/* Copyright (c) 2014, Intel Corporation All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: * Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * Redistributions in binary form must reproduce the above copyright notice, * this list of conditions and the following disclaimer in the documentation * and/or other materials provided with the distribution. * Neither the name of Intel Corporation nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "cache.h" #ifndef MEMCPY # define MEMCPY memcpy #endif #ifndef L # define L(label) .L##label #endif #ifndef cfi_startproc # define cfi_startproc .cfi_startproc #endif #ifndef cfi_endproc # define cfi_endproc .cfi_endproc #endif #ifndef cfi_rel_offset # define cfi_rel_offset(reg, off) .cfi_rel_offset reg, off #endif #ifndef cfi_restore # define cfi_restore(reg) .cfi_restore reg #endif #ifndef cfi_adjust_cfa_offset # define cfi_adjust_cfa_offset(off) .cfi_adjust_cfa_offset off #endif #ifndef ENTRY # define ENTRY(name) \ .type name, @function; \ .globl name; \ .p2align 4; \ name: \ cfi_startproc #endif #ifndef END # define END(name) \ cfi_endproc; \ .size name, .-name #endif #define DEST PARMS #define SRC DEST+4 #define LEN SRC+4 #define CFI_PUSH(REG) \ cfi_adjust_cfa_offset (4); \ cfi_rel_offset (REG, 0) #define CFI_POP(REG) \ cfi_adjust_cfa_offset (-4); \ cfi_restore (REG) #define PUSH(REG) pushl REG; CFI_PUSH (REG) #define POP(REG) popl REG; CFI_POP (REG) #define PARMS 8 /* Preserve EBX. */ #define ENTRANCE PUSH (%ebx); #define RETURN_END POP (%ebx); ret #define RETURN RETURN_END; CFI_PUSH (%ebx) .section .text.sse2,"ax",@progbits ENTRY (MEMCPY) ENTRANCE movl LEN(%esp), %ecx movl SRC(%esp), %eax movl DEST(%esp), %edx cmp %eax, %edx je L(return) cmp $16, %ecx jbe L(len_0_16_bytes) cmp $SHARED_CACHE_SIZE_HALF, %ecx jae L(large_page) movdqu (%eax), %xmm0 movdqu -16(%eax, %ecx), %xmm1 cmpl $32, %ecx movdqu %xmm0, (%edx) movdqu %xmm1, -16(%edx, %ecx) jbe L(return) movdqu 16(%eax), %xmm0 movdqu -32(%eax, %ecx), %xmm1 cmpl $64, %ecx movdqu %xmm0, 16(%edx) movdqu %xmm1, -32(%edx, %ecx) jbe L(return) movdqu 32(%eax), %xmm0 movdqu 48(%eax), %xmm1 movdqu -48(%eax, %ecx), %xmm2 movdqu -64(%eax, %ecx), %xmm3 cmpl $128, %ecx movdqu %xmm0, 32(%edx) movdqu %xmm1, 48(%edx) movdqu %xmm2, -48(%edx, %ecx) movdqu %xmm3, -64(%edx, %ecx) jbe L(return) /* Now the main loop: we align the address of the destination. */ leal 64(%edx), %ebx andl $-64, %ebx addl %edx, %ecx andl $-64, %ecx subl %edx, %eax /* We should stop two iterations before the termination (in order not to misprefetch). */ subl $64, %ecx cmpl %ebx, %ecx je L(main_loop_just_one_iteration) subl $64, %ecx cmpl %ebx, %ecx je L(main_loop_last_two_iterations) .p2align 4 L(main_loop_cache): prefetcht0 128(%ebx, %eax) movdqu (%ebx, %eax), %xmm0 movdqu 16(%ebx, %eax), %xmm1 movdqu 32(%ebx, %eax), %xmm2 movdqu 48(%ebx, %eax), %xmm3 movdqa %xmm0, (%ebx) movdqa %xmm1, 16(%ebx) movdqa %xmm2, 32(%ebx) movdqa %xmm3, 48(%ebx) lea 64(%ebx), %ebx cmpl %ebx, %ecx jne L(main_loop_cache) L(main_loop_last_two_iterations): movdqu (%ebx, %eax), %xmm0 movdqu 16(%ebx, %eax), %xmm1 movdqu 32(%ebx, %eax), %xmm2 movdqu 48(%ebx, %eax), %xmm3 movdqu 64(%ebx, %eax), %xmm4 movdqu 80(%ebx, %eax), %xmm5 movdqu 96(%ebx, %eax), %xmm6 movdqu 112(%ebx, %eax), %xmm7 movdqa %xmm0, (%ebx) movdqa %xmm1, 16(%ebx) movdqa %xmm2, 32(%ebx) movdqa %xmm3, 48(%ebx) movdqa %xmm4, 64(%ebx) movdqa %xmm5, 80(%ebx) movdqa %xmm6, 96(%ebx) movdqa %xmm7, 112(%ebx) jmp L(return) L(main_loop_just_one_iteration): movdqu (%ebx, %eax), %xmm0 movdqu 16(%ebx, %eax), %xmm1 movdqu 32(%ebx, %eax), %xmm2 movdqu 48(%ebx, %eax), %xmm3 movdqa %xmm0, (%ebx) movdqa %xmm1, 16(%ebx) movdqa %xmm2, 32(%ebx) movdqa %xmm3, 48(%ebx) jmp L(return) L(large_page): movdqu (%eax), %xmm0 movdqu 16(%eax), %xmm1 movdqu 32(%eax), %xmm2 movdqu 48(%eax), %xmm3 movdqu -64(%eax, %ecx), %xmm4 movdqu -48(%eax, %ecx), %xmm5 movdqu -32(%eax, %ecx), %xmm6 movdqu -16(%eax, %ecx), %xmm7 movdqu %xmm0, (%edx) movdqu %xmm1, 16(%edx) movdqu %xmm2, 32(%edx) movdqu %xmm3, 48(%edx) movdqu %xmm4, -64(%edx, %ecx) movdqu %xmm5, -48(%edx, %ecx) movdqu %xmm6, -32(%edx, %ecx) movdqu %xmm7, -16(%edx, %ecx) movdqu 64(%eax), %xmm0 movdqu 80(%eax), %xmm1 movdqu 96(%eax), %xmm2 movdqu 112(%eax), %xmm3 movdqu -128(%eax, %ecx), %xmm4 movdqu -112(%eax, %ecx), %xmm5 movdqu -96(%eax, %ecx), %xmm6 movdqu -80(%eax, %ecx), %xmm7 movdqu %xmm0, 64(%edx) movdqu %xmm1, 80(%edx) movdqu %xmm2, 96(%edx) movdqu %xmm3, 112(%edx) movdqu %xmm4, -128(%edx, %ecx) movdqu %xmm5, -112(%edx, %ecx) movdqu %xmm6, -96(%edx, %ecx) movdqu %xmm7, -80(%edx, %ecx) /* Now the main loop with non temporal stores. We align the address of the destination. */ leal 128(%edx), %ebx andl $-128, %ebx addl %edx, %ecx andl $-128, %ecx subl %edx, %eax .p2align 4 L(main_loop_large_page): movdqu (%ebx, %eax), %xmm0 movdqu 16(%ebx, %eax), %xmm1 movdqu 32(%ebx, %eax), %xmm2 movdqu 48(%ebx, %eax), %xmm3 movdqu 64(%ebx, %eax), %xmm4 movdqu 80(%ebx, %eax), %xmm5 movdqu 96(%ebx, %eax), %xmm6 movdqu 112(%ebx, %eax), %xmm7 movntdq %xmm0, (%ebx) movntdq %xmm1, 16(%ebx) movntdq %xmm2, 32(%ebx) movntdq %xmm3, 48(%ebx) movntdq %xmm4, 64(%ebx) movntdq %xmm5, 80(%ebx) movntdq %xmm6, 96(%ebx) movntdq %xmm7, 112(%ebx) lea 128(%ebx), %ebx cmpl %ebx, %ecx jne L(main_loop_large_page) sfence jmp L(return) L(len_0_16_bytes): testb $24, %cl jne L(len_9_16_bytes) testb $4, %cl .p2align 4,,5 jne L(len_5_8_bytes) testl %ecx, %ecx .p2align 4,,2 je L(return) movzbl (%eax), %ebx testb $2, %cl movb %bl, (%edx) je L(return) movzwl -2(%eax,%ecx), %ebx movw %bx, -2(%edx,%ecx) jmp L(return) L(len_9_16_bytes): movq (%eax), %xmm0 movq -8(%eax, %ecx), %xmm1 movq %xmm0, (%edx) movq %xmm1, -8(%edx, %ecx) jmp L(return) L(len_5_8_bytes): movl (%eax), %ebx movl %ebx, (%edx) movl -4(%eax,%ecx), %ebx movl %ebx, -4(%edx,%ecx) jmp L(return) L(return): movl %edx, %eax RETURN END (MEMCPY)