cannam@127: /* cannam@127: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@127: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@127: * cannam@127: * This program is free software; you can redistribute it and/or modify cannam@127: * it under the terms of the GNU General Public License as published by cannam@127: * the Free Software Foundation; either version 2 of the License, or cannam@127: * (at your option) any later version. cannam@127: * cannam@127: * This program is distributed in the hope that it will be useful, cannam@127: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@127: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@127: * GNU General Public License for more details. cannam@127: * cannam@127: * You should have received a copy of the GNU General Public License cannam@127: * along with this program; if not, write to the Free Software cannam@127: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@127: * cannam@127: */ cannam@127: cannam@127: /* This file was automatically generated --- DO NOT EDIT */ cannam@127: /* Generated on Sat Jul 30 16:45:28 EDT 2016 */ cannam@127: cannam@127: #include "codelet-dft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_twidsq_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 4 -dif -name q1bv_4 -include q1b.h -sign 1 */ cannam@127: cannam@127: /* cannam@127: * This function contains 44 FP additions, 32 FP multiplications, cannam@127: * (or, 36 additions, 24 multiplications, 8 fused multiply/add), cannam@127: * 38 stack variables, 0 constants, and 32 memory accesses cannam@127: */ cannam@127: #include "q1b.h" cannam@127: cannam@127: static void q1bv_4(R *ri, R *ii, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@127: { cannam@127: { cannam@127: INT m; cannam@127: R *x; cannam@127: x = ii; cannam@127: 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@127: V Tb, Tm, Tx, TI; cannam@127: { cannam@127: V Tc, T9, T3, TG, TA, TH, TD, Ta, T6, Td, Tn, To, Tq, Tr, Tf; cannam@127: V Tg; cannam@127: { cannam@127: V T1, T2, Ty, Tz, TB, TC, T4, T5; cannam@127: T1 = LD(&(x[0]), ms, &(x[0])); cannam@127: T2 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@127: Ty = LD(&(x[WS(vs, 3)]), ms, &(x[WS(vs, 3)])); cannam@127: Tz = LD(&(x[WS(vs, 3) + WS(rs, 2)]), ms, &(x[WS(vs, 3)])); cannam@127: TB = LD(&(x[WS(vs, 3) + WS(rs, 1)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: TC = LD(&(x[WS(vs, 3) + WS(rs, 3)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: T4 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@127: T5 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@127: Tc = LD(&(x[WS(vs, 1)]), ms, &(x[WS(vs, 1)])); cannam@127: T9 = VADD(T1, T2); cannam@127: T3 = VSUB(T1, T2); cannam@127: TG = VADD(Ty, Tz); cannam@127: TA = VSUB(Ty, Tz); cannam@127: TH = VADD(TB, TC); cannam@127: TD = VSUB(TB, TC); cannam@127: Ta = VADD(T4, T5); cannam@127: T6 = VSUB(T4, T5); cannam@127: Td = LD(&(x[WS(vs, 1) + WS(rs, 2)]), ms, &(x[WS(vs, 1)])); cannam@127: Tn = LD(&(x[WS(vs, 2)]), ms, &(x[WS(vs, 2)])); cannam@127: To = LD(&(x[WS(vs, 2) + WS(rs, 2)]), ms, &(x[WS(vs, 2)])); cannam@127: Tq = LD(&(x[WS(vs, 2) + WS(rs, 1)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: Tr = LD(&(x[WS(vs, 2) + WS(rs, 3)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: Tf = LD(&(x[WS(vs, 1) + WS(rs, 1)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: Tg = LD(&(x[WS(vs, 1) + WS(rs, 3)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: } cannam@127: { cannam@127: V Tk, Te, Tv, Tp, Tw, Ts, Tl, Th, T7, TE, Tu, TF; cannam@127: ST(&(x[0]), VADD(T9, Ta), ms, &(x[0])); cannam@127: Tk = VADD(Tc, Td); cannam@127: Te = VSUB(Tc, Td); cannam@127: Tv = VADD(Tn, To); cannam@127: Tp = VSUB(Tn, To); cannam@127: Tw = VADD(Tq, Tr); cannam@127: Ts = VSUB(Tq, Tr); cannam@127: Tl = VADD(Tf, Tg); cannam@127: Th = VSUB(Tf, Tg); cannam@127: ST(&(x[WS(rs, 3)]), VADD(TG, TH), ms, &(x[WS(rs, 1)])); cannam@127: T7 = BYTW(&(W[TWVL * 4]), VFNMSI(T6, T3)); cannam@127: TE = BYTW(&(W[TWVL * 4]), VFNMSI(TD, TA)); cannam@127: { cannam@127: V Tt, Ti, Tj, T8; cannam@127: T8 = BYTW(&(W[0]), VFMAI(T6, T3)); cannam@127: ST(&(x[WS(rs, 2)]), VADD(Tv, Tw), ms, &(x[0])); cannam@127: Tt = BYTW(&(W[TWVL * 4]), VFNMSI(Ts, Tp)); cannam@127: ST(&(x[WS(rs, 1)]), VADD(Tk, Tl), ms, &(x[WS(rs, 1)])); cannam@127: Ti = BYTW(&(W[TWVL * 4]), VFNMSI(Th, Te)); cannam@127: Tj = BYTW(&(W[0]), VFMAI(Th, Te)); cannam@127: ST(&(x[WS(vs, 3)]), T7, ms, &(x[WS(vs, 3)])); cannam@127: ST(&(x[WS(vs, 3) + WS(rs, 3)]), TE, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: ST(&(x[WS(vs, 1)]), T8, ms, &(x[WS(vs, 1)])); cannam@127: Tu = BYTW(&(W[0]), VFMAI(Ts, Tp)); cannam@127: ST(&(x[WS(vs, 3) + WS(rs, 2)]), Tt, ms, &(x[WS(vs, 3)])); cannam@127: TF = BYTW(&(W[0]), VFMAI(TD, TA)); cannam@127: ST(&(x[WS(vs, 3) + WS(rs, 1)]), Ti, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: ST(&(x[WS(vs, 1) + WS(rs, 1)]), Tj, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: } cannam@127: Tb = BYTW(&(W[TWVL * 2]), VSUB(T9, Ta)); cannam@127: Tm = BYTW(&(W[TWVL * 2]), VSUB(Tk, Tl)); cannam@127: Tx = BYTW(&(W[TWVL * 2]), VSUB(Tv, Tw)); cannam@127: ST(&(x[WS(vs, 1) + WS(rs, 2)]), Tu, ms, &(x[WS(vs, 1)])); cannam@127: TI = BYTW(&(W[TWVL * 2]), VSUB(TG, TH)); cannam@127: ST(&(x[WS(vs, 1) + WS(rs, 3)]), TF, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: } cannam@127: } cannam@127: ST(&(x[WS(vs, 2)]), Tb, ms, &(x[WS(vs, 2)])); cannam@127: ST(&(x[WS(vs, 2) + WS(rs, 1)]), Tm, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: ST(&(x[WS(vs, 2) + WS(rs, 2)]), Tx, ms, &(x[WS(vs, 2)])); cannam@127: ST(&(x[WS(vs, 2) + WS(rs, 3)]), TI, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: VTW(0, 1), cannam@127: VTW(0, 2), cannam@127: VTW(0, 3), cannam@127: {TW_NEXT, VL, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 4, XSIMD_STRING("q1bv_4"), twinstr, &GENUS, {36, 24, 8, 0}, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_q1bv_4) (planner *p) { cannam@127: X(kdft_difsq_register) (p, q1bv_4, &desc); cannam@127: } cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_twidsq_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 4 -dif -name q1bv_4 -include q1b.h -sign 1 */ cannam@127: cannam@127: /* cannam@127: * This function contains 44 FP additions, 24 FP multiplications, cannam@127: * (or, 44 additions, 24 multiplications, 0 fused multiply/add), cannam@127: * 22 stack variables, 0 constants, and 32 memory accesses cannam@127: */ cannam@127: #include "q1b.h" cannam@127: cannam@127: static void q1bv_4(R *ri, R *ii, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@127: { cannam@127: { cannam@127: INT m; cannam@127: R *x; cannam@127: x = ii; cannam@127: 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@127: V T3, T9, TA, TG, TD, TH, T6, Ta, Te, Tk, Tp, Tv, Ts, Tw, Th; cannam@127: V Tl; cannam@127: { cannam@127: V T1, T2, Ty, Tz; cannam@127: T1 = LD(&(x[0]), ms, &(x[0])); cannam@127: T2 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@127: T3 = VSUB(T1, T2); cannam@127: T9 = VADD(T1, T2); cannam@127: Ty = LD(&(x[WS(vs, 3)]), ms, &(x[WS(vs, 3)])); cannam@127: Tz = LD(&(x[WS(vs, 3) + WS(rs, 2)]), ms, &(x[WS(vs, 3)])); cannam@127: TA = VSUB(Ty, Tz); cannam@127: TG = VADD(Ty, Tz); cannam@127: } cannam@127: { cannam@127: V TB, TC, T4, T5; cannam@127: TB = LD(&(x[WS(vs, 3) + WS(rs, 1)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: TC = LD(&(x[WS(vs, 3) + WS(rs, 