Merge 13789 via bugfix_asm_pragmas

bitcoinknots · Jun 21, 2024 · 2dc1722 · 2dc1722
2 parents 5bb4fd2 + ea93c03
commit 2dc1722
Show file tree

Hide file tree

Showing 4 changed files with 48 additions and 0 deletions.
diff --git a/configure.ac b/configure.ac
@@ -506,6 +506,12 @@ AC_MSG_CHECKING([for SSE4.1 intrinsics])
 AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[
     #include <stdint.h>
     #include <immintrin.h>
+
+    #if defined(__clang__)
+    #pragma clang attribute push(__attribute__((__target__("sse4.1"))), apply_to = function)
+    #elif defined(__GNUC__)
+    #pragma GCC target ("sse4.1")
+    #endif
   ]],[[
     __m128i l = _mm_set1_epi32(0);
     return _mm_extract_epi32(l, 3);
@@ -521,6 +527,12 @@ AC_MSG_CHECKING([for AVX2 intrinsics])
 AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[
     #include <stdint.h>
     #include <immintrin.h>
+
+    #if defined(__clang__)
+    #pragma clang attribute push(__attribute__((__target__("avx,avx2"))), apply_to = function)
+    #elif defined(__GNUC__)
+    #pragma GCC target ("avx,avx2")
+    #endif
   ]],[[
     __m256i l = _mm256_set1_epi32(0);
     return _mm256_extract_epi32(l, 7);
@@ -536,6 +548,12 @@ AC_MSG_CHECKING([for x86 SHA-NI intrinsics])
 AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[
     #include <stdint.h>
     #include <immintrin.h>
+
+    #if defined(__clang__)
+    #pragma clang attribute push(__attribute__((__target__("sse4,sha"))), apply_to = function)
+    #elif defined(__GNUC__)
+    #pragma GCC target ("sse4,sha")
+    #endif
   ]],[[
     __m128i i = _mm_set1_epi32(0);
     __m128i j = _mm_set1_epi32(1);

diff --git a/src/crypto/sha256_avx2.cpp b/src/crypto/sha256_avx2.cpp
@@ -10,6 +10,12 @@
 #include <attributes.h>
 #include <crypto/common.h>
 
+#if defined(__clang__)
+#pragma clang attribute push(__attribute__((__target__("avx,avx2"))), apply_to = function)
+#elif defined(__GNUC__)
+#pragma GCC target ("avx,avx2")
+#endif
+
 namespace sha256d64_avx2 {
 namespace {
 
@@ -326,4 +332,8 @@ void Transform_8way(unsigned char* out, const unsigned char* in)
 
 }
 
+#if defined(__clang__)
+#pragma clang attribute pop
+#endif
+
 #endif
diff --git a/src/crypto/sha256_sse41.cpp b/src/crypto/sha256_sse41.cpp
@@ -10,6 +10,12 @@
 #include <attributes.h>
 #include <crypto/common.h>
 
+#if defined(__clang__)
+#pragma clang attribute push(__attribute__((__target__("sse4.1"))), apply_to = function)
+#elif defined(__GNUC__)
+#pragma GCC target ("sse4.1")
+#endif
+
 namespace sha256d64_sse41 {
 namespace {
 
@@ -318,4 +324,8 @@ void Transform_4way(unsigned char* out, const unsigned char* in)
 
 }
 
+#if defined(__clang__)
+#pragma clang attribute pop
+#endif
+
 #endif
diff --git a/src/crypto/sha256_x86_shani.cpp b/src/crypto/sha256_x86_shani.cpp
@@ -13,6 +13,12 @@
 
 #include <attributes.h>
 
+#if defined(__clang__)
+#pragma clang attribute push(__attribute__((__target__("sse4,sha"))), apply_to = function)
+#elif defined(__GNUC__)
+#pragma GCC target ("sse4,sha")
+#endif
+
 namespace {
 
 alignas(__m128i) const uint8_t MASK[16] = {0x03, 0x02, 0x01, 0x00, 0x07, 0x06, 0x05, 0x04, 0x0b, 0x0a, 0x09, 0x08, 0x0f, 0x0e, 0x0d, 0x0c};
@@ -355,4 +361,8 @@ void Transform_2way(unsigned char* out, const unsigned char* in)
 
 }
 
+#if defined(__clang__)
+#pragma clang attribute pop
+#endif
+
 #endif