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 4 -dif -sign 1 -name hc2cbdftv_4 -include hc2cbv.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 15 FP additions, 12 FP multiplications, Chris@10: * (or, 9 additions, 6 multiplications, 6 fused multiply/add), Chris@10: * 20 stack variables, 0 constants, and 8 memory accesses Chris@10: */ Chris@10: #include "hc2cbv.h" Chris@10: Chris@10: static void hc2cbdftv_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 6)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 6), MAKE_VOLATILE_STRIDE(16, rs)) { Chris@10: V T2, T3, T5, T6, Tf, T1, T9, Ta, T4, Tb, T7, Tc, Th, T8, Tg; Chris@10: V Te, Td, Ti, Tj; Chris@10: T2 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@10: T3 = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: T5 = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@10: T6 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@10: Tf = LDW(&(W[0])); Chris@10: T1 = LDW(&(W[TWVL * 4])); Chris@10: T9 = LDW(&(W[TWVL * 2])); Chris@10: Ta = VFMACONJ(T3, T2); Chris@10: T4 = VFNMSCONJ(T3, T2); Chris@10: Tb = VFMACONJ(T6, T5); Chris@10: T7 = VFNMSCONJ(T6, T5); Chris@10: Tc = VZMUL(T9, VSUB(Ta, Tb)); Chris@10: Th = VADD(Ta, Tb); Chris@10: T8 = VZMULI(T1, VFNMSI(T7, T4)); Chris@10: Tg = VZMULI(Tf, VFMAI(T7, T4)); Chris@10: Te = VCONJ(VSUB(Tc, T8)); Chris@10: Td = VADD(T8, Tc); Chris@10: Ti = VADD(Tg, Th); Chris@10: Tj = VCONJ(VSUB(Th, Tg)); Chris@10: ST(&(Rm[WS(rs, 1)]), Te, -ms, &(Rm[WS(rs, 1)])); Chris@10: ST(&(Rp[WS(rs, 1)]), Td, ms, &(Rp[WS(rs, 1)])); Chris@10: ST(&(Rp[0]), Ti, ms, &(Rp[0])); Chris@10: ST(&(Rm[0]), Tj, -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: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 4, XSIMD_STRING("hc2cbdftv_4"), twinstr, &GENUS, {9, 6, 6, 0} }; Chris@10: Chris@10: void XSIMD(codelet_hc2cbdftv_4) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cbdftv_4, &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 4 -dif -sign 1 -name hc2cbdftv_4 -include hc2cbv.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 15 FP additions, 6 FP multiplications, Chris@10: * (or, 15 additions, 6 multiplications, 0 fused multiply/add), Chris@10: * 22 stack variables, 0 constants, and 8 memory accesses Chris@10: */ Chris@10: #include "hc2cbv.h" Chris@10: Chris@10: static void hc2cbdftv_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 6)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 6), MAKE_VOLATILE_STRIDE(16, rs)) { Chris@10: V T5, Tc, T9, Td, T2, T4, T3, T6, T8, T7, Tj, Ti, Th, Tk, Tl; Chris@10: V Ta, Te, T1, Tb, Tf, Tg; Chris@10: T2 = LD(&(Rp[0]), ms, &(Rp[0])); Chris@10: T3 = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); Chris@10: T4 = VCONJ(T3); Chris@10: T5 = VSUB(T2, T4); Chris@10: Tc = VADD(T2, T4); Chris@10: T6 = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); Chris@10: T7 = LD(&(Rm[0]), -ms, &(Rm[0])); Chris@10: T8 = VCONJ(T7); Chris@10: T9 = VBYI(VSUB(T6, T8)); Chris@10: Td = VADD(T6, T8); Chris@10: Tj = VADD(Tc, Td); Chris@10: Th = LDW(&(W[0])); Chris@10: Ti = VZMULI(Th, VADD(T5, T9)); Chris@10: Tk = VADD(Ti, Tj); Chris@10: ST(&(Rp[0]), Tk, ms, &(Rp[0])); Chris@10: Tl = VCONJ(VSUB(Tj, Ti)); Chris@10: ST(&(Rm[0]), Tl, -ms, &(Rm[0])); Chris@10: T1 = LDW(&(W[TWVL * 4])); Chris@10: Ta = VZMULI(T1, VSUB(T5, T9)); Chris@10: Tb = LDW(&(W[TWVL * 2])); Chris@10: Te = VZMUL(Tb, VSUB(Tc, Td)); Chris@10: Tf = VADD(Ta, Te); Chris@10: ST(&(Rp[WS(rs, 1)]), Tf, ms, &(Rp[WS(rs, 1)])); Chris@10: Tg = VCONJ(VSUB(Te, Ta)); Chris@10: ST(&(Rm[WS(rs, 1)]), Tg, -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: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 4, XSIMD_STRING("hc2cbdftv_4"), twinstr, &GENUS, {15, 6, 0, 0} }; Chris@10: Chris@10: void XSIMD(codelet_hc2cbdftv_4) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cbdftv_4, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */