Pulled Mohammad's changes for creating variable sized arrays
Merge branch 'master' of /home/mghodrat/PairHMM/shared-repository into intel_pairhmm Conflicts: PairHMM_JNI/org_broadinstitute_sting_utils_pairhmm_VectorLoglessPairHMM.cc
This commit is contained in:
Karthik Gururaj
commit
a14a11c0cf
|
|
@ -1,10 +1,10 @@
|
||||||
#include "template.h"
|
#include "template.h"
|
||||||
|
|
||||||
#undef SIMD_TYPE
|
#undef SIMD_ENGINE
|
||||||
#undef SIMD_TYPE_SSE
|
#undef SIMD_ENGINE_SSE
|
||||||
|
|
||||||
#define SIMD_TYPE avx
|
#define SIMD_ENGINE avx
|
||||||
#define SIMD_TYPE_AVX
|
#define SIMD_ENGINE_AVX
|
||||||
|
|
||||||
#include "define-float.h"
|
#include "define-float.h"
|
||||||
#include "shift_template.c"
|
#include "shift_template.c"
|
||||||
|
|
|
||||||
|
|
@ -10,10 +10,10 @@ NUMBER compute_full_prob(testcase *tc, NUMBER *before_last_log = NULL)
|
||||||
|
|
||||||
Context<NUMBER> ctx;
|
Context<NUMBER> ctx;
|
||||||
|
|
||||||
NUMBER M[MROWS][MCOLS];
|
NUMBER M[ROWS][COLS];
|
||||||
NUMBER X[MROWS][MCOLS];
|
NUMBER X[ROWS][COLS];
|
||||||
NUMBER Y[MROWS][MCOLS];
|
NUMBER Y[ROWS][COLS];
|
||||||
NUMBER p[MROWS][6];
|
NUMBER p[ROWS][6];
|
||||||
|
|
||||||
p[0][MM] = ctx._(0.0);
|
p[0][MM] = ctx._(0.0);
|
||||||
p[0][GapM] = ctx._(0.0);
|
p[0][GapM] = ctx._(0.0);
|
||||||
|
|
|
||||||
|
|
@ -1,53 +1,51 @@
|
||||||
#include <iostream>
|
#include <iostream>
|
||||||
|
|
||||||
#ifdef PRECISION
|
#ifdef PRECISION
|
||||||
#undef PRECISION
|
#undef PRECISION
|
||||||
#undef MAIN_TYPE
|
#undef MAIN_TYPE
|
||||||
#undef MAIN_TYPE_SIZE
|
#undef MAIN_TYPE_SIZE
|
||||||
#undef UNION_TYPE
|
#undef UNION_TYPE
|
||||||
#undef IF_128
|
#undef IF_128
|
||||||
#undef IF_MAIN_TYPE
|
#undef IF_MAIN_TYPE
|
||||||
#undef SHIFT_CONST1
|
#undef SHIFT_CONST1
|
||||||
#undef SHIFT_CONST2
|
#undef SHIFT_CONST2
|
||||||
#undef SHIFT_CONST3
|
#undef SHIFT_CONST3
|
||||||
#undef _128_TYPE
|
#undef _128_TYPE
|
||||||
#undef _256_TYPE
|
#undef SIMD_TYPE
|
||||||
#undef AVX_LENGTH
|
#undef AVX_LENGTH
|
||||||
#undef MAVX_COUNT
|
#undef HAP_TYPE
|
||||||
#undef HAP_TYPE
|
#undef MASK_TYPE
|
||||||
#undef MASK_TYPE
|
#undef MASK_ALL_ONES
|
||||||
#undef MASK_ALL_ONES
|
|
||||||
|
|
||||||
#undef SET_VEC_ZERO(__vec)
|
#undef SET_VEC_ZERO(__vec)
|
||||||
#undef VEC_OR(__v1, __v2)
|
#undef VEC_OR(__v1, __v2)
|
||||||
#undef VEC_ADD(__v1, __v2)
|
#undef VEC_ADD(__v1, __v2)
|
||||||
#undef VEC_SUB(__v1, __v2)
|
#undef VEC_SUB(__v1, __v2)
|
||||||
#undef VEC_MUL(__v1, __v2)
|
#undef VEC_MUL(__v1, __v2)
|
||||||
#undef VEC_DIV(__v1, __v2)
|
#undef VEC_DIV(__v1, __v2)
|
||||||
#undef VEC_BLEND(__v1, __v2, __mask)
|
#undef VEC_BLEND(__v1, __v2, __mask)
|
||||||
#undef VEC_BLENDV(__v1, __v2, __maskV)
|
#undef VEC_BLENDV(__v1, __v2, __maskV)
|
||||||
#undef VEC_CAST_256_128(__v1)
|
#undef VEC_CAST_256_128(__v1)
|
||||||
#undef VEC_EXTRACT_128(__v1, __im)
|
#undef VEC_EXTRACT_128(__v1, __im)
|
||||||
#undef VEC_EXTRACT_UNIT(__v1, __im)
|
#undef VEC_EXTRACT_UNIT(__v1, __im)
|
||||||
#undef VEC_SET1_VAL128(__val)
|
#undef VEC_SET1_VAL128(__val)
|
||||||
#undef VEC_MOVE(__v1, __val)
|
#undef VEC_MOVE(__v1, __val)
|
||||||
#undef VEC_CAST_128_256(__v1)
|
#undef VEC_CAST_128_256(__v1)
|
||||||
#undef VEC_INSERT_VAL(__v1, __val, __pos)
|
#undef VEC_INSERT_VAL(__v1, __val, __pos)
|
||||||
#undef VEC_CVT_128_256(__v1)
|
#undef VEC_CVT_128_256(__v1)
|
||||||
#undef VEC_SET1_VAL(__val)
|
#undef VEC_SET1_VAL(__val)
|
||||||
#undef VEC_POPCVT_CHAR(__ch)
|
#undef VEC_POPCVT_CHAR(__ch)
|
||||||
#undef VEC_LDPOPCVT_CHAR(__addr)
|
#undef VEC_LDPOPCVT_CHAR(__addr)
|
||||||
#undef VEC_CMP_EQ(__v1, __v2)
|
#undef VEC_CMP_EQ(__v1, __v2)
|
||||||
#undef VEC_SET_LSE(__val)
|
#undef VEC_SET_LSE(__val)
|
||||||
#undef SHIFT_HAP(__v1, __val)
|
#undef SHIFT_HAP(__v1, __val)
|
||||||
#undef print256b(__v1)
|
#undef MASK_VEC
|
||||||
#undef MASK_VEC
|
#undef VEC_SSE_TO_AVX(__vsLow, __vsHigh, __vdst)
|
||||||
#undef VEC_SSE_TO_AVX(__vsLow, __vsHigh, __vdst)
|
#undef VEC_SHIFT_LEFT_1BIT(__vs)
|
||||||
#undef VEC_SHIFT_LEFT_1BIT(__vs)
|
#undef MASK_ALL_ONES
|
||||||
#undef MASK_ALL_ONES
|
#undef COMPARE_VECS(__v1, __v2)
|
||||||
#undef COMPARE_VECS(__v1, __v2)
|
#undef _256_INT_TYPE
|
||||||
#undef _256_INT_TYPE
|
#undef BITMASK_VEC
|
||||||
#undef BITMASK_VEC
|
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
#define PRECISION d
|
#define PRECISION d
|
||||||
|
|
@ -60,10 +58,9 @@
|
||||||
#define SHIFT_CONST2 1
|
#define SHIFT_CONST2 1
|
||||||
#define SHIFT_CONST3 8
|
#define SHIFT_CONST3 8
|
||||||
#define _128_TYPE __m128d
|
#define _128_TYPE __m128d
|
||||||
#define _256_TYPE __m256d
|
#define SIMD_TYPE __m256d
|
||||||
#define _256_INT_TYPE __m256i
|
#define _256_INT_TYPE __m256i
|
||||||
#define AVX_LENGTH 4
|
#define AVX_LENGTH 4
|
||||||
#define MAVX_COUNT (MROWS+7)/AVX_LENGTH
|
|
||||||
#define HAP_TYPE __m128i
|
#define HAP_TYPE __m128i
|
||||||
#define MASK_TYPE uint64_t
|
#define MASK_TYPE uint64_t
|
||||||
#define MASK_ALL_ONES 0xFFFFFFFFFFFFFFFF
|
#define MASK_ALL_ONES 0xFFFFFFFFFFFFFFFF
|
||||||
|
|
@ -135,12 +132,9 @@
|
||||||
#define SHIFT_HAP(__v1, __val) \
|
#define SHIFT_HAP(__v1, __val) \
|
||||||
__v1 = _mm_insert_epi32(_mm_slli_si128(__v1, 4), __val.i, 0)
|
__v1 = _mm_insert_epi32(_mm_slli_si128(__v1, 4), __val.i, 0)
|
||||||
|
|
||||||
#define print256b(__v1) \
|
|
||||||
print256bDP(__v1)
|
|
||||||
|
|
||||||
#define VEC_SSE_TO_AVX(__vsLow, __vsHigh, __vdst) \
|
#define VEC_SSE_TO_AVX(__vsLow, __vsHigh, __vdst) \
|
||||||
__vdst = _mm256_castpd128_pd256(__vsLow) ; \
|
__vdst = _mm256_castpd128_pd256(__vsLow) ; \
|
||||||
__vdst = _mm256_insertf128_pd(__vdst, __vsHigh, 1) ;
|
__vdst = _mm256_insertf128_pd(__vdst, __vsHigh, 1) ;
|
||||||
|
|
||||||
#define VEC_SHIFT_LEFT_1BIT(__vs) \
|
#define VEC_SHIFT_LEFT_1BIT(__vs) \
|
||||||
__vs = _mm_slli_epi64(__vs, 1)
|
__vs = _mm_slli_epi64(__vs, 1)
|
||||||
|
|
@ -156,15 +150,14 @@
|
||||||
exit(0) ; \
|
exit(0) ; \
|
||||||
} \
|
} \
|
||||||
} \
|
} \
|
||||||
}
|
}
|
||||||
|
|
||||||
class BitMaskVec_double {
|
class BitMaskVec_double {
|
||||||
|
|
||||||
MASK_VEC low_, high_ ;
|
MASK_VEC low_, high_ ;
|
||||||
_256_TYPE combined_ ;
|
SIMD_TYPE combined_ ;
|
||||||
|
|
||||||
public:
|
|
||||||
|
|
||||||
|
public:
|
||||||
inline MASK_TYPE& getLowEntry(int index) {
|
inline MASK_TYPE& getLowEntry(int index) {
|
||||||
return low_.masks[index] ;
|
return low_.masks[index] ;
|
||||||
}
|
}
|
||||||
|
|
@ -172,9 +165,8 @@ public:
|
||||||
return high_.masks[index] ;
|
return high_.masks[index] ;
|
||||||
}
|
}
|
||||||
|
|
||||||
inline const _256_TYPE& getCombinedMask() {
|
inline const SIMD_TYPE& getCombinedMask() {
|
||||||
VEC_SSE_TO_AVX(low_.vecf, high_.vecf, combined_) ;
|
VEC_SSE_TO_AVX(low_.vecf, high_.vecf, combined_) ;
|
||||||
|
|
||||||
return combined_ ;
|
return combined_ ;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
|
||||||
|
|
@ -1,53 +1,51 @@
|
||||||
#include <iostream>
|
#include <iostream>
|
||||||
|
|
||||||
#ifdef PRECISION
|
#ifdef PRECISION
|
||||||
#undef PRECISION
|
#undef PRECISION
|
||||||
#undef MAIN_TYPE
|
#undef MAIN_TYPE
|
||||||
#undef MAIN_TYPE_SIZE
|
#undef MAIN_TYPE_SIZE
|
||||||
#undef UNION_TYPE
|
#undef UNION_TYPE
|
||||||
#undef IF_128
|
#undef IF_128
|
||||||
#undef IF_MAIN_TYPE
|
#undef IF_MAIN_TYPE
|
||||||
#undef SHIFT_CONST1
|
#undef SHIFT_CONST1
|
||||||
#undef SHIFT_CONST2
|
#undef SHIFT_CONST2
|
||||||
#undef SHIFT_CONST3
|
#undef SHIFT_CONST3
|
||||||
#undef _128_TYPE
|
#undef _128_TYPE
|
||||||
#undef _256_TYPE
|
#undef SIMD_TYPE
|
||||||
#undef AVX_LENGTH
|
#undef AVX_LENGTH
|
||||||
#undef MAVX_COUNT
|
#undef HAP_TYPE
|
||||||
#undef HAP_TYPE
|
#undef MASK_TYPE
|
||||||
#undef MASK_TYPE
|
#undef MASK_ALL_ONES
|
||||||
#undef MASK_ALL_ONES
|
|
||||||
|
|
||||||
#undef SET_VEC_ZERO(__vec)
|
#undef SET_VEC_ZERO(__vec)
|
||||||
#undef VEC_OR(__v1, __v2)
|
#undef VEC_OR(__v1, __v2)
|
||||||
#undef VEC_ADD(__v1, __v2)
|
#undef VEC_ADD(__v1, __v2)
|
||||||
#undef VEC_SUB(__v1, __v2)
|
#undef VEC_SUB(__v1, __v2)
|
||||||
#undef VEC_MUL(__v1, __v2)
|
#undef VEC_MUL(__v1, __v2)
|
||||||
#undef VEC_DIV(__v1, __v2)
|
#undef VEC_DIV(__v1, __v2)
|
||||||
#undef VEC_BLEND(__v1, __v2, __mask)
|
#undef VEC_BLEND(__v1, __v2, __mask)
|
||||||
#undef VEC_BLENDV(__v1, __v2, __maskV)
|
#undef VEC_BLENDV(__v1, __v2, __maskV)
|
||||||
#undef VEC_CAST_256_128(__v1)
|
#undef VEC_CAST_256_128(__v1)
|
||||||
#undef VEC_EXTRACT_128(__v1, __im)
|
#undef VEC_EXTRACT_128(__v1, __im)
|
||||||
#undef VEC_EXTRACT_UNIT(__v1, __im)
|
#undef VEC_EXTRACT_UNIT(__v1, __im)
|
||||||
#undef VEC_SET1_VAL128(__val)
|
#undef VEC_SET1_VAL128(__val)
|
||||||
#undef VEC_MOVE(__v1, __val)
|
#undef VEC_MOVE(__v1, __val)
|
||||||
#undef VEC_CAST_128_256(__v1)
|
#undef VEC_CAST_128_256(__v1)
|
||||||
#undef VEC_INSERT_VAL(__v1, __val, __pos)
|
#undef VEC_INSERT_VAL(__v1, __val, __pos)
