[AMDGPU] Add f16 support (VI+)

Differential Revision: https://reviews.llvm.org/D25975

llvm-svn: 286753

1 files changed, 49 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/AMDGPU/llvm.floor.f16.ll b/llvm/test/CodeGen/AMDGPU/llvm.floor.f16.ll
new file mode 100644
index 00000000000..a7bdb1e9428
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/llvm.floor.f16.ll
@@ -0,0 +1,49 @@
+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
+; RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
+
+declare half @llvm.floor.f16(half %a)
+declare <2 x half> @llvm.floor.v2f16(<2 x half> %a)
+
+; GCN-LABEL: {{^}}floor_f16
+; GCN: buffer_load_ushort v[[A_F16:[0-9]+]]
+; SI:  v_cvt_f32_f16_e32 v[[A_F32:[0-9]+]], v[[A_F16]]
+; SI:  v_floor_f32_e32 v[[R_F32:[0-9]+]], v[[A_F32]]
+; SI:  v_cvt_f16_f32_e32 v[[R_F16:[0-9]+]], v[[R_F32]]
+; VI:  v_floor_f16_e32 v[[R_F16:[0-9]+]], v[[A_F16]]
+; GCN: buffer_store_short v[[R_F16]]
+; GCN: s_endpgm
+define void @floor_f16(
+    half addrspace(1)* %r,
+    half addrspace(1)* %a) {
+entry:
+  %a.val = load half, half addrspace(1)* %a
+  %r.val = call half @llvm.floor.f16(half %a.val)
+  store half %r.val, half addrspace(1)* %r
+  ret void
+}
+
+; GCN-LABEL: {{^}}floor_v2f16
+; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
+; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; SI:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
+; SI:  v_floor_f32_e32 v[[R_F32_0:[0-9]+]], v[[A_F32_0]]
+; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
+; SI:  v_floor_f32_e32 v[[R_F32_1:[0-9]+]], v[[A_F32_1]]
+; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+; VI:  v_floor_f16_e32 v[[R_F16_0:[0-9]+]], v[[A_V2_F16]]
+; VI:  v_floor_f16_e32 v[[R_F16_1:[0-9]+]], v[[A_F16_1]]
+; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: buffer_store_dword v[[R_V2_F16]]
+; GCN: s_endpgm
+define void @floor_v2f16(
+    <2 x half> addrspace(1)* %r,
+    <2 x half> addrspace(1)* %a) {
+entry:
+  %a.val = load <2 x half>, <2 x half> addrspace(1)* %a
+  %r.val = call <2 x half> @llvm.floor.v2f16(<2 x half> %a.val)
+  store <2 x half> %r.val, <2 x half> addrspace(1)* %r
+  ret void
+}