52 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9 %s3; RUN: llc -mtriple=amdgcn -mcpu=gfx900 < %s -debugify-and-strip-all-safe | FileCheck -check-prefixes=GFX9 %s4 5define amdgpu_kernel void @sdwa_test() local_unnamed_addr #0 {6; GFX9-LABEL: sdwa_test:7; GFX9: ; %bb.0: ; %bb8; GFX9-NEXT: v_add_u32_e32 v1, 20, v09; GFX9-NEXT: v_and_b32_e32 v1, 0xff, v110; GFX9-NEXT: v_add3_u32 v0, v0, v1, 1011; GFX9-NEXT: v_mov_b32_e32 v1, 012; GFX9-NEXT: global_store_dwordx2 v[0:1], v[0:1], off13; GFX9-NEXT: s_endpgm14bb:15 %tid = tail call i32 @llvm.amdgcn.workitem.id.x()16 %v0 = add i32 %tid, 1017 %v1 = add i32 %tid, 2018 %v2 = zext i32 %v0 to i6419 %v3 = zext i32 %v1 to i6420 %v.t = and i64 %v3, 25521 %v4 = add i64 %v2, %v.t22 store i64 %v4, ptr addrspace(1) poison23 ret void24}25 26define amdgpu_kernel void @test_add_co_sdwa(ptr addrspace(1) %arg, ptr addrspace(1) %arg1) #0 {27; GFX9-LABEL: test_add_co_sdwa:28; GFX9: ; %bb.0: ; %bb29; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x2430; GFX9-NEXT: v_lshlrev_b32_e32 v2, 2, v031; GFX9-NEXT: v_lshlrev_b32_e32 v3, 3, v032; GFX9-NEXT: s_waitcnt lgkmcnt(0)33; GFX9-NEXT: global_load_dword v4, v2, s[2:3]34; GFX9-NEXT: global_load_dwordx2 v[0:1], v3, s[0:1]35; GFX9-NEXT: s_waitcnt vmcnt(0)36; GFX9-NEXT: v_add_co_u32_sdwa v0, vcc, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_037; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc38; GFX9-NEXT: global_store_dwordx2 v3, v[0:1], s[0:1]39; GFX9-NEXT: s_endpgm40bb:41 %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()42 %tmp3 = getelementptr inbounds i32, ptr addrspace(1) %arg1, i32 %tmp43 %tmp4 = load i32, ptr addrspace(1) %tmp3, align 444 %tmp5 = and i32 %tmp4, 25545 %tmp6 = zext i32 %tmp5 to i6446 %tmp7 = getelementptr inbounds i64, ptr addrspace(1) %arg, i32 %tmp47 %tmp8 = load i64, ptr addrspace(1) %tmp7, align 848 %tmp9 = add nsw i64 %tmp8, %tmp649 store i64 %tmp9, ptr addrspace(1) %tmp7, align 850 ret void51}52