ztrsm_kernel_ln_2x2_penryn.s

Optimized GotoBLAS libraries

	PREFETCH (PREFETCHSIZE +  0) * SIZE(AA)

	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	movsd	-28 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movsd	-28 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	movsd	-26 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movsd	-26 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	movsd	-24 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movsd	-24 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	movsd	-22 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movsd	-22 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	movsd	-20 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movsd	-20 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	movsd	-18 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movsd	-18 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	movsd	-16 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movsd	-16 * SIZE(AA), %xmm0

	subl   $-16 * SIZE, AA
	subl   $-16 * SIZE, BB
	
	decl	%eax
	jne	.L141
	ALIGN_4
	
.L142:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K,  %eax
	subl	KK, %eax
#endif
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L144
	ALIGN_4

.L143:
	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	movsd	-30 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movsd	-30 * SIZE(AA), %xmm0

	addl	$2 * SIZE, AA
	addl	$2 * SIZE, BB
	decl	%eax
	jg	.L143
	ALIGN_4

.L144:
#if defined(LN) || defined(RT)
	movl	KK, %eax
	subl	$1, %eax

	movl	AORIG, AA
	sall	$ZBASE_SHIFT, %eax
	leal	(AA, %eax, 1), AA
	leal	(B,  %eax, 1), BB
#endif

	addps	%xmm2, %xmm4
	addps	%xmm3, %xmm5

	pshufd	 $0xb1, %xmm5, %xmm5
	pcmpeqb	%xmm0, %xmm0
	psllq	$63,   %xmm0

#ifndef CONJ
	shufps	$0xb1, %xmm0, %xmm0

	pxor	%xmm0, %xmm5
#else
#if defined(LN) || defined(LT)
	pxor	%xmm0, %xmm4
#else
	pxor	%xmm0, %xmm5
#endif
#endif
	
	addps	%xmm5, %xmm4

#if defined(LN) || defined(LT)
	movsd	-32 * SIZE(BB), %xmm2

	subps	%xmm4,  %xmm2
#else
	movsd	-32 * SIZE(AA), %xmm1

	subps	%xmm4,  %xmm1
#endif

#if defined(LN) || defined(LT)
	movaps	-32 * SIZE(AA), %xmm5

	pshufd	 $0x44, %xmm5, %xmm6
	pshufd	 $0x11, %xmm5, %xmm7

	pshufd	 $0xa0, %xmm2, %xmm4
	pshufd	 $0xf5, %xmm2, %xmm2

#ifndef CONJ
	xorps	 %xmm0, %xmm2
#else
	xorps	 %xmm0, %xmm4
#endif

	mulps	%xmm6,  %xmm4
	mulps	%xmm7,  %xmm2
	addps	%xmm4,  %xmm2
#endif

#if defined(RN) || defined(RT)
	movaps	-32 * SIZE(BB), %xmm4

	pshufd	 $0x44, %xmm4, %xmm6
	pshufd	 $0x11, %xmm4, %xmm7

	pshufd	 $0xa0, %xmm1, %xmm3
	pshufd	 $0xf5, %xmm1, %xmm1

#ifndef CONJ
	xorps	 %xmm0, %xmm1
#else
	xorps	 %xmm0, %xmm3
#endif

	mulps	%xmm6,  %xmm3
	mulps	%xmm7,  %xmm1

	addps	%xmm3,  %xmm1
#endif

#ifdef LN
	subl	$2 * SIZE, CO1
#endif

#if defined(LN) || defined(LT)
	movlps	%xmm2,  -32 * SIZE(BB)

	movlps	%xmm2,   0 * SIZE(CO1)
#else
	movlps	%xmm1,  -32 * SIZE(AA)

	movlps	%xmm1,   0 * SIZE(CO1)
#endif

#ifndef LN
	addl	$2 * SIZE, CO1
#endif

#if defined(LT) || defined(RN)
	movl	K,  %eax
	subl	KK, %eax
	sall	$ZBASE_SHIFT, %eax
	leal	(AA, %eax, 1), AA
	leal	(BB, %eax, 1), BB
#endif

#ifdef LN
	subl	$1, KK
#endif

#ifdef LT
	addl	$1, KK
#endif

#ifdef RT
	movl	K, %eax
	sall	$ZBASE_SHIFT, %eax
	addl	%eax, AORIG
#endif
	ALIGN_4

.L130:
	movl	M,  %ebx
	sarl	$1, %ebx
	jle	.L149
	ALIGN_4

.L110:
#ifdef LN
       movl	K, %eax
       sall	$1 + ZBASE_SHIFT, %eax
       subl	%eax, AORIG
#endif

