gemv_t_sse.s

Optimized GotoBLAS libraries

	addss	%xmm2, %xmm5

	movss	2 * SIZE(%ebp), %xmm1
	movss	3 * SIZE(%ebp), %xmm2

	addss	%xmm1, %xmm6
	addss	%xmm2, %xmm7

	movss	%xmm4, 0 * SIZE(%ebp)
	movss	%xmm5, 1 * SIZE(%ebp)
	movss	%xmm6, 2 * SIZE(%ebp)
	movss	%xmm7, 3 * SIZE(%ebp)
	addl	$4 * SIZE, %ebp

	decl	J
	jg	.L51

 	movl	N, %esi
	andl	$3, %esi
	jne	.L100

	jmp	.L99
	ALIGN_3

.L56:
	movss	0 * SIZE(%edx), %xmm1
	addl	%eax, %edx
	movss	0 * SIZE(%edx), %xmm2
	addl	%eax, %edx

	addss	%xmm1, %xmm4
	addss	%xmm2, %xmm5

	movss	0 * SIZE(%edx), %xmm1
	addl	%eax, %edx
	movss	0 * SIZE(%edx), %xmm2

	addss	%xmm1, %xmm6
	addss	%xmm2, %xmm7

	movss	%xmm4, 0 * SIZE(%ebp)
	addl	%eax, %ebp
	movss	%xmm5, 0 * SIZE(%ebp)
	addl	%eax, %ebp
	movss	%xmm6, 0 * SIZE(%ebp)
	addl	%eax, %ebp
	movss	%xmm7, 0 * SIZE(%ebp)
	addl	%eax, %ebp

	decl	J
	jg	.L51

 	movl	N, %esi
	andl	$3, %esi
	jne	.L100

	ALIGN_3

.L99:
	movl	A, %ebx
	addl	NLDA, %ebx
	movl	%ebx, A

	movl	IS, %esi
	addl	$P, %esi
	cmpl	M,  %esi
	jl	.L10
	ALIGN_3

.L999:
	movl	OLD_STACK, %esp
	popl	%ebx
	popl	%esi
	popl	%edi
	popl	%ebp
	ret
	ALIGN_3

.L100:
	movl	N, %esi
	andl	$3, %esi
	cmpl	$3, %esi
	jne	.L110
	ALIGN_3

.L101:
	movl	A, %ebx				# a_offset = a
	movl	LDA, %edx

	leal	(%ebx, %edx), %ecx		# a_offset2 = a + lda
	leal	(%ebx, %edx, 2), %eax
	addl	%edx, %eax

	movl	%eax, A

	movl	BUFFER, %esi

	pxor	%xmm4, %xmm4
	pxor	%xmm5, %xmm5
	pxor	%xmm6, %xmm6

	movaps	0 * SIZE(%esi), %xmm0
	movaps	4 * SIZE(%esi), %xmm2

	movl	MIN_M, %eax
	sarl	$3,  %eax
	jle	.L103
	ALIGN_3

.L102:
	movsd	 0 * SIZE(%ebx), %xmm1
	movhps	 2 * SIZE(%ebx), %xmm1
	mulps	%xmm0, %xmm1
	addps	%xmm1, %xmm4

	movsd	 0 * SIZE(%ecx), %xmm1
	movhps	 2 * SIZE(%ecx), %xmm1
	mulps	%xmm0, %xmm1
	addps	%xmm1, %xmm5

	movsd	 0 * SIZE(%ebx, %edx, 2), %xmm1
	movhps	 2 * SIZE(%ebx, %edx, 2), %xmm1
	mulps	%xmm0, %xmm1
	addps	%xmm1, %xmm6

	movaps	 8 * SIZE(%esi), %xmm0

	movsd	 4 * SIZE(%ebx), %xmm1
	movhps	 6 * SIZE(%ebx), %xmm1
	mulps	%xmm2, %xmm1
	addps	%xmm1, %xmm4

	movsd	 4 * SIZE(%ecx), %xmm1
	movhps	 6 * SIZE(%ecx), %xmm1
	mulps	%xmm2, %xmm1
	addps	%xmm1, %xmm5

	movsd	 4 * SIZE(%ebx, %edx, 2), %xmm1
	movhps	 6 * SIZE(%ebx, %edx, 2), %xmm1
	mulps	%xmm2, %xmm1
	addps	%xmm1, %xmm6

