AVX512 VPOPCNT: use the compiler's intrinsic

WojciechMula · Nov 22, 2017 · 92979cb · 92979cb
1 parent a51250e
commit 92979cb
Show file tree

Hide file tree

Showing 3 changed files with 1 addition and 18 deletions.
diff --git a/Makefile b/Makefile
@@ -24,8 +24,7 @@ FLAGS_SSE=$(FLAGS_INTEL) -mssse3 -DHAVE_SSE_INSTRUCTIONS
 FLAGS_AVX=$(FLAGS_INTEL) -mavx -DHAVE_AVX_INSTRUCTIONS
 FLAGS_AVX2=$(FLAGS_INTEL) -mavx2 -DHAVE_AVX2_INSTRUCTIONS
 FLAGS_AVX512BW=$(FLAGS_INTEL) -mavx512bw -DHAVE_AVX512BW_INSTRUCTIONS
-# Use the flag -mavx512bw bacuse my current GCC installation doesn't support AVX512 VPOPCNT yet.
-FLAGS_AVX512VPOPCNT=$(FLAGS_INTEL) -mavx512bw -DHAVE_AVX512VPOPCNT_INSTRUCTIONS
+FLAGS_AVX512VPOPCNT=$(FLAGS_INTEL) -mavx512bw -mavx512vpopcntdq -DHAVE_AVX512VPOPCNT_INSTRUCTIONS
 
 DEPS=popcnt-*.cpp function_registry.cpp sse_operators.cpp config.h
 ALL=speed_$(COMPILER) verify_$(COMPILER)

diff --git a/config.h b/config.h
@@ -5,7 +5,6 @@
 #   if !defined(HAVE_AVX512BW_INSTRUCTIONS)
 #       define HAVE_AVX512BW_INSTRUCTIONS
 #   endif
-#   define AVX512POPCNT_SW_EMULATOR
 #endif
 
 #if defined(HAVE_AVX512BW_INSTRUCTIONS)

diff --git a/popcnt-avx512-vpopcnt.cpp b/popcnt-avx512-vpopcnt.cpp
@@ -1,18 +1,3 @@
-#if defined(AVX512POPCNT_SW_EMULATOR) // waiting for SDE updates
-__m512i _mm512_popcnt_epi64(const __m512i& val) {
-
-    uint64_t tmp[8] __attribute__((aligned(64)));
-
-    _mm512_store_si512((__m512i*)tmp, val);
-
-    for (int i=0; i < 8; i++) {
-        tmp[i] = _mm_popcnt_u64(tmp[i]);
-    }
-
-    return _mm512_load_si512((__m512i*)tmp);
-}
-#endif
-
 uint64_t avx512_vpopcnt(const uint8_t* data, const size_t size) {
 
     const size_t chunks = size / 64;