From: Nathan Moinvaziri <nathan@nathanm.com>
Date: Fri, 7 Mar 2025 03:03:11 +0000 (-0800)
Subject: Disable MSVC optimizations for AVX512 GET_CHUNK_MAG #1883
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=f136e3c86b62e65e699ee8c99863f711693a6aec;p=thirdparty%2Fzlib-ng.git

Disable MSVC optimizations for AVX512 GET_CHUNK_MAG #1883

MSVC compiler (VS 17.11.x) incorrectly optimizes the GET_CHUNK_MAG code on
older versions. Appears to be resolved in VS 17.13.2. The compiler would
optimize the code in such a way that it would cause a decompression failure.
It only happens when /Os flag is set.
---

diff --git a/arch/x86/chunkset_avx512.c b/arch/x86/chunkset_avx512.c
index 9d28d33d..db8c1eb2 100644
--- a/arch/x86/chunkset_avx512.c
+++ b/arch/x86/chunkset_avx512.c
@@ -98,6 +98,10 @@ static inline uint8_t* CHUNKCOPY(uint8_t *out, uint8_t const *from, unsigned len
     return out;
 }
 
+/* MSVC compiler decompression bug when optimizing for size */
+#if defined(_MSC_VER) && _MSC_VER < 1943
+#  pragma optimize("", off)
+#endif
 static inline chunk_t GET_CHUNK_MAG(uint8_t *buf, uint32_t *chunk_rem, uint32_t dist) {
     lut_rem_pair lut_rem = perm_idx_lut[dist - 3];
     __m256i ret_vec;
@@ -128,6 +132,9 @@ static inline chunk_t GET_CHUNK_MAG(uint8_t *buf, uint32_t *chunk_rem, uint32_t
 
     return ret_vec;
 }
+#if defined(_MSC_VER) && _MSC_VER < 1943
+#  pragma optimize("", on)
+#endif
 
 static inline void storehalfchunk(uint8_t *out, halfchunk_t *chunk) {
     _mm_storeu_si128((__m128i *)out, *chunk);