mirrors
/
flac
mirror of https://github.com/xiph/flac
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

cpu.c: Merge ia32 and x86_64 CPU info functions 

After the removal of the OS SSE detection stuff ia32_cpu_info()
and x86_64_cpu_info() became very similar. Merging them makes
sense.

Patch-from: lvqcl <lvqcl.mail@gmail.com>
This commit is contained in:
Erik de Castro Lopo 2017-02-27 20:35:21 +11:00
parent 172146bd8a
commit 61e7dfa8bb
5 changed files with 178 additions and 214 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

303
src/libFLAC/cpu.c
View File

@ -39,17 +39,14 @@
#include <stdlib.h>
#include <string.h>
#if defined(_MSC_VER)
# include <intrin.h> /* for __cpuid() and _xgetbv() */
#endif
#if defined __GNUC__ && defined HAVE_CPUID_H
# include <cpuid.h> /* for __get_cpuid() and __get_cpuid_max() */
#if defined _MSC_VER
#include <intrin.h> /* for __cpuid() and _xgetbv() */
#elif defined __GNUC__ && defined HAVE_CPUID_H
#include <cpuid.h> /* for __get_cpuid() and __get_cpuid_max() */
#endif
#ifndef NDEBUG
#include <stdio.h>
#define dfprintf fprintf
#else
/* This is bad practice, it should be a static void empty function */
@ -57,34 +54,30 @@
#endif
#if defined FLAC__CPU_IA32
#if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && (defined FLAC__HAS_NASM || FLAC__HAS_X86INTRIN) && !defined FLAC__NO_ASM
/* these are flags in EDX of CPUID AX=00000001 */
static const uint32_t FLAC__CPUINFO_IA32_CPUID_CMOV = 0x00008000;
static const uint32_t FLAC__CPUINFO_IA32_CPUID_MMX = 0x00800000;
static const uint32_t FLAC__CPUINFO_IA32_CPUID_SSE = 0x02000000;
static const uint32_t FLAC__CPUINFO_IA32_CPUID_SSE2 = 0x04000000;
#endif
#if FLAC__HAS_X86INTRIN || FLAC__AVX_SUPPORTED || defined FLAC__HAS_NASM
/* these are flags in ECX of CPUID AX=00000001 */
static const uint32_t FLAC__CPUINFO_IA32_CPUID_SSE3 = 0x00000001;
static const uint32_t FLAC__CPUINFO_IA32_CPUID_SSSE3 = 0x00000200;
static const uint32_t FLAC__CPUINFO_IA32_CPUID_SSE41 = 0x00080000;
static const uint32_t FLAC__CPUINFO_IA32_CPUID_SSE42 = 0x00100000;
static const uint32_t FLAC__CPUINFO_X86_CPUID_CMOV = 0x00008000;
static const uint32_t FLAC__CPUINFO_X86_CPUID_MMX = 0x00800000;
static const uint32_t FLAC__CPUINFO_X86_CPUID_SSE = 0x02000000;
static const uint32_t FLAC__CPUINFO_X86_CPUID_SSE2 = 0x04000000;
/* these are flags in ECX of CPUID AX=00000001 */
static const uint32_t FLAC__CPUINFO_IA32_CPUID_OSXSAVE = 0x08000000;
static const uint32_t FLAC__CPUINFO_IA32_CPUID_AVX = 0x10000000;
static const uint32_t FLAC__CPUINFO_IA32_CPUID_FMA = 0x00001000;
static const uint32_t FLAC__CPUINFO_X86_CPUID_SSE3 = 0x00000001;
static const uint32_t FLAC__CPUINFO_X86_CPUID_SSSE3 = 0x00000200;
static const uint32_t FLAC__CPUINFO_X86_CPUID_SSE41 = 0x00080000;
static const uint32_t FLAC__CPUINFO_X86_CPUID_SSE42 = 0x00100000;
static const uint32_t FLAC__CPUINFO_X86_CPUID_OSXSAVE = 0x08000000;
static const uint32_t FLAC__CPUINFO_X86_CPUID_AVX = 0x10000000;
