From: Nathan Moinvaziri <nathan@nathanm.com>
Date: Fri, 2 Jan 2026 22:47:53 +0000 (-0800)
Subject: Use mm_blend_epi16 in crc32_(v)pclmulqdq final reduction
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=8eaa34a12c5511fe3c33cb0ea2db48409fa8e738;p=thirdparty%2Fzlib-ng.git

Use mm_blend_epi16 in crc32_(v)pclmulqdq final reduction

This is the preferred operation mentioned in
https://www.corsix.org/content/alternative-exposition-crc32_4k_pclmulqdq
---

diff --git a/arch/x86/crc32_pclmulqdq_tpl.h b/arch/x86/crc32_pclmulqdq_tpl.h
index 5271a6846..d18839bbb 100644
--- a/arch/x86/crc32_pclmulqdq_tpl.h
+++ b/arch/x86/crc32_pclmulqdq_tpl.h
@@ -210,7 +210,7 @@ static inline uint32_t fold_final(__m128i *xmm_crc0, __m128i *xmm_crc1, __m128i
     __m128i x_tmp0 = _mm_clmulepi64_si128(*xmm_crc3, barrett_k, 0x00);
     __m128i x_tmp1 = _mm_clmulepi64_si128(x_tmp0, barrett_k, 0x10);
 
-    x_tmp1 = _mm_and_si128(x_tmp1, _mm_setr_epi32(0, 0, ~0, 0));
+    x_tmp1 = _mm_blend_epi16(x_tmp1, _mm_setzero_si128(), 0xcf);
     x_tmp0 = _mm_xor_si128(x_tmp1, *xmm_crc3);
 
     __m128i x_res_a = _mm_clmulepi64_si128(x_tmp0, barrett_k, 0x01);