📄 zdot_sse.s

📁 Optimized GotoBLAS libraries
💻 S
字号:
/*********************************************************************//*                                                                   *//*             Optimized BLAS libraries                              *//*                     By Kazushige Goto <kgoto@tacc.utexas.edu>     *//*                                                                   *//* Copyright (c) The University of Texas, 2005. All rights reserved. *//* UNIVERSITY EXPRESSLY DISCLAIMS ANY AND ALL WARRANTIES CONCERNING  *//* THIS SOFTWARE AND DOCUMENTATION, INCLUDING ANY WARRANTIES OF      *//* MERCHANTABILITY, FITNESS FOR ANY PARTICULAR PURPOSE,              *//* NON-INFRINGEMENT AND WARRANTIES OF PERFORMANCE, AND ANY WARRANTY  *//* THAT MIGHT OTHERWISE ARISE FROM COURSE OF DEALING OR USAGE OF     *//* TRADE. NO WARRANTY IS EITHER EXPRESS OR IMPLIED WITH RESPECT TO   *//* THE USE OF THE SOFTWARE OR DOCUMENTATION.                         *//* Under no circumstances shall University be liable for incidental, *//* special, indirect, direct or consequential damages or loss of     *//* profits, interruption of business, or related expenses which may  *//* arise from use of Software or Documentation, including but not    *//* limited to those resulting from defects in Software and/or        *//* Documentation, or loss or inaccuracy of data of any kind.         *//*********************************************************************/#define ASSEMBLER#include "common.h"#if defined(F_INTERFACE) && defined(F_INTERFACE_F2C)#define RESULT	ARG1	/* rdi */#define N	ARG2	/* rsi */#define X	ARG3	/* rdx */#define INCX	ARG4	/* rcx */#ifndef WINDOWS_ABI#define Y	ARG5	/* r8  */#define INCY	ARG6	/* r9  */#else#define Y	%r10#define INCY	%r11#endif#else#define N	ARG1	/* rdi */#define X	ARG2	/* rsi */#define INCX	ARG3	/* rdx */#define Y	ARG4	/* rcx */#ifndef WINDOWS_ABI#define INCY	ARG5	/* r8  */#else#define INCY	%r10#endif#endif#ifdef BARCELONA#define PREFETCH_SIZE (16 * 8)#define movsd	     movlps#endif	PROLOGUE	PROFCODE#ifdef WINDOWS_ABI#if defined(F_INTERFACE) && defined(F_INTERFACE_F2C)		movq	40(%rsp), Y		movq	48(%rsp), INCY#else		movq	40(%rsp), INCY#endif#endif	SAVEREGISTERS#ifdef F_INTERFACE#ifndef USE64BITINT	movslq	(N), N			# N	movslq	(INCX),INCX		# INCX	movslq	(INCY),INCY		# INCY#else	movq	(N), N			# N	movq	(INCX),INCX		# INCX	movq	(INCY),INCY		# INCY#endif#endif	addq	INCX, INCX	addq	INCY, INCY	leaq	(, INCX, SIZE), INCX		leaq	(, INCY, SIZE), INCY		pxor	%xmm0, %xmm0	pxor	%xmm1, %xmm1	pxor	%xmm2, %xmm2	pxor	%xmm3, %xmm3	cmpq	$0, N	jle	.L999	cmpq	$2 * SIZE, INCX	jne	.L50	cmpq	$2 * SIZE, INCY	jne	.L50	movq	N,  %rax	sarq	$4, %rax	jle	.L15	movddup	 0 * SIZE(X), %xmm4	movddup	 2 * SIZE(X), %xmm6	movddup	 4 * SIZE(X), %xmm8	movddup	 6 * SIZE(X), %xmm10	movsd	 0 * SIZE(Y), %xmm5	movsd	 2 * SIZE(Y), %xmm7	movsd	 4 * SIZE(Y), %xmm9	movsd	 6 * SIZE(Y), %xmm11	movsd	 8 * SIZE(Y), %xmm12	movsd	10 * SIZE(Y), %xmm13	movsd	12 * SIZE(Y), %xmm14	movsd	14 * SIZE(Y), %xmm15	unpcklps	%xmm5, %xmm5	unpcklps	%xmm7, %xmm7	unpcklps	%xmm9, %xmm9	unpcklps	%xmm11, %xmm11	mulps	 %xmm4, %xmm5	mulps	 %xmm6, %xmm7	mulps	 %xmm8, %xmm9	mulps	 %xmm10, %xmm11	movddup	 8 * SIZE(X), %xmm4	movddup	10 * SIZE(X), %xmm6	movddup	12 * SIZE(X), %xmm8	movddup	14 * SIZE(X), %xmm10	decq	%rax	jle	.L12	ALIGN_3.L11:	unpcklps	%xmm12, %xmm12	unpcklps	%xmm13, %xmm13	unpcklps	%xmm14, %xmm14	unpcklps	%xmm15, %xmm15#ifdef BARCELONA	PREFETCH	(PREFETCH_SIZE + 0) * SIZE(X)#endif	addps	 %xmm5,  %xmm0	movsd	16 * SIZE(Y), %xmm5	mulps	 %xmm4, %xmm12	movddup	16 * SIZE(X), %xmm4	addps	 %xmm7,  %xmm1	movsd	18 * SIZE(Y), %xmm7	mulps	 %xmm6, %xmm13	movddup	18 * SIZE(X), %xmm6	addps	 %xmm9,  %xmm2	movsd	20 * SIZE(Y), %xmm9	mulps	 %xmm8, %xmm14	movddup	20 * SIZE(X), %xmm8	addps	 %xmm11, %xmm3	movsd	22 * SIZE(Y), %xmm11	mulps	 %xmm10, %xmm15	movddup	22 * SIZE(X), %xmm10	unpcklps	%xmm5, %xmm5	unpcklps	%xmm7, %xmm7	unpcklps	%xmm9, %xmm9	unpcklps	%xmm11, %xmm11#ifdef BARCELONA	PREFETCH	(PREFETCH_SIZE + 0) * SIZE(Y)#endif	addps	 %xmm12, %xmm0	movsd	24 * SIZE(Y), %xmm12	mulps	 %xmm4, %xmm5	movddup	24 * SIZE(X), %xmm4	addps	 %xmm13, %xmm1	movsd	26 * SIZE(Y), %xmm13	mulps	 %xmm6, %xmm7	movddup	26 * SIZE(X), %xmm6	addps	 %xmm14, %xmm2	movsd	28 * SIZE(Y), %xmm14	mulps	 %xmm8, %xmm9	movddup	28 * SIZE(X), %xmm8	addps	 %xmm15, %xmm3	movsd	30 * SIZE(Y), %xmm15	mulps	 %xmm10, %xmm11	movddup	30 * SIZE(X), %xmm10	unpcklps	%xmm12, %xmm12	unpcklps	%xmm13, %xmm13	unpcklps	%xmm14, %xmm14	unpcklps	%xmm15, %xmm15#ifdef