static const uint32_t FLAC__CPUINFO_X86_CPUID_FMA = 0x00001000;
/* these are flags in EBX of CPUID AX=00000007 */
static const uint32_t FLAC__CPUINFO_IA32_CPUID_AVX2 = 0x00000020;
#endif
static const uint32_t FLAC__CPUINFO_X86_CPUID_AVX2 = 0x00000020;
#if defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64
static uint32_t
cpu_xgetbv_x86(void)
{
#if (defined _MSC_VER || defined __INTEL_COMPILER) && FLAC__HAS_X86INTRIN && FLAC__AVX_SUPPORTED
#if (defined _MSC_VER || defined __INTEL_COMPILER) && FLAC__AVX_SUPPORTED
return (uint32_t)_xgetbv(0);
#elif defined __GNUC__
uint32_t lo, hi;
@ -94,144 +87,146 @@ cpu_xgetbv_x86(void)
return 0;
#endif
}
#endif
static void
ia32_cpu_info (FLAC__CPUInfo *info)
static uint32_t
cpu_have_cpuid(void)
{
#if !defined FLAC__CPU_IA32
(void) info;
#if defined FLAC__CPU_X86_64 || defined __i686__ || defined __SSE__ || (defined _M_IX86_FP && _M_IX86_FP > 0)
/* target CPU does have CPUID instruction */
return 1;
#elif defined FLAC__HAS_NASM
return FLAC__cpu_have_cpuid_asm_ia32();
#elif defined __GNUC__ && defined HAVE_CPUID_H
if (__get_cpuid_max(0, 0) != 0)
return 1;
else
return 0;
#elif defined _MSC_VER
FLAC__uint32 flags1, flags2;
__asm {
pushfd
pushfd
pop eax
mov flags1, eax
xor eax, 0x200000
push eax
popfd
pushfd
pop eax
mov flags2, eax
popfd
}
if (((flags1^flags2) & 0x200000) != 0)
return 1;
else
return 0;
#else
FLAC__bool ia32_osxsave = false;
FLAC__uint32 flags_eax, flags_ebx, flags_ecx, flags_edx;
#if !defined FLAC__NO_ASM && (defined FLAC__HAS_NASM || FLAC__HAS_X86INTRIN)
info->use_asm = true; /* we assume a minimum of 80386 with FLAC__CPU_IA32 */
#if defined FLAC__HAS_NASM
if(!FLAC__cpu_have_cpuid_asm_ia32())
return;
#endif
/* http://www.sandpile.org/x86/cpuid.htm */
if (FLAC__HAS_X86INTRIN) {
FLAC__cpu_info_x86(0, &flags_eax, &flags_ebx, &flags_ecx, &flags_edx);
info->ia32.intel = (flags_ebx == 0x756E6547 && flags_edx == 0x49656E69 && flags_ecx == 0x6C65746E) ? true : false; /* GenuineIntel */
FLAC__cpu_info_x86(1, &flags_eax, &flags_ebx, &flags_ecx, &flags_edx);
}
else {
#if FLAC__HAS_NASM
FLAC__cpu_info_asm_ia32(&flags_edx, &flags_ecx);
#endif
}
info->ia32.cmov = (flags_edx & FLAC__CPUINFO_IA32_CPUID_CMOV ) ? true : false;
info->ia32.mmx = (flags_edx & FLAC__CPUINFO_IA32_CPUID_MMX ) ? true : false;
info->ia32.sse = (flags_edx & FLAC__CPUINFO_IA32_CPUID_SSE ) ? true : false;
info->ia32.sse2 = (flags_edx & FLAC__CPUINFO_IA32_CPUID_SSE2 ) ? true : false;
info->ia32.sse3 = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_SSE3 ) ? true : false;
info->ia32.ssse3 = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_SSSE3) ? true : false;
info->ia32.sse41 = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_SSE41) ? true : false;
info->ia32.sse42 = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_SSE42) ? true : false;
if (FLAC__HAS_X86INTRIN && FLAC__AVX_SUPPORTED) {