|
||||||
#undef VEC_CVT_128_256(__v1)
|
#undef VEC_CVT_128_256(__v1)
|
||||||
#undef VEC_SET1_VAL(__val)
|
#undef VEC_SET1_VAL(__val)
|
||||||
#undef VEC_POPCVT_CHAR(__ch)
|
#undef VEC_POPCVT_CHAR(__ch)
|
||||||
#undef VEC_LDPOPCVT_CHAR(__addr)
|
#undef VEC_LDPOPCVT_CHAR(__addr)
|
||||||
#undef VEC_CMP_EQ(__v1, __v2)
|
#undef VEC_CMP_EQ(__v1, __v2)
|
||||||
#undef VEC_SET_LSE(__val)
|
#undef VEC_SET_LSE(__val)
|
||||||
#undef SHIFT_HAP(__v1, __val)
|
#undef SHIFT_HAP(__v1, __val)
|
||||||
#undef print256b(__v1)
|
#undef MASK_VEC
|
||||||
#undef MASK_VEC
|
#undef VEC_SSE_TO_AVX(__vsLow, __vsHigh, __vdst)
|
||||||
#undef VEC_SSE_TO_AVX(__vsLow, __vsHigh, __vdst)
|
#undef VEC_SHIFT_LEFT_1BIT(__vs)
|
||||||
#undef VEC_SHIFT_LEFT_1BIT(__vs)
|
#undef MASK_ALL_ONES
|
||||||
#undef MASK_ALL_ONES
|
#undef COMPARE_VECS(__v1, __v2)
|
||||||
#undef COMPARE_VECS(__v1, __v2)
|
#undef _256_INT_TYPE
|
||||||
#undef _256_INT_TYPE
|
#undef BITMASK_VEC
|
||||||
#undef BITMASK_VEC
|
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
#define PRECISION s
|
#define PRECISION s
|
||||||
|
|
@ -61,10 +59,9 @@
|
||||||
#define SHIFT_CONST2 3
|
#define SHIFT_CONST2 3
|
||||||
#define SHIFT_CONST3 4
|
#define SHIFT_CONST3 4
|
||||||
#define _128_TYPE __m128
|
#define _128_TYPE __m128
|
||||||
#define _256_TYPE __m256
|
#define SIMD_TYPE __m256
|
||||||
#define _256_INT_TYPE __m256i
|
#define _256_INT_TYPE __m256i
|
||||||
#define AVX_LENGTH 8
|
#define AVX_LENGTH 8
|
||||||
#define MAVX_COUNT (MROWS+7)/AVX_LENGTH
|
|
||||||
#define HAP_TYPE UNION_TYPE
|
#define HAP_TYPE UNION_TYPE
|
||||||
#define MASK_TYPE uint32_t
|
#define MASK_TYPE uint32_t
|
||||||
#define MASK_ALL_ONES 0xFFFFFFFF
|
#define MASK_ALL_ONES 0xFFFFFFFF
|
||||||
|
|
@ -136,12 +133,9 @@
|
||||||
#define SHIFT_HAP(__v1, __val) \
|
#define SHIFT_HAP(__v1, __val) \
|
||||||
_vector_shift_lastavxs(__v1, __val.f);
|
_vector_shift_lastavxs(__v1, __val.f);
|
||||||
|
|
||||||
#define print256b(__v1) \
|
|
||||||
print256bFP(__v1)
|
|
||||||
|
|
||||||
#define VEC_SSE_TO_AVX(__vsLow, __vsHigh, __vdst) \
|
#define VEC_SSE_TO_AVX(__vsLow, __vsHigh, __vdst) \
|
||||||
__vdst = _mm256_castps128_ps256(__vsLow) ; \
|
__vdst = _mm256_castps128_ps256(__vsLow) ; \
|
||||||
__vdst = _mm256_insertf128_ps(__vdst, __vsHigh, 1) ;
|
__vdst = _mm256_insertf128_ps(__vdst, __vsHigh, 1) ;
|
||||||
|
|
||||||
#define VEC_SHIFT_LEFT_1BIT(__vs) \
|
#define VEC_SHIFT_LEFT_1BIT(__vs) \
|
||||||
__vs = _mm_slli_epi32(__vs, 1)
|
__vs = _mm_slli_epi32(__vs, 1)
|
||||||
|
|
@ -156,14 +150,14 @@
|
||||||
exit(0) ; \
|
exit(0) ; \
|
||||||
} \
|
} \
|
||||||
} \
|
} \
|
||||||
}
|
}
|
||||||
|
|
||||||
class BitMaskVec_float {
|
class BitMaskVec_float {
|
||||||
|
|
||||||
MASK_VEC low_, high_ ;
|
MASK_VEC low_, high_ ;
|
||||||
_256_TYPE combined_ ;
|
SIMD_TYPE combined_ ;
|
||||||
|
|
||||||
public:
|
public:
|
||||||
|
|
||||||
inline MASK_TYPE& getLowEntry(int index) {
|
inline MASK_TYPE& getLowEntry(int index) {
|
||||||
return low_.masks[index] ;
|
return low_.masks[index] ;
|
||||||
|
|
@ -172,9 +166,8 @@ public:
|
||||||
return high_.masks[index] ;
|
return high_.masks[index] ;
|
||||||
}
|
}
|
||||||
|
|
||||||
inline const _256_TYPE& getCombinedMask() {
|
inline const SIMD_TYPE& getCombinedMask() {
|
||||||
VEC_SSE_TO_AVX(low_.vecf, high_.vecf, combined_) ;
|
VEC_SSE_TO_AVX(low_.vecf, high_.vecf, combined_) ;
|
||||||
|
|
||||||
return combined_ ;
|
return combined_ ;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
|
||||||
|
|
@ -1,53 +1,51 @@
|
||||||
#ifdef PRECISION
|
#ifdef PRECISION
|
||||||
#undef PRECISION
|
#undef PRECISION
|
||||||
#undef MAIN_TYPE
|
#undef MAIN_TYPE
|
||||||
#undef MAIN_TYPE_SIZE
|
#undef MAIN_TYPE_SIZE
|
||||||
#undef UNION_TYPE
|
#undef UNION_TYPE
|
||||||
#undef IF_128
|
#undef IF_128
|
||||||
#undef IF_MAIN_TYPE
|
#undef IF_MAIN_TYPE
|
||||||
#undef SHIFT_CONST1
|
#undef SHIFT_CONST1
|
||||||
#undef SHIFT_CONST2
|
#undef SHIFT_CONST2
|
||||||
#undef SHIFT_CONST3
|
#undef SHIFT_CONST3
|
||||||
#undef _128_TYPE
|
#undef _128_TYPE
|
||||||
#undef _256_TYPE
|
#undef SIMD_TYPE
|
||||||
#undef AVX_LENGTH
|
#undef AVX_LENGTH
|
||||||
#undef MAVX_COUNT
|
#undef HAP_TYPE
|
||||||
#undef HAP_TYPE
|
#undef MASK_TYPE
|
||||||
#undef MASK_TYPE
|
#undef MASK_ALL_ONES
|
||||||
#undef MASK_ALL_ONES
|
|
||||||
|
|
||||||
#undef VEC_EXTRACT_UNIT(__v1, __im)
|
#undef VEC_EXTRACT_UNIT(__v1, __im)
|
||||||
#undef VEC_INSERT_UNIT(__v1,__ins,__im)
|
#undef VEC_INSERT_UNIT(__v1,__ins,__im)
|
||||||
#undef SET_VEC_ZERO(__vec)
|
#undef SET_VEC_ZERO(__vec)
|
||||||
#undef VEC_OR(__v1, __v2)
|
#undef VEC_OR(__v1, __v2)
|
||||||
#undef VEC_ADD(__v1, __v2)
|
#undef VEC_ADD(__v1, __v2)
|
||||||
#undef VEC_SUB(__v1, __v2)
|
#undef VEC_SUB(__v1, __v2)
|
||||||
#undef VEC_MUL(__v1, __v2)
|
#undef VEC_MUL(__v1, __v2)
|
||||||
#undef VEC_DIV(__v1, __v2)
|
#undef VEC_DIV(__v1, __v2)
|
||||||
#undef VEC_BLEND(__v1, __v2, __mask)
|
#undef VEC_BLEND(__v1, __v2, __mask)
|
||||||
#undef VEC_BLENDV(__v1, __v2, __maskV)
|
#undef VEC_BLENDV(__v1, __v2, __maskV)
|
||||||
#undef VEC_CAST_256_128(__v1)
|
#undef VEC_CAST_256_128(__v1)
|
||||||
#undef VEC_EXTRACT_128(__v1, __im)
|
#undef VEC_EXTRACT_128(__v1, __im)
|
||||||
#undef VEC_EXTRACT_UNIT(__v1, __im)
|
#undef VEC_EXTRACT_UNIT(__v1, __im)
|
||||||
#undef VEC_SET1_VAL128(__val)
|
#undef VEC_SET1_VAL128(__val)
|
||||||
#undef VEC_MOVE(__v1, __val)
|
#undef VEC_MOVE(__v1, __val)
|
||||||
#undef VEC_CAST_128_256(__v1)
|
#undef VEC_CAST_128_256(__v1)
|
||||||
#undef VEC_INSERT_VAL(__v1, __val, __pos)
|
#undef VEC_INSERT_VAL(__v1, __val, __pos)
|
||||||
#undef VEC_CVT_128_256(__v1)
|
#undef VEC_CVT_128_256(__v1)
|
||||||
#undef VEC_SET1_VAL(__val)
|
#undef VEC_SET1_VAL(__val)
|
||||||
#undef VEC_POPCVT_CHAR(__ch)
|
#undef VEC_POPCVT_CHAR(__ch)
|
||||||
#undef VEC_LDPOPCVT_CHAR(__addr)
|
#undef VEC_LDPOPCVT_CHAR(__addr)
|
||||||
#undef VEC_CMP_EQ(__v1, __v2)
|
#undef VEC_CMP_EQ(__v1, __v2)
|
||||||
#undef VEC_SET_LSE(__val)
|
#undef VEC_SET_LSE(__val)
|
||||||
#undef SHIFT_HAP(__v1, __val)
|
#undef SHIFT_HAP(__v1, __val)
|
||||||
#undef print256b(__v1)
|
#undef MASK_VEC
|
||||||
#undef MASK_VEC
|
#undef VEC_SSE_TO_AVX(__vsLow, __vsHigh, __vdst)
|
||||||
#undef VEC_SSE_TO_AVX(__vsLow, __vsHigh, __vdst)
|
#undef VEC_SHIFT_LEFT_1BIT(__vs)
|
||||||
#undef VEC_SHIFT_LEFT_1BIT(__vs)
|
#undef MASK_ALL_ONES
|
||||||
#undef MASK_ALL_ONES
|
#undef COMPARE_VECS(__v1, __v2)
|
||||||
#undef COMPARE_VECS(__v1, __v2)
|
#undef _256_INT_TYPE
|
||||||
#undef _256_INT_TYPE
|
#undef BITMASK_VEC
|
||||||
#undef BITMASK_VEC
|
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
#define SSE
|
#define SSE
|
||||||
|
|
@ -62,10 +60,9 @@
|
||||||
#define SHIFT_CONST2 8
|
#define SHIFT_CONST2 8
|
||||||
#define SHIFT_CONST3 0
|
#define SHIFT_CONST3 0
|
||||||
#define _128_TYPE __m128d
|
#define _128_TYPE __m128d
|
||||||
#define _256_TYPE __m128d
|
#define SIMD_TYPE __m128d
|
||||||
#define _256_INT_TYPE __m128i
|
#define _256_INT_TYPE __m128i
|
||||||
#define AVX_LENGTH 2
|
#define AVX_LENGTH 2
|
||||||
#define MAVX_COUNT (MROWS+3)/AVX_LENGTH
|
|
||||||
#define HAP_TYPE __m128i
|
#define HAP_TYPE __m128i
|
||||||
#define MASK_TYPE uint64_t
|
#define MASK_TYPE uint64_t
|
||||||
#define MASK_ALL_ONES 0xFFFFFFFFFFFFFFFFL
|
#define MASK_ALL_ONES 0xFFFFFFFFFFFFFFFFL
|
||||||
|
|
@ -129,9 +126,7 @@
|
||||||
class BitMaskVec_sse_double {
|
class BitMaskVec_sse_double {
|
||||||
|
|
||||||
MASK_VEC combined_ ;
|
MASK_VEC combined_ ;
|
||||||
|
public:
|
||||||
public:
|
|
||||||
|
|
||||||
inline MASK_TYPE& getLowEntry(int index) {
|
inline MASK_TYPE& getLowEntry(int index) {
|
||||||
return combined_.masks[index] ;
|
return combined_.masks[index] ;
|
||||||
}
|
}
|
||||||
|
|
@ -139,7 +134,7 @@ public:
|
||||||
return combined_.masks[AVX_LENGTH/2+index] ;
|
return combined_.masks[AVX_LENGTH/2+index] ;
|
||||||
}
|
}
|
||||||
|
|
||||||
inline const _256_TYPE& getCombinedMask() {
|
inline const SIMD_TYPE& getCombinedMask() {
|
||||||
return combined_.vecf ;
|
return combined_.vecf ;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
|
||||||
|
|
@ -1,53 +1,51 @@
|
||||||
#ifdef PRECISION
|
#ifdef PRECISION
|
||||||
#undef PRECISION
|
#undef PRECISION
|
||||||
#undef MAIN_TYPE
|
#undef MAIN_TYPE
|
||||||
#undef MAIN_TYPE_SIZE
|
#undef MAIN_TYPE_SIZE
|
||||||
#undef UNION_TYPE
|
#undef UNION_TYPE
|
||||||
#undef IF_128
|
#undef IF_128
|
||||||
#undef IF_MAIN_TYPE
|
#undef IF_MAIN_TYPE
|
||||||
#undef SHIFT_CONST1
|
#undef SHIFT_CONST1
|
||||||
#undef SHIFT_CONST2
|
#undef SHIFT_CONST2
|
||||||
#undef SHIFT_CONST3
|
#undef SHIFT_CONST3
|
||||||
#undef _128_TYPE
|
#undef _128_TYPE
|
||||||
#undef _256_TYPE
|
#undef SIMD_TYPE
|
||||||
#undef AVX_LENGTH
|
#undef AVX_LENGTH
|
||||||
#undef MAVX_COUNT
|
#undef HAP_TYPE
|
||||||
#undef HAP_TYPE
|
#undef MASK_TYPE
