📄 iamax_sse.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 M ARG1 /* rdi */#define X ARG2 /* rsi */#define INCX ARG3 /* rdx */#define RET %rax#define I ARG4#define XX %r10#define MM %r11 #ifdef MIN#define maxps minps#define maxss minss#endif PROLOGUE PROFCODE#ifdef F_INTERFACE#ifndef USE64BITINT movslq (M), M movslq (INCX), INCX#else movq (M), M movq (INCX), INCX#endif#endif SAVEREGISTERS pxor %xmm0, %xmm0 /* Return Value(Float) */ xor RET, RET /* Return Value(Int) */ testq M, M jle .L999 leaq (, INCX, SIZE), INCX testq INCX, INCX jle .L999 movq M, MM movq X, XX#ifdef ABS pcmpeqb %xmm15, %xmm15 psrld $1, %xmm15 /* Generate ABS */#endif movss (X), %xmm0 addq INCX, X decq M shufps $0, %xmm0, %xmm0#ifdef ABS andps %xmm15, %xmm0#endif movaps %xmm0, %xmm1 movaps %xmm0, %xmm2 movaps %xmm0, %xmm3 /* Generating "seed value" */ cmpq $SIZE, INCX jne .L80 /* Incx != 1 goto L80 *//* Analigned Check */ testq $3, X /* 00000011 */ jne .L30 /* Purely Unaligned Mode */ cmpq $8, M jle .L30 /* if M <= 8 goto Unaligned mode */ testq $4, X /* bit test 000100 */ je .L05 movss 0 * SIZE(X), %xmm4#ifdef ABS andps %xmm15, %xmm4#endif maxss %xmm4, %xmm0 decq M addq $SIZE, X ALIGN_3.L05: testq $8, X je .L06 movsd 0 * SIZE(X), %xmm4 unpcklps %xmm4, %xmm4#ifdef ABS andps %xmm15, %xmm4#endif maxps %xmm4, %xmm1 subq $2, M addq $2 * SIZE, X ALIGN_3.L06: movq M, I sarq $4, I jle .L15 ALIGN_4 #define PREFETCHSIZE 256.L11:#if defined(OPTERON) || defined(BARCELONA) prefetch PREFETCHSIZE * SIZE(X)#endif#ifdef PENTIUM4 prefetcht2 PREFETCHSIZE * SIZE(X)#endif movaps 0 * SIZE(X), %xmm4#ifdef ABS andps %xmm15, %xmm4#endif maxps %xmm4, %xmm0 movaps 4 * SIZE(X), %xmm5#ifdef ABS andps %xmm15, %xmm5#endif maxps %xmm5, %xmm1 movaps 8 * SIZE(X), %xmm6#ifdef ABS andps %xmm15, %xmm6#endif maxps %xmm6, %xmm2 movaps 12 * SIZE(X), %xmm7#ifdef ABS andps %xmm15, %xmm7#endif maxps %xmm7, %xmm3 addq $16 * SIZE, X decq I jg .L11 ALIGN_4.L15: andq $15, M jle .L20 testq $8, M je .L16 movaps 0 * SIZE(X), %xmm4#ifdef ABS andps %xmm15, %xmm4#endif maxps %xmm4, %xmm0 movaps 4 * SIZE(X), %xmm5#ifdef ABS andps %xmm15, %xmm5#endif maxps %xmm5, %xmm1 addq $8 * SIZE, X ALIGN_3.L16: testq $4, M je .L17 movaps 0 * SIZE(X), %xmm6#ifdef ABS andps %xmm15, %xmm6#endif maxps %xmm6, %xmm2 addq $4 * SIZE, X ALIGN_3 .L17: testq $2, M je .L18 movsd 0 * SIZE(X), %xmm7 unpcklps %xmm7, %xmm7#ifdef ABS andps %xmm15, %xmm7#endif maxps %xmm7, %xmm3 addq $2 * SIZE, X .L18: testq $1, M je .L20 movss 0 * SIZE(X), %xmm4#ifdef ABS andps %xmm15, %xmm4#endif maxss %xmm4, %xmm0 ALIGN_3.L20: movq XX, X movq MM, M maxps %xmm1, %xmm0 maxps %xmm3, %xmm2 maxps %xmm2, %xmm0 movaps %xmm0, %xmm1 movhlps %xmm0, %xmm0 maxps %xmm1, %xmm0 movaps %xmm0, %xmm1 shufps $1, %xmm0, %xmm0 maxss %xmm1, %xmm0 shufps $0, %xmm0, %xmm0 testq $4, X je .L21 movss 0 * SIZE(X), %xmm1 decq M addq $SIZE, X#ifdef ABS andps %xmm15, %xmm1#endif incq RET comiss %xmm0, %xmm1 je .L999 ALIGN_3.L21: testq $8, X je .L22 movss 0 * SIZE(X), %xmm1 movss 1 * SIZE(X), %xmm2 subq $2, M addq $2 * SIZE, X#ifdef ABS andps %xmm15, %xmm1 andps %xmm15, %xmm2#endif incq RET comiss %xmm0, %xmm1 je .L999 incq RET comiss %xmm0, %xmm2 je .L999 ALIGN_3.L22: movq M, I sarq $3, I jle .L25 ALIGN_4 .L23:#if defined(OPTERON) || defined(BARCELONA) prefetch PREFETCHSIZE * SIZE(X)#endif#ifdef PENTIUM4 prefetcht2 PREFETCHSIZE * SIZE(X)#endif movaps 0 * SIZE(X), %xmm1#ifdef ABS andps %xmm15, %xmm1#endif cmpeqps %xmm0, %xmm1 movaps 4 * SIZE(X), %xmm3#ifdef ABS andps %xmm15, %xmm3#endif cmpeqps %xmm0, %xmm3 orps %xmm3, %xmm1 movmskps %xmm1, %r11 testq $15, %r11 jne .L24 addq $8 * SIZE, X addq $8, RET decq I jg .L23 jmp .L25 ALIGN_3.L24: movss 0 * SIZE(X), %xmm1 movss 1 * SIZE(X), %xmm2 movss 2 * SIZE(X), %xmm3 movss 3 * SIZE(X), %xmm4 movss 4 * SIZE(X), %xmm5 movss 5 * SIZE(X), %xmm6 movss 6 * SIZE(X), %xmm7 movss 7 * SIZE(X), %xmm8#ifdef ABS andps %xmm15, %xmm1 andps %xmm15, %xmm2 andps %xmm15, %xmm3 andps %xmm15, %xmm4 andps %xmm15, %xmm5 andps %xmm15, %xmm6 andps %xmm15, %xmm7 andps %xmm15, %xmm8#endif addq $8 * SIZE, X incq RET comiss %xmm0, %xmm1 je .L999 incq RET comiss %xmm0, %xmm2 je .L999 incq RET comiss %xmm0, %xmm3 je .L999 incq RET comiss %xmm0, %xmm4 je .L999 incq RET comiss %xmm0, %xmm5 je .L999 incq RET comiss %xmm0, %xmm6 je .L999 incq RET comiss %xmm0, %xmm7 je .L999 incq RET jmp .L999 ALIGN_4.L25: testq $4, M je .L26 movss 0 * SIZE(X), %xmm1 movss 1 * SIZE(X), %xmm2 movss 2 * SIZE(X), %xmm3 movss 3 * SIZE(X), %xmm4#ifdef ABS andps %xmm15, %xmm1 andps %xmm15, %xmm2 andps %xmm15, %xmm3 andps %xmm15, %xmm4#endif addq $4 * SIZE, X incq RET comiss %xmm0, %xmm1 je .L999 incq RET comiss %xmm0, %xmm2 je .L999 incq RET comiss %xmm0, %xmm3 je .L999 incq RET comiss %xmm0, %xmm4 je .L999 ALIGN_3 .L26: testq $2, M je .L27 movss 0 * SIZE(X), %xmm1 movss 1 * SIZE(X), %xmm2#ifdef ABS andps %xmm15, %xmm1 andps %xmm15, %xmm2#endif addq $2 * SIZE, X incq RET comiss %xmm0, %xmm1 je .L999 incq RET comiss %xmm0, %xmm2 je .L999 ALIGN_3 .L27: incq RET jmp .L999 ALIGN_3/* Unaligned Mode */.L30: movq M, I sarq $4, I jle .L35 ALIGN_4 .L31:#if defined(OPTERON) || defined(BARCELONA) prefetch PREFETCHSIZE * SIZE(X)#endif#ifdef PENTIUM4 prefetcht2 PREFETCHSIZE * SIZE(X)#endif movsd 0 * SIZE(X), %xmm4 movhps 2 * SIZE(X), %xmm4#ifdef ABS andps %xmm15, %xmm4#endif maxps %xmm4, %xmm0 movsd 4 * SIZE(X), %xmm5 movhps 6 * SIZE(X), %xmm5#ifdef ABS andps %xmm15, %xmm5#endif maxps %xmm5, %xmm1 movsd 8 * SIZE(X), %xmm6 movhps 10 * SIZE(X), %xmm6#ifdef ABS andps %xmm15, %xmm6#endif maxps %xmm6, %xmm2 movsd 12 * SIZE(X), %xmm7 movhps 14 * SIZE(X), %xmm7#ifdef ABS andps %xmm15, %xmm7#endif maxps %xmm7, %xmm3 addq $16 * SIZE, X decq I jg .L31 ALIGN_4.L35: andq $15, M jle .L40 testq $8, M je .L36 movsd 0 * SIZE(X), %xmm4 movhps 2 * SIZE(X), %xmm4#ifdef ABS andps %xmm15, %xmm4#endif maxps %xmm4, %xmm0 movsd 4 * SIZE(X), %xmm5 movhps 6 * SIZE(X), %xmm5#ifdef ABS andps %xmm15, %xmm5#endif maxps %xmm5, %xmm1 addq $8 * SIZE, X ALIGN_3.L36: testq $4, M je .L37 movsd 0 * SIZE(X), %xmm6 movhps 2 * SIZE(X), %xmm6#ifdef ABS andps %xmm15, %xmm6#endif maxps %xmm6, %xmm2 addq $4 * SIZE, X ALIGN_3 .L37: testq $2, M je .L38 movsd 0 * SIZE(X), %xmm7 unpcklps %xmm7, %xmm7#ifdef ABS andps %xmm15, %xmm7#endif maxps %xmm7, %xmm3 addq $2 * SIZE, X .L38: testq $1, M je .L40 movss 0 * SIZE(X), %xmm4#ifdef ABS andps %xmm15, %xmm4#endif
⌨️ 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 + -