ia32_osxsave = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_OSXSAVE) ? true : false;
info->ia32.avx = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_AVX ) ? true : false;
info->ia32.fma = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_FMA ) ? true : false;
FLAC__cpu_info_x86(7, &flags_eax, &flags_ebx, &flags_ecx, &flags_edx);
info->ia32.avx2 = (flags_ebx & FLAC__CPUINFO_IA32_CPUID_AVX2 ) ? true : false;
}
dfprintf(stderr, "CPU info (IA-32):\n");
dfprintf(stderr, " CMOV ....... %c\n", info->ia32.cmov ? 'Y' : 'n');
dfprintf(stderr, " MMX ........ %c\n", info->ia32.mmx ? 'Y' : 'n');
dfprintf(stderr, " SSE ........ %c\n", info->ia32.sse ? 'Y' : 'n');
dfprintf(stderr, " SSE2 ....... %c\n", info->ia32.sse2 ? 'Y' : 'n');
dfprintf(stderr, " SSE3 ....... %c\n", info->ia32.sse3 ? 'Y' : 'n');
dfprintf(stderr, " SSSE3 ...... %c\n", info->ia32.ssse3 ? 'Y' : 'n');
dfprintf(stderr, " SSE41 ...... %c\n", info->ia32.sse41 ? 'Y' : 'n');
dfprintf(stderr, " SSE42 ...... %c\n", info->ia32.sse42 ? 'Y' : 'n');
if (FLAC__HAS_X86INTRIN && FLAC__AVX_SUPPORTED) {
dfprintf(stderr, " AVX ........ %c\n", info->ia32.avx ? 'Y' : 'n');
dfprintf(stderr, " FMA ........ %c\n", info->ia32.fma ? 'Y' : 'n');
dfprintf(stderr, " AVX2 ....... %c\n", info->ia32.avx2 ? 'Y' : 'n');
}
/*
* now have to check for OS support of AVX instructions
*/
if (!FLAC__HAS_X86INTRIN || !info->ia32.avx || !ia32_osxsave || (cpu_xgetbv_x86() & 0x6) != 0x6) {
/* no OS AVX support */
info->ia32.avx = false;
info->ia32.avx2 = false;
info->ia32.fma = false;
}
if (FLAC__HAS_X86INTRIN && FLAC__AVX_SUPPORTED) {
dfprintf(stderr, " AVX OS sup . %c\n", info->ia32.avx ? 'Y' : 'n');
}
#else
info->use_asm = false;
#endif
return 0;
#endif
}
static void
x86_64_cpu_info (FLAC__CPUInfo *info)
cpuinfo_x86(FLAC__uint32 level, FLAC__uint32 *eax, FLAC__uint32 *ebx, FLAC__uint32 *ecx, FLAC__uint32 *edx)
{
#if !defined FLAC__NO_ASM && FLAC__HAS_X86INTRIN
#if defined _MSC_VER
int cpuinfo[4];
int ext = level & 0x80000000;
__cpuid(cpuinfo, ext);
if ((uint32_t)cpuinfo[0] >= level) {
#if FLAC__AVX_SUPPORTED
__cpuidex(cpuinfo, level, 0); /* for AVX2 detection */
#else
__cpuid(cpuinfo, level); /* some old compilers don't support __cpuidex */
#endif
*eax = cpuinfo[0]; *ebx = cpuinfo[1]; *ecx = cpuinfo[2]; *edx = cpuinfo[3];
return;
}
#elif defined __GNUC__ && defined HAVE_CPUID_H
FLAC__uint32 ext = level & 0x80000000;
__cpuid(ext, *eax, *ebx, *ecx, *edx);
if (*eax >= level) {
__cpuid_count(level, 0, *eax, *ebx, *ecx, *edx);
return;
}
#elif defined FLAC__HAS_NASM && defined FLAC__CPU_IA32
FLAC__cpu_info_asm_ia32(level, eax, ebx, ecx, edx);
return;
#endif
*eax = *ebx = *ecx = *edx = 0;
}
#endif
static void
x86_cpu_info (FLAC__CPUInfo *info)
{
#if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && (defined FLAC__HAS_NASM || FLAC__HAS_X86INTRIN) && !defined FLAC__NO_ASM
FLAC__bool x86_osxsave = false;
FLAC__bool os_avx = false;
FLAC__uint32 flags_eax, flags_ebx, flags_ecx, flags_edx;
info->use_asm = true;
