Commit 73307486 authored by Michael R. Crusoe's avatar Michael R. Crusoe 🏳️‍🌈
Browse files

New upstream version 1045

parent 60c6d42f

ChangeLog.txt

+8 −1
Original line number Diff line number Diff line 
2019-12-23  Martin C. Frith  <Martin C. Frith>



	* src/Alignment.cc, src/GappedXdropAligner.hh,

	src/GappedXdropAlignerDna.cc, src/mcf_simd.hh:

	Make it usable without SSE/SIMD

	[4270e27518ca] [tip]



2019-12-20  Martin C. Frith  <Martin C. Frith>



	* src/GappedXdropAligner.hh, src/GappedXdropAlignerDna.cc,

	src/mcf_simd.hh, test/last-test.out, test/last-test.sh:

	Maybe make lastal gapped alignment a bit faster

	[11810fcff80c] [tip]

	[11810fcff80c]



2019-12-19  Martin C. Frith  <Martin C. Frith>

src/Alignment.cc

+8 −2
Original line number Diff line number Diff line
@@ -347,11 +347,13 @@ void Alignment::extend( std::vector< SegmentPair >& chunks, 
				  del.openCost, del.growCost,

				  ins.openCost, ins.growCost,

				  gap.pairCost, gap.isAffine, maxDrop, smMax)

#if defined __SSE4_1__

    : isSimdMatrix ? aligner.alignDna(seq1 + start1, seq2 + start2,

				      isForward, sm,

				      del.openCost, del.growCost,

				      ins.openCost, ins.growCost,

				      maxDrop, smMax, alph.numbersToUppercase)

#endif

    :           aligner.align(seq1 + start1, seq2 + start2,

			      isForward, globality, sm,

			      del.openCost, del.growCost,
@@ -370,12 +372,16 @@ void Alignment::extend( std::vector< SegmentPair >& chunks, 
    if( isGreedy ){

      while( greedyAligner.getNextChunk( end1, end2, size ) )

	chunks.push_back( SegmentPair( end1 - size, end2 - size, size ) );

    } else if (isSimdMatrix && !pssm2 && !sm2qual) {

    }

#if defined __SSE4_1__

    else if (isSimdMatrix && !pssm2 && !sm2qual) {

      while (aligner.getNextChunkDna(end1, end2, size,

				     del.openCost, del.growCost,

				     ins.openCost, ins.growCost))

	chunks.push_back(SegmentPair(end1 - size, end2 - size, size));

    }else{

    }

#endif

    else {

      while( aligner.getNextChunk( end1, end2, size,

				   del.openCost, del.growCost,

				   ins.openCost, ins.growCost, gap.pairCost ) )

src/GappedXdropAligner.hh

+2 −1
Original line number Diff line number Diff line
@@ -317,7 +317,7 @@ class GappedXdropAligner { 
  void init3();



  // Everything below here is for alignDna & getNextChunkDna



#if defined __SSE4_1__

  std::vector<TinyScore> xTinyScores;

  std::vector<TinyScore> yTinyScores;

  std::vector<TinyScore> zTinyScores;
@@ -362,6 +362,7 @@ class GappedXdropAligner { 
    while (*x2 != target) ++x2;

    bestSeq1position = x2 - x2beg + seq1beg;

  }

#endif

};



}

src/GappedXdropAlignerDna.cc

+6 −2
Original line number Diff line number Diff line 
// Author: Martin C. Frith 2019

// SPDX-License-Identifier: GPL-3.0-or-later



#if defined __SSE4_1__



#include "GappedXdropAligner.hh"

#include "GappedXdropAlignerInl.hh"
@@ -109,7 +111,7 @@ int GappedXdropAligner::alignDna(const uchar *seq1, 
	SimdInt y = simdAdds1(simdLoad(y1+i), mDelGrowCost1);

	SimdInt z = simdAdds1(simdLoad(z1+i), mInsGrowCost1);

	SimdInt b = simdMin1(simdMin1(x, y), z);

	SimdInt isDrop = simdEq1(simdMin1(b, mBadScore), mBadScore);

	SimdInt isDrop = simdGe1(b, mBadScore);

	mBestScore = simdMin1(b, mBestScore);

	simdStore(x0+i, simdOr(simdSub1(b, s), isDrop));

	simdStore(y0+i, simdMin1(simdAdds1(b, mDelOpenCost), y));
@@ -166,7 +168,7 @@ int GappedXdropAligner::alignDna(const uchar *seq1, 
	SimdInt y = simdAdds1(simdLoad(y1+i), mDelGrowCost1);

	SimdInt z = simdAdds1(simdLoad(z1+i), mInsGrowCost1);

	SimdInt b = simdMin1(simdMin1(x, y), z);

	SimdInt isDrop = simdEq1(simdMin1(b, mBadScore), mBadScore);

	SimdInt isDrop = simdGe1(b, mBadScore);

	mBestScore = simdMin1(b, mBestScore);

	simdStore(x0+i, simdOr(simdSub1(b, s), isDrop));

	simdStore(y0+i, simdMin1(simdAdds1(b, mDelOpenCost), y));
@@ -273,3 +275,5 @@ bool GappedXdropAligner::getNextChunkDna(size_t &end1, 
}



}



#endif

src/mcf_simd.hh

+9 −74
Original line number Diff line number Diff line
@@ -4,7 +4,9 @@ 
#ifndef MCF_SIMD_HH

#define MCF_SIMD_HH



#if defined __SSE4_1__

#include <immintrin.h>

#endif



namespace mcf {
@@ -39,21 +41,12 @@ static inline SimdInt simdBlend(SimdInt x, SimdInt y, SimdInt mask) { 
}



const int simdLen = 8;

const int simdLen2 = 16;



static inline SimdInt simdSet(int i7, int i6, int i5, int i4,

			      int i3, int i2, int i1, int i0) {

  return _mm256_set_epi32(i7, i6, i5, i4, i3, i2, i1, i0);

}



static inline SimdInt simdSet2(short iF, short iE, short iD, short iC,

			       short iB, short iA, short i9, short i8,

			       short i7, short i6, short i5, short i4,

			       short i3, short i2, short i1, short i0) {

  return _mm256_set_epi16(iF, iE, iD, iC, iB, iA, i9, i8,

			  i7, i6, i5, i4, i3, i2, i1, i0);

}



static inline SimdInt simdSet1(char jF, char jE, char jD, char jC,

			       char jB, char jA, char j9, char j8,

			       char j7, char j6, char j5, char j4,
@@ -72,34 +65,22 @@ static inline SimdInt simdFill(int x) { 
  return _mm256_set1_epi32(x);

}



static inline SimdInt simdFill2(short x) {

  return _mm256_set1_epi16(x);

}



static inline SimdInt simdFill1(char x) {

  return _mm256_set1_epi8(x);

}



static inline SimdInt simdEq1(SimdInt x, SimdInt y) {

  return _mm256_cmpeq_epi8(x, y);

}



static inline SimdInt simdGt(SimdInt x, SimdInt y) {

  return _mm256_cmpgt_epi32(x, y);

}



static inline SimdInt simdGt2(SimdInt x, SimdInt y) {

  return _mm256_cmpgt_epi16(x, y);

static inline SimdInt simdGe1(SimdInt x, SimdInt y) {

  return _mm256_cmpeq_epi8(_mm256_min_epu8(x, y), y);

}



static inline SimdInt simdAdd(SimdInt x, SimdInt y) {

  return _mm256_add_epi32(x, y);

}



static inline SimdInt simdAdd2(SimdInt x, SimdInt y) {

  return _mm256_add_epi16(x, y);

}



static inline SimdInt simdAdd1(SimdInt x, SimdInt y) {

  return _mm256_add_epi8(x, y);

}
@@ -112,10 +93,6 @@ static inline SimdInt simdSub(SimdInt x, SimdInt y) { 
  return _mm256_sub_epi32(x, y);

}



static inline SimdInt simdSub2(SimdInt x, SimdInt y) {

  return _mm256_sub_epi16(x, y);

}



static inline SimdInt simdSub1(SimdInt x, SimdInt y) {

  return _mm256_sub_epi8(x, y);

}
@@ -128,10 +105,6 @@ static inline SimdInt simdMax(SimdInt x, SimdInt y) { 
  return _mm256_max_epi32(x, y);

}



static inline SimdInt simdMax2(SimdInt x, SimdInt y) {

  return _mm256_max_epi16(x, y);

}



static inline SimdInt simdMin1(SimdInt x, SimdInt y) {

  return _mm256_min_epu8(x, y);

}
@@ -144,14 +117,6 @@ static inline int simdHorizontalMax(SimdInt x) { 
  return _mm_cvtsi128_si32(z);

