zsymv_u_sse2.s

Optimized GotoBLAS libraries

	mulpd	  atemp3, a3
	addpd	  xt1,    xsum2
	addpd	  a3,     yy1
	MOVDDUP(4 * SIZE, A1, a3)

	movapd	  xtemp4, xt1
	movapd	14 * SIZE(XX), xtemp4
	mulpd	  a1,     xt1
	mulpd	  atemp2, a1
	ADD	  xt1,    xsum1
	addpd	  a1,     yy2
	MOVDDUP(6 * SIZE, A2, a1)

	movlpd	 yy2, 2 * SIZE(YY)
	movhpd	 yy2, 3 * SIZE(YY)
	movsd	 6 * SIZE(YY), yy2
	movhpd	 7 * SIZE(YY), yy2

	movapd	  xtemp2, xt1
	movapd	10 * SIZE(XX), xtemp2
	mulpd	  a2,     xt1
	mulpd	  atemp4, a2
	ADD	  xt1,    xsum2
	addpd	  a2,     yy1
	MOVDDUP(5 * SIZE, A1, a2)

	PREFETCH	PREFETCHSIZE(A2)

	movlpd	 yy1, 0 * SIZE(YY)
	movhpd	 yy1, 1 * SIZE(YY)
	movsd	 4 * SIZE(YY), yy1
	movhpd	 5 * SIZE(YY), yy1

	movapd	  xtemp1, xt1
	mulpd	  a3,     xt1
	mulpd	  atemp1, a3
	addpd	  xt1,    xsum1
	addpd	  a3,     yy1
	MOVDDUP(7 * SIZE, A2, a3)

	movapd	  xtemp3, xt1
	mulpd	  a1,     xt1
	mulpd	  atemp3, a1
	addpd	  xt1,    xsum2
	addpd	  a1,     yy2
	MOVDDUP(6 * SIZE, A1, a1)

	movapd	  xtemp2, xt1
	mulpd	  a2,     xt1
	mulpd	  atemp2, a2
	ADD	  xt1,    xsum1
	addpd	  a2,     yy1
	MOVDDUP(4 * SIZE, A2, a2)

	PREFETCHW	PREFETCHSIZE(YY)

	movapd	  xtemp4, xt1
	mulpd	  a3,     xt1
	mulpd	  atemp4, a3
	ADD	  xt1,    xsum2
	addpd	  a3,     yy2
	MOVDDUP(7 * SIZE, A1, a3)

	movapd	  xtemp3, xt1
	movapd	20 * SIZE(XX), xtemp3
	mulpd	  a1,     xt1
	mulpd	  atemp1, a1
	addpd	  xt1,    xsum1
	addpd	  a1,     yy2
	MOVDDUP(5 * SIZE, A2, a1)

	movapd	  xtemp1, xt1
	movapd	16 * SIZE(XX), xtemp1
	mulpd	  a2,     xt1
	mulpd	  atemp3, a2
	addpd	  xt1,    xsum2
	addpd	  a2,     yy1
	MOVDDUP(10 * SIZE, A2, a2)

	movapd	  xtemp4, xt1
	movapd	22 * SIZE(XX), xtemp4
	mulpd	  a3,     xt1
	mulpd	  atemp2, a3
	ADD	  xt1,    xsum1
	addpd	  a3,     yy2
	MOVDDUP( 9 * SIZE, A1, a3)

	movlpd	 yy2, 6 * SIZE(YY)
	movhpd	 yy2, 7 * SIZE(YY)
	movsd	10 * SIZE(YY), yy2
	movhpd	11 * SIZE(YY), yy2

	movapd	  xtemp2, xt1
	movapd	18 * SIZE(XX), xtemp2
	mulpd	  a1,     xt1
	mulpd	  atemp4, a1
	ADD	  xt1,    xsum2
	addpd	  a1,     yy1
	MOVDDUP( 8 * SIZE, A1, a1)

	movlpd	 yy1, 4 * SIZE(YY)
	movhpd	 yy1, 5 * SIZE(YY)
	movsd	 8 * SIZE(YY), yy1
	movhpd	 9 * SIZE(YY), yy1

	subq	 $-16 * SIZE, XX
	addq	 $  8 * SIZE, YY
	addq	 $  8 * SIZE, A1
	addq	 $  8 * SIZE, A2

	decq	 I
	jg	 .L12
	ALIGN_3

.L15:
	testq	$2, IS
	jle	.L18

	movapd	  xtemp1, xt1
	mulpd	  a1,     xt1
	mulpd	  atemp1, a1
	addpd	  xt1,    xsum1
	addpd	  a1,     yy1
	MOVDDUP(1 * SIZE, A1, a1)

	movapd	  xtemp3, xt1
	mulpd	  a2,     xt1
	mulpd	  atemp3, a2
	addpd	  xt1,    xsum2
	addpd	  a2,     yy2
	MOVDDUP(3 * SIZE, A2, a2)

