------- Comment #8 from nbenoit at tuxfamily dot org 2009-12-16 10:34 -------
I am confused, a performance regression is still noticeable:
* Intel Xeon E5320 (x86_64 arch but gcc machine is i686-pc-linux-gnu), with -O1
flag
GCC-4.4.2 7364 ms
GCC-trunk-r155286 9515 ms
* Intel Xeon 5160 (x86_64 arch and gcc machine is x86_64-linux-gnu), with -O1
flag
GCC-4.4.1 5960 ms
GCC-trunk-r155286 7355 ms
Here is a diff on the assembly generated for the Intel E5320:
$ diff 442/convol.s r155286/convol.s
11c11
< subl $8, %esp
---
> subl $12, %esp
13d12
< movl $H, %esi
17c16
< imull (%esi,%eax,4), %ebx
---
> imull H(,%eax,4), %ebx
22c21
< jg .L10
---
> setle %bl
24,25c23,25
< jle .L3
< .L10:
---
> setle -21(%ebp)
> testb %bl, -21(%ebp)
> jne .L3
28c28
< .L6:
---
> .L5:
31,32c31,32
< je .L5
< .L8:
---
> je .L4
> .L7:
34c34
< js .L6
---
> js .L5
40c40
< .L5:
---
> .L4:
43c43
< je .L7
---
> je .L6
46,47c46,47
< jmp .L8
< .L7:
---
> jmp .L7
> .L6:
50c50
< addl $8, %esp
---
> addl $12, %esp
60c60
< .ident "GCC: (GNU) 4.4.2"
---
> .ident "GCC: (GNU) 4.5.0 20091216 (experimental)"
--
nbenoit at tuxfamily dot org changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|RESOLVED |REOPENED
Resolution|FIXED |
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42027
