fzorb/REDACTED-rig
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Another dataset AVX2 init speedup (+3.8% faster on Zen3)

Browse source
This commit is contained in:
SChernykh 2020-12-19 19:46:31 +01:00
parent 633aaccd9c
commit 5efd00abec
4 changed files with 8 additions and 10 deletions
Download patch file Download diff file Expand all files Collapse all files

7
src/crypto/randomx/jit_compiler_x86_static.S
View file

@ -212,7 +212,7 @@ DECL(randomx_dataset_init_avx2_prologue):
mov rbp, rdx ;# block index
push rcx ;# max. block index
#endif
sub rsp, 32
sub rsp, 40
jmp randomx_dataset_init_avx2_prologue_loop_begin
#include "asm/program_sshash_avx2_constants.inc"
@ -240,9 +240,8 @@ randomx_dataset_init_avx2_prologue_loop_begin:
xor r15, r8
;# init AVX registers (lanes 1-4)
vpxor ymm0, ymm0, ymm0
movq xmm0, rbp
vpbroadcastq ymm0, xmm0
mov qword ptr [rsp+32], rbp
vbroadcastsd ymm0, qword ptr [rsp+32]
vpaddq ymm0, ymm0, ymmword ptr [r0_avx2_increments+rip]
;# ymm0 *= r0_avx2_mul