portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-09 21:50:45 +01:00
Code Releases Activity

fixed-size clean up

Browse Source
This commit is contained in:
nmeyer-ur 2020-05-09 22:53:42 +02:00
parent 291ee8c3d0
commit b2fd8b993a
1 changed files with 4 additions and 39 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

43
Grid/simd/Grid_a64fx-fixedsize.h
View File

@ -6,8 +6,8 @@
Copyright (C) 2020 Copyright (C) 2020
Authors: Nils Meyer <nils.meyer@ur.de> Regensburg University Author: Nils Meyer <nils.meyer@ur.de> Regensburg University
Richard Sandiford <richard.sandiford@arm.com> Arm Author: Richard Sandiford <richard.sandiford@arm.com> Arm
This program is free software; you can redistribute it and/or modify This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by it under the terms of the GNU General Public License as published by
@ -101,29 +101,14 @@ template <>
struct acle<double>{ struct acle<double>{
static inline pred pg1(){return svptrue_b64();} static inline pred pg1(){return svptrue_b64();}
static inline lutd tbl_swap(){ static inline lutd tbl_swap(){
/*
const uint64_t t[8] = {1, 0, 3, 2, 5, 4, 7, 6};
pred pg1 = svptrue_b64();
return svld1(pg1, t);
*/
const ulutd t = { .s = {1, 0, 3, 2, 5, 4, 7, 6} }; const ulutd t = { .s = {1, 0, 3, 2, 5, 4, 7, 6} };
return t.v; return t.v;
} }
static inline lutd tbl0(){ static inline lutd tbl0(){
/*
const uint64_t t[8] = {4, 5, 6, 7, 0, 1, 2, 3};
pred pg1 = svptrue_b64();
return svld1(pg1, t);
*/
const ulutd t = { .s = {4, 5, 6, 7, 0, 1, 2, 3} }; const ulutd t = { .s = {4, 5, 6, 7, 0, 1, 2, 3} };
return t.v; return t.v;
} }
static inline lutd tbl1(){ static inline lutd tbl1(){
/*
const uint64_t t[8] = {2, 3, 0, 1, 6, 7, 4, 5};
pred pg1 = svptrue_b64();
return svld1(pg1, t);
*/
const ulutd t = { .s = {2, 3, 0, 1, 6, 7, 4, 5} }; const ulutd t = { .s = {2, 3, 0, 1, 6, 7, 4, 5} };
return t.v; return t.v;
} }
@ -137,38 +122,18 @@ struct acle<float>{
static inline pred pg1(){return svptrue_b32();} static inline pred pg1(){return svptrue_b32();}
// exchange neighboring elements // exchange neighboring elements
static inline lutf tbl_swap(){ static inline lutf tbl_swap(){
/*
const uint32_t t[16] = {1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14};
pred pg1 = svptrue_b32();
return svld1(pg1, t);
*/
const ulutf t = { .s = {1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14} }; const ulutf t = { .s = {1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14} };
return t.v; return t.v;
} }
static inline lutf tbl0(){ static inline lutf tbl0(){
/*
const uint32_t t[16] = {8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7};
pred pg1 = svptrue_b32();
return svld1(pg1, t);
*/
const ulutf t = { .s = {8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7} }; const ulutf t = { .s = {8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7} };
return t.v; return t.v;
} }
static inline lutf tbl1(){ static inline lutf tbl1(){
/*
const uint32_t t[16] = {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11};
pred pg1 = svptrue_b32();
return svld1(pg1, t);
*/
const ulutf t = { .s = {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11} }; const ulutf t = { .s = {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11} };
return t.v; return t.v;
} }
static inline lutf tbl2(){ static inline lutf tbl2(){
/*
const uint32_t t[16] = {2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13};
pred pg1 = svptrue_b32();
return svld1(pg1, t);
*/
const ulutf t = { .s = {2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13} }; const ulutf t = { .s = {2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13} };
return t.v; return t.v;
} }
@ -400,8 +365,8 @@ struct MultComplex{
pred pg1 = acle<double>::pg1(); pred pg1 = acle<double>::pg1();
vecd z = acle<double>::zero(); vecd z = acle<double>::zero();
// using FCMLA // using FCMLA
vecd r_v = svcmla_x(pg1, z, a, b, 90); vecd r_v = svcmla_x(pg1, z, a, b, 0);
return svcmla_x(pg1, r_v, a, b, 0); return svcmla_x(pg1, r_v, a, b, 90);
} }
}; };