portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-06 20:25:56 +01:00
Code Releases Activity

revised sU addressing scheme

Browse Source
This commit is contained in:
nmeyer-ur 2020-04-09 14:44:25 +02:00
parent 9f224a1647
commit 326de36467
4 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Grid/simd/Fujitsu_A64FX_asm_double.h
View File

@ -283,7 +283,7 @@ asm ( \
// MULT_2SPIN
#define MULT_2SPIN_A64FXd(A) \
{ \
const auto & ref(U[sU][A]); \
const auto & ref(U[sU](A)); \
asm ( \
"ld1d { z24.d }, p5/z, [%[fetchptr], -6, mul vl] \n\t" \
"ld1d { z25.d }, p5/z, [%[fetchptr], -3, mul vl] \n\t" \

2
Grid/simd/Fujitsu_A64FX_asm_single.h
View File

@ -295,7 +295,7 @@ asm ( \
// MULT_2SPIN
#define MULT_2SPIN_A64FXf(A) \
{ \
const auto & ref(U[sU][A]); \
const auto & ref(U[sU](A)); \
asm ( \
"ld1d { z24.f }, p5/z, [%[fetchptr], -6, mul vl] \n\t" \
"ld1d { z25.f }, p5/z, [%[fetchptr], -3, mul vl] \n\t" \

2
Grid/simd/Fujitsu_A64FX_intrin_double.h
View File

@ -261,7 +261,7 @@ Author: Nils Meyer <nils.meyer@ur.de>
// MULT_2SPIN
#define MULT_2SPIN_A64FXd(A) \
{ \
const auto & ref(U[sU][A]); \
const auto & ref(U[sU](A)); \
U_00 = svld1(pg1, (float64_t*)((uint64_t)&ref[2][0] + -6 * 64)); \
U_10 = svld1(pg1, (float64_t*)((uint64_t)&ref[2][0] + -3 * 64)); \
U_20 = svld1(pg1, (float64_t*)((uint64_t)&ref[2][0] + 0 * 64)); \

2
Grid/simd/Fujitsu_A64FX_intrin_single.h
View File

@ -268,7 +268,7 @@ Author: Nils Meyer <nils.meyer@ur.de>
// MULT_2SPIN
#define MULT_2SPIN_A64FXf(A) \
{ \
const auto & ref(U[sU][A]); \
const auto & ref(U[sU](A)); \
U_00 = svld1(pg1, (float32_t*)((uint64_t)&ref[2][0] + -6 * 64)); \
U_10 = svld1(pg1, (float32_t*)((uint64_t)&ref[2][0] + -3 * 64)); \
U_20 = svld1(pg1, (float32_t*)((uint64_t)&ref[2][0] + 0 * 64)); \