📄 ztrsm_kernel_lt_4x2_sse.s

📁 Optimized GotoBLAS libraries
💻 S
📖 第 1 页 / 共 5 页
字号:
	movaps	 4 * SIZE(B), %xmm8	pshufd	 $0xee, %xmm8, %xmm9	pshufd	 $0xbb, %xmm8, %xmm10	pshufd	 $0xa0, %xmm5, %xmm4	pshufd	 $0xf5, %xmm5, %xmm5#ifndef CONJ	xorps	 %xmm15, %xmm5#else	xorps	 %xmm15, %xmm4#endif	mulps	%xmm9,  %xmm4	mulps	%xmm10, %xmm5	addps	%xmm4,  %xmm5#endif#ifdef RT	movaps	 4 * SIZE(B), %xmm8	pshufd	 $0xee, %xmm8, %xmm9	pshufd	 $0xbb, %xmm8, %xmm10	pshufd	 $0xa0, %xmm5, %xmm0	pshufd	 $0xf5, %xmm5, %xmm5#ifndef CONJ	xorps	 %xmm15, %xmm5#else	xorps	 %xmm15, %xmm0#endif	mulps	%xmm9,  %xmm0	mulps	%xmm10, %xmm5	addps	%xmm0,  %xmm5	pshufd	 $0x44, %xmm8, %xmm9	pshufd	 $0x11, %xmm8, %xmm10	pshufd	 $0xa0, %xmm5, %xmm0	pshufd	 $0xf5, %xmm5, %xmm2#ifndef CONJ	xorps	 %xmm15, %xmm2#else	xorps	 %xmm15, %xmm0#endif	mulps	%xmm9,  %xmm0	mulps	%xmm10, %xmm2	subps	%xmm0,  %xmm1	subps	%xmm2,  %xmm1	movaps	 0 * SIZE(B), %xmm8	pshufd	 $0x44, %xmm8, %xmm9	pshufd	 $0x11, %xmm8, %xmm10	pshufd	 $0xa0, %xmm1, %xmm4	pshufd	 $0xf5, %xmm1, %xmm1#ifndef CONJ	xorps	 %xmm15, %xmm1#else	xorps	 %xmm15, %xmm4#endif	mulps	%xmm9,  %xmm4	mulps	%xmm10, %xmm1	addps	%xmm4,  %xmm1#endif#ifdef LN	subq	$2 * SIZE, CO1	subq	$2 * SIZE, CO2#endif#if defined(LN) || defined(LT)	movaps	%xmm2,   0 * SIZE(B)	pshufd	$0x00, %xmm2, %xmm0	pshufd	$0x55, %xmm2, %xmm1	pshufd	$0xaa, %xmm2, %xmm4	pshufd	$0xff, %xmm2, %xmm5	movaps	%xmm0,   0 * SIZE(BO)	movaps	%xmm1,   4 * SIZE(BO)	movaps	%xmm4,   8 * SIZE(BO)	movaps	%xmm5,  12 * SIZE(BO)	movlps	%xmm2,   0 * SIZE(CO1)	movhps	%xmm2,   0 * SIZE(CO2)#else	movlps	%xmm1,   0 * SIZE(AO)	movlps	%xmm5,   2 * SIZE(AO)	movlps	%xmm1,   0 * SIZE(CO1)	movlps	%xmm5,   0 * SIZE(CO2)#endif#ifndef LN	addq	$2 * SIZE, CO1	addq	$2 * SIZE, CO2#endif#if defined(LT) || defined(RN)	movq	K,  %rax	subq	KK, %rax	salq	$ZBASE_SHIFT, %rax	leaq	(AO, %rax, 1), AO#ifdef LT	addq	$4 * SIZE, B#endif#endif#ifdef LN	subq	$1, KK	movq	BORIG, B#endif#ifdef LT	addq	$1, KK#endif#ifdef RT	movq	K, %rax	movq	BORIG, B	salq	$ZBASE_SHIFT, %rax	addq	%rax, AORIG#endif	ALIGN_4	.L39:#ifdef LN       leaq	(, K, SIZE), %rax       leaq	(B, %rax, 2 * COMPSIZE), B#endif#if defined(LT) || defined(RN)	movq	K,  %rax	subq	KK, %rax	leaq	(,%rax, SIZE), %rax	leaq	(B,  %rax, 2 * COMPSIZE), B#endif#ifdef RN	addq	$2, KK#endif#ifdef RT	subq	$2, KK#endif	decq	J			# j --	jg	.L01	