1// This file is dual licensed under the MIT and the University of Illinois Open
2// Source Licenses. See LICENSE.TXT for details.
3
4#include "../assembly.h"
5
6// di_int __ashrdi3(di_int input, int count);
7
8#ifdef __i386__
9#ifdef __SSE2__
10
11.text
12.balign 4
13DEFINE_COMPILERRT_FUNCTION(__ashrdi3)
14	movd	  12(%esp),		%xmm2	// Load count
15	movl	   8(%esp),		%eax
16#ifndef TRUST_CALLERS_USE_64_BIT_STORES
17	movd	   4(%esp),		%xmm0
18	movd	   8(%esp),		%xmm1
19	punpckldq	%xmm1,		%xmm0	// Load input
20#else
21	movq	   4(%esp),		%xmm0	// Load input
22#endif
23
24	psrlq		%xmm2,		%xmm0	// unsigned shift input by count
25
26	testl		%eax,		%eax	// check the sign-bit of the input
27	jns			1f					// early out for positive inputs
28
29	// If the input is negative, we need to construct the shifted sign bit
30	// to or into the result, as xmm does not have a signed right shift.
31	pcmpeqb		%xmm1,		%xmm1	// -1ULL
32	psrlq		$58,		%xmm1	// 0x3f
33	pandn		%xmm1,		%xmm2	// 63 - count
34	pcmpeqb		%xmm1,		%xmm1	// -1ULL
35	psubq		%xmm1,		%xmm2	// 64 - count
36	psllq		%xmm2,		%xmm1	// -1 << (64 - count) = leading sign bits
37	por			%xmm1,		%xmm0
38
39	// Move the result back to the general purpose registers and return
401:	movd		%xmm0,		%eax
41	psrlq		$32,		%xmm0
42	movd		%xmm0,		%edx
43	ret
44END_COMPILERRT_FUNCTION(__ashrdi3)
45
46#else // Use GPRs instead of SSE2 instructions, if they aren't available.
47
48.text
49.balign 4
50DEFINE_COMPILERRT_FUNCTION(__ashrdi3)
51	movl	  12(%esp),		%ecx	// Load count
52	movl	   8(%esp),		%edx	// Load high
53	movl	   4(%esp),		%eax	// Load low
54
55	testl		$0x20,		%ecx	// If count >= 32
56	jnz			1f					//    goto 1
57
58	shrdl		%cl, %edx,	%eax	// right shift low by count
59	sarl		%cl,		%edx	// right shift high by count
60	ret
61
621:	movl		%edx,		%eax	// Move high to low
63	sarl		$31,		%edx	// clear high
64	sarl		%cl,		%eax	// shift low by count - 32
65	ret
66END_COMPILERRT_FUNCTION(__ashrdi3)
67
68#endif // __SSE2__
69#endif // __i386__
70