|
||||||
#undef MASK_TYPE
|
#undef MASK_ALL_ONES
|
||||||
#undef MASK_ALL_ONES
|
|
||||||
|
|
||||||
#undef VEC_EXTRACT_UNIT(__v1, __im)
|
#undef VEC_EXTRACT_UNIT(__v1, __im)
|
||||||
#undef VEC_INSERT_UNIT(__v1,__ins,__im)
|
#undef VEC_INSERT_UNIT(__v1,__ins,__im)
|
||||||
#undef SET_VEC_ZERO(__vec)
|
#undef SET_VEC_ZERO(__vec)
|
||||||
#undef VEC_OR(__v1, __v2)
|
#undef VEC_OR(__v1, __v2)
|
||||||
#undef VEC_ADD(__v1, __v2)
|
#undef VEC_ADD(__v1, __v2)
|
||||||
#undef VEC_SUB(__v1, __v2)
|
#undef VEC_SUB(__v1, __v2)
|
||||||
#undef VEC_MUL(__v1, __v2)
|
#undef VEC_MUL(__v1, __v2)
|
||||||
#undef VEC_DIV(__v1, __v2)
|
#undef VEC_DIV(__v1, __v2)
|
||||||
#undef VEC_BLEND(__v1, __v2, __mask)
|
#undef VEC_BLEND(__v1, __v2, __mask)
|
||||||
#undef VEC_BLENDV(__v1, __v2, __maskV)
|
#undef VEC_BLENDV(__v1, __v2, __maskV)
|
||||||
#undef VEC_CAST_256_128(__v1)
|
#undef VEC_CAST_256_128(__v1)
|
||||||
#undef VEC_EXTRACT_128(__v1, __im)
|
#undef VEC_EXTRACT_128(__v1, __im)
|
||||||
#undef VEC_EXTRACT_UNIT(__v1, __im)
|
#undef VEC_EXTRACT_UNIT(__v1, __im)
|
||||||
#undef VEC_SET1_VAL128(__val)
|
#undef VEC_SET1_VAL128(__val)
|
||||||
#undef VEC_MOVE(__v1, __val)
|
#undef VEC_MOVE(__v1, __val)
|
||||||
#undef VEC_CAST_128_256(__v1)
|
#undef VEC_CAST_128_256(__v1)
|
||||||
#undef VEC_INSERT_VAL(__v1, __val, __pos)
|
#undef VEC_INSERT_VAL(__v1, __val, __pos)
|
||||||
#undef VEC_CVT_128_256(__v1)
|
#undef VEC_CVT_128_256(__v1)
|
||||||
#undef VEC_SET1_VAL(__val)
|
#undef VEC_SET1_VAL(__val)
|
||||||
#undef VEC_POPCVT_CHAR(__ch)
|
#undef VEC_POPCVT_CHAR(__ch)
|
||||||
#undef VEC_LDPOPCVT_CHAR(__addr)
|
#undef VEC_LDPOPCVT_CHAR(__addr)
|
||||||
#undef VEC_CMP_EQ(__v1, __v2)
|
#undef VEC_CMP_EQ(__v1, __v2)
|
||||||
#undef VEC_SET_LSE(__val)
|
#undef VEC_SET_LSE(__val)
|
||||||
#undef SHIFT_HAP(__v1, __val)
|
#undef SHIFT_HAP(__v1, __val)
|
||||||
#undef print256b(__v1)
|
#undef MASK_VEC
|
||||||
#undef MASK_VEC
|
#undef VEC_SSE_TO_AVX(__vsLow, __vsHigh, __vdst)
|
||||||
#undef VEC_SSE_TO_AVX(__vsLow, __vsHigh, __vdst)
|
#undef VEC_SHIFT_LEFT_1BIT(__vs)
|
||||||
#undef VEC_SHIFT_LEFT_1BIT(__vs)
|
#undef MASK_ALL_ONES
|
||||||
#undef MASK_ALL_ONES
|
#undef COMPARE_VECS(__v1, __v2)
|
||||||
#undef COMPARE_VECS(__v1, __v2)
|
#undef _256_INT_TYPE
|
||||||
#undef _256_INT_TYPE
|
#undef BITMASK_VEC
|
||||||
#undef BITMASK_VEC
|
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
#define SSE
|
#define SSE
|
||||||
|
|
@ -62,10 +60,10 @@
|
||||||
#define SHIFT_CONST2 4
|
#define SHIFT_CONST2 4
|
||||||
#define SHIFT_CONST3 0
|
#define SHIFT_CONST3 0
|
||||||
#define _128_TYPE __m128
|
#define _128_TYPE __m128
|
||||||
#define _256_TYPE __m128
|
#define SIMD_TYPE __m128
|
||||||
#define _256_INT_TYPE __m128i
|
#define _256_INT_TYPE __m128i
|
||||||
#define AVX_LENGTH 4
|
#define AVX_LENGTH 4
|
||||||
#define MAVX_COUNT (MROWS+3)/AVX_LENGTH
|
//#define MAVX_COUNT (MROWS+3)/AVX_LENGTH
|
||||||
#define HAP_TYPE UNION_TYPE
|
#define HAP_TYPE UNION_TYPE
|
||||||
#define MASK_TYPE uint32_t
|
#define MASK_TYPE uint32_t
|
||||||
#define MASK_ALL_ONES 0xFFFFFFFF
|
#define MASK_ALL_ONES 0xFFFFFFFF
|
||||||
|
|
@ -129,8 +127,7 @@ class BitMaskVec_sse_float {
|
||||||
|
|
||||||
MASK_VEC combined_ ;
|
MASK_VEC combined_ ;
|
||||||
|
|
||||||
public:
|
public:
|
||||||
|
|
||||||
inline MASK_TYPE& getLowEntry(int index) {
|
inline MASK_TYPE& getLowEntry(int index) {
|
||||||
return combined_.masks[index] ;
|
return combined_.masks[index] ;
|
||||||
}
|
}
|
||||||
|
|
@ -138,7 +135,7 @@ public:
|
||||||
return combined_.masks[AVX_LENGTH/2+index] ;
|
return combined_.masks[AVX_LENGTH/2+index] ;
|
||||||
}
|
}
|
||||||
|
|
||||||
inline const _256_TYPE& getCombinedMask() {
|
inline const SIMD_TYPE& getCombinedMask() {
|
||||||
return combined_.vecf ;
|
return combined_.vecf ;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
|
||||||
|
|
@ -68,7 +68,7 @@ JNIEXPORT void JNICALL Java_org_broadinstitute_sting_utils_pairhmm_VectorLogless
|
||||||
haplotypeBasesLength = env->GetArrayLength(haplotypeBasesGlobalRef);
|
haplotypeBasesLength = env->GetArrayLength(haplotypeBasesGlobalRef);
|
||||||
#ifdef ENABLE_ASSERTIONS
|
#ifdef ENABLE_ASSERTIONS
|
||||||
assert(haplotypeBasesArray && "haplotypeBasesArray not initialized in JNI");
|
assert(haplotypeBasesArray && "haplotypeBasesArray not initialized in JNI");
|
||||||
assert(haplotypeBasesLength < MCOLS);
|
//assert(haplotypeBasesLength < MCOLS);
|
||||||
#endif
|
#endif
|
||||||
#ifdef DEBUG0_1
|
#ifdef DEBUG0_1
|
||||||
cout << "JNI haplotype length "<<haplotypeBasesLength<<"\n";
|
cout << "JNI haplotype length "<<haplotypeBasesLength<<"\n";
|
||||||
|
|
@ -145,7 +145,7 @@ JNIEXPORT void JNICALL Java_org_broadinstitute_sting_utils_pairhmm_VectorLogless
|
||||||
assert(insertionGOPArray && "insertionGOP array not initialized in JNI");
|
assert(insertionGOPArray && "insertionGOP array not initialized in JNI");
|
||||||
assert(deletionGOPArray && "deletionGOP array not initialized in JNI");
|
assert(deletionGOPArray && "deletionGOP array not initialized in JNI");
|
||||||
assert(overallGCPArray && "overallGCP array not initialized in JNI");
|
assert(overallGCPArray && "overallGCP array not initialized in JNI");
|
||||||
assert(readLength < MROWS);
|
//assert(readLength < MROWS);
|
||||||
assert(readLength == env->GetArrayLength(readQuals));
|
assert(readLength == env->GetArrayLength(readQuals));
|
||||||
assert(readLength == env->GetArrayLength(insertionGOP));
|
assert(readLength == env->GetArrayLength(insertionGOP));
|
||||||
assert(readLength == env->GetArrayLength(deletionGOP));
|
assert(readLength == env->GetArrayLength(deletionGOP));
|
||||||
|
|
|
||||||
|
|
@ -4,27 +4,8 @@
|
||||||
#include <assert.h>
|
#include <assert.h>
|
||||||
#include <stdlib.h>
|
#include <stdlib.h>
|
||||||
|
|
||||||
//#define DEBUG
|
|
||||||
#define MUSTAFA
|
|
||||||
#define KARTHIK
|
|
||||||
|
|
||||||
/*
|
void CONCAT(CONCAT(precompute_masks_,SIMD_ENGINE), PRECISION)(const testcase& tc, int COLS, int numMaskVecs, MASK_TYPE (*maskArr)[NUM_DISTINCT_CHARS]) {
|
||||||
template <class T>
|
|
||||||
string getBinaryStr (T val, int numBitsToWrite) {
|
|
||||||
|
|
||||||
ostringstream oss ;
|
|
||||||
uint64_t mask = ((T) 0x1) << (numBitsToWrite-1) ;
|
|
||||||
for (int i=numBitsToWrite-1; i >= 0; --i) {
|
|
||||||
oss << ((val & mask) >> i) ;
|
|
||||||
mask >>= 1 ;
|
|
||||||
}
|
|
||||||
return oss.str() ;
|
|
||||||
}
|
|
||||||
*/
|
|
||||||
#ifdef MUSTAFA
|
|
||||||
|
|
||||||
|
|
||||||
void GEN_INTRINSIC(GEN_INTRINSIC(precompute_masks_,SIMD_TYPE), PRECISION)(const testcase& tc, int COLS, int numMaskVecs, MASK_TYPE (*maskArr)[NUM_DISTINCT_CHARS]) {
|
|
||||||
|
|
||||||
const int maskBitCnt = MAIN_TYPE_SIZE ;
|
const int maskBitCnt = MAIN_TYPE_SIZE ;
|
||||||
|
|
||||||
|
|
@ -58,7 +39,7 @@ void GEN_INTRINSIC(GEN_INTRINSIC(precompute_masks_,SIMD_TYPE), PRECISION)(const
|
||||||
|
|
||||||
}
|
}
|
||||||
|
|
||||||
void GEN_INTRINSIC(GEN_INTRINSIC(init_masks_for_row_,SIMD_TYPE), PRECISION)(const testcase& tc, char* rsArr, MASK_TYPE* lastMaskShiftOut, int beginRowIndex, int numRowsToProcess) {
|
void CONCAT(CONCAT(init_masks_for_row_,SIMD_ENGINE), PRECISION)(const testcase& tc, char* rsArr, MASK_TYPE* lastMaskShiftOut, int beginRowIndex, int numRowsToProcess) {
|
||||||
|
|
||||||
for (int ri=0; ri < numRowsToProcess; ++ri) {
|
for (int ri=0; ri < numRowsToProcess; ++ri) {
|
||||||
rsArr[ri] = ConvertChar::get(tc.rs[ri+beginRowIndex-1]) ;
|
rsArr[ri] = ConvertChar::get(tc.rs[ri+beginRowIndex-1]) ;
|
||||||
|
|
@ -69,7 +50,6 @@ void GEN_INTRINSIC(GEN_INTRINSIC(init_masks_for_row_,SIMD_TYPE), PRECISION)(cons
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
#define SET_MASK_WORD(__dstMask, __srcMask, __lastShiftOut, __shiftBy, __maskBitCnt){ \
|
#define SET_MASK_WORD(__dstMask, __srcMask, __lastShiftOut, __shiftBy, __maskBitCnt){ \
|
||||||
MASK_TYPE __bitMask = (((MASK_TYPE)0x1) << __shiftBy) - 1 ; \
|
MASK_TYPE __bitMask = (((MASK_TYPE)0x1) << __shiftBy) - 1 ; \
|
||||||
MASK_TYPE __nextShiftOut = (__srcMask & __bitMask) << (__maskBitCnt - __shiftBy) ; \
|
MASK_TYPE __nextShiftOut = (__srcMask & __bitMask) << (__maskBitCnt - __shiftBy) ; \
|
||||||
|
|
@ -78,7 +58,7 @@ void GEN_INTRINSIC(GEN_INTRINSIC(init_masks_for_row_,SIMD_TYPE), PRECISION)(cons
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
void GEN_INTRINSIC(GEN_INTRINSIC(update_masks_for_cols_, SIMD_TYPE), PRECISION)(int maskIndex, BITMASK_VEC& bitMaskVec, MASK_TYPE (*maskArr) [NUM_DISTINCT_CHARS], char* rsArr, MASK_TYPE* lastMaskShiftOut, int maskBitCnt) {
|
void CONCAT(CONCAT(update_masks_for_cols_,SIMD_ENGINE), PRECISION)(int maskIndex, BITMASK_VEC& bitMaskVec, MASK_TYPE (*maskArr) [NUM_DISTINCT_CHARS], char* rsArr, MASK_TYPE* lastMaskShiftOut, int maskBitCnt) {
|
||||||
|
|
||||||
for (int ei=0; ei < AVX_LENGTH/2; ++ei) {
|
