cannam@167: /* cannam@167: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@167: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@167: * cannam@167: * This program is free software; you can redistribute it and/or modify cannam@167: * it under the terms of the GNU General Public License as published by cannam@167: * the Free Software Foundation; either version 2 of the License, or cannam@167: * (at your option) any later version. cannam@167: * cannam@167: * This program is distributed in the hope that it will be useful, cannam@167: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@167: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@167: * GNU General Public License for more details. cannam@167: * cannam@167: * You should have received a copy of the GNU General Public License cannam@167: * along with this program; if not, write to the Free Software cannam@167: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@167: * cannam@167: */ cannam@167: cannam@167: /* This file was automatically generated --- DO NOT EDIT */ cannam@167: /* Generated on Thu May 24 08:06:13 EDT 2018 */ cannam@167: cannam@167: #include "dft/codelet-dft.h" cannam@167: cannam@167: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_twidsq_c.native -fma -simd -compact -variables 4 -pipeline-latency 8 -n 4 -dif -name q1fv_4 -include dft/simd/q1f.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 44 FP additions, 32 FP multiplications, cannam@167: * (or, 36 additions, 24 multiplications, 8 fused multiply/add), cannam@167: * 22 stack variables, 0 constants, and 32 memory accesses cannam@167: */ cannam@167: #include "dft/simd/q1f.h" cannam@167: cannam@167: static void q1fv_4(R *ri, R *ii, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@167: { cannam@167: { cannam@167: INT m; cannam@167: R *x; cannam@167: x = ri; cannam@167: for (m = mb, W = W + (mb * ((TWVL / VL) * 6)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 6), MAKE_VOLATILE_STRIDE(8, rs), MAKE_VOLATILE_STRIDE(8, vs)) { cannam@167: V T3, T9, TA, TG, TD, TH, T6, Ta, Te, Tk, Tp, Tv, Ts, Tw, Th; cannam@167: V Tl; cannam@167: { cannam@167: V T1, T2, Ty, Tz; cannam@167: T1 = LD(&(x[0]), ms, &(x[0])); cannam@167: T2 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@167: T3 = VSUB(T1, T2); cannam@167: T9 = VADD(T1, T2); cannam@167: Ty = LD(&(x[WS(vs, 3)]), ms, &(x[WS(vs, 3)])); cannam@167: Tz = LD(&(x[WS(vs, 3) + WS(rs, 2)]), ms, &(x[WS(vs, 3)])); cannam@167: TA = VSUB(Ty, Tz); cannam@167: TG = VADD(Ty, Tz); cannam@167: } cannam@167: { cannam@167: V TB, TC, T4, T5; cannam@167: TB = LD(&(x[WS(vs, 3) + WS(rs, 1)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: TC = LD(&(x[WS(vs, 3) + WS(rs, 3)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: TD = VSUB(TB, TC); cannam@167: TH = VADD(TB, TC); cannam@167: T4 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@167: T5 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@167: T6 = VSUB(T4, T5); cannam@167: Ta = VADD(T4, T5); cannam@167: } cannam@167: { cannam@167: V Tc, Td, Tn, To; cannam@167: Tc = LD(&(x[WS(vs, 1)]), ms, &(x[WS(vs, 1)])); cannam@167: Td = LD(&(x[WS(vs, 1) + WS(rs, 2)]), ms, &(x[WS(vs, 1)])); cannam@167: Te = VSUB(Tc, Td); cannam@167: Tk = VADD(Tc, Td); cannam@167: Tn = LD(&(x[WS(vs, 2)]), ms, &(x[WS(vs, 2)])); cannam@167: To = LD(&(x[WS(vs, 2) + WS(rs, 2)]), ms, &(x[WS(vs, 2)])); cannam@167: Tp = VSUB(Tn, To); cannam@167: Tv = VADD(Tn, To); cannam@167: } cannam@167: { cannam@167: V Tq, Tr, Tf, Tg; cannam@167: Tq = LD(&(x[WS(vs, 2) + WS(rs, 1)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: Tr = LD(&(x[WS(vs, 2) + WS(rs, 3)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: Ts = VSUB(Tq, Tr); cannam@167: Tw = VADD(Tq, Tr); cannam@167: Tf = LD(&(x[WS(vs, 1) + WS(rs, 1)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: Tg = LD(&(x[WS(vs, 1) + WS(rs, 3)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: Th = VSUB(Tf, Tg); cannam@167: Tl = VADD(Tf, Tg); cannam@167: } cannam@167: ST(&(x[0]), VADD(T9, Ta), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 1)]), VADD(Tk, Tl), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[WS(rs, 2)]), VADD(Tv, Tw), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 3)]), VADD(TG, TH), ms, &(x[WS(rs, 1)])); cannam@167: { cannam@167: V T7, Ti, Tt, TE; cannam@167: T7 = BYTWJ(&(W[0]), VFNMSI(T6, T3)); cannam@167: ST(&(x[WS(vs, 1)]), T7, ms, &(x[WS(vs, 1)])); cannam@167: Ti = BYTWJ(&(W[0]), VFNMSI(Th, Te)); cannam@167: ST(&(x[WS(vs, 1) + WS(rs, 1)]), Ti, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: Tt = BYTWJ(&(W[0]), VFNMSI(Ts, Tp)); cannam@167: ST(&(x[WS(vs, 1) + WS(rs, 2)]), Tt, ms, &(x[WS(vs, 1)])); cannam@167: TE = BYTWJ(&(W[0]), VFNMSI(TD, TA)); cannam@167: ST(&(x[WS(vs, 1) + WS(rs, 3)]), TE, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: } cannam@167: { cannam@167: V T8, Tj, Tu, TF; cannam@167: T8 = BYTWJ(&(W[TWVL * 4]), VFMAI(T6, T3)); cannam@167: ST(&(x[WS(vs, 3)]), T8, ms, &(x[WS(vs, 3)])); cannam@167: Tj = BYTWJ(&(W[TWVL * 4]), VFMAI(Th, Te)); cannam@167: ST(&(x[WS(vs, 3) + WS(rs, 1)]), Tj, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: Tu = BYTWJ(&(W[TWVL * 4]), VFMAI(Ts, Tp)); cannam@167: ST(&(x[WS(vs, 3) + WS(rs, 2)]), Tu, ms, &(x[WS(vs, 3)])); cannam@167: TF = BYTWJ(&(W[TWVL * 4]), VFMAI(TD, TA)); cannam@167: ST(&(x[WS(vs, 3) + WS(rs, 3)]), TF, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: } cannam@167: { cannam@167: V Tb, Tm, Tx, TI; cannam@167: Tb = BYTWJ(&(W[TWVL * 2]), VSUB(T9, Ta)); cannam@167: ST(&(x[WS(vs, 2)]), Tb, ms, &(x[WS(vs, 2)])); cannam@167: Tm = BYTWJ(&(W[TWVL * 2]), VSUB(Tk, Tl)); cannam@167: ST(&(x[WS(vs, 2) + WS(rs, 1)]), Tm, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: Tx = BYTWJ(&(W[TWVL * 2]), VSUB(Tv, Tw)); cannam@167: ST(&(x[WS(vs, 2) + WS(rs, 2)]), Tx, ms, &(x[WS(vs, 2)])); cannam@167: TI = BYTWJ(&(W[TWVL * 2]), VSUB(TG, TH)); cannam@167: ST(&(x[WS(vs, 2) + WS(rs, 3)]), TI, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: } cannam@167: } cannam@167: } cannam@167: VLEAVE(); cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: VTW(0, 1), cannam@167: VTW(0, 2), cannam@167: VTW(0, 3), cannam@167: {TW_NEXT, VL, 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 4, XSIMD_STRING("q1fv_4"), twinstr, &GENUS, {36, 24, 8, 0}, 0, 0, 0 }; cannam@167: cannam@167: void XSIMD(codelet_q1fv_4) (planner *p) { cannam@167: X(kdft_difsq_register) (p, q1fv_4, &desc); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_twidsq_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 4 -dif -name q1fv_4 -include dft/simd/q1f.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 44 FP additions, 24 FP multiplications, cannam@167: * (or, 44 additions, 24 multiplications, 0 fused multiply/add), cannam@167: * 22 stack variables, 0 constants, and 32 memory accesses cannam@167: */ cannam@167: #include "dft/simd/q1f.h" cannam@167: cannam@167: static void q1fv_4(R *ri, R *ii, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@167: { cannam@167: { cannam@167: INT m; cannam@167: R *x; cannam@167: x = ri; cannam@167: for (m = mb, W = W + (mb * ((TWVL / VL) * 6)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 6), MAKE_VOLATILE_STRIDE(8, rs), MAKE_VOLATILE_STRIDE(8, vs)) { cannam@167: V T3, T9, TA, TG, TD, TH, T6, Ta, Te, Tk, Tp, Tv, Ts, Tw, Th; cannam@167: V Tl; cannam@167: { cannam@167: V T1, T2, Ty, Tz; cannam@167: T1 = LD(&(x[0]), ms, &(x[0])); cannam@167: T2 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@167: T3 = VSUB(T1, T2); cannam@167: T9 = VADD(T1, T2); cannam@167: Ty = LD(&(x[WS(vs, 3)]), ms, &(x[WS(vs, 3)])); cannam@167: Tz = LD(&(x[WS(vs, 3) + WS(rs, 2)]), ms, &(x[WS(vs, 3)])); cannam@167: TA = VSUB(Ty, Tz); cannam@167: TG = VADD(Ty, Tz); cannam@167: } cannam@167: { cannam@167: V TB, TC, T4, T5; cannam@167: