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 08:05:58 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_twiddle_c.native -fma -simd -compact -variables 4 -pipeline-latency 8 -n 12 -name t1bv_12 -include dft/simd/t1b.h -sign 1 */ cannam@167: cannam@167: /* cannam@167: * This function contains 59 FP additions, 42 FP multiplications, cannam@167: * (or, 41 additions, 24 multiplications, 18 fused multiply/add), cannam@167: * 28 stack variables, 2 constants, and 24 memory accesses cannam@167: */ cannam@167: #include "dft/simd/t1b.h" cannam@167: cannam@167: static void t1bv_12(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@167: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@167: { cannam@167: INT m; cannam@167: R *x; cannam@167: x = ii; cannam@167: for (m = mb, W = W + (mb * ((TWVL / VL) * 22)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 22), MAKE_VOLATILE_STRIDE(12, rs)) { cannam@167: V T1, TK, T6, TA, Tq, TI, Tv, TE, T9, TL, Te, TB, Ti, TH, Tn; cannam@167: V TD; cannam@167: { cannam@167: V T5, T3, T4, T2; cannam@167: T1 = LD(&(x[0]), ms, &(x[0])); cannam@167: T4 = LD(&(x[WS(rs, 8)]), ms, &(x[0])); cannam@167: T5 = BYTW(&(W[TWVL * 14]), T4); cannam@167: T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@167: T3 = BYTW(&(W[TWVL * 6]), T2); cannam@167: TK = VSUB(T3, T5); cannam@167: T6 = VADD(T3, T5); cannam@167: TA = VFNMS(LDK(KP500000000), T6, T1); cannam@167: } cannam@167: { cannam@167: V Tu, Ts, Tp, Tt, Tr; cannam@167: Tp = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); cannam@167: Tq = BYTW(&(W[TWVL * 16]), Tp); cannam@167: Tt = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@167: Tu = BYTW(&(W[TWVL * 8]), Tt); cannam@167: Tr = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@167: Ts = BYTW(&(W[0]), Tr); cannam@167: TI = VSUB(Tu, Ts); cannam@167: Tv = VADD(Ts, Tu); cannam@167: TE = VFNMS(LDK(KP500000000), Tv, Tq); cannam@167: } cannam@167: { cannam@167: V Td, Tb, T8, Tc, Ta; cannam@167: T8 = LD(&(x[WS(rs, 6)]), ms, &(x[0])); cannam@167: T9 = BYTW(&(W[TWVL * 10]), T8); cannam@167: Tc = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@167: Td = BYTW(&(W[TWVL * 2]), Tc); cannam@167: Ta = LD(&(x[WS(rs, 10)]), ms, &(x[0])); cannam@167: Tb = BYTW(&(W[TWVL * 18]), Ta); cannam@167: TL = VSUB(Tb, Td); cannam@167: Te = VADD(Tb, Td); cannam@167: TB = VFNMS(LDK(KP500000000), Te, T9); cannam@167: } cannam@167: { cannam@167: V Tm, Tk, Th, Tl, Tj; cannam@167: Th = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@167: Ti = BYTW(&(W[TWVL * 4]), Th); cannam@167: Tl = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)])); cannam@167: Tm = BYTW(&(W[TWVL * 20]), Tl); cannam@167: Tj = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); cannam@167: Tk = BYTW(&(W[TWVL * 12]), Tj); cannam@167: TH = VSUB(Tk, Tm); cannam@167: Tn = VADD(Tk, Tm); cannam@167: TD = VFNMS(LDK(KP500000000), Tn, Ti); cannam@167: } cannam@167: { cannam@167: V Tg, Ty, Tx, Tz; cannam@167: { cannam@167: V T7, Tf, To, Tw; cannam@167: T7 = VADD(T1, T6); cannam@167: Tf = VADD(T9, Te); cannam@167: Tg = VSUB(T7, Tf); cannam@167: Ty = VADD(T7, Tf); cannam@167: To = VADD(Ti, Tn); cannam@167: Tw = VADD(Tq, Tv); cannam@167: Tx = VSUB(To, Tw); cannam@167: Tz = VADD(To, Tw); cannam@167: } cannam@167: ST(&(x[WS(rs, 3)]), VFNMSI(Tx, Tg), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[0]), VADD(Ty, Tz), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 9)]), VFMAI(Tx, Tg), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[WS(rs, 6)]), VSUB(Ty, Tz), ms, &(x[0])); cannam@167: } cannam@167: { cannam@167: V TS, TW, TV, TX; cannam@167: { cannam@167: V