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 -dif -sign 1 -name hc2cbdftv_8 -include hc2cbv.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 41 FP additions, 32 FP multiplications, Chris@10: * (or, 23 additions, 14 multiplications, 18 fused multiply/add), Chris@10: * 51 stack variables, 1 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "hc2cbv.h" Chris@10: Chris@10: static void hc2cbdftv_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: { 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 TJ, T4, Tf, TB, TD, TE, Tm, T1, Tj, TF, Tp, Tb, Tg, Tt, Tx; Chris@10: V T2, T3, Td, Te, T5, T6, T8, T9, Tn, T7, To, Ta, Tk, Tl, TG; Chris@10: V TL, Tq, Tc, Tu, Th, Tv, Ty, Tw, TC, Ti, TK, TA, Tz, TI, TH; Chris@10: V Ts, Tr, TN, TM; Chris@10: T2 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@10: T3 = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Td = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@10: Te = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: T5 = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@10: T6 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@10: T8 = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); Chris@10: T9 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@10: TJ = LDW(&(W[0])); Chris@10: Tk = VFMACONJ(T3, T2); Chris@10: T4 = VFNMSCONJ(T3, T2); Chris@10: Tl = VFMACONJ(Te, Td); Chris@10: Tf = VFNMSCONJ(Te, Td); Chris@10: Tn = VFMACONJ(T6, T5); Chris@10: T7 = VFNMSCONJ(T6, T5); Chris@10: To = VFMACONJ(T9, T8); Chris@10: Ta = VFMSCONJ(T9, T8); Chris@10: TB = LDW(&(W[TWVL * 8])); Chris@10: TD = LDW(&(W[TWVL * 6])); Chris@10: TE = VADD(Tk, Tl); Chris@10: Tm = VSUB(Tk, Tl); Chris@10: T1 = LDW(&(W[TWVL * 12])); Chris@10: Tj = LDW(&(W[TWVL * 10])); Chris@10: TF = VADD(Tn, To); Chris@10: Tp = VSUB(Tn, To); Chris@10: Tb = VADD(T7, Ta); Chris@10: Tg = VSUB(T7, Ta); Chris@10: Tt = LDW(&(W[TWVL * 4])); Chris@10: Tx = LDW(&(W[TWVL * 2])); Chris@10: TG = VZMUL(TD, VSUB(TE, TF)); Chris@10: TL = VADD(TE, TF); Chris@10: Tq = VZMUL(Tj, VFNMSI(Tp, Tm)); Chris@10: Tc = VFMA(LDK(KP707106781), Tb, T4); Chris@10: Tu = VFNMS(LDK(KP707106781), Tb, T4); Chris@10: Th = VFMA(LDK(KP707106781), Tg, Tf); Chris@10: Tv = VFNMS(LDK(KP707106781), Tg, Tf); Chris@10: Ty = VZMUL(Tx, VFMAI(Tp, Tm)); Chris@10: Tw = VZMULI(Tt, VFNMSI(Tv, Tu)); Chris@10: TC = VZMULI(TB, VFMAI(Tv, Tu)); Chris@10: Ti = VZMULI(T1, VFNMSI(Th, Tc)); Chris@10: TK = VZMULI(TJ, VFMAI(Th, Tc)); Chris@10: TA = VCONJ(VSUB(Ty, Tw)); Chris@10: Tz = VADD(Tw, Ty); Chris@10: TI = VCONJ(VSUB(TG, TC)); Chris@10: TH = VADD(TC, TG); Chris@10: Ts = VCONJ(VSUB(Tq, Ti)); Chris@10: Tr = VADD(Ti, Tq); Chris@10: TN = VCONJ(VSUB(TL, TK)); Chris@10: TM = VADD(TK, TL); Chris@10: ST(&(Rm[WS(rs, 1)]), TA, -ms, &(Rm[WS(rs, 1)])); Chris@10: ST(&(Rp[WS(rs, 1)]), Tz, ms, &(Rp[WS(rs, 1)])); Chris@10: ST(&(Rm[WS(rs, 2)]), TI, -ms, &(Rm[0])); Chris@10: ST(&(Rp[WS(rs, 2)]), TH, ms, &(Rp[0])); Chris@10: ST(&(Rm[WS(rs, 3)]), Ts, -ms, &(Rm[WS(rs, 1)])); Chris@10: ST(&(Rp[WS(rs, 3)]), Tr, ms, &(Rp[WS(rs, 1)])); Chris@10: ST(&(Rm[0]), TN, -ms, &(Rm[0])); Chris@10: ST(&(Rp[0]), TM, ms, &(Rp[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("hc2cbdftv_8"), twinstr, &GENUS, {23, 14, 18, 0} }; Chris@10: Chris@10: void XSIMD(codelet_hc2cbdftv_8) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cbdftv_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 -dif -sign 1 -name hc2cbdftv_8 -include hc2cbv.