📄 dot_sse2_barcelona.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 */#define INCY	ARG5	/* r8  */#define PREFETCH	prefetch#define PREFETCHSIZE	(8 * 16)	PROLOGUE	PROFCODE#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	testq	$SIZE, Y	je	.L10	movsd	  0 * SIZE(X), %xmm0	mulsd	  0 * SIZE(Y), %xmm0	addq	$1 * SIZE, X	addq	$1 * SIZE, Y	decq	N	jle	.L999	ALIGN_2.L10:	testq	$SIZE, X	jne	.L20	movq	N,    %rax	andq	$-16, %rax	jle	.L14	leaq	(X, %rax, 8), X	leaq	(Y, %rax, 8), Y	negq	%rax	movapd	  0 * SIZE(X, %rax, 8), %xmm4	movapd	  2 * SIZE(Y, %rax, 8), %xmm5	movapd	  4 * SIZE(X, %rax, 8), %xmm6	movapd	  6 * SIZE(Y, %rax, 8), %xmm7	movapd	  8 * SIZE(X, %rax, 8), %xmm8	movapd	 10 * SIZE(Y, %rax, 8), %xmm9	movapd	 12 * SIZE(X, %rax, 8), %xmm10	movapd	 14 * SIZE(Y, %rax, 8), %xmm11	mulpd	  0 * SIZE(Y, %rax, 8), %xmm4	mulpd	  2 * SIZE(X, %rax, 8), %xmm5	mulpd	  4 * SIZE(Y, %rax, 8), %xmm6	mulpd	  6 * SIZE(X, %rax, 8), %xmm7	subq	$-16, %rax	jge	.L12	ALIGN_3.L11:	PREFETCH	(PREFETCHSIZE + 0) * SIZE(X, %rax, 8)	PREFETCH	(PREFETCHSIZE + 8) * SIZE(Y, %rax, 8)	mulpd	 -8 * SIZE(Y, %rax, 8), %xmm8	addpd	%xmm4,  %xmm0	movapd	  0 * SIZE(X, %rax, 8), %xmm4	mulpd	 -6 * SIZE(X, %rax, 8), %xmm9	addpd	%xmm5,  %xmm1	movapd	  2 * SIZE(Y, %rax, 8), %xmm5	mulpd	 -4 * SIZE(Y, %rax, 8), %xmm10	addpd	%xmm6,  %xmm2	movapd	  4 * SIZE(X, %rax, 8), %xmm6	mulpd	 -2 * SIZE(X, %rax, 8), %xmm11	addpd	%xmm7,  %xmm3	movapd	  6 * SIZE(Y, %rax, 8), %xmm7	mulpd	  0 * SIZE(Y, %rax, 8), %xmm4	addpd	%xmm8,  %xmm0	movapd	  8 * SIZE(X, %rax, 8), %xmm8	mulpd	  2 * SIZE(X, %rax, 8), %xmm5	addpd	%xmm9,  %xmm1	movapd	 10 * SIZE(Y, %rax, 8), %xmm9	mulpd	  4 * SIZE(Y, %rax, 8), %xmm6	addpd	%xmm10, %xmm2	movapd	 12 * SIZE(X, %rax, 8), %xmm10	mulpd	  6 * SIZE(X, %rax, 8), %xmm7	addpd	%xmm11, %xmm3	movapd	 14 * SIZE(Y, %rax, 8), %xmm11	subq	$-16, %rax	jl,pt	.L11	ALIGN_3.L12:	mulpd	 -8 * SIZE(Y, %rax, 8), %xmm8	addpd	%xmm4,  %xmm0	mulpd	 -6 * SIZE(X, %rax, 8), %xmm9	addpd	%xmm5,  %xmm1	mulpd	 -4 * SIZE(Y, %rax, 8), %xmm10	addpd	%xmm6,  %xmm2	mulpd	 -2 * SIZE(X, %rax, 8), %xmm11	addpd	%xmm7,  %xmm3	addpd	%xmm8,  %xmm0	addpd	%xmm9,  %xmm1	addpd	%xmm10, %xmm2	addpd	%xmm11, %xmm3	ALIGN_3.L14:	testq	$15, N	jle	.L999	testq	$8, N	jle	.L15	movapd	  0 * SIZE(X), %xmm4	movapd	  2 * SIZE(X), %xmm5	movapd	  4 * SIZE(X), %xmm6	movapd	  6 * SIZE(X), %xmm7	mulpd	  0 * SIZE(Y), %xmm4	mulpd	  2 * SIZE(Y), %xmm5	mulpd	  4 * SIZE(Y), %xmm6	mulpd	  6 * SIZE(Y), %xmm7	addpd	%xmm4,  %xmm0	addpd	%xmm5,  %xmm1	addpd	%xmm6,  %xmm2	addpd	%xmm7,  %xmm3	addq	$8 * SIZE, X	addq	$8 * SIZE, Y	ALIGN_3.L15:	testq	$4, N	jle	.L16	movapd	  0 * SIZE(X), %xmm4	movapd	  2 * SIZE(X), %xmm5	mulpd	  0 * SIZE(Y), %xmm4	mulpd	  2 * SIZE(Y), %xmm5	addpd	%xmm4,  %xmm0	addpd	%xmm5,  %xmm1	addq	$4 * SIZE, X	addq	$4 * SIZE, Y	ALIGN_3.L16:	testq	$2, N	jle	.L17	movapd	  0 * SIZE(X), %xmm4	mulpd	  0 * SIZE(Y), %xmm4	addpd	%xmm4,  %xmm2	addq	$2 * SIZE, X	addq	$2 * SIZE, Y	ALIGN_3.L17:	testq	$1, N	jle	.L999	movsd	  0 * SIZE(X), %xmm4	mulsd	  0 * SIZE(Y), %xmm4	addsd	%xmm4,  %xmm3	jmp	.L999	ALIGN_3.L20:	movq	N,    %rax	andq	$-16, %rax	jle	.L24	leaq	(X, %rax, 8), X	leaq	(Y, %rax, 8), Y	negq	%rax	movlpd	  0 * SIZE(X, %rax, 8), %xmm4	movhpd	  1 * SIZE(X, %rax, 8), %xmm4	movlpd	  2 * SIZE(X, %rax, 8), %xmm5	movhpd	  3 * SIZE(X, %rax, 8), %xmm5	movlpd	  4 * SIZE(X, %rax, 8), %xmm6	movhpd	  5 * SIZE(X, %rax, 8), %xmm6	movlpd	  6 * SIZE(X, %rax, 8), %xmm7	movhpd	  7 * SIZE(X, %rax, 8), %xmm7	movlpd	  8 * SIZE(X, %rax, 8), %xmm8	movhpd	  9 * SIZE(X, %rax, 8), %xmm8	movlpd	 10 * SIZE(X, %rax, 8), %xmm9	movhpd	 11 * SIZE(X, %rax, 8), %xmm9	movlpd	 12 * SIZE(X, %rax, 8), %xmm10	movhpd	 13 * SIZE(X, %rax, 8), %xmm10	movlpd	 14 * SIZE(X, %rax, 8), %xmm11	mulpd	  0 * SIZE(Y, %rax, 8), %xmm4	mulpd	  2 * SIZE(Y, %rax, 8), %xmm5	mulpd	  4 * SIZE(Y, %rax, 8), %xmm6	mulpd	  6 * SIZE(Y, %rax, 8), %xmm7	subq	$-16, %rax	jge	.L22	ALIGN_3.L21:	PREFETCH	(PREFETCHSIZE + 0) * SIZE(X, %rax, 8)	mulpd	 -8 * SIZE(Y, %rax, 8), %xmm8	addpd	%xmm4,  %xmm0	movhpd	 -1 * SIZE(X, %rax, 8), %xmm11	movlpd	  0 * SIZE(X, %rax, 8), %xmm4	mulpd	 -6 * SIZE(Y, %rax, 8), %xmm9	addpd	%xmm5,  %xmm1	movhpd	  1 * SIZE(X, %rax, 8), %xmm4	movlpd	  2 * SIZE(X, %rax, 8), %xmm5	PREFETCH	(PREFETCHSIZE + 0) * SIZE(Y, %rax, 8)	mulpd	 -4 * SIZE(Y, %rax, 8), %xmm10	addpd	%xmm6,  %xmm2	movlpd	  4 * SIZE(X, %rax, 8), %xmm6	movhpd	  3 * SIZE(X, %rax, 8), %xmm5	mulpd	 -2 * SIZE(Y, %rax, 8), %xmm11	addpd	%xmm7,  %xmm3	movlpd	  6 * SIZE(X, %rax, 8), %xmm7	movhpd	  5 * SIZE(X, %rax, 8), %xmm6	mulpd	  0 * SIZE(Y, %rax, 8), %xmm4	addpd	%xmm8,  %xmm0	movhpd	  7 * SIZE(X, %rax, 8), %xmm7	movlpd	  8 * SIZE(X, %rax, 8), %xmm8	PREFETCH	(PREFETCHSIZE + 8) * SIZE(X, %rax, 8)	mulpd	  2 * SIZE(Y, %rax, 8), %xmm5	addpd	%xmm9,  %xmm1	movhpd	  9 * SIZE(X, %rax, 8), %xmm8	movlpd	 10 * SIZE(X, %rax, 8), %xmm9	mulpd	  4 * SIZE(Y, %rax, 8), %xmm6	addpd	%xmm10, %xmm2	movhpd	 11 * SIZE(X, %rax, 8), %xmm9	movlpd	 12 * SIZE(X, %rax, 8), %xmm10	PREFETCH	(PREFETCHSIZE + 8) * SIZE(Y, %rax, 8)	mulpd	  6 * SIZE(Y, %rax, 8), %xmm7	addpd	%xmm11, %xmm3	movhpd	 13 * SIZE(X, %rax, 8), %xmm10	movlpd	 14 * SIZE(X, %rax, 8), %xmm11	subq	$-16, %rax	jl,pt	.L21	ALIGN_3.L22:	movhpd	 -1 * SIZE(X, %rax, 8), %xmm11	mulpd	 -8 * SIZE(Y), %xmm8	mulpd	 -6 * SIZE(Y), %xmm9	mulpd	 -4 * SIZE(Y), %xmm10	mulpd	 -2 * SIZE(Y), %xmm11	addpd	%xmm4,  %xmm0	addpd	%xmm5,  %xmm1	addpd	%xmm6,  %xmm2	addpd	%xmm7,  %xmm3	addpd	%xmm8,  %xmm0	addpd	%xmm9,  %xmm1	addpd	%xmm10, %xmm2	addpd	%xmm11, %xmm3	ALIGN_3.L24:	testq	$15, N	jle	.L999	testq	$8, N	jle	.L25	movlpd	  0 * SIZE(X), %xmm4	movhpd	  1 * SIZE(X), %xmm4	movlpd	  2 * SIZE(X), %xmm5	movhpd	  3 * SIZE(X), %xmm5	movlpd	  4 * SIZE(X), %xmm6	movhpd	  5 * SIZE(X), %xmm6	movlpd	  6 * SIZE(X), %xmm7	movhpd	  7 * SIZE(X), %xmm7	mulpd	  0 * SIZE(Y), %xmm4	mulpd	  2 * SIZE(Y), %xmm5	mulpd	  4 * SIZE(Y), %xmm6	mulpd	  6 * SIZE(Y), %xmm7	addpd	%xmm4,  %xmm0	addpd	%xmm5,  %xmm1	addpd	%xmm6,  %xmm2	addpd	%xmm7,  %xmm3	addq	$8 * SIZE, X	addq	$8 * SIZE, Y	ALIGN_3.L25:	testq	$4, N	jle	.L26	movlpd	  0 * SIZE(X), %xmm4	movhpd	  1 * SIZE(X), %xmm4	movlpd	  2 * SIZE(X), %xmm5	movhpd	  3 * SIZE(X), %xmm5	mulpd	  0 * SIZE(Y), %xmm4	mulpd	  2 * SIZE(Y), %xmm5	addpd	%xmm4,  %xmm0	addpd	%xmm5,  %xmm1	addq	$4 * SIZE, X	addq	$4 * SIZE, Y	ALIGN_3.L26:	testq	$2, N	jle	.L27	movlpd	  0 * SIZE(X), %xmm4	movhpd	  1 * SIZE(X), %xmm4	mulpd	  0 * SIZE(Y), %xmm4	addpd	%xmm4,  %xmm2	addq	$2 * SIZE, X	addq	$2 * SIZE, Y	ALIGN_3.L27:	testq	$1, N	jle	.L999	movsd	  0 * SIZE(X), %xmm4	mulsd	  0 * SIZE(Y), %xmm4	addsd	%xmm4,  %xmm3	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:	movsd	0 * SIZE(X), %xmm4	addq	INCX, X	mulsd	0 * SIZE(Y), %xmm4	addq	INCY, Y	movsd	0 * SIZE(X), %xmm5	addq	INCX, X	mulsd	0 * SIZE(Y), %xmm5	addq	INCY, Y	movsd	0 * SIZE(X), %xmm6	addq	INCX, X	mulsd	0 * SIZE(Y), %xmm6	addq	INCY, Y	movsd	0 * SIZE(X), %xmm7	addq	INCX, X	mulsd	0 * SIZE(Y), %xmm7	addq	INCY, Y	addsd	%xmm4, %xmm0	addsd	%xmm5, %xmm1	addsd	%xmm6, %xmm2	addsd	%xmm7, %xmm3	decq	%rax	jg	.L53	ALIGN_3.L55:	movq	N, %rax	andq	$3,   %rax	jle	.L999	ALIGN_3.L56:	movsd	0 * SIZE(X), %xmm4	addq	INCX, X	mulsd	0 * SIZE(Y), %xmm4	addq	INCY, Y	addsd	%xmm4, %xmm0	decq	%rax	jg	.L56	ALIGN_3.L999:	addpd	%xmm1, %xmm0	addpd	%xmm3, %xmm2	addpd	%xmm2, %xmm0#ifndef HAVE_SSE3	movapd	%xmm0, %xmm1	unpckhpd	%xmm0, %xmm0	addsd	%xmm1, %xmm0#else	haddpd	%xmm0, %xmm0#endif	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 + -