ALIGN_4.L40:	testq	$1, N	je	.L999	ALIGN_4#ifdef LN	movq	OFFSET, %rax	addq	M, %rax	movq	%rax, KK#endif		leaq	BUFFER, BO#ifdef RT       movq	K, %rax       salq	$ZBASE_SHIFT, %rax       subq	%rax, B#endif#if defined(LN) || defined(RT)	movq	KK, %rax	movq	B, BORIG	salq	$ZBASE_SHIFT, %rax	leaq	(B,  %rax, 1), B	leaq	(BO, %rax, 4), BO#endif	#if defined(LT)	movq	OFFSET, %rax	movq	%rax, KK#endif#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	sarq	$2, %rax	jle	.L43	ALIGN_4	.L42:	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	movaps	%xmm0,  0 * SIZE(BO)	movaps	%xmm1,  4 * SIZE(BO)	movaps	%xmm2,  8 * SIZE(BO)	movaps	%xmm3, 12 * SIZE(BO)	pshufd	 $0x00, %xmm7, %xmm4	pshufd	 $0x55, %xmm7, %xmm5	pshufd	 $0xaa, %xmm7, %xmm6	pshufd	 $0xff, %xmm7, %xmm7	movaps	%xmm4, 16 * SIZE(BO)	movaps	%xmm5, 20 * SIZE(BO)	movaps	%xmm6, 24 * SIZE(BO)	movaps	%xmm7, 28 * SIZE(BO)	addq	$ 8 * SIZE, B	addq	$32 * SIZE, BO	decq	%rax	jne	.L42	ALIGN_4.L43:#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	andq	$3, %rax	BRANCH	jle	.L50	ALIGN_4.L44:	movlps	 0 * SIZE(B), %xmm3	pshufd	 $0x00, %xmm3, %xmm0	pshufd	 $0x55, %xmm3, %xmm1	movaps	%xmm0,  0 * SIZE(BO)	movaps	%xmm1,  4 * SIZE(BO)	addq	$2 * SIZE, B	addq	$8 * SIZE, BO	decq	%rax	jne	.L44	ALIGN_4	.L50:#if defined(LT) || defined(RN)	movq	A, AO#else	movq	A, AORIG#endif#ifdef RT       subq	LDC, C#endif	movq	C, CO1			# coffset1 = c#ifndef RT	addq	LDC, C#endif	movq	M,  I	sarq	$2, I		# i = (m >> 2)	jle	.L60	ALIGN_4.L51:#ifdef LN       movq	K, %rax       salq	$2 + ZBASE_SHIFT, %rax       subq	%rax, AORIG#endif#if defined(LN) || defined(RT)	movq	KK, %rax	movq	AORIG, AO	salq	$ZBASE_SHIFT, %rax	leaq	(AO, %rax, 4), AO#endif	leaq	BUFFER, BO#if defined(LN) || defined(RT)	movq	KK, %rax	salq	$ZBASE_SHIFT, %rax	leaq	(BO, %rax, 4), BO#endif		movaps	 0 * SIZE(AO), %xmm8	movaps	16 * SIZE(AO), %xmm10	movaps	32 * SIZE(AO), %xmm12	movaps	48 * SIZE(AO), %xmm14	movaps	 0 * SIZE(BO), %xmm9	movaps	16 * SIZE(BO), %xmm11	movaps	32 * SIZE(BO), %xmm13	movaps	48 * SIZE(BO), %xmm15	pxor	%xmm0, %xmm0	pxor	%xmm1, %xmm1	pxor	%xmm2, %xmm2	pxor	%xmm3, %xmm3	pxor	%xmm4, %xmm4	pxor	%xmm5, %xmm5	pxor	%xmm6, %xmm6	pxor	%xmm7, %xmm7	PREFETCHW      4 * SIZE(CO1)#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	sarq	$3, %rax	je	.L55	ALIGN_4.L52:	mulps	%xmm8, %xmm9#if defined(OPTERON) && defined(HAVE_PREFETCH)	