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