Chris@10: /* Chris@10: * Copyright (c) 2003, 2007-11 Matteo Frigo Chris@10: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology Chris@10: * Chris@10: * This program is free software; you can redistribute it and/or modify Chris@10: * it under the terms of the GNU General Public License as published by Chris@10: * the Free Software Foundation; either version 2 of the License, or Chris@10: * (at your option) any later version. Chris@10: * Chris@10: * This program is distributed in the hope that it will be useful, Chris@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@10: * GNU General Public License for more details. Chris@10: * Chris@10: * You should have received a copy of the GNU General Public License Chris@10: * along with this program; if not, write to the Free Software Chris@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@10: * Chris@10: */ Chris@10: Chris@10: /* This file was automatically generated --- DO NOT EDIT */ Chris@10: /* Generated on Sun Nov 25 07:39:33 EST 2012 */ Chris@10: Chris@10: #include "codelet-dft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* 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 */ Chris@10: Chris@10: /* Chris@10: * This function contains 44 FP additions, 32 FP multiplications, Chris@10: * (or, 36 additions, 24 multiplications, 8 fused multiply/add), Chris@10: * 38 stack variables, 0 constants, and 32 memory accesses Chris@10: */ Chris@10: #include "q1b.h" Chris@10: Chris@10: static void q1bv_4(R *ri, R *ii, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) Chris@10: { Chris@10: { Chris@10: INT m; Chris@10: R *x; Chris@10: x = ii; Chris@10: 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)) { Chris@10: V Tb, Tm, Tx, TI; Chris@10: { Chris@10: V Tc, T9, T3, TG, TA, TH, TD, Ta, T6, Td, Tn, To, Tq, Tr, Tf; Chris@10: V Tg; Chris@10: { Chris@10: V T1, T2, Ty, Tz, TB, TC, T4, T5; Chris@10: T1 = LD(&(x[0]), ms, &(x[0])); Chris@10: T2 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); Chris@10: Ty = LD(&(x[WS(vs, 3)]), ms, &(x[WS(vs, 3)])); Chris@10: Tz = LD(&(x[WS(vs, 3) + WS(rs, 2)]), ms, &(x[WS(vs, 3)])); Chris@10: TB = LD(&(x[WS(vs, 3) + WS(rs, 1)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); Chris@10: TC = LD(&(x[WS(vs, 3) + WS(rs, 3)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); Chris@10: T4 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); Chris@10: T5 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); Chris@10: Tc = LD(&(x[WS(vs, 1)]), ms, &(x[WS(vs, 1)])); Chris@10: T9 = VADD(T1, T2); Chris@10: T3 = VSUB(T1, T2); Chris@10: TG = VADD(Ty, Tz); Chris@10: TA = VSUB(Ty, Tz); Chris@10: TH = VADD(TB, TC); Chris@10: TD = VSUB(TB, TC); Chris@10: Ta = VADD(T4, T5); Chris@10: T6 = VSUB(T4, T5); Chris@10: Td = LD(&(x[WS(vs, 1) + WS(rs, 2)]), ms, &(x[WS(vs, 1)])); Chris@10: Tn = LD(&(x[WS(vs, 2)]), ms, &(x[WS(vs, 2)])); Chris@10: To = LD(&(x[WS(vs, 2) + WS(rs, 2)]), ms, &(x[WS(vs, 2)])); Chris@10: Tq = LD(&(x[WS(vs, 2) + WS(rs, 1)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); Chris@10: Tr = LD(&(x[WS(vs, 2) + WS(rs, 3)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); Chris@10: Tf = LD(&(x[WS(vs, 1) + WS(rs, 1)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); Chris@10: Tg = LD(&(x[WS(vs, 1) + WS(rs, 3)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); Chris@10: } Chris@10: { Chris@10: V Tk, Te, Tv, Tp, Tw, Ts, Tl, Th, T7, TE, Tu, TF; Chris@10: ST(&(x[0]), VADD(T9, Ta), ms, &(x[0])); Chris@10: Tk = VADD(Tc, Td); Chris@10: Te = VSUB(Tc, Td); Chris@10: Tv = VADD(Tn, To); Chris@10: Tp = VSUB(Tn, To); Chris@10: Tw = VADD(Tq, Tr); Chris@10: Ts = VSUB(Tq, Tr); Chris@10: Tl = VADD(Tf, Tg); Chris@10: Th = VSUB(Tf, Tg); Chris@10: ST(&(x[WS(rs, 3)]), VADD(TG, TH), ms, &(x[WS(rs, 1)])); Chris@10: T7 = BYTW(&(W[TWVL * 4]), VFNMSI(T6, T3)); Chris@10: TE = BYTW(&(W[TWVL * 4]), VFNMSI(TD, TA)); Chris@10: { Chris@10: V Tt, Ti, Tj, T8; Chris@10: T8 = BYTW(&(W[0]), VFMAI(T6, T3)); Chris@10: ST(&(x[WS(rs, 2)]), VADD(Tv, Tw), ms, &(x[0])); Chris@10: Tt = BYTW(&(W[TWVL * 4]), VFNMSI(Ts, Tp)); Chris@10: ST(&(x[WS(rs, 1)]), VADD(Tk, Tl), ms, &(x[WS(rs, 1)])); Chris@10: Ti = BYTW(&(W[TWVL * 4]), VFNMSI(Th, Te)); Chris@10: Tj = BYTW(&(W[0]), VFMAI(Th, Te)); Chris@10: ST(&(x[WS(vs, 3)]), T7, ms, &(x[WS(vs, 3)])); Chris@10: ST(&(x[WS(vs, 3) + WS(rs, 3)]), TE, ms, &(x[WS(vs, 3) + WS(rs, 1)])); Chris@10: ST(&(x[WS(vs, 1)]), T8, ms, &(x[WS(vs, 1)])); Chris@10: Tu = BYTW(&(W[0]), VFMAI(Ts, Tp)); Chris@10: ST(&(x[WS(vs, 3) + WS(rs, 2)]), Tt, ms, &(x[WS(vs, 3)])); Chris@10: TF = BYTW(&(W[0]), VFMAI(TD, TA)); Chris@10: ST(&(x[WS(vs, 3) + WS(rs, 1)]), Ti, ms, &(x[WS(vs, 3) + WS(rs, 1)])); Chris@10: ST(&(x[WS(vs, 1) + WS(rs, 1)]), Tj, ms, &(x[WS(vs, 1) + WS(rs, 1)])); Chris@10: } Chris@10: Tb = BYTW(&(W[TWVL * 2]), VSUB(T9, Ta)); Chris@10: Tm = BYTW(&(W[TWVL * 2]), VSUB(Tk, Tl)); Chris@10: Tx = BYTW(&(W[TWVL * 2]), VSUB(Tv, Tw)); Chris@10: ST(&(x[WS(vs, 1) + WS(rs, 2)]), Tu, ms, &(x[WS(vs, 1)])); Chris@10: TI = BYTW(&(W[TWVL * 2]), VSUB(TG, TH)); Chris@10: ST(&(x[WS(vs, 1) + WS(rs, 3)]), TF, ms, &(x[WS(vs, 1) + WS(rs, 1)])); Chris@10: } Chris@10: } Chris@10: ST(&(x[WS(vs, 2)]), Tb, ms, &(x[WS(vs, 2)])); Chris@10: ST(&(x[WS(vs, 2) + WS(rs, 1)]), Tm, ms, &(x[WS(vs, 2) + WS(rs, 1)])); Chris@10: ST(&(x[WS(vs, 2) + WS(rs, 2)]), Tx, ms, &(x[WS(vs, 2)])); Chris@10: ST(&(x[WS(vs, 2) + WS(rs, 3)]), TI, ms, &(x[WS(vs, 2) + WS(rs, 1)])); Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: VTW(0, 1), Chris@10: VTW(0, 2), Chris@10: VTW(0, 3), Chris@10: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 4, XSIMD_STRING("q1bv_4"), twinstr, &GENUS, {36, 24, 8, 0}, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_q1bv_4) (planner *p) { Chris@10: X(kdft_difsq_register) (p, q1bv_4, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* 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 */ Chris@10: Chris@10: /* Chris@10: * This function contains 44 FP additions, 24 FP multiplications, Chris@10: * (or, 44 additions, 24 multiplications, 0 fused multiply/add), Chris@10: * 22 stack variables, 0 constants, and 32 memory accesses Chris@10: */ Chris@10: #include "q1b.h" Chris@10: Chris@10: static void q1bv_4(R *ri, R *ii, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) Chris@10: { Chris@10: { Chris@10: INT m; Chris@10: R *x; Chris@10: x = ii; Chris@10: 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)) { Chris@10: V T3, T9, TA, TG, TD, TH, T6, Ta, Te, Tk, Tp, Tv, Ts, Tw, Th; Chris@10: V Tl; Chris@10: { Chris@10: V T1, T2, Ty, Tz; Chris@10: T1 = LD(&(x[0]), ms, &(x[0])); Chris@10: T2 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); Chris@10: T3 = VSUB(T1, T2); Chris@10: T9 = VADD(T1, T2); Chris@10: Ty = LD(&(x[WS(vs, 3)]), ms, &(x[WS(vs, 3)])); Chris@10: Tz = LD(&(x[WS(vs, 3) + WS(rs, 2)]), ms, &(x[WS(vs, 3)])); Chris@10: TA = VSUB(Ty, Tz); Chris@10: TG = VADD(Ty, Tz); Chris@10: } Chris@10: { Chris@10: V TB, TC, T4, T5; Chris@10: TB = LD(&(x[WS(vs, 3) + WS(rs, 1)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); Chris@10: TC = LD(&(x[WS(vs, 3) + WS(rs, 3)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); Chris@10: TD = VBYI(VSUB(TB, TC)); Chris@10: TH = VADD(TB, TC); Chris@10: T4 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); Chris@10: T5 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); Chris@10: T6 = VBYI(VSUB(T4, T5)); Chris@10: Ta = VADD(T4, T5); Chris@10: } Chris@10: { Chris@10: V Tc, Td, Tn, To; Chris@10: Tc = LD(&(x[WS(vs, 1)]), ms, &(x[WS(vs, 1)])); Chris@10: Td = LD(&(x[WS(vs, 1) + WS(rs, 2)]), ms, &(x[WS(vs, 1)])); Chris@10: Te = VSUB(Tc, Td); Chris@10: Tk = VADD(Tc, Td); Chris@10: Tn = LD(&(x[WS(vs, 2)]), ms, &(x[WS(vs, 2)])); Chris@10: To = LD(&(x[WS(vs, 2) + WS(rs, 2)]), ms, &(x[WS(vs, 2)])); Chris@10: Tp = VSUB(Tn, To); Chris@10: Tv = VADD(Tn, To); Chris@10: } Chris@10: { Chris@10: V Tq, Tr, Tf, Tg; Chris@10: Tq = LD(&(x[WS(vs, 2) + WS(rs, 1)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); Chris@10: Tr = LD(&(x[WS(vs, 2) + WS(rs, 3)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); Chris@10: Ts = VBYI(VSUB(Tq, Tr)); Chris@10: Tw = VADD(Tq, Tr); Chris@10: Tf = LD(&(x[WS(vs, 1) + WS(rs, 1)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); Chris@10: Tg = LD(&(x[WS(vs, 1) + WS(rs, 3)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); Chris@10: Th = VBYI(VSUB(Tf, Tg)); Chris@10: Tl = VADD(Tf, Tg); Chris@10: } Chris@10: ST(&(x[0]), VADD(T9, Ta), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 1)]), VADD(Tk, Tl), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 2)]), VADD(Tv, Tw), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 3)]), VADD(TG, TH), ms, &(x[WS(rs, 1)])); Chris@10: { Chris@10: V T7, Ti, Tt, TE; Chris@10: T7 = BYTW(&(W[TWVL * 4]), VSUB(T3, T6)); Chris@10: ST(&(x[WS(vs, 3)]), T7, ms, &(x[WS(vs, 3)])); Chris@10: Ti = BYTW(&(W[TWVL * 4]), VSUB(Te, Th)); Chris@10: ST(&(x[WS(vs, 3) + WS(rs, 1)]), Ti, ms, &(x[WS(vs, 3) + WS(rs, 1)])); Chris@10: Tt = BYTW(&(W[TWVL * 4]), VSUB(Tp, Ts)); Chris@10: ST(&(x[WS(vs, 3) + WS(rs, 2)]), Tt, ms, &(x[WS(vs, 3)])); Chris@10: TE = BYTW(&(W[TWVL * 4]), VSUB(TA, TD)); Chris@10: ST(&(x[WS(vs, 3) + WS(rs, 3)]), TE, ms, &(x[WS(vs, 3) + WS(rs, 1)])); Chris@10: } Chris@10: { Chris@10: V T8, Tj, Tu, TF; Chris@10: T8 = BYTW(&(W[0]), VADD(T3, T6)); Chris@10: ST(&(x[WS(vs, 1)]), T8, ms, &(x[WS(vs, 1)])); Chris@10: Tj = BYTW(&(W[0]), VADD(Te, Th)); Chris@10: ST(&(x[WS(vs, 1) + WS(rs, 1)]), Tj, ms, &(x[WS(vs, 1) + WS(rs, 1)])); Chris@10: Tu = BYTW(&(W[0]), VADD(Tp, Ts)); Chris@10: ST(&(x[WS(vs, 1) + WS(rs, 2)]), Tu, ms, &(x[WS(vs, 1)])); Chris@10: TF = BYTW(&(W[0]), VADD(TA, TD)); Chris@10: ST(&(x[WS(vs, 1) + WS(rs, 3)]), TF, ms, &(x[WS(vs, 1) + WS(rs, 1)])); Chris@10: } Chris@10: { Chris@10: V Tb, Tm, Tx, TI; Chris@10: Tb = BYTW(&(W[TWVL * 2]), VSUB(T9, Ta)); Chris@10: ST(&(x[WS(vs, 2)]), Tb, ms, &(x[WS(vs, 2)])); Chris@10: Tm = BYTW(&(W[TWVL * 2]), VSUB(Tk, Tl)); Chris@10: ST(&(x[WS(vs, 2) + WS(rs, 1)]), Tm, ms, &(x[WS(vs, 2) + WS(rs, 1)])); Chris@10: Tx = BYTW(&(W[TWVL * 2]), VSUB(Tv, Tw)); Chris@10: ST(&(x[WS(vs, 2) + WS(rs, 2)]), Tx, ms, &(x[WS(vs, 2)])); Chris@10: TI = BYTW(&(W[TWVL * 2]), VSUB(TG, TH)); Chris@10: ST(&(x[WS(vs, 2) + WS(rs, 3)]), TI, ms, &(x[WS(vs, 2) + WS(rs, 1)])); Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: VTW(0, 1), Chris@10: VTW(0, 2), Chris@10: VTW(0, 3), Chris@10: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 4, XSIMD_STRING("q1bv_4"), twinstr, &GENUS, {44, 24, 0, 0}, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_q1bv_4) (planner *p) { Chris@10: X(kdft_difsq_register) (p, q1bv_4, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */