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 07:42:29 EST 2012 */ Chris@10: Chris@10: #include "codelet-rdft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2cdft_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -trivial-stores -variables 32 -no-generate-bytw -n 8 -dit -name hc2cfdftv_8 -include hc2cfv.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 41 FP additions, 40 FP multiplications, Chris@10: * (or, 23 additions, 22 multiplications, 18 fused multiply/add), Chris@10: * 52 stack variables, 2 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "hc2cfv.h" Chris@10: Chris@10: static void hc2cfdftv_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 14)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 14), MAKE_VOLATILE_STRIDE(32, rs)) { Chris@10: V T3, Tc, Tl, Ts, Tf, Tg, Te, Tp, T7, Ta, T1, T2, Tb, Tj, Tk; Chris@10: V Ti, Tr, T5, T6, T4, T9, Th, Tq, TC, T8, Td, TF, Tm, TG, TD; Chris@10: V Tt, Tu, Tn, TH, TL, TE, TK, Tz, Tv, Ty, To, TJ, TI, TN, TM; Chris@10: V TB, TA, Tx, Tw; Chris@10: T1 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@10: T2 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@10: Tb = LDW(&(W[0])); Chris@10: Tj = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Tk = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Ti = LDW(&(W[TWVL * 12])); Chris@10: Tr = LDW(&(W[TWVL * 10])); Chris@10: T5 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@10: T6 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@10: T3 = VFMACONJ(T2, T1); Chris@10: Tc = VZMULIJ(Tb, VFNMSCONJ(T2, T1)); Chris@10: T4 = LDW(&(W[TWVL * 6])); Chris@10: T9 = LDW(&(W[TWVL * 8])); Chris@10: Tl = VZMULIJ(Ti, VFNMSCONJ(Tk, Tj)); Chris@10: Ts = VZMULJ(Tr, VFMACONJ(Tk, Tj)); Chris@10: Tf = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Tg = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Te = LDW(&(W[TWVL * 4])); Chris@10: Tp = LDW(&(W[TWVL * 2])); Chris@10: T7 = VZMULJ(T4, VFMACONJ(T6, T5)); Chris@10: Ta = VZMULIJ(T9, VFNMSCONJ(T6, T5)); Chris@10: Th = VZMULIJ(Te, VFNMSCONJ(Tg, Tf)); Chris@10: Tq = VZMULJ(Tp, VFMACONJ(Tg, Tf)); Chris@10: TC = VADD(T3, T7); Chris@10: T8 = VSUB(T3, T7); Chris@10: Td = VSUB(Ta, Tc); Chris@10: TF = VADD(Tc, Ta); Chris@10: Tm = VSUB(Th, Tl); Chris@10: TG = VADD(Th, Tl); Chris@10: TD = VADD(Tq, Ts); Chris@10: Tt = VSUB(Tq, Ts); Chris@10: Tu = VSUB(Tm, Td); Chris@10: Tn = VADD(Td, Tm); Chris@10: TH = VSUB(TF, TG); Chris@10: TL = VADD(TF, TG); Chris@10: TE = VSUB(TC, TD); Chris@10: TK = VADD(TC, TD); Chris@10: Tz = VFMA(LDK(KP707106781), Tu, Tt); Chris@10: Tv = VFNMS(LDK(KP707106781), Tu, Tt); Chris@10: Ty = VFNMS(LDK(KP707106781), Tn, T8); Chris@10: To = VFMA(LDK(KP707106781), Tn, T8); Chris@10: TJ = VCONJ(VMUL(LDK(KP500000000), VFNMSI(TH, TE))); Chris@10: TI = VMUL(LDK(KP500000000), VFMAI(TH, TE)); Chris@10: TN = VCONJ(VMUL(LDK(KP500000000), VADD(TL, TK))); Chris@10: TM = VMUL(LDK(KP500000000), VSUB(TK, TL)); Chris@10: TB = VMUL(LDK(KP500000000), VFMAI(Tz, Ty)); Chris@10: TA = VCONJ(VMUL(LDK(KP500000000), VFNMSI(Tz, Ty))); Chris@10: Tx = VCONJ(VMUL(LDK(KP500000000), VFMAI(Tv, To))); Chris@10: Tw = VMUL(LDK(KP500000000), VFNMSI(Tv, To)); Chris@10: ST(&(Rm[WS(rs, 1)]), TJ, -ms, &(Rm[WS(rs, 1)])); Chris@10: ST(&(Rp[WS(rs, 2)]), TI, ms, &(Rp[0])); Chris@10: ST(&(Rm[WS(rs, 3)]), TN, -ms, &(Rm[WS(rs, 1)])); Chris@10: ST(&(Rp[0]), TM, ms, &(Rp[0])); Chris@10: ST(&(Rp[WS(rs, 3)]), TB, ms, &(Rp[WS(rs, 1)])); Chris@10: ST(&(Rm[WS(rs, 2)]), TA, -ms, &(Rm[0])); Chris@10: ST(&(Rm[0]), Tx, -ms, &(Rm[0])); Chris@10: ST(&(Rp[WS(rs, 1)]), Tw, ms, &(Rp[WS(rs, 1)])); Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: VTW(1, 1), Chris@10: VTW(1, 2), Chris@10: VTW(1, 3), Chris@10: VTW(1, 4), Chris@10: VTW(1, 5), Chris@10: VTW(1, 6), Chris@10: VTW(1, 7), Chris@10: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 8, XSIMD_STRING("hc2cfdftv_8"), twinstr, &GENUS, {23, 22, 18, 0} }; Chris@10: Chris@10: void XSIMD(codelet_hc2cfdftv_8) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdftv_8, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2cdft_c.native -simd -compact -variables 4 -pipeline-latency 8 -trivial-stores -variables 32 -no-generate-bytw -n 8 -dit -name hc2cfdftv_8 -include hc2cfv.