From e0a9e4e145dda2886c842b418571f0217cb2fd60 Mon Sep 17 00:00:00 2001
From: Ash Vardanian <1983160+ashvardanian@users.noreply.github.com>
Date: Sun, 1 Dec 2024 08:34:01 +0000
Subject: [PATCH] Fix: Missing `_mm_cvtsi128_si64x` in Clang

---
 include/stringzilla/stringzilla.h | 8 ++++----
 scripts/test.cpp                  | 2 --
 2 files changed, 4 insertions(+), 6 deletions(-)

diff --git a/include/stringzilla/stringzilla.h b/include/stringzilla/stringzilla.h
index fc5ab517..054f02e8 100644
--- a/include/stringzilla/stringzilla.h
+++ b/include/stringzilla/stringzilla.h
@@ -4022,7 +4022,7 @@ SZ_PUBLIC sz_u64_t sz_checksum_avx2(sz_cptr_t text, sz_size_t length) {
         __m128i low_xmm = _mm256_castsi256_si128(sums_vec.ymm);
         __m128i high_xmm = _mm256_extracti128_si256(sums_vec.ymm, 1);
         __m128i sums_xmm = _mm_add_epi64(low_xmm, high_xmm);
-        sz_u64_t low = (sz_u64_t)_mm_cvtsi128_si64x(sums_xmm);
+        sz_u64_t low = (sz_u64_t)_mm_cvtsi128_si64(sums_xmm);
         sz_u64_t high = (sz_u64_t)_mm_extract_epi64(sums_xmm, 1);
         sz_u64_t result = low + high;
         if (length) result += sz_checksum_serial(text, length);
@@ -4073,7 +4073,7 @@ SZ_PUBLIC sz_u64_t sz_checksum_avx2(sz_cptr_t text, sz_size_t length) {
         __m128i low_xmm = _mm256_castsi256_si128(sums_vec.ymm);
         __m128i high_xmm = _mm256_extracti128_si256(sums_vec.ymm, 1);
         __m128i sums_xmm = _mm_add_epi64(low_xmm, high_xmm);
-        sz_u64_t low = (sz_u64_t)_mm_cvtsi128_si64x(sums_xmm);
+        sz_u64_t low = (sz_u64_t)_mm_cvtsi128_si64(sums_xmm);
         sz_u64_t high = (sz_u64_t)_mm_extract_epi64(sums_xmm, 1);
         result += low + high;
         return result;
@@ -5306,7 +5306,7 @@ SZ_PUBLIC sz_u64_t sz_checksum_avx512(sz_cptr_t text, sz_size_t length) {
         __mmask16 mask = _sz_u16_mask_until(length);
         text_vec.xmms[0] = _mm_maskz_loadu_epi8(mask, text);
         sums_vec.xmms[0] = _mm_sad_epu8(text_vec.xmms[0], _mm_setzero_si128());
-        sz_u64_t low = (sz_u64_t)_mm_cvtsi128_si64x(sums_vec.xmms[0]);
+        sz_u64_t low = (sz_u64_t)_mm_cvtsi128_si64(sums_vec.xmms[0]);
         sz_u64_t high = (sz_u64_t)_mm_extract_epi64(sums_vec.xmms[0], 1);
         return low + high;
     }
@@ -5318,7 +5318,7 @@ SZ_PUBLIC sz_u64_t sz_checksum_avx512(sz_cptr_t text, sz_size_t length) {
         __m128i low_xmm = _mm256_castsi256_si128(sums_vec.ymms[0]);
         __m128i high_xmm = _mm256_extracti128_si256(sums_vec.ymms[0], 1);
         __m128i sums_xmm = _mm_add_epi64(low_xmm, high_xmm);
-        sz_u64_t low = (sz_u64_t)_mm_cvtsi128_si64x(sums_xmm);
+        sz_u64_t low = (sz_u64_t)_mm_cvtsi128_si64(sums_xmm);
         sz_u64_t high = (sz_u64_t)_mm_extract_epi64(sums_xmm, 1);
         return low + high;
     }
diff --git a/scripts/test.cpp b/scripts/test.cpp
index 104ca032..47ef46d2 100644
--- a/scripts/test.cpp
+++ b/scripts/test.cpp
@@ -1546,12 +1546,10 @@ int main(int argc, char const **argv) {
     std::printf("- Uses NEON: %s \n", SZ_USE_ARM_NEON ? "yes" : "no");
     std::printf("- Uses SVE: %s \n", SZ_USE_ARM_SVE ? "yes" : "no");
 
-#if 0
     // Basic utilities
     test_arithmetical_utilities();
     test_memory_utilities();
     test_replacements();
-#endif
 
 // Compatibility with STL
 #if SZ_DETECT_CPP_17 && __cpp_lib_string_view