@82: for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(16, is), MAKE_VOLATILE_STRIDE(16, os)) { Chris@82: V T3, Tj, Te, Tk, Ta, Tn, Tf, Tm, Tr, Tu; Chris@82: { Chris@82: V T1, T2, Tc, Td; Chris@82: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@82: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@82: T3 = VSUB(T1, T2); Chris@82: Tj = VADD(T1, T2); Chris@82: Tc = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@82: Td = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@82: Te = VSUB(Tc, Td); Chris@82: Tk = VADD(Tc, Td); Chris@82: { Chris@82: V T4, T5, T6, T7, T8, T9; Chris@82: T4 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@82: T5 = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@82: T6 = VSUB(T4, T5); Chris@82: T7 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@82: T8 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@82: T9 = VSUB(T7, T8); Chris@82: Ta = VADD(T6, T9); Chris@82: Tn = VADD(T7, T8); Chris@82: Tf = VSUB(T9, T6); Chris@82: Tm = VADD(T4, T5); Chris@82: } Chris@82: } Chris@82: { Chris@82: V Ts, Tb, Tg, Tp, Tq, Tt; Chris@82: Tb = VFMA(LDK(KP707106781), Ta, T3); Chris@82: Tg = VFNMS(LDK(KP707106781), Tf, Te); Chris@82: Tr = VFNMSI(Tg, Tb); Chris@82: STM2(&(xo[2]), Tr, ovs, &(xo[2])); Chris@82: Ts = VFMAI(Tg, Tb); Chris@82: STM2(&(xo[14]), Ts, ovs, &(xo[2])); Chris@82: Tp = VSUB(Tj, Tk); Chris@82: Tq = VSUB(Tn, Tm); Chris@82: Tt = VFNMSI(Tq, Tp); Chris@82: STM2(&(xo[12]), Tt, ovs, &(xo[0])); Chris@82: STN2(&(xo[12]), Tt, Ts, ovs); Chris@82: Tu = VFMAI(Tq, Tp); Chris@82: STM2(&(xo[4]), Tu, ovs, &(xo[0])); Chris@82: } Chris@82: { Chris@82: V Tv, Th, Ti, Tw; Chris@82: Th = VFNMS(LDK(KP707106781), Ta, T3); Chris@82: Ti = VFMA(LDK(KP707106781), Tf, Te); Chris@82: Tv = VFNMSI(Ti, Th); Chris@82: STM2(&(xo[10]), Tv, ovs, &(xo[2])); Chris@82: Tw = VFMAI(Ti, Th); Chris@82: STM2(&(xo[6]), Tw, ovs, &(xo[2])); Chris@82: STN2(&(xo[4]), Tu, Tw, ovs); Chris@82: { Chris@82: V Tl, To, Tx, Ty; Chris@82: Tl = VADD(Tj, Tk); Chris@82: To = VADD(Tm, Tn); Chris@82: Tx = VSUB(Tl, To); Chris@82: STM2(&(xo[8]), Tx, ovs, &(xo[0])); Chris@82: STN2(&(xo[8]), Tx, Tv, ovs); Chris@82: Ty = VADD(Tl, To); Chris@82: STM2(&(xo[0]), Ty, ovs, &(xo[0])); Chris@82: STN2(&(xo[0]), Ty, Tr, ovs); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: VLEAVE(); Chris@82: } Chris@82: Chris@82: static const kdft_desc desc = { 8, XSIMD_STRING("n2fv_8"), {16, 0, 10, 0}, &GENUS, 0, 2, 0, 0 }; Chris@82: Chris@82: void XSIMD(codelet_n2fv_8) (planner *p) { Chris@82: X(kdft_register) (p, n2fv_8, &desc); Chris@82: } Chris@82: Chris@82: #else Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_notw_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 8 -name n2fv_8 -with-ostride 2 -include dft/simd/n2f.h -store-multiple 2 */ Chris@82: Chris@82: /* Chris@82: * This function contains 26 FP additions, 2 FP multiplications, Chris@82: * (or, 26 additions, 2 multiplications, 0 fused multiply/add), Chris@82: * 24 stack variables, 1 constants, and 20 memory accesses Chris@82: */ Chris@82: #include "dft/simd/n2f.h" Chris@82: Chris@82: static void n2fv_8(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@82: { Chris@82: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@82: { Chris@82: INT i; Chris@82: const R *xi; Chris@82: R *xo; Chris@82: xi = ri; Chris@82: xo = ro; Chris@82: for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(16, is), MAKE_VOLATILE_STRIDE(16, os)) { Chris@82: V T3, Tj, Tf, Tk, Ta, Tn, Tc, Tm, Ts, Tu; Chris@82: { Chris@82: V T1, T2, Td, Te; Chris@82: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@82: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@82: T3 = VSUB(T1, T2); Chris@82: Tj = VADD(T1, T2); Chris@82: Td = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@82: Te = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@82: Tf = VSUB(Td, Te); Chris@82: Tk = VADD(Td, Te); Chris@82: { Chris@82: V T4, T5, T6, T7, T8, T9; Chris@82: T4 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@82: T5 = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@82: T6 = VSUB(T4, T5); Chris@82: T7 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@82: T8 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@82: T9 = VSUB(T7, T8); Chris@82: Ta = VMUL(LDK(KP707106781), VADD(T6, T9)); Chris@82: Tn = VADD(T7, T8); Chris@82: Tc = VMUL(LDK(KP707106781), VSUB(T9, T6)); Chris@82: Tm = VADD(T4, T5); Chris@82: } Chris@82: } Chris@82: { Chris@82: V Tr, Tb, Tg, Tp, Tq, Tt; Chris@82: Tb = VADD(T3, Ta); Chris@82: Tg = VBYI(VSUB(Tc, Tf)); Chris@82: Tr = VSUB(Tb, Tg); Chris@82: STM2(&(xo[14]), Tr, ovs, &(xo[2])); Chris@82: Ts = VADD(Tb, Tg); Chris@82: STM2(&(xo[2]), Ts, ovs, &(xo[2])); Chris@82: Tp = VSUB(Tj, Tk); Chris@82: Tq = VBYI(VSUB(Tn, Tm)); Chris@82: Tt = VSUB(Tp, Tq); Chris@82: STM2(&(xo[12]), Tt, ovs, &(xo[0])); Chris@82: STN2(&(xo[12]), Tt, Tr, ovs); Chris@82: Tu = VADD(Tp, Tq); Chris@82: STM2(&(xo[4]), Tu, ovs, &(xo[0])); Chris@82: } Chris@82: { Chris@82: V Tv, Th, Ti, Tw; Chris@82: Th = VSUB(T3, Ta); Chris@82: Ti = VBYI(VADD(Tf, Tc)); Chris@82: Tv = VSUB(Th, Ti); Chris@82: STM2(&(xo[10]), Tv, ovs, &(xo[2])); Chris@82: Tw = VADD(Th, Ti); Chris@82: STM2(&(xo[6]), Tw, ovs, &(xo[2])); Chris@82: STN2(&(xo[4]), Tu, Tw, ovs); Chris@82: { Chris@82: V Tl, To, Tx, Ty; Chris@82: Tl = VADD(Tj, Tk); Chris@82: To = VADD(Tm, Tn); Chris@82: Tx = VSUB(Tl, To); Chris@82: STM2(&(xo[8]), Tx, ovs, &(xo[0])); Chris@82: STN2(&(xo[8]), Tx, Tv, ovs); Chris@82: Ty = VADD(Tl, To); Chris@82: STM2(&(xo[0]), Ty, ovs, &(xo[0])); Chris@82: STN2(&(xo[0]), Ty, Ts, ovs); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: VLEAVE(); Chris@82: } Chris@82: Chris@82: static const kdft_desc desc = { 8, XSIMD_STRING("n2fv_8"), {26, 2, 0, 0}, &GENUS, 0, 2, 0, 0 }; Chris@82: Chris@82: void XSIMD(codelet_n2fv_8) (planner *p) { Chris@82: X(kdft_register) (p, n2fv_8, &desc); Chris@82: } Chris@82: Chris@82: #endif