Implement __stosb intrinsic as a volatile memset

[agutowski]

Summary

We need __stosb to be an intrinsic, because SecureZeroMemory function uses it without including intrin.h. Implementing it as a volatile memset is not consistent with MSDN specification, but it gives us target-independent IR while keeping the most important properties of __stosb.

[hans]

lib/CodeGen/CGBuiltin.cpp

7604

Nit: I'd suggest capital w for "We" and ending the sentence with a period. I see many other comments in the file don't do this, but we can at least set a good example :-)

7610

Hmm, does the __stosb intrinsic require Dest to be non-null (e.g. would Dest=NULL, Count=0 be OK?) I'm not even sure what llvm's memset requires actually.

7613

Why is it returning Dest here, and not the result of Builder.CreateMemSet?

[agutowski]

fix return value and comment

[agutowski]

lib/CodeGen/CGBuiltin.cpp

7613

My mistake (that's what memset returns), thanks!

[agutowski]

remove nullptr check; don't repeat EmitScalarExpr

[agutowski]

lib/CodeGen/CGBuiltin.cpp

7610

I can't find any guarantee that memset accepts Dest=NULL and Count=0. So I guess we can either add a branch here, checking if the pointer is NULL (and that Count=0?) or assume that memset won't do anything strange. I vote for the latter (the current code does it).

[majnemer]

lib/CodeGen/CGBuiltin.cpp

7610

I think the approach you've taken here is fine, we shouldn't need a branch.

[agutowski]

rebase

[majnemer]

lib/CodeGen/CGBuiltin.cpp

7777

I think we should manually force the alignment to 1 instead of trying to be clever.

[agutowski]

force the alignment to 1

[majnemer]

LGTM

include/clang/Basic/BuiltinsX86.def

 
 TARGET_HEADER_BUILTIN(_AddressOfReturnAddress, "v*", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
 
+TARGET_HEADER_BUILTIN(__stosb, "vUc*Ucz", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+
 #undef BUILTIN
 #undef TARGET_BUILTIN
 #undef TARGET_HEADER_BUILTIN

lib/CodeGen/CGBuiltin.cpp

     Value *F = CGM.getIntrinsic(Intrinsic::addressofreturnaddress);
     return Builder.CreateCall(F);
   }
+  case X86::BI__stosb: {
+    // We treat __stosb as a volatile memset - it may not generate "rep stosb"
+    // instruction, but it will create a memset that won't be optimized away.
+    return Builder.CreateMemSet(Ops[0], Ops[1], Ops[2], 1, true);
   }
+  }
 }
 
 

lib/Headers/intrin.h

                         : "%edi", "%esi", "%ecx");
 }
 static __inline__ void __DEFAULT_FN_ATTRS
-__stosb(unsigned char *__dst, unsigned char __x, size_t __n) {
-  __asm__("rep stosb" : : "D"(__dst), "a"(__x), "c"(__n)
-                        : "%edi", "%ecx");
-}
-static __inline__ void __DEFAULT_FN_ATTRS
 __stosd(unsigned long *__dst, unsigned long __x, size_t __n) {
   __asm__("rep stosl" : : "D"(__dst), "a"(__x), "c"(__n)
                         : "%edi", "%ecx");

test/CodeGen/ms-intrinsics.c

 
 #include <intrin.h>
 
+#if defined(__i386__) || defined(__x86_64__)
+void test__stosb(unsigned char *Dest, unsigned char Data, size_t Count) {
+  return __stosb(Dest, Data, Count);
+}
+
+// CHECK-I386: define{{.*}}void @test__stosb
+// CHECK-I386:   tail call void @llvm.memset.p0i8.i32(i8* %Dest, i8 %Data, i32 %Count, i32 1, i1 true)
+// CHECK-I386:   ret void
+// CHECK-I386: }
+
+// CHECK-X64: define{{.*}}void @test__stosb
+// CHECK-X64:   tail call void @llvm.memset.p0i8.i64(i8* %Dest, i8 %Data, i64 %Count, i32 1, i1 true)
+// CHECK-X64:   ret void
+// CHECK-X64: }
+#endif
+
 void *test_ReturnAddress() {
   return _ReturnAddress();
 }

test/Headers/ms-intrin.cpp

   __movsd(0, 0, 0);
   __movsw(0, 0, 0);
 
-  __stosb(0, 0, 0);
   __stosd(0, 0, 0);
   __stosw(0, 0, 0);