📄 zgemm_kernel_1x1.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 STACK 16#define ARGS 8 #define KK 0 + STACK(%esp)#define KKK 4 + STACK(%esp)#define STACK_M 4 + STACK + ARGS(%esp)#define STACK_N 8 + STACK + ARGS(%esp)#define STACK_K 12 + STACK + ARGS(%esp)#ifdef DOUBLE#define ALPHA_R 16 + STACK + ARGS(%esp)#define ALPHA_I 24 + STACK + ARGS(%esp)#define STACK_A 32 + STACK + ARGS(%esp)#define STACK_B 36 + STACK + ARGS(%esp)#define STACK_C 40 + STACK + ARGS(%esp)#define STACK_LDC 44 + STACK + ARGS(%esp)#define OFFSET 48 + STACK + ARGS(%esp)#else#define ALPHA_R 16 + STACK + ARGS(%esp)#define ALPHA_I 20 + STACK + ARGS(%esp)#define STACK_A 24 + STACK + ARGS(%esp)#define STACK_B 28 + STACK + ARGS(%esp)#define STACK_C 32 + STACK + ARGS(%esp)#define STACK_LDC 36 + STACK + ARGS(%esp)#define OFFSET 40 + STACK + ARGS(%esp)#endif PROLOGUE subl $ARGS, %esp pushl %ebp pushl %edi pushl %esi pushl %ebx PROFCODE#define M %esi#define K %edi#define A %ebx#define B %ecx#define C %edx#define LDC %ebp#if defined(TRMMKERNEL) && !defined(LEFT) movl OFFSET, %eax negl %eax movl %eax, KK#endif movl STACK_K, K movl STACK_LDC, LDC sall $ZBASE_SHIFT, LDC cmpl $0, STACK_N jle .L29 cmpl $0, STACK_M jle .L29 ALIGN_4.L30:#if defined(TRMMKERNEL) && defined(LEFT) movl OFFSET, %eax movl %eax, KK#endif movl STACK_A, A movl STACK_C, C movl STACK_M, M ALIGN_4.L34:#if !defined(TRMMKERNEL) || \ (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movl STACK_B, B#else movl STACK_B, B movl KK, %eax leal (, %eax, SIZE), %eax leal (A, %eax, 2), A leal (B, %eax, 2), B#endif fldz fldz fldz fldz FLD 4 * SIZE(B) # B5 FLD 4 * SIZE(A) # A5 FLD 0 * SIZE(B) # B0 FLD 0 * SIZE(A) # A0#ifndef TRMMKERNEL movl K, %eax#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA)) movl K, %eax subl KK, %eax movl %eax, KKK #else movl KK, %eax#ifdef LEFT addl $1, %eax#else addl $1, %eax#endif movl %eax, KKK#endif#ifdef HAVE_SSE prefetcht2 2 * SIZE(C)#endif sarl $2, %eax je .L37 ALIGN_4#define PREFETCH_OFFSET 40.L38:#ifdef HAVE_SSE prefetchnta (PREFETCH_OFFSET) * SIZE(B)#ifdef CORE_KATMAI prefetcht0 (PREFETCH_OFFSET) * SIZE(A)#endif#endif fmul %st, %st(1) FMUL 1 * SIZE(B) fxch %st(1) faddp %st, %st(5) FLD 0 * SIZE(B) fxch %st(1)#if defined(NN) || defined(CN) faddp %st, %st(4)#else fsubrp %st, %st(4)#endif FLD 1 * SIZE(A) fmul %st, %st(1) FMUL 1 * SIZE(B) fxch %st(1)#if defined(NN) || defined(NC) faddp %st, %st(7)#else fsubrp %st, %st(7)#endif FLD 2 * SIZE(B) fxch %st(1)#if defined(NN) || defined(CC) fsubrp %st, %st(6)#else faddp %st, %st(6)#endif FLD 2 * SIZE(A) fmul %st, %st(1) FMUL 3 * SIZE(B) fxch %st(1) faddp %st, %st(5) FLD 2 * SIZE(B) fxch %st(1)#if defined(NN) || defined(CN) faddp %st, %st(4)#else fsubrp %st, %st(4)#endif FLD 3 * SIZE(A) fmul %st, %st(1) FMUL 3 * SIZE(B) fxch %st(1)#if defined(NN) || defined(NC) faddp %st, %st(7)#else fsubrp %st, %st(7)#endif FLD 8 * SIZE(B) fxch %st(1)#if defined(NN) || defined(CC) fsubrp %st, %st(6)#else faddp %st, %st(6)#endif FLD 8 * SIZE(A) fxch %st(2)#ifdef HAVE_SSE#ifdef DOUBLE prefetchnta (PREFETCH_OFFSET + 4) * SIZE(B)#ifdef CORE_KATMAI prefetcht0 (PREFETCH_OFFSET + 4) * SIZE(A)#endif#endif#endif fmul %st, %st(3) FMUL 5 * SIZE(B) fxch %st(3) faddp %st, %st(5) FLD 4 * SIZE(B) fxch %st(3)#if defined(NN) || defined(CN) faddp %st, %st(4)#else fsubrp %st, %st(4)#endif FLD 5 * SIZE(A) fmul %st, %st(3) FMUL 5 * SIZE(B) fxch %st(3)#if defined(NN) || defined(NC) faddp %st, %st(7)#else fsubrp %st, %st(7)#endif FLD 6 * SIZE(B) fxch %st(3)#if defined(NN) || defined(CC) fsubrp %st, %st(6)#else faddp %st, %st(6)#endif FLD 6 * SIZE(A) fmul %st, %st(3) FMUL 7 * SIZE(B) fxch %st(3) faddp %st, %st(5) FLD 6 * SIZE(B) fxch %st(3)#if defined(NN) || defined(CN) faddp %st, %st(4)#else fsubrp %st, %st(4)#endif FLD 7 * SIZE(A) fmul %st, %st(3) FMUL 7 * SIZE(B) fxch %st(3)#if defined(NN) || defined(NC) faddp %st, %st(7)#else fsubrp %st, %st(7)#endif FLD 12 * SIZE(B) fxch %st(3)#if defined(NN) || defined(CC) fsubrp %st, %st(6)#else faddp %st, %st(6)#endif FLD 12 * SIZE(A) fxch %st(2) subl $-8 * SIZE, B subl $-8 * SIZE, A decl %eax jg .L38 ALIGN_4.L37:#ifndef TRMMKERNEL movl K, %eax#else movl KKK, %eax#endif andl $3, %eax jle .L43 ALIGN_2.L54: fmul %st, %st(1) FMUL 1 * SIZE(B) fxch %st(1) faddp %st, %st(5) FLD 0 * SIZE(B) fxch %st(1)#if defined(NN) || defined(CN) faddp %st, %st(4)#else fsubrp %st, %st(4)#endif FLD 1 * SIZE(A) fmul %st, %st(1) FMUL 1 * SIZE(B) fxch %st(1)#if defined(NN) || defined(NC) faddp %st, %st(7)#else fsubrp %st, %st(7)#endif FLD 2 * SIZE(B) fxch %st(1)#if defined(NN) || defined(CC) fsubrp %st, %st(6)#else faddp %st, %st(6)#endif FLD 2 * SIZE(A) addl $2 * SIZE, A addl $2 * SIZE, B decl %eax jg .L54 ALIGN_3.L43: ffreep %st(0) ffreep %st(0) ffreep %st(0) ffreep %st(0) FLD ALPHA_R fxch %st(3) FLD ALPHA_I fxch %st(5) faddp %st, %st(2) # ctemp3 += ctemp4 faddp %st, %st(2) # ctemp1 += ctemp2 fld %st(0) # copy ctemp2 fmul %st(4) # ctemp3 *= alpha_i fld %st(2) # copy ctemp1 fmul %st(4) # ctemp1 *= alpha_r fsubp %st(1) # ctemp2 -= ctemp4#ifndef TRMMKERNEL FADD 0 * SIZE(C)#endif FST 0 * SIZE(C) fmulp %st(2) # ctemp3 *= alpha_i fmulp %st(2) # ctemp1 *= alpha_r faddp %st, %st(1) # ctemp1 += ctemp3#ifndef TRMMKERNEL FADD 1 * SIZE(C)#endif FST 1 * SIZE(C) addl $2 * SIZE, C#if (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \ (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) movl K, %eax subl KKK, %eax leal (,%eax, SIZE), %eax leal (A, %eax, 2), A leal (B, %eax, 2), B#endif#if defined(TRMMKERNEL) && defined(LEFT) addl $1, KK#endif decl M jg .L34 ALIGN_2.L33:#if defined(TRMMKERNEL) && !defined(LEFT) addl $1, KK#endif movl B, STACK_B addl LDC, STACK_C decl STACK_N jg .L30 ALIGN_2.L29: popl %ebx popl %esi popl %edi popl %ebp addl $ARGS, %esp 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 + -