[AMDGPU] Expand `br_cc` for f16

Differential Revision: https://reviews.llvm.org/D26732

llvm-svn: 287199

1 files changed, 112 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/AMDGPU/br_cc.f16.ll b/llvm/test/CodeGen/AMDGPU/br_cc.f16.ll
new file mode 100644
index 00000000000..6cf3fdad3e3
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/br_cc.f16.ll
@@ -0,0 +1,112 @@
+; 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
+
+; GCN-LABEL: {{^}}br_cc_f16
+; GCN: buffer_load_ushort v[[A_F16:[0-9]+]]
+; GCN: buffer_load_ushort v[[B_F16:[0-9]+]]
+
+; SI:  v_cvt_f32_f16_e32 v[[A_F32:[0-9]+]], v[[A_F16]]
+; SI:  v_cvt_f32_f16_e32 v[[B_F32:[0-9]+]], v[[B_F16]]
+; SI:  v_cmp_nlt_f32_e32 vcc, v[[A_F32]], v[[B_F32]]
+; VI:  v_cmp_nlt_f16_e32 vcc, v[[A_F16]], v[[B_F16]]
+; GCN: s_cbranch_vccnz
+
+; GCN: one{{$}}
+; SI:  v_cvt_f16_f32_e32 v[[A_F16:[0-9]+]], v[[A_F32]]
+; GCN: buffer_store_short v[[A_F16]]
+; GCN: s_endpgm
+
+; GCN: two{{$}}
+; SI:  v_cvt_f16_f32_e32 v[[B_F16:[0-9]+]], v[[B_F32]]
+; GCN: buffer_store_short v[[B_F16]]
+; GCN: s_endpgm
+define void @br_cc_f16(
+    half addrspace(1)* %r,
+    half addrspace(1)* %a,
+    half addrspace(1)* %b) {
+entry:
+  %a.val = load half, half addrspace(1)* %a
+  %b.val = load half, half addrspace(1)* %b
+  %fcmp = fcmp olt half %a.val, %b.val
+  br i1 %fcmp, label %one, label %two
+
+one:
+  store half %a.val, half addrspace(1)* %r
+  ret void
+
+two:
+  store half %b.val, half addrspace(1)* %r
+  ret void
+}
+
+; GCN-LABEL: {{^}}br_cc_f16_imm_a
+; GCN: v_mov_b32_e32 v[[A_F16:[0-9]+]], 0x380{{0|1}}{{$}}
+; SI:  v_cvt_f32_f16_e32 v[[A_F32:[0-9]+]], v[[A_F16]]
+; GCN: buffer_load_ushort v[[B_F16:[0-9]+]]
+
+; SI:  v_cvt_f32_f16_e32 v[[B_F32:[0-9]+]], v[[B_F16]]
+; SI:  v_cmp_ngt_f32_e32 vcc, v[[B_F32]], v[[A_F32]]
+; VI:  v_cmp_nle_f16_e32 vcc, v[[A_F16]], v[[B_F16]]
+; GCN: s_cbranch_vccnz
+
+; GCN: one{{$}}
+; VI:  v_mov_b32_e32 v[[A_F16:[0-9]+]], 0x3800{{$}}
+; GCN: buffer_store_short v[[A_F16]]
+; GCN: s_endpgm
+
+; GCN: two{{$}}
+; SI:  v_cvt_f16_f32_e32 v[[B_F16:[0-9]+]], v[[B_F32]]
+; GCN: buffer_store_short v[[B_F16]]
+; GCN: s_endpgm
+define void @br_cc_f16_imm_a(
+    half addrspace(1)* %r,
+    half addrspace(1)* %b) {
+entry:
+  %b.val = load half, half addrspace(1)* %b
+  %fcmp = fcmp olt half 0xH3800, %b.val
+  br i1 %fcmp, label %one, label %two
+
+one:
+  store half 0xH3800, half addrspace(1)* %r
+  ret void
+
+two:
+  store half %b.val, half addrspace(1)* %r
+  ret void
+}
+
+; GCN-LABEL: {{^}}br_cc_f16_imm_b
+; GCN: v_mov_b32_e32 v[[B_F16:[0-9]+]], {{0x37ff|0x3800}}{{$}}
+; SI:  v_cvt_f32_f16_e32 v[[B_F32:[0-9]+]], v[[B_F16]]
+; GCN: buffer_load_ushort v[[A_F16:[0-9]+]]
+
+; SI:  v_cvt_f32_f16_e32 v[[A_F32:[0-9]+]], v[[A_F16]]
+; SI:  v_cmp_nlt_f32_e32 vcc, v[[A_F32]], v[[B_F32]]
+; VI:  v_cmp_nge_f16_e32 vcc, v[[B_F16]], v[[A_F16]]
+; GCN: s_cbranch_vccnz
+
+; GCN: one{{$}}
+; SI:  v_cvt_f16_f32_e32 v[[A_F16:[0-9]+]], v[[A_F32]]
+; GCN: buffer_store_short v[[A_F16]]
+; GCN: s_endpgm
+
+; GCN: two{{$}}
+; VI:  v_mov_b32_e32 v[[B_F16:[0-9]+]], 0x3800{{$}}
+; GCN: buffer_store_short v[[B_F16]]
+; GCN: s_endpgm
+define void @br_cc_f16_imm_b(
+    half addrspace(1)* %r,
+    half addrspace(1)* %a) {
+entry:
+  %a.val = load half, half addrspace(1)* %a
+  %fcmp = fcmp olt half %a.val, 0xH3800
+  br i1 %fcmp, label %one, label %two
+
+one:
+  store half %a.val, half addrspace(1)* %r
+  ret void
+
+two:
+  store half 0xH3800, half addrspace(1)* %r
+  ret void
+}