diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll | 302 |
1 files changed, 152 insertions, 150 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll b/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll index ab867b0..6f1675e 100644 --- a/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll +++ b/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fmax.ll @@ -4228,55 +4228,56 @@ define bfloat @buffer_fat_ptr_agent_atomic_fmax_ret_bf16__offset__amdgpu_no_fine ; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0 ; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0 ; GFX12-TRUE16-NEXT: s_addk_co_i32 s16, 0x200 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v5, 16, v0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_and_b32 s4, s16, -4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, s4 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v5, s4 ; GFX12-TRUE16-NEXT: s_and_b32 s4, s16, 3 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_lshl_b32 s4, s4, 3 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_lshl_b32 s5, 0xffff, s4 -; GFX12-TRUE16-NEXT: buffer_load_b32 v1, v4, s[0:3], null offen +; GFX12-TRUE16-NEXT: buffer_load_b32 v2, v5, s[0:3], null offen ; GFX12-TRUE16-NEXT: s_not_b32 s6, s5 ; GFX12-TRUE16-NEXT: s_mov_b32 s5, 0 ; GFX12-TRUE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v0, s4, v1 +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, s4, v2 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.h, v0.l ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v0, v0, v5 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX12-TRUE16-NEXT: v_bfe_u32 v2, v0, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v3, 0x400000, v0 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v0 -; GFX12-TRUE16-NEXT: v_add3_u32 v2, v2, v0, 0x7fff +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 16, v1 +; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v1, v1, v3 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.h, v3.l +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX12-TRUE16-NEXT: v_bfe_u32 v4, v1, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v1 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX12-TRUE16-NEXT: v_add3_u32 v4, v4, v1, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc_lo -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v2.h, 0 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v2.l, v0.h +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v1, v4, v6, vcc_lo +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.l, v1.h ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v0, s4, v2 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v1, s4, v3 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe -; GFX12-TRUE16-NEXT: v_and_or_b32 v0, v1, s6, v0 +; GFX12-TRUE16-NEXT: v_and_or_b32 v1, v2, s6, v1 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0 -; GFX12-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[2:3], v4, s[0:3], null offen th:TH_ATOMIC_RETURN +; GFX12-TRUE16-NEXT: v_dual_mov_b32 v4, v2 :: v_dual_mov_b32 v3, v1 +; GFX12-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[3:4], v5, s[0:3], null offen th:TH_ATOMIC_RETURN ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_DEV -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v3 ; GFX12-TRUE16-NEXT: s_or_b32 s5, vcc_lo, s5 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s5 ; GFX12-TRUE16-NEXT: s_cbranch_execnz .LBB13_1 ; GFX12-TRUE16-NEXT: ; %bb.2: ; %atomicrmw.end ; GFX12-TRUE16-NEXT: s_or_b32 exec_lo, exec_lo, s5 -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v0, s4, v2 +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v0, s4, v3 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 ; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31] ; @@ -4385,53 +4386,54 @@ define bfloat @buffer_fat_ptr_agent_atomic_fmax_ret_bf16__offset__amdgpu_no_fine ; GFX11-TRUE16: ; %bb.0: ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-TRUE16-NEXT: s_addk_i32 s16, 0x200 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v5, 16, v0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_b32 s4, s16, -4 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1) -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, s4 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v5, s4 ; GFX11-TRUE16-NEXT: s_and_b32 s4, s16, 3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_lshl_b32 s4, s4, 3 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_lshl_b32 s5, 0xffff, s4 -; GFX11-TRUE16-NEXT: buffer_load_b32 v1, v4, s[0:3], 0 offen +; GFX11-TRUE16-NEXT: buffer_load_b32 v2, v5, s[0:3], 0 offen ; GFX11-TRUE16-NEXT: s_not_b32 s6, s5 ; GFX11-TRUE16-NEXT: s_mov_b32 s5, 0 ; GFX11-TRUE16-NEXT: .p2align 6 ; GFX11-TRUE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v0, s4, v1 +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v1, s4, v2 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, v0.l ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 16, v1 +; GFX11-TRUE16-NEXT: v_max_f32_e32 v1, v1, v3 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, v3.l +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-TRUE16-NEXT: v_bfe_u32 v4, v1, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v1 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX11-TRUE16-NEXT: v_add3_u32 v4, v4, v1, 0x7fff ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX11-TRUE16-NEXT: v_max_f32_e32 v0, v0, v5 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_bfe_u32 v2, v0, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v3, 0x400000, v0 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v0 -; GFX11-TRUE16-NEXT: v_add3_u32 v2, v2, v0, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc_lo -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v2.h, 0 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v2.l, v0.h +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v1, v4, v6, vcc_lo +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, v1.h ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v0, s4, v2 -; GFX11-TRUE16-NEXT: v_and_or_b32 v0, v1, s6, v0 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v1, s4, v3 +; GFX11-TRUE16-NEXT: v_and_or_b32 v1, v2, s6, v1 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_dual_mov_b32 v3, v1 :: v_dual_mov_b32 v2, v0 -; GFX11-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[2:3], v4, s[0:3], 0 offen glc +; GFX11-TRUE16-NEXT: v_dual_mov_b32 v4, v2 :: v_dual_mov_b32 v3, v1 +; GFX11-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[3:4], v5, s[0:3], 0 offen glc ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl1_inv ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v3 ; GFX11-TRUE16-NEXT: s_or_b32 s5, vcc_lo, s5 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s5 ; GFX11-TRUE16-NEXT: s_cbranch_execnz .LBB13_1 ; GFX11-TRUE16-NEXT: ; %bb.2: ; %atomicrmw.end ; GFX11-TRUE16-NEXT: s_or_b32 exec_lo, exec_lo, s5 -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v0, s4, v2 +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v0, s4, v3 ; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-FAKE16-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_bf16__offset__amdgpu_no_fine_grained_memory: @@ -4749,48 +4751,49 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_bf16__offset__amdgpu_no_fine ; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0 ; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0 ; GFX12-TRUE16-NEXT: s_addk_co_i32 s16, 0x200 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_and_b32 s4, s16, -4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, s4 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, s4 ; GFX12-TRUE16-NEXT: s_and_b32 s4, s16, 3 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_lshl_b32 s4, s4, 3 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_lshl_b32 s5, 0xffff, s4 -; GFX12-TRUE16-NEXT: buffer_load_b32 v1, v2, s[0:3], null offen +; GFX12-TRUE16-NEXT: buffer_load_b32 v2, v3, s[0:3], null offen ; GFX12-TRUE16-NEXT: s_not_b32 s6, s5 ; GFX12-TRUE16-NEXT: s_mov_b32 s5, 0 ; GFX12-TRUE16-NEXT: .LBB14_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v0, s4, v1 +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v1, s4, v2 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.l, 0 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.h, v0.l ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v0, v0, v3 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX12-TRUE16-NEXT: v_bfe_u32 v4, v0, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v5, 0x400000, v0 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v0 -; GFX12-TRUE16-NEXT: v_add3_u32 v4, v4, v0, 0x7fff +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 16, v1 +; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v1, v1, v4 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.h, v4.l +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v1, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v1 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v1, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v0, v4, v5, vcc_lo -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.l, v0.h +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v1, v5, v6, vcc_lo +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.l, v1.h ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v0, s4, v4 