for (int ei=0; ei < AVX_LENGTH/2; ++ei) {
|
||||||
SET_MASK_WORD(bitMaskVec.getLowEntry(ei), maskArr[maskIndex][rsArr[ei]],
|
SET_MASK_WORD(bitMaskVec.getLowEntry(ei), maskArr[maskIndex][rsArr[ei]],
|
||||||
|
|
@ -92,48 +72,14 @@ void GEN_INTRINSIC(GEN_INTRINSIC(update_masks_for_cols_, SIMD_TYPE), PRECISION)(
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
//void GEN_INTRINSIC(computeDistVec, PRECISION) (BITMASK_VEC& bitMaskVec, _256_TYPE& distm, _256_TYPE& _1_distm, _256_TYPE& distmChosen, const _256_TYPE& distmSel, int firstRowIndex, int lastRowIndex) {
|
inline void CONCAT(CONCAT(computeDistVec,SIMD_ENGINE), PRECISION) (BITMASK_VEC& bitMaskVec, SIMD_TYPE& distm, SIMD_TYPE& _1_distm, SIMD_TYPE& distmChosen) {
|
||||||
|
|
||||||
inline void GEN_INTRINSIC(GEN_INTRINSIC(computeDistVec, SIMD_TYPE), PRECISION) (BITMASK_VEC& bitMaskVec, _256_TYPE& distm, _256_TYPE& _1_distm, _256_TYPE& distmChosen) {
|
|
||||||
//#define computeDistVec() {
|
|
||||||
|
|
||||||
#ifdef DEBUGG
|
|
||||||
long long *temp1 = (long long *)(&maskV);
|
|
||||||
double *temp2 = (double *)(&distm);
|
|
||||||
double *temp3 = (double *)(&_1_distm);
|
|
||||||
printf("***\n%lx\n%lx\n%f\n%f\n%f\n%f\n***\n", temp1[0], temp1[1], temp2[0], temp2[1], temp3[0], temp3[1]);
|
|
||||||
#endif
|
|
||||||
|
|
||||||
distmChosen = VEC_BLENDV(distm, _1_distm, bitMaskVec.getCombinedMask()) ;
|
distmChosen = VEC_BLENDV(distm, _1_distm, bitMaskVec.getCombinedMask()) ;
|
||||||
|
|
||||||
/*COMPARE_VECS(distmChosen, distmSel, firstRowIndex, lastRowIndex) ;*/
|
|
||||||
|
|
||||||
bitMaskVec.shift_left_1bit() ;
|
bitMaskVec.shift_left_1bit() ;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
template<class NUMBER> void CONCAT(CONCAT(initializeVectors,SIMD_ENGINE), PRECISION)(int ROWS, int COLS, NUMBER* shiftOutM, NUMBER *shiftOutX, NUMBER *shiftOutY, Context<NUMBER> ctx, testcase *tc, SIMD_TYPE *p_MM, SIMD_TYPE *p_GAPM, SIMD_TYPE *p_MX, SIMD_TYPE *p_XX, SIMD_TYPE *p_MY, SIMD_TYPE *p_YY, SIMD_TYPE *distm1D)
|
||||||
/*
|
|
||||||
template<class NUMBER>
|
|
||||||
struct HmmData {
|
|
||||||
int ROWS ;
|
|
||||||
int COLS ;
|
|
||||||
|
|
||||||
NUMBER shiftOutM[MROWS+MCOLS+AVX_LENGTH], shiftOutX[MROWS+MCOLS+AVX_LENGTH], shiftOutY[MROWS+MCOLS+AVX_LENGTH] ;
|
|
||||||
Context<NUMBER> ctx ;
|
|
||||||
testcase* tc ;
|
|
||||||
_256_TYPE p_MM[MAVX_COUNT], p_GAPM[MAVX_COUNT], p_MX[MAVX_COUNT], p_XX[MAVX_COUNT], p_MY[MAVX_COUNT], p_YY[MAVX_COUNT], distm1D[MAVX_COUNT] ;
|
|
||||||
_256_TYPE pGAPM, pMM, pMX, pXX, pMY, pYY ;
|
|
||||||
|
|
||||||
UNION_TYPE M_t, M_t_1, M_t_2, X_t, X_t_1, X_t_2, Y_t, Y_t_1, Y_t_2, M_t_y, M_t_1_y ;
|
|
||||||
UNION_TYPE rs , rsN ;
|
|
||||||
_256_TYPE distmSel;
|
|
||||||
_256_TYPE distm, _1_distm;
|
|
||||||
|
|
||||||
} ;
|
|
||||||
*/
|
|
||||||
#endif // MUSTAFA
|
|
||||||
|
|
||||||
template<class NUMBER> void GEN_INTRINSIC(GEN_INTRINSIC(initializeVectors, SIMD_TYPE), PRECISION)(int ROWS, int COLS, NUMBER* shiftOutM, NUMBER *shiftOutX, NUMBER *shiftOutY, Context<NUMBER> ctx, testcase *tc, _256_TYPE *p_MM, _256_TYPE *p_GAPM, _256_TYPE *p_MX, _256_TYPE *p_XX, _256_TYPE *p_MY, _256_TYPE *p_YY, _256_TYPE *distm1D)
|
|
||||||
{
|
{
|
||||||
NUMBER zero = ctx._(0.0);
|
NUMBER zero = ctx._(0.0);
|
||||||
NUMBER init_Y = ctx.INITIAL_CONSTANT / (tc->haplen);
|
NUMBER init_Y = ctx.INITIAL_CONSTANT / (tc->haplen);
|
||||||
|
|
@ -158,7 +104,6 @@ template<class NUMBER> void GEN_INTRINSIC(GEN_INTRINSIC(initializeVectors, SIMD_
|
||||||
*ptr_p_MY = ctx._(0.0);
|
*ptr_p_MY = ctx._(0.0);
|
||||||
*ptr_p_GAPM = ctx._(0.0);
|
*ptr_p_GAPM = ctx._(0.0);
|
||||||
|
|
||||||
|
|
||||||
for (int r = 1; r < ROWS; r++)
|
for (int r = 1; r < ROWS; r++)
|
||||||
{
|
{
|
||||||
int _i = tc->i[r-1] & 127;
|
int _i = tc->i[r-1] & 127;
|
||||||
|
|
@ -169,14 +114,8 @@ template<class NUMBER> void GEN_INTRINSIC(GEN_INTRINSIC(initializeVectors, SIMD_
|
||||||
*(ptr_p_GAPM+r-1) = ctx._(1.0) - ctx.ph2pr[_c];
|
*(ptr_p_GAPM+r-1) = ctx._(1.0) - ctx.ph2pr[_c];
|
||||||
*(ptr_p_MX+r-1) = ctx.ph2pr[_i];
|
*(ptr_p_MX+r-1) = ctx.ph2pr[_i];
|
||||||
*(ptr_p_XX+r-1) = ctx.ph2pr[_c];
|
*(ptr_p_XX+r-1) = ctx.ph2pr[_c];
|
||||||
#ifdef KARTHIK
|
|
||||||
*(ptr_p_MY+r-1) = ctx.ph2pr[_d];
|
*(ptr_p_MY+r-1) = ctx.ph2pr[_d];
|
||||||
*(ptr_p_YY+r-1) = ctx.ph2pr[_c];
|
*(ptr_p_YY+r-1) = ctx.ph2pr[_c];
|
||||||
#else
|
|
||||||
*(ptr_p_MY+r-1) = (r == ROWS - 1) ? ctx._(1.0) : ctx.ph2pr[_d];
|
|
||||||
*(ptr_p_YY+r-1) = (r == ROWS - 1) ? ctx._(1.0) : ctx.ph2pr[_c];
|
|
||||||
#endif
|
|
||||||
|
|
||||||
}
|
}
|
||||||
|
|
||||||
NUMBER *ptr_distm1D = (NUMBER *)distm1D;
|
NUMBER *ptr_distm1D = (NUMBER *)distm1D;
|
||||||
|
|
@ -188,13 +127,13 @@ template<class NUMBER> void GEN_INTRINSIC(GEN_INTRINSIC(initializeVectors, SIMD_
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
template<class NUMBER> inline void GEN_INTRINSIC(GEN_INTRINSIC(stripINITIALIZATION, SIMD_TYPE), PRECISION)(
|
template<class NUMBER> inline void CONCAT(CONCAT(stripeINITIALIZATION,SIMD_ENGINE), PRECISION)(
|
||||||
int stripIdx, Context<NUMBER> ctx, testcase *tc, _256_TYPE &pGAPM, _256_TYPE &pMM, _256_TYPE &pMX, _256_TYPE &pXX, _256_TYPE &pMY, _256_TYPE &pYY,
|
int stripeIdx, Context<NUMBER> ctx, testcase *tc, SIMD_TYPE &pGAPM, SIMD_TYPE &pMM, SIMD_TYPE &pMX, SIMD_TYPE &pXX, SIMD_TYPE &pMY, SIMD_TYPE &pYY,
|
||||||
_256_TYPE &rs, UNION_TYPE &rsN, _256_TYPE &distm, _256_TYPE &_1_distm, _256_TYPE *distm1D, _256_TYPE N_packed256, _256_TYPE *p_MM , _256_TYPE *p_GAPM ,
|
SIMD_TYPE &rs, UNION_TYPE &rsN, SIMD_TYPE &distm, SIMD_TYPE &_1_distm, SIMD_TYPE *distm1D, SIMD_TYPE N_packed256, SIMD_TYPE *p_MM , SIMD_TYPE *p_GAPM ,
|
||||||
_256_TYPE *p_MX, _256_TYPE *p_XX , _256_TYPE *p_MY, _256_TYPE *p_YY, UNION_TYPE &M_t_2, UNION_TYPE &X_t_2, UNION_TYPE &M_t_1, UNION_TYPE &X_t_1,
|
SIMD_TYPE *p_MX, SIMD_TYPE *p_XX , SIMD_TYPE *p_MY, SIMD_TYPE *p_YY, UNION_TYPE &M_t_2, UNION_TYPE &X_t_2, UNION_TYPE &M_t_1, UNION_TYPE &X_t_1,
|
||||||
UNION_TYPE &Y_t_2, UNION_TYPE &Y_t_1, UNION_TYPE &M_t_1_y, NUMBER* shiftOutX, NUMBER* shiftOutM)
|
UNION_TYPE &Y_t_2, UNION_TYPE &Y_t_1, UNION_TYPE &M_t_1_y, NUMBER* shiftOutX, NUMBER* shiftOutM)
|
||||||
{
|
{
|
||||||
int i = stripIdx;
|
int i = stripeIdx;
|
||||||
pGAPM = p_GAPM[i];
|
pGAPM = p_GAPM[i];
|
||||||
pMM = p_MM[i];
|
pMM = p_MM[i];
|
||||||
pMX = p_MX[i];
|
pMX = p_MX[i];
|
||||||
|
|
@ -206,17 +145,12 @@ template<class NUMBER> inline void GEN_INTRINSIC(GEN_INTRINSIC(stripINITIALIZATI
|
||||||
NUMBER init_Y = ctx.INITIAL_CONSTANT / (tc->haplen);
|
NUMBER init_Y = ctx.INITIAL_CONSTANT / (tc->haplen);
|
||||||
UNION_TYPE packed1; packed1.d = VEC_SET1_VAL(1.0);
|
UNION_TYPE packed1; packed1.d = VEC_SET1_VAL(1.0);
|
||||||
UNION_TYPE packed3; packed3.d = VEC_SET1_VAL(3.0);
|
UNION_TYPE packed3; packed3.d = VEC_SET1_VAL(3.0);
|
||||||
/* compare rs and N */
|
|
||||||
#ifndef MUSTAFA
|
|
||||||
rs = VEC_LDPOPCVT_CHAR((tc->irs+i*AVX_LENGTH));
|
|
||||||
rsN.d = VEC_CMP_EQ(N_packed256, rs);
|
|
||||||
#endif
|
|
||||||
distm = distm1D[i];
|
distm = distm1D[i];
|
||||||
_1_distm = VEC_SUB(packed1.d, distm);
|
_1_distm = VEC_SUB(packed1.d, distm);
|
||||||
|
|
||||||
#ifdef KARTHIK
|
|
||||||
distm = VEC_DIV(distm, packed3.d);
|
distm = VEC_DIV(distm, packed3.d);
|
||||||
#endif
|
|
||||||
/* initialize M_t_2, M_t_1, X_t_2, X_t_1, Y_t_2, Y_t_1 */
|
/* initialize M_t_2, M_t_1, X_t_2, X_t_1, Y_t_2, Y_t_1 */
|
||||||
M_t_2.d = VEC_SET1_VAL(zero);
|
M_t_2.d = VEC_SET1_VAL(zero);
|
||||||
X_t_2.d = VEC_SET1_VAL(zero);
|
X_t_2.d = VEC_SET1_VAL(zero);
|
||||||
|
|
@ -236,13 +170,11 @@ template<class NUMBER> inline void GEN_INTRINSIC(GEN_INTRINSIC(stripINITIALIZATI
|
||||||
M_t_1_y = M_t_1;
|
M_t_1_y = M_t_1;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
inline SIMD_TYPE CONCAT(CONCAT(computeDISTM,SIMD_ENGINE), PRECISION)(int d, int COLS, testcase * tc, HAP_TYPE &hap, SIMD_TYPE rs, UNION_TYPE rsN, SIMD_TYPE N_packed256,
|
||||||
|
SIMD_TYPE distm, SIMD_TYPE _1_distm)
|
||||||
inline _256_TYPE GEN_INTRINSIC(GEN_INTRINSIC(computeDISTM, SIMD_TYPE), PRECISION)(int d, int COLS, testcase * tc, HAP_TYPE &hap, _256_TYPE rs, UNION_TYPE rsN, _256_TYPE N_packed256,
|
|
||||||
_256_TYPE distm, _256_TYPE _1_distm)
|
|
||||||
{
|
{
|
||||||
UNION_TYPE hapN, rshap;
|
UNION_TYPE hapN, rshap;
|
||||||
_256_TYPE cond;
|
SIMD_TYPE cond;
|
||||||
IF_32 shiftInHap;
|
IF_32 shiftInHap;
|
||||||
|
|
||||||
int *hap_ptr = tc->ihap;
|
int *hap_ptr = tc->ihap;
|
||||||
|
|
@ -251,7 +183,7 @@ inline _256_TYPE GEN_INTRINSIC(GEN_INTRINSIC(computeDISTM, SIMD_TYPE), PRECISION
|
||||||
|
|
||||||
/* shift hap */
|
/* shift hap */
|
||||||
SHIFT_HAP(hap, shiftInHap);
|
SHIFT_HAP(hap, shiftInHap);
|
||||||
_256_TYPE hapF = VEC_CVT_128_256(hap);
|
SIMD_TYPE hapF = VEC_CVT_128_256(hap);
|
||||||
|
|
||||||
rshap.d = VEC_CMP_EQ(rs, hapF);
|
rshap.d = VEC_CMP_EQ(rs, hapF);
|
||||||