info->use_asm = true; /* we assume a minimum of 80386 */
if (!cpu_have_cpuid())
return;
/* http://www.sandpile.org/x86/cpuid.htm */
FLAC__cpu_info_x86(0, &flags_eax, &flags_ebx, &flags_ecx, &flags_edx);
cpuinfo_x86(0, &flags_eax, &flags_ebx, &flags_ecx, &flags_edx);
info->x86.intel = (flags_ebx == 0x756E6547 && flags_edx == 0x49656E69 && flags_ecx == 0x6C65746E) ? true : false; /* GenuineIntel */
FLAC__cpu_info_x86(1, &flags_eax, &flags_ebx, &flags_ecx, &flags_edx);
info->x86.sse3 = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_SSE3 ) ? true : false;
info->x86.ssse3 = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_SSSE3) ? true : false;
info->x86.sse41 = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_SSE41) ? true : false;
info->x86.sse42 = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_SSE42) ? true : false;
cpuinfo_x86(1, &flags_eax, &flags_ebx, &flags_ecx, &flags_edx);
info->x86.cmov = (flags_edx & FLAC__CPUINFO_X86_CPUID_CMOV ) ? true : false;
info->x86.mmx = (flags_edx & FLAC__CPUINFO_X86_CPUID_MMX ) ? true : false;
info->x86.sse = (flags_edx & FLAC__CPUINFO_X86_CPUID_SSE ) ? true : false;
info->x86.sse2 = (flags_edx & FLAC__CPUINFO_X86_CPUID_SSE2 ) ? true : false;
info->x86.sse3 = (flags_ecx & FLAC__CPUINFO_X86_CPUID_SSE3 ) ? true : false;
info->x86.ssse3 = (flags_ecx & FLAC__CPUINFO_X86_CPUID_SSSE3) ? true : false;
info->x86.sse41 = (flags_ecx & FLAC__CPUINFO_X86_CPUID_SSE41) ? true : false;
info->x86.sse42 = (flags_ecx & FLAC__CPUINFO_X86_CPUID_SSE42) ? true : false;
if (FLAC__AVX_SUPPORTED) {
x86_osxsave = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_OSXSAVE) ? true : false;
info->x86.avx = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_AVX ) ? true : false;
info->x86.fma = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_FMA ) ? true : false;
FLAC__cpu_info_x86(7, &flags_eax, &flags_ebx, &flags_ecx, &flags_edx);
info->x86.avx2 = (flags_ebx & FLAC__CPUINFO_IA32_CPUID_AVX2 ) ? true : false;
x86_osxsave = (flags_ecx & FLAC__CPUINFO_X86_CPUID_OSXSAVE) ? true : false;
info->x86.avx = (flags_ecx & FLAC__CPUINFO_X86_CPUID_AVX ) ? true : false;
info->x86.fma = (flags_ecx & FLAC__CPUINFO_X86_CPUID_FMA ) ? true : false;
cpuinfo_x86(7, &flags_eax, &flags_ebx, &flags_ecx, &flags_edx);
info->x86.avx2 = (flags_ebx & FLAC__CPUINFO_X86_CPUID_AVX2 ) ? true : false;
}
#if defined FLAC__CPU_IA32
dfprintf(stderr, "CPU info (IA-32):\n");
#else
dfprintf(stderr, "CPU info (x86-64):\n");
dfprintf(stderr, " SSE3 ....... %c\n", info->x86.sse3 ? 'Y' : 'n');
dfprintf(stderr, " SSSE3 ...... %c\n", info->x86.ssse3 ? 'Y' : 'n');
dfprintf(stderr, " SSE41 ...... %c\n", info->x86.sse41 ? 'Y' : 'n');
dfprintf(stderr, " SSE42 ...... %c\n", info->x86.sse42 ? 'Y' : 'n');
#endif
dfprintf(stderr, " CMOV ....... %c\n", info->x86.cmov ? 'Y' : 'n');
dfprintf(stderr, " MMX ........ %c\n", info->x86.mmx ? 'Y' : 'n');
dfprintf(stderr, " SSE ........ %c\n", info->x86.sse ? 'Y' : 'n');
