zzh
/
gatk-3.8
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Check only for SSE 4.1 (rather than SSE 4.2) when trying to use the SSE 

implementation of PairHMM
This commit is contained in:
Karthik Gururaj 2014-02-07 15:19:55 -08:00
parent dc44b64ad8
commit 20a46e4098
3 changed files with 21 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
protected/java/src/org/broadinstitute/sting/utils/pairhmm/VectorLoglessPairHMM.java
View File

@ -77,8 +77,9 @@ import java.io.OutputStream;
public class VectorLoglessPairHMM extends JNILoglessPairHMM { public class VectorLoglessPairHMM extends JNILoglessPairHMM {
//For machine capabilities //For machine capabilities
public static final long sse42Mask = 1; public static final long sse41Mask = 1;
public static final long avxMask = 2; public static final long sse42Mask = 2;
public static final long avxMask = 4;
public static final long enableAll = 0xFFFFFFFFFFFFFFFFl; public static final long enableAll = 0xFFFFFFFFFFFFFFFFl;
//Used to copy references to byteArrays to JNI from reads //Used to copy references to byteArrays to JNI from reads

18
public/c++/VectorPairHMM/utils.cc
View File

@ -22,6 +22,18 @@ bool is_avx_supported()
return ((ecx >> 28)&1) == 1; return ((ecx >> 28)&1) == 1;
} }
bool is_sse41_supported()
{
int ecx = 0, edx = 0, ebx = 0;
__asm__("cpuid"
: "=b" (ebx),
"=c" (ecx),
"=d" (edx)
: "a" (1)
);
return ((ecx >> 19)&1) == 1;
}
bool is_sse42_supported() bool is_sse42_supported()
{ {
int ecx = 0, edx = 0, ebx = 0; int ecx = 0, edx = 0, ebx = 0;
@ -41,6 +53,8 @@ uint64_t get_machine_capabilities()
machine_mask |= (1 << AVX_CUSTOM_IDX); machine_mask |= (1 << AVX_CUSTOM_IDX);
if(is_sse42_supported()) if(is_sse42_supported())
machine_mask |= (1 << SSE42_CUSTOM_IDX); machine_mask |= (1 << SSE42_CUSTOM_IDX);
if(is_sse41_supported())
machine_mask |= (1 << SSE41_CUSTOM_IDX);
return machine_mask; return machine_mask;
} }
@ -54,9 +68,9 @@ void initialize_function_pointers(uint64_t mask)
g_compute_full_prob_double = compute_full_prob_avxd<double>; g_compute_full_prob_double = compute_full_prob_avxd<double>;
} }
else else
if(is_sse42_supported() && (mask & (1<< SSE42_CUSTOM_IDX))) if(is_sse41_supported() && (mask & (1<< SSE41_CUSTOM_IDX)))
{ {
cout << "Using SSE4.2 accelerated implementation of PairHMM\n"; cout << "Using SSE4.1 accelerated implementation of PairHMM\n";
g_compute_full_prob_float = compute_full_prob_sses<float>; g_compute_full_prob_float = compute_full_prob_sses<float>;
g_compute_full_prob_double = compute_full_prob_ssed<double>; g_compute_full_prob_double = compute_full_prob_ssed<double>;
} }

3
public/c++/VectorPairHMM/utils.h
View File

@ -32,7 +32,8 @@ uint64_t diff_time(struct timespec& prev_time);
//bit 0 is sse4.2, bit 1 is AVX //bit 0 is sse4.2, bit 1 is AVX
enum ProcessorCapabilitiesEnum enum ProcessorCapabilitiesEnum
{ {
SSE42_CUSTOM_IDX=0, SSE41_CUSTOM_IDX=0,
SSE42_CUSTOM_IDX,
AVX_CUSTOM_IDX AVX_CUSTOM_IDX
}; };
#define ENABLE_ALL_HARDWARE_FEATURES 0xFFFFFFFFFFFFFFFFull #define ENABLE_ALL_HARDWARE_FEATURES 0xFFFFFFFFFFFFFFFFull