Skip to content

Commits on Source 4

Show whitespace changes
Inline Side-by-side
ChangeLog.txt
View file @ 852e4d6a
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>
......
debian/changelog
View file @ 852e4d6a
last-align (1045-1) UNRELEASED; urgency=medium
* Team upload.
* New upstream version
* Updated simde patches
-- Michael R. Crusoe <michael.crusoe@gmail.com> Mon, 30 Dec 2019 18:44:40 +0100
last-align (1044-2) unstable; urgency=medium
* Team upload.
......
debian/patches/simde
View file @ 852e4d6a
--- last-align.orig/src/mcf_simd.hh
+++ last-align/src/mcf_simd.hh
@@ -4,38 +4,36 @@
@@ -4,47 +4,43 @@
#ifndef MCF_SIMD_HH
#define MCF_SIMD_HH
-#if defined __SSE4_1__
-#include <immintrin.h>
-#endif
+#include "../debian/include/simde/x86/avx2.h"
namespace mcf {
......@@ -47,7 +49,6 @@
}
const int simdLen = 8;
@@ -43,14 +41,14 @@
static inline SimdInt simdSet(int i7, int i6, int i5, int i4,
int i3, int i2, int i1, int i0) {
......@@ -55,16 +56,8 @@
+ return simde_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,
+ return simde_mm256_set_epi16(iF, iE, iD, iC, iB, iA, i9, i8,
i7, i6, i5, i4, i3, i2, i1, i0);
}
@@ -62,265 +60,108 @@
static inline SimdInt simdSet1(char jF, char jE, char jD, char jC,
@@ -55,207 +51,80 @@
char iB, char iA, char i9, char i8,
char i7, char i6, char i5, char i4,
char i3, char i2, char i1, char i0) {
......@@ -77,129 +70,69 @@
static inline SimdInt simdFill(int x) {
- return _mm256_set1_epi32(x);
+ return simde_mm256_set1_epi32(x);
}
static inline SimdInt simdFill2(short x) {
- return _mm256_set1_epi16(x);
+ return simde_mm256_set1_epi16(x);
}
static inline SimdInt simdFill1(char x) {
-}
-
-static inline SimdInt simdFill1(char x) {
- return _mm256_set1_epi8(x);
+ return simde_mm256_set1_epi8(x);
}
static inline SimdInt simdEq1(SimdInt x, SimdInt y) {
- return _mm256_cmpeq_epi8(x, y);
+ return simde_mm256_cmpeq_epi8(x, y);
}
static inline SimdInt simdGt(SimdInt x, SimdInt y) {
-}
-
-static inline SimdInt simdGt(SimdInt x, SimdInt y) {
- return _mm256_cmpgt_epi32(x, y);
+ return simde_mm256_cmpgt_epi32(x, y);
}
static inline SimdInt simdGt2(SimdInt x, SimdInt y) {
- return _mm256_cmpgt_epi16(x, y);
+ return simde_mm256_cmpgt_epi16(x, y);
}
static inline SimdInt simdAdd(SimdInt x, SimdInt 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);
+ return simde_mm256_add_epi32(x, y);
}
static inline SimdInt simdAdd2(SimdInt x, SimdInt y) {
- return _mm256_add_epi16(x, y);
+ return simde_mm256_add_epi16(x, y);
}
static inline SimdInt simdAdd1(SimdInt x, SimdInt y) {
-}
-
-static inline SimdInt simdAdd1(SimdInt x, SimdInt y) {
- return _mm256_add_epi8(x, y);
+ return simde_mm256_add_epi8(x, y);
}
static inline SimdInt simdAdds1(SimdInt x, SimdInt y) {
-}
-
-static inline SimdInt simdAdds1(SimdInt x, SimdInt y) {
- return _mm256_adds_epu8(x, y);
+ return simde_mm256_adds_epu8(x, y);
}
static inline SimdInt simdSub(SimdInt x, SimdInt y) {
-}
-
-static inline SimdInt simdSub(SimdInt x, SimdInt y) {
- return _mm256_sub_epi32(x, y);
+ return simde_mm256_sub_epi32(x, y);
}
static inline SimdInt simdSub2(SimdInt x, SimdInt y) {
- return _mm256_sub_epi16(x, y);
+ return simde_mm256_sub_epi16(x, y);
}
static inline SimdInt simdSub1(SimdInt x, SimdInt y) {
-}
-
-static inline SimdInt simdSub1(SimdInt x, SimdInt y) {
- return _mm256_sub_epi8(x, y);
+ return simde_mm256_sub_epi8(x, y);
}
static inline SimdInt simdLeft(SimdInt x, int bits) {
-}
-
-static inline SimdInt simdLeft(SimdInt x, int bits) {
- return _mm256_slli_epi32(x, bits);
+ return simde_mm256_slli_epi32(x, bits);
}
static inline SimdInt simdMax(SimdInt x, SimdInt y) {
-}
-
-static inline SimdInt simdMax(SimdInt x, SimdInt y) {
- return _mm256_max_epi32(x, y);
+ return simde_mm256_max_epi32(x, y);
}
static inline SimdInt simdMax2(SimdInt x, SimdInt y) {
- return _mm256_max_epi16(x, y);
+ return simde_mm256_max_epi16(x, y);
}
static inline SimdInt simdMin1(SimdInt x, SimdInt y) {
-}
-
-static inline SimdInt simdMin1(SimdInt x, SimdInt y) {
- return _mm256_min_epu8(x, y);
+ return simde_mm256_min_epu8(x, y);
}
static inline int simdHorizontalMax(SimdInt x) {
-}
-
-static inline int simdHorizontalMax(SimdInt x) {
- __m128i z = _mm256_castsi256_si128(x);
- z = _mm_max_epi32(z, _mm256_extracti128_si256(x, 1));
- z = _mm_max_epi32(z, _mm_shuffle_epi32(z, 0x4E));
- z = _mm_max_epi32(z, _mm_shuffle_epi32(z, 0xB1));
- return _mm_cvtsi128_si32(z);
+ simde__m128i z = simde_mm256_castsi256_si128(x);
+ z = simde_mm_max_epi32(z, simde_mm256_extracti128_si256(x, 1));
+ z = simde_mm_max_epi32(z, simde_mm_shuffle_epi32(z, 0x4E));
+ z = simde_mm_max_epi32(z, simde_mm_shuffle_epi32(z, 0xB1));
+ return simde_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);
+ simde__m128i z = simde_mm256_castsi256_si128(x);
+ z = simde_mm_max_epi16(z, simde_mm256_extracti128_si256(x, 1));
+ z = simde_mm_sub_epi16(simde_mm_set1_epi16(32767), z);
+ z = simde_mm_minpos_epu16(z);
+ return 32767 - simde_mm_extract_epi16(z, 0);
}
static inline int simdHorizontalMin1(SimdInt x) {
-}
-
-static inline int simdHorizontalMin1(SimdInt x) {
- __m128i z = _mm256_castsi256_si128(x);
- z = _mm_min_epu8(z, _mm256_extracti128_si256(x, 1));
- z = _mm_min_epu8(z, _mm_srli_epi16(z, 8));
- z = _mm_minpos_epu16(z);
- return _mm_extract_epi16(z, 0);
+ simde__m128i z = simde_mm256_castsi256_si128(x);
+ z = simde_mm_min_epu8(z, simde_mm256_extracti128_si256(x, 1));
+ z = simde_mm_min_epu8(z, simde_mm_srli_epi16(z, 8));
+ z = simde_mm_minpos_epu16(z);
+ return simde_mm_extract_epi16(z, 0);
}
static inline SimdInt simdChoose1(SimdInt items, SimdInt choices) {
-}
-
-static inline SimdInt simdChoose1(SimdInt items, SimdInt choices) {
- return _mm256_shuffle_epi8(items, choices);
-}
-
......@@ -234,16 +167,9 @@
-}
-
-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);
+ return simde_mm256_shuffle_epi8(items, choices);
}
-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,
......@@ -256,98 +182,97 @@
-
-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 simde_mm256_set1_epi32(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 simde_mm256_set1_epi8(x);
}
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 simdAdd(SimdInt x, SimdInt y) {
+ return simde_mm256_cmpgt_epi32(x, y);
}
static inline SimdInt simdGe1(SimdInt x, SimdInt y) {
- return _mm_cmpeq_epi8(_mm_min_epu8(x, y), y);
+ return simde_mm256_cmpeq_epi8(simde_mm256_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 simde_mm256_add_epi32(x, y);
}
static inline SimdInt simdAdd1(SimdInt x, SimdInt y) {
- return _mm_add_epi8(x, y);
-}
-
-static inline SimdInt simdAdds1(SimdInt x, SimdInt y) {
+ return simde_mm256_add_epi8(x, y);
}
static inline SimdInt simdAdds1(SimdInt x, SimdInt y) {
- return _mm_adds_epu8(x, y);
-}
-
-static inline SimdInt simdSub(SimdInt x, SimdInt y) {
+ return simde_mm256_adds_epu8(x, y);
}
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 simde_mm256_sub_epi32(x, y);
}
static inline SimdInt simdSub1(SimdInt x, SimdInt y) {
- return _mm_sub_epi8(x, y);
-}
-
-static inline SimdInt simdLeft(SimdInt x, int bits) {
+ return simde_mm256_sub_epi8(x, y);
}
static inline SimdInt simdLeft(SimdInt x, int bits) {
- return _mm_slli_epi32(x, bits);
-}
-
-static inline SimdInt simdMax(SimdInt x, SimdInt y) {
+ return simde_mm256_slli_epi32(x, bits);
}
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 simde_mm256_max_epi32(x, y);
}
static inline SimdInt simdMin1(SimdInt x, SimdInt y) {