dfprintf(stderr, " SSE2 ....... %c\n", info->x86.sse2 ? 'Y' : 'n');
dfprintf(stderr, " SSE3 ....... %c\n", info->x86.sse3 ? 'Y' : 'n');
dfprintf(stderr, " SSSE3 ...... %c\n", info->x86.ssse3 ? 'Y' : 'n');
dfprintf(stderr, " SSE41 ...... %c\n", info->x86.sse41 ? 'Y' : 'n');
dfprintf(stderr, " SSE42 ...... %c\n", info->x86.sse42 ? 'Y' : 'n');
if (FLAC__AVX_SUPPORTED) {
dfprintf(stderr, " AVX ........ %c\n", info->x86.avx ? 'Y' : 'n');
dfprintf(stderr, " FMA ........ %c\n", info->x86.fma ? 'Y' : 'n');
dfprintf(stderr, " AVX2 ....... %c\n", info->x86.avx2 ? 'Y' : 'n');
dfprintf(stderr, " AVX ........ %c\n", info->x86.avx ? 'Y' : 'n');
dfprintf(stderr, " FMA ........ %c\n", info->x86.fma ? 'Y' : 'n');
dfprintf(stderr, " AVX2 ....... %c\n", info->x86.avx2 ? 'Y' : 'n');
}
/*
* now have to check for OS support of AVX instructions
*/
if (!info->x86.avx || !x86_osxsave || (cpu_xgetbv_x86() & 0x6) != 0x6) {
if (FLAC__AVX_SUPPORTED && info->x86.avx && x86_osxsave && (cpu_xgetbv_x86() & 0x6) == 0x6) {
os_avx = true;
}
if (os_avx) {
dfprintf(stderr, " AVX OS sup . %c\n", info->x86.avx ? 'Y' : 'n');
}
if (!os_avx) {
/* no OS AVX support */
info->x86.avx = false;
info->x86.avx2 = false;
info->x86.fma = false;
}
if (FLAC__AVX_SUPPORTED) {
dfprintf(stderr, " AVX OS sup . %c\n", info->x86.avx ? 'Y' : 'n');
}
#else
/* Silence compiler warnings. */
(void) info;
#if defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64
if (0) cpu_xgetbv_x86 ();
#endif
info->use_asm = false;
#endif
}
@ -245,51 +240,15 @@ void FLAC__cpu_info (FLAC__CPUInfo *info)
info->type = FLAC__CPUINFO_TYPE_X86_64;
#else
info->type = FLAC__CPUINFO_TYPE_UNKNOWN;
info->use_asm = false;
#endif
switch (info->type) {
case FLAC__CPUINFO_TYPE_IA32:
ia32_cpu_info (info);
break;
case FLAC__CPUINFO_TYPE_IA32: /* fallthrough */
case FLAC__CPUINFO_TYPE_X86_64:
x86_64_cpu_info (info);
x86_cpu_info (info);
break;
default:
info->use_asm = false;
break;
}
}
#if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && FLAC__HAS_X86INTRIN
void FLAC__cpu_info_x86(FLAC__uint32 level, FLAC__uint32 *eax, FLAC__uint32 *ebx, FLAC__uint32 *ecx, FLAC__uint32 *edx)
{
#if defined _MSC_VER || defined __INTEL_COMPILER
int cpuinfo[4];
int ext = level & 0x80000000;
__cpuid(cpuinfo, ext);
if((uint32_t)cpuinfo[0] >= level) {
#if FLAC__AVX_SUPPORTED
__cpuidex(cpuinfo, level, 0); /* for AVX2 detection */
#else
__cpuid(cpuinfo, level); /* some old compilers don't support __cpuidex */
#endif
*eax = cpuinfo[0]; *ebx = cpuinfo[1]; *ecx = cpuinfo[2]; *edx = cpuinfo[3];
return;
}
#elif defined __GNUC__ && defined HAVE_CPUID_H
FLAC__uint32 ext = level & 0x80000000;
__cpuid(ext, *eax, *ebx, *ecx, *edx);
if (*eax >= level) {
__cpuid_count(level, 0, *eax, *ebx, *ecx, *edx);
return;
}
#endif
*eax = *ebx = *ecx = *edx = 0;
}
#endif /* (FLAC__CPU_IA32 || FLAC__CPU_X86_64) && FLAC__HAS_X86INTRIN */

42
src/libFLAC/ia32/cpu_asm.nasm
View File

