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 10 -dit -name hc2cfdftv_10 -include hc2cfv.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 61 FP additions, 60 FP multiplications, cannam@95: * (or, 33 additions, 32 multiplications, 28 fused multiply/add), cannam@95: * 77 stack variables, 5 constants, and 20 memory accesses cannam@95: */ cannam@95: #include "hc2cfv.h" cannam@95: cannam@95: static void hc2cfdftv_10(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@95: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DVK(KP618033988, +0.618033988749894848204586834365638117720309180); cannam@95: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 18)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 18), MAKE_VOLATILE_STRIDE(40, rs)) { cannam@95: V T5, T6, Tw, Tr, Tc, Tj, Tl, Tm, Tk, Ts, Tg, Ty, T3, T4, T1; cannam@95: V T2, Tv, Tq, Ta, Tb, T9, Ti, Te, Tf, Td, Tx, Tn, Tt, Th, TQ; cannam@95: V TT, Tz, T7, TR, To, Tu, TU; cannam@95: T1 = LD(&(Rp[0]), ms, &(Rp[0])); cannam@95: T2 = LD(&(Rm[0]), -ms, &(Rm[0])); cannam@95: Tv = LDW(&(W[0])); cannam@95: T5 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); cannam@95: T6 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); cannam@95: Tq = LDW(&(W[TWVL * 6])); cannam@95: Ta = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); cannam@95: Tb = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: T9 = LDW(&(W[TWVL * 2])); cannam@95: Ti = LDW(&(W[TWVL * 4])); cannam@95: Tw = VZMULIJ(Tv, VFNMSCONJ(T2, T1)); cannam@95: Te = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); cannam@95: Tf = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: Tr = VZMULJ(Tq, VFMACONJ(T6, T5)); cannam@95: Td = LDW(&(W[TWVL * 12])); cannam@95: Tx = LDW(&(W[TWVL * 10])); cannam@95: Tc = VZMULJ(T9, VFMACONJ(Tb, Ta)); cannam@95: Tj = VZMULIJ(Ti, VFNMSCONJ(Tb, Ta)); cannam@95: Tl = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0])); cannam@95: Tm = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0])); cannam@95: Tk = LDW(&(W[TWVL * 14])); cannam@95: Ts = LDW(&(W[TWVL * 16])); cannam@95: Tg = VZMULIJ(Td, VFNMSCONJ(Tf, Te)); cannam@95: Ty = VZMULJ(Tx, VFMACONJ(Tf, Te)); cannam@95: T3 = VFMACONJ(T2, T1); cannam@95: T4 = LDW(&(W[TWVL * 8])); cannam@95: Tn = VZMULJ(Tk, VFMACONJ(Tm, Tl)); cannam@95: Tt = VZMULIJ(Ts, VFNMSCONJ(Tm, Tl)); cannam@95: Th = VSUB(Tc, Tg); cannam@95: TQ = VADD(Tc, Tg); cannam@95: TT = VADD(Tw, Ty); cannam@95: Tz = VSUB(Tw, Ty); cannam@95: T7 = VZMULIJ(T4, VFNMSCONJ(T6, T5)); cannam@95: TR = VADD(Tj, Tn); cannam@95: To = VSUB(Tj, Tn); cannam@95: Tu = VSUB(Tr, Tt); cannam@95: TU = VADD(Tr, Tt); cannam@95: { cannam@95: V TP, T8, TS, T11, Tp, TH, TA, TG, TV, T12, TE, TB, TM, TI, TZ; cannam@95: V TW, T17, T13, TD, TC, TY, TX, TL, TF, T10, T16, TN, TO, TK, TJ; cannam@95: V T18, T19, T15, T14; cannam@95: TP = VADD(T3, T7); cannam@95: T8 = VSUB(T3, T7); cannam@95: TS = VADD(TQ, TR); cannam@95: T11 = VSUB(TQ, TR); cannam@95: Tp = VSUB(Th, To); cannam@95: TH = VADD(Th, To); cannam@95: TA = VSUB(Tu, Tz); cannam@95: TG = VADD(Tz, Tu); cannam@95: TV = VADD(TT, TU); cannam@95: T12 = VSUB(TU, TT); cannam@95: TE = VSUB(Tp, TA); cannam@95: TB = VADD(Tp, TA); cannam@95: TM = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), TG, TH)); cannam@95: TI = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), TH, TG)); cannam@95: TZ = VSUB(TS, TV); cannam@95: TW = VADD(TS, TV); cannam@95: T17 = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), T11, T12)); cannam@95: T13 = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), T12, T11)); cannam@95: TD = VFNMS(LDK(KP250000000), TB, T8); cannam@95: TC = VMUL(LDK(KP500000000), VADD(T8, TB)); cannam@95: TY = VFNMS(LDK(KP250000000), TW, TP); cannam@95: TX = VCONJ(VMUL(LDK(KP500000000), VADD(TP, TW))); cannam@95: TL = VFMA(LDK(KP559016994), TE, TD); cannam@95: TF = VFNMS(LDK(KP559016994), TE, TD); cannam@95: ST(&(Rp[0]), TC, ms, &(Rp[0])); cannam@95: T10 = VFMA(LDK(KP559016994), TZ, TY); cannam@95: T16 = VFNMS(LDK(KP559016994), TZ, TY); cannam@95: ST(&(Rm[WS(rs, 4)]), TX, -ms, &(Rm[0])); cannam@95: TN = VCONJ(VMUL(LDK(KP500000000), VFNMSI(TM, TL))); cannam@95: TO = VMUL(LDK(KP500000000), VFMAI(TM, TL)); cannam@95: TK = VMUL(LDK(KP500000000), VFMAI(TI, TF)); cannam@95: TJ = VCONJ(VMUL(LDK(KP500000000), VFNMSI(TI, TF))); cannam@95: T18 = VMUL(LDK(KP500000000), VFNMSI(T17, T16)); cannam@95: T19 = VCONJ(VMUL(LDK(KP500000000), VFMAI(T17, T16))); cannam@95: T15 = VCONJ(VMUL(LDK(KP500000000), VFMAI(T13, T10))); cannam@95: T14 = VMUL(LDK(KP500000000), VFNMSI(T13, T10)); cannam@95: ST(&(Rm[WS(rs, 3)]), TN, -ms, &(Rm[WS(rs, 1)])); cannam@95: ST(&(Rp[WS(rs, 4)]), TO, ms, &(Rp[0])); cannam@95: ST(&(Rp[WS(rs, 2)]), TK, ms, &(Rp[0])); cannam@95: ST(&(Rm[WS(rs, 1)]), TJ, -ms, &(Rm[WS(rs, 1)])); cannam@95: ST(&(Rp[WS(rs, 3)]), T18, ms, &(Rp[WS(rs, 1)])); cannam@95: ST(&(Rm[WS(rs, 2)]), T19, -ms, &(Rm[0])); cannam@95: ST(&(Rm[0]), T15, -ms, &(Rm[0])); cannam@95: ST(&(Rp[WS(rs, 1)]), T14, ms, &(Rp[WS(rs, 1)])); cannam@95: } 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: VTW(1, 6), cannam@95: VTW(1, 7), cannam@95: VTW(1, 8), cannam@95: VTW(1, 9), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 10, XSIMD_STRING("hc2cfdftv_10"), twinstr, &GENUS, {33, 32, 28, 0} }; cannam@95: cannam@95: void XSIMD(codelet_hc2cfdftv_10) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cfdftv_10, &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 10 -dit -name hc2cfdftv_10 -include hc2cfv.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 61 FP additions, 38 FP multiplications, cannam@95: * (or, 55 additions, 32 multiplications, 6 fused multiply/add), cannam@95: * 82 stack variables, 5 constants, and 20 memory accesses cannam@95: */ cannam@95: #include "hc2cfv.h" cannam@95: cannam@95: static void hc2cfdftv_10(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP125000000, +0.125000000000000000000000000000000000000000000); cannam@95: DVK(KP279508497, +0.279508497187473712051146708591409529430077295); cannam@95: DVK(KP587785252, +0.587785252292473129168705954639072768597652438); cannam@95: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@95: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 18)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 18), MAKE_VOLATILE_STRIDE(40, rs)) { cannam@95: V Tl, Tt, Tu, TY, TZ, T10, Tz, TE, TF, TV, TW, TX, Ta, TU, TN; cannam@95: V TR, TH, TQ, TK, TL, TM, TI, TG, TJ, TT, TO, TP, TS, T18, T1c; cannam@95: V T12, T1b, T15, T16, T17, T14, T11, T13, T1e, T19, T1a, T1d; cannam@95: { cannam@95: V T1, T3, Ty, T8, T7, TB, Tf, Ts, Tk, Tw, Tq, TD, T2, Tx, T6; cannam@95: V TA, Tc, Te, Td, Tb, Tr, Tj, Ti, Th, Tg, Tv, Tn, Tp, To, Tm; cannam@95: V TC, T4, T9, T5; cannam@95: T1 = LD(&(Rp[0]), ms, &(Rp[0])); cannam@95: T2 = LD(&(Rm[0]), -ms, &(Rm[0])); cannam@95: T3 = VCONJ(T2); cannam@95: Tx = LDW(&(W[0])); cannam@95: Ty = VZMULIJ(Tx, VSUB(T3, T1)); cannam@95: T8 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); cannam@95: T6 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); cannam@95: T7 = VCONJ(T6); cannam@95: TA = LDW(&(W[TWVL * 6])); cannam@95: TB = VZMULJ(TA, VADD(T7, T8)); cannam@95: Tc = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); cannam@95: Td = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: Te = VCONJ(Td); cannam@95: Tb = LDW(&(W[TWVL * 2])); cannam@95: Tf = VZMULJ(Tb, VADD(Tc, Te)); cannam@95: Tr = LDW(&(W[TWVL * 4])); cannam@95: Ts = VZMULIJ(Tr, VSUB(Te, Tc)); cannam@95: Tj = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); cannam@95: Th = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: Ti = VCONJ(Th); cannam@95: Tg = LDW(&(W[TWVL * 12])); cannam@95: Tk = VZMULIJ(Tg, VSUB(Ti, Tj)); cannam@95: Tv = LDW(&(W[TWVL * 10])); cannam@95: Tw = VZMULJ(Tv, VADD(Ti, Tj)); cannam@95: Tn = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0])); cannam@95: To = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0])); cannam@95: Tp = VCONJ(To); cannam@95: Tm = LDW(&(W[TWVL * 14])); cannam@95: Tq = VZMULJ(Tm, VADD(Tn, Tp)); cannam@95: TC = LDW(&(W[TWVL * 16])); cannam@95: TD = VZMULIJ(TC, VSUB(Tp, Tn)); cannam@95: Tl = VSUB(Tf, Tk); cannam@95: Tt = VSUB(Tq, Ts); cannam@95: Tu = VADD(Tl, Tt); cannam@95: TY = VADD(Ty, Tw); cannam@95: TZ = VADD(TB, TD); cannam@95: T10 = VADD(TY, TZ); cannam@95: Tz = VSUB(Tw, Ty); cannam@95: TE = VSUB(TB, TD); cannam@95: TF = VADD(Tz, TE); cannam@95: TV = VADD(Tf, Tk); cannam@95: TW = VADD(Ts, Tq); cannam@95: TX = VADD(TV, TW); cannam@95: T4 = VADD(T1, T3); cannam@95: T5 = LDW(&(W[TWVL * 8])); cannam@95: T9 = VZMULIJ(T5, VSUB(T7, T8)); cannam@95: Ta = VSUB(T4, T9); cannam@95: TU = VADD(T4, T9); cannam@95: } cannam@95: TL = VSUB(Tl, Tt); cannam@95: TM = VSUB(TE, Tz); cannam@95: TN = VMUL(LDK(KP500000000), VBYI(VFMA(LDK(KP951056516), TL, VMUL(LDK(KP587785252), TM)))); cannam@95: TR = VMUL(LDK(KP500000000), VBYI(VFNMS(LDK(KP587785252), TL, VMUL(LDK(KP951056516), TM)))); cannam@95: TI = VMUL(LDK(KP279508497), VSUB(Tu, TF)); cannam@95: TG = VADD(Tu, TF); cannam@95: TJ = VFNMS(LDK(KP125000000), TG, VMUL(LDK(KP500000000), Ta)); cannam@95: TH = VCONJ(VMUL(LDK(KP500000000), VADD(Ta, TG))); cannam@95: TQ = VSUB(TJ, TI); cannam@95: TK = VADD(TI, TJ); cannam@95: ST(&(Rm[WS(rs, 4)]), TH, -ms, &(Rm[0])); cannam@95: TT = VCONJ(VADD(TQ, TR)); cannam@95: ST(&(Rm[WS(rs, 2)]), TT, -ms, &(Rm[0])); cannam@95: TO = VSUB(TK, TN); cannam@95: ST(&(Rp[WS(rs, 1)]), TO, ms, &(Rp[WS(rs, 1)])); cannam@95: TP = VCONJ(VADD(TK, TN)); cannam@95: ST(&(Rm[0]), TP, -ms, &(Rm[0])); cannam@95: TS = VSUB(TQ, TR); cannam@95: ST(&(Rp[WS(rs, 3)]), TS, ms, &(Rp[WS(rs, 1)])); cannam@95: T16 = VSUB(TZ, TY); cannam@95: T17 = VSUB(TV, TW); cannam@95: T18 = VMUL(LDK(KP500000000), VBYI(VFNMS(LDK(KP587785252), T17, VMUL(LDK(KP951056516), T16)))); cannam@95: T1c = VMUL(LDK(KP500000000), VBYI(VFMA(LDK(KP951056516), T17, VMUL(LDK(KP587785252), T16)))); cannam@95: T14 = VMUL(LDK(KP279508497), VSUB(TX, T10)); cannam@95: T11 = VADD(TX, T10); cannam@95: T13 = VFNMS(LDK(KP125000000), T11, VMUL(LDK(KP500000000), TU)); cannam@95: T12 = VMUL(LDK(KP500000000), VADD(TU, T11)); cannam@95: T1b = VADD(T14, T13); cannam@95: T15 = VSUB(T13, T14); cannam@95: ST(&(Rp[0]), T12, ms, &(Rp[0])); cannam@95: T1e = VADD(T1b, T1c); cannam@95: ST(&(Rp[WS(rs, 4)]), T1e, ms, &(Rp[0])); cannam@95: T19 = VCONJ(VSUB(T15, T18)); cannam@95: ST(&(Rm[WS(rs, 1)]), T19, -ms, &(Rm[WS(rs, 1)])); cannam@95: T1a = VADD(T15, T18); cannam@95: ST(&(Rp[WS(rs, 2)]), T1a, ms, &(Rp[0])); cannam@95: T1d = VCONJ(VSUB(T1b, T1c)); cannam@95: ST(&(Rm[WS(rs, 3)]), T1d, -ms, &(Rm[WS(rs, 1)])); 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: VTW(1, 6), cannam@95: VTW(1, 7), cannam@95: VTW(1, 8), cannam@95: VTW(1, 9), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 10, XSIMD_STRING("hc2cfdftv_10"), twinstr, &GENUS, {55, 32, 6, 0} }; cannam@95: cannam@95: void XSIMD(codelet_hc2cfdftv_10) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cfdftv_10, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */