📄 dot_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"#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	PROLOGUE	PROFCODE#ifdef WINDOWS_ABI	movq	40(%rsp), INCY#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	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	$SIZE, INCX	jne	.L50	cmpq	$SIZE, INCY	jne	.L50	cmpq	$3, N	jle	.L17	testq	$SIZE, Y	je	.L05	movss	0 * SIZE(X), %xmm0	mulss	0 * SIZE(Y), %xmm0	addq	$1 * SIZE, X	addq	$1 * SIZE, Y	decq	N	ALIGN_2.L05:	testq	$2 * SIZE, Y	je	.L10	movsd	0 * SIZE(X), %xmm4	movsd	0 * SIZE(Y), %xmm1	mulps	%xmm4, %xmm1	addq	$2 * SIZE, X	addq	$2 * SIZE, Y	subq	$2, N	jle	.L999	ALIGN_2.L10:	testq	$3 * SIZE, X	jne	.L20	movq	N,  %rax	sarq	$5, %rax	jle	.L14	movaps	 0 * SIZE(X), %xmm4	movaps	 4 * SIZE(X), %xmm5	movaps	 8 * SIZE(X), %xmm6	movaps	12 * SIZE(X), %xmm7	movaps	 0 * SIZE(Y), %xmm8	movaps	 4 * SIZE(Y), %xmm9	movaps	 8 * SIZE(Y), %xmm10	movaps	12 * SIZE(Y), %xmm11	movaps	16 * SIZE(X), %xmm12	movaps	20 * SIZE(X), %xmm13	movaps	24 * SIZE(X), %xmm14	movaps	28 * SIZE(X), %xmm15	mulps	%xmm8,  %xmm4	movaps	16 * SIZE(Y), %xmm8	mulps	%xmm9,  %xmm5	movaps	20 * SIZE(Y), %xmm9	mulps	%xmm10, %xmm6	movaps	24 * SIZE(Y), %xmm10	mulps	%xmm11, %xmm7	movaps	28 * SIZE(Y), %xmm11	decq	%rax	jle	.L12	ALIGN_3.L11:	addps	%xmm4,  %xmm0	movaps	32 * SIZE(X), %xmm4	mulps	%xmm8,  %xmm12	addps	%xmm5,  %xmm1	movaps	36 * SIZE(X), %xmm5	mulps	%xmm9,  %xmm13	addps	%xmm6,  %xmm2	movaps	40 * SIZE(X), %xmm6	mulps	%xmm10, %xmm14	addps	%xmm7,  %xmm3	movaps	44 * SIZE(X), %xmm7	mulps	%xmm11, %xmm15	movaps	32 * SIZE(Y), %xmm8	movaps	36 * SIZE(Y), %xmm9	movaps	40 * SIZE(Y), %xmm10	movaps	44 * SIZE(Y), %xmm11	addps	%xmm12,  %xmm0	movaps	48 * SIZE(X), %xmm12	mulps	%xmm8,  %xmm4	addps	%xmm13,  %xmm1	movaps	52 * SIZE(X), %xmm13	mulps	%xmm9,  %xmm5	addps	%xmm14,  %xmm2	movaps	56 * SIZE(X), %xmm14	mulps	%xmm10, %xmm6	addps	%xmm15,  %xmm3	movaps	60 * SIZE(X), %xmm15	mulps	%xmm11, %xmm7	movaps	48 * SIZE(Y), %xmm8	movaps	52 * SIZE(Y), %xmm9	movaps	56 * SIZE(Y), %xmm10	movaps	60 * SIZE(Y), %xmm11	addq	$32 * SIZE, X	addq	$32 * SIZE, Y	decq	%rax	jg	.L11	ALIGN_3.L12:	addps	%xmm4,  %xmm0	mulps	%xmm8,  %xmm12	addps	%xmm5,  %xmm1	mulps	%xmm9,  %xmm13	addps	%xmm6,  %xmm2	mulps	%xmm10, %xmm14	addps	%xmm7,  %xmm3	mulps	%xmm11, %xmm15	addps	%xmm12,  %xmm0	addps	%xmm13,  %xmm1	addps	%xmm14,  %xmm2	addps	%xmm15,  %xmm3	addq	$32 * SIZE, X	addq	$32 * SIZE, Y	ALIGN_3.L14:	testq	$31, N	jle	.L999	movq	N,  %rax	andq	$16, %rax	jle	.L15	movaps	 0 * SIZE(X), %xmm4	movaps	 4 * SIZE(X), %xmm5	movaps	 8 * SIZE(X), %xmm6	movaps	12 * SIZE(X), %xmm7	movaps	 0 * SIZE(Y), %xmm8	movaps	 4 * SIZE(Y), %xmm9	movaps	 8 * SIZE(Y), %xmm10	movaps	12 * SIZE(Y), %xmm11	mulps	%xmm8,  %xmm4	mulps	%xmm9,  %xmm5	mulps	%xmm10, %xmm6	mulps	%xmm11, %xmm7	addps	%xmm4,  %xmm0	addps	%xmm5,  %xmm1	addps	%xmm6,  %xmm2	addps	%xmm7,  %xmm3	addq	$16 * SIZE, X	addq	$16 * SIZE, Y	ALIGN_3.L15:	movq	N,  %rax	andq	$8, %rax	jle	.L16	movaps	 0 * SIZE(X), %xmm4	movaps	 4 * SIZE(X), %xmm5	movaps	 0 * SIZE(Y), %xmm8	movaps	 4 * SIZE(Y), %xmm9	mulps	%xmm8,  %xmm4	mulps	%xmm9,  %xmm5	addps	%xmm4,  %xmm0	addps	%xmm5,  %xmm1	addq	$8 * SIZE, X	addq	$8 * SIZE, Y	ALIGN_3.L16:	movq	N,  %rax	andq	$4, %rax	jle	.L17	movaps	 0 * SIZE(X), %xmm4	movaps	 0 * SIZE(Y), %xmm8	mulps	%xmm8,  %xmm4	addps	%xmm4,  %xmm2	addq	$4 * SIZE, X	addq	$4 * SIZE, Y	ALIGN_3.L17:	movq	N,  %rax	andq	$2, %rax	jle	.L18	movsd	 0 * SIZE(X), %xmm4	movsd	 0 * SIZE(Y), %xmm8	mulps	%xmm8,  %xmm4	addps	%xmm4,  %xmm3	addq	$2 * SIZE, X	addq	$2 * SIZE, Y	ALIGN_3.L18:	movq	N,  %rax	andq	$1, %rax	jle	.L999	movss	 0 * SIZE(X), %xmm4	movss	 0 * SIZE(Y), %xmm8	mulss	%xmm8,  %xmm4	addss	%xmm4,  %xmm0	jmp	.L999	ALIGN_3.L20:	movq	N,  %rax	sarq	$5, %rax	jle	.L24#ifdef HAVE_SSE3	lddqu	 0 * SIZE(X), %xmm4	lddqu	 4 * SIZE(X), %xmm5	lddqu	 8 * SIZE(X), %xmm6	lddqu	12 * SIZE(X), %xmm7#else	movsd	 0 * SIZE(X), %xmm4	movhps	 2 * SIZE(X), %xmm4	movsd	 4 * SIZE(X), %xmm5	movhps	 6 * SIZE(X), %xmm5	movsd	 8 * SIZE(X), %xmm6	movhps	10 * SIZE(X), %xmm6	movsd	12 * SIZE(X), %xmm7	movhps	14 * SIZE(X), %xmm7#endif	movaps	 0 * SIZE(Y), %xmm8	movaps	 4 * SIZE(Y), %xmm9	movaps	 8 * SIZE(Y), %xmm10	movaps	12 * SIZE(Y), %xmm11#ifdef HAVE_SSE3	lddqu	16 * SIZE(X), %xmm12	lddqu	20 * SIZE(X), %xmm13	lddqu	24 * SIZE(X), %xmm14	lddqu	28 * SIZE(X), %xmm15#else	movsd	16 * SIZE(X), %xmm12	movhps	18 * SIZE(X), %xmm12	movsd	20 * SIZE(X), %xmm13	movhps	22 * SIZE(X), %xmm13	movsd	24 * SIZE(X), %xmm14	movhps	26 * SIZE(X), %xmm14	movsd	28 * SIZE(X), %xmm15	movhps	30 * SIZE(X), %xmm15#endif	mulps	%xmm8,  %xmm4	movaps	16 * SIZE(Y), %xmm8	mulps	%xmm9,  %xmm5	movaps	20 * SIZE(Y), %xmm9	mulps	%xmm10, %xmm6	movaps	24 * SIZE(Y), %xmm10	mulps	%xmm11, %xmm7	movaps	28 * SIZE(Y), %xmm11	decq	%rax	jle	.L22	ALIGN_3.L21:	addps	%xmm4,  %xmm0#ifdef HAVE_SSE3	lddqu	32 * SIZE(X), %xmm4#else	movsd	32 * SIZE(X), %xmm4	movhps	34 * SIZE(X), %xmm4#endif	mulps	%xmm8,  %xmm12	addps	%xmm5,  %xmm1#ifdef HAVE_SSE3	lddqu	36 * SIZE(X), %xmm5#else	movsd	36 * SIZE(X), %xmm5	movhps	38 * SIZE(X), %xmm5#endif	mulps	%xmm9,  %xmm13	addps	%xmm6,  %xmm2#ifdef HAVE_SSE3	lddqu	40 * SIZE(X), %xmm6#else	movsd	40 * SIZE(X), %xmm6	movhps	42 * SIZE(X), %xmm6#endif	mulps	%xmm10, %xmm14	addps	%xmm7,  %xmm3#ifdef HAVE_SSE3	lddqu	44 * SIZE(X), %xmm7#else	movsd	44 * SIZE(X), %xmm7	movhps	46 * SIZE(X), %xmm7#endif	mulps	%xmm11, %xmm15	movaps	32 * SIZE(Y), %xmm8	movaps	36 * SIZE(Y), %xmm9	movaps	40 * SIZE(Y), %xmm10	movaps	44 * SIZE(Y), %xmm11	addps	%xmm12,  %xmm0#ifdef HAVE_SSE3	lddqu	48 * SIZE(X), %xmm12#else	movsd	48 * SIZE(X), %xmm12	movhps	50 * SIZE(X), %xmm12#endif	mulps	%xmm8,  %xmm4	addps	%xmm13,  %xmm1#ifdef HAVE_SSE3	lddqu	52 * SIZE(X), %xmm13#else	movsd	52 * SIZE(X), %xmm13	movhps	54 * SIZE(X), %xmm13#endif	mulps	%xmm9,  %xmm5	addps	%xmm14,  %xmm2#ifdef HAVE_SSE3	lddqu	56 * SIZE(X), %xmm14#else	movsd	56 * SIZE(X), %xmm14	movhps	58 * SIZE(X), %xmm14#endif	mulps	%xmm10, %xmm6	addps	%xmm15,  %xmm3#ifdef HAVE_SSE3	lddqu	60 * SIZE(X), %xmm15#else	movsd	60 * SIZE(X), %xmm15	movhps	62 * SIZE(X), %xmm15#endif	mulps	%xmm11, %xmm7	movaps	48 * SIZE(Y), %xmm8	movaps	52 * SIZE(Y), %xmm9	movaps	56 * SIZE(Y), %xmm10	movaps	60 * SIZE(Y), %xmm11	addq	$32 * SIZE, X	addq	$32 * SIZE, Y	decq	%rax	jg	.L21	ALIGN_3.L22:	addps	%xmm4,  %xmm0	mulps	%xmm8,  %xmm12	addps	%xmm5,  %xmm1	mulps	%xmm9,  %xmm13	addps	%xmm6,  %xmm2	mulps	%xmm10, %xmm14	addps	%xmm7,  %xmm3	mulps	%xmm11, %xmm15	addps	%xmm12,  %xmm0	addps	%xmm13,  %xmm1	addps	%xmm14,  %xmm2	addps	%xmm15,  %xmm3	addq	$32 * SIZE, X	addq	$32 * SIZE, Y	ALIGN_3.L24:	testq	$31, N	jle	.L999	movq	N,  %rax	andq	$16, %rax	jle	.L25#ifdef HAVE_SSE3	lddqu	 0 * SIZE(X), %xmm4	lddqu	 4 * SIZE(X), %xmm5	lddqu	 8 * SIZE(X), %xmm6	lddqu	12 * SIZE(X), %xmm7#else	movsd	 0 * SIZE(X), %xmm4	movhps	 2 * SIZE(X), %xmm4	movsd	 4 * SIZE(X), %xmm5	movhps	 6 * SIZE(X), %xmm5	movsd	 8 * SIZE(X), %xmm6	movhps	10 * SIZE(X), %xmm6	movsd	12 * SIZE(X), %xmm7	movhps	14 * SIZE(X), %xmm7#endif	movaps	 0 * SIZE(Y), %xmm8	movaps	 4 * SIZE(Y), %xmm9	movaps	 8 * SIZE(Y), %xmm10	movaps	12 * SIZE(Y), %xmm11	mulps	%xmm8,  %xmm4	mulps	%xmm9,  %xmm5	mulps	%xmm10, %xmm6	mulps	%xmm11, %xmm7	addps	%xmm4,  %xmm0	addps	%xmm5,  %xmm1	addps	%xmm6,  %xmm2	addps	%xmm7,  %xmm3	addq	$16 * SIZE, X	addq	$16 * SIZE, Y	ALIGN_3.L25:	movq	N,  %rax	andq	$8, %rax	jle	.L26#ifdef HAVE_SSE3	lddqu	 0 * SIZE(X), %xmm4	lddqu	 4 * SIZE(X), %xmm5#else	movsd	 0 * SIZE(X), %xmm4	movhps	 2 * SIZE(X), %xmm4	movsd	 4 * SIZE(X), %xmm5	movhps	 6 * SIZE(X), %xmm5#endif	movaps	 0 * SIZE(Y), %xmm8	movaps	 4 * SIZE(Y), %xmm9	mulps	%xmm8,  %xmm4	mulps	%xmm9,  %xmm5	addps	%xmm4,  %xmm0	addps	%xmm5,  %xmm1	addq	$8 * SIZE, X	addq	$8 * SIZE, Y	ALIGN_3.L26:	movq	N,  %rax	andq	$4, %rax	jle	.L27#ifdef HAVE_SSE3	lddqu	 0 * SIZE(X), %xmm4#else	movsd	 0 * SIZE(X), %xmm4	movhps	 2 * SIZE(X), %xmm4#endif	movaps	 0 * SIZE(Y), %xmm8	mulps	%xmm8,  %xmm4	addps	%xmm4,  %xmm2	addq	$4 * SIZE, X	addq	$4 * SIZE, Y	ALIGN_3.L27:	movq	N,  %rax	andq	$2, %rax	jle	.L28	movsd	 0 * SIZE(X), %xmm4	movsd	 0 * SIZE(Y), %xmm8	mulps	%xmm8,  %xmm4	addps	%xmm4,  %xmm3	addq	$2 * SIZE, X	addq	$2 * SIZE, Y	ALIGN_3.L28:	movq	N,  %rax	andq	$1, %rax	jle	.L999	movss	 0 * SIZE(X), %xmm4	movss	 0 * SIZE(Y), %xmm8	mulss	%xmm8,  %xmm4	addss	%xmm4,  %xmm0	jmp	.L999	ALIGN_3.L50:#ifdef F_INTERFACE	testq	INCX, INCX	jge	.L51	movq	N, %rax	decq	%rax	imulq	INCX, %rax	subq	%rax, X	ALIGN_3.L51:	testq	INCY, INCY	jge	.L52	movq	N, %rax	decq	%rax	imulq	INCY, %rax	subq	%rax, Y	ALIGN_3.L52:#endif	movq	N,  %rax	sarq	$2, %rax	jle	.L55	ALIGN_3.L53:	movss	0 * SIZE(X), %xmm4	addq	INCX, X	mulss	0 * SIZE(Y), %xmm4	addq	INCY, Y	movss	0 * SIZE(X), %xmm5	addq	INCX, X	mulss	0 * SIZE(Y), %xmm5	addq	INCY, Y	movss	0 * SIZE(X), %xmm6	addq	INCX, X	mulss	0 * SIZE(Y), %xmm6	addq	INCY, Y	movss	0 * SIZE(X), %xmm7	addq	INCX, X	mulss	0 * SIZE(Y), %xmm7	addq	INCY, Y	addss	%xmm4, %xmm0	addss	%xmm5, %xmm1	addss	%xmm6, %xmm2	addss	%xmm7, %xmm3	decq	%rax	jg	.L53	ALIGN_3.L55:	movq	N, %rax	andq	$3,   %rax	jle	.L999	ALIGN_3.L56:	movss	0 * SIZE(X), %xmm4	addq	INCX, X	mulss	0 * SIZE(Y), %xmm4	addq	INCY, Y	addss	%xmm4, %xmm0	decq	%rax	jg	.L56	ALIGN_3.L999:	addps	%xmm1, %xmm0	addps	%xmm3, %xmm2	addps	%xmm2, %xmm0#ifndef HAVE_SSE3	movhlps	%xmm0, %xmm1	addps	%xmm1, %xmm0		movaps	%xmm0, %xmm1	shufps  $1, %xmm0, %xmm0	addss	 %xmm1, %xmm0#else	haddps	%xmm0, %xmm0	haddps	%xmm0, %xmm0#endif#if !defined(DOUBLE) && defined(F_INTERFACE) && defined(NEED_F2CCONV)	cvtss2sd	%xmm0, %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 + -