diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/global-saddr-load.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/global-saddr-load.ll | 88 |
1 files changed, 60 insertions, 28 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/global-saddr-load.ll b/llvm/test/CodeGen/AMDGPU/global-saddr-load.ll index 1602e31d6147..4aba073d70cf 100644 --- a/llvm/test/CodeGen/AMDGPU/global-saddr-load.ll +++ b/llvm/test/CodeGen/AMDGPU/global-saddr-load.ll @@ -4393,21 +4393,37 @@ define amdgpu_ps <2 x half> @global_load_saddr_i16_d16hi_zero_hi(ptr addrspace(1 ; GCN-NEXT: v_mov_b32_e32 v0, v1 ; GCN-NEXT: ; return to shader part epilog ; -; GFX11-LABEL: global_load_saddr_i16_d16hi_zero_hi: -; GFX11: ; %bb.0: -; GFX11-NEXT: v_mov_b32_e32 v1, 0 -; GFX11-NEXT: global_load_d16_hi_b16 v1, v0, s[2:3] -; GFX11-NEXT: s_waitcnt vmcnt(0) -; GFX11-NEXT: v_mov_b32_e32 v0, v1 -; GFX11-NEXT: ; return to shader part epilog +; GFX11-TRUE16-LABEL: global_load_saddr_i16_d16hi_zero_hi: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX11-TRUE16-NEXT: global_load_d16_hi_b16 v1, v0, s[2:3] +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v0, v1 +; GFX11-TRUE16-NEXT: ; return to shader part epilog ; -; GFX12-SDAG-LABEL: global_load_saddr_i16_d16hi_zero_hi: -; GFX12-SDAG: ; %bb.0: -; GFX12-SDAG-NEXT: v_mov_b32_e32 v1, 0 -; GFX12-SDAG-NEXT: global_load_d16_hi_b16 v1, v0, s[2:3] -; GFX12-SDAG-NEXT: s_wait_loadcnt 0x0 -; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, v1 -; GFX12-SDAG-NEXT: ; return to shader part epilog +; GFX11-FAKE16-LABEL: global_load_saddr_i16_d16hi_zero_hi: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v1, 0 +; GFX11-FAKE16-NEXT: global_load_d16_hi_b16 v1, v0, s[2:3] +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, v1 +; GFX11-FAKE16-NEXT: ; return to shader part epilog +; +; GFX12-SDAG-TRUE16-LABEL: global_load_saddr_i16_d16hi_zero_hi: +; GFX12-SDAG-TRUE16: ; %bb.0: +; GFX12-SDAG-TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX12-SDAG-TRUE16-NEXT: global_load_d16_hi_b16 v1, v0, s[2:3] +; GFX12-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-SDAG-TRUE16-NEXT: v_mov_b32_e32 v0, v1 +; GFX12-SDAG-TRUE16-NEXT: ; return to shader part epilog +; +; GFX12-SDAG-FAKE16-LABEL: global_load_saddr_i16_d16hi_zero_hi: +; GFX12-SDAG-FAKE16: ; %bb.0: +; GFX12-SDAG-FAKE16-NEXT: v_mov_b32_e32 v1, 0 +; GFX12-SDAG-FAKE16-NEXT: global_load_d16_hi_b16 v1, v0, s[2:3] +; GFX12-SDAG-FAKE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-SDAG-FAKE16-NEXT: v_mov_b32_e32 v0, v1 +; GFX12-SDAG-FAKE16-NEXT: ; return to shader part epilog ; ; GFX12-GISEL-TRUE16-LABEL: global_load_saddr_i16_d16hi_zero_hi: ; GFX12-GISEL-TRUE16: ; %bb.0: @@ -4439,21 +4455,37 @@ define amdgpu_ps <2 x half> @global_load_saddr_i16_d16hi_zero_hi_immneg128(ptr a ; GCN-NEXT: v_mov_b32_e32 v0, v1 ; GCN-NEXT: ; return to shader part epilog ; -; GFX11-LABEL: global_load_saddr_i16_d16hi_zero_hi_immneg128: -; GFX11: ; %bb.0: -; GFX11-NEXT: v_mov_b32_e32 v1, 0 -; GFX11-NEXT: global_load_d16_hi_b16 v1, v0, s[2:3] offset:-128 -; GFX11-NEXT: s_waitcnt vmcnt(0) -; GFX11-NEXT: v_mov_b32_e32 v0, v1 -; GFX11-NEXT: ; return to shader part epilog +; GFX11-TRUE16-LABEL: global_load_saddr_i16_d16hi_zero_hi_immneg128: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX11-TRUE16-NEXT: global_load_d16_hi_b16 v1, v0, s[2:3] offset:-128 +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v0, v1 +; GFX11-TRUE16-NEXT: ; return to shader part epilog ; -; GFX12-SDAG-LABEL: global_load_saddr_i16_d16hi_zero_hi_immneg128: -; GFX12-SDAG: ; %bb.0: -; GFX12-SDAG-NEXT: v_mov_b32_e32 v1, 0 -; GFX12-SDAG-NEXT: global_load_d16_hi_b16 v1, v0, s[2:3] offset:-128 -; GFX12-SDAG-NEXT: s_wait_loadcnt 0x0 -; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, v1 -; GFX12-SDAG-NEXT: ; return to shader part epilog +; GFX11-FAKE16-LABEL: global_load_saddr_i16_d16hi_zero_hi_immneg128: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v1, 0 +; GFX11-FAKE16-NEXT: global_load_d16_hi_b16 v1, v0, s[2:3] offset:-128 +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v0, v1 +; GFX11-FAKE16-NEXT: ; return to shader part epilog +; +; GFX12-SDAG-TRUE16-LABEL: global_load_saddr_i16_d16hi_zero_hi_immneg128: +; GFX12-SDAG-TRUE16: ; %bb.0: +; GFX12-SDAG-TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX12-SDAG-TRUE16-NEXT: global_load_d16_hi_b16 v1, v0, s[2:3] offset:-128 +; GFX12-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-SDAG-TRUE16-NEXT: v_mov_b32_e32 v0, v1 +; GFX12-SDAG-TRUE16-NEXT: ; return to shader part epilog +; +; GFX12-SDAG-FAKE16-LABEL: global_load_saddr_i16_d16hi_zero_hi_immneg128: +; GFX12-SDAG-FAKE16: ; %bb.0: +; GFX12-SDAG-FAKE16-NEXT: v_mov_b32_e32 v1, 0 +; GFX12-SDAG-FAKE16-NEXT: global_load_d16_hi_b16 v1, v0, s[2:3] offset:-128 +; GFX12-SDAG-FAKE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-SDAG-FAKE16-NEXT: v_mov_b32_e32 v0, v1 +; GFX12-SDAG-FAKE16-NEXT: ; return to shader part epilog ; ; GFX12-GISEL-TRUE16-LABEL: global_load_saddr_i16_d16hi_zero_hi_immneg128: ; GFX12-GISEL-TRUE16: ; %bb.0: |