BARCELONA	PREFETCH	(PREFETCH_SIZE + 16) * SIZE(X)#endif	addps	 %xmm5, %xmm0	movsd	32 * SIZE(Y), %xmm5	mulps	 %xmm4, %xmm12	movddup	32 * SIZE(X), %xmm4	addps	 %xmm7, %xmm1	movsd	34 * SIZE(Y), %xmm7	mulps	 %xmm6, %xmm13	movddup	34 * SIZE(X), %xmm6	addps	 %xmm9, %xmm2	movsd	36 * SIZE(Y), %xmm9	mulps	 %xmm8, %xmm14	movddup	36 * SIZE(X), %xmm8	addps	 %xmm11, %xmm3	movsd	38 * SIZE(Y), %xmm11	mulps	 %xmm10, %xmm15	movddup	38 * SIZE(X), %xmm10	unpcklps	%xmm5, %xmm5	unpcklps	%xmm7, %xmm7	unpcklps	%xmm9, %xmm9	unpcklps	%xmm11, %xmm11#ifdef BARCELONA	PREFETCH	(PREFETCH_SIZE + 16) * SIZE(Y)#endif	addps	 %xmm12, %xmm0	movsd	40 * SIZE(Y), %xmm12	mulps	 %xmm4, %xmm5	movddup	40 * SIZE(X), %xmm4	addps	 %xmm13, %xmm1	movsd	42 * SIZE(Y), %xmm13	mulps	 %xmm6, %xmm7	movddup	42 * SIZE(X), %xmm6	addps	 %xmm14, %xmm2	movsd	44 * SIZE(Y), %xmm14	mulps	 %xmm8, %xmm9	movddup	44 * SIZE(X), %xmm8	addps	 %xmm15, %xmm3	movsd	46 * SIZE(Y), %xmm15	mulps	 %xmm10, %xmm11	movddup	46 * SIZE(X), %xmm10	addq	$32 * SIZE, X	addq	$32 * SIZE, Y	decq	%rax	jg	.L11	ALIGN_3.L12:	unpcklps	%xmm12, %xmm12	unpcklps	%xmm13, %xmm13	unpcklps	%xmm14, %xmm14	unpcklps	%xmm15, %xmm15	addps	 %xmm5,  %xmm0	movsd	16 * SIZE(Y), %xmm5	mulps	 %xmm4, %xmm12	movddup	16 * SIZE(X), %xmm4	addps	 %xmm7,  %xmm1	movsd	18 * SIZE(Y), %xmm7	mulps	 %xmm6, %xmm13	movddup	18 * SIZE(X), %xmm6	addps	 %xmm9,  %xmm2	movsd	20 * SIZE(Y), %xmm9	mulps	 %xmm8, %xmm14	movddup	20 * SIZE(X), %xmm8	addps	 %xmm11, %xmm3	movsd	22 * SIZE(Y), %xmm11	mulps	 %xmm10, %xmm15	movddup	22 * SIZE(X), %xmm10	unpcklps	%xmm5, %xmm5	unpcklps	%xmm7, %xmm7	unpcklps	%xmm9, %xmm9	unpcklps	%xmm11, %xmm11	addps	 %xmm12, %xmm0	movsd	24 * SIZE(Y), %xmm12	mulps	 %xmm4, %xmm5	movddup	24 * SIZE(X), %xmm4	addps	 %xmm13, %xmm1	movsd	26 * SIZE(Y), %xmm13	mulps	 %xmm6, %xmm7	movddup	26 * SIZE(X), %xmm6	addps	 %xmm14, %xmm2	movsd	28 * SIZE(Y), %xmm14	mulps	 %xmm8, %xmm9	movddup	28 * SIZE(X), %xmm8	addps	 %xmm15, %xmm3	movsd	30 * SIZE(Y), %xmm15	mulps	 %xmm10, %xmm11	movddup	30 * SIZE(X), %xmm10	unpcklps	%xmm12, %xmm12	unpcklps	%xmm13, %xmm13	unpcklps	%xmm14, %xmm14	unpcklps	%xmm15, %xmm15	addps	 %xmm5, %xmm0	mulps	 %xmm4, %xmm12	addps	 %xmm7, %xmm1	mulps	 %xmm6, %xmm13	addps	 %xmm9, %xmm2	mulps	 %xmm8, %xmm14	addps	 %xmm11, %xmm3	mulps	 %xmm10, %xmm15	addps	 %xmm12, %xmm0	addps	 %xmm13, %xmm1	addps	 %xmm14, %xmm2	addps	 %xmm15, %xmm3	addq	$32 * SIZE, X	addq	$32 * SIZE, Y	ALIGN_3.L15:	testq	$15, N	jle	.L999	testq	$8, N	jle	.L16	movddup	 0 * SIZE(X), %xmm4	movddup	 2 * SIZE(X), %xmm6	movddup	 4 * SIZE(X), %xmm8	movddup	 6 * SIZE(X), %xmm10	movsd	 0 * SIZE(Y), %xmm5	movsd	 2 * SIZE(Y), %xmm7	movsd	 4 * SIZE(Y), %xmm9	movsd	 6 * SIZE(Y), %xmm11	movsd	 8 * SIZE(Y), %xmm12	movsd	10 * SIZE(Y), %xmm13	movsd	12 * SIZE(Y), %xmm14	movsd	14 * SIZE(Y), %xmm15	unpcklps	%xmm5, %xmm5	unpcklps	%xmm7, %xmm7	unpcklps	%xmm9, %xmm9	unpcklps	%xmm11, %xmm11	mulps	 %xmm4, %xmm5	movddup	 8 * SIZE(X), %xmm4	mulps	 %xmm6, %xmm7	movddup	10 * SIZE(X), %xmm6	mulps	 %xmm8, %xmm9	movddup	12 * SIZE(X), %xmm8	mulps	 %xmm10, %xmm11	movddup	14 * SIZE(X), %xmm10	unpcklps	%xmm12, %xmm12	unpcklps	%xmm13, %xmm13	unpcklps	%xmm14, %xmm14	unpcklps	%xmm15, %xmm15	addps	 %xmm5,  %xmm0	mulps	 %xmm4, %xmm12	addps	 %xmm7,  %xmm1	mulps	 %xmm6, %xmm13	addps	 %xmm9,  %xmm2	mulps	 %xmm8, %xmm14	addps	 %xmm11, %xmm3	mulps	 %xmm10, %xmm15	addps	 %xmm12, %xmm0	addps	 %xmm13, %xmm1	addps	 %xmm14, %xmm2	addps	 %xmm15, %xmm3	addq	$16 * SIZE, X	addq	$16 * SIZE, Y	ALIGN_3.L16:	testq	$4, N	jle	.L17	movddup	 0 * SIZE(X), %xmm4	movddup	 2 * SIZE(X), %xmm6	movddup	 4 * SIZE(X), %xmm8	movddup	 6 * SIZE(X), %xmm10	movsd	 0 * SIZE(Y), %xmm5	movsd	 2 * SIZE(Y), %xmm7	movsd	 4 * SIZE(Y), %xmm9	movsd	 6 * SIZE(Y), %xmm11	unpcklps	%xmm5, %xmm5	unpcklps	%xmm7, %xmm7	unpcklps	%xmm9, %xmm9	unpcklps	%xmm11, %xmm11	mulps	 %xmm4, %xmm5	mulps	 %xmm6, %xmm7	mulps	 %xmm8, %xmm9	mulps	 %xmm10, %xmm11	addps	 %xmm5,  %xmm0	addps	 %xmm7,  %xmm1	addps	 %xmm9,  %xmm2	addps	 %xmm11, %xmm3	addq	$8 * SIZE, X	addq	$8 * SIZE, Y	ALIGN_3.L17:	testq	$2, N	jle	.L18	movddup	 0 * SIZE(X), %xmm4	movddup	 2 * SIZE(X), %xmm6	movsd	 0 * SIZE(Y), %xmm5	movsd	 2 * SIZE(Y), %xmm7	unpcklps	%xmm5, %xmm5	unpcklps	%xmm7, %xmm7	mulps	 %xmm4, %xmm5	mulps	 %xmm6, %xmm7	addps	 %xmm5,  %xmm0	addps	 %xmm7,  %xmm1	addq	$4 * SIZE, X	addq	$4 * SIZE, Y	ALIGN_3.L18:	testq	$1, N	jle	.L999	movddup	 0 * SIZE(X), %xmm4	movsd	 0 * SIZE(Y), %xmm5	unpcklps	%xmm5, %xmm5	mulps	 %xmm4, %xmm5	addps	 %xmm5,  %xmm0	jmp	.L999	ALIGN_3.L50:#ifdef F_INTERFACE	testq	INCX, INCX		# if (incx < 0)	jge	.L51	movq	N, %rax		# n	decq	%rax			# n - 1	imulq	INCX, %rax		# (n - 1) * incx	subq	%rax, X	ALIGN_3.L51:	testq	INCY, INCY		# if (incy < 0)	jge	.L60	movq	N, %rax	decq	%rax			# (n - 1)	imulq	INCY, %rax		# (n - 1) * incy	subq	%rax, Y	ALIGN_3.L60:#endif	movq	N,  %rax	sarq	$3, %rax	jle	.L65	ALIGN_3.L62:	movddup	 0 * SIZE(X), %xmm4	movsd	 0 * SIZE(Y), %xmm5	unpcklps	%xmm5, %xmm5	mulps	 %xmm4, %xmm5	addps	 %xmm5, %xmm0	addq	INCX, X	addq	INCY, Y	movddup	 0 * SIZE(X), %xmm4	movsd	 0 * SIZE(Y), %xmm5	unpcklps	%xmm5, %xmm5	mulps	 %xmm4, %xmm5	addps	 %xmm5, %xmm1	addq	INCX, X	addq	INCY, Y	movddup	 0 * SIZE(X), %xmm4	movsd	 0 * SIZE(Y), %xmm5	unpcklps	%xmm5, %xmm5	mulps	 %xmm4, %xmm5	addps	 %xmm5, %xmm2	addq	INCX, X	addq	INCY, Y	movddup	 0 * SIZE(X), %xmm4	movsd	 0 * SIZE(Y), %xmm5	unpcklps	%xmm5, %xmm5	mulps	 %xmm4, %xmm5	addps	 %xmm5, %xmm3	addq	INCX, X	addq	INCY, Y	movddup	 0 * SIZE(X), %xmm4	movsd	 0 * SIZE(Y), %xmm5	unpcklps	%xmm5, %xmm5	mulps	 %xmm4, %xmm5	addps	 %xmm5, %xmm0	addq	INCX, X	addq	INCY, Y	movddup	 0 * SIZE(X), %xmm4	movsd	 0 * SIZE(Y), %xmm5	unpcklps	%xmm5, %xmm5	mulps	 %xmm4, %xmm5	addps	 %xmm5, %xmm1	addq	INCX, X	addq	INCY, Y	movddup	 0 * SIZE(X), %xmm4	movsd	 0 * SIZE(Y), %xmm5	unpcklps	%xmm5, %xmm5	mulps	 %xmm4, %xmm5	addps	 %xmm5, %xmm2	addq	INCX, X	addq	INCY, Y	movddup	 0 * SIZE(X), %xmm4	movsd	 0 * SIZE(Y), %xmm5	unpcklps	%xmm5, %xmm5	mulps	 %xmm4, %xmm5	addps	 %xmm5, %xmm3	addq	INCX, X	addq	INCY, Y	decq	%rax	jg	.L62	ALIGN_3.L65:	movq	N, %rax	andq	$7,   %rax	jle	.L999	ALIGN_3.L66:	movddup	 0 * SIZE(X), %xmm4	/* a b a b */	movsd	 0 * SIZE(Y), %xmm5	/* c d x x */	unpcklps	%xmm5, %xmm5	mulps	 %xmm4, %xmm5	addps	 %xmm5, %xmm0	addq	INCX, X	addq	INCY, Y	decq	%rax	jg	.L66	ALIGN_3.L999:	addps	%xmm1, %xmm0	addps	%xmm3, %xmm2	addps	%xmm2, %xmm0	movhlps	%xmm0, %xmm1	/* a b c d */	pshufd	$ 1, %xmm0, %xmm2	pshufd	$11, %xmm0, %xmm3#ifndef CONJ	subss	 %xmm3, %xmm0	addss	 %xmm2, %xmm1#else	addss	 %xmm3, %xmm0	subss	 %xmm2, %xmm1#endif#if defined(F_INTERFACE) && defined(F_INTERFACE_F2C)	movss	%xmm0, 0 * SIZE(RESULT)	movss	%xmm1, 1 * SIZE(RESULT)#elif defined(PACKED_RETURN) || defined(F_SUN)	unpcklps       %xmm1, %xmm0	#endif	RESTOREREGISTERS	ret	EPILOGUE
⌨️ 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 + -