portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-09-20 17:25:37 +01:00
Code Releases Activity

unions for tables eliminate explicit loads, gcc does not complain

Browse Source
This commit is contained in:
nmeyer-ur 2020-05-09 21:21:57 +02:00
parent 55a55660cb
commit e1a5b3ea49
1 changed files with 39 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
Grid/simd/Grid_a64fx-fixedsize.h
View File

@ -70,6 +70,17 @@ typedef svuint64_t lutd __attribute__((arm_sve_vector_bits(512))); // LUTs for
NAMESPACE_BEGIN(Grid); NAMESPACE_BEGIN(Grid);
NAMESPACE_BEGIN(Optimization); NAMESPACE_BEGIN(Optimization);
// convenience union types for tables eliminate loads
union ulutf {
lutf v;
uint32_t s[16];
};
union ulutd {
lutd v;
uint64_t s[8];
};
template <typename T> template <typename T>
struct acle{}; struct acle{};
@ -77,19 +88,31 @@ 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}; const uint64_t t[8] = {1, 0, 3, 2, 5, 4, 7, 6};
pred pg1 = svptrue_b64(); pred pg1 = svptrue_b64();
return svld1(pg1, t); return svld1(pg1, t);
*/
const ulutd t = { .s = {1, 0, 3, 2, 5, 4, 7, 6} };
return t.v;
} }
static inline lutd tbl0(){ static inline lutd tbl0(){
/*
const uint64_t t[8] = {4, 5, 6, 7, 0, 1, 2, 3}; const uint64_t t[8] = {4, 5, 6, 7, 0, 1, 2, 3};
pred pg1 = svptrue_b64(); pred pg1 = svptrue_b64();
return svld1(pg1, t); return svld1(pg1, t);
*/
const ulutd t = { .s = {4, 5, 6, 7, 0, 1, 2, 3} };
return t.v;
} }
static inline lutd tbl1(){ static inline lutd tbl1(){
/*
const uint64_t t[8] = {2, 3, 0, 1, 6, 7, 4, 5}; const uint64_t t[8] = {2, 3, 0, 1, 6, 7, 4, 5};
pred pg1 = svptrue_b64(); pred pg1 = svptrue_b64();
return svld1(pg1, t); return svld1(pg1, t);
*/
const ulutd t = { .s = {2, 3, 0, 1, 6, 7, 4, 5} };
return t.v;
} }
static inline pred pg_even(){return svzip1_b64(svptrue_b64(), svpfalse_b());} static inline pred pg_even(){return svzip1_b64(svptrue_b64(), svpfalse_b());}
static inline pred pg_odd() {return svzip1_b64(svpfalse_b(), svptrue_b64());} static inline pred pg_odd() {return svzip1_b64(svpfalse_b(), svptrue_b64());}
@ -101,24 +124,40 @@ 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}; 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(); pred pg1 = svptrue_b32();
return svld1(pg1, t); return svld1(pg1, t);
*/
const ulutf t = { .s = {1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14} };
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}; 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(); pred pg1 = svptrue_b32();
return svld1(pg1, t); return svld1(pg1, t);
*/
const ulutf t = { .s = {8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7} };
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}; 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(); pred pg1 = svptrue_b32();
return svld1(pg1, t); return svld1(pg1, t);
*/
const ulutf t = { .s = {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11} };
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}; 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(); pred pg1 = svptrue_b32();
return svld1(pg1, t); return svld1(pg1, t);
*/
const ulutf t = { .s = {2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13} };
return t.v;
} }
static inline pred pg_even(){return svzip1_b32(svptrue_b32(), svpfalse_b());} static inline pred pg_even(){return svzip1_b32(svptrue_b32(), svpfalse_b());}
static inline pred pg_odd() {return svzip1_b32(svpfalse_b(), svptrue_b32());} static inline pred pg_odd() {return svzip1_b32(svpfalse_b(), svptrue_b32());}