hapN.d = VEC_CMP_EQ(N_packed256, hapF);
|
hapN.d = VEC_CMP_EQ(N_packed256, hapF);
|
||||||
|
|
@ -261,27 +193,19 @@ inline _256_TYPE GEN_INTRINSIC(GEN_INTRINSIC(computeDISTM, SIMD_TYPE), PRECISION
|
||||||
cond = VEC_OR(cond, hapN.d);
|
cond = VEC_OR(cond, hapN.d);
|
||||||
|
|
||||||
/* distm1D = (cond) ? 1-distm1D : distm1D; */
|
/* distm1D = (cond) ? 1-distm1D : distm1D; */
|
||||||
_256_TYPE distmSel = VEC_BLENDV(distm, _1_distm, cond);
|
SIMD_TYPE distmSel = VEC_BLENDV(distm, _1_distm, cond);
|
||||||
|
|
||||||
return distmSel;
|
return distmSel;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
inline void GEN_INTRINSIC(GEN_INTRINSIC(computeMXY, SIMD_TYPE), PRECISION)(UNION_TYPE &M_t, UNION_TYPE &X_t, UNION_TYPE &Y_t, UNION_TYPE &M_t_y,
|
inline void CONCAT(CONCAT(computeMXY,SIMD_ENGINE), PRECISION)(UNION_TYPE &M_t, UNION_TYPE &X_t, UNION_TYPE &Y_t, UNION_TYPE &M_t_y,
|
||||||
UNION_TYPE M_t_2, UNION_TYPE X_t_2, UNION_TYPE Y_t_2, UNION_TYPE M_t_1, UNION_TYPE X_t_1, UNION_TYPE M_t_1_y, UNION_TYPE Y_t_1,
|
UNION_TYPE M_t_2, UNION_TYPE X_t_2, UNION_TYPE Y_t_2, UNION_TYPE M_t_1, UNION_TYPE X_t_1, UNION_TYPE M_t_1_y, UNION_TYPE Y_t_1,
|
||||||
_256_TYPE pMM, _256_TYPE pGAPM, _256_TYPE pMX, _256_TYPE pXX, _256_TYPE pMY, _256_TYPE pYY, _256_TYPE distmSel)
|
SIMD_TYPE pMM, SIMD_TYPE pGAPM, SIMD_TYPE pMX, SIMD_TYPE pXX, SIMD_TYPE pMY, SIMD_TYPE pYY, SIMD_TYPE distmSel)
|
||||||
{
|
{
|
||||||
/* Compute M_t <= distm * (p_MM*M_t_2 + p_GAPM*X_t_2 + p_GAPM*Y_t_2) */
|
/* Compute M_t <= distm * (p_MM*M_t_2 + p_GAPM*X_t_2 + p_GAPM*Y_t_2) */
|
||||||
M_t.d = VEC_MUL(VEC_ADD(VEC_ADD(VEC_MUL(M_t_2.d, pMM), VEC_MUL(X_t_2.d, pGAPM)), VEC_MUL(Y_t_2.d, pGAPM)), distmSel);
|
M_t.d = VEC_MUL(VEC_ADD(VEC_ADD(VEC_MUL(M_t_2.d, pMM), VEC_MUL(X_t_2.d, pGAPM)), VEC_MUL(Y_t_2.d, pGAPM)), distmSel);
|
||||||
|
|
||||||
#ifdef DEBUG
|
|
||||||
double *temp1 = (double *)(&pGAPM);
|
|
||||||
double *temp2 = (double *)(&pMM);
|
|
||||||
double *temp3 = (double *)(&distmSel);
|
|
||||||
printf("%f\n%f\n%f\n%f\n%f\n%f\n", temp1[0], temp1[1], temp2[0], temp2[1], temp3[0], temp3[1]);
|
|
||||||
//printf("%f\n%f\n%f\n%f\n", X_t_2.f[0], X_t_2.f[1], Y_t_2.f[0], Y_t_2.f[1]);
|
|
||||||
printf("%f\n%f\n----------------------------------------------------------------------------\n", M_t.f[0], M_t.f[1]);
|
|
||||||
#endif
|
|
||||||
M_t_y = M_t;
|
M_t_y = M_t;
|
||||||
|
|
||||||
/* Compute X_t */
|
/* Compute X_t */
|
||||||
|
|
@ -291,56 +215,52 @@ inline void GEN_INTRINSIC(GEN_INTRINSIC(computeMXY, SIMD_TYPE), PRECISION)(UNION
|
||||||
Y_t.d = VEC_ADD(VEC_MUL(M_t_1_y.d, pMY) , VEC_MUL(Y_t_1.d, pYY));
|
Y_t.d = VEC_ADD(VEC_MUL(M_t_1_y.d, pMY) , VEC_MUL(Y_t_1.d, pYY));
|
||||||
}
|
}
|
||||||
|
|
||||||
template<class NUMBER> NUMBER GEN_INTRINSIC(GEN_INTRINSIC(compute_full_prob_,SIMD_TYPE), PRECISION) (testcase *tc, NUMBER *before_last_log = NULL)
|
template<class NUMBER> NUMBER CONCAT(CONCAT(compute_full_prob_,SIMD_ENGINE), PRECISION) (testcase *tc, NUMBER *before_last_log = NULL)
|
||||||
{
|
{
|
||||||
_256_TYPE p_MM [MAVX_COUNT], p_GAPM [MAVX_COUNT], p_MX [MAVX_COUNT];
|
int ROWS = tc->rslen + 1;
|
||||||
_256_TYPE p_XX [MAVX_COUNT], p_MY [MAVX_COUNT], p_YY [MAVX_COUNT];
|
int COLS = tc->haplen + 1;
|
||||||
_256_TYPE distm1D[MAVX_COUNT];
|
int MAVX_COUNT = (ROWS+AVX_LENGTH-1)/AVX_LENGTH;
|
||||||
NUMBER shiftOutM[MROWS+MCOLS+AVX_LENGTH], shiftOutX[MROWS+MCOLS+AVX_LENGTH], shiftOutY[MROWS+MCOLS+AVX_LENGTH];
|
|
||||||
|
SIMD_TYPE p_MM [MAVX_COUNT], p_GAPM [MAVX_COUNT], p_MX [MAVX_COUNT];
|
||||||
|
SIMD_TYPE p_XX [MAVX_COUNT], p_MY [MAVX_COUNT], p_YY [MAVX_COUNT];
|
||||||
|
SIMD_TYPE distm1D[MAVX_COUNT];
|
||||||
|
NUMBER shiftOutM[ROWS+COLS+AVX_LENGTH], shiftOutX[ROWS+COLS+AVX_LENGTH], shiftOutY[ROWS+COLS+AVX_LENGTH];
|
||||||
UNION_TYPE M_t, M_t_1, M_t_2, X_t, X_t_1, X_t_2, Y_t, Y_t_1, Y_t_2, M_t_y, M_t_1_y;
|
UNION_TYPE M_t, M_t_1, M_t_2, X_t, X_t_1, X_t_2, Y_t, Y_t_1, Y_t_2, M_t_y, M_t_1_y;
|
||||||
_256_TYPE pGAPM, pMM, pMX, pXX, pMY, pYY;
|
SIMD_TYPE pGAPM, pMM, pMX, pXX, pMY, pYY;
|
||||||
|
|
||||||
struct timeval start, end;
|
struct timeval start, end;
|
||||||
NUMBER result_avx2;
|
NUMBER result_avx2;
|
||||||
Context<NUMBER> ctx;
|
Context<NUMBER> ctx;
|
||||||
UNION_TYPE rs , rsN;
|
UNION_TYPE rs , rsN;
|
||||||
HAP_TYPE hap;
|
HAP_TYPE hap;
|
||||||
_256_TYPE distmSel, distmChosen ;
|
SIMD_TYPE distmSel, distmChosen ;
|
||||||
_256_TYPE distm, _1_distm;
|
SIMD_TYPE distm, _1_distm;
|
||||||
|
|
||||||
int r, c;
|
int r, c;
|
||||||
int ROWS = tc->rslen + 1;
|
|
||||||
int COLS = tc->haplen + 1;
|
|
||||||
int AVX_COUNT = (ROWS+7)/8;
|
|
||||||
NUMBER zero = ctx._(0.0);
|
NUMBER zero = ctx._(0.0);
|
||||||
UNION_TYPE packed1; packed1.d = VEC_SET1_VAL(1.0);
|
UNION_TYPE packed1; packed1.d = VEC_SET1_VAL(1.0);
|
||||||
_256_TYPE N_packed256 = VEC_POPCVT_CHAR('N');
|
SIMD_TYPE N_packed256 = VEC_POPCVT_CHAR('N');
|
||||||
NUMBER init_Y = ctx.INITIAL_CONSTANT / (tc->haplen);
|
NUMBER init_Y = ctx.INITIAL_CONSTANT / (tc->haplen);
|
||||||
int remainingRows = (ROWS-1) % AVX_LENGTH;
|
int remainingRows = (ROWS-1) % AVX_LENGTH;
|
||||||
int strip_cnt = ((ROWS-1) / AVX_LENGTH) + (remainingRows!=0);
|
int stripe_cnt = ((ROWS-1) / AVX_LENGTH) + (remainingRows!=0);
|
||||||
|
|
||||||
#ifdef MUSTAFA
|
|
||||||
const int maskBitCnt = MAIN_TYPE_SIZE ;
|
const int maskBitCnt = MAIN_TYPE_SIZE ;
|
||||||
const int numMaskVecs = (COLS+ROWS+maskBitCnt-1)/maskBitCnt ; // ceil function
|
const int numMaskVecs = (COLS+ROWS+maskBitCnt-1)/maskBitCnt ; // ceil function
|
||||||
|
|
||||||
MASK_TYPE maskArr[numMaskVecs][NUM_DISTINCT_CHARS] ;
|
MASK_TYPE maskArr[numMaskVecs][NUM_DISTINCT_CHARS] ;
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(precompute_masks_,SIMD_TYPE), PRECISION)(*tc, COLS, numMaskVecs, maskArr) ;
|
CONCAT(CONCAT(precompute_masks_,SIMD_ENGINE), PRECISION)(*tc, COLS, numMaskVecs, maskArr) ;
|
||||||
|
|
||||||
char rsArr[AVX_LENGTH] ;
|
char rsArr[AVX_LENGTH] ;
|
||||||
MASK_TYPE lastMaskShiftOut[AVX_LENGTH] ;
|
MASK_TYPE lastMaskShiftOut[AVX_LENGTH] ;
|
||||||
#endif
|
CONCAT(CONCAT(initializeVectors,SIMD_ENGINE), PRECISION)<NUMBER>(ROWS, COLS, shiftOutM, shiftOutX, shiftOutY,
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(initializeVectors,SIMD_TYPE), PRECISION)<NUMBER>(ROWS, COLS, shiftOutM, shiftOutX, shiftOutY,
|
|
||||||
ctx, tc, p_MM, p_GAPM, p_MX, p_XX, p_MY, p_YY, distm1D);
|
ctx, tc, p_MM, p_GAPM, p_MX, p_XX, p_MY, p_YY, distm1D);
|
||||||
|
|
||||||
//for (int __ii=0; __ii < 10; ++__ii)
|
for (int i=0;i<stripe_cnt-1;i++)
|
||||||
for (int i=0;i<strip_cnt-1;i++)
|
|
||||||
{
|
{
|
||||||
//STRIP_INITIALIZATION
|
//STRIPE_INITIALIZATION
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(stripINITIALIZATION,SIMD_TYPE), PRECISION)(i, ctx, tc, pGAPM, pMM, pMX, pXX, pMY, pYY, rs.d, rsN, distm, _1_distm, distm1D, N_packed256, p_MM , p_GAPM ,
|
CONCAT(CONCAT(stripeINITIALIZATION,SIMD_ENGINE), PRECISION)(i, ctx, tc, pGAPM, pMM, pMX, pXX, pMY, pYY, rs.d, rsN, distm, _1_distm, distm1D, N_packed256, p_MM , p_GAPM ,
|
||||||
p_MX, p_XX , p_MY, p_YY, M_t_2, X_t_2, M_t_1, X_t_1, Y_t_2, Y_t_1, M_t_1_y, shiftOutX, shiftOutM);
|
p_MX, p_XX , p_MY, p_YY, M_t_2, X_t_2, M_t_1, X_t_1, Y_t_2, Y_t_1, M_t_1_y, shiftOutX, shiftOutM);
|
||||||
#ifdef MUSTAFA
|
CONCAT(CONCAT(init_masks_for_row_,SIMD_ENGINE), PRECISION)(*tc, rsArr, lastMaskShiftOut, i*AVX_LENGTH+1, AVX_LENGTH) ;
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(init_masks_for_row_,SIMD_TYPE), PRECISION)(*tc, rsArr, lastMaskShiftOut, i*AVX_LENGTH+1, AVX_LENGTH) ;
|
|
||||||
#endif
|
|
||||||
// Since there are no shift intrinsics in AVX, keep the masks in 2 SSE vectors
|
// Since there are no shift intrinsics in AVX, keep the masks in 2 SSE vectors
|
||||||
|
|
||||||
BITMASK_VEC bitMaskVec ;
|
BITMASK_VEC bitMaskVec ;
|
||||||
|
|
@ -348,28 +268,20 @@ template<class NUMBER> NUMBER GEN_INTRINSIC(GEN_INTRINSIC(compute_full_prob_,SIM
|
||||||
for (int begin_d=1;begin_d<COLS+AVX_LENGTH;begin_d+=MAIN_TYPE_SIZE)
|
for (int begin_d=1;begin_d<COLS+AVX_LENGTH;begin_d+=MAIN_TYPE_SIZE)
|
||||||
{
|
{
|
||||||
int numMaskBitsToProcess = std::min(MAIN_TYPE_SIZE, COLS+AVX_LENGTH-begin_d) ;
|
int numMaskBitsToProcess = std::min(MAIN_TYPE_SIZE, COLS+AVX_LENGTH-begin_d) ;
|
||||||
#ifdef MUSTAFA
|
CONCAT(CONCAT(update_masks_for_cols_,SIMD_ENGINE), PRECISION)((begin_d-1)/MAIN_TYPE_SIZE, bitMaskVec, maskArr, rsArr, lastMaskShiftOut, maskBitCnt) ;
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(update_masks_for_cols_,SIMD_TYPE), PRECISION)((begin_d-1)/MAIN_TYPE_SIZE, bitMaskVec, maskArr, rsArr, lastMaskShiftOut, maskBitCnt) ;
|
|
||||||
#endif
|
|
||||||
|
|
||||||
// if (d % MAIN_TYPE_SIZE == 1)
|
|
||||||
|
|
||||||
for (int mbi=0; mbi < numMaskBitsToProcess; ++mbi) {
|
for (int mbi=0; mbi < numMaskBitsToProcess; ++mbi) {
|
||||||
#ifdef MUSTAFA
|
