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 4 -dif -sign 1 -name hc2cbdftv_4 -include hc2cbv.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 15 FP additions, 12 FP multiplications, cannam@95: * (or, 9 additions, 6 multiplications, 6 fused multiply/add), cannam@95: * 20 stack variables, 0 constants, and 8 memory accesses cannam@95: */ cannam@95: #include "hc2cbv.h" cannam@95: cannam@95: static void hc2cbdftv_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: V T2, T3, T5, T6, Tf, T1, T9, Ta, T4, Tb, T7, Tc, Th, T8, Tg; cannam@95: V Te, Td, Ti, Tj; cannam@95: T2 = LD(&(Rp[0]), ms, &(Rp[0])); cannam@95: T3 = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: T5 = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); cannam@95: T6 = LD(&(Rm[0]), -ms, &(Rm[0])); cannam@95: Tf = LDW(&(W[0])); cannam@95: T1 = LDW(&(W[TWVL * 4])); cannam@95: T9 = LDW(&(W[TWVL * 2])); cannam@95: Ta = VFMACONJ(T3, T2); cannam@95: T4 = VFNMSCONJ(T3, T2); cannam@95: Tb = VFMACONJ(T6, T5); cannam@95: T7 = VFNMSCONJ(T6, T5); cannam@95: Tc = VZMUL(T9, VSUB(Ta, Tb)); cannam@95: Th = VADD(Ta, Tb); cannam@95: T8 = VZMULI(T1, VFNMSI(T7, T4)); cannam@95: Tg = VZMULI(Tf, VFMAI(T7, T4)); cannam@95: Te = VCONJ(VSUB(Tc, T8)); cannam@95: Td = VADD(T8, Tc); cannam@95: Ti = VADD(Tg, Th); cannam@95: Tj = VCONJ(VSUB(Th, Tg)); cannam@95: ST(&(Rm[WS(rs, 1)]), Te, -ms, &(Rm[WS(rs, 1)])); cannam@95: ST(&(Rp[WS(rs, 1)]), Td, ms, &(Rp[WS(rs, 1)])); cannam@95: ST(&(Rp[0]), Ti, ms, &(Rp[0])); cannam@95: ST(&(Rm[0]), Tj, -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: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 4, XSIMD_STRING("hc2cbdftv_4"), twinstr, &GENUS, {9, 6, 6, 0} }; cannam@95: cannam@95: void XSIMD(codelet_hc2cbdftv_4) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cbdftv_4, &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 4 -dif -sign 1 -name hc2cbdftv_4 -include hc2cbv.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 15 FP additions, 6 FP multiplications, cannam@95: * (or, 15 additions, 6 multiplications, 0 fused multiply/add), cannam@95: * 22 stack variables, 0 constants, and 8 memory accesses cannam@95: */ cannam@95: #include "hc2cbv.h" cannam@95: cannam@95: static void hc2cbdftv_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: V T5, Tc, T9, Td, T2, T4, T3, T6, T8, T7, Tj, Ti, Th, Tk, Tl; cannam@95: V Ta, Te, T1, Tb, Tf, Tg; cannam@95: T2 = LD(&(Rp[0]), ms, &(Rp[0])); cannam@95: T3 = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: T4 = VCONJ(T3); cannam@95: T5 = VSUB(T2, T4); cannam@95: Tc = VADD(T2, T4); cannam@95: T6 = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); cannam@95: T7 = LD(&(Rm[0]), -ms, &(Rm[0])); cannam@95: T8 = VCONJ(T7); cannam@95: T9 = VBYI(VSUB(T6, T8)); cannam@95: Td = VADD(T6, T8); cannam@95: Tj = VADD(Tc, Td); cannam@95: Th = LDW(&(W[0])); cannam@95: Ti = VZMULI(Th, VADD(T5, T9)); cannam@95: Tk = VADD(Ti, Tj); cannam@95: ST(&(Rp[0]), Tk, ms, &(Rp[0])); cannam@95: Tl = VCONJ(VSUB(Tj, Ti)); cannam@95: ST(&(Rm[0]), Tl, -ms, &(Rm[0])); cannam@95: T1 = LDW(&(W[TWVL * 4])); cannam@95: Ta = VZMULI(T1, VSUB(T5, T9)); cannam@95: Tb = LDW(&(W[TWVL * 2])); cannam@95: Te = VZMUL(Tb, VSUB(Tc, Td)); cannam@95: Tf = VADD(Ta, Te); cannam@95: ST(&(Rp[WS(rs, 1)]), Tf, ms, &(Rp[WS(rs, 1)])); cannam@95: Tg = VCONJ(VSUB(Te, Ta)); cannam@95: ST(&(Rm[WS(rs, 1)]), Tg, -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: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 4, XSIMD_STRING("hc2cbdftv_4"), twinstr, &GENUS, {15, 6, 0, 0} }; cannam@95: cannam@95: void XSIMD(codelet_hc2cbdftv_4) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cbdftv_4, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */