zzh
/
minimap2
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

r339: improved SIMD detection 

old code does not check AVX2
This commit is contained in:
Heng Li 2017-09-05 13:10:30 -04:00
parent 46e8b6a4f9
commit 8b9f2aaf04
2 changed files with 40 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

60
ksw2_dispatch.c
View File

@ -5,30 +5,48 @@
#define SIMD_SSE 0x1 #define SIMD_SSE 0x1
#define SIMD_SSE2 0x2 #define SIMD_SSE2 0x2
#define SIMD_SSE3 0x4 #define SIMD_SSE3 0x4
#define SIMD_SSE4_1 0x8 #define SIMD_SSSE3 0x8
#define SIMD_SSE4_2 0x10 #define SIMD_SSE4_1 0x10
#define SIMD_AVX 0x20 #define SIMD_SSE4_2 0x20
#define SIMD_AVX2 0x40 #define SIMD_AVX 0x40
#define SIMD_AVX512F 0x80 #define SIMD_AVX2 0x80
#define SIMD_AVX512F 0x100
unsigned x86_simd(void) #ifndef _MSC_VER
// adapted from https://github.com/01org/linux-sgx/blob/master/common/inc/internal/linux/cpuid_gnu.h
void __cpuidex(int cpuid[4], int func_id, int subfunc_id)
{ {
unsigned eax, ebx, ecx, edx, flag = 0; #if defined(__x86_64__)
#ifdef _MSC_VER asm volatile ("cpuid"
int cpuid[4]; : "=a" (cpuid[0]), "=b" (cpuid[1]), "=c" (cpuid[2]), "=d" (cpuid[3])
__cpuid(cpuid, 1); : "0" (func_id), "2" (subfunc_id));
eax = cpuid[0], ebx = cpuid[1], ecx = cpuid[2], edx = cpuid[3]; #else // on 32bit, ebx can NOT be used as PIC code
#else asm volatile ("xchgl %%ebx, %1; cpuid; xchgl %%ebx, %1"
asm volatile("cpuid" : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx) : "a" (1)); : "=a" (cpuid[0]), "=r" (cpuid[1]), "=c" (cpuid[2]), "=d" (cpuid[3])
: "0" (func_id), "2" (subfunc_id));
#endif #endif
if (edx>>25&1) flag |= SIMD_SSE; }
if (edx>>26&1) flag |= SIMD_SSE2; #endif
if (ecx>>0 &1) flag |= SIMD_SSE3;
if (ecx>>19&1) flag |= SIMD_SSE4_1; int x86_simd(void)
if (ecx>>20&1) flag |= SIMD_SSE4_2; {
if (ecx>>28&1) flag |= SIMD_AVX; int flag = 0, cpuid[4], max_id;
if (ebx>>5 &1) flag |= SIMD_AVX2; __cpuidex(cpuid, 0, 0);
if (ebx>>16&1) flag |= SIMD_AVX512F; max_id = cpuid[0];
if (max_id == 0) return 0;
__cpuidex(cpuid, 1, 0);
if (cpuid[3]>>25&1) flag |= SIMD_SSE;
if (cpuid[3]>>26&1) flag |= SIMD_SSE2;
if (cpuid[2]>>0 &1) flag |= SIMD_SSE3;
if (cpuid[2]>>9 &1) flag |= SIMD_SSSE3;
if (cpuid[2]>>19&1) flag |= SIMD_SSE4_1;
if (cpuid[2]>>20&1) flag |= SIMD_SSE4_2;
if (cpuid[2]>>28&1) flag |= SIMD_AVX;
if (max_id >= 7) {
__cpuidex(cpuid, 7, 0);
if (cpuid[1]>>5 &1) flag |= SIMD_AVX2;
if (cpuid[1]>>16&1) flag |= SIMD_AVX512F;
}
return flag; return flag;
} }

2
main.c
View File

@ -6,7 +6,7 @@
#include "mmpriv.h" #include "mmpriv.h"
#include "getopt.h" #include "getopt.h"
#define MM_VERSION "2.1-r338-dirty" #define MM_VERSION "2.1-r339-dirty"
#ifdef __linux__ #ifdef __linux__
#include <sys/resource.h> #include <sys/resource.h>