CONCAT(CONCAT(computeDistVec,SIMD_ENGINE), PRECISION) (bitMaskVec, distm, _1_distm, distmChosen) ;
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(computeDistVec,SIMD_TYPE), PRECISION) (bitMaskVec, distm, _1_distm, distmChosen) ;
|
|
||||||
#else
|
|
||||||
distmChosen = GEN_INTRINSIC(GEN_INTRINSIC(computeDISTM,SIMD_TYPE), PRECISION)(begin_d+mbi, COLS, tc, hap, rs.d, rsN, N_packed256, distm, _1_distm);
|
|
||||||
#endif
|
|
||||||
int ShiftIdx = begin_d + mbi + AVX_LENGTH;
|
int ShiftIdx = begin_d + mbi + AVX_LENGTH;
|
||||||
|
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(computeMXY,SIMD_TYPE), PRECISION)(M_t, X_t, Y_t, M_t_y, M_t_2, X_t_2, Y_t_2, M_t_1, X_t_1, M_t_1_y, Y_t_1,
|
CONCAT(CONCAT(computeMXY,SIMD_ENGINE), PRECISION)(M_t, X_t, Y_t, M_t_y, M_t_2, X_t_2, Y_t_2, M_t_1, X_t_1, M_t_1_y, Y_t_1,
|
||||||
pMM, pGAPM, pMX, pXX, pMY, pYY, distmChosen);
|
pMM, pGAPM, pMX, pXX, pMY, pYY, distmChosen);
|
||||||
|
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift, SIMD_TYPE), PRECISION)(M_t, shiftOutM[ShiftIdx], shiftOutM[begin_d+mbi]);
|
CONCAT(CONCAT(_vector_shift,SIMD_ENGINE), PRECISION)(M_t, shiftOutM[ShiftIdx], shiftOutM[begin_d+mbi]);
|
||||||
|
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift, SIMD_TYPE), PRECISION)(X_t, shiftOutX[ShiftIdx], shiftOutX[begin_d+mbi]);
|
CONCAT(CONCAT(_vector_shift,SIMD_ENGINE), PRECISION)(X_t, shiftOutX[ShiftIdx], shiftOutX[begin_d+mbi]);
|
||||||
|
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift, SIMD_TYPE), PRECISION)(Y_t_1, shiftOutY[ShiftIdx], shiftOutY[begin_d+mbi]);
|
CONCAT(CONCAT(_vector_shift,SIMD_ENGINE), PRECISION)(Y_t_1, shiftOutY[ShiftIdx], shiftOutY[begin_d+mbi]);
|
||||||
|
|
||||||
M_t_2 = M_t_1; M_t_1 = M_t; X_t_2 = X_t_1; X_t_1 = X_t;
|
M_t_2 = M_t_1; M_t_1 = M_t; X_t_2 = X_t_1; X_t_1 = X_t;
|
||||||
Y_t_2 = Y_t_1; Y_t_1 = Y_t; M_t_1_y = M_t_y;
|
Y_t_2 = Y_t_1; Y_t_1 = Y_t; M_t_1_y = M_t_y;
|
||||||
|
|
@ -377,17 +289,16 @@ template<class NUMBER> NUMBER GEN_INTRINSIC(GEN_INTRINSIC(compute_full_prob_,SIM
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
int i = strip_cnt-1;
|
int i = stripe_cnt-1;
|
||||||
{
|
{
|
||||||
//STRIP_INITIALIZATION
|
//STRIPE_INITIALIZATION
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(stripINITIALIZATION,SIMD_TYPE), PRECISION)(i, ctx, tc, pGAPM, pMM, pMX, pXX, pMY, pYY, rs.d, rsN, distm, _1_distm, distm1D, N_packed256, p_MM , p_GAPM ,
|
CONCAT(CONCAT(stripeINITIALIZATION,SIMD_ENGINE), PRECISION)(i, ctx, tc, pGAPM, pMM, pMX, pXX, pMY, pYY, rs.d, rsN, distm, _1_distm, distm1D, N_packed256, p_MM , p_GAPM ,
|
||||||
p_MX, p_XX , p_MY, p_YY, M_t_2, X_t_2, M_t_1, X_t_1, Y_t_2, Y_t_1, M_t_1_y, shiftOutX, shiftOutM);
|
p_MX, p_XX , p_MY, p_YY, M_t_2, X_t_2, M_t_1, X_t_1, Y_t_2, Y_t_1, M_t_1_y, shiftOutX, shiftOutM);
|
||||||
|
|
||||||
if (remainingRows==0) remainingRows=AVX_LENGTH;
|
if (remainingRows==0) remainingRows=AVX_LENGTH;
|
||||||
#ifdef MUSTAFA
|
CONCAT(CONCAT(init_masks_for_row_,SIMD_ENGINE), PRECISION)(*tc, rsArr, lastMaskShiftOut, i*AVX_LENGTH+1, remainingRows) ;
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(init_masks_for_row_,SIMD_TYPE), PRECISION)(*tc, rsArr, lastMaskShiftOut, i*AVX_LENGTH+1, remainingRows) ;
|
|
||||||
#endif
|
SIMD_TYPE sumM, sumX;
|
||||||
_256_TYPE sumM, sumX;
|
|
||||||
sumM = VEC_SET1_VAL(zero);
|
sumM = VEC_SET1_VAL(zero);
|
||||||
sumX = VEC_SET1_VAL(zero);
|
sumX = VEC_SET1_VAL(zero);
|
||||||
|
|
||||||
|
|
@ -396,31 +307,24 @@ template<class NUMBER> NUMBER GEN_INTRINSIC(GEN_INTRINSIC(compute_full_prob_,SIM
|
||||||
|
|
||||||
for (int begin_d=1;begin_d<COLS+remainingRows-1;begin_d+=MAIN_TYPE_SIZE)
|
for (int begin_d=1;begin_d<COLS+remainingRows-1;begin_d+=MAIN_TYPE_SIZE)
|
||||||
{
|
{
|
||||||
|
|
||||||
int numMaskBitsToProcess = std::min(MAIN_TYPE_SIZE, COLS+remainingRows-1-begin_d) ;
|
int numMaskBitsToProcess = std::min(MAIN_TYPE_SIZE, COLS+remainingRows-1-begin_d) ;
|
||||||
#ifdef MUSTAFA
|
CONCAT(CONCAT(update_masks_for_cols_,SIMD_ENGINE),PRECISION)((begin_d-1)/MAIN_TYPE_SIZE, bitMaskVec, maskArr, rsArr, lastMaskShiftOut, maskBitCnt) ;
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(update_masks_for_cols_, SIMD_TYPE),PRECISION)((begin_d-1)/MAIN_TYPE_SIZE, bitMaskVec, maskArr, rsArr, lastMaskShiftOut, maskBitCnt) ;
|
|
||||||
#endif
|
|
||||||
|
|
||||||
for (int mbi=0; mbi < numMaskBitsToProcess; ++mbi) {
|
for (int mbi=0; mbi < numMaskBitsToProcess; ++mbi) {
|
||||||
|
|
||||||
#ifdef MUSTAFA
|
CONCAT(CONCAT(computeDistVec,SIMD_ENGINE), PRECISION) (bitMaskVec, distm, _1_distm, distmChosen) ;
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(computeDistVec, SIMD_TYPE), PRECISION) (bitMaskVec, distm, _1_distm, distmChosen) ;
|
|
||||||
#else
|
|
||||||
distmChosen = GEN_INTRINSIC(GEN_INTRINSIC(computeDISTM,SIMD_TYPE), PRECISION)(begin_d+mbi, COLS, tc, hap, rs.d, rsN, N_packed256, distm, _1_distm);
|
|
||||||
#endif
|
|
||||||
int ShiftIdx = begin_d + mbi +AVX_LENGTH;
|
int ShiftIdx = begin_d + mbi +AVX_LENGTH;
|
||||||
|
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(computeMXY, SIMD_TYPE), PRECISION)(M_t, X_t, Y_t, M_t_y, M_t_2, X_t_2, Y_t_2, M_t_1, X_t_1, M_t_1_y, Y_t_1,
|
CONCAT(CONCAT(computeMXY,SIMD_ENGINE), PRECISION)(M_t, X_t, Y_t, M_t_y, M_t_2, X_t_2, Y_t_2, M_t_1, X_t_1, M_t_1_y, Y_t_1,
|
||||||
pMM, pGAPM, pMX, pXX, pMY, pYY, distmChosen);
|
pMM, pGAPM, pMX, pXX, pMY, pYY, distmChosen);
|
||||||
|
|
||||||
sumM = VEC_ADD(sumM, M_t.d);
|
sumM = VEC_ADD(sumM, M_t.d);
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift_last, SIMD_TYPE), PRECISION)(M_t, shiftOutM[ShiftIdx]);
|
CONCAT(CONCAT(_vector_shift_last,SIMD_ENGINE), PRECISION)(M_t, shiftOutM[ShiftIdx]);
|
||||||
|
|
||||||
sumX = VEC_ADD(sumX, X_t.d);
|
sumX = VEC_ADD(sumX, X_t.d);
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift_last, SIMD_TYPE), PRECISION)(X_t, shiftOutX[ShiftIdx]);
|
CONCAT(CONCAT(_vector_shift_last,SIMD_ENGINE), PRECISION)(X_t, shiftOutX[ShiftIdx]);
|
||||||
|
|
||||||
GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift_last, SIMD_TYPE), PRECISION)(Y_t_1, shiftOutY[ShiftIdx]);
|
CONCAT(CONCAT(_vector_shift_last,SIMD_ENGINE), PRECISION)(Y_t_1, shiftOutY[ShiftIdx]);
|
||||||
|
|
||||||
M_t_2 = M_t_1; M_t_1 = M_t; X_t_2 = X_t_1; X_t_1 = X_t;
|
M_t_2 = M_t_1; M_t_1 = M_t; X_t_2 = X_t_1; X_t_1 = X_t;
|
||||||
Y_t_2 = Y_t_1; Y_t_1 = Y_t; M_t_1_y = M_t_y;
|
Y_t_2 = Y_t_1; Y_t_1 = Y_t; M_t_1_y = M_t_y;
|
||||||
|
|
@ -431,7 +335,6 @@ template<class NUMBER> NUMBER GEN_INTRINSIC(GEN_INTRINSIC(compute_full_prob_,SIM
|
||||||
sumMX.d = VEC_ADD(sumM, sumX);
|
sumMX.d = VEC_ADD(sumM, sumX);
|
||||||
result_avx2 = sumMX.f[remainingRows-1];
|
result_avx2 = sumMX.f[remainingRows-1];
|
||||||
}
|
}
|
||||||
//printf("result_avx2: %f\n", result_avx2);
|
|
||||||
return result_avx2;
|
return result_avx2;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
|
||||||
|
|
@ -1,10 +1,9 @@
|
||||||
#include "headers.h"
|
#include "headers.h"
|
||||||
|
|
||||||
#include "template.h"
|
#include "template.h"
|
||||||
#include "vector_defs.h"
|
#include "vector_defs.h"
|
||||||
|
|
||||||
#define SIMD_TYPE avx
|
#define SIMD_ENGINE avx
|
||||||
#define SIMD_TYPE_AVX
|
#define SIMD_ENGINE_AVX
|
||||||
|
|
||||||
|
|
||||||
#define BATCH_SIZE 10000
|
#define BATCH_SIZE 10000
|
||||||
|
|
@ -13,48 +12,12 @@
|
||||||
double getCurrClk();
|
double getCurrClk();
|
||||||
int thread_level_parallelism_enabled = false ;
|
int thread_level_parallelism_enabled = false ;
|
||||||
|
|
||||||
void print128b_F(__m128 x)
|
|
||||||
{
|
|
||||||
float *p = (float *)(&x);
|
|
||||||
for (int i=3;i>=0;i--)
|
|
||||||
printf("%f ", *(p+i));
|
|
||||||
printf("\n");
|
|
||||||
}
|
|
||||||
|
|
||||||
void print128b_D(__m128d x)
|
|
||||||
{
|
|
||||||
double *p = (double *)(&x);
|
|
||||||
for (int i=1;i>=0;i--)
|
|
||||||
printf("%f ", *(p+i));
|
|
||||||
printf("\n");
|
|
||||||
}
|
|
||||||
|
|
||||||
int main()
|
int main()
|
||||||
{
|
{
|
||||||
/*
|
|
||||||
IF_128f x;
|
|
||||||
x.f = _mm_set_ps(1.0, 2.0, 3.0, 4.0);
|
|
||||||
IF_32 shiftIn, shiftOut;
|
|
||||||
shiftIn.f = 5.0f;
|
|
||||||
print128b_F(x.f);
|
|
||||||
GEN_INTRINSIC(_vector_shift, s)(x, shiftIn, shiftOut);
|
|
||||||
print128b_F(x.f);
|
|
||||||
|
|
||||||
IF_128d y;
|
|
||||||
y.f = _mm_set_pd(10.0, 11.0);
|
|
||||||
IF_64 shiftInd, shiftOutd;
|
|
||||||
shiftInd.f = 12.0;
|
|
||||||
print128b_D(y.f);
|
|
||||||
GEN_INTRINSIC(_vector_shift, d)(y, shiftInd, shiftOutd);
|
|
||||||
print128b_D(y.f);
|
|
||||||
|
|
||||||
exit(0);
|
|
||||||
*/
|
|
||||||
|
|
||||||
testcase* tc = new testcase[BATCH_SIZE];
|
testcase* tc = new testcase[BATCH_SIZE];
|
||||||
float result[BATCH_SIZE], result_avxf;
|
float result[BATCH_SIZE], result_avxf;
|
||||||
double result_avxd;
|
double result_avxd;
|
||||||
//struct timeval start, end;
|
|
||||||
double lastClk = 0.0 ;
|
double lastClk = 0.0 ;
|
||||||