TQ, TR, TT, TU; cannam@167: TQ = VSUB(TA, TB); cannam@167: TR = VADD(TH, TI); cannam@167: TS = VFNMS(LDK(KP866025403), TR, TQ); cannam@167: TW = VFMA(LDK(KP866025403), TR, TQ); cannam@167: TT = VSUB(TD, TE); cannam@167: TU = VSUB(TK, TL); cannam@167: TV = VFMA(LDK(KP866025403), TU, TT); cannam@167: TX = VFNMS(LDK(KP866025403), TU, TT); cannam@167: } cannam@167: ST(&(x[WS(rs, 1)]), VFMAI(TV, TS), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[WS(rs, 7)]), VFNMSI(TX, TW), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[WS(rs, 11)]), VFNMSI(TV, TS), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[WS(rs, 5)]), VFMAI(TX, TW), ms, &(x[WS(rs, 1)])); cannam@167: } cannam@167: { cannam@167: V TG, TO, TN, TP; cannam@167: { cannam@167: V TC, TF, TJ, TM; cannam@167: TC = VADD(TA, TB); cannam@167: TF = VADD(TD, TE); cannam@167: TG = VSUB(TC, TF); cannam@167: TO = VADD(TC, TF); cannam@167: TJ = VSUB(TH, TI); cannam@167: TM = VADD(TK, TL); cannam@167: TN = VMUL(LDK(KP866025403), VSUB(TJ, TM)); cannam@167: TP = VMUL(LDK(KP866025403), VADD(TM, TJ)); cannam@167: } cannam@167: ST(&(x[WS(rs, 10)]), VFNMSI(TN, TG), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 4)]), VFMAI(TP, TO), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 2)]), VFMAI(TN, TG), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 8)]), VFNMSI(TP, TO), ms, &(x[0])); 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: VTW(0, 4), cannam@167: VTW(0, 5), cannam@167: VTW(0, 6), cannam@167: VTW(0, 7), cannam@167: VTW(0, 8), cannam@167: VTW(0, 9), cannam@167: VTW(0, 10), cannam@167: VTW(0, 11), cannam@167: {TW_NEXT, VL, 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 12, XSIMD_STRING("t1bv_12"), twinstr, &GENUS, {41, 24, 18, 0}, 0, 0, 0 }; cannam@167: cannam@167: void XSIMD(codelet_t1bv_12) (planner *p) { cannam@167: X(kdft_dit_register) (p, t1bv_12, &desc); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 12 -name t1bv_12 -include dft/simd/t1b.h -sign 1 */ cannam@167: cannam@167: /* cannam@167: * This function contains 59 FP additions, 30 FP multiplications, cannam@167: * (or, 55 additions, 26 multiplications, 4 fused multiply/add), cannam@167: * 28 stack variables, 2 constants, and 24 memory accesses cannam@167: */ cannam@167: #include "dft/simd/t1b.h" cannam@167: cannam@167: static void t1bv_12(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@167: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@167: { cannam@167: INT m; cannam@167: R *x; cannam@167: x = ii; cannam@167: for (m = mb, W = W + (mb * ((TWVL / VL) * 22)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 22), MAKE_VOLATILE_STRIDE(12, rs)) { cannam@167: V T1, Tt, T6, T7, TB, Tq, TC, TD, T9, Tu, Te, Tf, Tx, Tl, Ty; cannam@167: V Tz; cannam@167: { cannam@167: V T5, T3, T4, T2; cannam@167: T1 = LD(&(x[0]), ms, &(x[0])); cannam@167: T4 = LD(&(x[WS(rs, 8)]), ms, &(x[0])); cannam@167: T5 = BYTW(&(W[TWVL * 14]), T4); cannam@167: T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@167: T3 = BYTW(&(W[TWVL * 6]), T2); cannam@167: Tt = VSUB(T3, T5); cannam@167: T6 = VADD(T3, T5); cannam@167: T7 = VFNMS(LDK(KP500000000), T6, T1); cannam@167: } cannam@167: { cannam@167: V Tn, Tp, Tm, TA, To; cannam@167: Tm = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@167: Tn = BYTW(&(W[0]), Tm); cannam@167: TA = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); cannam@167: TB = BYTW(&(W[TWVL * 16]), TA); cannam@167: To = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@167: Tp = BYTW(&(W[TWVL * 8]), To); cannam@167: Tq = VSUB(Tn, Tp); cannam@167: TC = VADD(Tn, Tp); cannam@167: TD = VFNMS(LDK(KP500000000), TC, TB); cannam@167: } cannam@167: { cannam@167: V Td, Tb, T8, Tc, Ta; cannam@167: T8 = LD(&(x[WS(rs, 6)]), ms, &(x[0])); cannam@167: T9 = BYTW(&(W[TWVL * 10]), T8); cannam@167: Tc = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@167: Td = BYTW(&(W[TWVL * 2]), Tc); cannam@167: Ta = LD(&(x[WS(rs, 10)]), ms, &(x[0])); cannam@167: Tb = BYTW(&(W[TWVL * 18]), Ta); cannam@167: Tu = VSUB(Tb, Td); cannam@167: Te = VADD(Tb, Td); cannam@167: Tf = VFNMS(LDK(KP500000000), Te, T9); cannam@167: } cannam@167: { cannam@167: V Ti, Tk, Th, Tw, Tj; cannam@167: Th = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); cannam@167: Ti = BYTW(&(W[TWVL * 12]), Th); cannam@167: Tw = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@167: Tx = BYTW(&(W[TWVL * 4]), Tw); cannam@167: Tj = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)])); cannam@167: Tk = BYTW(&(W[TWVL * 20]), Tj); cannam@167: Tl = VSUB(Ti, Tk); cannam@167: Ty = VADD(Ti, Tk); cannam@167: Tz = VFNMS(LDK(KP500000000), Ty, Tx); cannam@167: } cannam@167: { cannam@167: V Ts, TG, TF, TH; cannam@167: { cannam@167: V Tg, Tr, Tv, TE; cannam@167: Tg = VSUB(T7, Tf); cannam@167: Tr = VMUL(LDK(KP866025403), VSUB(Tl, Tq)); cannam@167: Ts = VSUB(Tg, Tr); cannam@167: TG = VADD(Tg, Tr); cannam@167: Tv = VMUL(LDK(KP866025403), VSUB(Tt, Tu)); cannam@167: TE = VSUB(Tz, TD); cannam@167: TF = VBYI(VADD(Tv, TE)); cannam@167: TH = VBYI(VSUB(TE, Tv)); cannam@167: } cannam@167: ST(&(x[WS(rs, 11)]), VSUB(Ts, TF), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[WS(rs, 5)]), VADD(TG, TH), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[WS(rs, 1)]), VADD(Ts, TF), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[WS(rs, 7)]), VSUB(TG, TH), ms, &(x[WS(rs, 1)])); cannam@167: } cannam@167: { cannam@167: V TS, TW, TV, TX; cannam@167: { cannam@167: V TQ, TR, TT, TU; cannam@167: TQ = VADD(T1, T6); cannam@167: TR = VADD(T9, Te); cannam@167: TS = VSUB(TQ, TR); cannam@167: TW = VADD(TQ, TR); cannam@167: TT = VADD(Tx, Ty); cannam@167: TU = VADD(TB, TC); cannam@167: TV = VBYI(VSUB(TT, TU)); cannam@167: TX = VADD(TT, TU); cannam@167: } cannam@167: ST(&(x[WS(rs, 3)]), VSUB(TS, TV), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[0]), VADD(TW, TX), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 9)]), VADD(TS, TV), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[WS(rs, 6)]), VSUB(TW, TX), ms, &(x[0])); cannam@167: } cannam@167: { cannam@167: V TK, TO, TN, TP; cannam@167: { cannam@167: V TI, TJ, TL, TM; cannam@167: TI = VADD(Tl, Tq); cannam@167: TJ = VADD(Tt, Tu); cannam@167: TK = VBYI(VMUL(LDK(KP866025403), VSUB(TI, TJ))); cannam@167: TO = VBYI(VMUL(LDK(KP866025403), VADD(TJ, TI))); cannam@167: TL = VADD(T7, Tf); cannam@167: TM = VADD(Tz, TD); cannam@167: TN = VSUB(TL, TM); cannam@167: TP = VADD(TL, TM); cannam@167: } cannam@167: ST(&(x[WS(rs, 2)]), VADD(TK, TN), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 8)]), VSUB(TP, TO), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 10)]), VSUB(TN, TK), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 4)]), VADD(TO, TP), ms, &(x[0])); 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: VTW(0, 4), cannam@167: VTW(0, 5), cannam@167: VTW(0, 6), cannam@167: VTW(0, 7), cannam@167: VTW(0, 8), cannam@167: VTW(0, 9), cannam@167: VTW(0, 10), cannam@167: VTW(0, 11), cannam@167: {TW_NEXT, VL, 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 12, XSIMD_STRING("t1bv_12"), twinstr, &GENUS, {55, 26, 4, 0}, 0, 0, 0 }; cannam@167: cannam@167: void XSIMD(codelet_t1bv_12) (planner *p) { cannam@167: X(kdft_dit_register) (p, t1bv_12, &desc); cannam@167: } cannam@167: #endif