	movapd	  xtemp2, xt1
	mulpd	  a1,     xt1
	mulpd	  atemp2, a1
	ADD	  xt1,    xsum1
	addpd	  a1,     yy1
	MOVDDUP(2 * SIZE, A1, a1)

	movapd	  xtemp4, xt1
	mulpd	  a2,     xt1
	mulpd	  atemp4, a2
	ADD	  xt1,    xsum2
	addpd	  a2,     yy2
	MOVDDUP(0 * SIZE, A2, a2)

	movapd	  xtemp3, xt1
	mulpd	  a1,     xt1
	mulpd	  atemp1, a1
	addpd	  xt1,    xsum1
	addpd	  a1,     yy2
	MOVDDUP(3 * SIZE, A1, a1)

	movapd	  xtemp1, xt1
	mulpd	  a2,     xt1
	mulpd	  atemp3, a2
	addpd	  xt1,    xsum2
	addpd	  a2,     yy1
	MOVDDUP(1 * SIZE, A2, a2)

	movapd	  xtemp4, xt1
	mulpd	  a1,     xt1
	mulpd	  atemp2, a1
	ADD	  xt1,    xsum1
	addpd	  a1,     yy2

	movlpd	 yy2, 2 * SIZE(YY)
	movhpd	 yy2, 3 * SIZE(YY)
	movsd	 6 * SIZE(YY), yy2
	movhpd	 7 * SIZE(YY), yy2

	movapd	  xtemp2, xt1
	mulpd	  a2,     xt1
	mulpd	  atemp4, a2
	ADD	  xt1,    xsum2
	addpd	  a2,     yy1

	movlpd	 yy1, 0 * SIZE(YY)
	movhpd	 yy1, 1 * SIZE(YY)
	movsd	 4 * SIZE(YY), yy1
	movhpd	 5 * SIZE(YY), yy1

	addq	 $4 * SIZE, YY
	addq	 $4 * SIZE, A1
	addq	 $4 * SIZE, A2
	ALIGN_3

.L18:
	MOVDDUP(0 * SIZE, A1, a1)
	MOVDDUP(0 * SIZE, A2, a2)

	mulpd	  atemp1, a1
	mulpd	  atemp1, a2
	addpd	  a1,     xsum1
	addpd	  a2,     xsum2

#ifndef HEMV
	MOVDDUP(1 * SIZE, A1, a1)
	MOVDDUP(1 * SIZE, A2, a2)

	mulpd	  atemp2, a1
	mulpd	  atemp2, a2
	addpd	  a1,     xsum1
	addpd	  a2,     xsum2
#else
	MOVDDUP(1 * SIZE, A2, a2)

	mulpd	  atemp2, a2
	subpd	  a2,     xsum2
#endif

	MOVDDUP(0 * SIZE, A2, a1)
	MOVDDUP(2 * SIZE, A2, a2)

	mulpd	  atemp3, a1
	mulpd	  atemp3, a2
	addpd	  a1,     xsum1
	addpd	  a2,     xsum2

#ifndef HEMV
	MOVDDUP(1 * SIZE, A2, a1)
	MOVDDUP(3 * SIZE, A2, a2)

	mulpd	  atemp4, a1
	mulpd	  atemp4, a2
	addpd	  a1,     xsum1
	addpd	  a2,     xsum2
#else
	MOVDDUP(1 * SIZE, A2, a1)

	mulpd	  atemp4, a1
	addpd	  a1,     xsum1
#endif

	addpd	 xsum1, yy1
	addpd	 xsum2, yy2

	movlpd	 yy1, 0 * SIZE(YY)
	movhpd	 yy1, 1 * SIZE(YY)
	movlpd	 yy2, 2 * SIZE(YY)
	movhpd	 yy2, 3 * SIZE(YY)

	addq	 $2, IS

	movq	 IS, I
	addq	 $2, I
	cmpq	 M, I
	jle	 .L11
	ALIGN_3

.L20:
	testq	$1, M
	jle	.L990

	movq	A,  A1
	leaq	(, IS, 4), I

	movapd		0 * SIZE(NEW_X, I, SIZE), atemp1
	movapd		2 * SIZE(NEW_X, I, SIZE), atemp2

	pxor		xsum1, xsum1
	pxor		xsum2, xsum2

	MOVDDUP(0 * SIZE, A1, a1)
	MOVDDUP(1 * SIZE, A1, a2)

	movapd	 0 * SIZE(NEW_X), xtemp1
	movapd	 2 * SIZE(NEW_X), xtemp2
	movapd	 4 * SIZE(NEW_X), xtemp3
	movapd	 6 * SIZE(NEW_X), xtemp4

	movsd	 0 * SIZE(NEW_Y), yy1
	movhpd	 1 * SIZE(NEW_Y), yy1
	movsd	 2 * SIZE(NEW_Y), yy2
	movhpd	 3 * SIZE(NEW_Y), yy2

