trsm_kernel_rt_4x4_sse.s

Optimized GotoBLAS libraries

	movaps	%xmm2,   4 * SIZE(BB)
#else
	movss	%xmm0,   0 * SIZE(AA)
	movss	%xmm1,   1 * SIZE(AA)
#endif

#ifdef LN
	subl	$1 * SIZE, CO1
#endif

#if defined(LN) || defined(LT)
	pshufd	$1, %xmm1, %xmm3

	movss	%xmm1, 0 * SIZE(CO1)
	movss	%xmm3, 0 * SIZE(CO1, LDC)
#else
	movss	%xmm0, 0 * SIZE(CO1)
	movss	%xmm1, 0 * SIZE(CO1, LDC)
#endif

#ifndef LN
	addl	$1 * SIZE, CO1
#endif

#if defined(LT) || defined(RN)
	movl	K,  %eax
	subl	KK, %eax
	leal	(AA, %eax, SIZE), AA
#ifdef LT
	addl	$2 * SIZE, B
#endif
#endif

#ifdef LN
	subl	$1, KK
	movl	BORIG, B
#endif

#ifdef LT
	addl	$1, KK
#endif

#ifdef RT
	movl	K, %eax
	movl	BORIG, B
	sall	$BASE_SHIFT, %eax
	addl	%eax, AORIG
#endif
	ALIGN_4

.L79:
#ifdef LN
       movl	K, %eax
       leal	(, %eax, SIZE), %eax
       leal 	(B, %eax, 2), B
#endif

#if defined(LT) || defined(RN)
	movl	K,  %eax
	subl	KK, %eax
	leal	(,%eax, SIZE), %eax
	leal	(B,  %eax, 2), B
#endif

#ifdef RN
	addl	$2, KK
#endif

#ifdef RT
	subl	$2, KK
#endif
	ALIGN_4

.L80:
	movl	N, %eax
	sarl	$2, %eax
	movl	%eax, J
	jle	.L999

.L01:
#ifdef LN
	movl	OFFSET, %eax
	addl	M, %eax
	movl	%eax, KK
#endif	

	leal	BUFFER, %ecx

#ifdef RT
       movl	K, %eax
       sall	$2 + BASE_SHIFT, %eax
       subl	%eax, B
#endif

#if defined(LN) || defined(RT)
	movl	KK, %eax
	movl	B, BORIG
        sall	$2 + BASE_SHIFT, %eax
	leal	(B,  %eax, 1), B
	leal	(BB, %eax, 4), BB
#endif	

#ifdef LT
	movl	OFFSET, %eax
	movl	%eax, KK
#endif

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	sarl	$1, %eax
	jle	.L05
	ALIGN_4
	
.L02:
	movaps	 0 * SIZE(B), %xmm3
	movaps	 4 * SIZE(B), %xmm7

	pshufd	 $0x00, %xmm3, %xmm0
	pshufd	 $0x55, %xmm3, %xmm1
	pshufd	 $0xaa, %xmm3, %xmm2
	pshufd	 $0xff, %xmm3, %xmm3

	pshufd	 $0x00, %xmm7, %xmm4
	pshufd	 $0x55, %xmm7, %xmm5
	pshufd	 $0xaa, %xmm7, %xmm6
	pshufd	 $0xff, %xmm7, %xmm7

	movaps	%xmm0,  0 * SIZE(BB)
	movaps	%xmm1,  4 * SIZE(BB)
	movaps	%xmm2,  8 * SIZE(BB)
	movaps	%xmm3, 12 * SIZE(BB)
	movaps	%xmm4, 16 * SIZE(BB)
	movaps	%xmm5, 20 * SIZE(BB)
	movaps	%xmm6, 24 * SIZE(BB)
	movaps	%xmm7, 28 * SIZE(BB)

	addl	$ 8 * SIZE, B
	addl	$32 * SIZE, %ecx
	decl	%eax
	jne	.L02
	ALIGN_2

.L05:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$1, %eax
	BRANCH
	jle	.L10

	movaps	 0 * SIZE(B), %xmm3

	pshufd	 $0x00, %xmm3, %xmm0
	pshufd	 $0x55, %xmm3, %xmm1
	pshufd	 $0xaa, %xmm3, %xmm2
	pshufd	 $0xff, %xmm3, %xmm3

	movaps	%xmm0,  0 * SIZE(BB)
	movaps	%xmm1,  4 * SIZE(BB)
	movaps	%xmm2,  8 * SIZE(BB)
	movaps	%xmm3, 12 * SIZE(BB)

	addl	$4 * SIZE, B
	ALIGN_4
	
.L10:
#if defined(LT) || defined(RN)
	movl	A, AA
#else
	movl	A, %eax
	movl	%eax, AORIG
#endif

	leal	(, LDC, 4), %eax

#ifdef RT
	subl	%eax, C
#endif
	movl	C, CO1
#ifndef RT
	addl	%eax, C
#endif

	movl	M,  %ebx
	sarl	$2, %ebx	# i = (m >> 2)
	jle	.L20
	ALIGN_4

.L11:
#ifdef LN
       movl	K, %eax
       sall	$2 + BASE_SHIFT, %eax
       subl	%eax, AORIG
#endif

