From 972e73490640298e8cdcb5060ef25bc203a93737 Mon Sep 17 00:00:00 2001
From: Eric Biggers <ebiggers@google.com>
Date: Tue, 22 Oct 2024 22:43:32 -0700
Subject: [PATCH] lib/x86: Specify evex512 and no-evex512 when appropriate

libdeflate failed to build with -mno-evex512 with clang 18+ or gcc 14+
because that flag causes the evex512 feature (which is new in those
compiler versions) to have to be explicitly added to functions that use
AVX512 with 512-bit vectors.  Add this feature where appropriate.

For the same compiler versions, also add no-evex512 to the functions
that use AVX512 + 256-bit vectors so that they are ready for AVX10/256.

Resolves https://github.com/ebiggers/libdeflate/issues/394
---
 lib/x86/adler32_impl.h | 4 ++--
 lib/x86/cpu_features.h | 8 ++++++++
 lib/x86/crc32_impl.h   | 4 ++--
 3 files changed, 12 insertions(+), 4 deletions(-)

diff --git a/lib/x86/adler32_impl.h b/lib/x86/adler32_impl.h
index 458a9e09..100a2301 100644
--- a/lib/x86/adler32_impl.h
+++ b/lib/x86/adler32_impl.h
@@ -82,7 +82,7 @@
  */
 #  define adler32_x86_avx512_vl256_vnni	adler32_x86_avx512_vl256_vnni
 #  define SUFFIX				   _avx512_vl256_vnni
-#  define ATTRIBUTES		_target_attribute("avx512bw,avx512vl,avx512vnni")
+#  define ATTRIBUTES		_target_attribute("avx512bw,avx512vl,avx512vnni" NO_EVEX512)
 #  define VL			32
 #  define USE_VNNI		1
 #  define USE_AVX512		1
@@ -95,7 +95,7 @@
  */
 #  define adler32_x86_avx512_vl512_vnni	adler32_x86_avx512_vl512_vnni
 #  define SUFFIX				   _avx512_vl512_vnni
-#  define ATTRIBUTES		_target_attribute("avx512bw,avx512vnni")
+#  define ATTRIBUTES		_target_attribute("avx512bw,avx512vnni" EVEX512)
 #  define VL			64
 #  define USE_VNNI		1
 #  define USE_AVX512		1
diff --git a/lib/x86/cpu_features.h b/lib/x86/cpu_features.h
index 8dda21fd..e70bc80f 100644
--- a/lib/x86/cpu_features.h
+++ b/lib/x86/cpu_features.h
@@ -164,6 +164,14 @@ static inline u32 get_x86_cpu_features(void) { return 0; }
 #  define HAVE_AVXVNNI(features)	((features) & X86_CPU_FEATURE_AVXVNNI)
 #endif
 
+#if GCC_PREREQ(14, 0) || CLANG_PREREQ(18, 0, 18000000)
+#  define EVEX512	",evex512"	/* needed to override potential -mno-evex512 */
+#  define NO_EVEX512	",no-evex512"	/* needed for AVX10/256 compatibility */
+#else
+#  define EVEX512	""
+#  define NO_EVEX512	""
+#endif
+
 #endif /* ARCH_X86_32 || ARCH_X86_64 */
 
 #endif /* LIB_X86_CPU_FEATURES_H */
diff --git a/lib/x86/crc32_impl.h b/lib/x86/crc32_impl.h
index 54996d9c..1747aa85 100644
--- a/lib/x86/crc32_impl.h
+++ b/lib/x86/crc32_impl.h
@@ -106,7 +106,7 @@ static const u8 MAYBE_UNUSED shift_tab[48] = {
  */
 #  define crc32_x86_vpclmulqdq_avx512_vl256  crc32_x86_vpclmulqdq_avx512_vl256
 #  define SUFFIX				      _vpclmulqdq_avx512_vl256
-#  define ATTRIBUTES		_target_attribute("vpclmulqdq,pclmul,avx512bw,avx512vl")
+#  define ATTRIBUTES		_target_attribute("vpclmulqdq,pclmul,avx512bw,avx512vl" NO_EVEX512)
 #  define VL			32
 #  define USE_SSE4_1		1
 #  define USE_AVX512		1
@@ -119,7 +119,7 @@ static const u8 MAYBE_UNUSED shift_tab[48] = {
  */
 #  define crc32_x86_vpclmulqdq_avx512_vl512  crc32_x86_vpclmulqdq_avx512_vl512
 #  define SUFFIX				      _vpclmulqdq_avx512_vl512
-#  define ATTRIBUTES		_target_attribute("vpclmulqdq,pclmul,avx512bw,avx512vl")
+#  define ATTRIBUTES		_target_attribute("vpclmulqdq,pclmul,avx512bw,avx512vl" EVEX512)
 #  define VL			64
 #  define USE_SSE4_1		1
 #  define USE_AVX512		1