65 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx908 -amdgpu-atomic-optimizer-strategy=None < %s | FileCheck --check-prefix=GCN %s3 4; Test using saddr addressing mode of global_* flat atomic instructions.5 6; --------------------------------------------------------------------------------7; amdgcn global atomic fadd8; --------------------------------------------------------------------------------9 10define amdgpu_ps void @global_fadd_saddr_f32_nortn(ptr addrspace(1) inreg %sbase, i32 %voffset, float %data) {11; GCN-LABEL: global_fadd_saddr_f32_nortn:12; GCN: ; %bb.0:13; GCN-NEXT: global_atomic_add_f32 v0, v1, s[2:3]14; GCN-NEXT: s_waitcnt vmcnt(0)15; GCN-NEXT: buffer_wbinvl116; GCN-NEXT: s_endpgm17 %zext.offset = zext i32 %voffset to i6418 %gep0 = getelementptr inbounds i8, ptr addrspace(1) %sbase, i64 %zext.offset19 %ret = atomicrmw fadd ptr addrspace(1) %gep0, float %data syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0, !amdgpu.ignore.denormal.mode !020 ret void21}22 23define amdgpu_ps void @global_fadd_saddr_f32_nortn_neg128(ptr addrspace(1) inreg %sbase, i32 %voffset, float %data) {24; GCN-LABEL: global_fadd_saddr_f32_nortn_neg128:25; GCN: ; %bb.0:26; GCN-NEXT: global_atomic_add_f32 v0, v1, s[2:3] offset:-12827; GCN-NEXT: s_waitcnt vmcnt(0)28; GCN-NEXT: buffer_wbinvl129; GCN-NEXT: s_endpgm30 %zext.offset = zext i32 %voffset to i6431 %gep0 = getelementptr inbounds i8, ptr addrspace(1) %sbase, i64 %zext.offset32 %gep1 = getelementptr inbounds i8, ptr addrspace(1) %gep0, i64 -12833 %ret = atomicrmw fadd ptr addrspace(1) %gep1, float %data syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0, !amdgpu.ignore.denormal.mode !034 ret void35}36 37define amdgpu_ps void @global_fadd_saddr_v2f16_nortn(ptr addrspace(1) inreg %sbase, i32 %voffset, <2 x half> %data) {38; GCN-LABEL: global_fadd_saddr_v2f16_nortn:39; GCN: ; %bb.0:40; GCN-NEXT: global_atomic_pk_add_f16 v0, v1, s[2:3]41; GCN-NEXT: s_waitcnt vmcnt(0)42; GCN-NEXT: buffer_wbinvl143; GCN-NEXT: s_endpgm44 %zext.offset = zext i32 %voffset to i6445 %gep0 = getelementptr inbounds i8, ptr addrspace(1) %sbase, i64 %zext.offset46 %ret = atomicrmw fadd ptr addrspace(1) %gep0, <2 x half> %data syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !047 ret void48}49 50define amdgpu_ps void @global_fadd_saddr_v2f16_nortn_neg128(ptr addrspace(1) inreg %sbase, i32 %voffset, <2 x half> %data) {51; GCN-LABEL: global_fadd_saddr_v2f16_nortn_neg128:52; GCN: ; %bb.0:53; GCN-NEXT: global_atomic_pk_add_f16 v0, v1, s[2:3] offset:-12854; GCN-NEXT: s_waitcnt vmcnt(0)55; GCN-NEXT: buffer_wbinvl156; GCN-NEXT: s_endpgm57 %zext.offset = zext i32 %voffset to i6458 %gep0 = getelementptr inbounds i8, ptr addrspace(1) %sbase, i64 %zext.offset59 %gep1 = getelementptr inbounds i8, ptr addrspace(1) %gep0, i64 -12860 %ret = atomicrmw fadd ptr addrspace(1) %gep1, <2 x half> %data syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !061 ret void62}63 64!0 = !{}65