diff --git a/Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h b/Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h index 44d4bb46..cdf1ade5 100644 --- a/Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h +++ b/Grid/qcd/action/fermion/implementation/WilsonKernelsHandImplementation.h @@ -91,29 +91,29 @@ Author: paboyle Chimu_31=ref()(3)(1);\ Chimu_32=ref()(3)(2);\ std::cout << std::endl << "DEBUG -- LOAD_CHIMU" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chimu_00); \ + svst1(pg1, &debugreg.v, Chimu_00); \ std::cout << "Chimu_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chimu_01); \ + svst1(pg1, &debugreg.v, Chimu_01); \ std::cout << "Chimu_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chimu_02); \ + svst1(pg1, &debugreg.v, Chimu_02); \ std::cout << "Chimu_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chimu_10); \ + svst1(pg1, &debugreg.v, Chimu_10); \ std::cout << "Chimu_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chimu_11); \ + svst1(pg1, &debugreg.v, Chimu_11); \ std::cout << "Chimu_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chimu_12); \ + svst1(pg1, &debugreg.v, Chimu_12); \ std::cout << "Chimu_12 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chimu_20); \ + svst1(pg1, &debugreg.v, Chimu_20); \ std::cout << "Chimu_20 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chimu_21); \ + svst1(pg1, &debugreg.v, Chimu_21); \ std::cout << "Chimu_21 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chimu_22); \ + svst1(pg1, &debugreg.v, Chimu_22); \ std::cout << "Chimu_22 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chimu_30); \ + svst1(pg1, &debugreg.v, Chimu_30); \ std::cout << "Chimu_30 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chimu_31); \ + svst1(pg1, &debugreg.v, Chimu_31); \ std::cout << "Chimu_31 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chimu_32); \ + svst1(pg1, &debugreg.v, Chimu_32); \ std::cout << "Chimu_32 -- " << debugreg << std::endl; \ } @@ -126,17 +126,17 @@ Author: paboyle Chi_11 = ref()(1)(1);\ Chi_12 = ref()(1)(2);\ std::cout << std::endl << "DEBUG -- LOAD_CHI" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_00); \ + svst1(pg1, &debugreg.v, Chi_00); \ std::cout << "Chi_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_01); \ + svst1(pg1, &debugreg.v, Chi_01); \ std::cout << "Chi_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_02); \ + svst1(pg1, &debugreg.v, Chi_02); \ std::cout << "Chi_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_10); \ + svst1(pg1, &debugreg.v, Chi_10); \ std::cout << "Chi_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_11); \ + svst1(pg1, &debugreg.v, Chi_11); \ std::cout << "Chi_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_12); \ + svst1(pg1, &debugreg.v, Chi_12); \ std::cout << "Chi_12 -- " << debugreg << std::endl; \ } @@ -171,34 +171,34 @@ Author: paboyle UChi_02+= U_20*Chi_02;\ UChi_12+= U_20*Chi_12;\ std::cout << std::endl << "DEBUG -- MULT_2SPIN" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, UChi_00); \ + svst1(pg1, &debugreg.v, UChi_00); \ std::cout << "UChi_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, UChi_01); \ + svst1(pg1, &debugreg.v, UChi_01); \ std::cout << "UChi_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, UChi_02); \ + svst1(pg1, &debugreg.v, UChi_02); \ std::cout << "UChi_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, UChi_10); \ + svst1(pg1, &debugreg.v, UChi_10); \ std::cout << "UChi_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, UChi_11); \ + svst1(pg1, &debugreg.v, UChi_11); \ std::cout << "UChi_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, UChi_12); \ + svst1(pg1, &debugreg.v, UChi_12); \ std::cout << "UChi_12 -- " << debugreg << std::endl; \ } #define PERMUTE_DIR(dir) \ std::cout << std::endl << "DEBUG -- PERM PRE" << std::endl; \ -svst1(pg1, (float64_t*)&debugreg.v, Chi_00); \ +svst1(pg1, &debugreg.v, Chi_00); \ std::cout << "Chi_00 -- " << debugreg << std::endl; \ -svst1(pg1, (float64_t*)&debugreg.v, Chi_01); \ +svst1(pg1, &debugreg.v, Chi_01); \ std::cout << "Chi_01 -- " << debugreg << std::endl; \ -svst1(pg1, (float64_t*)&debugreg.v, Chi_02); \ +svst1(pg1, &debugreg.v, Chi_02); \ std::cout << "Chi_02 -- " << debugreg << std::endl; \ -svst1(pg1, (float64_t*)&debugreg.v, Chi_10); \ +svst1(pg1, &debugreg.v, Chi_10); \ std::cout << "Chi_10 -- " << debugreg << std::endl; \ -svst1(pg1, (float64_t*)&debugreg.v, Chi_11); \ +svst1(pg1, &debugreg.v, Chi_11); \ std::cout << "Chi_11 -- " << debugreg << std::endl; \ -svst1(pg1, (float64_t*)&debugreg.v, Chi_12); \ +svst1(pg1, &debugreg.v, Chi_12); \ std::cout << "Chi_12 -- " << debugreg << std::endl; \ permute##dir(Chi_00,Chi_00);\ permute##dir(Chi_01,Chi_01);\ @@ -207,17 +207,17 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ permute##dir(Chi_11,Chi_11);\ permute##dir(Chi_12,Chi_12);\ std::cout << std::endl << "DEBUG -- PERM POST" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_00); \ + svst1(pg1, &debugreg.v, Chi_00); \ std::cout << "Chi_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_01); \ + svst1(pg1, &debugreg.v, Chi_01); \ std::cout << "Chi_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_02); \ + svst1(pg1, &debugreg.v, Chi_02); \ std::cout << "Chi_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_10); \ + svst1(pg1, &debugreg.v, Chi_10); \ std::cout << "Chi_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_11); \ + svst1(pg1, &debugreg.v, Chi_11); \ std::cout << "Chi_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_12); \ + svst1(pg1, &debugreg.v, Chi_12); \ std::cout << "Chi_12 -- " << debugreg << std::endl; // hspin(0)=fspin(0)+timesI(fspin(3)); @@ -230,17 +230,17 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ Chi_11 = Chimu_11+timesI(Chimu_21);\ Chi_12 = Chimu_12+timesI(Chimu_22);\ std::cout << std::endl << "DEBUG -- XP_PROJ" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_00); \ + svst1(pg1, &debugreg.v, Chi_00); \ std::cout << "Chi_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_01); \ + svst1(pg1, &debugreg.v, Chi_01); \ std::cout << "Chi_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_02); \ + svst1(pg1, &debugreg.v, Chi_02); \ std::cout << "Chi_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_10); \ + svst1(pg1, &debugreg.v, Chi_10); \ std::cout << "Chi_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_11); \ + svst1(pg1, &debugreg.v, Chi_11); \ std::cout << "Chi_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_12); \ + svst1(pg1, &debugreg.v, Chi_12); \ std::cout << "Chi_12 -- " << debugreg << std::endl; #define YP_PROJ \ @@ -251,17 +251,17 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ Chi_11 = Chimu_11+Chimu_21;\ Chi_12 = Chimu_12+Chimu_22;\ std::cout << std::endl << "DEBUG -- YP_PROJ" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_00); \ + svst1(pg1, &debugreg.v, Chi_00); \ std::cout << "Chi_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_01); \ + svst1(pg1, &debugreg.v, Chi_01); \ std::cout << "Chi_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_02); \ + svst1(pg1, &debugreg.v, Chi_02); \ std::cout << "Chi_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_10); \ + svst1(pg1, &debugreg.v, Chi_10); \ std::cout << "Chi_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_11); \ + svst1(pg1, &debugreg.v, Chi_11); \ std::cout << "Chi_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_12); \ + svst1(pg1, &debugreg.v, Chi_12); \ std::cout << "Chi_12 -- " << debugreg << std::endl; #define ZP_PROJ \ @@ -272,17 +272,17 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ Chi_11 = Chimu_11-timesI(Chimu_31); \ Chi_12 = Chimu_12-timesI(Chimu_32);\ std::cout << std::endl << "DEBUG -- ZP_PROJ" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_00); \ + svst1(pg1, &debugreg.v, Chi_00); \ std::cout << "Chi_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_01); \ + svst1(pg1, &debugreg.v, Chi_01); \ std::cout << "Chi_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_02); \ + svst1(pg1, &debugreg.v, Chi_02); \ std::cout << "Chi_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_10); \ + svst1(pg1, &debugreg.v, Chi_10); \ std::cout << "Chi_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_11); \ + svst1(pg1, &debugreg.v, Chi_11); \ std::cout << "Chi_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_12); \ + svst1(pg1, &debugreg.v, Chi_12); \ std::cout << "Chi_12 -- " << debugreg << std::endl; #define TP_PROJ \ @@ -293,17 +293,17 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ Chi_11 = Chimu_11+Chimu_31; \ Chi_12 = Chimu_12+Chimu_32;\ std::cout << std::endl << "DEBUG -- TP_PROJ" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_00); \ + svst1(pg1, &debugreg.v, Chi_00); \ std::cout << "Chi_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_01); \ + svst1(pg1, &debugreg.v, Chi_01); \ std::cout << "Chi_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_02); \ + svst1(pg1, &debugreg.v, Chi_02); \ std::cout << "Chi_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_10); \ + svst1(pg1, &debugreg.v, Chi_10); \ std::cout << "Chi_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_11); \ + svst1(pg1, &debugreg.v, Chi_11); \ std::cout << "Chi_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_12); \ + svst1(pg1, &debugreg.v, Chi_12); \ std::cout << "Chi_12 -- " << debugreg << std::endl; @@ -317,17 +317,17 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ Chi_11 = Chimu_11-timesI(Chimu_21);\ Chi_12 = Chimu_12-timesI(Chimu_22);\ std::cout << std::endl << "DEBUG -- XM_PROJ" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_00); \ + svst1(pg1, &debugreg.v, Chi_00); \ std::cout << "Chi_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_01); \ + svst1(pg1, &debugreg.v, Chi_01); \ std::cout << "Chi_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_02); \ + svst1(pg1, &debugreg.v, Chi_02); \ std::cout << "Chi_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_10); \ + svst1(pg1, &debugreg.v, Chi_10); \ std::cout << "Chi_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_11); \ + svst1(pg1, &debugreg.v, Chi_11); \ std::cout << "Chi_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_12); \ + svst1(pg1, &debugreg.v, Chi_12); \ std::cout << "Chi_12 -- " << debugreg << std::endl; #define YM_PROJ \ @@ -338,17 +338,17 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ Chi_11 = Chimu_11-Chimu_21;\ Chi_12 = Chimu_12-Chimu_22;\ std::cout << std::endl << "DEBUG -- YM_PROJ" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_00); \ + svst1(pg1, &debugreg.v, Chi_00); \ std::cout << "Chi_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_01); \ + svst1(pg1, &debugreg.v, Chi_01); \ std::cout << "Chi_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_02); \ + svst1(pg1, &debugreg.v, Chi_02); \ std::cout << "Chi_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_10); \ + svst1(pg1, &debugreg.v, Chi_10); \ std::cout << "Chi_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_11); \ + svst1(pg1, &debugreg.v, Chi_11); \ std::cout << "Chi_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_12); \ + svst1(pg1, &debugreg.v, Chi_12); \ std::cout << "Chi_12 -- " << debugreg << std::endl; #define ZM_PROJ \ @@ -359,17 +359,17 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ Chi_11 = Chimu_11+timesI(Chimu_31); \ Chi_12 = Chimu_12+timesI(Chimu_32);\ std::cout << std::endl << "DEBUG -- ZM_PROJ" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_00); \ + svst1(pg1, &debugreg.v, Chi_00); \ std::cout << "Chi_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_01); \ + svst1(pg1, &debugreg.v, Chi_01); \ std::cout << "Chi_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_02); \ + svst1(pg1, &debugreg.v, Chi_02); \ std::cout << "Chi_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_10); \ + svst1(pg1, &debugreg.v, Chi_10); \ std::cout << "Chi_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_11); \ + svst1(pg1, &debugreg.v, Chi_11); \ std::cout << "Chi_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_12); \ + svst1(pg1, &debugreg.v, Chi_12); \ std::cout << "Chi_12 -- " << debugreg << std::endl; #define TM_PROJ \ @@ -380,17 +380,17 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ Chi_11 = Chimu_11-Chimu_31; \ Chi_12 = Chimu_12-Chimu_32;\ std::cout << std::endl << "DEBUG -- TM_PROJ" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_00); \ + svst1(pg1, &debugreg.v, Chi_00); \ std::cout << "Chi_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_01); \ + svst1(pg1, &debugreg.v, Chi_01); \ std::cout << "Chi_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_02); \ + svst1(pg1, &debugreg.v, Chi_02); \ std::cout << "Chi_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_10); \ + svst1(pg1, &debugreg.v, Chi_10); \ std::cout << "Chi_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_11); \ + svst1(pg1, &debugreg.v, Chi_11); \ std::cout << "Chi_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, Chi_12); \ + svst1(pg1, &debugreg.v, Chi_12); \ std::cout << "Chi_12 -- " << debugreg << std::endl; // fspin(0)=hspin(0); @@ -411,29 +411,29 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ result_31 = timesMinusI(UChi_01);\ result_32 = timesMinusI(UChi_02);\ std::cout << std::endl << "DEBUG -- XP_RECON" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_00); \ + svst1(pg1, &debugreg.v, result_00); \ std::cout << "result_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_01); \ + svst1(pg1, &debugreg.v, result_01); \ std::cout << "result_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_02); \ + svst1(pg1, &debugreg.v, result_02); \ std::cout << "result_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_10); \ + svst1(pg1, &debugreg.v, result_10); \ std::cout << "result_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_11); \ + svst1(pg1, &debugreg.v, result_11); \ std::cout << "result_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_12); \ + svst1(pg1, &debugreg.v, result_12); \ std::cout << "result_12 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_20); \ + svst1(pg1, &debugreg.v, result_20); \ std::cout << "result_20 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_21); \ + svst1(pg1, &debugreg.v, result_21); \ std::cout << "result_21 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_22); \ + svst1(pg1, &debugreg.v, result_22); \ std::cout << "result_22 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_30); \ + svst1(pg1, &debugreg.v, result_30); \ std::cout << "result_30 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_31); \ + svst1(pg1, &debugreg.v, result_31); \ std::cout << "result_31 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_32); \ + svst1(pg1, &debugreg.v, result_32); \ std::cout << "result_32 -- " << debugreg << std::endl; #define XP_RECON_ACCUM\ @@ -450,29 +450,29 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ result_31-=timesI(UChi_01);\ result_32-=timesI(UChi_02);\ std::cout << std::endl << "DEBUG -- XP_RECON_ACCUM" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_00); \ + svst1(pg1, &debugreg.v, result_00); \ std::cout << "result_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_01); \ + svst1(pg1, &debugreg.v, result_01); \ std::cout << "result_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_02); \ + svst1(pg1, &debugreg.v, result_02); \ std::cout << "result_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_10); \ + svst1(pg1, &debugreg.v, result_10); \ std::cout << "result_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_11); \ + svst1(pg1, &debugreg.v, result_11); \ std::cout << "result_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_12); \ + svst1(pg1, &debugreg.v, result_12); \ std::cout << "result_12 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_20); \ + svst1(pg1, &debugreg.v, result_20); \ std::cout << "result_20 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_21); \ + svst1(pg1, &debugreg.v, result_21); \ std::cout << "result_21 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_22); \ + svst1(pg1, &debugreg.v, result_22); \ std::cout << "result_22 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_30); \ + svst1(pg1, &debugreg.v, result_30); \ std::cout << "result_30 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_31); \ + svst1(pg1, &debugreg.v, result_31); \ std::cout << "result_31 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_32); \ + svst1(pg1, &debugreg.v, result_32); \ std::cout << "result_32 -- " << debugreg << std::endl; #define XM_RECON\ @@ -489,29 +489,29 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ result_31 = timesI(UChi_01);\ result_32 = timesI(UChi_02);\ std::cout << std::endl << "DEBUG -- XM_RECON" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_00); \ + svst1(pg1, &debugreg.v, result_00); \ std::cout << "result_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_01); \ + svst1(pg1, &debugreg.v, result_01); \ std::cout << "result_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_02); \ + svst1(pg1, &debugreg.v, result_02); \ std::cout << "result_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_10); \ + svst1(pg1, &debugreg.v, result_10); \ std::cout << "result_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_11); \ + svst1(pg1, &debugreg.v, result_11); \ std::cout << "result_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_12); \ + svst1(pg1, &debugreg.v, result_12); \ std::cout << "result_12 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_20); \ + svst1(pg1, &debugreg.v, result_20); \ std::cout << "result_20 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_21); \ + svst1(pg1, &debugreg.v, result_21); \ std::cout << "result_21 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_22); \ + svst1(pg1, &debugreg.v, result_22); \ std::cout << "result_22 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_30); \ + svst1(pg1, &debugreg.v, result_30); \ std::cout << "result_30 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_31); \ + svst1(pg1, &debugreg.v, result_31); \ std::cout << "result_31 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_32); \ + svst1(pg1, &debugreg.v, result_32); \ std::cout << "result_32 -- " << debugreg << std::endl; #define XM_RECON_ACCUM\ @@ -528,29 +528,29 