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v1, s4, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe -; GFX12-TRUE16-NEXT: v_and_or_b32 v0, v1, s6, v0 +; GFX12-TRUE16-NEXT: v_and_or_b32 v1, v2, s6, v1 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v4, v0 -; GFX12-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v2, s[0:3], null offen th:TH_ATOMIC_RETURN +; GFX12-TRUE16-NEXT: v_dual_mov_b32 v5, v2 :: v_dual_mov_b32 v4, v1 +; GFX12-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v3, s[0:3], null offen th:TH_ATOMIC_RETURN ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_DEV -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v1 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v1, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v4 ; GFX12-TRUE16-NEXT: s_or_b32 s5, vcc_lo, s5 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s5 @@ -4903,46 +4906,47 @@ define void @buffer_fat_ptr_agent_atomic_fmax_noret_bf16__offset__amdgpu_no_fine ; GFX11-TRUE16: ; %bb.0: ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-TRUE16-NEXT: s_addk_i32 s16, 0x200 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_b32 s4, s16, -4 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1) -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, s4 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, s4 ; GFX11-TRUE16-NEXT: s_and_b32 s4, s16, 3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_lshl_b32 s4, s4, 3 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_lshl_b32 s5, 0xffff, s4 -; GFX11-TRUE16-NEXT: buffer_load_b32 v1, v2, s[0:3], 0 offen +; GFX11-TRUE16-NEXT: buffer_load_b32 v2, v3, s[0:3], 0 offen ; GFX11-TRUE16-NEXT: s_not_b32 s6, s5 ; GFX11-TRUE16-NEXT: s_mov_b32 s5, 0 ; GFX11-TRUE16-NEXT: .p2align 6 ; GFX11-TRUE16-NEXT: .LBB14_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v0, s4, v1 +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v1, s4, v2 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.l, 0 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.h, v0.l ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 16, v1 +; GFX11-TRUE16-NEXT: v_max_f32_e32 v1, v1, v4 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.h, v4.l +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v1, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v1 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v1, 0x7fff ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX11-TRUE16-NEXT: v_max_f32_e32 v0, v0, v3 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_bfe_u32 v4, v0, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v5, 0x400000, v0 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v0 -; GFX11-TRUE16-NEXT: v_add3_u32 v4, v4, v0, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v0, v4, v5, vcc_lo -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.l, v0.h +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v1, v5, v6, vcc_lo +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.l, v1.h ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v0, s4, v4 -; GFX11-TRUE16-NEXT: v_and_or_b32 v0, v1, s6, v0 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v1, s4, v4 +; GFX11-TRUE16-NEXT: v_and_or_b32 v1, v2, s6, v1 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v4, v0 -; GFX11-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v2, s[0:3], 0 offen glc +; GFX11-TRUE16-NEXT: v_dual_mov_b32 v5, v2 :: v_dual_mov_b32 v4, v1 +; GFX11-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v3, s[0:3], 0 offen glc ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl1_inv ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v1 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s5, vcc_lo, s5 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s5 @@ -5256,15 +5260,15 @@ define bfloat @buffer_fat_ptr_agent_atomic_fmax_ret_bf16__offset__waterfall__amd ; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0 ; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0 ; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0 -; GFX12-TRUE16-NEXT: v_add_nc_u32_e32 v4, 0x200, v4 +; GFX12-TRUE16-NEXT: v_add_nc_u32_e32 v6, 0x200, v4 ; GFX12-TRUE16-NEXT: s_mov_b32 s1, exec_lo ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_and_b32_e32 v6, 3, v4 -; GFX12-TRUE16-NEXT: v_and_b32_e32 v8, -4, v4 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v7, 3, v6 +; GFX12-TRUE16-NEXT: v_and_b32_e32 v4, 3, v6 +; GFX12-TRUE16-NEXT: v_and_b32_e32 v10, -4, v6 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 3, v4 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v6, v7, 0xffff -; GFX12-TRUE16-NEXT: v_not_b32_e32 v9, v6 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v7, v4, 0xffff +; GFX12-TRUE16-NEXT: v_not_b32_e32 v11, v7 ; GFX12-TRUE16-NEXT: .LBB15_1: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: v_readfirstlane_b32 s4, v0 ; GFX12-TRUE16-NEXT: v_readfirstlane_b32 s5, v1 @@ -5278,39 +5282,38 @@ define bfloat @buffer_fat_ptr_agent_atomic_fmax_ret_bf16__offset__waterfall__amd ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_and_saveexec_b32 s0, s0 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 -; GFX12-TRUE16-NEXT: buffer_load_b32 v6, v8, s[4:7], null offen +; GFX12-TRUE16-NEXT: buffer_load_b32 v7, v10, s[4:7], null offen ; GFX12-TRUE16-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX12-TRUE16-NEXT: s_cbranch_execnz .LBB15_1 ; GFX12-TRUE16-NEXT: ; %bb.2: ; GFX12-TRUE16-NEXT: s_mov_b32 exec_lo, s1 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v10, 16, v5 ; GFX12-TRUE16-NEXT: s_mov_b32 s1, 0 ; GFX12-TRUE16-NEXT: .LBB15_3: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: ; Child Loop BB15_4 Depth 2 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v7, v6 +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v6, v4, v7 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v8.l, 0 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v8.h, v5.l ; GFX12-TRUE16-NEXT: s_mov_b32 s2, exec_lo ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v6, 16, v6 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v4, v4, v10 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v11, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v6, v6, v8 +; GFX12-TRUE16-NEXT: v_bfe_u32 v9, v6, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v12, 0x400000, v6 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v6, v6 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add3_u32 v9, v9, v6, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v11, vcc_lo -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v6, v9, v12, vcc_lo +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v6.l, v6.h +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v6.h, v8.l +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v6, v4, v6 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v7, v5 -; GFX12-TRUE16-NEXT: v_and_or_b32 v5, v6, v9, v4 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v5 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v5, v6 +; GFX12-TRUE16-NEXT: v_and_or_b32 v6, v7, v11, v6 +; GFX12-TRUE16-NEXT: v_dual_mov_b32 v9, v7 :: v_dual_mov_b32 v8, v6 ; GFX12-TRUE16-NEXT: .LBB15_4: ; Parent Loop BB15_3 Depth=1 ; GFX12-TRUE16-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX12-TRUE16-NEXT: v_readfirstlane_b32 s4, v0 @@ -5325,14 +5328,14 @@ define bfloat @buffer_fat_ptr_agent_atomic_fmax_ret_bf16__offset__waterfall__amd ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_and_saveexec_b32 s0, s0 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 -; GFX12-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v8, s[4:7], null offen th:TH_ATOMIC_RETURN +; GFX12-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[8:9], v10, s[4:7], null offen th:TH_ATOMIC_RETURN ; GFX12-TRUE16-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX12-TRUE16-NEXT: s_cbranch_execnz .LBB15_4 ; GFX12-TRUE16-NEXT: ; %bb.5: ; in Loop: Header=BB15_3 Depth=1 ; GFX12-TRUE16-NEXT: s_mov_b32 exec_lo, s2 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v6 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v6, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v8, v7 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v7, v8 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_DEV ; GFX12-TRUE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -5340,7 +5343,7 @@ define bfloat @buffer_fat_ptr_agent_atomic_fmax_ret_bf16__offset__waterfall__amd ; GFX12-TRUE16-NEXT: s_cbranch_execnz .LBB15_3 ; GFX12-TRUE16-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX12-TRUE16-NEXT: s_or_b32 exec_lo, exec_lo, s1 -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v0, v7, v4 +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v0, v4, v8 ; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 ; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31] ; @@ -5516,16 +5519,16 @@ define bfloat @buffer_fat_ptr_agent_atomic_fmax_ret_bf16__offset__waterfall__amd ; GFX11-TRUE16-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_bf16__offset__waterfall__amdgpu_no_fine_grained_memory: ; GFX11-TRUE16: ; %bb.0: ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_add_nc_u32_e32 v4, 0x200, v4 +; GFX11-TRUE16-NEXT: v_add_nc_u32_e32 v6, 0x200, v4 ; GFX11-TRUE16-NEXT: s_mov_b32 s1, 0 ; GFX11-TRUE16-NEXT: s_mov_b32 s2, exec_lo ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_and_b32_e32 v6, 3, v4 -; GFX11-TRUE16-NEXT: v_and_b32_e32 v8, -4, v4 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v7, 3, v6 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v4, 3, v6 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v10, -4, v6 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 3, v4 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v6, v7, 0xffff -; GFX11-TRUE16-NEXT: v_not_b32_e32 v9, v6 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v7, v4, 0xffff +; GFX11-TRUE16-NEXT: v_not_b32_e32 v11, v7 ; GFX11-TRUE16-NEXT: .LBB15_1: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: v_readfirstlane_b32 s4, v0 ; GFX11-TRUE16-NEXT: v_readfirstlane_b32 s5, v1 @@ -5537,39 +5540,38 @@ define bfloat @buffer_fat_ptr_agent_atomic_fmax_ret_bf16__offset__waterfall__amd ; GFX11-TRUE16-NEXT: s_and_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_saveexec_b32 s0, s0 -; GFX11-TRUE16-NEXT: buffer_load_b32 v6, v8, s[4:7], 0 offen +; GFX11-TRUE16-NEXT: buffer_load_b32 v7, v10, s[4:7], 0 offen ; GFX11-TRUE16-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX11-TRUE16-NEXT: s_cbranch_execnz .LBB15_1 ; GFX11-TRUE16-NEXT: ; %bb.2: ; GFX11-TRUE16-NEXT: s_mov_b32 exec_lo, s2 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v10, 16, v5 ; GFX11-TRUE16-NEXT: s_set_inst_prefetch_distance 0x1 ; GFX11-TRUE16-NEXT: .p2align 6 ; GFX11-TRUE16-NEXT: .LBB15_3: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: ; Child Loop BB15_4 Depth 2 ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v7, v6 +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v6, v4, v7 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v8.l, 0 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v8.h, v5.l ; GFX11-TRUE16-NEXT: s_mov_b32 s2, exec_lo ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v6, 16, v6 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX11-TRUE16-NEXT: v_max_f32_e32 v4, v4, v10 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v11, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v11, vcc_lo -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX11-TRUE16-NEXT: v_max_f32_e32 v6, v6, v8 +; GFX11-TRUE16-NEXT: v_bfe_u32 v9, v6, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v12, 0x400000, v6 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v6, v6 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add3_u32 v9, v9, v6, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v6, v9, v12, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v6.l, v6.h +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v6.h, v8.l +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v6, v4, v6 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v7, v5 -; GFX11-TRUE16-NEXT: v_and_or_b32 v5, v6, v9, v4 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v5 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v5, v6 +; GFX11-TRUE16-NEXT: v_and_or_b32 v6, v7, v11, v6 +; GFX11-TRUE16-NEXT: v_dual_mov_b32 v9, v7 :: v_dual_mov_b32 v8, v6 ; GFX11-TRUE16-NEXT: .LBB15_4: ; Parent Loop BB15_3 Depth=1 ; GFX11-TRUE16-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX11-TRUE16-NEXT: v_readfirstlane_b32 s4, v0 @@ -5583,14 +5585,14 @@ define bfloat @buffer_fat_ptr_agent_atomic_fmax_ret_bf16__offset__waterfall__amd ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_saveexec_b32 s0, s0 ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) -; GFX11-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[4:5], v8, s[4:7], 0 offen glc +; GFX11-TRUE16-NEXT: buffer_atomic_cmpswap_b32 v[8:9], v10, s[4:7], 0 offen glc ; GFX11-TRUE16-NEXT: s_xor_b32 exec_lo, exec_lo, s0 ; GFX11-TRUE16-NEXT: s_cbranch_execnz .LBB15_4 ; GFX11-TRUE16-NEXT: ; %bb.5: ; in Loop: Header=BB15_3 Depth=1 ; GFX11-TRUE16-NEXT: s_mov_b32 exec_lo, s2 ; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v6 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v6, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v8, v7 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v7, v8 ; GFX11-TRUE16-NEXT: buffer_gl1_inv ; GFX11-TRUE16-NEXT: buffer_gl0_inv ; GFX11-TRUE16-NEXT: s_or_b32 s1, vcc_lo, s1 @@ -5600,7 +5602,7 @@ define bfloat @buffer_fat_ptr_agent_atomic_fmax_ret_bf16__offset__waterfall__amd ; GFX11-TRUE16-NEXT: ; %bb.6: ; %atomicrmw.end ; GFX11-TRUE16-NEXT: s_set_inst_prefetch_distance 0x2 ; GFX11-TRUE16-NEXT: s_or_b32 exec_lo, exec_lo, s1 -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v0, v7, v4 +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v0, v4, v8 ; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-FAKE16-LABEL: buffer_fat_ptr_agent_atomic_fmax_ret_bf16__offset__waterfall__amdgpu_no_fine_grained_memory: |