@ -61,37 +61,57 @@ cident FLAC__cpu_have_cpuid_asm_ia32
popfd
ret
; **********************************************************************
;
; void FLAC__cpu_info_asm_ia32(FLAC__uint32 *flags_edx, FLAC__uint32 *flags_ecx)
; void FLAC__cpu_info_asm_ia32(FLAC__uint32 level, FLAC__uint32 *eax, FLAC__uint32 *ebx, FLAC__uint32 *ecx, FLAC__uint32 *edx)
;
cident FLAC__cpu_info_asm_ia32
;[esp + 8] == flags_edx
;[esp + 12] == flags_ecx
;[esp + 8] == level
;[esp + 12] == flags_eax
;[esp + 16] == flags_ebx
;[esp + 20] == flags_ecx
;[esp + 24] == flags_edx
push ebx
call FLAC__cpu_have_cpuid_asm_ia32
test eax, eax
jz .no_cpuid
mov eax, 0
mov eax, [esp + 8]
and eax, 0x80000000
cpuid
cmp eax, 1
cmp eax, [esp + 8]
jb .no_cpuid
xor ecx, ecx
mov eax, 1
mov eax, [esp + 8]
cpuid
mov ebx, [esp + 8]
mov [ebx], edx
mov ebx, [esp + 12]
push ebx
;[esp + 16] == flags_eax
mov ebx, [esp + 16]
mov [ebx], eax
pop eax
;[esp + 16] == flags_ebx
mov ebx, [esp + 16]
mov [ebx], eax
mov ebx, [esp + 20]
mov [ebx], ecx
mov ebx, [esp + 24]
mov [ebx], edx
jmp .end
.no_cpuid:
xor eax, eax
mov ebx, [esp + 8]
mov [ebx], eax
mov ebx, [esp + 12]
mov [ebx], eax
mov ebx, [esp + 16]
mov [ebx], eax
mov ebx, [esp + 20]
mov [ebx], eax
mov ebx, [esp + 24]
mov [ebx], eax
.end:
pop ebx
ret

17
src/libFLAC/include/private/cpu.h
View File

@ -164,18 +164,6 @@ typedef struct {
FLAC__bool sse;
FLAC__bool sse2;
FLAC__bool sse3;
FLAC__bool ssse3;
FLAC__bool sse41;
FLAC__bool sse42;
FLAC__bool avx;
FLAC__bool avx2;
FLAC__bool fma;
} FLAC__CPUInfo_IA32;
typedef struct {
FLAC__bool intel;
FLAC__bool sse3;
FLAC__bool ssse3;
FLAC__bool sse41;
@ -189,7 +177,6 @@ typedef struct {
typedef struct {
FLAC__bool use_asm;
FLAC__CPUInfo_Type type;
FLAC__CPUInfo_IA32 ia32;
FLAC__CPUInfo_x86 x86;
} FLAC__CPUInfo;
@ -197,8 +184,6 @@ void FLAC__cpu_info(FLAC__CPUInfo *info);
FLAC__uint32 FLAC__cpu_have_cpuid_asm_ia32(void);
void FLAC__cpu_info_asm_ia32(FLAC__uint32 *flags_edx, FLAC__uint32 *flags_ecx);
void FLAC__cpu_info_x86(FLAC__uint32 level, FLAC__uint32 *eax, FLAC__uint32 *ebx, FLAC__uint32 *ecx, FLAC__uint32 *edx);
void FLAC__cpu_info_asm_ia32(FLAC__uint32 level, FLAC__uint32 *eax, FLAC__uint32 *ebx, FLAC__uint32 *ecx, FLAC__uint32 *edx);
#endif

4
src/libFLAC/stream_decoder.c
View File

@ -384,7 +384,7 @@ static FLAC__StreamDecoderInitStatus init_stream_internal_(
FLAC__ASSERT(decoder->private_->cpuinfo.type == FLAC__CPUINFO_TYPE_IA32);
#ifdef FLAC__HAS_NASM
decoder->private_->local_lpc_restore_signal_64bit = FLAC__lpc_restore_signal_wide_asm_ia32; /* OPT_IA32: was really necessary for GCC < 4.9 */
if(decoder->private_->cpuinfo.ia32.mmx) {
if (decoder->private_->cpuinfo.x86.mmx) {
decoder->private_->local_lpc_restore_signal = FLAC__lpc_restore_signal_asm_ia32;
decoder->private_->local_lpc_restore_signal_16bit = FLAC__lpc_restore_signal_asm_ia32_mmx;
}
@ -395,7 +395,7 @@ static FLAC__StreamDecoderInitStatus init_stream_internal_(
#endif