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 41 FP additions, 23 FP multiplications, Chris@10: * (or, 41 additions, 23 multiplications, 0 fused multiply/add), Chris@10: * 57 stack variables, 3 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "hc2cfv.h" Chris@10: Chris@10: static void hc2cfdftv_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@10: DVK(KP353553390, +0.353553390593273762200422181052424519642417969); Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 14)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 14), MAKE_VOLATILE_STRIDE(32, rs)) { Chris@10: V Ta, TE, Tr, TF, Tl, TK, Tw, TG, T1, T6, T3, T8, T2, T7, T4; Chris@10: V T9, T5, To, Tq, Tn, Tp, Tc, Th, Te, Tj, Td, Ti, Tf, Tk, Tb; Chris@10: V Tg, Tt, Tv, Ts, Tu, Ty, Tz, Tm, Tx, TC, TD, TA, TB, TI, TO; Chris@10: V TL, TP, TH, TJ, TM, TR, TN, TQ; Chris@10: T1 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@10: T6 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@10: T2 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@10: T3 = VCONJ(T2); Chris@10: T7 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@10: T8 = VCONJ(T7); Chris@10: T4 = VADD(T1, T3); Chris@10: T5 = LDW(&(W[TWVL * 6])); Chris@10: T9 = VZMULJ(T5, VADD(T6, T8)); Chris@10: Ta = VADD(T4, T9); Chris@10: TE = VMUL(LDK(KP500000000), VSUB(T4, T9)); Chris@10: Tn = LDW(&(W[0])); Chris@10: To = VZMULIJ(Tn, VSUB(T3, T1)); Chris@10: Tp = LDW(&(W[TWVL * 8])); Chris@10: Tq = VZMULIJ(Tp, VSUB(T8, T6)); Chris@10: Tr = VADD(To, Tq); Chris@10: TF = VSUB(To, Tq); Chris@10: Tc = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Th = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Td = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Te = VCONJ(Td); Chris@10: Ti = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Tj = VCONJ(Ti); Chris@10: Tb = LDW(&(W[TWVL * 2])); Chris@10: Tf = VZMULJ(Tb, VADD(Tc, Te)); Chris@10: Tg = LDW(&(W[TWVL * 10])); Chris@10: Tk = VZMULJ(Tg, VADD(Th, Tj)); Chris@10: Tl = VADD(Tf, Tk); Chris@10: TK = VSUB(Tf, Tk); Chris@10: Ts = LDW(&(W[TWVL * 4])); Chris@10: Tt = VZMULIJ(Ts, VSUB(Te, Tc)); Chris@10: Tu = LDW(&(W[TWVL * 12])); Chris@10: Tv = VZMULIJ(Tu, VSUB(Tj, Th)); Chris@10: Tw = VADD(Tt, Tv); Chris@10: TG = VSUB(Tv, Tt); Chris@10: Tm = VADD(Ta, Tl); Chris@10: Tx = VADD(Tr, Tw); Chris@10: Ty = VCONJ(VMUL(LDK(KP500000000), VSUB(Tm, Tx))); Chris@10: Tz = VMUL(LDK(KP500000000), VADD(Tm, Tx)); Chris@10: ST(&(Rm[WS(rs, 3)]), Ty, -ms, &(Rm[WS(rs, 1)])); Chris@10: ST(&(Rp[0]), Tz, ms, &(Rp[0])); Chris@10: TA = VSUB(Ta, Tl); Chris@10: TB = VBYI(VSUB(Tw, Tr)); Chris@10: TC = VCONJ(VMUL(LDK(KP500000000), VSUB(TA, TB))); Chris@10: TD = VMUL(LDK(KP500000000), VADD(TA, TB)); Chris@10: ST(&(Rm[WS(rs, 1)]), TC, -ms, &(Rm[WS(rs, 1)])); Chris@10: ST(&(Rp[WS(rs, 2)]), TD, ms, &(Rp[0])); Chris@10: TH = VMUL(LDK(KP353553390), VADD(TF, TG)); Chris@10: TI = VADD(TE, TH); Chris@10: TO = VSUB(TE, TH); Chris@10: TJ = VMUL(LDK(KP707106781), VSUB(TG, TF)); Chris@10: TL = VMUL(LDK(KP500000000), VBYI(VSUB(TJ, TK))); Chris@10: TP = VMUL(LDK(KP500000000), VBYI(VADD(TK, TJ))); Chris@10: TM = VCONJ(VSUB(TI, TL)); Chris@10: ST(&(Rm[0]), TM, -ms, &(Rm[0])); Chris@10: TR = VADD(TO, TP); Chris@10: ST(&(Rp[WS(rs, 3)]), TR, ms, &(Rp[WS(rs, 1)])); Chris@10: TN = VADD(TI, TL); Chris@10: ST(&(Rp[WS(rs, 1)]), TN, ms, &(Rp[WS(rs, 1)])); Chris@10: TQ = VCONJ(VSUB(TO, TP)); Chris@10: ST(&(Rm[WS(rs, 2)]), TQ, -ms, &(Rm[0])); Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: VTW(1, 1), Chris@10: VTW(1, 2), Chris@10: VTW(1, 3), Chris@10: VTW(1, 4), Chris@10: VTW(1, 5), Chris@10: VTW(1, 6), Chris@10: VTW(1, 7), Chris@10: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 8, XSIMD_STRING("hc2cfdftv_8"), twinstr, &GENUS, {41, 23, 0, 0} }; Chris@10: Chris@10: void XSIMD(codelet_hc2cfdftv_8) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdftv_8, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */