Chris@82: /* Chris@82: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@82: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@82: * Chris@82: * This program is free software; you can redistribute it and/or modify Chris@82: * it under the terms of the GNU General Public License as published by Chris@82: * the Free Software Foundation; either version 2 of the License, or Chris@82: * (at your option) any later version. Chris@82: * Chris@82: * This program is distributed in the hope that it will be useful, Chris@82: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@82: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@82: * GNU General Public License for more details. Chris@82: * Chris@82: * You should have received a copy of the GNU General Public License Chris@82: * along with this program; if not, write to the Free Software Chris@82: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@82: * Chris@82: */ Chris@82: Chris@82: /* This file was automatically generated --- DO NOT EDIT */ Chris@82: /* Generated on Thu May 24 08:08:11 EDT 2018 */ Chris@82: Chris@82: #include "rdft/codelet-rdft.h" Chris@82: Chris@82: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_hc2cdft_c.native -fma -simd -compact -variables 4 -pipeline-latency 8 -trivial-stores -variables 32 -no-generate-bytw -n 6 -dif -sign 1 -name hc2cbdftv_6 -include rdft/simd/hc2cbv.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 29 FP additions, 24 FP multiplications, Chris@82: * (or, 17 additions, 12 multiplications, 12 fused multiply/add), Chris@82: * 38 stack variables, 2 constants, and 12 memory accesses Chris@82: */ Chris@82: #include "rdft/simd/hc2cbv.h" Chris@82: Chris@82: static void hc2cbdftv_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@82: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@82: { Chris@82: INT m; Chris@82: 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@82: V T4, Te, Tj, Tp, Tb, To, Th, Ti, Ta, Tg, T7, Tf, T2, T3, T8; Chris@82: V T9, T5, T6, Tx, Tw, Tv, Ty, Tz, Tq, Ts, Tn, Tr, Tt, Tu, Tc; Chris@82: V Tk, T1, Td, Tl, Tm; Chris@82: T2 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@82: T3 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@82: T4 = VFNMSCONJ(T3, T2); Chris@82: Te = VFMACONJ(T3, T2); Chris@82: T8 = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@82: T9 = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@82: Ta = VFMSCONJ(T9, T8); Chris@82: Tg = VFMACONJ(T9, T8); Chris@82: T5 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@82: T6 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@82: T7 = VFNMSCONJ(T6, T5); Chris@82: Tf = VFMACONJ(T6, T5); Chris@82: Tj = VMUL(LDK(KP866025403), VSUB(Tf, Tg)); Chris@82: Tp = VMUL(LDK(KP866025403), VSUB(T7, Ta)); Chris@82: Tb = VADD(T7, Ta); Chris@82: To = VFNMS(LDK(KP500000000), Tb, T4); Chris@82: Th = VADD(Tf, Tg); Chris@82: Ti = VFNMS(LDK(KP500000000), Th, Te); Chris@82: Tx = VADD(Te, Th); Chris@82: Tv = LDW(&(W[0])); Chris@82: Tw = VZMULI(Tv, VFMAI(Tp, To)); Chris@82: Ty = VADD(Tw, Tx); Chris@82: ST(&(Rp[0]), Ty, ms, &(Rp[0])); Chris@82: Tz = VCONJ(VSUB(Tx, Tw)); Chris@82: ST(&(Rm[0]), Tz, -ms, &(Rm[0])); Chris@82: Tn = LDW(&(W[TWVL * 8])); Chris@82: Tq = VZMULI(Tn, VFNMSI(Tp, To)); Chris@82: Tr = LDW(&(W[TWVL * 6])); Chris@82: Ts = VZMUL(Tr, VFMAI(Tj, Ti)); Chris@82: Tt = VADD(Tq, Ts); Chris@82: ST(&(Rp[WS(rs, 2)]), Tt, ms, &(Rp[0])); Chris@82: Tu = VCONJ(VSUB(Ts, Tq)); Chris@82: ST(&(Rm[WS(rs, 2)]), Tu, -ms, &(Rm[0])); Chris@82: T1 = LDW(&(W[TWVL * 4])); Chris@82: Tc = VZMULI(T1, VADD(T4, Tb)); Chris@82: Td = LDW(&(W[TWVL * 2])); Chris@82: Tk = VZMUL(Td, VFNMSI(Tj, Ti)); Chris@82: Tl = VADD(Tc, Tk); Chris@82: ST(&(Rp[WS(rs, 1)]), Tl, ms, &(Rp[WS(rs, 1)])); Chris@82: Tm = VCONJ(VSUB(Tk, Tc)); Chris@82: ST(&(Rm[WS(rs, 1)]), Tm, -ms, &(Rm[WS(rs, 