1276789Sdim// This file is dual licensed under the MIT and the University of Illinois Open 2276789Sdim// Source Licenses. See LICENSE.TXT for details. 3276789Sdim 4276789Sdim#include "../assembly.h" 5276789Sdim 6276789Sdim// di_int __ashldi3(di_int input, int count); 7276789Sdim 8276789Sdim// This routine has some extra memory traffic, loading the 64-bit input via two 9276789Sdim// 32-bit loads, then immediately storing it back to the stack via a single 64-bit 10276789Sdim// store. This is to avoid a write-small, read-large stall. 11276789Sdim// However, if callers of this routine can be safely assumed to store the argument 12276789Sdim// via a 64-bt store, this is unnecessary memory traffic, and should be avoided. 13276789Sdim// It can be turned off by defining the TRUST_CALLERS_USE_64_BIT_STORES macro. 14276789Sdim 15276789Sdim#ifdef __i386__ 16276789Sdim#ifdef __SSE2__ 17276789Sdim 18276789Sdim.text 19276789Sdim.balign 4 20276789SdimDEFINE_COMPILERRT_FUNCTION(__ashldi3) 21276789Sdim movd 12(%esp), %xmm2 // Load count 22276789Sdim#ifndef TRUST_CALLERS_USE_64_BIT_STORES 23276789Sdim movd 4(%esp), %xmm0 24276789Sdim movd 8(%esp), %xmm1 25276789Sdim punpckldq %xmm1, %xmm0 // Load input 26276789Sdim#else 27276789Sdim movq 4(%esp), %xmm0 // Load input 28276789Sdim#endif 29276789Sdim psllq %xmm2, %xmm0 // shift input by count 30276789Sdim movd %xmm0, %eax 31276789Sdim psrlq $32, %xmm0 32276789Sdim movd %xmm0, %edx 33276789Sdim ret 34276789SdimEND_COMPILERRT_FUNCTION(__ashldi3) 35276789Sdim 36276789Sdim#else // Use GPRs instead of SSE2 instructions, if they aren't available. 37276789Sdim 38276789Sdim.text 39276789Sdim.balign 4 40276789SdimDEFINE_COMPILERRT_FUNCTION(__ashldi3) 41276789Sdim movl 12(%esp), %ecx // Load count 42276789Sdim movl 8(%esp), %edx // Load high 43276789Sdim movl 4(%esp), %eax // Load low 44276789Sdim 45276789Sdim testl $0x20, %ecx // If count >= 32 46276789Sdim jnz 1f // goto 1 47276789Sdim shldl %cl, %eax, %edx // left shift high by count 48276789Sdim shll %cl, %eax // left shift low by count 49276789Sdim ret 50276789Sdim 51276789Sdim1: movl %eax, %edx // Move low to high 52276789Sdim xorl %eax, %eax // clear low 53276789Sdim shll %cl, %edx // shift high by count - 32 54276789Sdim ret 55276789SdimEND_COMPILERRT_FUNCTION(__ashldi3) 56276789Sdim 57276789Sdim#endif // __SSE2__ 58276789Sdim#endif // __i386__ 59