#if defined(LN) || defined(RT)
	movl	KK, %eax
	movl	AORIG, AA
	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 4), AA
#endif

	leal	BUFFER, BB

#if defined(LN) || defined(RT)
	movl	KK, %eax
	sall	$2 + BASE_SHIFT, %eax
	leal	(BB, %eax, 4), BB
#endif	

	movaps	 0 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movaps	16 * SIZE(AA), %xmm1
	pxor	%xmm5, %xmm5
	movaps	 0 * SIZE(BB), %xmm2
	pxor	%xmm6, %xmm6
	movaps	16 * SIZE(BB), %xmm3
	pxor	%xmm7, %xmm7

	leal	(LDC, LDC, 2), %eax

	PREFETCHW	3 * SIZE(CO1)
	PREFETCHW	3 * SIZE(CO1, LDC)
	PREFETCHW	3 * SIZE(CO1, LDC, 2)
	PREFETCHW	3 * SIZE(CO1, %eax)

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	sarl	$3, %eax
	je	.L15
	ALIGN_4

.L12:
	KERNEL1(0 * 16)
	KERNEL2(0 * 16)
	KERNEL3(0 * 16)
	KERNEL4(0 * 16)
	KERNEL5(0 * 16)
	KERNEL6(0 * 16)
	KERNEL7(0 * 16)
	KERNEL8(0 * 16)

	addl   $128 * SIZE, BB
	addl   $32 * SIZE, AA
	decl   %eax
	jne    .L12
	ALIGN_4

.L15:
#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	andl	$7, %eax		# if (k & 1)
	BRANCH
	je .L18
	ALIGN_4

.L16:
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	movaps	 4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm5
	movaps	 8 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	mulps	12 * SIZE(BB), %xmm0
	addps	%xmm2, %xmm6
	movaps	16 * SIZE(BB), %xmm2
	addps	%xmm0, %xmm7
	movaps	 4 * SIZE(AA), %xmm0

	addl	$ 4 * SIZE, AA
	addl	$16 * SIZE, BB
	decl	%eax
	jg	.L16
	ALIGN_4

.L18:
#if defined(LN) || defined(RT)
	movl	KK, %eax
#ifdef LN
	subl	$4, %eax
#else
	subl	$4, %eax
#endif

	movl	AORIG, AA
	movl	BORIG, B
	leal	BUFFER, BB

	sall	$2 + BASE_SHIFT, %eax
	leal	(AA, %eax, 1), AA
	leal	(B,  %eax, 1), B
	leal	(BB, %eax, 4), BB
#endif

#if defined(LN) || defined(LT)
	movaps	 %xmm4, %xmm0
	unpcklps %xmm6, %xmm4
	unpckhps %xmm6, %xmm0

	movaps	 %xmm5, %xmm1
	unpcklps %xmm7, %xmm5
	unpckhps %xmm7, %xmm1

	movaps	 %xmm4, %xmm6
	unpcklps %xmm5, %xmm4
	unpckhps %xmm5, %xmm6

	movaps	 %xmm0, %xmm2
	unpcklps %xmm1, %xmm0
	unpckhps %xmm1, %xmm2

	movaps	 0 * SIZE(B), %xmm1
	movaps	 4 * SIZE(B), %xmm3
	movaps	 8 * SIZE(B), %xmm5
	movaps	12 * SIZE(B), %xmm7

	subps	%xmm4,  %xmm1
	subps	%xmm6,  %xmm3
	subps	%xmm0,  %xmm5
	subps	%xmm2,  %xmm7
#else
	movaps	 0 * SIZE(AA), %xmm0
	movaps	 4 * SIZE(AA), %xmm1
	movaps	 8 * SIZE(AA), %xmm2
	movaps	12 * SIZE(AA), %xmm3

	subps	%xmm4, %xmm0
	subps	%xmm5, %xmm1
	subps	%xmm6, %xmm2
	subps	%xmm7, %xmm3
#endif