#if defined(LN) || defined(RT)
	movl	KK, %eax
	movl	AORIG, AA
	sall	$1 + ZBASE_SHIFT, %eax
	addl	%eax, AA
#endif

	movl	B, BB

#if defined(LN) || defined(RT)
	movl	KK, %eax
	sall	$ZBASE_SHIFT, %eax
	addl	%eax, BB
#endif	

	movaps	-32 * SIZE(AA), %xmm0
	pxor	%xmm2, %xmm2
	movsd	-32 * SIZE(BB), %xmm1
	pxor	%xmm3, %xmm3
	movhps	-30 * SIZE(BB), %xmm1
	pxor	%xmm4, %xmm4
#ifdef LN
	prefetcht0	-4 * SIZE(CO1)
#else
	prefetcht0	 3 * SIZE(CO1)
#endif
	pxor	%xmm5, %xmm5
	pxor	%xmm6, %xmm6
	pxor	%xmm7, %xmm7

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	sarl	$3, %eax
	je	.L112
	ALIGN_4

.L111:
	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	PREFETCH (PREFETCHSIZE +  0) * SIZE(AA)
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	mulps	%xmm0, %xmm3
	movaps	-28 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm4
	pshufd	$0xaa, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0xff, %xmm1, %xmm3
	movaps	-28 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movaps	-24 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	mulps	%xmm0, %xmm3
	movaps	-20 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm4
	pshufd	$0xaa, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0xff, %xmm1, %xmm3
	movaps	-24 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movaps	-16 * SIZE(AA), %xmm0

	PREFETCH (PREFETCHSIZE + 16) * SIZE(AA)

	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	mulps	%xmm0, %xmm3
	movaps	-12 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm4
	pshufd	$0xaa, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0xff, %xmm1, %xmm3
	movaps	-20 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movaps	 -8 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	mulps	%xmm0, %xmm3
	movaps	 -4 * SIZE(AA), %xmm0

	addps	%xmm2, %xmm4
	pshufd	$0xaa, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0xff, %xmm1, %xmm3
	movaps	-16 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movaps	  0 * SIZE(AA), %xmm0

	subl   $-32 * SIZE, AA
	subl   $-16 * SIZE, BB
	
	decl   %eax
	jne	.L111
	ALIGN_4
	
.L112:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K,  %eax
	subl	KK, %eax
#endif
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L114
	ALIGN_4

.L113:
	addps	%xmm2, %xmm4
	pshufd	$0x00, %xmm1, %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm3, %xmm5
	pshufd	$0x55, %xmm1, %xmm3
	movsd	-30 * SIZE(BB), %xmm1
	mulps	%xmm0, %xmm3
	movaps	-28 * SIZE(AA), %xmm0

	addl	$4 * SIZE, AA
	addl	$2 * SIZE, BB
	decl	%eax
	jg	.L113
	ALIGN_4

.L114:
#if defined(LN) || defined(RT)
	movl	KK, %eax
#ifdef LN
	subl	$2, %eax
#else
	subl	$1, %eax
#endif

	movl	AORIG, AA
	sall	$ZBASE_SHIFT, %eax
	leal	(AA, %eax, 2), AA
	leal	(B,  %eax, 1), BB
#endif

	addps	%xmm2, %xmm4
	addps	%xmm3, %xmm5

	pshufd	 $0xb1, %xmm5, %xmm5
	pcmpeqb	%xmm0, %xmm0
	psllq	$63,   %xmm0

#ifndef CONJ
	shufps	$0xb1, %xmm0, %xmm0

	pxor	%xmm0, %xmm5
#else
#if defined(LN) || defined(LT)
	pxor	%xmm0, %xmm4
#else
	pxor	%xmm0, %xmm5
#endif
#endif
	
	addps	%xmm5, %xmm4

#if defined(LN) || defined(LT)
	movaps	%xmm4, %xmm5
	unpcklpd %xmm6, %xmm4
	unpckhpd %xmm6, %xmm5

	movsd	-32 * SIZE(BB), %xmm2
	movsd	-30 * SIZE(BB), %xmm3

	subps	%xmm4,  %xmm2
	subps	%xmm5,  %xmm3
#else
	movaps	-32 * SIZE(AA), %xmm1

	subps	%xmm4,  %xmm1
#endif

#ifdef LN
	movaps	-28 * SIZE(AA), %xmm5

	pshufd	 $0xee, %xmm5, %xmm6
	pshufd	 $0xbb, %xmm5, %xmm7

	pshufd	 $0xa0, %xmm3, %xmm4
	pshufd	 $0xf5, %xmm3, %xmm3

#ifndef CONJ
	xorps	 %xmm0, %xmm3
#else
	xorps	 %xmm0, %xmm4
#endif

	mulps	%xmm6,  %xmm4
	mulps	%xmm7, %xmm3
	addps	%xmm4,  %xmm3

	pshufd	 $0x44, %xmm5, %xmm6
	pshufd	 $0x11, %xmm5, %xmm7

	pshufd	 $0xa0, %xmm3, %xmm4
	pshufd	 $0xf5, %xmm3, %xmm1

#ifndef CONJ
	xorps	 %xmm0, %xmm1
#else
	xorps	 %xmm0, %xmm4
#endif

	mulps	%xmm6,  %xmm4
	mulps	%xmm7, %xmm1
	subps	%xmm4,  %xmm2
	subps	%xmm1,  %xmm2

