fix for #2721 #2723
fix for #2721 #2723
Conversation
| self.reserve_unused_temp_gpr(GPR::R9); | ||
| self.move_location(Size::S64, loc, Location::GPR(GPR::R9)); | ||
| self.move_location( | ||
| self.assembler.emit_xchg( |
There was a problem hiding this comment.
On x86, the xchg instruction always does an atomic swap, which is very slow. Try to use a normal mov instead.
There was a problem hiding this comment.
I know, but I need to save the value of the register.
There was a problem hiding this comment.
It's probably cheaper to add another pair of push/pop for r9 than to use xchg:
push rax (dummy)
push r9
mov r9, #imm64
mov [sp + 8], r9
pop r9
There was a problem hiding this comment.
No it's not.
I have done this simple program:
#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <time.h>
int main(int argc, const char* argv)
{
struct timespec t1, t2;
clock_gettime(CLOCK_MONOTONIC, &t1);
#define NUM 100000
for(int i=0; i<NUM; ++i)
asm volatile ("\n\
push %%r9\n\
mov $0x123456789abcdef, %%r9\n\
xchg (%%rsp), %%r9\n\
pop %%r9\n\
"::);
clock_gettime(CLOCK_MONOTONIC, &t2);
uint64_t t = t1.tv_sec*1000000000LL + t1.tv_nsec;
t = (t2.tv_sec*1000000000LL + t2.tv_nsec) - t;
printf("Xchg: %llu nsec for %d occurences\n", t, NUM);
for(int i=0; i<NUM; ++i)
asm volatile ("\n\
push %%rax\n\
push %%r9\n\
mov $0x123456789abcdef, %%r9\n\
mov 0x8(%%rsp), %%r9\n\
pop %%r9\n\
pop %%rax\n\
"::);
clock_gettime(CLOCK_MONOTONIC, &t2);
t = t1.tv_sec*1000000000LL + t1.tv_nsec;
t = (t2.tv_sec*1000000000LL + t2.tv_nsec) - t;
printf("Push: %llu nsec for %d occurences\n", t, NUM);
}
And while I have some variation, the xchg method is always faster:
seb@Seb-PC:~/git/wasmer$ ./test
Xchg: 1911800 nsec for 100000 occurences
Push: 2216700 nsec for 100000 occurences
seb@Seb-PC:~/git/wasmer$ ./test
Xchg: 1156300 nsec for 100000 occurences
Push: 1452800 nsec for 100000 occurences
seb@Seb-PC:~/git/wasmer$ ./test
Xchg: 1156000 nsec for 100000 occurences
Push: 1452800 nsec for 100000 occurences
seb@Seb-PC:~/git/wasmer$ ./test
Xchg: 1345700 nsec for 100000 occurences
Push: 1668700 nsec for 100000 occurences
seb@Seb-PC:~/git/wasmer$ ./test
Xchg: 1499400 nsec for 100000 occurences
Push: 1898400 nsec for 100000 occurences
seb@Seb-PC:~/git/wasmer$ ./test
Xchg: 1185400 nsec for 100000 occurences
Push: 1493200 nsec for 100000 occurences
seb@Seb-PC:~/git/wasmer$ ./test
Xchg: 1276400 nsec for 100000 occurences
Push: 1598700 nsec for 100000 occurences
seb@Seb-PC:~/git/wasmer$ ./test
Xchg: 1276800 nsec for 100000 occurences
Push: 1596000 nsec for 100000 occurences
The implicit LOCK is not that slow, compare to a write+read in memory that the push/pop.
|
Please retain the authorship for the test case. Either cherry-pick the commit from my branch or if you'd rather not do so for some reason, I can submit a PR with the changes I've made as a separate PR once the fix here lands. |
|
Ok, I'll see what I can do so you keep the authorship, no problem. |
I'll let you do a PR when this fix is merged. |
|
bors r+ |
ptitSeb
changed the title
2725: Regression test for #2721 r=syrusakbary a=nagisa The wasm snippet in the test corresponds to roughly: ```rust let name = "bananapeach"; banana(a, b, c, name.len() as _, name.as_ptr() as _, f, g, h); ``` however sometime between 2.0 and 2.1 the name pointer is no longer being passed through as an argument. Instead a 0 gets passed in. To make things weirder, if `name.as_ptr()` is passed through multiple times, the second time the pointer will get passed correctly. The fix for this issue is in #2723. Co-authored-by: Simonas Kazlauskas <[email protected]> Co-authored-by: Syrus Akbary <[email protected]>
Co-authored-by: ptitSeb <[email protected]>
Description
Fix for #2721, regression introduce with Windows backend for Singlepass.