lshrdi3.S revision 214152
138032Speter// This file is distributed under the University of Illinois Open Source 2182352Sgshapiro// License. See LICENSE.TXT for details. 364565Sgshapiro 438032Speter#include "../assembly.h" 538032Speter 638032Speter// di_int __lshrdi3(di_int input, int count); 738032Speter 838032Speter// This routine has some extra memory traffic, loading the 64-bit input via two 938032Speter// 32-bit loads, then immediately storing it back to the stack via a single 64-bit 1038032Speter// store. This is to avoid a write-small, read-large stall. 1138032Speter// However, if callers of this routine can be safely assumed to store the argument 1238032Speter// via a 64-bt store, this is unnecessary memory traffic, and should be avoided. 1338032Speter// It can be turned off by defining the TRUST_CALLERS_USE_64_BIT_STORES macro. 1490795Sgshapiro 1590795Sgshapiro#ifdef __i386__ 1690795Sgshapiro#ifdef __SSE2__ 17141862Sgshapiro 1864565Sgshapiro.text 1964565Sgshapiro.align 4 2064565SgshapiroDEFINE_COMPILERRT_FUNCTION(__lshrdi3) 2190795Sgshapiro movd 12(%esp), %xmm2 // Load count 2238032Speter#ifndef TRUST_CALLERS_USE_64_BIT_STORES 23249729Sgshapiro movd 4(%esp), %xmm0 2464565Sgshapiro movd 8(%esp), %xmm1 25111367Sgshapiro punpckldq %xmm1, %xmm0 // Load input 2638032Speter#else 2738032Speter movq 4(%esp), %xmm0 // Load input 2838032Speter#endif 2964565Sgshapiro psrlq %xmm2, %xmm0 // shift input by count 3064565Sgshapiro movd %xmm0, %eax 3164565Sgshapiro psrlq $32, %xmm0 3264565Sgshapiro movd %xmm0, %edx 3364565Sgshapiro ret 3464565Sgshapiro 3564565Sgshapiro#else // Use GPRs instead of SSE2 instructions, if they aren't available. 3664565Sgshapiro 3764565Sgshapiro.text 3864565Sgshapiro.align 4 3964565SgshapiroDEFINE_COMPILERRT_FUNCTION(__lshrdi3) 4038032Speter movl 12(%esp), %ecx // Load count 4138032Speter movl 8(%esp), %edx // Load high 4238032Speter movl 4(%esp), %eax // Load low 4338032Speter 4438032Speter testl $0x20, %ecx // If count >= 32 45173343Sgshapiro jnz 1f // goto 1 4638032Speter 4738032Speter shrdl %cl, %edx, %eax // right shift low by count 4890795Sgshapiro shrl %cl, %edx // right shift high by count 4942580Speter ret 5064565Sgshapiro 5138032Speter1: movl %edx, %eax // Move high to low 5238032Speter xorl %edx, %edx // clear high 5390795Sgshapiro shrl %cl, %eax // shift low by count - 32 5438032Speter ret 55141862Sgshapiro 56141862Sgshapiro#endif // __SSE2__ 5764565Sgshapiro#endif // __i386__ 5864565Sgshapiro