[fast-isel] Fold "urem x, pow2" -> "and x, pow2-1". This should fix the 271%

execution-time regression for nsieve-bits on the ARMv7 -O0 -g nightly tester.
This may also improve compile-time on architectures that would otherwise 
generate a libcall for urem (e.g., ARM) or fall back to the DAG selector.
rdar://10810716

llvm-svn: 153230

2 files changed, 16 insertions, 0 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp b/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp
index 9f4a44a4927..4db10b75868 100644
--- a/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp
@@ -395,6 +395,13 @@ bool FastISel::SelectBinaryOp(const User *I, unsigned ISDOpcode) {
       ISDOpcode = ISD::SRA;
     }
 
+    // Transform "urem x, pow2" -> "and x, pow2-1".
+    if (ISDOpcode == ISD::UREM && isa<BinaryOperator>(I) &&
+        isPowerOf2_64(Imm)) {
+      --Imm;
+      ISDOpcode = ISD::AND;
+    }
+
     unsigned ResultReg = FastEmit_ri_(VT.getSimpleVT(), ISDOpcode, Op0,
                                       Op0IsKill, Imm, VT.getSimpleVT());
     if (ResultReg == 0) return false;
diff --git a/llvm/test/CodeGen/ARM/fast-isel.ll b/llvm/test/CodeGen/ARM/fast-isel.ll
index 905543a54cf..417e2d9e410 100644
--- a/llvm/test/CodeGen/ARM/fast-isel.ll
+++ b/llvm/test/CodeGen/ARM/fast-isel.ll
@@ -217,3 +217,12 @@ entry:
 ; THUMB: vcmpe.f32 s0, #0
   ret i1 %4
 }
+
+; ARM: @urem_fold
+; THUMB: @urem_fold
+; ARM: and r0, r0, #31
+; THUMB: and r0, r0, #31
+define i32 @urem_fold(i32 %a) nounwind {
+  %rem = urem i32 %a, 32
+  ret i32 %rem
+}