Fix handling of __clock() builtin for "generic" targets.

cbackend.cpp

@@ -3704,6 +3704,7 @@ void CWriter::lowerIntrinsics(llvm::Function &F) {
           case llvm::Intrinsic::sadd_with_overflow:
           case llvm::Intrinsic::trap:
           case llvm::Intrinsic::objectsize:
+          case llvm::Intrinsic::readcyclecounter:
               // We directly implement these intrinsics
             break;
           default:
@@ -4056,6 +4057,9 @@ bool CWriter::visitBuiltinCall(llvm::CallInst &I, llvm::Intrinsic::ID ID,
     return true;
   case llvm::Intrinsic::objectsize:
     return true;
+  case llvm::Intrinsic::readcyclecounter:
+    Out << "__clock()";
+    return true;
   }
 }
 

examples/intrinsics/generic-16.h

@@ -1759,3 +1759,23 @@ static FORCEINLINE uint64_t __atomic_cmpxchg(uint64_t *p, uint64_t cmpval,
     return __sync_val_compare_and_swap(p, cmpval, newval);
 #endif
 }
+
+#ifdef WIN32
+#include <windows.h>
+#define __clock __rdtsc
+#else // WIN32
+static FORCEINLINE uint64_t __clock() {
+  uint32_t low, high;
+#ifdef __x86_64
+  __asm__ __volatile__ ("xorl %%eax,%%eax \n    cpuid"
+                        ::: "%rax", "%rbx", "%rcx", "%rdx" );
+#else
+  __asm__ __volatile__ ("xorl %%eax,%%eax \n    cpuid"
+                        ::: "%eax", "%ebx", "%ecx", "%edx" );
+#endif
+  __asm__ __volatile__ ("rdtsc" : "=a" (low), "=d" (high));
+  return (uint64_t)high << 32 | low;
+}
+
+#endif // !WIN32
+

examples/intrinsics/generic-32.h

@@ -1827,3 +1827,23 @@ static FORCEINLINE uint64_t __atomic_cmpxchg(uint64_t *p, uint64_t cmpval,
     return __sync_val_compare_and_swap(p, cmpval, newval);
 #endif
 }
+
+#ifdef WIN32
+#include <windows.h>
+#define __clock __rdtsc
+#else // WIN32
+static FORCEINLINE uint64_t __clock() {
+  uint32_t low, high;
+#ifdef __x86_64
+  __asm__ __volatile__ ("xorl %%eax,%%eax \n    cpuid"
+                        ::: "%rax", "%rbx", "%rcx", "%rdx" );
+#else
+  __asm__ __volatile__ ("xorl %%eax,%%eax \n    cpuid"
+                        ::: "%eax", "%ebx", "%ecx", "%edx" );
+#endif
+  __asm__ __volatile__ ("rdtsc" : "=a" (low), "=d" (high));
+  return (uint64_t)high << 32 | low;
+}
+#endif // WIN32
+
+#undef FORCEINLINE

examples/intrinsics/generic-64.h

@@ -1960,3 +1960,23 @@ static FORCEINLINE uint64_t __atomic_cmpxchg(uint64_t *p, uint64_t cmpval,
     return __sync_val_compare_and_swap(p, cmpval, newval);
 #endif
 }
+
+#ifdef WIN32
+#include <windows.h>
+#define __clock __rdtsc
+#else // WIN32
+static FORCEINLINE uint64_t __clock() {
+  uint32_t low, high;
+#ifdef __x86_64
+  __asm__ __volatile__ ("xorl %%eax,%%eax \n    cpuid"
+                        ::: "%rax", "%rbx", "%rcx", "%rdx" );
+#else
+  __asm__ __volatile__ ("xorl %%eax,%%eax \n    cpuid"
+                        ::: "%eax", "%ebx", "%ecx", "%edx" );
+#endif
+  __asm__ __volatile__ ("rdtsc" : "=a" (low), "=d" (high));
+  return (uint64_t)high << 32 | low;
+}
+#endif
+
+#undef FORCEINLINE

examples/intrinsics/knc.h

@@ -2121,9 +2121,24 @@ static FORCEINLINE uint64_t __atomic_cmpxchg(uint64_t *p, uint64_t cmpval,
 #endif
 }
 
+#ifdef WIN32
+#include <windows.h>
+#define __clock __rdtsc
+#else // WIN32
+static FORCEINLINE uint64_t __clock() {
+  uint32_t low, high;
+#ifdef __x86_64
+  __asm__ __volatile__ ("xorl %%eax,%%eax \n    cpuid"
+                        ::: "%rax", "%rbx", "%rcx", "%rdx" );
+#else
+  __asm__ __volatile__ ("xorl %%eax,%%eax \n    cpuid"
+                        ::: "%eax", "%ebx", "%ecx", "%edx" );
+#endif
+  __asm__ __volatile__ ("rdtsc" : "=a" (low), "=d" (high));
+  return (uint64_t)high << 32 | low;
+}
+#endif // !WIN32
+
 #undef FORCEINLINE
 #undef PRE_ALIGN
 #undef POST_ALIGN
-
-
-

examples/intrinsics/knc2x.h

@@ -2055,7 +2055,24 @@ static FORCEINLINE void __aos_to_soa4_float(float *ptr, __vec32_f *out0, __vec32
 }
 */
 
+#ifdef WIN32
+#include <windows.h>
+#define __clock __rdtsc
+#else // WIN32
+static FORCEINLINE uint64_t __clock() {
+  uint32_t low, high;
+#ifdef __x86_64
+  __asm__ __volatile__ ("xorl %%eax,%%eax \n    cpuid"
+                        ::: "%rax", "%rbx", "%rcx", "%rdx" );
+#else
+  __asm__ __volatile__ ("xorl %%eax,%%eax \n    cpuid"
+                        ::: "%eax", "%ebx", "%ecx", "%edx" );
+#endif
+  __asm__ __volatile__ ("rdtsc" : "=a" (low), "=d" (high));
+  return (uint64_t)high << 32 | low;
+}
+#endif // !WIN32
+
 #undef FORCEINLINE
 #undef PRE_ALIGN
 #undef POST_ALIGN
-

examples/intrinsics/sse4.h

@@ -4000,6 +4000,22 @@ static FORCEINLINE uint64_t __atomic_cmpxchg(uint64_t *p, uint64_t cmpval,
 #endif
 }
 
+#ifdef WIN32
+#include <windows.h>
+#define __clock __rdtsc
+#else // WIN32
+static FORCEINLINE uint64_t __clock() {
+  uint32_t low, high;
+#ifdef __x86_64
+  __asm__ __volatile__ ("xorl %%eax,%%eax \n    cpuid"
+                        ::: "%rax", "%rbx", "%rcx", "%rdx" );
+#else
+  __asm__ __volatile__ ("xorl %%eax,%%eax \n    cpuid"
+                        ::: "%eax", "%ebx", "%ecx", "%edx" );
+#endif
+  __asm__ __volatile__ ("rdtsc" : "=a" (low), "=d" (high));
+  return (uint64_t)high << 32 | low;
+}
+#endif // !WIN32
+
 #undef FORCEINLINE
-
-