	movaps	12 * SIZE(%esi), %xmm2

	addl	$8 * SIZE, %ecx
	addl	$8 * SIZE, %ebx
	addl	$8 * SIZE, %esi

	decl	%eax
	jg	.L102
	ALIGN_3

.L103:
	movl	MIN_M, %eax
	andl	$7,  %eax
	je	.L105
	ALIGN_3

.L104:
	movss	0 * SIZE(%ebx), %xmm1
	mulss	%xmm0, %xmm1
	addss	%xmm1, %xmm4

	movss	0 * SIZE(%ecx), %xmm1
	addl	$SIZE, %ecx
	mulss	%xmm0, %xmm1
	addss	%xmm1, %xmm5

	movss	0 * SIZE(%ebx, %edx, 2), %xmm1
	addl	$SIZE, %ebx
	mulss	%xmm0, %xmm1
	addss	%xmm1, %xmm6

	movss	 1 * SIZE(%esi), %xmm0
	addl	$SIZE, %esi

	decl	%eax
	jg	.L104
	ALIGN_3

.L105:
	movaps	%xmm4, %xmm0
	shufps	$0xe, %xmm4, %xmm4
	addps	 %xmm0, %xmm4
	
	movaps	%xmm5, %xmm0
	shufps	$0xe, %xmm5, %xmm5
	addps	 %xmm0, %xmm5

	movaps	%xmm6, %xmm0
	shufps	$0xe, %xmm6, %xmm6
	addps	 %xmm0, %xmm6

	movaps	%xmm4, %xmm0
	shufps	$0x39, %xmm4, %xmm4
	addss	 %xmm0, %xmm4

	movaps	%xmm5, %xmm0
	shufps	$0x39, %xmm5, %xmm5
	addss	 %xmm0, %xmm5

	movaps	%xmm6, %xmm0
	shufps	$0x39, %xmm6, %xmm6
	addss	 %xmm0, %xmm6

	mulss	%xmm3, %xmm4
	mulss	%xmm3, %xmm5
	mulss	%xmm3, %xmm6

	movl	INCY, %eax
	movl	%ebp, %edx
	cmpl	$SIZE, %eax
	jne	.L106

	movss	0 * SIZE(%ebp), %xmm1
	movss	1 * SIZE(%ebp), %xmm2

	addss	%xmm1, %xmm4
	addss	%xmm2, %xmm5

	movss	2 * SIZE(%ebp), %xmm1

	addss	%xmm1, %xmm6

	movss	%xmm4, 0 * SIZE(%ebp)
	movss	%xmm5, 1 * SIZE(%ebp)
	movss	%xmm6, 2 * SIZE(%ebp)
	jmp	.L99
	ALIGN_3

.L106:
	movss	0 * SIZE(%edx), %xmm1
	addl	%eax, %edx
	movss	0 * SIZE(%edx), %xmm2
	addl	%eax, %edx

	addss	%xmm1, %xmm4
	addss	%xmm2, %xmm5

	movss	0 * SIZE(%edx), %xmm1
	addss	%xmm1, %xmm6

	movss	%xmm4, 0 * SIZE(%ebp)
	addl	%eax, %ebp
	movss	%xmm5, 0 * SIZE(%ebp)
	addl	%eax, %ebp
	movss	%xmm6, 0 * SIZE(%ebp)
	jmp	.L99
	ALIGN_3

.L110:
	cmpl	$2, %esi
	jne	.L120
	ALIGN_3

.L111:
	movl	A, %ebx				# a_offset = a
	movl	LDA, %edx

	leal	(%ebx, %edx), %ecx		# a_offset2 = a + lda
	leal	(%ebx, %edx, 2), %eax

	movl	%eax, A

	movl	BUFFER, %esi

	pxor	%xmm4, %xmm4
	pxor	%xmm5, %xmm5

	movaps	0 * SIZE(%esi), %xmm0
	movaps	4 * SIZE(%esi), %xmm2

	movl	MIN_M, %eax
	sarl	$3,  %eax
	jle	.L113
	ALIGN_3

.L112:
	movsd	 0 * SIZE(%ebx), %xmm1
	movhps	 2 * SIZE(%ebx), %xmm1
	mulps	%xmm0, %xmm1
	addps	%xmm1, %xmm4

	movsd	 0 * SIZE(%ecx), %xmm1
	movhps	 2 * SIZE(%ecx), %xmm1
	mulps	%xmm0, %xmm1
	addps	%xmm1, %xmm5

	movaps	 8 * SIZE(%esi), %xmm0

	movsd	 4 * SIZE(%ebx), %xmm1
	movhps	 6 * SIZE(%ebx), %xmm1
	mulps	%xmm2, %xmm1
	addps	%xmm1, %xmm4

	movsd	 4 * SIZE(%ecx), %xmm1
	movhps	 6 * SIZE(%ecx), %xmm1
	mulps	%xmm2, %xmm1
	addps	%xmm1, %xmm5