	movaps	-32 * SIZE(AA), %xmm5

	pshufd	 $0x44, %xmm5, %xmm6
	pshufd	 $0x11, %xmm5, %xmm7

	pshufd	 $0xa0, %xmm2, %xmm4
	pshufd	 $0xf5, %xmm2, %xmm2

#ifndef CONJ
	xorps	 %xmm0, %xmm2
#else
	xorps	 %xmm0, %xmm4
#endif

	mulps	%xmm6,  %xmm4
	mulps	%xmm7, %xmm2
	addps	%xmm4,  %xmm2
#endif

#ifdef LT
	movaps	-32 * SIZE(AA), %xmm5

	pshufd	 $0x44, %xmm5, %xmm6
	pshufd	 $0x11, %xmm5, %xmm7

	pshufd	 $0xa0, %xmm2, %xmm4
	pshufd	 $0xf5, %xmm2, %xmm2

#ifndef CONJ
	xorps	 %xmm0, %xmm2
#else
	xorps	 %xmm0, %xmm4
#endif

	mulps	%xmm6,  %xmm4
	mulps	%xmm7, %xmm2
	addps	%xmm4,  %xmm2

	pshufd	 $0xee, %xmm5, %xmm6
	pshufd	 $0xbb, %xmm5, %xmm7

	pshufd	 $0xa0, %xmm2, %xmm4
	pshufd	 $0xf5, %xmm2, %xmm1

#ifndef CONJ
	xorps	 %xmm0, %xmm1
#else
	xorps	 %xmm0, %xmm4
#endif

	mulps	%xmm6,  %xmm4
	mulps	%xmm7, %xmm1
	subps	%xmm4,  %xmm3
	subps	%xmm1,  %xmm3

	movaps	-28 * SIZE(AA), %xmm5

	pshufd	 $0xee, %xmm5, %xmm6
	pshufd	 $0xbb, %xmm5, %xmm7

	pshufd	 $0xa0, %xmm3, %xmm4
	pshufd	 $0xf5, %xmm3, %xmm3

#ifndef CONJ
	xorps	 %xmm0, %xmm3
#else
	xorps	 %xmm0, %xmm4
#endif

	mulps	%xmm6,  %xmm4
	mulps	%xmm7,  %xmm3
	addps	%xmm4,  %xmm3
#endif

#if defined(RN) || defined(RT)
	movaps	-32 * SIZE(BB), %xmm4

	pshufd	 $0x44, %xmm4, %xmm6
	pshufd	 $0x11, %xmm4, %xmm7

	pshufd	 $0xa0, %xmm1, %xmm3
	pshufd	 $0xf5, %xmm1, %xmm1

#ifndef CONJ
	xorps	 %xmm0, %xmm1
#else
	xorps	 %xmm0, %xmm3
#endif

	mulps	%xmm6,  %xmm3
	mulps	%xmm7,  %xmm1

	addps	%xmm3,  %xmm1
#endif

#ifdef LN
	subl	$4 * SIZE, CO1
#endif

#if defined(LN) || defined(LT)
	movlps	%xmm2, -32 * SIZE(BB)
	movlps	%xmm3, -30 * SIZE(BB)

	movlps	%xmm2,   0 * SIZE(CO1)
	movlps	%xmm3,   2 * SIZE(CO1)
#else
	movaps	%xmm1, -32 * SIZE(AA)

	movlps	%xmm1,   0 * SIZE(CO1)
	movhps	%xmm1,   2 * SIZE(CO1)
#endif

#ifndef LN
	addl	$4 * SIZE, CO1
#endif

#if defined(LT) || defined(RN)
	movl	K,  %eax
	subl	KK, %eax
	sall	$ZBASE_SHIFT, %eax
	leal	(AA, %eax, 2), AA
	leal	(BB, %eax, 1), BB
#endif

#ifdef LN
	subl	$2, KK
#endif

#ifdef LT
	addl	$2, KK
#endif

#ifdef RT
	movl	K, %eax
	sall	$1 + ZBASE_SHIFT, %eax
	addl	%eax, AORIG
#endif

	decl	%ebx			# i --
	jg	.L110
	ALIGN_4

.L149:
#ifdef LN
       movl	K, %eax
       sall	$ZBASE_SHIFT, %eax
       addl	%eax, B
#endif

#if defined(LT) || defined(RN)
	movl	BB, B
#endif

#ifdef RN
	addl	$1, KK
#endif

#ifdef RT
	subl	$1, KK
#endif
	ALIGN_4

.L999:
	popl	%ebx
	popl	%esi
	popl	%edi
	popl	%ebp

	addl	$ARGS, %esp
	ret

	EPILOGUE