cannam@95: /* cannam@95: * Copyright (c) 2003, 2007-11 Matteo Frigo cannam@95: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology cannam@95: * cannam@95: * This program is free software; you can redistribute it and/or modify cannam@95: * it under the terms of the GNU General Public License as published by cannam@95: * the Free Software Foundation; either version 2 of the License, or cannam@95: * (at your option) any later version. cannam@95: * cannam@95: * This program is distributed in the hope that it will be useful, cannam@95: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@95: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@95: * GNU General Public License for more details. cannam@95: * cannam@95: * You should have received a copy of the GNU General Public License cannam@95: * along with this program; if not, write to the Free Software cannam@95: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@95: * cannam@95: */ cannam@95: cannam@95: /* This file was automatically generated --- DO NOT EDIT */ cannam@95: /* Generated on Sun Nov 25 07:42:29 EST 2012 */ cannam@95: cannam@95: #include "codelet-rdft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* 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 */ cannam@95: cannam@95: /* cannam@95: * This function contains 29 FP additions, 30 FP multiplications, cannam@95: * (or, 17 additions, 18 multiplications, 12 fused multiply/add), cannam@95: * 38 stack variables, 2 constants, and 12 memory accesses cannam@95: */ cannam@95: #include "hc2cfv.h" cannam@95: cannam@95: static void hc2cfdftv_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: V T5, T6, T3, Tj, T4, T9, Te, Th, T1, T2, Ti, Tc, Td, Tb, Tg; cannam@95: V T7, Ta, Tt, Tk, Tr, T8, Ts, Tf, Tx, Tu, To, Tl, Tw, Tv, Tn; cannam@95: V Tm, Tz, Ty, Tp, Tq; cannam@95: T1 = LD(&(Rp[0]), ms, &(Rp[0])); cannam@95: T2 = LD(&(Rm[0]), -ms, &(Rm[0])); cannam@95: Ti = LDW(&(W[0])); cannam@95: Tc = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); cannam@95: Td = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); cannam@95: Tb = LDW(&(W[TWVL * 8])); cannam@95: Tg = LDW(&(W[TWVL * 6])); cannam@95: T5 = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); cannam@95: T6 = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: T3 = VFMACONJ(T2, T1); cannam@95: Tj = VZMULIJ(Ti, VFNMSCONJ(T2, T1)); cannam@95: T4 = LDW(&(W[TWVL * 4])); cannam@95: T9 = LDW(&(W[TWVL * 2])); cannam@95: Te = VZMULIJ(Tb, VFNMSCONJ(Td, Tc)); cannam@95: Th = VZMULJ(Tg, VFMACONJ(Td, Tc)); cannam@95: T7 = VZMULIJ(T4, VFNMSCONJ(T6, T5)); cannam@95: Ta = VZMULJ(T9, VFMACONJ(T6, T5)); cannam@95: Tt = VADD(Tj, Th); cannam@95: Tk = VSUB(Th, Tj); cannam@95: Tr = VADD(T3, T7); cannam@95: T8 = VSUB(T3, T7); cannam@95: Ts = VADD(Ta, Te); cannam@95: Tf = VSUB(Ta, Te); cannam@95: Tx = VMUL(LDK(KP866025403), VSUB(Tt, Ts)); cannam@95: Tu = VADD(Ts, Tt); cannam@95: To = VMUL(LDK(KP866025403), VSUB(Tk, Tf)); cannam@95: Tl = VADD(Tf, Tk); cannam@95: Tw = VFNMS(LDK(KP500000000), Tu, Tr); cannam@95: Tv = VCONJ(VMUL(LDK(KP500000000), VADD(Tr, Tu))); cannam@95: Tn = VFNMS(LDK(KP500000000), Tl, T8); cannam@95: Tm = VMUL(LDK(KP500000000), VADD(T8, Tl)); cannam@95: Tz = VMUL(LDK(KP500000000), VFMAI(Tx, Tw)); cannam@95: Ty = VCONJ(VMUL(LDK(KP500000000), VFNMSI(Tx, Tw))); cannam@95: ST(&(Rm[WS(rs, 2)]), Tv, -ms, &(Rm[0])); cannam@95: Tp = VMUL(LDK(KP500000000), VFNMSI(To, Tn)); cannam@95: Tq = VCONJ(VMUL(LDK(KP500000000), VFMAI(To, Tn))); cannam@95: ST(&(Rp[0]), Tm, ms, &(Rp[0])); cannam@95: ST(&(Rp[WS(rs, 1)]), Tz, ms, &(Rp[WS(rs, 1)])); cannam@95: ST(&(Rm[0]), Ty, -ms, &(Rm[0])); cannam@95: ST(&(Rm[WS(rs, 1)]), Tq, -ms, &(Rm[WS(rs, 1)])); cannam@95: ST(&(Rp[WS(rs, 2)]), Tp, ms, &(Rp[0])); cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(1, 1), cannam@95: VTW(1, 2), cannam@95: VTW(1, 3), cannam@95: VTW(1, 4), cannam@95: VTW(1, 5), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 6, XSIMD_STRING("hc2cfdftv_6"), twinstr, &GENUS, {17, 18, 12, 0} }; cannam@95: cannam@95: void XSIMD(codelet_hc2cfdftv_6) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cfdftv_6, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* 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 */ cannam@95: cannam@95: /* cannam@95: * This function contains 29 FP additions, 20 FP multiplications, cannam@95: * (or, 27 additions, 18 multiplications, 2 fused multiply/add), cannam@95: * 42 stack variables, 3 constants, and 12 memory accesses cannam@95: */ cannam@95: #include "hc2cfv.h" cannam@95: cannam@95: static void hc2cfdftv_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: V Ta, Tu, Tn, Tw, Ti, Tv, T1, T8, Tg, Tf, T7, T3, Te, T6, T2; cannam@95: V T4, T9, T5, Tk, Tm, Tj, Tl, Tc, Th, Tb, Td, Tr, Tp, Tq, To; cannam@95: V Tt, Ts, TA, Ty, Tz, Tx, TC, TB; cannam@95: T1 = LD(&(Rp[0]), ms, &(Rp[0])); cannam@95: T8 = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); cannam@95: Tg = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); cannam@95: Te = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); cannam@95: Tf = VCONJ(Te); cannam@95: T6 = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: T7 = VCONJ(T6); cannam@95: T2 = LD(&(Rm[0]), -ms, &(Rm[0])); cannam@95: T3 = VCONJ(T2); cannam@95: T4 = VADD(T1, T3); cannam@95: T5 = LDW(&(W[TWVL * 4])); cannam@95: T9 = VZMULIJ(T5, VSUB(T7, T8)); cannam@95: Ta = VADD(T4, T9); cannam@95: Tu = VSUB(T4, T9); cannam@95: Tj = LDW(&(W[0])); cannam@95: Tk = VZMULIJ(Tj, VSUB(T3, T1)); cannam@95: Tl = LDW(&(W[TWVL * 6])); cannam@95: Tm = VZMULJ(Tl, VADD(Tf, Tg)); cannam@95: Tn = VADD(Tk, Tm); cannam@95: Tw = VSUB(Tm, Tk); cannam@95: Tb = LDW(&(W[TWVL * 2])); cannam@95: Tc = VZMULJ(Tb, VADD(T7, T8)); cannam@95: Td = LDW(&(W[TWVL * 8])); cannam@95: Th = VZMULIJ(Td, VSUB(Tf, Tg)); cannam@95: Ti = VADD(Tc, Th); cannam@95: Tv = VSUB(Tc, Th); cannam@95: Tr = VMUL(LDK(KP500000000), VBYI(VMUL(LDK(KP866025403), VSUB(Tn, Ti)))); cannam@95: To = VADD(Ti, Tn); cannam@95: Tp = VMUL(LDK(KP500000000), VADD(Ta, To)); cannam@95: Tq = VFNMS(LDK(KP250000000), To, VMUL(LDK(KP500000000), Ta)); cannam@95: ST(&(Rp[0]), Tp, ms, &(Rp[0])); cannam@95: Tt = VCONJ(VADD(Tq, Tr)); cannam@95: ST(&(Rm[WS(rs, 1)]), Tt, -ms, &(Rm[WS(rs, 1)])); cannam@95: Ts = VSUB(Tq, Tr); cannam@95: ST(&(Rp[WS(rs, 2)]), Ts, ms, &(Rp[0])); cannam@95: TA = VMUL(LDK(KP500000000), VBYI(VMUL(LDK(KP866025403), VSUB(Tw, Tv)))); cannam@95: Tx = VADD(Tv, Tw); cannam@95: Ty = VCONJ(VMUL(LDK(KP500000000), VADD(Tu, Tx))); cannam@95: Tz = VFNMS(LDK(KP250000000), Tx, VMUL(LDK(KP500000000), Tu)); cannam@95: ST(&(Rm[WS(rs, 2)]), Ty, -ms, &(Rm[0])); cannam@95: TC = VADD(Tz, TA); cannam@95: ST(&(Rp[WS(rs, 1)]), TC, ms, &(Rp[WS(rs, 1)])); cannam@95: TB = VCONJ(VSUB(Tz, TA)); cannam@95: ST(&(Rm[0]), TB, -ms, &(Rm[0])); cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(1, 1), cannam@95: VTW(1, 2), cannam@95: VTW(1, 3), cannam@95: VTW(1, 4), cannam@95: VTW(1, 5), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 6, XSIMD_STRING("hc2cfdftv_6"), twinstr, &GENUS, {27, 18, 2, 0} }; cannam@95: cannam@95: void XSIMD(codelet_hc2cfdftv_6) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cfdftv_6, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */