Optimise sswap for POWER9

Use lxvd2x instruction instead of lxvw4x.
lxvd2x performs far better on the new POWER architecture than lxvw4x.
This commit is contained in:
Matt Brown 2017-06-14 16:23:20 +10:00
parent be55f96cbd
commit 6f4eca5ea4
1 changed files with 32 additions and 32 deletions

View File

@ -42,43 +42,43 @@ static void sswap_kernel_32 (long n, float *x, float *y)
".p2align 5 \n" ".p2align 5 \n"
"1: \n\t" "1: \n\t"
"lxvw4x 32, 0, %4 \n\t" "lxvd2x 32, 0, %4 \n\t"
"lxvw4x 33, %5, %4 \n\t" "lxvd2x 33, %5, %4 \n\t"
"lxvw4x 34, %6, %4 \n\t" "lxvd2x 34, %6, %4 \n\t"
"lxvw4x 35, %7, %4 \n\t" "lxvd2x 35, %7, %4 \n\t"
"lxvw4x 36, %8, %4 \n\t" "lxvd2x 36, %8, %4 \n\t"
"lxvw4x 37, %9, %4 \n\t" "lxvd2x 37, %9, %4 \n\t"
"lxvw4x 38, %10, %4 \n\t" "lxvd2x 38, %10, %4 \n\t"
"lxvw4x 39, %11, %4 \n\t" "lxvd2x 39, %11, %4 \n\t"
"lxvw4x 40, 0, %3 \n\t" "lxvd2x 40, 0, %3 \n\t"
"lxvw4x 41, %5, %3 \n\t" "lxvd2x 41, %5, %3 \n\t"
"lxvw4x 42, %6, %3 \n\t" "lxvd2x 42, %6, %3 \n\t"
"lxvw4x 43, %7, %3 \n\t" "lxvd2x 43, %7, %3 \n\t"
"lxvw4x 44, %8, %3 \n\t" "lxvd2x 44, %8, %3 \n\t"
"lxvw4x 45, %9, %3 \n\t" "lxvd2x 45, %9, %3 \n\t"
"lxvw4x 46, %10, %3 \n\t" "lxvd2x 46, %10, %3 \n\t"
"lxvw4x 47, %11, %3 \n\t" "lxvd2x 47, %11, %3 \n\t"
"stxvw4x 32, 0, %3 \n\t" "stxvd2x 32, 0, %3 \n\t"
"stxvw4x 33, %5, %3 \n\t" "stxvd2x 33, %5, %3 \n\t"
"stxvw4x 34, %6, %3 \n\t" "stxvd2x 34, %6, %3 \n\t"
"stxvw4x 35, %7, %3 \n\t" "stxvd2x 35, %7, %3 \n\t"
"stxvw4x 36, %8, %3 \n\t" "stxvd2x 36, %8, %3 \n\t"
"stxvw4x 37, %9, %3 \n\t" "stxvd2x 37, %9, %3 \n\t"
"stxvw4x 38, %10, %3 \n\t" "stxvd2x 38, %10, %3 \n\t"
"stxvw4x 39, %11, %3 \n\t" "stxvd2x 39, %11, %3 \n\t"
"addi %3, %3, 128 \n\t" "addi %3, %3, 128 \n\t"
"stxvw4x 40, 0, %4 \n\t" "stxvd2x 40, 0, %4 \n\t"
"stxvw4x 41, %5, %4 \n\t" "stxvd2x 41, %5, %4 \n\t"
"stxvw4x 42, %6, %4 \n\t" "stxvd2x 42, %6, %4 \n\t"
"stxvw4x 43, %7, %4 \n\t" "stxvd2x 43, %7, %4 \n\t"
"stxvw4x 44, %8, %4 \n\t" "stxvd2x 44, %8, %4 \n\t"
"stxvw4x 45, %9, %4 \n\t" "stxvd2x 45, %9, %4 \n\t"
"stxvw4x 46, %10, %4 \n\t" "stxvd2x 46, %10, %4 \n\t"
"stxvw4x 47, %11, %4 \n\t" "stxvd2x 47, %11, %4 \n\t"
"addi %4, %4, 128 \n\t" "addi %4, %4, 128 \n\t"