x86: Enable Intel CET

When Intel CET is enabled, we need to include <cet.h> in assembly codes
to mark Intel CET support and place _CET_ENDBR at the function entry.

Author: hjl-tools

common.h

@@ -416,6 +416,15 @@ please https://github.com/xianyi/OpenBLAS/issues/246
 #include "common_alpha.h"
 #endif
 
+#if (defined(ARCH_X86) || defined(ARCH_X86_64)) && defined(__CET__) && defined(__has_include)
+#if __has_include(<cet.h>)
+#include <cet.h>
+#endif
+#endif
+#ifndef _CET_ENDBR
+#define _CET_ENDBR
+#endif
+
 #ifdef ARCH_X86
 #include "common_x86.h"
 #endif

common_x86.h

@@ -340,7 +340,8 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 	.align 16; \
 	.globl REALNAME ;\
        .type REALNAME, @function; \
-REALNAME:
+REALNAME: \
+	_CET_ENDBR
 
 #ifdef PROFILE
 #define PROFCODE call mcount

common_x86_64.h

@@ -451,7 +451,8 @@ static __inline unsigned int blas_quickdivide(unsigned int x, unsigned int y){
 	.align 512; \
 	.globl REALNAME ;\
        .type REALNAME, @function; \
-REALNAME:
+REALNAME: \
+	_CET_ENDBR
 
 #ifdef PROFILE
 #define PROFCODE call *mcount@GOTPCREL(%rip)