Skip to content

Commit 54fd976

Browse files
committed
Add Zen5 tunings
1 parent dc89435 commit 54fd976

File tree

2 files changed

+5
-5
lines changed

2 files changed

+5
-5
lines changed

src/encoder_avx_base.h

+4-4
Original file line numberDiff line numberDiff line change
@@ -217,12 +217,12 @@ HEDLEY_ALWAYS_INLINE void do_encode_avx2(int line_size, int* colOffset, const ui
217217
// duplicate halves
218218
data1A = _mm256_inserti128_si256(dataA, _mm256_castsi256_si128(dataA), 1);
219219
data1B = _mm256_inserti128_si256(dataB, _mm256_castsi256_si128(dataB), 1);
220-
#if defined(__tune_znver2__) || defined(__tune_znver3__) || defined(__tune_znver4__)
221-
data2A = _mm256_permute2x128_si256(dataA, dataA, 0x11);
222-
data2B = _mm256_permute2x128_si256(dataB, dataB, 0x11);
223-
#else
220+
#if defined(__tune_znver1__) || defined(__tune_bdver4__)
224221
data2A = _mm256_permute4x64_epi64(dataA, 0xee);
225222
data2B = _mm256_permute4x64_epi64(dataB, 0xee);
223+
#else
224+
data2A = _mm256_permute2x128_si256(dataA, dataA, 0x11);
225+
data2B = _mm256_permute2x128_si256(dataB, dataB, 0x11);
226226
#endif
227227

228228
shuf1A = _mm256_load_si256(lookupsAVX2->shufExpand + m1);

src/encoder_sse_base.h

+1-1
Original file line numberDiff line numberDiff line change
@@ -351,7 +351,7 @@ HEDLEY_ALWAYS_INLINE void do_encode_sse(int line_size, int* colOffset, const uin
351351
#if defined(__POPCNT__) && !defined(__tune_btver1__)
352352
if(use_isa & ISA_FEATURE_POPCNT) {
353353
shuf2Len = popcnt32(maskA) + 16;
354-
# if defined(__tune_znver4__) || defined(__tune_znver3__) || defined(__tune_znver2__) || defined(__tune_znver1__) || defined(__tune_btver2__)
354+
# if defined(__tune_znver6__) || defined(__tune_znver5__) || defined(__tune_znver4__) || defined(__tune_znver3__) || defined(__tune_znver2__) || defined(__tune_znver1__) || defined(__tune_btver2__)
355355
shuf1Len = popcnt32(m1) + 8;
356356
shuf3Len = popcnt32(m3) + shuf2Len + 8;
357357
# else

0 commit comments

Comments
 (0)