3)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: TD = VBYI(VSUB(TB, TC)); cannam@127: TH = VADD(TB, TC); cannam@127: T4 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@127: T5 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@127: T6 = VBYI(VSUB(T4, T5)); cannam@127: Ta = VADD(T4, T5); cannam@127: } cannam@127: { cannam@127: V Tc, Td, Tn, To; cannam@127: Tc = LD(&(x[WS(vs, 1)]), ms, &(x[WS(vs, 1)])); cannam@127: Td = LD(&(x[WS(vs, 1) + WS(rs, 2)]), ms, &(x[WS(vs, 1)])); cannam@127: Te = VSUB(Tc, Td); cannam@127: Tk = VADD(Tc, Td); cannam@127: Tn = LD(&(x[WS(vs, 2)]), ms, &(x[WS(vs, 2)])); cannam@127: To = LD(&(x[WS(vs, 2) + WS(rs, 2)]), ms, &(x[WS(vs, 2)])); cannam@127: Tp = VSUB(Tn, To); cannam@127: Tv = VADD(Tn, To); cannam@127: } cannam@127: { cannam@127: V Tq, Tr, Tf, Tg; cannam@127: Tq = LD(&(x[WS(vs, 2) + WS(rs, 1)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: Tr = LD(&(x[WS(vs, 2) + WS(rs, 3)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: Ts = VBYI(VSUB(Tq, Tr)); cannam@127: Tw = VADD(Tq, Tr); cannam@127: Tf = LD(&(x[WS(vs, 1) + WS(rs, 1)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: Tg = LD(&(x[WS(vs, 1) + WS(rs, 3)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: Th = VBYI(VSUB(Tf, Tg)); cannam@127: Tl = VADD(Tf, Tg); cannam@127: } cannam@127: ST(&(x[0]), VADD(T9, Ta), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 1)]), VADD(Tk, Tl), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 2)]), VADD(Tv, Tw), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 3)]), VADD(TG, TH), ms, &(x[WS(rs, 1)])); cannam@127: { cannam@127: V T7, Ti, Tt, TE; cannam@127: T7 = BYTW(&(W[TWVL * 4]), VSUB(T3, T6)); cannam@127: ST(&(x[WS(vs, 3)]), T7, ms, &(x[WS(vs, 3)])); cannam@127: Ti = BYTW(&(W[TWVL * 4]), VSUB(Te, Th)); cannam@127: ST(&(x[WS(vs, 3) + WS(rs, 1)]), Ti, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: Tt = BYTW(&(W[TWVL * 4]), VSUB(Tp, Ts)); cannam@127: ST(&(x[WS(vs, 3) + WS(rs, 2)]), Tt, ms, &(x[WS(vs, 3)])); cannam@127: TE = BYTW(&(W[TWVL * 4]), VSUB(TA, TD)); cannam@127: ST(&(x[WS(vs, 3) + WS(rs, 3)]), TE, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: } cannam@127: { cannam@127: V T8, Tj, Tu, TF; cannam@127: T8 = BYTW(&(W[0]), VADD(T3, T6)); cannam@127: ST(&(x[WS(vs, 1)]), T8, ms, &(x[WS(vs, 1)])); cannam@127: Tj = BYTW(&(W[0]), VADD(Te, Th)); cannam@127: ST(&(x[WS(vs, 1) + WS(rs, 1)]), Tj, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: Tu = BYTW(&(W[0]), VADD(Tp, Ts)); cannam@127: ST(&(x[WS(vs, 1) + WS(rs, 2)]), Tu, ms, &(x[WS(vs, 1)])); cannam@127: TF = BYTW(&(W[0]), VADD(TA, TD)); cannam@127: ST(&(x[WS(vs, 1) + WS(rs, 3)]), TF, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: } cannam@127: { cannam@127: V Tb, Tm, Tx, TI; cannam@127: Tb = BYTW(&(W[TWVL * 2]), VSUB(T9, Ta)); cannam@127: ST(&(x[WS(vs, 2)]), Tb, ms, &(x[WS(vs, 2)])); cannam@127: Tm = BYTW(&(W[TWVL * 2]), VSUB(Tk, Tl)); cannam@127: ST(&(x[WS(vs, 2) + WS(rs, 1)]), Tm, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: Tx = BYTW(&(W[TWVL * 2]), VSUB(Tv, Tw)); cannam@127: ST(&(x[WS(vs, 2) + WS(rs, 2)]), Tx, ms, &(x[WS(vs, 2)])); cannam@127: TI = BYTW(&(W[TWVL * 2]), VSUB(TG, TH)); cannam@127: ST(&(x[WS(vs, 2) + WS(rs, 3)]), TI, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: } cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: VTW(0, 1), cannam@127: VTW(0, 2), cannam@127: VTW(0, 3), cannam@127: {TW_NEXT, VL, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 4, XSIMD_STRING("q1bv_4"), twinstr, &GENUS, {44, 24, 0, 0}, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_q1bv_4) (planner *p) { cannam@127: X(kdft_difsq_register) (p, q1bv_4, &desc); cannam@127: } cannam@127: #endif /* HAVE_FMA */