Re: Ideas for Ada 200X

[Ole-Hjalmar Kristensen <oleh@vlinux.voxelvision.no>]

Bill Findlay <yaldnifw@blueyonder.co.uk> writes:

> On 4/6/03 15:21, in article 7vllwh53ei.fsf@vlinux.voxelvision.no,
> "Ole-Hjalmar Kristensen" <oleh@vlinux.voxelvision.no> wrote:
> 
> > Bill Findlay <yaldnifw@blueyonder.co.uk> writes:
> > 
> >> On 4/6/03 11:10, in article 7vu1b640gf.fsf@vlinux.voxelvision.no,
> >> "Ole-Hjalmar Kristensen" <oleh@vlinux.voxelvision.no> wrote:
> >> 
> >>> I think the difference between C++ and Ada is the compiler. I run both
> >>> through gcc, and got identical results.
> >> 
> >> They are far from identical, AFAICS.
> >> You show the Ada code as being 10 times faster, at 0.200s user-mode CPU time
> >> for C++ matrix_c, 0.030s for Ada matrix_fun, and 0.010s for matrix_proc :
> >> 
> 
> > 
> > Well, yes, but the clock time is the same. It's an otherwise unloaded
> > system, and I get repeatable results, so I suspect there's something
> > funny going on with the user time measurements. Besides, a user time
> > of 0.01 seconds would indicate 10**10 additions per second, and I
> > don't believe my computer is quite as fast as that.
> 
> Even the 0.2s real time indicates a rate of order 10**9 per second.
> Is your computer really that fast, or could the Ada optimizer have turned
> most of the additions into zero-time operations? 8-)
> 
> -- 
> Bill-Findlay chez blue-yonder.co.uk ("-" => "")
> 
> 

I suspect that the matrices are small enough that both code and data
are predominantly in the cache, and in that case, it should be that
fast. Here is the assembler listing, and it certainly looks like it's
adding:


	.file	"matrix_proc.adb"
gcc2_compiled.:
___gnu_compiled_ada:
.text
	.align 4
_matrix_proc__add.0:
	pushl %ebp
	movl %esp,%ebp
	subl $8,%esp
	pushl %edi
	pushl %esi
	movl $400,%edi
	pushl %ebx
	movl 12(%ebp),%esi
	movl %ecx,-4(%ebp)
	movl $1,-8(%ebp)
	movl 8(%ebp),%ecx
	.align 2,0x90
L6:
	movl $1,%ebx
	leal 4(%edi),%edx
	.align 2,0x90
L10:
	movl -404(%ecx,%edx),%eax
	addl -404(%esi,%edx),%eax
	movl %eax,-404(%ecx,%edx)
	movl -400(%ecx,%edx),%eax
	addl -400(%esi,%edx),%eax
	movl %eax,-400(%ecx,%edx)
	movl -396(%ecx,%edx),%eax
	addl -396(%esi,%edx),%eax
	movl %eax,-396(%ecx,%edx)
	movl -392(%ecx,%edx),%eax
	addl -392(%esi,%edx),%eax
	movl %eax,-392(%ecx,%edx)
	movl -388(%ecx,%edx),%eax
	addl -388(%esi,%edx),%eax
	movl %eax,-388(%ecx,%edx)
	addl $5,%ebx
	addl $20,%edx
	cmpl $100,%ebx
	jle L10
	addl $400,%edi
	incl -8(%ebp)
	cmpl $100,-8(%ebp)
	jle L6
	leal -20(%ebp),%esp
	popl %ebx
	popl %esi
	popl %edi
	movl %ebp,%esp
	popl %ebp
	ret
	.align 4
.globl __ada_matrix_proc
__ada_matrix_proc:
	pushl %ebp
	movl $80008,%eax
	movl %esp,%ebp
	call __alloca
	pushl %edi
	pushl %esi
	pushl %ebx
	leal -40000(%ebp),%edx
	movl $1,-80004(%ebp)
	leal -80000(%ebp),%ebx
	.align 2,0x90
L18:
	movl $1,-80008(%ebp)
	movl $400,%edi
	.align 2,0x90
L21:
	movl $1,%esi
	leal 4(%edi),%ecx
	.align 2,0x90
L24:
	movl -404(%edx,%ecx),%eax
	addl -404(%ebx,%ecx),%eax
	movl %eax,-404(%edx,%ecx)
	movl -400(%edx,%ecx),%eax
	addl -400(%ebx,%ecx),%eax
	movl %eax,-400(%edx,%ecx)
	movl -396(%edx,%ecx),%eax
	addl -396(%ebx,%ecx),%eax
	movl %eax,-396(%edx,%ecx)
	movl -392(%edx,%ecx),%eax
	addl -392(%ebx,%ecx),%eax
	movl %eax,-392(%edx,%ecx)
	movl -388(%edx,%ecx),%eax
	addl -388(%ebx,%ecx),%eax
	movl %eax,-388(%edx,%ecx)
	addl $5,%esi
	addl $20,%ecx
	cmpl $100,%esi
	jle L24
	addl $400,%edi
	incl -80008(%ebp)
	cmpl $100,-80008(%ebp)
	jle L21
	incl -80004(%ebp)
	cmpl $10000,-80004(%ebp)
	jle L18
	leal -80020(%ebp),%esp
	popl %ebx
	popl %esi
	popl %edi
	movl %ebp,%esp
	popl %ebp
	ret

-- 
Ole-Hj. Kristensen

******************************************************************************
* You cannot consistently believe this sentence.
******************************************************************************