1)])); Chris@82: } Chris@82: } Chris@82: VLEAVE(); Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: VTW(1, 1), Chris@82: VTW(1, 2), Chris@82: VTW(1, 3), Chris@82: VTW(1, 4), Chris@82: VTW(1, 5), Chris@82: {TW_NEXT, VL, 0} Chris@82: }; Chris@82: Chris@82: static const hc2c_desc desc = { 6, XSIMD_STRING("hc2cbdftv_6"), twinstr, &GENUS, {17, 12, 12, 0} }; Chris@82: Chris@82: void XSIMD(codelet_hc2cbdftv_6) (planner *p) { Chris@82: X(khc2c_register) (p, hc2cbdftv_6, &desc, HC2C_VIA_DFT); Chris@82: } Chris@82: #else Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_hc2cdft_c.native -simd -compact -variables 4 -pipeline-latency 8 -trivial-stores -variables 32 -no-generate-bytw -n 6 -dif -sign 1 -name hc2cbdftv_6 -include rdft/simd/hc2cbv.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 29 FP additions, 14 FP multiplications, Chris@82: * (or, 27 additions, 12 multiplications, 2 fused multiply/add), Chris@82: * 41 stack variables, 2 constants, and 12 memory accesses Chris@82: */ Chris@82: #include "rdft/simd/hc2cbv.h" Chris@82: Chris@82: static void hc2cbdftv_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@82: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@82: { Chris@82: INT m; Chris@82: 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@82: V T5, Th, Te, Ts, Tk, Tm, T2, T4, T3, T6, Tc, T8, Tb, T7, Ta; Chris@82: V T9, Td, Ti, Tj, TA, Tf, Tn, Tv, Tt, Tz, T1, Tl, Tg, Tu, Tr; Chris@82: V Tq, Ty, To, Tp, TC, TB, Tx, Tw; Chris@82: T2 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@82: T3 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); Chris@82: T4 = VCONJ(T3); Chris@82: T5 = VSUB(T2, T4); Chris@82: Th = VADD(T2, T4); Chris@82: T6 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); Chris@82: Tc = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@82: T7 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@82: T8 = VCONJ(T7); Chris@82: Ta = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@82: Tb = VCONJ(Ta); Chris@82: T9 = VSUB(T6, T8); Chris@82: Td = VSUB(Tb, Tc); Chris@82: Te = VADD(T9, Td); Chris@82: Ts = VBYI(VMUL(LDK(KP866025403), VSUB(T9, Td))); Chris@82: Ti = VADD(T6, T8); Chris@82: Tj = VADD(Tb, Tc); Chris@82: Tk = VADD(Ti, Tj); Chris@82: Tm = VBYI(VMUL(LDK(KP866025403), VSUB(Ti, Tj))); Chris@82: TA = VADD(Th, Tk); Chris@82: T1 = LDW(&(W[TWVL * 4])); Chris@82: Tf = VZMULI(T1, VADD(T5, Te)); Chris@82: Tl = VFNMS(LDK(KP500000000), Tk, Th); Chris@82: Tg = LDW(&(W[TWVL * 2])); Chris@82: Tn = VZMUL(Tg, VSUB(Tl, Tm)); Chris@82: Tu = LDW(&(W[TWVL * 6])); Chris@82: Tv = VZMUL(Tu, VADD(Tm, Tl)); Chris@82: Tr = VFNMS(LDK(KP500000000), Te, T5); Chris@82: Tq = LDW(&(W[TWVL * 8])); Chris@82: Tt = VZMULI(Tq, VSUB(Tr, Ts)); Chris@82: Ty = LDW(&(W[0])); Chris@82: Tz = VZMULI(Ty, VADD(Ts, Tr)); Chris@82: To = VADD(Tf, Tn); Chris@82: ST(&(Rp[WS(rs, 1)]), To, ms, &(Rp[WS(rs, 1)])); Chris@82: Tp = VCONJ(VSUB(Tn, Tf)); Chris@82: ST(&(Rm[WS(rs, 1)]), Tp, -ms, &(Rm[WS(rs, 1)])); Chris@82: TC = VCONJ(VSUB(TA, Tz)); Chris@82: ST(&(Rm[0]), TC, -ms, &(Rm[0])); Chris@82: TB = VADD(Tz, TA); Chris@82: ST(&(Rp[0]), TB, ms, &(Rp[0])); Chris@82: Tx = VCONJ(VSUB(Tv, Tt)); Chris@82: ST(&(Rm[WS(rs, 2)]), Tx, -ms, &(Rm[0])); Chris@82: Tw = VADD(Tt, Tv); Chris@82: ST(&(Rp[WS(rs, 2)]), Tw, ms, &(Rp[0])); Chris@82: } Chris@82: } Chris@82: VLEAVE(); Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: VTW(1, 1), Chris@82: VTW(1, 2), Chris@82: VTW(1, 3), Chris@82: VTW(1, 4), Chris@82: VTW(1, 5), Chris@82: {TW_NEXT, VL, 0} Chris@82: }; Chris@82: Chris@82: static const hc2c_desc desc = { 6, XSIMD_STRING("hc2cbdftv_6"), twinstr, &GENUS, {27, 12, 2, 0} }; Chris@82: Chris@82: void XSIMD(codelet_hc2cbdftv_6) (planner *p) { Chris@82: X(khc2c_register) (p, hc2cbdftv_6, &desc, HC2C_VIA_DFT); Chris@82: } Chris@82: #endif