#if FLAC__HAS_X86INTRIN && ! defined FLAC__INTEGER_ONLY_LIBRARY
# if defined FLAC__SSE4_1_SUPPORTED
if(decoder->private_->cpuinfo.ia32.sse41) {
if (decoder->private_->cpuinfo.x86.sse41) {
decoder->private_->local_lpc_restore_signal = FLAC__lpc_restore_signal_intrin_sse41;
decoder->private_->local_lpc_restore_signal_16bit = FLAC__lpc_restore_signal_16_intrin_sse41;
decoder->private_->local_lpc_restore_signal_64bit = FLAC__lpc_restore_signal_wide_intrin_sse41;

26
src/libFLAC/stream_encoder.c
View File

@ -889,7 +889,7 @@ static FLAC__StreamEncoderInitStatus init_stream_internal_(
# ifdef FLAC__CPU_IA32
FLAC__ASSERT(encoder->private_->cpuinfo.type == FLAC__CPUINFO_TYPE_IA32);
# ifdef FLAC__HAS_NASM
if(encoder->private_->cpuinfo.ia32.sse) {
if (encoder->private_->cpuinfo.x86.sse) {
if(encoder->protected_->max_lpc_order < 4)
encoder->private_->local_lpc_compute_autocorrelation = FLAC__lpc_compute_autocorrelation_asm_ia32_sse_lag_4_old;
else if(encoder->protected_->max_lpc_order < 8)
@ -905,7 +905,7 @@ static FLAC__StreamEncoderInitStatus init_stream_internal_(
encoder->private_->local_lpc_compute_autocorrelation = FLAC__lpc_compute_autocorrelation_asm_ia32;
encoder->private_->local_lpc_compute_residual_from_qlp_coefficients_64bit = FLAC__lpc_compute_residual_from_qlp_coefficients_wide_asm_ia32; /* OPT_IA32: was really necessary for GCC < 4.9 */
if(encoder->private_->cpuinfo.ia32.mmx) {
if (encoder->private_->cpuinfo.x86.mmx) {
encoder->private_->local_lpc_compute_residual_from_qlp_coefficients = FLAC__lpc_compute_residual_from_qlp_coefficients_asm_ia32;
encoder->private_->local_lpc_compute_residual_from_qlp_coefficients_16bit = FLAC__lpc_compute_residual_from_qlp_coefficients_asm_ia32_mmx;
}
@ -914,13 +914,13 @@ static FLAC__StreamEncoderInitStatus init_stream_internal_(
encoder->private_->local_lpc_compute_residual_from_qlp_coefficients_16bit = FLAC__lpc_compute_residual_from_qlp_coefficients_asm_ia32;
}
if(encoder->private_->cpuinfo.ia32.mmx && encoder->private_->cpuinfo.ia32.cmov)
if (encoder->private_->cpuinfo.x86.mmx && encoder->private_->cpuinfo.x86.cmov)
encoder->private_->local_fixed_compute_best_predictor = FLAC__fixed_compute_best_predictor_asm_ia32_mmx_cmov;
# endif /* FLAC__HAS_NASM */
# if FLAC__HAS_X86INTRIN
# if defined FLAC__SSE_SUPPORTED
if(encoder->private_->cpuinfo.ia32.sse) {
if(encoder->private_->cpuinfo.ia32.sse42 || !encoder->private_->cpuinfo.ia32.intel) { /* use new autocorrelation functions */
if (encoder->private_->cpuinfo.x86.sse) {
if (encoder->private_->cpuinfo.x86.sse42 || !encoder->private_->cpuinfo.x86.intel) { /* use new autocorrelation functions */
if(encoder->protected_->max_lpc_order < 4)
encoder->private_->local_lpc_compute_autocorrelation = FLAC__lpc_compute_autocorrelation_intrin_sse_lag_4_new;
else if(encoder->protected_->max_lpc_order < 8)
@ -948,19 +948,19 @@ static FLAC__StreamEncoderInitStatus init_stream_internal_(
# endif
# ifdef FLAC__SSE2_SUPPORTED
if(encoder->private_->cpuinfo.ia32.sse2) {
