mirror of
https://github.com/wownero/RandomX.git
synced 2026-10-21 01:08:48 +02:00
initBlock asm version (disabled)
This commit is contained in:
tevador
@@ -0,0 +1,155 @@
|
||||
prefetcht0 byte ptr [rbp]
|
||||
xor r9, r9
|
||||
xor r10, r10
|
||||
xor r11, r11
|
||||
xor r12, r12
|
||||
xor r13, r13
|
||||
xor r14, r14
|
||||
xor r15, r15
|
||||
initBlock_loop:
|
||||
;# c0
|
||||
mov rbx, r8
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r8+r9]
|
||||
call squareHash
|
||||
mov r9, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
;# c1
|
||||
mov rbx, r9
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r9+r10]
|
||||
call squareHash
|
||||
mov r10, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
;# c2
|
||||
mov rbx, r10
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r10+r11]
|
||||
call squareHash
|
||||
mov r11, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
;# c3
|
||||
mov rbx, r11
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r11+r12]
|
||||
call squareHash
|
||||
mov r12, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
;# c4
|
||||
mov rbx, r12
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r12+r13]
|
||||
call squareHash
|
||||
mov r13, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
;# c5
|
||||
mov rbx, r13
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r13+r14]
|
||||
call squareHash
|
||||
mov r14, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
;# c6
|
||||
mov rbx, r14
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r14+r15]
|
||||
call squareHash
|
||||
mov r15, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
;# c7
|
||||
mov rbx, r15
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r15+r8]
|
||||
call squareHash
|
||||
mov r8, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
sub rsi, 1
|
||||
jnz initBlock_loop
|
||||
mov qword ptr [rbp+0], r8
|
||||
mov qword ptr [rbp+8], r9
|
||||
mov qword ptr [rbp+16], r10
|
||||
mov qword ptr [rbp+24], r11
|
||||
mov qword ptr [rbp+32], r12
|
||||
mov qword ptr [rbp+40], r13
|
||||
mov qword ptr [rbp+48], r14
|
||||
mov qword ptr [rbp+56], r15
|
||||
+2
-2
@@ -40,7 +40,7 @@ along with RandomX. If not, see<http://www.gnu.org/licenses/>.
|
||||
|
||||
namespace RandomX {
|
||||
|
||||
#if !defined(_M_X64)
|
||||
#if true //RANDOMX_ARGON_GROWTH != 0 || (!defined(_M_X64) && !defined(__x86_64__))
|
||||
static FORCE_INLINE uint8_t* selectMixBlock(const Cache& cache, uint64_t& currentIndex, uint64_t& nextIndex) {
|
||||
uint8_t* mixBlock;
|
||||
if (RANDOMX_ARGON_GROWTH == 0) {
|
||||
@@ -75,7 +75,7 @@ namespace RandomX {
|
||||
|
||||
uint8_t* mixBlock;
|
||||
|
||||
for (auto i = 0; i < RANDOMX_CACHE_ACCESSES / 8; ++i) {
|
||||
for (auto i = 0; i < iterations; ++i) {
|
||||
mixBlock = selectMixBlock(cache, c0, c1);
|
||||
mixCache(mixBlock, c0, c1, c2, c3, c4, c5, c6, c7);
|
||||
|
||||
|
||||
+1
-1
@@ -25,7 +25,7 @@ along with RandomX. If not, see<http://www.gnu.org/licenses/>.
