The Gaudi Framework  master (fb0007c6)
Loading...
Searching...
No Matches
Functions
instrset_detect.cpp File Reference
#include "instrset.h"
Include dependency graph for instrset_detect.cpp:

Go to the source code of this file.

Functions

int instrset_detect (void)
 
bool hasFMA3 (void)
 
bool hasFMA4 (void)
 
bool hasXOP (void)
 
bool hasF16C (void)
 
bool hasAVX512ER (void)
 
bool hasAVX512VBMI (void)
 
bool hasAVX512VBMI2 (void)
 

Function Documentation

◆ hasAVX512ER()

bool hasAVX512ER ( void )

Definition at line 142 of file instrset_detect.cpp.

142 {
143 if ( instrset_detect() < 9 ) return false; // must have AVX512F
144 int abcd[4]; // cpuid results
145 cpuid( abcd, 7 ); // call cpuid function 7
146 return ( ( abcd[1] & ( 1 << 27 ) ) != 0 ); // ebx bit 27 indicates AVX512ER
147 }
instrset_detect
int instrset_detect(void)
Definition instrset_detect.cpp:63

◆ hasAVX512VBMI()

bool hasAVX512VBMI ( void )

Definition at line 150 of file instrset_detect.cpp.

150 {
151 if ( instrset_detect() < 10 ) return false; // must have AVX512BW
152 int abcd[4]; // cpuid results
153 cpuid( abcd, 7 ); // call cpuid function 7
154 return ( ( abcd[2] & ( 1 << 1 ) ) != 0 ); // ecx bit 1 indicates AVX512VBMI
155 }

◆ hasAVX512VBMI2()

bool hasAVX512VBMI2 ( void )

Definition at line 158 of file instrset_detect.cpp.

158 {
159 if ( instrset_detect() < 10 ) return false; // must have AVX512BW
160 int abcd[4]; // cpuid results
161 cpuid( abcd, 7 ); // call cpuid function 7
162 return ( ( abcd[2] & ( 1 << 6 ) ) != 0 ); // ecx bit 6 indicates AVX512VBMI2
163 }

◆ hasF16C()

bool hasF16C ( void )

Definition at line 134 of file instrset_detect.cpp.

134 {
135 if ( instrset_detect() < 7 ) return false; // must have AVX
136 int abcd[4]; // cpuid results
137 cpuid( abcd, 1 ); // call cpuid function 1
138 return ( ( abcd[2] & ( 1 << 29 ) ) != 0 ); // ecx bit 29 indicates F16C
139 }

◆ hasFMA3()

bool hasFMA3 ( void )

Definition at line 110 of file instrset_detect.cpp.

110 {
111 if ( instrset_detect() < 7 ) return false; // must have AVX
112 int abcd[4]; // cpuid results
113 cpuid( abcd, 1 ); // call cpuid function 1
114 return ( ( abcd[2] & ( 1 << 12 ) ) != 0 ); // ecx bit 12 indicates FMA3
115 }

◆ hasFMA4()

bool hasFMA4 ( void )

Definition at line 118 of file instrset_detect.cpp.

118 {
119 if ( instrset_detect() < 7 ) return false; // must have AVX
120 int abcd[4]; // cpuid results
121 cpuid( abcd, 0x80000001 ); // call cpuid function 0x80000001
122 return ( ( abcd[2] & ( 1 << 16 ) ) != 0 ); // ecx bit 16 indicates FMA4
123 }

◆ hasXOP()

bool hasXOP ( void )

Definition at line 126 of file instrset_detect.cpp.

126 {
127 if ( instrset_detect() < 7 ) return false; // must have AVX
128 int abcd[4]; // cpuid results
129 cpuid( abcd, 0x80000001 ); // call cpuid function 0x80000001
130 return ( ( abcd[2] & ( 1 << 11 ) ) != 0 ); // ecx bit 11 indicates XOP
131 }

◆ instrset_detect()

int instrset_detect ( void )

Definition at line 63 of file instrset_detect.cpp.

63 {
64
65 static int iset = -1; // remember value for next call
66 if ( iset >= 0 ) {
67 return iset; // called before
68 }
69 iset = 0; // default value
70 int abcd[4] = { 0, 0, 0, 0 }; // cpuid results
71 cpuid( abcd, 0 ); // call cpuid function 0
72 if ( abcd[0] == 0 ) return iset; // no further cpuid function supported
73 cpuid( abcd, 1 ); // call cpuid function 1 for feature flags
74 if ( ( abcd[3] & ( 1 << 0 ) ) == 0 ) return iset; // no floating point
75 if ( ( abcd[3] & ( 1 << 23 ) ) == 0 ) return iset; // no MMX
76 if ( ( abcd[3] & ( 1 << 15 ) ) == 0 ) return iset; // no conditional move
77 if ( ( abcd[3] & ( 1 << 24 ) ) == 0 ) return iset; // no FXSAVE
78 if ( ( abcd[3] & ( 1 << 25 ) ) == 0 ) return iset; // no SSE
79 iset = 1; // 1: SSE supported
80 if ( ( abcd[3] & ( 1 << 26 ) ) == 0 ) return iset; // no SSE2
81 iset = 2; // 2: SSE2 supported
82 if ( ( abcd[2] & ( 1 << 0 ) ) == 0 ) return iset; // no SSE3
83 iset = 3; // 3: SSE3 supported
84 if ( ( abcd[2] & ( 1 << 9 ) ) == 0 ) return iset; // no SSSE3
85 iset = 4; // 4: SSSE3 supported
86 if ( ( abcd[2] & ( 1 << 19 ) ) == 0 ) return iset; // no SSE4.1
87 iset = 5; // 5: SSE4.1 supported
88 if ( ( abcd[2] & ( 1 << 23 ) ) == 0 ) return iset; // no POPCNT
89 if ( ( abcd[2] & ( 1 << 20 ) ) == 0 ) return iset; // no SSE4.2
90 iset = 6; // 6: SSE4.2 supported
91 if ( ( abcd[2] & ( 1 << 27 ) ) == 0 ) return iset; // no OSXSAVE
92 if ( ( xgetbv( 0 ) & 6 ) != 6 ) return iset; // AVX not enabled in O.S.
93 if ( ( abcd[2] & ( 1 << 28 ) ) == 0 ) return iset; // no AVX
94 iset = 7; // 7: AVX supported
95 cpuid( abcd, 7 ); // call cpuid leaf 7 for feature flags
96 if ( ( abcd[1] & ( 1 << 5 ) ) == 0 ) return iset; // no AVX2
97 iset = 8;
98 if ( ( abcd[1] & ( 1 << 16 ) ) == 0 ) return iset; // no AVX512
99 cpuid( abcd, 0xD ); // call cpuid leaf 0xD for feature flags
100 if ( ( abcd[0] & 0x60 ) != 0x60 ) return iset; // no AVX512
101 iset = 9;
102 cpuid( abcd, 7 ); // call cpuid leaf 7 for feature flags
103 if ( ( abcd[1] & ( 1 << 31 ) ) == 0 ) return iset; // no AVX512VL
104 if ( ( abcd[1] & 0x40020000 ) != 0x40020000 ) return iset; // no AVX512BW, AVX512DQ
105 iset = 10;
106 return iset;
107 }