From 6ec1ca7c85a4009b77e149a798a331592b96ea42 Mon Sep 17 00:00:00 2001
From: Hennadii Stepanov <32963518+hebasto@users.noreply.github.com>
Date: Mon, 12 Feb 2024 14:26:00 +0000
Subject: [PATCH 1/2] build: Fix test for SSE4.1 intrinsics

This change uses the `_mm_blend_epi16` SSE4.1 function used in our code
and fixes false-positive cases, for example, when CXXFLAGS="-mno-sse4.1"
provided.
---
 configure.ac | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/configure.ac b/configure.ac
index 4f715158731..2c73e0920c9 100644
--- a/configure.ac
+++ b/configure.ac
@@ -504,11 +504,12 @@ TEMP_CXXFLAGS="$CXXFLAGS"
 CXXFLAGS="$SSE41_CXXFLAGS $CXXFLAGS"
 AC_MSG_CHECKING([for SSE4.1 intrinsics])
 AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[
-    #include <stdint.h>
     #include <immintrin.h>
   ]],[[
-    __m128i l = _mm_set1_epi32(0);
-    return _mm_extract_epi32(l, 3);
+    __m128i a = _mm_set1_epi32(0);
+    __m128i b = _mm_set1_epi32(1);
+    __m128i r = _mm_blend_epi16(a, b, 0xFF);
+    return _mm_extract_epi32(r, 3);
   ]])],
  [ AC_MSG_RESULT([yes]); enable_sse41=yes; AC_DEFINE([ENABLE_SSE41], [1], [Define this symbol to build code that uses SSE4.1 intrinsics]) ],
  [ AC_MSG_RESULT([no])]

From d440f13db02c82c842000abe4fe4d0c721a4ad3b Mon Sep 17 00:00:00 2001
From: Hennadii Stepanov <32963518+hebasto@users.noreply.github.com>
Date: Mon, 12 Feb 2024 14:26:16 +0000
Subject: [PATCH 2/2] crypto: Guard code with `ENABLE_SSE41` macro

The code in `sha_x86_shani.cpp` uses the `_mm_blend_epi16` function from
the SSE4.1 instruction set. However, it is possible that SHA-NI is
enabled even when SSE4.1 is disabled.

This changes avoid compilation errors in such a condition.
---
 src/Makefile.am                 | 10 +++++-----
 src/crypto/sha256.cpp           |  2 +-
 src/crypto/sha256_x86_shani.cpp |  2 +-
 3 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/src/Makefile.am b/src/Makefile.am
index b5d5c4652ab..9c1d890f5e5 100644
--- a/src/Makefile.am
+++ b/src/Makefile.am
@@ -53,15 +53,15 @@ LIBBITCOIN_CRYPTO = $(LIBBITCOIN_CRYPTO_BASE)
 if ENABLE_SSE41
 LIBBITCOIN_CRYPTO_SSE41 = crypto/libbitcoin_crypto_sse41.la
 LIBBITCOIN_CRYPTO += $(LIBBITCOIN_CRYPTO_SSE41)
+if ENABLE_X86_SHANI
+LIBBITCOIN_CRYPTO_X86_SHANI = crypto/libbitcoin_crypto_x86_shani.la
+LIBBITCOIN_CRYPTO += $(LIBBITCOIN_CRYPTO_X86_SHANI)
+endif
 endif
 if ENABLE_AVX2
 LIBBITCOIN_CRYPTO_AVX2 = crypto/libbitcoin_crypto_avx2.la
 LIBBITCOIN_CRYPTO += $(LIBBITCOIN_CRYPTO_AVX2)
 endif
-if ENABLE_X86_SHANI
-LIBBITCOIN_CRYPTO_X86_SHANI = crypto/libbitcoin_crypto_x86_shani.la
-LIBBITCOIN_CRYPTO += $(LIBBITCOIN_CRYPTO_X86_SHANI)
-endif
 if ENABLE_ARM_SHANI
 LIBBITCOIN_CRYPTO_ARM_SHANI = crypto/libbitcoin_crypto_arm_shani.la
 LIBBITCOIN_CRYPTO += $(LIBBITCOIN_CRYPTO_ARM_SHANI)
@@ -612,7 +612,7 @@ crypto_libbitcoin_crypto_x86_shani_la_LDFLAGS = $(AM_LDFLAGS) -static
 crypto_libbitcoin_crypto_x86_shani_la_CXXFLAGS = $(AM_CXXFLAGS) $(PIE_FLAGS) -static
 crypto_libbitcoin_crypto_x86_shani_la_CPPFLAGS = $(AM_CPPFLAGS)
 crypto_libbitcoin_crypto_x86_shani_la_CXXFLAGS += $(X86_SHANI_CXXFLAGS)
-crypto_libbitcoin_crypto_x86_shani_la_CPPFLAGS += -DENABLE_X86_SHANI
+crypto_libbitcoin_crypto_x86_shani_la_CPPFLAGS += -DENABLE_SSE41 -DENABLE_X86_SHANI
 crypto_libbitcoin_crypto_x86_shani_la_SOURCES = crypto/sha256_x86_shani.cpp
 
 # See explanation for -static in crypto_libbitcoin_crypto_base_la's LDFLAGS and
diff --git a/src/crypto/sha256.cpp b/src/crypto/sha256.cpp
index 4c7bb6f20ff..bb98cd09d2d 100644
--- a/src/crypto/sha256.cpp
+++ b/src/crypto/sha256.cpp
@@ -623,7 +623,7 @@ std::string SHA256AutoDetect(sha256_implementation::UseImplementation use_implem
         }
     }
 
-#if defined(ENABLE_X86_SHANI)
+#if defined(ENABLE_SSE41) && defined(ENABLE_X86_SHANI)
     if (have_x86_shani) {
         Transform = sha256_x86_shani::Transform;
         TransformD64 = TransformD64Wrapper<sha256_x86_shani::Transform>;
diff --git a/src/crypto/sha256_x86_shani.cpp b/src/crypto/sha256_x86_shani.cpp
index 79871bfcc11..74718281939 100644
--- a/src/crypto/sha256_x86_shani.cpp
+++ b/src/crypto/sha256_x86_shani.cpp
@@ -6,7 +6,7 @@
 // Written and placed in public domain by Jeffrey Walton.
 // Based on code from Intel, and by Sean Gulley for the miTLS project.
 
-#ifdef ENABLE_X86_SHANI
+#if defined(ENABLE_SSE41) && defined(ENABLE_X86_SHANI)
 
 #include <stdint.h>
 #include <immintrin.h>