rpg@rpg$ cat test.c
#include <stdio.h>
#include <stdlib.h>

typedef float v4sf __attribute__ ((vector_size (16)));

union f4vector
{
  v4sf v;
  float f[4];
};

const v4sf half = {0.5,0.5,0.5,0.5};
int main()
{
	union f4vector b = {(float)rand(),(float)rand(),(float)rand(),(float)rand()};
	union f4vector c = {(float)rand(),(float)rand(),(float)rand(),(float)rand()};
	union f4vector data = {0.34, 0.29, 0.97, 0.0};
	union f4vector data2 = {0.0, 0.0, 0.0, 0.0};
	int i;
	for (i=0;i<1024*1024*100;i++)
	{
		//~ b = (float)clock();
		data2.v = (data.v+half)*c.v+b.v-half;
	}
	//~ float data3[4];
	//~ data3 = data2;
	printf("%f %f %f\n", data2.f[0],data2.f[1],data2.f[2]);

}rpg@rpg$ gcc -S -O2 test.c
rpg@rpg$ cat test.s
	.file	"test.c"
	.section	.rodata.str1.1,"aMS",@progbits,1
.LC4:
	.string	"%f %f %f\n"
	.text
	.p2align 4,,15
.globl main
	.type	main, @function
main:
	pushl	%ebp
	movl	%esp, %ebp
	andl	$-16, %esp
	pushl	%edi
	pushl	%esi
	pushl	%ebx
	subl	$164, %esp
	call	rand
	movl	%eax, %edi
	call	rand
	movl	%eax, %esi
	call	rand
	movl	%eax, %ebx
	call	rand
	movl	%edi, 44(%esp)
	fildl	44(%esp)
	fstps	48(%esp)
	movl	%esi, 44(%esp)
	fildl	44(%esp)
	fstps	52(%esp)
	movl	%ebx, 44(%esp)
	fildl	44(%esp)
	movl	%eax, 44(%esp)
	movl	48(%esp), %eax
	fstps	56(%esp)
	fildl	44(%esp)
	fstps	60(%esp)
	movl	%eax, 144(%esp)
	movl	52(%esp), %eax
	movl	%eax, 148(%esp)
	movl	56(%esp), %eax
	movl	%eax, 152(%esp)
	movl	60(%esp), %eax
	movl	%eax, 156(%esp)
	call	rand
	movl	%eax, %edi
	call	rand
	movl	%eax, %esi
	call	rand
	movl	%eax, %ebx
	call	rand
	movl	%edi, 44(%esp)
	fildl	44(%esp)
	fstps	48(%esp)
	movl	%esi, 44(%esp)
	fildl	44(%esp)
	fstps	52(%esp)
	movl	%ebx, 44(%esp)
	fildl	44(%esp)
	movl	%eax, 44(%esp)
	movl	48(%esp), %eax
	fstps	56(%esp)
	fildl	44(%esp)
	fstps	60(%esp)
	movl	%eax, 128(%esp)
	movl	52(%esp), %eax
	movl	%eax, 132(%esp)
	movl	56(%esp), %eax
	flds	132(%esp)
	movl	%eax, 136(%esp)
	movl	60(%esp), %eax
	flds	136(%esp)
	movl	%eax, 140(%esp)
	flds	140(%esp)
	flds	.LC0
	fmuls	128(%esp)
	fstps	48(%esp)
	fxch	%st(2)
	movl	48(%esp), %eax
	fmuls	.LC1
	fstps	52(%esp)
	fmuls	.LC2
	fstps	56(%esp)
	flds	.LC3
	fmul	%st, %st(1)
	fxch	%st(1)
	fstps	60(%esp)
	movl	%eax, 112(%esp)
	flds	112(%esp)
	fadds	144(%esp)
	movl	52(%esp), %eax
	fstps	48(%esp)
	movl	%eax, 116(%esp)
	flds	148(%esp)
	fadds	116(%esp)
	movl	56(%esp), %eax
	fstps	52(%esp)
	movl	%eax, 120(%esp)
	flds	152(%esp)
	fadds	120(%esp)
	movl	60(%esp), %eax
	fstps	56(%esp)
	movl	%eax, 124(%esp)
	flds	156(%esp)
	fadds	124(%esp)
	movl	48(%esp), %eax
	fstps	60(%esp)
	movl	%eax, 96(%esp)
	movl	52(%esp), %eax
	flds	96(%esp)
	fsub	%st(1), %st
	movl	%eax, 100(%esp)
	movl	56(%esp), %eax
	fstps	48(%esp)
	flds	100(%esp)
	fsub	%st(1), %st
	movl	%eax, 104(%esp)
	movl	60(%esp), %eax
	fstps	52(%esp)
	flds	104(%esp)
	fsub	%st(1), %st
	movl	%eax, 108(%esp)
	fstps	56(%esp)
	fsubrs	108(%esp)
	fstps	60(%esp)
	movl	48(%esp), %eax
	movl	$.LC4, 4(%esp)
	movl	$1, (%esp)