- return _mm_min_epu8(x, y);
-}
-
-static inline int simdHorizontalMax(SimdInt x) {
+ return simde_mm256_min_epu8(x, y);
}
static inline int simdHorizontalMax(SimdInt x) {
- x = simdMax(x, _mm_shuffle_epi32(x, 0x4E));
- x = simdMax(x, _mm_shuffle_epi32(x, 0xB1));
- 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) {
+ simde__m128i z = simde_mm256_castsi256_si128(x);
+ z = simde_mm_max_epi32(z, simde_mm256_extracti128_si256(x, 1));
+ z = simde_mm_max_epi32(z, simde_mm_shuffle_epi32(z, 0x4E));
+ z = simde_mm_max_epi32(z, simde_mm_shuffle_epi32(z, 0xB1));
+ return simde_mm_cvtsi128_si32(z);
}
static inline int simdHorizontalMin1(SimdInt x) {
- x = _mm_min_epu8(x, _mm_srli_epi16(x, 8));
- x = _mm_minpos_epu16(x); // SSE4.1
- return _mm_extract_epi16(x, 0);
-}
-
-static inline SimdInt simdChoose1(SimdInt items, SimdInt choices) {
- return _mm_shuffle_epi8(items, choices);
-}
-
+ simde__m128i z = simde_mm256_castsi256_si128(x);
+ z = simde_mm_min_epu8(z, simde_mm256_extracti128_si256(x, 1));
+ z = simde_mm_min_epu8(z, simde_mm_srli_epi16(z, 8));
+ z = simde_mm_minpos_epu16(z);
+ return simde_mm_extract_epi16(z, 0);
}
static inline SimdInt simdChoose1(SimdInt items, SimdInt choices) {
- return _mm_shuffle_epi8(items, choices); // SSSE3
+ return simde_mm256_shuffle_epi8(items, choices);
}
-#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; }
......@@ -493,7 +418,16 @@
.SUFFIXES:
--- last-align.orig/src/GappedXdropAlignerDna.cc
+++ last-align/src/GappedXdropAlignerDna.cc
@@ -41,12 +41,10 @@
@@ -1,8 +1,6 @@
// Author: Martin C. Frith 2019
// SPDX-License-Identifier: GPL-3.0-or-later
-#if defined __SSE4_1__
-
#include "GappedXdropAligner.hh"
#include "GappedXdropAlignerInl.hh"
@@ -43,12 +41,10 @@
const SimdInt scorer4x4 =
simdSet1(
......@@ -506,7 +440,7 @@
scorer[3][3], scorer[3][2], scorer[3][1], scorer[3][0],
scorer[2][3], scorer[2][2], scorer[2][1], scorer[2][0],
scorer[1][3], scorer[1][2], scorer[1][1], scorer[1][0],
@@ -125,8 +123,6 @@
@@ -127,8 +123,6 @@
for (int i = 0; i < numCells; i += simdBytes) {
SimdInt s = simdSet1(
......@@ -515,7 +449,7 @@
scorer[s1[31]][s2[31]],
scorer[s1[30]][s2[30]],
scorer[s1[29]][s2[29]],
@@ -143,7 +139,6 @@
@@ -145,7 +139,6 @@
scorer[s1[18]][s2[18]],
scorer[s1[17]][s2[17]],
scorer[s1[16]][s2[16]],
......@@ -523,7 +457,7 @@
scorer[s1[15]][s2[15]],
scorer[s1[14]][s2[14]],
scorer[s1[13]][s2[13]],
@@ -159,7 +154,6 @@
@@ -161,7 +154,6 @@
scorer[s1[3]][s2[3]],
scorer[s1[2]][s2[2]],
scorer[s1[1]][s2[1]],
......@@ -531,3 +465,58 @@
scorer[s1[0]][s2[0]]);
SimdInt x = simdAdds1(simdLoad(x2+i), mScoreRise12);
@@ -275,5 +267,3 @@
}
}
-
-#endif
--- last-align.orig/src/Alignment.cc
+++ last-align/src/Alignment.cc
@@ -347,13 +347,11 @@
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,
@@ -373,14 +371,12 @@
while( greedyAligner.getNextChunk( end1, end2, size ) )
chunks.push_back( SegmentPair( end1 - size, end2 - size, size ) );
}
-#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));
}
-#endif
else {
while( aligner.getNextChunk( end1, end2, size,
del.openCost, del.growCost,
--- last-align.orig/src/GappedXdropAligner.hh
+++ last-align/src/GappedXdropAligner.hh
@@ -317,7 +317,6 @@
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;
@@ -362,7 +361,6 @@
while (*x2 != target) ++x2;
bestSeq1position = x2 - x2beg + seq1beg;
}
-#endif
};
}
src/Alignment.cc
View file @ 852e4d6a
......@@ -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
View file @ 852e4d6a
......@@ -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
View file @ 852e4d6a
// 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
View file @ 852e4d6a
......@@ -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; }
......