PREFETCH	(PREFETCHSIZE +  0) * SIZE(AO)#endif	mulps	 4 * SIZE(BO), %xmm8	addps	%xmm9, %xmm0	movaps	 0 * SIZE(BO), %xmm9	addps	%xmm8, %xmm1	movaps	 4 * SIZE(AO), %xmm8	mulps	%xmm8, %xmm9	mulps	 4 * SIZE(BO), %xmm8	addps	%xmm9, %xmm4	movaps	 8 * SIZE(BO), %xmm9	addps	%xmm8, %xmm5	movaps	 8 * SIZE(AO), %xmm8	mulps	%xmm8, %xmm9	mulps	12 * SIZE(BO), %xmm8	addps	%xmm9, %xmm0	movaps	 8 * SIZE(BO), %xmm9	addps	%xmm8, %xmm1	movaps	12 * SIZE(AO), %xmm8	mulps	%xmm8, %xmm9	mulps	12 * SIZE(BO), %xmm8	addps	%xmm9, %xmm4	movaps	64 * SIZE(BO), %xmm9	addps	%xmm8, %xmm5	movaps	64 * SIZE(AO), %xmm8#if defined(OPTERON) && defined(HAVE_PREFETCH)	PREFETCH	(PREFETCHSIZE + 16) * SIZE(AO)#endif	mulps	%xmm10, %xmm11	mulps	20 * SIZE(BO), %xmm10	addps	%xmm11, %xmm0	movaps	16 * SIZE(BO), %xmm11	addps	%xmm10, %xmm1	movaps	20 * SIZE(AO), %xmm10	mulps	%xmm10, %xmm11	mulps	20 * SIZE(BO), %xmm10	addps	%xmm11, %xmm4	movaps	24 * SIZE(BO), %xmm11	addps	%xmm10, %xmm5	movaps	24 * SIZE(AO), %xmm10	mulps	%xmm10, %xmm11	mulps	28 * SIZE(BO), %xmm10	addps	%xmm11, %xmm0	movaps	24 * SIZE(BO), %xmm11	addps	%xmm10, %xmm1	movaps	28 * SIZE(AO), %xmm10	mulps	%xmm10, %xmm11	mulps	28 * SIZE(BO), %xmm10	addps	%xmm11, %xmm4	movaps	 80 * SIZE(BO), %xmm11	addps	%xmm10, %xmm5	movaps	 80 * SIZE(AO), %xmm10#if defined(OPTERON) && defined(HAVE_PREFETCH)	PREFETCH	(PREFETCHSIZE + 32) * SIZE(AO)#endif	mulps	%xmm12, %xmm13	mulps	36 * SIZE(BO), %xmm12	addps	%xmm13, %xmm0	movaps	32 * SIZE(BO), %xmm13	addps	%xmm12, %xmm1	movaps	36 * SIZE(AO), %xmm12	mulps	%xmm12, %xmm13	mulps	36 * SIZE(BO), %xmm12	addps	%xmm13, %xmm4	movaps	40 * SIZE(BO), %xmm13	addps	%xmm12, %xmm5	movaps	40 * SIZE(AO), %xmm12	mulps	%xmm12, %xmm13	mulps	44 * SIZE(BO), %xmm12	addps	%xmm13, %xmm0	movaps	40 * SIZE(BO), %xmm13	addps	%xmm12, %xmm1	movaps	44 * SIZE(AO), %xmm12	mulps	%xmm12, %xmm13	mulps	44 * SIZE(BO), %xmm12	addps	%xmm13, %xmm4	movaps	 96 * SIZE(BO), %xmm13	addps	%xmm12, %xmm5	movaps	 96 * SIZE(AO), %xmm12#if defined(OPTERON) && defined(HAVE_PREFETCH)	PREFETCH	(PREFETCHSIZE + 48) * SIZE(AO)#endif	mulps	%xmm14, %xmm15	mulps	52 * SIZE(BO), %xmm14	addps	%xmm15, %xmm0	movaps	48 * SIZE(BO), %xmm15	addps	%xmm14, %xmm1	movaps	52 * SIZE(AO), %xmm14	mulps	%xmm14, %xmm15	mulps	52 * SIZE(BO), %xmm14	addps	%xmm15, %xmm4	movaps	56 * SIZE(BO), %xmm15	addps	%xmm14, %xmm5	movaps	56 * SIZE(AO), %xmm14	