Fix for Bug 26903, adds support to inline __builtin_mempcpy

Patch by Sunita Marathe

Differential Revision: http://reviews.llvm.org/D21920

llvm-svn: 275284

4 files changed, 50 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/Generic/mempcpy_call.ll b/llvm/test/CodeGen/Generic/mempcpy_call.ll
new file mode 100644
index 00000000000..b88f20160af
--- /dev/null
+++ b/llvm/test/CodeGen/Generic/mempcpy_call.ll
@@ -0,0 +1,16 @@
+; RUN: llc < %s -O2 | FileCheck %s
+
+; This test just checks that mempcpy is lowered as memcpy.
+; The test to check that the return value of mempcpy is the dst pointer adjusted
+; by the copy size is done by Codegen/X86/mempcpy_ret_val.ll
+
+; CHECK-LABEL: CALL_MEMPCPY:
+; CHECK: callq memcpy
+;
+define void @CALL_MEMPCPY(i8* %DST, i8* %SRC, i64 %N) {
+entry:
+  %call = tail call i8* @mempcpy(i8* %DST, i8* %SRC, i64 %N)
+  ret void
+}
+
+declare i8* @mempcpy(i8*, i8*, i64)
diff --git a/llvm/test/CodeGen/X86/mempcpy_ret_val.ll b/llvm/test/CodeGen/X86/mempcpy_ret_val.ll
new file mode 100644
index 00000000000..d914f15101b
--- /dev/null
+++ b/llvm/test/CodeGen/X86/mempcpy_ret_val.ll
@@ -0,0 +1,28 @@
+; RUN: llc < %s -O2 | FileCheck %s
+
+@G = common global i8* null, align 8
+
+; 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.
+
+; CHECK-LABEL: RET_MEMPCPY:
+; CHECK: movq [[REG:%r[a-z0-9]+]], {{.*}}G
+; CHECK: callq memcpy
+; CHECK: movq [[REG]], %rax
+;
+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(...)