xref: /freebsd/contrib/llvm-project/compiler-rt/lib/builtins/i386/ashrdi3.S (revision 1d386b48a555f61cb7325543adbbb5c3f3407a66)
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 __ashrdi3(di_int input, int count);
8
9#ifdef __i386__
10#ifdef __SSE2__
11
12.text
13.balign 4
14DEFINE_COMPILERRT_FUNCTION(__ashrdi3)
15	movd	  12(%esp),		%xmm2	// Load count
16	movl	   8(%esp),		%eax
17#ifndef TRUST_CALLERS_USE_64_BIT_STORES
18	movd	   4(%esp),		%xmm0
19	movd	   8(%esp),		%xmm1
20	punpckldq	%xmm1,		%xmm0	// Load input
21#else
22	movq	   4(%esp),		%xmm0	// Load input
23#endif
24
25	psrlq		%xmm2,		%xmm0	// unsigned shift input by count
26
27	testl		%eax,		%eax	// check the sign-bit of the input
28	jns			1f					// early out for positive inputs
29
30	// If the input is negative, we need to construct the shifted sign bit
31	// to or into the result, as xmm does not have a signed right shift.
32	pcmpeqb		%xmm1,		%xmm1	// -1ULL
33	psrlq		$58,		%xmm1	// 0x3f
34	pandn		%xmm1,		%xmm2	// 63 - count
35	pcmpeqb		%xmm1,		%xmm1	// -1ULL
36	psubq		%xmm1,		%xmm2	// 64 - count
37	psllq		%xmm2,		%xmm1	// -1 << (64 - count) = leading sign bits
38	por			%xmm1,		%xmm0
39
40	// Move the result back to the general purpose registers and return
411:	movd		%xmm0,		%eax
42	psrlq		$32,		%xmm0
43	movd		%xmm0,		%edx
44	ret
45END_COMPILERRT_FUNCTION(__ashrdi3)
46
47#else // Use GPRs instead of SSE2 instructions, if they aren't available.
48
49.text
50.balign 4
51DEFINE_COMPILERRT_FUNCTION(__ashrdi3)
52	movl	  12(%esp),		%ecx	// Load count
53	movl	   8(%esp),		%edx	// Load high
54	movl	   4(%esp),		%eax	// Load low
55
56	testl		$0x20,		%ecx	// If count >= 32
57	jnz			1f					//    goto 1
58
59	shrdl		%cl, %edx,	%eax	// right shift low by count
60	sarl		%cl,		%edx	// right shift high by count
61	ret
62
631:	movl		%edx,		%eax	// Move high to low
64	sarl		$31,		%edx	// clear high
65	sarl		%cl,		%eax	// shift low by count - 32
66	ret
67END_COMPILERRT_FUNCTION(__ashrdi3)
68
69#endif // __SSE2__
70#endif // __i386__
71
72NO_EXEC_STACK_DIRECTIVE
73
74