mulps	%xmm14, %xmm15	mulps	60 * SIZE(BO), %xmm14	addps	%xmm15, %xmm0	movaps	56 * SIZE(BO), %xmm15	addps	%xmm14, %xmm1	movaps	60 * SIZE(AO), %xmm14	mulps	%xmm14, %xmm15	mulps	60 * SIZE(BO), %xmm14	addps	%xmm15, %xmm4	movaps	112 * SIZE(BO), %xmm15	addps	%xmm14, %xmm5	movaps	112 * SIZE(AO), %xmm14	addq   $64 * SIZE, AO	addq   $64 * SIZE, BO	decq   %rax	jne    .L52	ALIGN_4.L55:#if defined(LT) || defined(RN)	movq	KK, %rax#else	movq	K, %rax	subq	KK, %rax#endif	movaps	POSINV,  %xmm15	andq	$7, %rax		# if (k & 1)	BRANCH	je .L58	ALIGN_4.L56:	mulps	%xmm8, %xmm9	mulps	 4 * SIZE(BO), %xmm8	addps	%xmm9, %xmm0	movaps	 0 * SIZE(BO), %xmm9	addps	%xmm8, %xmm1	movaps	 4 * SIZE(AO), %xmm8	mulps	%xmm8, %xmm9	mulps	 4 * SIZE(BO), %xmm8	addps	%xmm9, %xmm4	movaps	 8 * SIZE(BO), %xmm9	addps	%xmm8, %xmm5	movaps	 8 * SIZE(AO), %xmm8	addq	$ 8 * SIZE, AO		# aoffset  += 4	addq	$ 8 * SIZE, BO		# boffset1 += 8	decq	%rax	jg	.L56	ALIGN_4.L58:	shufps	$0xb1, %xmm1, %xmm1	shufps	$0xb1, %xmm5, %xmm5#if defined(LN) || defined(LT)#ifndef CONJ	xorps	 %xmm15, %xmm1	xorps	 %xmm15, %xmm5#else	xorps	 %xmm15, %xmm0	xorps	 %xmm15, %xmm4#endif#else	xorps	 %xmm15, %xmm1	xorps	 %xmm15, %xmm5#endif	addps	%xmm1, %xmm0	addps	%xmm5, %xmm4#if defined(LN) || defined(RT)	movq	KK, %rax#ifdef LN	subq	$4, %rax#else	subq	$1, %rax#endif	movq	AORIG, AO	movq	BORIG, B	leaq	BUFFER, BO	salq	$ZBASE_SHIFT, %rax	leaq	(AO, %rax, 4), AO	leaq	(B,  %rax, 1), B	leaq	(BO, %rax, 4), BO#endif#if defined(LN) || defined(LT)	movaps	%xmm0, %xmm1	unpcklpd %xmm2, %xmm0	unpckhpd %xmm2, %xmm1	movaps	%xmm4, %xmm5	unpcklpd %xmm6, %xmm4	unpckhpd %xmm6, %xmm5	movlps	 0 * SIZE(B), %xmm2	movlps	 2 * SIZE(B), %xmm3	movlps	 4 * SIZE(B), %xmm6	movlps	 6 * SIZE(B), %xmm7	subps	%xmm0,  %xmm2	subps	%xmm1,  %xmm3	subps	%xmm4,  %xmm6	subps	%xmm5,  %xmm7#else	movaps	 0 * SIZE(AO), %xmm1	movaps	 4 * SIZE(AO), %xmm3	subps	%xmm0,  %xmm1	subps	%xmm4,  %xmm3#endif#ifdef LN	movaps	28 * SIZE(AO), %xmm8	pshufd	 $0xee, %xmm8, %xmm9	pshufd	 $0xbb, %xmm8, %xmm10	pshufd	 $0xa0, %xmm7, %xmm0	pshufd	 $0xf5, %xmm7, %xmm7#ifndef CONJ	xorps	 %xmm15, %xmm7#else	xorps	 %xmm15, %xmm0#endif	mulps	%xmm9,  %xmm0	mulps	%xmm10, %xmm7	addps	%xmm0,  %xmm7	pshufd	 $0x44, %xmm8, %xmm9	pshufd	 $0x11, %xmm8, %xmm10	pshufd	 $0xa0, %xmm7, %xmm0	pshufd	 $0xf5, %xmm7, %xmm1#ifndef CONJ	xorps	 %xmm15, %xmm1#else	xorps	 %xmm15, %xmm0#endif	