diff options
Diffstat (limited to 'test/CodeGen/AMDGPU/hsa-fp-mode.ll')
| -rw-r--r-- | test/CodeGen/AMDGPU/hsa-fp-mode.ll | 31 |
1 files changed, 21 insertions, 10 deletions
diff --git a/test/CodeGen/AMDGPU/hsa-fp-mode.ll b/test/CodeGen/AMDGPU/hsa-fp-mode.ll index 51d6aee25f456..b1901cf894b08 100644 --- a/test/CodeGen/AMDGPU/hsa-fp-mode.ll +++ b/test/CodeGen/AMDGPU/hsa-fp-mode.ll @@ -4,7 +4,7 @@ ; GCN: float_mode = 192 ; GCN: enable_dx10_clamp = 1 ; GCN: enable_ieee_mode = 1 -define void @test_default_ci(float addrspace(1)* %out0, double addrspace(1)* %out1) #0 { +define amdgpu_kernel void @test_default_ci(float addrspace(1)* %out0, double addrspace(1)* %out1) #0 { store float 0.0, float addrspace(1)* %out0 store double 0.0, double addrspace(1)* %out1 ret void @@ -14,7 +14,7 @@ define void @test_default_ci(float addrspace(1)* %out0, double addrspace(1)* %ou ; GCN: float_mode = 192 ; GCN: enable_dx10_clamp = 1 ; GCN: enable_ieee_mode = 1 -define void @test_default_vi(float addrspace(1)* %out0, double addrspace(1)* %out1) #1 { +define amdgpu_kernel void @test_default_vi(float addrspace(1)* %out0, double addrspace(1)* %out1) #1 { store float 0.0, float addrspace(1)* %out0 store double 0.0, double addrspace(1)* %out1 ret void @@ -24,7 +24,7 @@ define void @test_default_vi(float addrspace(1)* %out0, double addrspace(1)* %ou ; GCN: float_mode = 192 ; GCN: enable_dx10_clamp = 1 ; GCN: enable_ieee_mode = 1 -define void @test_f64_denormals(float addrspace(1)* %out0, double addrspace(1)* %out1) #2 { +define amdgpu_kernel void @test_f64_denormals(float addrspace(1)* %out0, double addrspace(1)* %out1) #2 { store float 0.0, float addrspace(1)* %out0 store double 0.0, double addrspace(1)* %out1 ret void @@ -34,7 +34,7 @@ define void @test_f64_denormals(float addrspace(1)* %out0, double addrspace(1)* ; GCN: float_mode = 48 ; GCN: enable_dx10_clamp = 1 ; GCN: enable_ieee_mode = 1 -define void @test_f32_denormals(float addrspace(1)* %out0, double addrspace(1)* %out1) #3 { +define amdgpu_kernel void @test_f32_denormals(float addrspace(1)* %out0, double addrspace(1)* %out1) #3 { store float 0.0, float addrspace(1)* %out0 store double 0.0, double addrspace(1)* %out1 ret void @@ -44,7 +44,7 @@ define void @test_f32_denormals(float addrspace(1)* %out0, double addrspace(1)* ; GCN: float_mode = 240 ; GCN: enable_dx10_clamp = 1 ; GCN: enable_ieee_mode = 1 -define void @test_f32_f64_denormals(float addrspace(1)* %out0, double addrspace(1)* %out1) #4 { +define amdgpu_kernel void @test_f32_f64_denormals(float addrspace(1)* %out0, double addrspace(1)* %out1) #4 { store float 0.0, float addrspace(1)* %out0 store double 0.0, double addrspace(1)* %out1 ret void @@ -54,7 +54,17 @@ define void @test_f32_f64_denormals(float addrspace(1)* %out0, double addrspace( ; GCN: float_mode = 0 ; GCN: enable_dx10_clamp = 1 ; GCN: enable_ieee_mode = 1 -define void @test_no_denormals(float addrspace(1)* %out0, double addrspace(1)* %out1) #5 { +define amdgpu_kernel void @test_no_denormals(float addrspace(1)* %out0, double addrspace(1)* %out1) #5 { + store float 0.0, float addrspace(1)* %out0 + store double 0.0, double addrspace(1)* %out1 + ret void +} + +; GCN-LABEL: {{^}}test_no_dx10_clamp_vi: +; GCN: float_mode = 192 +; GCN: enable_dx10_clamp = 0 +; GCN: enable_ieee_mode = 1 +define amdgpu_kernel void @test_no_dx10_clamp_vi(float addrspace(1)* %out0, double addrspace(1)* %out1) #6 { store float 0.0, float addrspace(1)* %out0 store double 0.0, double addrspace(1)* %out1 ret void @@ -62,7 +72,8 @@ define void @test_no_denormals(float addrspace(1)* %out0, double addrspace(1)* % attributes #0 = { nounwind "target-cpu"="kaveri" } attributes #1 = { nounwind "target-cpu"="fiji" } -attributes #2 = { nounwind "target-features"="-fp32-denormals,+fp64-denormals" } -attributes #3 = { nounwind "target-features"="+fp32-denormals,-fp64-denormals" } -attributes #4 = { nounwind "target-features"="+fp32-denormals,+fp64-denormals" } -attributes #5 = { nounwind "target-features"="-fp32-denormals,-fp64-denormals" } +attributes #2 = { nounwind "target-features"="-fp32-denormals,+fp64-fp16-denormals" } +attributes #3 = { nounwind "target-features"="+fp32-denormals,-fp64-fp16-denormals" } +attributes #4 = { nounwind "target-features"="+fp32-denormals,+fp64-fp16-denormals" } +attributes #5 = { nounwind "target-features"="-fp32-denormals,-fp64-fp16-denormals" } +attributes #6 = { nounwind "target-cpu"="fiji" "target-features"="-dx10-clamp" } |