	movaps	12 * SIZE(%esi), %xmm2
	addl	$8 * SIZE, %ebx
	addl	$8 * SIZE, %ecx
	addl	$8 * SIZE, %esi

	decl	%eax
	jg	.L112
	ALIGN_3

.L113:
	movl	MIN_M, %eax
	andl	$7,  %eax
	je	.L115
	ALIGN_3

.L114:
	movss	0 * SIZE(%ebx), %xmm1
	addl	$SIZE, %ebx
	mulss	%xmm0, %xmm1
	addss	%xmm1, %xmm4

	movss	0 * SIZE(%ecx), %xmm1
	addl	$SIZE, %ecx
	mulss	%xmm0, %xmm1
	addss	%xmm1, %xmm5

	movss	 1 * SIZE(%esi), %xmm0
	addl	$SIZE, %esi

	decl	%eax
	jg	.L114
	ALIGN_3

.L115:
	movaps	%xmm4, %xmm0
	shufps	$0xe, %xmm4, %xmm4
	addps	 %xmm0, %xmm4

	movaps	%xmm5, %xmm0
	shufps	$0xe, %xmm5, %xmm5
	addps	 %xmm0, %xmm5

	movaps	%xmm4, %xmm0
	shufps	$0x39, %xmm4, %xmm4
	addss	 %xmm0, %xmm4

	movaps	%xmm5, %xmm0
	shufps	$0x39, %xmm5, %xmm5
	addss	 %xmm0, %xmm5

	mulss	%xmm3, %xmm4
	mulss	%xmm3, %xmm5

	movl	INCY, %eax
	movl	%ebp, %edx
	cmpl	$SIZE, %eax
	jne	.L116

	movss	0 * SIZE(%ebp), %xmm1
	movss	1 * SIZE(%ebp), %xmm2

	addss	%xmm1, %xmm4
	addss	%xmm2, %xmm5

	movss	%xmm4, 0 * SIZE(%ebp)
	movss	%xmm5, 1 * SIZE(%ebp)
	jmp	.L99
	ALIGN_3

.L116:
	movss	0 * SIZE(%edx), %xmm1
	addl	%eax, %edx
	movss	0 * SIZE(%edx), %xmm2

	addss	%xmm1, %xmm4
	addss	%xmm2, %xmm5

	movss	%xmm4, 0 * SIZE(%ebp)
	addl	%eax, %ebp
	movss	%xmm5, 0 * SIZE(%ebp)
	jmp	.L99
	ALIGN_3

.L120:
	movl	A, %ebx				# a_offset = a
	movl	LDA, %edx

	leal	(%ebx, %edx), %ecx		# a_offset2 = a + lda
	leal	(%ebx, %edx, 1), %eax

	movl	%eax, A

	movl	BUFFER, %esi

	pxor	%xmm4, %xmm4

	movaps	0 * SIZE(%esi), %xmm0
	movaps	4 * SIZE(%esi), %xmm2

	movl	MIN_M, %eax
	sarl	$3,  %eax
	jle	.L123
	ALIGN_3

.L122:
	movsd	 0 * SIZE(%ebx), %xmm1
	movhps	 2 * SIZE(%ebx), %xmm1
	mulps	%xmm0, %xmm1
	addps	%xmm1, %xmm4

	movaps	 8 * SIZE(%esi), %xmm0

	movsd	 4 * SIZE(%ebx), %xmm1
	movhps	 6 * SIZE(%ebx), %xmm1
	mulps	%xmm2, %xmm1
	addps	%xmm1, %xmm4

	movaps	12 * SIZE(%esi), %xmm2

	addl	$8 * SIZE, %ebx
	addl	$8 * SIZE, %esi

	decl	%eax
	jg	.L122
	ALIGN_3

.L123:
	movl	MIN_M, %eax
	andl	$7,  %eax
	je	.L125
	ALIGN_3

.L124:
	movss	0 * SIZE(%ebx), %xmm1
	addl	$SIZE, %ebx
	mulss	%xmm0, %xmm1
	addss	%xmm1, %xmm4

	movss	 1 * SIZE(%esi), %xmm0
	addl	$SIZE, %esi

	decl	%eax
	jg	.L124
	ALIGN_3

.L125:
	movaps	%xmm4, %xmm0
	shufps	$0xe, %xmm4, %xmm4
	addps	 %xmm0, %xmm4
	
	movaps	%xmm4, %xmm0
	shufps	$0x39, %xmm4, %xmm4
	addss	 %xmm0, %xmm4

	mulss	%xmm3, %xmm4

	movss	0 * SIZE(%ebp), %xmm1
	addss	%xmm1, %xmm4
	movss	%xmm4, 0 * SIZE(%ebp)
	jmp	.L99

	EPILOGUE