|
||||
|
||||
namespace RandomX {
|
||||
|
||||
#if defined(_M_X64)
|
||||
#if false //RANDOMX_ARGON_GROWTH == 0 && (defined(_M_X64) || defined(__x86_64__))
|
||||
extern "C"
|
||||
#endif
|
||||
void initBlock(const Cache& cache, uint8_t* out, uint64_t blockNumber, unsigned iterations);
|
||||
|
||||
@@ -10,8 +10,31 @@
|
||||
#define DECL(x) x
|
||||
#endif
|
||||
|
||||
#include "configuration.h"
|
||||
|
||||
.global DECL(squareHash)
|
||||
.global DECL(initBlock)
|
||||
|
||||
DECL(squareHash):
|
||||
mov rcx, rdi
|
||||
#include "asm/squareHash.inc"
|
||||
|
||||
DECL(initBlock):
|
||||
push rbx
|
||||
push rbp
|
||||
push r12
|
||||
push r13
|
||||
push r14
|
||||
push r15
|
||||
mov rdi, qword ptr [rdi]
|
||||
mov rbp, rsi
|
||||
mov r8, rdx
|
||||
mov rsi, rcx
|
||||
#include "asm/initBlock.inc"
|
||||
pop r15
|
||||
pop r14
|
||||
pop r13
|
||||
pop r12
|
||||
pop rbp
|
||||
pop rbx
|
||||
ret
|
||||
+2
-156
@@ -22,165 +22,11 @@ initBlock PROC
|
||||
push r13
|
||||
push r14
|
||||
push r15
|
||||
mov rsi, r9
|
||||
mov rdi, qword ptr [rcx]
|
||||
mov rbp, rdx
|
||||
prefetcht0 byte ptr [rbp]
|
||||
; r8 = blockNumber
|
||||
xor r9, r9
|
||||
xor r10, r10
|
||||
xor r11, r11
|
||||
xor r12, r12
|
||||
xor r13, r13
|
||||
xor r14, r14
|
||||
xor r15, r15
|
||||
initBlock_loop:
|
||||
; c0
|
||||
mov rbx, r8
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r8+r9]
|
||||
call squareHash
|
||||
mov r9, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
; c1
|
||||
mov rbx, r9
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r9+r10]
|
||||
call squareHash
|
||||
mov r10, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
; c2
|
||||
mov rbx, r10
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r10+r11]
|
||||
call squareHash
|
||||
mov r11, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
; c3
|
||||
mov rbx, r11
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r11+r12]
|
||||
call squareHash
|
||||
mov r12, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
; c4
|
||||
mov rbx, r12
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r12+r13]
|
||||
call squareHash
|
||||
mov r13, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
; c5
|
||||
mov rbx, r13
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r13+r14]
|
||||
call squareHash
|
||||
mov r14, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
; c6
|
||||
mov rbx, r14
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r14+r15]
|
||||
call squareHash
|
||||
mov r15, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
; c7
|
||||
mov rbx, r15
|
||||
and rbx, 4194303
|
||||
shl rbx, 6
|
||||
add rbx, rdi
|
||||
prefetchnta byte ptr [rbx]
|
||||
lea rcx, [r15+r8]
|
||||
call squareHash
|
||||
mov r8, rax
|
||||
xor r8, qword ptr [rbx+0]
|
||||
xor r9, qword ptr [rbx+8]
|
||||
xor r10, qword ptr [rbx+16]
|
||||
xor r11, qword ptr [rbx+24]
|
||||
xor r12, qword ptr [rbx+32]
|
||||
xor r13, qword ptr [rbx+40]
|
||||
xor r14, qword ptr [rbx+48]
|
||||
xor r15, qword ptr [rbx+56]
|
||||
sub rsi, 1
|
||||
jnz initBlock_loop
|
||||
mov qword ptr [rbp+0], r8
|
||||
mov qword ptr [rbp+8], r9
|
||||
mov qword ptr [rbp+16], r10
|
||||
mov qword ptr [rbp+24], r11
|
||||
mov qword ptr [rbp+32], r12
|
||||
mov qword ptr [rbp+40], r13
|
||||
mov qword ptr [rbp+48], r14
|
||||
mov qword ptr [rbp+56], r15
|
||||
mov rsi, r9
|
||||
include asm/initBlock.inc
|
||||
pop r15
|
||||
pop r14
|
||||
pop r13
|
||||
|
||||
Reference in New Issue
Block a user