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ result_31+= timesI(UChi_01);\ result_32+= timesI(UChi_02);\ std::cout << std::endl << "DEBUG -- XM_RECON_ACCUM" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_00); \ + svst1(pg1, &debugreg.v, result_00); \ std::cout << "result_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_01); \ + svst1(pg1, &debugreg.v, result_01); \ std::cout << "result_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_02); \ + svst1(pg1, &debugreg.v, result_02); \ std::cout << "result_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_10); \ + svst1(pg1, &debugreg.v, result_10); \ std::cout << "result_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_11); \ + svst1(pg1, &debugreg.v, result_11); \ std::cout << "result_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_12); \ + svst1(pg1, &debugreg.v, result_12); \ std::cout << "result_12 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_20); \ + svst1(pg1, &debugreg.v, result_20); \ std::cout << "result_20 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_21); \ + svst1(pg1, &debugreg.v, result_21); \ std::cout << "result_21 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_22); \ + svst1(pg1, &debugreg.v, result_22); \ std::cout << "result_22 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_30); \ + svst1(pg1, &debugreg.v, result_30); \ std::cout << "result_30 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_31); \ + svst1(pg1, &debugreg.v, result_31); \ std::cout << "result_31 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_32); \ + svst1(pg1, &debugreg.v, result_32); \ std::cout << "result_32 -- " << debugreg << std::endl; #define YP_RECON_ACCUM\ @@ -567,29 +567,29 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ result_31-= UChi_01;\ result_32-= UChi_02;\ std::cout << std::endl << "DEBUG -- YP_RECON_ACCUM" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_00); \ + svst1(pg1, &debugreg.v, result_00); \ std::cout << "result_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_01); \ + svst1(pg1, &debugreg.v, result_01); \ std::cout << "result_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_02); \ + svst1(pg1, &debugreg.v, result_02); \ std::cout << "result_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_10); \ + svst1(pg1, &debugreg.v, result_10); \ std::cout << "result_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_11); \ + svst1(pg1, &debugreg.v, result_11); \ std::cout << "result_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_12); \ + svst1(pg1, &debugreg.v, result_12); \ std::cout << "result_12 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_20); \ + svst1(pg1, &debugreg.v, result_20); \ std::cout << "result_20 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_21); \ + svst1(pg1, &debugreg.v, result_21); \ std::cout << "result_21 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_22); \ + svst1(pg1, &debugreg.v, result_22); \ std::cout << "result_22 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_30); \ + svst1(pg1, &debugreg.v, result_30); \ std::cout << "result_30 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_31); \ + svst1(pg1, &debugreg.v, result_31); \ std::cout << "result_31 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_32); \ + svst1(pg1, &debugreg.v, result_32); \ std::cout << "result_32 -- " << debugreg << std::endl; #define YM_RECON_ACCUM\ @@ -606,29 +606,29 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ result_31+= UChi_01;\ result_32+= UChi_02;\ std::cout << std::endl << "DEBUG -- YM_RECON_ACCUM" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_00); \ + svst1(pg1, &debugreg.v, result_00); \ std::cout << "result_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_01); \ + svst1(pg1, &debugreg.v, result_01); \ std::cout << "result_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_02); \ + svst1(pg1, &debugreg.v, result_02); \ std::cout << "result_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_10); \ + svst1(pg1, &debugreg.v, result_10); \ std::cout << "result_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_11); \ + svst1(pg1, &debugreg.v, result_11); \ std::cout << "result_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_12); \ + svst1(pg1, &debugreg.v, result_12); \ std::cout << "result_12 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_20); \ + svst1(pg1, &debugreg.v, result_20); \ std::cout << "result_20 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_21); \ + svst1(pg1, &debugreg.v, result_21); \ std::cout << "result_21 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_22); \ + svst1(pg1, &debugreg.v, result_22); \ std::cout << "result_22 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_30); \ + svst1(pg1, &debugreg.v, result_30); \ std::cout << "result_30 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_31); \ + svst1(pg1, &debugreg.v, result_31); \ std::cout << "result_31 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_32); \ + svst1(pg1, &debugreg.v, result_32); \ std::cout << "result_32 -- " << debugreg << std::endl; #define ZP_RECON_ACCUM\ @@ -645,29 +645,29 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ result_31+= timesI(UChi_11); \ result_32+= timesI(UChi_12);\ std::cout << std::endl << "DEBUG -- ZP_RECON_ACCUM" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_00); \ + svst1(pg1, &debugreg.v, result_00); \ std::cout << "result_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_01); \ + svst1(pg1, &debugreg.v, result_01); \ std::cout << "result_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_02); \ + svst1(pg1, &debugreg.v, result_02); \ std::cout << "result_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_10); \ + svst1(pg1, &debugreg.v, result_10); \ std::cout << "result_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_11); \ + svst1(pg1, &debugreg.v, result_11); \ std::cout << "result_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_12); \ + svst1(pg1, &debugreg.v, result_12); \ std::cout << "result_12 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_20); \ + svst1(pg1, &debugreg.v, result_20); \ std::cout << "result_20 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_21); \ + svst1(pg1, &debugreg.v, result_21); \ std::cout << "result_21 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_22); \ + svst1(pg1, &debugreg.v, result_22); \ std::cout << "result_22 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_30); \ + svst1(pg1, &debugreg.v, result_30); \ std::cout << "result_30 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_31); \ + svst1(pg1, &debugreg.v, result_31); \ std::cout << "result_31 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_32); \ + svst1(pg1, &debugreg.v, result_32); \ std::cout << "result_32 -- " << debugreg << std::endl; #define ZM_RECON_ACCUM\ @@ -684,29 +684,29 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ result_31-= timesI(UChi_11); \ result_32-= timesI(UChi_12);\ std::cout << std::endl << "DEBUG -- ZM_RECON_ACCUM" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_00); \ + svst1(pg1, &debugreg.v, result_00); \ std::cout << "result_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_01); \ + svst1(pg1, &debugreg.v, result_01); \ std::cout << "result_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_02); \ + svst1(pg1, &debugreg.v, result_02); \ std::cout << "result_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_10); \ + svst1(pg1, &debugreg.v, result_10); \ std::cout << "result_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_11); \ + svst1(pg1, &debugreg.v, result_11); \ std::cout << "result_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_12); \ + svst1(pg1, &debugreg.v, result_12); \ std::cout << "result_12 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_20); \ + svst1(pg1, &debugreg.v, result_20); \ std::cout << "result_20 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_21); \ + svst1(pg1, &debugreg.v, result_21); \ std::cout << "result_21 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_22); \ + svst1(pg1, &debugreg.v, result_22); \ std::cout << "result_22 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_30); \ + svst1(pg1, &debugreg.v, result_30); \ std::cout << "result_30 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_31); \ + svst1(pg1, &debugreg.v, result_31); \ std::cout << "result_31 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_32); \ + svst1(pg1, &debugreg.v, result_32); \ std::cout << "result_32 -- " << debugreg << std::endl; #define TP_RECON_ACCUM\ @@ -723,29 +723,29 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ result_31+= UChi_11; \ result_32+= UChi_12;\ std::cout << std::endl << "DEBUG -- TP_RECON_ACCUM" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_00); \ + svst1(pg1, &debugreg.v, result_00); \ std::cout << "result_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_01); \ + svst1(pg1, &debugreg.v, result_01); \ std::cout << "result_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_02); \ + svst1(pg1, &debugreg.v, result_02); \ std::cout << "result_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_10); \ + svst1(pg1, &debugreg.v, result_10); \ std::cout << "result_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_11); \ + svst1(pg1, &debugreg.v, result_11); \ std::cout << "result_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_12); \ + svst1(pg1, &debugreg.v, result_12); \ std::cout << "result_12 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_20); \ + svst1(pg1, &debugreg.v, result_20); \ std::cout << "result_20 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_21); \ + svst1(pg1, &debugreg.v, result_21); \ std::cout << "result_21 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_22); \ + svst1(pg1, &debugreg.v, result_22); \ std::cout << "result_22 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_30); \ + svst1(pg1, &debugreg.v, result_30); \ std::cout << "result_30 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_31); \ + svst1(pg1, &debugreg.v, result_31); \ std::cout << "result_31 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_32); \ + svst1(pg1, &debugreg.v, result_32); \ std::cout << "result_32 -- " << debugreg << std::endl; #define TM_RECON_ACCUM\ @@ -762,29 +762,29 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ result_31-= UChi_11; \ result_32-= UChi_12;\ std::cout << std::endl << "DEBUG -- TM_RECON_ACCUM" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_00); \ + svst1(pg1, &debugreg.v, result_00); \ std::cout << "result_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_01); \ + svst1(pg1, &debugreg.v, result_01); \ std::cout << "result_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_02); \ + svst1(pg1, &debugreg.v, result_02); \ std::cout << "result_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_10); \ + svst1(pg1, &debugreg.v, result_10); \ std::cout << "result_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_11); \ + svst1(pg1, &debugreg.v, result_11); \ std::cout << "result_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_12); \ + svst1(pg1, &debugreg.v, result_12); \ std::cout << "result_12 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_20); \ + svst1(pg1, &debugreg.v, result_20); \ std::cout << "result_20 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_21); \ + svst1(pg1, &debugreg.v, result_21); \ std::cout << "result_21 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_22); \ + svst1(pg1, &debugreg.v, result_22); \ std::cout << "result_22 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_30); \ + svst1(pg1, &debugreg.v, result_30); \ std::cout << "result_30 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_31); \ + svst1(pg1, &debugreg.v, result_31); \ std::cout << "result_31 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_32); \ + svst1(pg1, &debugreg.v, result_32); \ std::cout << "result_32 -- " << debugreg << std::endl; #define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON) \ @@ -866,29 +866,29 @@ std::cout << "Chi_12 -- " << debugreg << std::endl; \ ref()(3)(1)+=result_31; \ ref()(3)(2)+=result_32; \ std::cout << std::endl << "DEBUG -- RESULT" << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_00); \ + svst1(pg1, &debugreg.v, result_00); \ std::cout << "result_00 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_01); \ + svst1(pg1, &debugreg.v, result_01); \ std::cout << "result_01 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_02); \ + svst1(pg1, &debugreg.v, result_02); \ std::cout << "result_02 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_10); \ + svst1(pg1, &debugreg.v, result_10); \ std::cout << "result_10 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_11); \ + svst1(pg1, &debugreg.v, result_11); \ std::cout << "result_11 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_12); \ + svst1(pg1, &debugreg.v, result_12); \ std::cout << "result_12 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_20); \ + svst1(pg1, &debugreg.v, result_20); \ std::cout << "result_20 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_21); \ + svst1(pg1, &debugreg.v, result_21); \ std::cout << "result_21 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_22); \ + svst1(pg1, &debugreg.v, result_22); \ std::cout << "result_22 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_30); \ + svst1(pg1, &debugreg.v, result_30); \ std::cout << "result_30 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_31); \ + svst1(pg1, &debugreg.v, result_31); \ std::cout << "result_31 -- " << debugreg << std::endl; \ - svst1(pg1, (float64_t*)&debugreg.v, result_32); \ + svst1(pg1, &debugreg.v, result_32); \ std::cout << "result_32 -- " << debugreg << std::endl; \ }