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 6 -dit -name hc2cfdftv_6 -include hc2cfv.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 29 FP additions, 30 FP multiplications, Chris@10: * (or, 17 additions, 18 multiplications, 12 fused multiply/add), Chris@10: * 38 stack variables, 2 constants, and 12 memory accesses Chris@10: */ Chris@10: #include "hc2cfv.h" Chris@10: Chris@10: static void hc2cfdftv_6(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(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 10)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 10), MAKE_VOLATILE_STRIDE(24, rs)) { Chris@10: V T5, T6, T3, Tj, T4, T9, Te, Th, T1, T2, Ti, Tc, Td, Tb, Tg; Chris@10: V T7, Ta, Tt, Tk, Tr, T8, Ts, Tf, Tx, Tu, To, Tl, Tw, Tv, Tn; Chris@10: V Tm, Tz, Ty, Tp, Tq; Chris@10: T1 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@10: T2 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@10: Ti = LDW(&(W[0])); Chris@10: Tc = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@10: Td = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@10: Tb = LDW(&(W[TWVL * 8])); Chris@10: Tg = LDW(&(W[TWVL * 6])); Chris@10: T5 = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@10: T6 = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: T3 = VFMACONJ(T2, T1); Chris@10: Tj = VZMULIJ(Ti, VFNMSCONJ(T2, T1)); Chris@10: T4 = LDW(&(W[TWVL * 4])); Chris@10: T9 = LDW(&(W[TWVL * 2])); Chris@10: Te = VZMULIJ(Tb, VFNMSCONJ(Td, Tc)); Chris@10: Th = VZMULJ(Tg, VFMACONJ(Td, Tc)); Chris@10: T7 = VZMULIJ(T4, VFNMSCONJ(T6, T5)); Chris@10: Ta = VZMULJ(T9, VFMACONJ(T6, T5)); Chris@10: Tt = VADD(Tj, Th); Chris@10: Tk = VSUB(Th, Tj); Chris@10: Tr = VADD(T3, T7); Chris@10: T8 = VSUB(T3, T7); Chris@10: Ts = VADD(Ta, Te); Chris@10: Tf = VSUB(Ta, Te); Chris@10: Tx = VMUL(LDK(KP866025403), VSUB(Tt, Ts)); Chris@10: Tu = VADD(Ts, Tt); Chris@10: To = VMUL(LDK(KP866025403), VSUB(Tk, Tf)); Chris@10: Tl = VADD(Tf, Tk); Chris@10: Tw = VFNMS(LDK(KP500000000), Tu, Tr); Chris@10: Tv = VCONJ(VMUL(LDK(KP500000000), VADD(Tr, Tu))); Chris@10: Tn = VFNMS(LDK(KP500000000), Tl, T8); Chris@10: Tm = VMUL(LDK(KP500000000), VADD(T8, Tl)); Chris@10: Tz = VMUL(LDK(KP500000000), VFMAI(Tx, Tw)); Chris@10: Ty = VCONJ(VMUL(LDK(KP500000000), VFNMSI(Tx, Tw))); Chris@10: ST(&(Rm[WS(rs, 2)]), Tv, -ms, &(Rm[0])); Chris@10: Tp = VMUL(LDK(KP500000000), VFNMSI(To, Tn)); Chris@10: Tq = VCONJ(VMUL(LDK(KP500000000), VFMAI(To, Tn))); Chris@10: ST(&(Rp[0]), Tm, ms, &(Rp[0])); Chris@10: ST(&(Rp[WS(rs, 1)]), Tz, ms, &(Rp[WS(rs, 1)])); Chris@10: ST(&(Rm[0]), Ty, -ms, &(Rm[0])); Chris@10: ST(&(Rm[WS(rs, 1)]), Tq, -ms, &(Rm[WS(rs, 1)])); Chris@10: ST(&(Rp[WS(rs, 2)]), Tp, 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: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 6, XSIMD_STRING("hc2cfdftv_6"), twinstr, &GENUS, {17, 18, 12, 0} }; Chris@10: Chris@10: void XSIMD(codelet_hc2cfdftv_6) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdftv_6, &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 6 -dit -name hc2cfdftv_6 -include hc2cfv.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 29 FP additions, 20 FP multiplications, Chris@10: * (or, 27 additions, 18 multiplications, 2 fused multiply/add), Chris@10: * 42 stack variables, 3 constants, and 12 memory accesses Chris@10: */ Chris@10: #include "hc2cfv.h" Chris@10: Chris@10: static void hc2cfdftv_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 10)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 10), MAKE_VOLATILE_STRIDE(24, rs)) { Chris@10: V Ta, Tu, Tn, Tw, Ti, Tv, T1, T8, Tg, Tf, T7, T3, Te, T6, T2; Chris@10: V T4, T9, T5, Tk, Tm, Tj, Tl, Tc, Th, Tb, Td, Tr, Tp, Tq, To; Chris@10: V Tt, Ts, TA, Ty, Tz, Tx, TC, TB; Chris@10: T1 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@10: T8 = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@10: Tg = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@10: Te = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@10: Tf = VCONJ(Te); Chris@10: T6 = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: T7 = VCONJ(T6); Chris@10: T2 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@10: T3 = VCONJ(T2); Chris@10: T4 = VADD(T1, T3); Chris@10: T5 = LDW(&(W[TWVL * 4])); Chris@10: T9 = VZMULIJ(T5, VSUB(T7, T8)); Chris@10: Ta = VADD(T4, T9); Chris@10: Tu = VSUB(T4, T9); Chris@10: Tj = LDW(&(W[0])); Chris@10: Tk = VZMULIJ(Tj, VSUB(T3, T1)); Chris@10: Tl = LDW(&(W[TWVL * 6])); Chris@10: Tm = VZMULJ(Tl, VADD(Tf, Tg)); Chris@10: Tn = VADD(Tk, Tm); Chris@10: Tw = VSUB(Tm, Tk); Chris@10: Tb = LDW(&(W[TWVL * 2])); Chris@10: Tc = VZMULJ(Tb, VADD(T7, T8)); Chris@10: Td = LDW(&(W[TWVL * 8])); Chris@10: Th = VZMULIJ(Td, VSUB(Tf, Tg)); Chris@10: Ti = VADD(Tc, Th); Chris@10: Tv = VSUB(Tc, Th); Chris@10: Tr = VMUL(LDK(KP500000000), VBYI(VMUL(LDK(KP866025403), VSUB(Tn, Ti)))); Chris@10: To = VADD(Ti, Tn); Chris@10: Tp = VMUL(LDK(KP500000000), VADD(Ta, To)); Chris@10: Tq = VFNMS(LDK(KP250000000), To, VMUL(LDK(KP500000000), Ta)); Chris@10: ST(&(Rp[0]), Tp, ms, &(Rp[0])); Chris@10: Tt = VCONJ(VADD(Tq, Tr)); Chris@10: ST(&(Rm[WS(rs, 1)]), Tt, -ms, &(Rm[WS(rs, 1)])); Chris@10: Ts = VSUB(Tq, Tr); Chris@10: ST(&(Rp[WS(rs, 2)]), Ts, ms, &(Rp[0])); Chris@10: TA = VMUL(LDK(KP500000000), VBYI(VMUL(LDK(KP866025403), VSUB(Tw, Tv)))); Chris@10: Tx = VADD(Tv, Tw); Chris@10: Ty = VCONJ(VMUL(LDK(KP500000000), VADD(Tu, Tx))); Chris@10: Tz = VFNMS(LDK(KP250000000), Tx, VMUL(LDK(KP500000000), Tu)); Chris@10: ST(&(Rm[WS(rs, 2)]), Ty, -ms, &(Rm[0])); Chris@10: TC = VADD(Tz, TA); Chris@10: ST(&(Rp[WS(rs, 1)]), TC, ms, &(Rp[WS(rs, 1)])); Chris@10: TB = VCONJ(VSUB(Tz, TA)); Chris@10: ST(&(Rm[0]), TB, -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: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 6, XSIMD_STRING("hc2cfdftv_6"), twinstr, &GENUS, {27, 18, 2, 0} }; Chris@10: Chris@10: void XSIMD(codelet_hc2cfdftv_6) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdftv_6, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */