diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/llvm.round.f64.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/llvm.round.f64.ll | 20 |
1 files changed, 9 insertions, 11 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.round.f64.ll b/llvm/test/CodeGen/AMDGPU/llvm.round.f64.ll index 8bb7274..76b97e8 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.round.f64.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.round.f64.ll @@ -78,7 +78,6 @@ define amdgpu_kernel void @v_round_f64(ptr addrspace(1) %out, ptr addrspace(1) % ; SI-NEXT: buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64 ; SI-NEXT: s_mov_b32 s2, -1 ; SI-NEXT: s_mov_b32 s3, 0xfffff -; SI-NEXT: v_mov_b32_e32 v8, 0x3ff00000 ; SI-NEXT: s_waitcnt vmcnt(0) ; SI-NEXT: v_bfe_u32 v4, v3, 20, 11 ; SI-NEXT: v_add_i32_e32 v6, vcc, 0xfffffc01, v4 @@ -93,12 +92,12 @@ define amdgpu_kernel void @v_round_f64(ptr addrspace(1) %out, ptr addrspace(1) % ; SI-NEXT: v_cndmask_b32_e32 v5, v5, v3, vcc ; SI-NEXT: v_cndmask_b32_e32 v4, v4, v2, vcc ; SI-NEXT: v_add_f64 v[6:7], v[2:3], -v[4:5] -; SI-NEXT: s_brev_b32 s2, -2 +; SI-NEXT: v_mov_b32_e32 v2, 0x3ff00000 ; SI-NEXT: v_cmp_ge_f64_e64 vcc, |v[6:7]|, 0.5 -; SI-NEXT: v_cndmask_b32_e32 v2, 0, v8, vcc -; SI-NEXT: v_bfi_b32 v3, s2, v2, v3 -; SI-NEXT: v_mov_b32_e32 v2, v1 -; SI-NEXT: v_add_f64 v[2:3], v[4:5], v[2:3] +; SI-NEXT: s_brev_b32 s2, -2 +; SI-NEXT: v_cndmask_b32_e32 v2, 0, v2, vcc +; SI-NEXT: v_bfi_b32 v2, s2, v2, v3 +; SI-NEXT: v_add_f64 v[2:3], v[4:5], v[1:2] ; SI-NEXT: s_mov_b64 s[2:3], s[6:7] ; SI-NEXT: buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64 ; SI-NEXT: s_endpgm @@ -113,16 +112,15 @@ define amdgpu_kernel void @v_round_f64(ptr addrspace(1) %out, ptr addrspace(1) % ; CI-NEXT: s_waitcnt lgkmcnt(0) ; CI-NEXT: s_mov_b64 s[4:5], s[2:3] ; CI-NEXT: buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64 -; CI-NEXT: v_mov_b32_e32 v8, 0x3ff00000 ; CI-NEXT: s_brev_b32 s2, -2 ; CI-NEXT: s_waitcnt vmcnt(0) ; CI-NEXT: v_trunc_f64_e32 v[4:5], v[2:3] ; CI-NEXT: v_add_f64 v[6:7], v[2:3], -v[4:5] +; CI-NEXT: v_mov_b32_e32 v2, 0x3ff00000 ; CI-NEXT: v_cmp_ge_f64_e64 vcc, |v[6:7]|, 0.5 -; CI-NEXT: v_cndmask_b32_e32 v2, 0, v8, vcc -; CI-NEXT: v_bfi_b32 v3, s2, v2, v3 -; CI-NEXT: v_mov_b32_e32 v2, v1 -; CI-NEXT: v_add_f64 v[2:3], v[4:5], v[2:3] +; CI-NEXT: v_cndmask_b32_e32 v2, 0, v2, vcc +; CI-NEXT: v_bfi_b32 v2, s2, v2, v3 +; CI-NEXT: v_add_f64 v[2:3], v[4:5], v[1:2] ; CI-NEXT: s_mov_b64 s[2:3], s[6:7] ; CI-NEXT: buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64 ; CI-NEXT: s_endpgm |