TB = LD(&(x[WS(vs, 3) + WS(rs, 1)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: TC = LD(&(x[WS(vs, 3) + WS(rs, 3)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: TD = VBYI(VSUB(TB, TC)); cannam@167: TH = VADD(TB, TC); cannam@167: T4 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@167: T5 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@167: T6 = VBYI(VSUB(T4, T5)); cannam@167: Ta = VADD(T4, T5); cannam@167: } cannam@167: { cannam@167: V Tc, Td, Tn, To; cannam@167: Tc = LD(&(x[WS(vs, 1)]), ms, &(x[WS(vs, 1)])); cannam@167: Td = LD(&(x[WS(vs, 1) + WS(rs, 2)]), ms, &(x[WS(vs, 1)])); cannam@167: Te = VSUB(Tc, Td); cannam@167: Tk = VADD(Tc, Td); cannam@167: Tn = LD(&(x[WS(vs, 2)]), ms, &(x[WS(vs, 2)])); cannam@167: To = LD(&(x[WS(vs, 2) + WS(rs, 2)]), ms, &(x[WS(vs, 2)])); cannam@167: Tp = VSUB(Tn, To); cannam@167: Tv = VADD(Tn, To); cannam@167: } cannam@167: { cannam@167: V Tq, Tr, Tf, Tg; cannam@167: Tq = LD(&(x[WS(vs, 2) + WS(rs, 1)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: Tr = LD(&(x[WS(vs, 2) + WS(rs, 3)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: Ts = VBYI(VSUB(Tq, Tr)); cannam@167: Tw = VADD(Tq, Tr); cannam@167: Tf = LD(&(x[WS(vs, 1) + WS(rs, 1)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: Tg = LD(&(x[WS(vs, 1) + WS(rs, 3)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: Th = VBYI(VSUB(Tf, Tg)); cannam@167: Tl = VADD(Tf, Tg); cannam@167: } cannam@167: ST(&(x[0]), VADD(T9, Ta), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 1)]), VADD(Tk, Tl), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[WS(rs, 2)]), VADD(Tv, Tw), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 3)]), VADD(TG, TH), ms, &(x[WS(rs, 1)])); cannam@167: { cannam@167: V T7, Ti, Tt, TE; cannam@167: T7 = BYTWJ(&(W[0]), VSUB(T3, T6)); cannam@167: ST(&(x[WS(vs, 1)]), T7, ms, &(x[WS(vs, 1)])); cannam@167: Ti = BYTWJ(&(W[0]), VSUB(Te, Th)); cannam@167: ST(&(x[WS(vs, 1) + WS(rs, 1)]), Ti, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: Tt = BYTWJ(&(W[0]), VSUB(Tp, Ts)); cannam@167: ST(&(x[WS(vs, 1) + WS(rs, 2)]), Tt, ms, &(x[WS(vs, 1)])); cannam@167: TE = BYTWJ(&(W[0]), VSUB(TA, TD)); cannam@167: ST(&(x[WS(vs, 1) + WS(rs, 3)]), TE, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: } cannam@167: { cannam@167: V T8, Tj, Tu, TF; cannam@167: T8 = BYTWJ(&(W[TWVL * 4]), VADD(T3, T6)); cannam@167: ST(&(x[WS(vs, 3)]), T8, ms, &(x[WS(vs, 3)])); cannam@167: Tj = BYTWJ(&(W[TWVL * 4]), VADD(Te, Th)); cannam@167: ST(&(x[WS(vs, 3) + WS(rs, 1)]), Tj, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: Tu = BYTWJ(&(W[TWVL * 4]), VADD(Tp, Ts)); cannam@167: ST(&(x[WS(vs, 3) + WS(rs, 2)]), Tu, ms, &(x[WS(vs, 3)])); cannam@167: TF = BYTWJ(&(W[TWVL * 4]), VADD(TA, TD)); cannam@167: ST(&(x[WS(vs, 3) + WS(rs, 3)]), TF, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: } cannam@167: { cannam@167: V Tb, Tm, Tx, TI; cannam@167: Tb = BYTWJ(&(W[TWVL * 2]), VSUB(T9, Ta)); cannam@167: ST(&(x[WS(vs, 2)]), Tb, ms, &(x[WS(vs, 2)])); cannam@167: Tm = BYTWJ(&(W[TWVL * 2]), VSUB(Tk, Tl)); cannam@167: ST(&(x[WS(vs, 2) + WS(rs, 1)]), Tm, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: Tx = BYTWJ(&(W[TWVL * 2]), VSUB(Tv, Tw)); cannam@167: ST(&(x[WS(vs, 2) + WS(rs, 2)]), Tx, ms, &(x[WS(vs, 2)])); cannam@167: TI = BYTWJ(&(W[TWVL * 2]), VSUB(TG, TH)); cannam@167: ST(&(x[WS(vs, 2) + WS(rs, 3)]), TI, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: } cannam@167: } cannam@167: } cannam@167: VLEAVE(); cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: VTW(0, 1), cannam@167: VTW(0, 2), cannam@167: VTW(0, 3), cannam@167: {TW_NEXT, VL, 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 4, XSIMD_STRING("q1fv_4"), twinstr, &GENUS, {44, 24, 0, 0}, 0, 0, 0 }; cannam@167: cannam@167: void XSIMD(codelet_q1fv_4) (planner *p) { cannam@167: X(kdft_difsq_register) (p, q1fv_4, &desc); cannam@167: } cannam@167: #endif