[llvm] r292328 - DAG: Consider nnan in isKnownNeverNaN

[Matt Arsenault via llvm-commits]

Author: arsenm
Date: Tue Jan 17 20:10:08 2017
New Revision: 292328

URL: http://llvm.org/viewvc/llvm-project?rev=292328&view=rev
Log:
DAG: Consider nnan in isKnownNeverNaN

Modified:
    llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
    llvm/trunk/test/CodeGen/AMDGPU/fmed3.ll

Modified: llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp?rev=292328&r1=292327&r2=292328&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp Tue Jan 17 20:10:08 2017
@@ -3061,6 +3061,9 @@ bool SelectionDAG::isKnownNeverNaN(SDVal
   if (getTarget().Options.NoNaNsFPMath)
     return true;
 
+  if (const BinaryWithFlagsSDNode *BF = dyn_cast<BinaryWithFlagsSDNode>(Op))
+    return BF->Flags.hasNoNaNs();
+
   // If the value is a constant, we can obviously see if it is a NaN or not.
   if (const ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(Op))
     return !C->getValueAPF().isNaN();

Modified: llvm/trunk/test/CodeGen/AMDGPU/fmed3.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/fmed3.ll?rev=292328&r1=292327&r2=292328&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/fmed3.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/fmed3.ll Tue Jan 17 20:10:08 2017
@@ -7,6 +7,22 @@ declare float @llvm.maxnum.f32(float, fl
 declare double @llvm.minnum.f64(double, double) #0
 declare double @llvm.maxnum.f64(double, double) #0
 
+; GCN-LABEL: {{^}}v_test_nnan_input_fmed3_r_i_i_f32:
+; GCN: v_add_f32_e32 [[ADD:v[0-9]+]], 1.0, v{{[0-9]+}}
+; GCN: v_med3_f32 v{{[0-9]+}}, [[ADD]], 2.0, 4.0
+define void @v_test_nnan_input_fmed3_r_i_i_f32(float addrspace(1)* %out, float addrspace(1)* %aptr) #1 {
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %gep0 = getelementptr float, float addrspace(1)* %aptr, i32 %tid
+  %outgep = getelementptr float, float addrspace(1)* %out, i32 %tid
+  %a = load float, float addrspace(1)* %gep0
+  %a.add = fadd nnan float %a, 1.0
+  %max = call float @llvm.maxnum.f32(float %a.add, float 2.0)
+  %med = call float @llvm.minnum.f32(float %max, float 4.0)
+
+  store float %med, float addrspace(1)* %outgep
+  ret void
+}
+
 ; GCN-LABEL: {{^}}v_test_fmed3_r_i_i_f32:
 ; NOSNAN: v_med3_f32 v{{[0-9]+}}, v{{[0-9]+}}, 2.0, 4.0