if (encoder->private_->cpuinfo.x86.sse2) {
encoder->private_->local_lpc_compute_residual_from_qlp_coefficients = FLAC__lpc_compute_residual_from_qlp_coefficients_intrin_sse2;
encoder->private_->local_lpc_compute_residual_from_qlp_coefficients_16bit = FLAC__lpc_compute_residual_from_qlp_coefficients_16_intrin_sse2;
}
# endif
# ifdef FLAC__SSE4_1_SUPPORTED
if(encoder->private_->cpuinfo.ia32.sse41) {
if (encoder->private_->cpuinfo.x86.sse41) {
encoder->private_->local_lpc_compute_residual_from_qlp_coefficients = FLAC__lpc_compute_residual_from_qlp_coefficients_intrin_sse41;
encoder->private_->local_lpc_compute_residual_from_qlp_coefficients_64bit = FLAC__lpc_compute_residual_from_qlp_coefficients_wide_intrin_sse41;
}
# endif
# ifdef FLAC__AVX2_SUPPORTED
if(encoder->private_->cpuinfo.ia32.avx2) {
if (encoder->private_->cpuinfo.x86.avx2) {
encoder->private_->local_lpc_compute_residual_from_qlp_coefficients_16bit = FLAC__lpc_compute_residual_from_qlp_coefficients_16_intrin_avx2;
encoder->private_->local_lpc_compute_residual_from_qlp_coefficients = FLAC__lpc_compute_residual_from_qlp_coefficients_intrin_avx2;
encoder->private_->local_lpc_compute_residual_from_qlp_coefficients_64bit = FLAC__lpc_compute_residual_from_qlp_coefficients_wide_intrin_avx2;
@ -968,13 +968,13 @@ static FLAC__StreamEncoderInitStatus init_stream_internal_(
# endif
# ifdef FLAC__SSE2_SUPPORTED
if (encoder->private_->cpuinfo.ia32.sse2) {
if (encoder->private_->cpuinfo.x86.sse2) {
encoder->private_->local_fixed_compute_best_predictor = FLAC__fixed_compute_best_predictor_intrin_sse2;
encoder->private_->local_fixed_compute_best_predictor_wide = FLAC__fixed_compute_best_predictor_wide_intrin_sse2;
}
# endif
# ifdef FLAC__SSSE3_SUPPORTED
if (encoder->private_->cpuinfo.ia32.ssse3) {
if (encoder->private_->cpuinfo.x86.ssse3) {
encoder->private_->local_fixed_compute_best_predictor = FLAC__fixed_compute_best_predictor_intrin_ssse3;
encoder->private_->local_fixed_compute_best_predictor_wide = FLAC__fixed_compute_best_predictor_wide_intrin_ssse3;
}
@ -1041,15 +1041,15 @@ static FLAC__StreamEncoderInitStatus init_stream_internal_(
if(encoder->private_->cpuinfo.use_asm) {
# if defined FLAC__CPU_IA32
# ifdef FLAC__SSE2_SUPPORTED
if(encoder->private_->cpuinfo.ia32.sse2)
if (encoder->private_->cpuinfo.x86.sse2)
encoder->private_->local_precompute_partition_info_sums = FLAC__precompute_partition_info_sums_intrin_sse2;
# endif
# ifdef FLAC__SSSE3_SUPPORTED
if(encoder->private_->cpuinfo.ia32.ssse3)
if (encoder->private_->cpuinfo.x86.ssse3)
encoder->private_->local_precompute_partition_info_sums = FLAC__precompute_partition_info_sums_intrin_ssse3;
# endif
# ifdef FLAC__AVX2_SUPPORTED
if(encoder->private_->cpuinfo.ia32.avx2)
if (encoder->private_->cpuinfo.x86.avx2)
encoder->private_->local_precompute_partition_info_sums = FLAC__precompute_partition_info_sums_intrin_avx2;
# endif
# elif defined FLAC__CPU_X86_64