1// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 2// See https://llvm.org/LICENSE.txt for license information. 3// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 4 5#include "../assembly.h" 6 7// di_int __lshrdi3(di_int input, int count); 8 9// This routine has some extra memory traffic, loading the 64-bit input via two 10// 32-bit loads, then immediately storing it back to the stack via a single 64-bit 11// store. This is to avoid a write-small, read-large stall. 12// However, if callers of this routine can be safely assumed to store the argument 13// via a 64-bt store, this is unnecessary memory traffic, and should be avoided. 14// It can be turned off by defining the TRUST_CALLERS_USE_64_BIT_STORES macro. 15 16#ifdef __i386__ 17#ifdef __SSE2__ 18 19.text 20.balign 4 21DEFINE_COMPILERRT_FUNCTION(__lshrdi3) 22 movd 12(%esp), %xmm2 // Load count 23#ifndef TRUST_CALLERS_USE_64_BIT_STORES 24 movd 4(%esp), %xmm0 25 movd 8(%esp), %xmm1 26 punpckldq %xmm1, %xmm0 // Load input 27#else 28 movq 4(%esp), %xmm0 // Load input 29#endif 30 psrlq %xmm2, %xmm0 // shift input by count 31 movd %xmm0, %eax 32 psrlq $32, %xmm0 33 movd %xmm0, %edx 34 ret 35END_COMPILERRT_FUNCTION(__lshrdi3) 36 37#else // Use GPRs instead of SSE2 instructions, if they aren't available. 38 39.text 40.balign 4 41DEFINE_COMPILERRT_FUNCTION(__lshrdi3) 42 movl 12(%esp), %ecx // Load count 43 movl 8(%esp), %edx // Load high 44 movl 4(%esp), %eax // Load low 45 46 testl $0x20, %ecx // If count >= 32 47 jnz 1f // goto 1 48 49 shrdl %cl, %edx, %eax // right shift low by count 50 shrl %cl, %edx // right shift high by count 51 ret 52 531: movl %edx, %eax // Move high to low 54 xorl %edx, %edx // clear high 55 shrl %cl, %eax // shift low by count - 32 56 ret 57END_COMPILERRT_FUNCTION(__lshrdi3) 58 59#endif // __SSE2__ 60#endif // __i386__ 61 62NO_EXEC_STACK_DIRECTIVE 63 64