	movl	%eax, 64(%esp)
	movl	52(%esp), %eax
	movl	%eax, 68(%esp)
	movl	56(%esp), %eax
	movl	%eax, 72(%esp)
	movl	60(%esp), %eax
	flds	72(%esp)
	fstpl	24(%esp)
	movl	%eax, 76(%esp)
	flds	68(%esp)
	fstpl	16(%esp)
	flds	64(%esp)
	fstpl	8(%esp)
	call	__printf_chk
	addl	$164, %esp
	popl	%ebx
	popl	%esi
	popl	%edi
	movl	%ebp, %esp
	popl	%ebp
	ret
	.size	main, .-main
.globl half
	.section	.rodata
	.align 16
	.type	half, @object
	.size	half, 16
half:
	.long	1056964608
	.long	1056964608
	.long	1056964608
	.long	1056964608
	.section	.rodata.cst4,"aM",@progbits,4
	.align 4
.LC0:
	.long	1062668862
	.align 4
.LC1:
	.long	1061830000
	.align 4
.LC2:
	.long	1069295862
	.align 4
.LC3:
	.long	1056964608
	.ident	"GCC: (Ubuntu/Linaro 4.5.2-8ubuntu4) 4.5.2"
	.section	.note.GNU-stack,"",@progbits
rpg@rpg$ gcc -S -O2 -msse test.c
rpg@rpg$ cat test.s
	.file	"test.c"
	.section	.rodata.str1.1,"aMS",@progbits,1
.LC2:
	.string	"%f %f %f\n"
	.text
	.p2align 4,,15
.globl main
	.type	main, @function
main:
	pushl	%ebp
	movl	%esp, %ebp
	andl	$-16, %esp
	pushl	%edi
	pushl	%esi
	pushl	%ebx
	subl	$100, %esp
	call	rand
	movl	%eax, %ebx
	call	rand
	movl	%eax, 48(%esp)
	call	rand
	movl	%eax, 72(%esp)
	call	rand
	movl	%eax, 68(%esp)
	call	rand
	movl	%eax, 44(%esp)
	call	rand
	movl	%eax, %edi
	call	rand
	movl	%eax, %esi
	call	rand
	movl	44(%esp), %edx
	movl	48(%esp), %ecx
	movl	%edx, 76(%esp)
	fildl	76(%esp)
	fstps	64(%esp)
	movl	%edi, 76(%esp)
	fildl	76(%esp)
	fstps	60(%esp)
	movl	%esi, 76(%esp)
	fildl	76(%esp)
	fstps	56(%esp)
	movl	%eax, 76(%esp)
	fildl	76(%esp)
	fstps	52(%esp)
	movl	%ebx, 76(%esp)
	fildl	76(%esp)
	movss	64(%esp), %xmm3
	movl	%ecx, 76(%esp)
	fstps	64(%esp)
	movss	60(%esp), %xmm4
	movss	52(%esp), %xmm2
	movss	56(%esp), %xmm1
	fildl	76(%esp)
	unpcklps	%xmm4, %xmm3
	fstps	60(%esp)
	fildl	72(%esp)
	unpcklps	%xmm2, %xmm1
	fstps	72(%esp)
	fildl	68(%esp)
	fstps	68(%esp)
	movaps	%xmm1, %xmm0
	movaps	%xmm3, %xmm1
	movlhps	%xmm0, %xmm1
	movss	68(%esp), %xmm3
	movss	72(%esp), %xmm0
	movss	64(%esp), %xmm4
	mulps	.LC0, %xmm1
	unpcklps	%xmm3, %xmm0
	movss	60(%esp), %xmm3
	movaps	%xmm0, %xmm2
	unpcklps	%xmm3, %xmm4
	movaps	%xmm4, %xmm0
	movlhps	%xmm2, %xmm0
	addps	%xmm1, %xmm0
	subps	.LC1, %xmm0
	movaps	%xmm0, 80(%esp)
	flds	88(%esp)
	fstpl	24(%esp)
	movl	$.LC2, 4(%esp)
	flds	84(%esp)
	fstpl	16(%esp)
	movl	$1, (%esp)
	flds	80(%esp)
	fstpl	8(%esp)
	call	__printf_chk
	addl	$100, %esp
	popl	%ebx
	popl	%esi
	popl	%edi
	movl	%ebp, %esp
	popl	%ebp
	ret
	.size	main, .-main
.globl half
	.section	.rodata
	.align 16
	.type	half, @object
	.size	half, 16
half:
	.long	1056964608
	.long	1056964608
	.long	1056964608
	.long	1056964608
	.section	.rodata.cst16,"aM",@progbits,16
	.align 16
.LC0:
	.long	1062668862
	.long	1061830000
	.long	1069295862
	.long	1056964608
	.align 16
.LC1:
	.long	1056964608
	.long	1056964608
	.long	1056964608
	.long	1056964608
	.ident	"GCC: (Ubuntu/Linaro 4.5.2-8ubuntu4) 4.5.2"
	.section	.note.GNU-stack,"",@progbits
rpg@rpg$