1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libjpeg/simd/jsimdcpu.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,105 @@
1.4 +;
1.5 +; jsimdcpu.asm - SIMD instruction support check
1.6 +;
1.7 +; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
1.8 +;
1.9 +; Based on
1.10 +; x86 SIMD extension for IJG JPEG library
1.11 +; Copyright (C) 1999-2006, MIYASAKA Masaru.
1.12 +; For conditions of distribution and use, see copyright notice in jsimdext.inc
1.13 +;
1.14 +; This file should be assembled with NASM (Netwide Assembler),
1.15 +; can *not* be assembled with Microsoft's MASM or any compatible
1.16 +; assembler (including Borland's Turbo Assembler).
1.17 +; NASM is available from http://nasm.sourceforge.net/ or
1.18 +; http://sourceforge.net/project/showfiles.php?group_id=6208
1.19 +;
1.20 +; [TAB8]
1.21 +
1.22 +%include "jsimdext.inc"
1.23 +
1.24 +; --------------------------------------------------------------------------
1.25 + SECTION SEG_TEXT
1.26 + BITS 32
1.27 +;
1.28 +; Check if the CPU supports SIMD instructions
1.29 +;
1.30 +; GLOBAL(unsigned int)
1.31 +; jpeg_simd_cpu_support (void)
1.32 +;
1.33 +
1.34 + align 16
1.35 + global EXTN(jpeg_simd_cpu_support)
1.36 +
1.37 +EXTN(jpeg_simd_cpu_support):
1.38 + push ebx
1.39 +; push ecx ; need not be preserved
1.40 +; push edx ; need not be preserved
1.41 +; push esi ; unused
1.42 + push edi
1.43 +
1.44 + xor edi,edi ; simd support flag
1.45 +
1.46 + pushfd
1.47 + pop eax
1.48 + mov edx,eax
1.49 + xor eax, 1<<21 ; flip ID bit in EFLAGS
1.50 + push eax
1.51 + popfd
1.52 + pushfd
1.53 + pop eax
1.54 + xor eax,edx
1.55 + jz short .return ; CPUID is not supported
1.56 +
1.57 + ; Check for MMX instruction support
1.58 + xor eax,eax
1.59 + cpuid
1.60 + test eax,eax
1.61 + jz short .return
1.62 +
1.63 + xor eax,eax
1.64 + inc eax
1.65 + cpuid
1.66 + mov eax,edx ; eax = Standard feature flags
1.67 +
1.68 + test eax, 1<<23 ; bit23:MMX
1.69 + jz short .no_mmx
1.70 + or edi, byte JSIMD_MMX
1.71 +.no_mmx:
1.72 + test eax, 1<<25 ; bit25:SSE
1.73 + jz short .no_sse
1.74 + or edi, byte JSIMD_SSE
1.75 +.no_sse:
1.76 + test eax, 1<<26 ; bit26:SSE2
1.77 + jz short .no_sse2
1.78 + or edi, byte JSIMD_SSE2
1.79 +.no_sse2:
1.80 +
1.81 + ; Check for 3DNow! instruction support
1.82 + mov eax, 0x80000000
1.83 + cpuid
1.84 + cmp eax, 0x80000000
1.85 + jbe short .return
1.86 +
1.87 + mov eax, 0x80000001
1.88 + cpuid
1.89 + mov eax,edx ; eax = Extended feature flags
1.90 +
1.91 + test eax, 1<<31 ; bit31:3DNow!(vendor independent)
1.92 + jz short .no_3dnow
1.93 + or edi, byte JSIMD_3DNOW
1.94 +.no_3dnow:
1.95 +
1.96 +.return:
1.97 + mov eax,edi
1.98 +
1.99 + pop edi
1.100 +; pop esi ; unused
1.101 +; pop edx ; need not be preserved
1.102 +; pop ecx ; need not be preserved
1.103 + pop ebx
1.104 + ret
1.105 +
1.106 +; For some reason, the OS X linker does not honor the request to align the
1.107 +; segment unless we do this.
1.108 + align 16
