1// Copyright 2011 Google Inc. All Rights Reserved.
2//
3// Use of this source code is governed by a BSD-style license
4// that can be found in the COPYING file in the root of the source
5// tree. An additional intellectual property rights grant can be found
6// in the file PATENTS. All contributing project authors may
7// be found in the AUTHORS file in the root of the source tree.
8// -----------------------------------------------------------------------------
9//
10// CPU detection
11//
12// Author: Christian Duvivier (cduvivier@google.com)
13
14#include "./dsp.h"
15
16#if defined(__ANDROID__)
17#include "cpu-features.h"
18#endif
19
20//------------------------------------------------------------------------------
21// SSE2 detection.
22//
23
24// apple/darwin gcc-4.0.1 defines __PIC__, but not __pic__ with -fPIC.
25#if (defined(__pic__) || defined(__PIC__)) && defined(__i386__)
26static WEBP_INLINE void GetCPUInfo(int cpu_info[4], int info_type) {
27  __asm__ volatile (
28    "mov %%ebx, %%edi\n"
29    "cpuid\n"
30    "xchg %%edi, %%ebx\n"
31    : "=a"(cpu_info[0]), "=D"(cpu_info[1]), "=c"(cpu_info[2]), "=d"(cpu_info[3])
32    : "a"(info_type), "c"(0));
33}
34#elif defined(__x86_64__) && \
35      (defined(__code_model_medium__) || defined(__code_model_large__)) && \
36      defined(__PIC__)
37static WEBP_INLINE void GetCPUInfo(int cpu_info[4], int info_type) {
38  __asm__ volatile (
39    "xchg{q}\t{%%rbx}, %q1\n"
40    "cpuid\n"
41    "xchg{q}\t{%%rbx}, %q1\n"
42    : "=a"(cpu_info[0]), "=&r"(cpu_info[1]), "=c"(cpu_info[2]),
43      "=d"(cpu_info[3])
44    : "a"(info_type), "c"(0));
45}
46#elif defined(__i386__) || defined(__x86_64__)
47static WEBP_INLINE void GetCPUInfo(int cpu_info[4], int info_type) {
48  __asm__ volatile (
49    "cpuid\n"
50    : "=a"(cpu_info[0]), "=b"(cpu_info[1]), "=c"(cpu_info[2]), "=d"(cpu_info[3])
51    : "a"(info_type), "c"(0));
52}
53#elif (defined(_M_X64) || defined(_M_IX86)) && \
54      defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 150030729  // >= VS2008 SP1
55#include <intrin.h>
56#define GetCPUInfo(info, type) __cpuidex(info, type, 0)  // set ecx=0
57#elif defined(WEBP_MSC_SSE2)
58#define GetCPUInfo __cpuid
59#endif
60
61// NaCl has no support for xgetbv or the raw opcode.
62#if !defined(__native_client__) && (defined(__i386__) || defined(__x86_64__))
63static WEBP_INLINE uint64_t xgetbv(void) {
64  const uint32_t ecx = 0;
65  uint32_t eax, edx;
66  // Use the raw opcode for xgetbv for compatibility with older toolchains.
67  __asm__ volatile (
68    ".byte 0x0f, 0x01, 0xd0\n"
69    : "=a"(eax), "=d"(edx) : "c" (ecx));
70  return ((uint64_t)edx << 32) | eax;
71}
72#elif (defined(_M_X64) || defined(_M_IX86)) && \
73      defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 160040219  // >= VS2010 SP1
74#include <immintrin.h>
75#define xgetbv() _xgetbv(0)
76#elif defined(_MSC_VER) && defined(_M_IX86)
77static WEBP_INLINE uint64_t xgetbv(void) {
78  uint32_t eax_, edx_;
79  __asm {
80    xor ecx, ecx  // ecx = 0
81    // Use the raw opcode for xgetbv for compatibility with older toolchains.
82    __asm _emit 0x0f __asm _emit 0x01 __asm _emit 0xd0
83    mov eax_, eax
84    mov edx_, edx
85  }
86  return ((uint64_t)edx_ << 32) | eax_;
87}
88#else
89#define xgetbv() 0U  // no AVX for older x64 or unrecognized toolchains.
90#endif
91
92#if defined(__i386__) || defined(__x86_64__) || defined(WEBP_MSC_SSE2)
93static int x86CPUInfo(CPUFeature feature) {
94  int max_cpuid_value;
95  int cpu_info[4];
96
97  // get the highest feature value cpuid supports
98  GetCPUInfo(cpu_info, 0);
99  max_cpuid_value = cpu_info[0];
100  if (max_cpuid_value < 1) {
101    return 0;
102  }
103
104  GetCPUInfo(cpu_info, 1);
105  if (feature == kSSE2) {
106    return 0 != (cpu_info[3] & 0x04000000);
107  }
108  if (feature == kSSE3) {
109    return 0 != (cpu_info[2] & 0x00000001);
110  }
111  if (feature == kSSE4_1) {
112    return 0 != (cpu_info[2] & 0x00080000);
113  }
114  if (feature == kAVX) {
115    // bits 27 (OSXSAVE) & 28 (256-bit AVX)
116    if ((cpu_info[2] & 0x18000000) == 0x18000000) {
117      // XMM state and YMM state enabled by the OS.
118      return (xgetbv() & 0x6) == 0x6;
119    }
120  }
121  if (feature == kAVX2) {
122    if (x86CPUInfo(kAVX) && max_cpuid_value >= 7) {
123      GetCPUInfo(cpu_info, 7);
124      return ((cpu_info[1] & 0x00000020) == 0x00000020);
125    }
126  }
127  return 0;
128}
129VP8CPUInfo VP8GetCPUInfo = x86CPUInfo;
130#elif defined(WEBP_ANDROID_NEON)  // NB: needs to be before generic NEON test.
131static int AndroidCPUInfo(CPUFeature feature) {
132  const AndroidCpuFamily cpu_family = android_getCpuFamily();
133  const uint64_t cpu_features = android_getCpuFeatures();
134  if (feature == kNEON) {
135    return (cpu_family == ANDROID_CPU_FAMILY_ARM &&
136            0 != (cpu_features & ANDROID_CPU_ARM_FEATURE_NEON));
137  }
138  return 0;
139}
140VP8CPUInfo VP8GetCPUInfo = AndroidCPUInfo;
141#elif defined(WEBP_USE_NEON)
142// define a dummy function to enable turning off NEON at runtime by setting
143// VP8DecGetCPUInfo = NULL
144static int armCPUInfo(CPUFeature feature) {
145  (void)feature;
146  return 1;
147}
148VP8CPUInfo VP8GetCPUInfo = armCPUInfo;
149#elif defined(WEBP_USE_MIPS32) || defined(WEBP_USE_MIPS_DSP_R2)
150static int mipsCPUInfo(CPUFeature feature) {
151  if ((feature == kMIPS32) || (feature == kMIPSdspR2)) {
152    return 1;
153  } else {
154    return 0;
155  }
156
157}
158VP8CPUInfo VP8GetCPUInfo = mipsCPUInfo;
159#else
160VP8CPUInfo VP8GetCPUInfo = NULL;
161#endif
162
163