double aggregateTimeRead = 0.0;
|
double aggregateTimeRead = 0.0;
|
||||||
double aggregateTimeCompute = 0.0;
|
double aggregateTimeCompute = 0.0;
|
||||||
|
|
@ -63,11 +26,10 @@ int main()
|
||||||
// Need to call it once to initialize the static array
|
// Need to call it once to initialize the static array
|
||||||
ConvertChar::init() ;
|
ConvertChar::init() ;
|
||||||
|
|
||||||
|
// char* ompEnvVar = getenv("OMP_NUM_THREADS") ;
|
||||||
// char* ompEnvVar = getenv("OMP_NUM_THREADS") ;
|
// if (ompEnvVar != NULL && ompEnvVar != "" && ompEnvVar != "1" ) {
|
||||||
// if (ompEnvVar != NULL && ompEnvVar != "" && ompEnvVar != "1" ) {
|
// thread_level_parallelism_enabled = true ;
|
||||||
// thread_level_parallelism_enabled = true ;
|
// }
|
||||||
// }
|
|
||||||
|
|
||||||
bool noMoreData = false;
|
bool noMoreData = false;
|
||||||
int count =0;
|
int count =0;
|
||||||
|
|
@ -83,47 +45,34 @@ int main()
|
||||||
noMoreData = true;
|
noMoreData = true;
|
||||||
break;
|
break;
|
||||||
}
|
}
|
||||||
//gettimeofday(&end, NULL);
|
|
||||||
aggregateTimeRead += (getCurrClk() - lastClk) ;
|
aggregateTimeRead += (getCurrClk() - lastClk) ;
|
||||||
//((end.tv_sec * 1000000 + end.tv_usec) - (start.tv_sec * 1000000 + start.tv_usec));
|
|
||||||
|
|
||||||
//gettimeofday(&start, NULL);
|
|
||||||
lastClk = getCurrClk() ;
|
lastClk = getCurrClk() ;
|
||||||
|
|
||||||
//#pragma omp parallel for schedule(dynamic) if(thread_level_parallelism_enabled)
|
//#pragma omp parallel for schedule(dynamic) if(thread_level_parallelism_enabled)
|
||||||
for (int b=0;b<read_count;b++)
|
for (int b=0;b<read_count;b++)
|
||||||
{
|
{
|
||||||
result_avxf = GEN_INTRINSIC(GEN_INTRINSIC(compute_full_prob_, SIMD_TYPE), s)<float>(&tc[b]);
|
result_avxf = CONCAT(CONCAT(compute_full_prob_,SIMD_ENGINE), s)<float>(&tc[b]);
|
||||||
|
|
||||||
#ifdef RUN_HYBRID
|
#ifdef RUN_HYBRID
|
||||||
#define MIN_ACCEPTED 1e-28f
|
#define MIN_ACCEPTED 1e-28f
|
||||||
if (result_avxf < MIN_ACCEPTED) {
|
if (result_avxf < MIN_ACCEPTED) {
|
||||||
//printf("**************** RUNNING DOUBLE ******************\n");
|
|
||||||
count++;
|
count++;
|
||||||
result_avxd = GEN_INTRINSIC(GEN_INTRINSIC(compute_full_prob_, SIMD_TYPE), d)<double>(&tc[b]);
|
result_avxd = CONCAT(CONCAT(compute_full_prob_,SIMD_ENGINE), d)<double>(&tc[b]);
|
||||||
result[b] = log10(result_avxd) - log10(ldexp(1.0, 1020.f));
|
result[b] = log10(result_avxd) - log10(ldexp(1.0, 1020.f));
|
||||||
}
|
}
|
||||||
else
|
else
|
||||||
result[b] = log10f(result_avxf) - log10f(ldexpf(1.f, 120.f));
|
result[b] = log10f(result_avxf) - log10f(ldexpf(1.f, 120.f));
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
#ifndef RUN_HYBRID
|
#ifndef RUN_HYBRID
|
||||||
result[b] = log10f(result_avxf) - log10f(ldexpf(1.f, 120.f));
|
result[b] = log10f(result_avxf) - log10f(ldexpf(1.f, 120.f));
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
}
|
}
|
||||||
//gettimeofday(&end, NULL);
|
|
||||||
aggregateTimeCompute += (getCurrClk() - lastClk) ;
|
aggregateTimeCompute += (getCurrClk() - lastClk) ;
|
||||||
//((end.tv_sec * 1000000 + end.tv_usec) - (start.tv_sec * 1000000 + start.tv_usec));
|
|
||||||
|
|
||||||
//gettimeofday(&start, NULL);
|
|
||||||
lastClk = getCurrClk() ;
|
lastClk = getCurrClk() ;
|
||||||
//for (int b=0;b<read_count;b++)
|
for (int b=0;b<read_count;b++)
|
||||||
//printf("%E\n", result[b]);
|
printf("%E\n", result[b]);
|
||||||
//gettimeofday(&end, NULL);
|
|
||||||
aggregateTimeWrite += (getCurrClk() - lastClk) ;
|
aggregateTimeWrite += (getCurrClk() - lastClk) ;
|
||||||
//((end.tv_sec * 1000000 + end.tv_usec) - (start.tv_sec * 1000000 + start.tv_usec));
|
|
||||||
|
|
||||||
}
|
}
|
||||||
|
|
||||||
delete tc;
|
delete tc;
|
||||||
|
|
|
||||||
|
|
@ -1,10 +1,9 @@
|
||||||
#ifdef PRECISION
|
#ifdef PRECISION
|
||||||
|
|
||||||
#ifdef SIMD_TYPE_AVX
|
#ifdef SIMD_ENGINE_AVX
|
||||||
|
|
||||||
inline void GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift,SIMD_TYPE), PRECISION) (UNION_TYPE &x, MAIN_TYPE shiftIn, MAIN_TYPE &shiftOut)
|
inline void CONCAT(CONCAT(_vector_shift,SIMD_ENGINE), PRECISION) (UNION_TYPE &x, MAIN_TYPE shiftIn, MAIN_TYPE &shiftOut)
|
||||||
{
|
{
|
||||||
|
|
||||||
IF_128 xlow , xhigh;
|
IF_128 xlow , xhigh;
|
||||||
/* cast x to xlow */
|
/* cast x to xlow */
|
||||||
xlow.f = VEC_CAST_256_128(x.d);
|
xlow.f = VEC_CAST_256_128(x.d);
|
||||||
|
|
@ -33,9 +32,8 @@ inline void GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift,SIMD_TYPE), PRECISION) (UN
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
inline void GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift_last, SIMD_TYPE), PRECISION) (UNION_TYPE &x, MAIN_TYPE shiftIn)
|
inline void CONCAT(CONCAT(_vector_shift_last,SIMD_ENGINE), PRECISION) (UNION_TYPE &x, MAIN_TYPE shiftIn)
|
||||||
{
|
{
|
||||||
|
|
||||||
IF_128 xlow , xhigh;
|
IF_128 xlow , xhigh;
|
||||||
/* cast x to xlow */
|
/* cast x to xlow */
|
||||||
xlow.f = VEC_CAST_256_128(x.d);
|
xlow.f = VEC_CAST_256_128(x.d);
|
||||||
|
|
@ -61,9 +59,9 @@ inline void GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift_last, SIMD_TYPE), PRECISIO
|
||||||
|
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
#ifdef SIMD_TYPE_SSE
|
#ifdef SIMD_ENGINE_SSE
|
||||||
|
|
||||||
inline void GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift, SIMD_TYPE), PRECISION) (UNION_TYPE &x, MAIN_TYPE shiftIn, MAIN_TYPE &shiftOut)
|
inline void CONCAT(CONCAT(_vector_shift,SIMD_ENGINE), PRECISION) (UNION_TYPE &x, MAIN_TYPE shiftIn, MAIN_TYPE &shiftOut)
|
||||||
{
|
{
|
||||||
IF_MAIN_TYPE tempIn, tempOut;
|
IF_MAIN_TYPE tempIn, tempOut;
|
||||||
tempIn.f = shiftIn;
|
tempIn.f = shiftIn;
|
||||||
|
|
@ -76,7 +74,7 @@ inline void GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift, SIMD_TYPE), PRECISION) (U
|
||||||
x.i = VEC_INSERT_UNIT(x.i , tempIn.i, SHIFT_CONST3);
|
x.i = VEC_INSERT_UNIT(x.i , tempIn.i, SHIFT_CONST3);
|
||||||
}
|
}
|
||||||
|
|
||||||
inline void GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift_last, SIMD_TYPE), PRECISION) (UNION_TYPE &x, MAIN_TYPE shiftIn)
|
inline void CONCAT(CONCAT(_vector_shift_last,SIMD_ENGINE), PRECISION) (UNION_TYPE &x, MAIN_TYPE shiftIn)
|
||||||
{
|
{
|
||||||
IF_MAIN_TYPE temp; temp.f = shiftIn;
|
IF_MAIN_TYPE temp; temp.f = shiftIn;
|
||||||
/* shift */
|
/* shift */
|
||||||
|
|
|
||||||
|
|
@ -1,10 +1,10 @@
|
||||||
#include "template.h"
|
#include "template.h"
|
||||||
|
|
||||||
#undef SIMD_TYPE
|
#undef SIMD_ENGINE
|
||||||
#undef SIMD_TYPE_AVX
|
#undef SIMD_ENGINE_AVX
|
||||||
|
|
||||||
#define SIMD_TYPE sse
|
#define SIMD_ENGINE sse
|
||||||
#define SIMD_TYPE_SSE
|
#define SIMD_ENGINE_SSE
|
||||||
|
|
||||||
#include "define-sse-float.h"
|
#include "define-sse-float.h"
|
||||||
#include "shift_template.c"
|
#include "shift_template.c"
|
||||||
|
|
|
||||||
|
|
@ -10,11 +10,11 @@
|
||||||
#define MY 4
|
#define MY 4
|
||||||
#define YY 5
|
#define YY 5
|
||||||
|
|
||||||
#define MROWS 500
|
//#define MROWS 500
|
||||||
#define MCOLS 1000
|
//#define MCOLS 1000
|
||||||
|
|
||||||
#define CAT(X,Y) X####Y
|
#define CAT(X,Y) X####Y
|
||||||
#define GEN_INTRINSIC(X,Y) CAT(X,Y)
|
#define CONCAT(X,Y) CAT(X,Y)
|
||||||
|
|
||||||
#define ALIGNED __attribute__((aligned(32)))
|
#define ALIGNED __attribute__((aligned(32)))
|
||||||
|
|
||||||
|
|
|
||||||
|
|
@ -96,7 +96,7 @@ int read_testcase(testcase *tc, FILE* ifp)
|
||||||
|
|
||||||
tc->haplen = strlen(tc->hap);
|
tc->haplen = strlen(tc->hap);
|
||||||
tc->rslen = strlen(tc->rs);
|
tc->rslen = strlen(tc->rs);
|
||||||
assert(tc->rslen < MROWS);
|
//assert(tc->rslen < MROWS);
|
||||||
tc->ihap = (int *) malloc(tc->haplen*sizeof(int));
|
tc->ihap = (int *) malloc(tc->haplen*sizeof(int));
|
||||||
tc->irs = (int *) malloc(tc->rslen*sizeof(int));
|
tc->irs = (int *) malloc(tc->rslen*sizeof(int));
|
||||||
|
|
||||||
|
|
@ -216,7 +216,7 @@ int read_mod_testcase(ifstream& fptr, testcase* tc, bool reformat)
|
||||||
//cout << "Lengths "<<tc->haplen <<" "<<tc->rslen<<"\n";
|
//cout << "Lengths "<<tc->haplen <<" "<<tc->rslen<<"\n";
|
||||||
memcpy(tc->rs, tokens[1].c_str(),tokens[1].size());
|
memcpy(tc->rs, tokens[1].c_str(),tokens[1].size());
|
||||||
assert(tokens.size() == 2 + 4*(tc->rslen));
|
assert(tokens.size() == 2 + 4*(tc->rslen));
|
||||||
assert(tc->rslen < MROWS);
|
//assert(tc->rslen < MROWS);
|
||||||
for(unsigned j=0;j<tc->rslen;++j)
|
for(unsigned j=0;j<tc->rslen;++j)
|
||||||
tc->q[j] = (char)convToInt(tokens[2+0*tc->rslen+j]);
|
tc->q[j] = (char)convToInt(tokens[2+0*tc->rslen+j]);
|
||||||
for(unsigned j=0;j<tc->rslen;++j)
|
for(unsigned j=0;j<tc->rslen;++j)
|
||||||
|
|
|
||||||
|
|
@ -1,9 +1,9 @@
|
||||||
#undef SIMD_TYPE
|
#undef SIMD_ENGINE
|
||||||
#undef SIMD_TYPE_AVX
|
#undef SIMD_ENGINE_AVX
|
||||||
#undef SIMD_TYPE_SSE
|
#undef SIMD_ENGINE_SSE
|
||||||
|
|
||||||
#define SIMD_TYPE avx