mulps	%xmm9,  %xmm0	mulps	%xmm10, %xmm1	subps	%xmm0,  %xmm6	subps	%xmm1,  %xmm6	movaps	24 * SIZE(AO), %xmm8	pshufd	 $0xee, %xmm8, %xmm9	pshufd	 $0xbb, %xmm8, %xmm10	pshufd	 $0xa0, %xmm7, %xmm0	pshufd	 $0xf5, %xmm7, %xmm1#ifndef CONJ	xorps	 %xmm15, %xmm1#else	xorps	 %xmm15, %xmm0#endif	mulps	%xmm9,  %xmm0	mulps	%xmm10, %xmm1	subps	%xmm0,  %xmm3	subps	%xmm1,  %xmm3	pshufd	 $0x44, %xmm8, %xmm9	pshufd	 $0x11, %xmm8, %xmm10	pshufd	 $0xa0, %xmm7, %xmm0	pshufd	 $0xf5, %xmm7, %xmm1#ifndef CONJ	xorps	 %xmm15, %xmm1#else	xorps	 %xmm15, %xmm0#endif	mulps	%xmm9,  %xmm0	mulps	%xmm10, %xmm1	subps	%xmm0,  %xmm2	subps	%xmm1,  %xmm2	movaps	20 * SIZE(AO), %xmm8	pshufd	 $0x44, %xmm8, %xmm9	pshufd	 $0x11, %xmm8, %xmm10	pshufd	 $0xa0, %xmm6, %xmm0	pshufd	 $0xf5, %xmm6, %xmm6#ifndef CONJ	xorps	 %xmm15, %xmm6#else	xorps	 %xmm15, %xmm0#endif	mulps	%xmm9,  %xmm0	mulps	%xmm10, %xmm6	addps	%xmm0,  %xmm6	movaps	16 * SIZE(AO), %xmm8	pshufd	 $0xee, %xmm8, %xmm9	pshufd	 $0xbb, %xmm8, %xmm10	pshufd	 $0xa0, %xmm6, %xmm0	pshufd	 $0xf5, %xmm6, %xmm1#ifndef CONJ	xorps	 %xmm15, %xmm1#else	xorps	 %xmm15, %xmm0#endif	mulps	%xmm9,  %xmm0	mulps	%xmm10, %xmm1	subps	%xmm0,  %xmm3	subps	%xmm1,  %xmm3	pshufd	 $0x44, %xmm8, %xmm9	pshufd	 $0x11, %xmm8, %xmm10	pshufd	 $0xa0, %xmm6, %xmm0	pshufd	 $0xf5, %xmm6, %xmm1#ifndef CONJ	xorps	 %xmm15, %xmm1#else	xorps	 %xmm15, %xmm0#endif	mulps	%xmm9,  %xmm0	mulps	%xmm10, %xmm1	subps	%xmm0,  %xmm2	subps	%xmm1,  %xmm2	movaps	 8 * SIZE(AO), %xmm8	pshufd	 $0xee, %xmm8, %xmm9	pshufd	 $0xbb, %xmm8, %xmm10	pshufd	 $0xa0, %xmm3, %xmm0	pshufd	 $0xf5, %xmm3, %xmm3#ifndef CONJ	xorps	 %xmm15, %xmm3#else	xorps	 %xmm15, %xmm0#endif	mulps	%xmm9,  %xmm0	mulps	%xmm10, %xmm3	addps	%xmm0,  %xmm3	pshufd	 $0x44, %xmm8, %xmm9	pshufd	 $0x11, %xmm8, %xmm10	pshufd	 $0xa0, %xmm3, %xmm0	pshufd	 $0xf5, %xmm3, %xmm1#ifndef CONJ	xorps	 %xmm15, %xmm1#else	xorps	 %xmm15, %xmm0#endif	mulps	%xmm9,  %xmm0	mulps	%xmm10, %xmm1	subps	%xmm0,  %xmm2	subps	%xmm1,  %xmm2	movaps	 0 * SIZE(AO), %xmm8	pshufd	 $0x44, %xmm8, %xmm9	pshufd	 $0x11, %xmm8, %xmm10	pshufd	 $0xa0, %xmm2, %xmm0	pshufd	 $0xf5, %xmm2, %xmm2#ifndef CONJ	xorps	 %xmm15, %xmm2#else	xorps	 %xmm15, %xmm0#endif	mulps	%xmm9,  %xmm0	mulps	%xmm10, %xmm2	addps	%xmm0,  %xmm2
⌨️ 快捷键说明

复制代码 Ctrl + C
搜索代码 Ctrl + F
全屏模式 F11
切换主题 Ctrl + Shift + D
显示快捷键 ?
增大字号 Ctrl + =
减小字号 Ctrl + -