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 41 FP additions, 16 FP multiplications, Chris@10: * (or, 41 additions, 16 multiplications, 0 fused multiply/add), Chris@10: * 55 stack variables, 1 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "hc2cbv.h" Chris@10: Chris@10: static void hc2cbdftv_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: { 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 T5, Tj, Tq, TI, Te, Tk, Tt, TJ, T2, Tg, T4, Ti, T3, Th, To; Chris@10: V Tp, T6, Tc, T8, Tb, T7, Ta, T9, Td, Tr, Ts, TP, Tu, Tm, TO; Chris@10: V Tn, Tf, Tl, T1, TN, Tv, TR, Tw, TQ, TC, TK, TA, TG, TB, TH; Chris@10: V Ty, Tz, Tx, TF, TD, TM, TE, TL; Chris@10: T2 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@10: Tg = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@10: T3 = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: T4 = VCONJ(T3); Chris@10: Th = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: Ti = VCONJ(Th); Chris@10: T5 = VSUB(T2, T4); Chris@10: Tj = VSUB(Tg, Ti); Chris@10: To = VADD(T2, T4); Chris@10: Tp = VADD(Tg, Ti); Chris@10: Tq = VSUB(To, Tp); Chris@10: TI = VADD(To, Tp); Chris@10: T6 = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Tc = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); Chris@10: T7 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@10: T8 = VCONJ(T7); Chris@10: Ta = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@10: Tb = VCONJ(Ta); Chris@10: T9 = VSUB(T6, T8); Chris@10: Td = VSUB(Tb, Tc); Chris@10: Te = VMUL(LDK(KP707106781), VADD(T9, Td)); Chris@10: Tk = VMUL(LDK(KP707106781), VSUB(T9, Td)); Chris@10: Tr = VADD(T6, T8); Chris@10: Ts = VADD(Tb, Tc); Chris@10: Tt = VBYI(VSUB(Tr, Ts)); Chris@10: TJ = VADD(Tr, Ts); Chris@10: TP = VADD(TI, TJ); Chris@10: Tn = LDW(&(W[TWVL * 10])); Chris@10: Tu = VZMUL(Tn, VSUB(Tq, Tt)); Chris@10: Tf = VADD(T5, Te); Chris@10: Tl = VBYI(VADD(Tj, Tk)); Chris@10: T1 = LDW(&(W[TWVL * 12])); Chris@10: Tm = VZMULI(T1, VSUB(Tf, Tl)); Chris@10: TN = LDW(&(W[0])); Chris@10: TO = VZMULI(TN, VADD(Tl, Tf)); Chris@10: Tv = VADD(Tm, Tu); Chris@10: ST(&(Rp[WS(rs, 3)]), Tv, ms, &(Rp[WS(rs, 1)])); Chris@10: TR = VCONJ(VSUB(TP, TO)); Chris@10: ST(&(Rm[0]), TR, -ms, &(Rm[0])); Chris@10: Tw = VCONJ(VSUB(Tu, Tm)); Chris@10: ST(&(Rm[WS(rs, 3)]), Tw, -ms, &(Rm[WS(rs, 1)])); Chris@10: TQ = VADD(TO, TP); Chris@10: ST(&(Rp[0]), TQ, ms, &(Rp[0])); Chris@10: TB = LDW(&(W[TWVL * 2])); Chris@10: TC = VZMUL(TB, VADD(Tq, Tt)); Chris@10: TH = LDW(&(W[TWVL * 6])); Chris@10: TK = VZMUL(TH, VSUB(TI, TJ)); Chris@10: Ty = VBYI(VSUB(Tk, Tj)); Chris@10: Tz = VSUB(T5, Te); Chris@10: Tx = LDW(&(W[TWVL * 4])); Chris@10: TA = VZMULI(Tx, VADD(Ty, Tz)); Chris@10: TF = LDW(&(W[TWVL * 8])); Chris@10: TG = VZMULI(TF, VSUB(Tz, Ty)); Chris@10: TD = VADD(TA, TC); Chris@10: ST(&(Rp[WS(rs, 1)]), TD, ms, &(Rp[WS(rs, 1)])); Chris@10: TM = VCONJ(VSUB(TK, TG)); Chris@10: ST(&(Rm[WS(rs, 2)]), TM, -ms, &(Rm[0])); Chris@10: TE = VCONJ(VSUB(TC, TA)); Chris@10: ST(&(Rm[WS(rs, 1)]), TE, -ms, &(Rm[WS(rs, 1)])); Chris@10: TL = VADD(TG, TK); Chris@10: ST(&(Rp[WS(rs, 2)]), TL, ms, &(Rp[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("hc2cbdftv_8"), twinstr, &GENUS, {41, 16, 0, 0} }; Chris@10: Chris@10: void XSIMD(codelet_hc2cbdftv_8) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cbdftv_8, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */