Recommitting r275284: add support to inline __builtin_mempcpy

Patch by Sunita Marathe

Third try, now following fixes to MSan to handle mempcy in such a way that this commit won't break the MSan buildbots. (Thanks, Evegenii!)

llvm-svn: 277189

3 files changed, 34 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/X86/mempcpy.ll b/llvm/test/CodeGen/X86/mempcpy.ll
new file mode 100644
index 00000000000..1c737b64402
--- /dev/null
+++ b/llvm/test/CodeGen/X86/mempcpy.ll
@@ -0,0 +1,28 @@
+;  RUN: llc < %s -mtriple=x86_64-unknown-linux -O2 | FileCheck %s
+;  RUN: llc < %s -mtriple=i686-unknown-linux -O2 | FileCheck %s
+
+; This test checks that:
+; (1)  mempcpy is lowered as memcpy, and 
+; (2)  its return value is DST+N i.e. the dst pointer adjusted by the copy size.
+; To keep the testing of (2) independent of the exact instructions used to
+; adjust the dst pointer, DST+N is explicitly computed and stored to a global
+; variable G before the mempcpy call. This instance of DST+N causes the repeat
+; DST+N done in the context of the return value of mempcpy to be redundant, and
+; the first instance to be reused as the return value. This allows the check for
+; (2) to be expressed as verifying that the MOV to store DST+N to G and
+; the MOV to copy DST+N to %rax use the same source register.
+@G = common global i8* null, align 8
+
+; CHECK-LABEL: RET_MEMPCPY:
+; CHECK: mov{{.*}} [[REG:%[er][a-z0-9]+]], {{.*}}G
+; CHECK: call{{.*}} {{.*}}memcpy
+; CHECK: mov{{.*}} [[REG]], %{{[er]}}ax
+;
+define i8* @RET_MEMPCPY(i8* %DST, i8* %SRC, i64 %N) {
+  %add.ptr = getelementptr inbounds i8, i8* %DST, i64 %N
+  store i8* %add.ptr, i8** @G, align 8
+  %call = tail call i8* @mempcpy(i8* %DST, i8* %SRC, i64 %N) 
+  ret i8* %call
+}
+
+declare i8* @mempcpy(i8*, i8*, i64)
diff --git a/llvm/test/Transforms/InferFunctionAttrs/annotate.ll b/llvm/test/Transforms/InferFunctionAttrs/annotate.ll
index 039114d0cb9..64676bf310b 100644
--- a/llvm/test/Transforms/InferFunctionAttrs/annotate.ll
+++ b/llvm/test/Transforms/InferFunctionAttrs/annotate.ll
@@ -499,6 +499,9 @@ declare i32 @memcmp(i8*, i8*, i64)
 ; CHECK: declare i8* @memcpy(i8*, i8* nocapture readonly, i64) [[G0]]
 declare i8* @memcpy(i8*, i8*, i64)
 
+; CHECK: declare i8* @mempcpy(i8*, i8* nocapture readonly, i64) [[G0]]
+declare i8* @mempcpy(i8*, i8*, i64)
+
 ; CHECK: declare i8* @memmove(i8*, i8* nocapture readonly, i64) [[G0]]
 declare i8* @memmove(i8*, i8*, i64)
 
diff --git a/llvm/test/Transforms/InferFunctionAttrs/no-proto.ll b/llvm/test/Transforms/InferFunctionAttrs/no-proto.ll
index 256f5c356b8..25a4805c367 100644
--- a/llvm/test/Transforms/InferFunctionAttrs/no-proto.ll
+++ b/llvm/test/Transforms/InferFunctionAttrs/no-proto.ll
@@ -480,6 +480,9 @@ declare void @memcmp(...)
 ; CHECK: declare void @memcpy(...)
 declare void @memcpy(...)
 
+; CHECK: declare void @mempcpy(...)
+declare void @mempcpy(...)
+
 ; CHECK: declare void @memmove(...)
 declare void @memmove(...)