Re: More linpack on i386-linux analysis

[N8TM at aol dot com]

In a message dated 7/12/98 7:47:22 AM Pacific Daylight Time, tm@netcom.com
writes:

> I've been looking at the inner loop of daxpy() in linpack
>  for i386-linux:


Here's what egcs-19980707 is giving me for unrolled inner loop of daxpy:

	.p2align 4,,7
L377:
	movl 16(%ebp),%eax
	fld %st(0)
	fmuls (%eax,%ecx,4)
	fadds (%esi,%edx,4)
	addl 20(%ebp),%ecx
	fstps (%esi,%edx,4)
	fld %st(0)
	fmuls (%eax,%ecx,4)
	addl %edi,%edx
	fadds (%esi,%edx,4)
	addl 20(%ebp),%ecx
	fstps (%esi,%edx,4)
	fld %st(0)
	fmuls (%eax,%ecx,4)
	addl %edi,%edx
	fadds (%esi,%edx,4)
	addl 20(%ebp),%ecx
	fstps (%esi,%edx,4)
	fld %st(0)
	fmuls (%eax,%ecx,4)
	addl %edi,%edx
	fadds (%esi,%edx,4)
	addl $4,%ebx
	addl 20(%ebp),%ecx
	fstps (%esi,%edx,4)
	addl %edi,%edx
	cmpl 8(%ebp),%ebx
	jl L377

and for the unit stride case:

	.p2align 4,,7
L382:
	fld %st(0)
	fmuls (%ecx)
	fadds (%esi)
	fstps (%esi)
	fld %st(0)
	fmuls 4(%ecx)
	fadds 4(%esi)
	fstps 4(%esi)
	fld %st(0)
	fmuls 8(%ecx)
	fadds 8(%esi)
	fstps 8(%esi)
	fld %st(0)
	fmuls 12(%ecx)
	fadds 12(%esi)
	addl $4,%ebx
	addl $16,%ecx
	fstps 12(%esi)
	addl $16,%esi
	cmpl 8(%ebp),%ebx
	jl L382