1353358Sdim// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 2353358Sdim// See https://llvm.org/LICENSE.txt for license information. 3353358Sdim// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 4276789Sdim 5276789Sdim#include "../assembly.h" 6276789Sdim 7276789Sdim// di_int __lshrdi3(di_int input, int count); 8276789Sdim 9276789Sdim// This routine has some extra memory traffic, loading the 64-bit input via two 10276789Sdim// 32-bit loads, then immediately storing it back to the stack via a single 64-bit 11276789Sdim// store. This is to avoid a write-small, read-large stall. 12276789Sdim// However, if callers of this routine can be safely assumed to store the argument 13276789Sdim// via a 64-bt store, this is unnecessary memory traffic, and should be avoided. 14276789Sdim// It can be turned off by defining the TRUST_CALLERS_USE_64_BIT_STORES macro. 15276789Sdim 16276789Sdim#ifdef __i386__ 17276789Sdim#ifdef __SSE2__ 18276789Sdim 19276789Sdim.text 20276789Sdim.balign 4 21276789SdimDEFINE_COMPILERRT_FUNCTION(__lshrdi3) 22276789Sdim movd 12(%esp), %xmm2 // Load count 23276789Sdim#ifndef TRUST_CALLERS_USE_64_BIT_STORES 24276789Sdim movd 4(%esp), %xmm0 25276789Sdim movd 8(%esp), %xmm1 26276789Sdim punpckldq %xmm1, %xmm0 // Load input 27276789Sdim#else 28276789Sdim movq 4(%esp), %xmm0 // Load input 29276789Sdim#endif 30276789Sdim psrlq %xmm2, %xmm0 // shift input by count 31276789Sdim movd %xmm0, %eax 32276789Sdim psrlq $32, %xmm0 33276789Sdim movd %xmm0, %edx 34276789Sdim ret 35276789SdimEND_COMPILERRT_FUNCTION(__lshrdi3) 36276789Sdim 37276789Sdim#else // Use GPRs instead of SSE2 instructions, if they aren't available. 38276789Sdim 39276789Sdim.text 40276789Sdim.balign 4 41276789SdimDEFINE_COMPILERRT_FUNCTION(__lshrdi3) 42276789Sdim movl 12(%esp), %ecx // Load count 43276789Sdim movl 8(%esp), %edx // Load high 44276789Sdim movl 4(%esp), %eax // Load low 45353358Sdim 46276789Sdim testl $0x20, %ecx // If count >= 32 47276789Sdim jnz 1f // goto 1 48276789Sdim 49276789Sdim shrdl %cl, %edx, %eax // right shift low by count 50276789Sdim shrl %cl, %edx // right shift high by count 51276789Sdim ret 52353358Sdim 53276789Sdim1: movl %edx, %eax // Move high to low 54276789Sdim xorl %edx, %edx // clear high 55276789Sdim shrl %cl, %eax // shift low by count - 32 56276789Sdim ret 57276789SdimEND_COMPILERRT_FUNCTION(__lshrdi3) 58276789Sdim 59276789Sdim#endif // __SSE2__ 60276789Sdim#endif // __i386__ 61309124Sdim 62309124SdimNO_EXEC_STACK_DIRECTIVE 63309124Sdim 64