[llvm-branch-commits] [llvm] AMDGPU/GlobalISel: Add tests for missing readanylane combines (PR #145910)

[Petar Avramovic via llvm-branch-commits]

================
@@ -0,0 +1,166 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -global-isel -mtriple=amdgcn-amd-amdpal -mcpu=gfx1010 -new-reg-bank-select < %s | FileCheck %s
+
+define amdgpu_ps void @readanylane_to_virtual_vgpr(ptr addrspace(1) inreg %ptr0, ptr addrspace(1) inreg %ptr1) {
+; CHECK-LABEL: readanylane_to_virtual_vgpr:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    v_mov_b32_e32 v0, 0
+; CHECK-NEXT:    global_load_dword v1, v0, s[0:1] glc dlc
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    global_store_dword v0, v1, s[2:3]
+; CHECK-NEXT:    s_endpgm
+  %load = load volatile float, ptr addrspace(1) %ptr0
+  store float %load, ptr addrspace(1) %ptr1
+  ret void
+}
+
+define amdgpu_ps float @readanylane_to_physical_vgpr(ptr addrspace(1) inreg %ptr) {
+; CHECK-LABEL: readanylane_to_physical_vgpr:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    v_mov_b32_e32 v0, 0
+; CHECK-NEXT:    global_load_dword v0, v0, s[0:1] glc dlc
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    v_readfirstlane_b32 s0, v0
+; CHECK-NEXT:    v_mov_b32_e32 v0, s0
+; CHECK-NEXT:    ; return to shader part epilog
+  %load = load volatile float, ptr addrspace(1) %ptr
+  ret float %load
+}
+
+define amdgpu_ps void @readanylane_to_bitcast_to_virtual_vgpr(ptr addrspace(1) inreg %ptr0, ptr addrspace(1) inreg %ptr1) {
+; CHECK-LABEL: readanylane_to_bitcast_to_virtual_vgpr:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    v_mov_b32_e32 v0, 0
+; CHECK-NEXT:    global_load_dword v1, v0, s[0:1] glc dlc
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    v_readfirstlane_b32 s0, v1
+; CHECK-NEXT:    v_mov_b32_e32 v1, s0
+; CHECK-NEXT:    global_store_dword v0, v1, s[2:3]
+; CHECK-NEXT:    s_endpgm
+  %load = load volatile <2 x i16>, ptr addrspace(1) %ptr0
+  %bitcast = bitcast <2 x i16> %load to i32
+  store i32 %bitcast, ptr addrspace(1) %ptr1
+  ret void
+}
+
+define amdgpu_ps float @readanylane_to_bitcast_to_physical_vgpr(ptr addrspace(1) inreg %ptr0, ptr addrspace(1) inreg %ptr1) {
+; CHECK-LABEL: readanylane_to_bitcast_to_physical_vgpr:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    v_mov_b32_e32 v0, 0
+; CHECK-NEXT:    global_load_dword v0, v0, s[0:1] glc dlc
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    v_readfirstlane_b32 s0, v0
+; CHECK-NEXT:    v_mov_b32_e32 v0, s0
+; CHECK-NEXT:    ; return to shader part epilog
+  %load = load volatile <2 x i16>, ptr addrspace(1) %ptr0
+  %bitcast = bitcast <2 x i16> %load to float
+  ret float %bitcast
+}
+
+define amdgpu_ps void @unmerge_readanylane_merge_to_virtual_vgpr(ptr addrspace(1) inreg %ptr0, ptr addrspace(1) inreg %ptr1) {
+; CHECK-LABEL: unmerge_readanylane_merge_to_virtual_vgpr:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    v_mov_b32_e32 v2, 0
+; CHECK-NEXT:    global_load_dwordx2 v[0:1], v2, s[0:1] glc dlc
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    v_readfirstlane_b32 s0, v0
+; CHECK-NEXT:    v_readfirstlane_b32 s1, v1
+; CHECK-NEXT:    v_mov_b32_e32 v0, s0
+; CHECK-NEXT:    v_mov_b32_e32 v1, s1
+; CHECK-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
+; CHECK-NEXT:    s_endpgm
+  %load = load volatile i64, ptr addrspace(1) %ptr0
+  store i64 %load, ptr addrspace(1) %ptr1
+  ret void
+}
+
+;define amdgpu_ps double @unmerge_readanylane_merge_to_physical_vgpr(ptr addrspace(1) inreg %ptr0, ptr addrspace(1) inreg %ptr1) {
+;  %load = load volatile double, ptr addrspace(1) %ptr0
+;  ret double %load
+;}
+
+define amdgpu_ps void @unmerge_readanylane_merge_bitcast_to_virtual_vgpr(ptr addrspace(1) inreg %ptr0, ptr addrspace(1) inreg %ptr1) {
+; CHECK-LABEL: unmerge_readanylane_merge_bitcast_to_virtual_vgpr:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    v_mov_b32_e32 v2, 0
+; CHECK-NEXT:    global_load_dwordx2 v[0:1], v2, s[0:1] glc dlc
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    v_readfirstlane_b32 s0, v0
+; CHECK-NEXT:    v_readfirstlane_b32 s1, v1
+; CHECK-NEXT:    v_mov_b32_e32 v0, s0
+; CHECK-NEXT:    v_mov_b32_e32 v1, s1
+; CHECK-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
+; CHECK-NEXT:    s_endpgm
+  %load = load volatile <2 x i32>, ptr addrspace(1) %ptr0
+  %bitcast = bitcast <2 x i32> %load to double
+  store double %bitcast, ptr addrspace(1) %ptr1
+  ret void
+}
+
+;define amdgpu_ps double @unmerge_readanylane_merge_bitcast_to_physical_vgpr(ptr addrspace(1) inreg %ptr0, ptr addrspace(1) inreg %ptr1) {
+;  %load = load volatile <2 x i32>, ptr addrspace(1) %ptr0
+;  %bitcast = bitcast <2 x i32> %load to double
+;  ret double %bitcast
+;}
----------------
petar-avramovic wrote:

There is no combine happening in commented out tests. It is path where there is copy to physical vgpr, but only one that exists is calling convention with float return, and there is none for larger physical vgprs (they return to sgpr) so I edited mir test with `SI_RETURN_TO_EPILOG implicit $vgpr0_vgpr1` to test that combine works.

https://github.com/llvm/llvm-project/pull/145910