	movq		NEW_X, XX
	movq		NEW_Y, YY

	movq	IS,  I
	sarq	$1,  I
	jle	.L28
	ALIGN_3

.L22:
	movapd	  xtemp1, xt1
	movapd	 8 * SIZE(XX), xtemp1
	mulpd	  a1,     xt1
	mulpd	  atemp1, a1
	addpd	  xt1,    xsum1
	addpd	  a1,     yy1
	MOVDDUP(2 * SIZE, A1, a1)

	movapd	  xtemp2, xt1
	movapd	10 * SIZE(XX), xtemp2
	mulpd	  a2,     xt1
	mulpd	  atemp2, a2
	ADD	  xt1,    xsum2
	addpd	  a2,     yy1
	MOVDDUP(3 * SIZE, A1, a2)

	movlpd	 yy1, 0 * SIZE(YY)
	movhpd	 yy1, 1 * SIZE(YY)
	movsd	 4 * SIZE(YY), yy1
	movhpd	 5 * SIZE(YY), yy1

	movapd	  xtemp3, xt1
	movapd	12 * SIZE(XX), xtemp3
	mulpd	  a1,     xt1
	mulpd	  atemp1, a1
	addpd	  xt1,    xsum1
	addpd	  a1,     yy2
	MOVDDUP(4 * SIZE, A1, a1)

	movapd	  xtemp4, xt1
	movapd	14 * SIZE(XX), xtemp4
	mulpd	  a2,     xt1
	mulpd	  atemp2, a2
	ADD	  xt1,    xsum2
	addpd	  a2,     yy2
	MOVDDUP(5 * SIZE, A1, a2)

	movlpd	 yy2, 2 * SIZE(YY)
	movhpd	 yy2, 3 * SIZE(YY)
	movsd	 6 * SIZE(YY), yy2
	movhpd	 7 * SIZE(YY), yy2

	addq	 $8 * SIZE, XX
	addq	 $4 * SIZE, YY
	addq	 $4 * SIZE, A1

	decq	 I
	jg	 .L22
	ALIGN_3

.L28:
	MOVDDUP(0 * SIZE, A1, a1)

#ifndef HEMV
	MOVDDUP(1 * SIZE, A1, a2)

	mulpd	  atemp1, a1
	mulpd	  atemp2, a2
	addpd	  a1,     xsum1
	addpd	  a2,     xsum2

#else
	mulpd	  atemp1, a1
	addpd	  a1,     xsum1
#endif

	addpd	 xsum2, xsum1
	addpd	 xsum1, yy1

	movlpd	 yy1, 0 * SIZE(YY)
	movhpd	 yy1, 1 * SIZE(YY)
	ALIGN_3

.L990:
	cmpq   $2 * SIZE, INCY
	je    .L999

	movq	M,  %rax
	sarq	$2, %rax
	jle	.L997
	ALIGN_3

.L996:
	movapd	 0 * SIZE(NEW_Y), %xmm0
	movapd	 2 * SIZE(NEW_Y), %xmm1
	movapd	 4 * SIZE(NEW_Y), %xmm2
	movapd	 6 * SIZE(NEW_Y), %xmm3

	movsd	%xmm0,  0 * SIZE(Y)
	movhpd	%xmm0,  1 * SIZE(Y)
	addq	INCY, Y
	movsd	%xmm1,  0 * SIZE(Y)
	movhpd	%xmm1,  1 * SIZE(Y)
	addq	INCY, Y
	movsd	%xmm2,  0 * SIZE(Y)
	movhpd	%xmm2,  1 * SIZE(Y)
	addq	INCY, Y
	movsd	%xmm3,  0 * SIZE(Y)
	movhpd	%xmm3,  1 * SIZE(Y)
	addq	INCY, Y

	addq	$8 * SIZE, NEW_Y
	decq	%rax
	jg	.L996
	ALIGN_3

.L997:
	movq	M, %rax
	andq	$3, %rax
	jle	.L999
	ALIGN_3

.L998:
	movapd	0 * SIZE(NEW_Y), %xmm0

	movsd	%xmm0,  0 * SIZE(Y)
	movhpd	%xmm0,  1 * SIZE(Y)
	addq	INCY, Y

	addq	$2 * SIZE, NEW_Y

	decq	%rax
	jg	.L998
	ALIGN_3

.L999:
	movq	  0(%rsp), %rbx
	movq	  8(%rsp), %rbp
	movq	 16(%rsp), %r12
	movq	 24(%rsp), %r13
	movq	 32(%rsp), %r14
	movq	 40(%rsp), %r15

#ifdef WINDOWS_ABI
	movq	 48(%rsp), %rdi
	movq	 56(%rsp), %rsi
	movups	 64(%rsp), %xmm6
	movups	 80(%rsp), %xmm7
	movups	 96(%rsp), %xmm8
	movups	112(%rsp), %xmm9
	movups	128(%rsp), %xmm10
	movups	144(%rsp), %xmm11
	movups	160(%rsp), %xmm12
	movups	176(%rsp), %xmm13
	movups	192(%rsp), %xmm14
	movups	208(%rsp), %xmm15
#endif

	addq	$STACKSIZE, %rsp
	ret
	EPILOGUE