#ifdef LN
	movaps	 12 * SIZE(AA), %xmm4
	pshufd	 $0xff, %xmm4, %xmm6
	mulps	 %xmm6, %xmm7
	pshufd	 $0xaa, %xmm4, %xmm6
	mulps	 %xmm7, %xmm6
	subps	 %xmm6, %xmm5
	pshufd	 $0x55, %xmm4, %xmm6
	mulps	 %xmm7, %xmm6
	subps	 %xmm6, %xmm3
	pshufd	 $0x00, %xmm4, %xmm6
	mulps	 %xmm7, %xmm6
	subps	 %xmm6, %xmm1

	movaps	  8 * SIZE(AA), %xmm4
	pshufd	 $0xaa, %xmm4, %xmm6
	mulps	 %xmm6, %xmm5
	pshufd	 $0x55, %xmm4, %xmm6
	mulps	 %xmm5, %xmm6
	subps	 %xmm6, %xmm3
	pshufd	 $0x00, %xmm4, %xmm6
	mulps	 %xmm5, %xmm6
	subps	 %xmm6, %xmm1

	movaps	  4 * SIZE(AA), %xmm4
	pshufd	 $0x55, %xmm4, %xmm6
	mulps	 %xmm6, %xmm3
	pshufd	 $0x00, %xmm4, %xmm6
	mulps	 %xmm3, %xmm6
	subps	 %xmm6, %xmm1

	movaps	  0 * SIZE(AA), %xmm4
	pshufd	 $0x00, %xmm4, %xmm6
	mulps	 %xmm6, %xmm1
#endif

#ifdef LT
	movaps	 0 * SIZE(AA), %xmm4
	pshufd	 $0x00, %xmm4, %xmm6
	mulps	 %xmm6, %xmm1

	pshufd	 $0x55, %xmm4, %xmm6
	mulps	 %xmm1, %xmm6
	subps	 %xmm6, %xmm3
	pshufd	 $0xaa, %xmm4, %xmm6
	mulps	 %xmm1, %xmm6
	subps	 %xmm6, %xmm5
	pshufd	 $0xff, %xmm4, %xmm6
	mulps	 %xmm1, %xmm6
	subps	 %xmm6, %xmm7

	movaps	 4 * SIZE(AA), %xmm4
	pshufd	 $0x55, %xmm4, %xmm6
	mulps	 %xmm6, %xmm3
	pshufd	 $0xaa, %xmm4, %xmm6
	mulps	 %xmm3, %xmm6
	subps	 %xmm6, %xmm5
	pshufd	 $0xff, %xmm4, %xmm6
	mulps	 %xmm3, %xmm6
	subps	 %xmm6, %xmm7

	movaps	 8 * SIZE(AA), %xmm4
	pshufd	 $0xaa, %xmm4, %xmm6
	mulps	 %xmm6, %xmm5
	pshufd	 $0xff, %xmm4, %xmm6
	mulps	 %xmm5, %xmm6
	subps	 %xmm6, %xmm7

	movaps	12 * SIZE(AA), %xmm4
	pshufd	 $0xff, %xmm4, %xmm6
	mulps	 %xmm6, %xmm7
#endif

#ifdef RN
	movaps	 0 * SIZE(B), %xmm6
	pshufd	 $0x00, %xmm6, %xmm7
	mulps	 %xmm7, %xmm0
	pshufd	 $0x55, %xmm6, %xmm7
	mulps	 %xmm0, %xmm7
	subps	 %xmm7, %xmm1
	pshufd	 $0xaa, %xmm6, %xmm7
	mulps	 %xmm0, %xmm7
	subps	 %xmm7, %xmm2
	pshufd	 $0xff, %xmm6, %xmm7
	mulps	 %xmm0, %xmm7
	subps	 %xmm7, %xmm3

	movaps	 4 * SIZE(B), %xmm6
	pshufd	 $0x55, %xmm6, %xmm7
	mulps	 %xmm7, %xmm1
	pshufd	 $0xaa, %xmm6, %xmm7
	mulps	 %xmm1, %xmm7
	subps	 %xmm7, %xmm2
	pshufd	 $0xff, %xmm6, %xmm7
	mulps	 %xmm1, %xmm7
	subps	 %xmm7, %xmm3

	movaps	 8 * SIZE(B), %xmm6
	pshufd	 $0xaa, %xmm6, %xmm7
	mulps	 %xmm7, %xmm2
	pshufd	 $0xff, %xmm6, %xmm7
	mulps	 %xmm2, %xmm7
	subps	 %xmm7, %xmm3

	movaps	 12 * SIZE(B), %xmm6
	pshufd	 $0xff, %xmm6, %xmm7
	mulps	 %xmm7, %xmm3
#endif

#ifdef RT
	movaps	 12 * SIZE(B), %xmm6
	pshufd	 $0xff, %xmm6, %xmm7
	mulps	 %xmm7, %xmm3
	pshufd	 $0xaa, %xmm6, %xmm7
	mulps	 %xmm3, %xmm7
	subps	 %xmm7, %xmm2
	pshufd	 $0x55, %xmm6, %xmm7
	mulps	 %xmm3, %xmm7
	subps	 %xmm7, %xmm1
	pshufd	 $0x00, %xmm6, %xmm7
	mulps	 %xmm3, %xmm7
	subps	 %xmm7, %xmm0

	movaps	  8 * SIZE(B), %xmm6
	pshufd	 $0xaa, %xmm6, %xmm7
	mulps	 %xmm7, %xmm2
	pshufd	 $0x55, %xmm6, %xmm7
	mulps	 %xmm2, %xmm7
	subps	 %xmm7, %xmm1
	pshufd	 $0x00, %xmm6, %xmm7
	mulps	 %xmm2, %xmm7
	subps	 %xmm7, %xmm0

	movaps	  4 * SIZE(B), %xmm6
	pshufd	 $0x55, %xmm6, %xmm7
	mulps	 %xmm7, %xmm1
	pshufd	 $0x00, %xmm6, %xmm7
	mulps	 %xmm1, %xmm7
	subps	 %xmm7, %xmm0

	movaps	  0 * SIZE(B), %xmm6
	pshufd	 $0x00, %xmm6, %xmm7
	mulps	 %xmm7, %xmm0
#endif

#if defined(LN) || defined(LT)
	movaps	%xmm1,   0 * SIZE(B)
	movaps	%xmm3,   4 * SIZE(B)
	movaps	%xmm5,   8 * SIZE(B)
	movaps	%xmm7,  12 * SIZE(B)

	pshufd	$0x00, %xmm1, %xmm0
	pshufd	$0x55, %xmm1, %xmm2
	pshufd	$0xaa, %xmm1, %xmm4
	pshufd	$0xff, %xmm1, %xmm6
	movaps	%xmm0,   0 * SIZE(BB)
	movaps	%xmm2,   4 * SIZE(BB)
	movaps	%xmm4,   8 * SIZE(BB)
	movaps	%xmm6,  12 * SIZE(BB)

	pshufd	$0x00, %xmm3, %xmm0
	pshufd	$0x55, %xmm3, %xmm2
	pshufd	$0xaa, %xmm3, %xmm4
	pshufd	$0xff, %xmm3, %xmm6
	movaps	%xmm0,  16 * SIZE(BB)
	movaps	%xmm2,  20 * SIZE(BB)
	movaps	%xmm4,  24 * SIZE(BB)
	movaps	%xmm6,  28 * SIZE(BB)

	pshufd	$0x00, %xmm5, %xmm0
	pshufd	$0x55, %xmm5, %xmm2
	pshufd	$0xaa, %xmm5, %xmm4
	pshufd	$0xff, %xmm5, %xmm6
	movaps	%xmm0,  32 * SIZE(BB)
	movaps	%xmm2,  36 * SIZE(BB)
	movaps	%xmm4,  40 * SIZE(BB)
	movaps	%xmm6,  44 * SIZE(BB)

	pshufd	$0x00, %xmm7, %xmm0
	pshufd	$0x55, %xmm7, %xmm2
	pshufd	$0xaa, %xmm7, %xmm4
	pshufd	$0xff, %xmm7, %xmm6
	movaps	%xmm0,  48 * SIZE(BB)
	movaps	%xmm2,  52 * SIZE(BB)
	movaps	%xmm4,  56 * SIZE(BB)
	movaps	%xmm6,  60 * SIZE(BB)
#else
	movaps	%xmm0,   0 * SIZE(AA)
	movaps	%xmm1,   4 * SIZE(AA)
	movaps	%xmm2,   8 * SIZE(AA)
	movaps	%xmm3,  12 * SIZE(AA)
#endif

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

	leal	(LDC, LDC, 2), %eax

#if defined(LN) || defined(LT)
	movaps	 %xmm1, %xmm0
	unpcklps %xmm5, %xmm1
	unpckhps %xmm5, %xmm0

	movaps	 %xmm3, %xmm4
	unpcklps %xmm7, %xmm3
	unpckhps %xmm7, %xmm4

	movaps	 %xmm1, %xmm2
	unpcklps %xmm3, %xmm1
	unpckhps %xmm3, %xmm2

	movaps	 %xmm0, %xmm6
	unpcklps %xmm4, %xmm0
	unpckhps %xmm4, %xmm6

	movlps	%xmm1, 0 * SIZE(CO1)
	movhps	%xmm1, 2 * SIZE(CO1)
	movlps	%xmm2, 0 * SIZE(CO1, LDC, 1)
	movhps	%xmm2, 2 * SIZE(CO1, LDC, 1)
	movlps	%xmm0, 0 * SIZE(CO1, LDC, 2)
	movhps	%xmm0, 2 * SIZE(CO1, LDC, 2)
	movlps	%xmm6, 0 * SIZE(CO1, %eax, 1)
	movhps	%xmm6, 2 * SIZE(CO1, %eax, 1)
#else
	movlps	%xmm0, 0 * SIZE(CO1)
	movhps	%xmm0, 2 * SIZE(CO1)
	movlps	%xmm1, 0 * SIZE(CO1, LDC, 1)
	movhps	%xmm1, 2 * SIZE(CO1, LDC, 1)
	movlps	%xmm2, 0 * SIZE(CO1, LDC, 2)
	movhps	%xmm2, 2 * SIZE(CO1, LDC, 2)
	movlps	%xmm3, 0 * SIZE(CO1, %eax, 1)
	movhps	%xmm3, 2 * SIZE(CO1, %eax, 1)
#endif

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

#if defined(LT) || defined(RN)
	movl	K,  %eax
	subl	KK, %eax
	leal	(,%eax, SIZE), %eax
	leal	(AA, %eax, 4), AA
#ifdef LT
	addl	$16 * SIZE, B
#endif
#endif

#ifdef LN
	subl	$4, KK
	movl	BORIG, B
#endif

#ifdef LT
	addl	$4, KK
#endif

#ifdef RT
	movl	K, %eax
	movl	BORIG, B
	sall	$2 + BASE_SHIFT, %eax
	addl	%eax, AORIG
#endif

	decl	%ebx			# i --
	jg	.L11
	ALIGN_4

.L20:
	testl	$2, M
	je	.L30

#ifdef LN
       movl	K, %eax
       sall	$1 + BASE_SHIFT, %eax
       subl	%eax, AORIG
#endif

#if defined(LN) || defined(RT)
	movl	KK, %eax
	movl	AORIG, AA
	leal	(, %eax, SIZE), %eax
	leal	(AA, %eax, 2), AA
#endif

	leal	BUFFER, BB

#if defined(LN) || defined(RT)
	movl	KK, %eax
	sall	$2 + BASE_SHIFT, %eax
	leal	(BB, %eax, 4), BB
#endif	

	movsd	 0 * SIZE(AA), %xmm0
	pxor	%xmm4, %xmm4
	movsd	 8 * SIZE(AA), %xmm1
	pxor	%xmm5, %xmm5
	movsd	 0 * SIZE(BB), %xmm2
	pxor	%xmm6, %xmm6
	movsd	16 * SIZE(BB), %xmm3
	pxor	%xmm7, %xmm7

#if defined(LT) || defined(RN)
	movl	KK, %eax
#else
	movl	K, %eax
	subl	KK, %eax
#endif
	sarl	$3, %eax
	je	.L25
	ALIGN_4

.L22:
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
#if defined(OPTERON) || defined(BARCELONA)
	prefetcht0 (PREFETCHSIZE  + 0) * SIZE(AA)
#endif
	movsd	 4 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm5
	movsd	 8 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm6
	movsd	12 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movsd	 2 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm7
	movsd	32 * SIZE(BB), %xmm2

	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm4
	movsd	20 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm5
	movsd	24 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm6
	movsd	28 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	movsd	 4 * SIZE(AA), %xmm0
	addps	%xmm3, %xmm7
	movsd	48 * SIZE(BB), %xmm3

	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm4
	movsd	36 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm5
	movsd	40 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	addps	%xmm2, %xmm6
	movsd	44 * SIZE(BB), %xmm2
	mulps	%xmm0, %xmm2
	movsd	 6 * SIZE(AA), %xmm0
	addps	%xmm2, %xmm7
	movsd	64 * SIZE(BB), %xmm2

	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm4
	movsd	52 * SIZE(BB), %xmm3
	mulps	%xmm0, %xmm3
	addps	%xmm3, %xmm5
	movsd	56 * SIZE(BB), %xmm3