}



static inline int simdHorizontalMax2(SimdInt x) {

  __m128i z = _mm256_castsi256_si128(x);

  z = _mm_max_epi16(z, _mm256_extracti128_si256(x, 1));

  z = _mm_sub_epi16(_mm_set1_epi16(32767), z);

  z = _mm_minpos_epu16(z);

  return 32767 - _mm_extract_epi16(z, 0);

}



static inline int simdHorizontalMin1(SimdInt x) {

  __m128i z = _mm256_castsi256_si128(x);

  z = _mm_min_epu8(z, _mm256_extracti128_si256(x, 1));
@@ -195,17 +160,11 @@ static inline SimdInt simdBlend(SimdInt x, SimdInt y, SimdInt mask) { 
}



const int simdLen = 4;

const int simdLen2 = 8;



static inline SimdInt simdSet(int i3, int i2, int i1, int i0) {

  return _mm_set_epi32(i3, i2, i1, i0);

}



static inline SimdInt simdSet2(short i7, short i6, short i5, short i4,

			       short i3, short i2, short i1, short i0) {

  return _mm_set_epi16(i7, i6, i5, i4, i3, i2, i1, i0);

}



static inline SimdInt simdSet1(char iF, char iE, char iD, char iC,

			       char iB, char iA, char i9, char i8,

			       char i7, char i6, char i5, char i4,
@@ -218,34 +177,22 @@ static inline SimdInt simdFill(int x) { 
  return _mm_set1_epi32(x);

}



static inline SimdInt simdFill2(short x) {

  return _mm_set1_epi16(x);

}



static inline SimdInt simdFill1(char x) {

  return _mm_set1_epi8(x);

}



static inline SimdInt simdEq1(SimdInt x, SimdInt y) {

  return _mm_cmpeq_epi8(x, y);

}



static inline SimdInt simdGt(SimdInt x, SimdInt y) {

  return _mm_cmpgt_epi32(x, y);

}



static inline SimdInt simdGt2(SimdInt x, SimdInt y) {

  return _mm_cmpgt_epi16(x, y);

static inline SimdInt simdGe1(SimdInt x, SimdInt y) {

  return _mm_cmpeq_epi8(_mm_min_epu8(x, y), y);

}



static inline SimdInt simdAdd(SimdInt x, SimdInt y) {

  return _mm_add_epi32(x, y);

}



static inline SimdInt simdAdd2(SimdInt x, SimdInt y) {

  return _mm_add_epi16(x, y);

}



static inline SimdInt simdAdd1(SimdInt x, SimdInt y) {

  return _mm_add_epi8(x, y);

}
@@ -258,10 +205,6 @@ static inline SimdInt simdSub(SimdInt x, SimdInt y) { 
  return _mm_sub_epi32(x, y);

}



static inline SimdInt simdSub2(SimdInt x, SimdInt y) {

  return _mm_sub_epi16(x, y);

}



static inline SimdInt simdSub1(SimdInt x, SimdInt y) {

  return _mm_sub_epi8(x, y);

}
@@ -274,10 +217,6 @@ static inline SimdInt simdMax(SimdInt x, SimdInt y) { 
  return _mm_max_epi32(x, y);  // SSE4.1

}



static inline SimdInt simdMax2(SimdInt x, SimdInt y) {

  return _mm_max_epi16(x, y);

}



static inline SimdInt simdMin1(SimdInt x, SimdInt y) {

  return _mm_min_epu8(x, y);

}
@@ -288,12 +227,6 @@ static inline int simdHorizontalMax(SimdInt x) { 
  return _mm_cvtsi128_si32(x);

}



static inline int simdHorizontalMax2(SimdInt x) {

  x = simdSub2(simdFill2(32767), x);

  x = _mm_minpos_epu16(x);  // SSE4.1

  return 32767 - _mm_extract_epi16(x, 0);

}



static inline int simdHorizontalMin1(SimdInt x) {

  x = _mm_min_epu8(x, _mm_srli_epi16(x, 8));

  x = _mm_minpos_epu16(x);  // SSE4.1
@@ -301,13 +234,15 @@ static inline int simdHorizontalMin1(SimdInt x) { 
}



static inline SimdInt simdChoose1(SimdInt items, SimdInt choices) {

  return _mm_shuffle_epi8(items, choices);

  return _mm_shuffle_epi8(items, choices);  // SSSE3

}



#else



typedef int SimdInt;

const int simdBytes = 1;

const int simdLen = 1;

static inline int simdZero() { return 0; }

static inline int simdSet(int x) { return x; }

static inline int simdFill(int x) { return x; }

static inline int simdLoad(const int *p) { return *p; }