|
#define SIMD_ENGINE avx
|
||||||
#define SIMD_TYPE_AVX
|
#define SIMD_ENGINE_AVX
|
||||||
|
|
||||||
#include "define-float.h"
|
#include "define-float.h"
|
||||||
#include "vector_function_prototypes.h"
|
#include "vector_function_prototypes.h"
|
||||||
|
|
@ -11,11 +11,11 @@
|
||||||
#include "define-double.h"
|
#include "define-double.h"
|
||||||
#include "vector_function_prototypes.h"
|
#include "vector_function_prototypes.h"
|
||||||
|
|
||||||
#undef SIMD_TYPE
|
#undef SIMD_ENGINE
|
||||||
#undef SIMD_TYPE_AVX
|
#undef SIMD_ENGINE_AVX
|
||||||
|
|
||||||
#define SIMD_TYPE sse
|
#define SIMD_ENGINE sse
|
||||||
#define SIMD_TYPE_SSE
|
#define SIMD_ENGINE_SSE
|
||||||
|
|
||||||
|
|
||||||
#include "define-sse-float.h"
|
#include "define-sse-float.h"
|
||||||
|
|
@ -24,7 +24,7 @@
|
||||||
#include "define-sse-double.h"
|
#include "define-sse-double.h"
|
||||||
#include "vector_function_prototypes.h"
|
#include "vector_function_prototypes.h"
|
||||||
|
|
||||||
#undef SIMD_TYPE
|
#undef SIMD_ENGINE
|
||||||
#undef SIMD_TYPE_AVX
|
#undef SIMD_ENGINE_AVX
|
||||||
#undef SIMD_TYPE_SSE
|
#undef SIMD_ENGINE_SSE
|
||||||
|
|
||||||
|
|
|
||||||
|
|
@ -1,19 +1,19 @@
|
||||||
inline void GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift,SIMD_TYPE), PRECISION) (UNION_TYPE &x, MAIN_TYPE shiftIn, MAIN_TYPE &shiftOut);
|
inline void CONCAT(CONCAT(_vector_shift,SIMD_ENGINE), PRECISION) (UNION_TYPE &x, MAIN_TYPE shiftIn, MAIN_TYPE &shiftOut);
|
||||||
inline void GEN_INTRINSIC(GEN_INTRINSIC(_vector_shift_last,SIMD_TYPE), PRECISION) (UNION_TYPE &x, MAIN_TYPE shiftIn);
|
inline void CONCAT(CONCAT(_vector_shift_last,SIMD_ENGINE), PRECISION) (UNION_TYPE &x, MAIN_TYPE shiftIn);
|
||||||
inline void GEN_INTRINSIC(GEN_INTRINSIC(precompute_masks_,SIMD_TYPE), PRECISION)(const testcase& tc, int COLS, int numMaskVecs, MASK_TYPE (*maskArr)[NUM_DISTINCT_CHARS]);
|
inline void CONCAT(CONCAT(precompute_masks_,SIMD_ENGINE), PRECISION)(const testcase& tc, int COLS, int numMaskVecs, MASK_TYPE (*maskArr)[NUM_DISTINCT_CHARS]);
|
||||||
inline void GEN_INTRINSIC(GEN_INTRINSIC(init_masks_for_row_,SIMD_TYPE), PRECISION)(const testcase& tc, char* rsArr, MASK_TYPE* lastMaskShiftOut, int beginRowIndex, int numRowsToProcess);
|
inline void CONCAT(CONCAT(init_masks_for_row_,SIMD_ENGINE), PRECISION)(const testcase& tc, char* rsArr, MASK_TYPE* lastMaskShiftOut, int beginRowIndex, int numRowsToProcess);
|
||||||
inline void GEN_INTRINSIC(GEN_INTRINSIC(update_masks_for_cols_,SIMD_TYPE), PRECISION)(int maskIndex, MASK_VEC& currMaskVecLow, MASK_VEC& currMaskVecHigh, MASK_TYPE (*maskArr) [NUM_DISTINCT_CHARS], char* rsArr, MASK_TYPE* lastMaskShiftOut, MASK_TYPE maskBitCnt);
|
inline void CONCAT(CONCAT(update_masks_for_cols_,SIMD_ENGINE), PRECISION)(int maskIndex, MASK_VEC& currMaskVecLow, MASK_VEC& currMaskVecHigh, MASK_TYPE (*maskArr) [NUM_DISTINCT_CHARS], char* rsArr, MASK_TYPE* lastMaskShiftOut, MASK_TYPE maskBitCnt);
|
||||||
inline void GEN_INTRINSIC(GEN_INTRINSIC(computeDistVec,SIMD_TYPE), PRECISION) (MASK_VEC& currMaskVecLow, MASK_VEC& currMaskVecHigh, _256_TYPE& distm, _256_TYPE& _1_distm, _256_TYPE& distmChosen);
|
inline void CONCAT(CONCAT(computeDistVec,SIMD_ENGINE), PRECISION) (MASK_VEC& currMaskVecLow, MASK_VEC& currMaskVecHigh, SIMD_TYPE& distm, SIMD_TYPE& _1_distm, SIMD_TYPE& distmChosen);
|
||||||
template<class NUMBER> inline void GEN_INTRINSIC(GEN_INTRINSIC(initializeVectors,SIMD_TYPE), PRECISION)(int ROWS, int COLS, NUMBER* shiftOutM, NUMBER *shiftOutX, NUMBER *shiftOutY, Context<NUMBER> ctx, testcase *tc, _256_TYPE *p_MM, _256_TYPE *p_GAPM, _256_TYPE *p_MX, _256_TYPE *p_XX, _256_TYPE *p_MY, _256_TYPE *p_YY, _256_TYPE *distm1D);
|
template<class NUMBER> inline void CONCAT(CONCAT(initializeVectors,SIMD_ENGINE), PRECISION)(int ROWS, int COLS, NUMBER* shiftOutM, NUMBER *shiftOutX, NUMBER *shiftOutY, Context<NUMBER> ctx, testcase *tc, SIMD_TYPE *p_MM, SIMD_TYPE *p_GAPM, SIMD_TYPE *p_MX, SIMD_TYPE *p_XX, SIMD_TYPE *p_MY, SIMD_TYPE *p_YY, SIMD_TYPE *distm1D);
|
||||||
template<class NUMBER> inline void GEN_INTRINSIC(GEN_INTRINSIC(stripINITIALIZATION,SIMD_TYPE), PRECISION)(
|
template<class NUMBER> inline void CONCAT(CONCAT(stripINITIALIZATION,SIMD_ENGINE), PRECISION)(
|
||||||
int stripIdx, Context<NUMBER> ctx, testcase *tc, _256_TYPE &pGAPM, _256_TYPE &pMM, _256_TYPE &pMX, _256_TYPE &pXX, _256_TYPE &pMY, _256_TYPE &pYY,
|
int stripIdx, Context<NUMBER> ctx, testcase *tc, SIMD_TYPE &pGAPM, SIMD_TYPE &pMM, SIMD_TYPE &pMX, SIMD_TYPE &pXX, SIMD_TYPE &pMY, SIMD_TYPE &pYY,
|
||||||
_256_TYPE &rs, UNION_TYPE &rsN, _256_TYPE &distm, _256_TYPE &_1_distm, _256_TYPE *distm1D, _256_TYPE N_packed256, _256_TYPE *p_MM , _256_TYPE *p_GAPM ,
|
SIMD_TYPE &rs, UNION_TYPE &rsN, SIMD_TYPE &distm, SIMD_TYPE &_1_distm, SIMD_TYPE *distm1D, SIMD_TYPE N_packed256, SIMD_TYPE *p_MM , SIMD_TYPE *p_GAPM ,
|
||||||
_256_TYPE *p_MX, _256_TYPE *p_XX , _256_TYPE *p_MY, _256_TYPE *p_YY, UNION_TYPE &M_t_2, UNION_TYPE &X_t_2, UNION_TYPE &M_t_1, UNION_TYPE &X_t_1,
|
SIMD_TYPE *p_MX, SIMD_TYPE *p_XX , SIMD_TYPE *p_MY, SIMD_TYPE *p_YY, UNION_TYPE &M_t_2, UNION_TYPE &X_t_2, UNION_TYPE &M_t_1, UNION_TYPE &X_t_1,
|
||||||
UNION_TYPE &Y_t_2, UNION_TYPE &Y_t_1, UNION_TYPE &M_t_1_y, NUMBER* shiftOutX, NUMBER* shiftOutM);
|
UNION_TYPE &Y_t_2, UNION_TYPE &Y_t_1, UNION_TYPE &M_t_1_y, NUMBER* shiftOutX, NUMBER* shiftOutM);
|
||||||
inline _256_TYPE GEN_INTRINSIC(GEN_INTRINSIC(computeDISTM,SIMD_TYPE), PRECISION)(int d, int COLS, testcase * tc, HAP_TYPE &hap, _256_TYPE rs, UNION_TYPE rsN, _256_TYPE N_packed256,
|
inline SIMD_TYPE CONCAT(CONCAT(computeDISTM,SIMD_ENGINE), PRECISION)(int d, int COLS, testcase * tc, HAP_TYPE &hap, SIMD_TYPE rs, UNION_TYPE rsN, SIMD_TYPE N_packed256,
|
||||||
_256_TYPE distm, _256_TYPE _1_distm);
|
SIMD_TYPE distm, SIMD_TYPE _1_distm);
|
||||||
inline void GEN_INTRINSIC(GEN_INTRINSIC(computeMXY,SIMD_TYPE), PRECISION)(UNION_TYPE &M_t, UNION_TYPE &X_t, UNION_TYPE &Y_t, UNION_TYPE &M_t_y,
|
inline void CONCAT(CONCAT(computeMXY,SIMD_ENGINE), PRECISION)(UNION_TYPE &M_t, UNION_TYPE &X_t, UNION_TYPE &Y_t, UNION_TYPE &M_t_y,
|
||||||
UNION_TYPE M_t_2, UNION_TYPE X_t_2, UNION_TYPE Y_t_2, UNION_TYPE M_t_1, UNION_TYPE X_t_1, UNION_TYPE M_t_1_y, UNION_TYPE Y_t_1,
|
UNION_TYPE M_t_2, UNION_TYPE X_t_2, UNION_TYPE Y_t_2, UNION_TYPE M_t_1, UNION_TYPE X_t_1, UNION_TYPE M_t_1_y, UNION_TYPE Y_t_1,
|
||||||
_256_TYPE pMM, _256_TYPE pGAPM, _256_TYPE pMX, _256_TYPE pXX, _256_TYPE pMY, _256_TYPE pYY, _256_TYPE distmSel);
|
SIMD_TYPE pMM, SIMD_TYPE pGAPM, SIMD_TYPE pMX, SIMD_TYPE pXX, SIMD_TYPE pMY, SIMD_TYPE pYY, SIMD_TYPE distmSel);
|
||||||
template<class NUMBER> NUMBER GEN_INTRINSIC(GEN_INTRINSIC(compute_full_prob_,SIMD_TYPE), PRECISION) (testcase *tc, NUMBER *before_last_log = NULL);
|
template<class NUMBER> NUMBER CONCAT(CONCAT(compute_full_prob_,SIMD_ENGINE), PRECISION) (testcase *tc, NUMBER *before_last_log = NULL);
|
||||||
|
|
||||||
|
|
|
||||||
12
build.xml
12
build.xml
|
|
@ -272,16 +272,16 @@
|
||||||
|
|
||||||
<!-- Comment out the following lines to build the GATK without a network connection, assuming you have all of the libraries cached already -->
|
<!-- Comment out the following lines to build the GATK without a network connection, assuming you have all of the libraries cached already -->
|
||||||
|
|
||||||
<!--<get src="http://repo1.maven.org/maven2/org/apache/ivy/ivy/${ivy.install.version}/${ivy.jar.file}"-->
|
<get src="http://repo1.maven.org/maven2/org/apache/ivy/ivy/${ivy.install.version}/${ivy.jar.file}"
|
||||||
<!--dest="${ivy.jar.dir}/${ivy.jar.file}"-->
|
dest="${ivy.jar.dir}/${ivy.jar.file}"
|
||||||
<!--usetimestamp="true"/>-->
|
usetimestamp="true"/>
|
||||||
<taskdef resource="org/apache/ivy/ant/antlib.xml"
|
<taskdef resource="org/apache/ivy/ant/antlib.xml"
|
||||||
uri="antlib:org.apache.ivy.ant"
|
uri="antlib:org.apache.ivy.ant"
|
||||||
classpath="${ivy.jar.dir}/${ivy.jar.file}"/>
|
classpath="${ivy.jar.dir}/${ivy.jar.file}"/>
|
||||||
|
|
||||||
<!--<get src="http://repo1.maven.org/maven2/org/apache/maven/maven-ant-tasks/${maven-ant-tasks.install.version}/${maven-ant-tasks.jar.file}"-->
|
<get src="http://repo1.maven.org/maven2/org/apache/maven/maven-ant-tasks/${maven-ant-tasks.install.version}/${maven-ant-tasks.jar.file}"
|
||||||
<!--dest="${ivy.jar.dir}/${maven-ant-tasks.jar.file}"-->
|
dest="${ivy.jar.dir}/${maven-ant-tasks.jar.file}"
|
||||||
<!--usetimestamp="true"/>-->
|
usetimestamp="true"/>
|
||||||
<taskdef resource="org/apache/maven/artifact/ant/antlib.xml"
|
<taskdef resource="org/apache/maven/artifact/ant/antlib.xml"
|
||||||
uri="antlib:antlib:org.apache.maven.artifact.ant"
|
uri="antlib:antlib:org.apache.maven.artifact.ant"
|
||||||
classpath="${ivy.jar.dir}/${maven-ant-tasks.jar.file}"/>
|
classpath="${ivy.jar.dir}/${maven-ant-tasks.jar.file}"/>
|
||||||
|
|
|
||||||
Loading…
Reference in New Issue