38 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 32; RUN: llc -mtriple amdgcn-amdhsa -mcpu=fiji -amdgpu-scalarize-global-loads < %s | FileCheck -check-prefix=GCN %s3 4declare i32 @llvm.amdgcn.workitem.id.x()5declare i32 @llvm.amdgcn.readfirstlane(i32)6 7define amdgpu_kernel void @readfirstlane_uniform(ptr addrspace(1) noalias nocapture readonly, ptr addrspace(1) noalias nocapture readonly) {8; GCN-LABEL: readfirstlane_uniform:9; GCN: ; %bb.0:10; GCN-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x011; GCN-NEXT: s_add_i32 s12, s12, s1712; GCN-NEXT: v_readfirstlane_b32 s4, v013; GCN-NEXT: s_mov_b32 s5, 014; GCN-NEXT: s_lshr_b32 flat_scratch_hi, s12, 815; GCN-NEXT: s_lshl_b64 s[4:5], s[4:5], 216; GCN-NEXT: s_waitcnt lgkmcnt(0)17; GCN-NEXT: s_add_u32 s0, s0, s418; GCN-NEXT: s_addc_u32 s1, s1, s519; GCN-NEXT: s_load_dword s4, s[0:1], 0x020; GCN-NEXT: s_add_u32 s0, s2, 4021; GCN-NEXT: s_addc_u32 s1, s3, 022; GCN-NEXT: v_mov_b32_e32 v0, s023; GCN-NEXT: s_mov_b32 flat_scratch_lo, s1324; GCN-NEXT: v_mov_b32_e32 v1, s125; GCN-NEXT: s_waitcnt lgkmcnt(0)26; GCN-NEXT: v_mov_b32_e32 v2, s427; GCN-NEXT: flat_store_dword v[0:1], v228; GCN-NEXT: s_endpgm29 %tid = tail call i32 @llvm.amdgcn.workitem.id.x()30 %scalar = tail call i32 @llvm.amdgcn.readfirstlane(i32 %tid)31 %idx = zext i32 %scalar to i6432 %gep0 = getelementptr inbounds float, ptr addrspace(1) %0, i64 %idx33 %val = load float, ptr addrspace(1) %gep0, align 434 %gep1 = getelementptr inbounds float, ptr addrspace(1) %1, i64 1035 store float %val, ptr addrspace(1) %gep1, align 436 ret void37}38