[mips] clz is defined to give 32 for zero. Similarly, dclz gives 64.

Summary:
While debugging another issue, I noticed that Mips currently specifies that the
count leading zero builtins are undefined when the input is zero. The
architecture specifications say that the clz and dclz instructions write 32 or
64 respectively when given zero.

This doesn't fix any bugs that I'm aware of but it may improve optimisation in
some cases.

Differential Revision: http://reviews.llvm.org/D4431

llvm-svn: 212618

2 files changed, 34 insertions, 0 deletions

diff --git a/clang/lib/Basic/Targets.cpp b/clang/lib/Basic/Targets.cpp
index f048fa82274..75e154d7d5e 100644
--- a/clang/lib/Basic/Targets.cpp
+++ b/clang/lib/Basic/Targets.cpp
@@ -5513,6 +5513,8 @@ public:
     if (RegNo == 1) return 5;
     return -1;
   }
+
+  bool isCLZForZeroUndef() const override { return false; }
 };
 
 const Builtin::Info MipsTargetInfoBase::BuiltinInfo[] = {
diff --git a/clang/test/CodeGen/mips-count-builtins.c b/clang/test/CodeGen/mips-count-builtins.c
new file mode 100644
index 00000000000..7c9a2574809
--- /dev/null
+++ b/clang/test/CodeGen/mips-count-builtins.c
@@ -0,0 +1,32 @@
+// RUN: %clang_cc1 %s -triple mips-unknown-linux-gnu -emit-llvm -o - | FileCheck %s
+//
+// Test that the ctlz and cttz builtins are defined for zero.
+// Based on count-builtin.c
+
+int leading, trailing, pop;
+
+void test_i16(short P) {
+  leading = __builtin_clzs(P);
+  trailing = __builtin_ctzs(P);
+
+// CHECK: @test_i16
+// CHECK: call i16 @llvm.ctlz.i16(i16 {{.*}}, i1 false)
+// CHECK: call i16 @llvm.cttz.i16(i16 {{.*}}, i1 false)
+}
+
+void test_i32(int P) {
+  leading = __builtin_clz(P);
+  trailing = __builtin_ctz(P);
+
+// CHECK: @test_i32
+// CHECK: call i32 @llvm.ctlz.i32(i32 {{.*}}, i1 false)
+// CHECK: call i32 @llvm.cttz.i32(i32 {{.*}}, i1 false)
+}
+
+void test_i64(float P) {
+  leading = __builtin_clzll(P);
+  trailing = __builtin_ctzll(P);
+// CHECK: @test_i64
+// CHECK: call i64 @llvm.ctlz.i64(i64 {{.*}}, i1 false)
+// CHECK: call i64 @llvm.cttz.i64(i64 {{.*}}, i1 false)
+}