Fix combine2x128

diff --git a/src/nfa/shufti.c b/src/nfa/shufti.c
index 578904784997a7ca53e453bc32b85fe604d0ca96..2e63be9fdca846c4f1a5ac0b559d088e3197eb48 100644 (file)
--- a/src/nfa/shufti.c
+++ b/src/nfa/shufti.c
@@ -308,7 +308,7 @@ const u8 *fwdBlockShort(m256 mask, m128 chars, const u8 *buf,
                         const m256 low4bits) {
     // do the hi and lo shuffles in the one avx register
     m256 c = set2x128(chars);
-    c = _mm256_srlv_epi64(c, _mm256_set_epi64x(0, 0, 4, 4));
+    c = _mm256_srlv_epi64(c, _mm256_set_epi64x(4, 4, 0, 0));
     c = and256(c, low4bits);
     m256 c_shuf = vpshufb(mask, c);
     m128 t = and128(movdq_hi(c_shuf), cast256to128(c_shuf));
@@ -440,7 +440,7 @@ const u8 *revBlockShort(m256 mask, m128 chars, const u8 *buf,
                         const m256 low4bits) {
     // do the hi and lo shuffles in the one avx register
     m256 c = set2x128(chars);
-    c = _mm256_srlv_epi64(c, _mm256_set_epi64x(0, 0, 4, 4));
+    c = _mm256_srlv_epi64(c, _mm256_set_epi64x(4, 4, 0, 0));
     c = and256(c, low4bits);
     m256 c_shuf = vpshufb(mask, c);
     m128 t = and128(movdq_hi(c_shuf), cast256to128(c_shuf));
@@ -565,7 +565,7 @@ const u8 *fwdBlockShort2(m256 mask1, m256 mask2, m128 chars, const u8 *buf,
                          const m256 low4bits) {
     // do the hi and lo shuffles in the one avx register
     m256 c = set2x128(chars);
-    c = _mm256_srlv_epi64(c, _mm256_set_epi64x(0, 0, 4, 4));
+    c = _mm256_srlv_epi64(c, _mm256_set_epi64x(4, 4, 0, 0));
     c = and256(c, low4bits);
     m256 c_shuf1 = vpshufb(mask1, c);
     m256 c_shuf2 = rshift128_m256(vpshufb(mask2, c), 1);

diff --git a/src/rose/rose_build_bytecode.cpp b/src/rose/rose_build_bytecode.cpp
index 43df79628242bff72db7c09dd338668e5bbfa521..f074973d75e4a1eb841aa6bae5d3966f9bce978a 100644 (file)
--- a/src/rose/rose_build_bytecode.cpp
+++ b/src/rose/rose_build_bytecode.cpp
@@ -3054,8 +3054,8 @@ bool makeRoleShufti(const vector<LookEntry> &look,
             neg_mask &= 0xffff;
             array<u8, 32> nib_mask;
             array<u8, 16> bucket_select_mask_16;
-            copy(hi_mask.begin(), hi_mask.begin() + 16, nib_mask.begin());
-            copy(lo_mask.begin(), lo_mask.begin() + 16, nib_mask.begin() + 16);
+            copy(lo_mask.begin(), lo_mask.begin() + 16, nib_mask.begin());
+            copy(hi_mask.begin(), hi_mask.begin() + 16, nib_mask.begin() + 16);
             copy(bucket_select_lo.begin(), bucket_select_lo.begin() + 16,
                  bucket_select_mask_16.begin());
             auto ri = make_unique<RoseInstrCheckShufti16x8>

diff --git a/src/util/simd_utils.h b/src/util/simd_utils.h
index afa8c7f8e55b2ac1e94ac5bb0a1bf0aab7faa863..35e1a390539ea10f3c5fb9cecf832787b1b88b2c 100644 (file)
--- a/src/util/simd_utils.h
+++ b/src/util/simd_utils.h
@@ -658,8 +658,8 @@ m128 movdq_lo(m256 x) {
 }
 
 static really_inline
-m256 combine2x128(m128 a, m128 b) {
-    m256 rv = {a, b};
+m256 combine2x128(m128 hi, m128 lo) {
+    m256 rv = {lo, hi};
     return rv;
 }
 
@@ -712,7 +712,7 @@ m256 combine2x128(m128 hi, m128 lo) {
 #if defined(_mm256_set_m128i)
     return _mm256_set_m128i(hi, lo);
 #else
-    return insert128to256(cast128to256(hi), lo, 1);
+    return insert128to256(cast128to256(lo), hi, 1);
 #endif
 }
 #endif //AVX2