From 0b839b9d7e3a2e3d8ac23c0ac5679edf89d670d2 Mon Sep 17 00:00:00 2001
From: Andrew Noyes <andrew@weaselab.dev>
Date: Sat, 29 Jun 2024 22:20:50 -0700
Subject: [PATCH] Fixes for symbol multi-versioning with avx512f

---
 ConflictSet.cpp    | 30 +++++++++++++++++++++++-------
 symbol-imports.txt |  2 ++
 2 files changed, 25 insertions(+), 7 deletions(-)

diff --git a/ConflictSet.cpp b/ConflictSet.cpp
index 5e9625f..d2bdd1f 100644
--- a/ConflictSet.cpp
+++ b/ConflictSet.cpp
@@ -38,6 +38,12 @@ limitations under the License.
 #include <arm_neon.h>
 #endif
 
+#if defined(__has_feature)
+#  if __has_feature(thread_sanitizer)
+#define __SANITIZE_THREAD__
+#  endif
+#endif
+
 #include <memcheck.h>
 
 using namespace weaselab;
@@ -1767,11 +1773,17 @@ downLeftSpine:
 }
 
 #ifdef HAS_AVX
-__attribute__((target("default"))) uint32_t compare16_32bit(const uint32_t *vs,
-                                                            uint32_t rv) {
+#ifndef __SANITIZE_THREAD__
+__attribute__((target("default")))
+#endif
+uint32_t compare16_32bit(const InternalVersionT *vs,
+                                                            InternalVersionT rv) {
+  uint32_t compared = 0;
   __m128i w[4];
   memcpy(w, vs, sizeof(w));
-  const auto rvVec = _mm_set1_epi32(rv);
+  uint32_t r;
+  memcpy(&r, &rv, sizeof(r));
+  const auto rvVec = _mm_set1_epi32(r);
   const auto zero = _mm_setzero_si128();
   for (int i = 0; i < 4; ++i) {
     compared |=
@@ -1781,14 +1793,18 @@ __attribute__((target("default"))) uint32_t compare16_32bit(const uint32_t *vs,
   return compared;
 }
 
-__attribute__((target("avx512f"))) uint32_t compare16_32bit(const uint32_t *vs,
-                                                            uint32_t rv) {
-  __mm512i w;
+#ifndef __SANITIZE_THREAD__
+__attribute__((target("avx512f"))) uint32_t compare16_32bit(const InternalVersionT *vs,
+                                                            InternalVersionT rv) {
+  __m512i w;
   memcpy(&w, vs, sizeof(w));
-  return _mm512_cmpgt_epi32_mask(_mm512_sub_epi32(w, _mm512_set1_epi32(rv)),
+  uint32_t r;
+  memcpy(&r, &rv, sizeof(r));
+  return _mm512_cmpgt_epi32_mask(_mm512_sub_epi32(w, _mm512_set1_epi32(r)),
                                  _mm512_setzero());
 }
 #endif
+#endif
 
 // Returns true if v[i] <= readVersion for all i such that begin <= is[i] < end
 // Preconditions: begin <= end, end - begin < 256
diff --git a/symbol-imports.txt b/symbol-imports.txt
index bbed523..0a97ef0 100644
--- a/symbol-imports.txt
+++ b/symbol-imports.txt
@@ -1,4 +1,6 @@
 _GLOBAL_OFFSET_TABLE_
+__cpu_indicator_init
+__cpu_model
 __stack_chk_fail@GLIBC_2.4
 __tls_get_addr@GLIBC_2.3
 abort@GLIBC_2.2.5