add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@74508 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 20 insertions, 0 deletions

diff --git a/lib/Target/X86/README.txt b/lib/Target/X86/README.txt
index 4464878ce2..711f1d2345 100644
--- a/lib/Target/X86/README.txt
+++ b/lib/Target/X86/README.txt
@@ -1932,3 +1932,23 @@ Replacing an icmp+select with a shift should always be considered profitable in
 instcombine.
 
 //===---------------------------------------------------------------------===//
+
+It looks like we don't have patterns (or they aren't matching) for adc with
+immediate:
+
+define i64 @f1(i64 %a) nounwind {
+   %tmp = sub i64 %a, 734439407618
+   ret i64 %tmp
+}
+$ llvm-as < t.ll | llc -march=x86
+
+_f1:
+	movl	4(%esp), %eax
+	addl	$4294967294, %eax
+	movl	$4294967124, %edx
+	adcl	8(%esp), %edx
+	ret
+
+There is no need to clobber %edx there.
+
+//===---------------------------------------------------------------------===//