NVPTX: support fpext/fptrunc to and from f16.

llvm-svn: 213377

diff --git a/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp b/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
index 645a9bb5c5e3ea4477d0c027be8fc508ea21185b..05bad16ddd89305ca44f27e55f44ad86e78fb9aa 100644 (file)
--- a/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
+++ b/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
@@ -203,8 +203,11 @@ NVPTXTargetLowering::NVPTXTargetLowering(NVPTXTargetMachine &TM)
   setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::Other, Custom);
 
   // Turn FP extload into load/fextend
+  setLoadExtAction(ISD::EXTLOAD, MVT::f16, Expand);
   setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
   // Turn FP truncstore into trunc + store.
+  setTruncStoreAction(MVT::f32, MVT::f16, Expand);
+  setTruncStoreAction(MVT::f64, MVT::f16, Expand);
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
 
   // PTX does not support load / store predicate registers

diff --git a/llvm/test/CodeGen/NVPTX/half.ll b/llvm/test/CodeGen/NVPTX/half.ll
index ba18dc3c1797b0552cbfca3ea29132b00d4d797d..aa08cc78e91a781a380eb578b607091342764120 100644 (file)
--- a/llvm/test/CodeGen/NVPTX/half.ll
+++ b/llvm/test/CodeGen/NVPTX/half.ll
@@ -28,3 +28,43 @@ define void @test_bitcast_to_half(half addrspace(1)* %out, i16 addrspace(1)* %in
   store half %val_fp, half addrspace(1)* %out
   ret void
 }
+
+define void @test_extend32(half addrspace(1)* %in, float addrspace(1)* %out) {
+; CHECK-LABEL: @test_extend32
+; CHECK: cvt.f32.f16
+
+  %val16 = load half addrspace(1)* %in
+  %val32 = fpext half %val16 to float
+  store float %val32, float addrspace(1)* %out
+  ret void
+}
+
+define void @test_extend64(half addrspace(1)* %in, double addrspace(1)* %out) {
+; CHECK-LABEL: @test_extend64
+; CHECK: cvt.f64.f16
+
+  %val16 = load half addrspace(1)* %in
+  %val64 = fpext half %val16 to double
+  store double %val64, double addrspace(1)* %out
+  ret void
+}
+
+define void @test_trunc32(float addrspace(1)* %in, half addrspace(1)* %out) {
+; CHECK-LABEL: test_trunc32
+; CHECK: cvt.rn.f16.f32
+
+  %val32 = load float addrspace(1)* %in
+  %val16 = fptrunc float %val32 to half
+  store half %val16, half addrspace(1)* %out
+  ret void
+}
+
+define void @test_trunc64(double addrspace(1)* %in, half addrspace(1)* %out) {
+; CHECK-LABEL: @test_trunc64
+; CHECK: cvt.rn.f16.f64
+
+  %val32 = load double addrspace(1)* %in
+  %val16 = fptrunc double %val32 to half
+  store half %val16, half addrspace(1)* %out
+  ret void
+}