OSDN Git Service

[AMDGPU] adjusted test checks because minnum with NaN gets simplified
authorSanjay Patel <spatel@rotateright.com>
Sun, 15 Jul 2018 15:14:40 +0000 (15:14 +0000)
committerSanjay Patel <spatel@rotateright.com>
Sun, 15 Jul 2018 15:14:40 +0000 (15:14 +0000)
This was improved with rL337127, but I missed the failure in this test.
I'm not sure what the expected result will be, so I've generalized it
and added a FIXME comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@337128 91177308-0d34-0410-b5e6-96231b3b80d8

test/CodeGen/AMDGPU/fcanonicalize-elimination.ll

index 48e4828..a4db46b 100644 (file)
@@ -379,11 +379,12 @@ define amdgpu_kernel void @test_fold_canonicalize_minnum_value_f32(float addrspa
   ret void
 }
 
+; FIXME: Should there be more checks here? minnum with NaN operand is simplified away.
+
 ; GCN-LABEL: test_fold_canonicalize_sNaN_value_f32:
-; GCN:  v_min_f32_e32 [[V0:v[0-9]+]], 0x7f800001, v{{[0-9]+}}
-; GCN-FLUSH:  v_mul_f32_e32 [[RESULT:v[0-9]+]], 1.0, [[V0]]
-; GCN-DENORM: v_max_f32_e32 [[RESULT:v[0-9]+]], [[V0]], [[V0]]
-; GCN:  {{flat|global}}_store_dword v[{{[0-9:]+}}], [[RESULT]]
+; VI:   v_add_u32_e32 v{{[0-9]+}}
+; GFX9:        v_add_co_u32_e32 v{{[0-9]+}}
+; GCN:  {{flat|global}}_store_dword v[{{[0-9:]+}}]
 define amdgpu_kernel void @test_fold_canonicalize_sNaN_value_f32(float addrspace(1)* %arg) {
   %id = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr inbounds float, float addrspace(1)* %arg, i32 %id