diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/flat-atomicrmw-fmin.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/flat-atomicrmw-fmin.ll | 1477 |
1 files changed, 142 insertions, 1335 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/flat-atomicrmw-fmin.ll b/llvm/test/CodeGen/AMDGPU/flat-atomicrmw-fmin.ll index fdfbb42..216ea07 100644 --- a/llvm/test/CodeGen/AMDGPU/flat-atomicrmw-fmin.ll +++ b/llvm/test/CodeGen/AMDGPU/flat-atomicrmw-fmin.ll @@ -20,28 +20,10 @@ define float @flat_agent_atomic_fmin_ret_f32(ptr %ptr, float %val) #0 { ; GFX12-NEXT: s_wait_samplecnt 0x0 ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 -; GFX12-NEXT: flat_load_b32 v3, v[0:1] -; GFX12-NEXT: v_max_num_f32_e32 v2, v2, v2 -; GFX12-NEXT: s_mov_b32 s0, 0 -; GFX12-NEXT: .LBB0_1: ; %atomicrmw.start -; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX12-NEXT: v_mov_b32_e32 v4, v3 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-NEXT: v_max_num_f32_e32 v3, v4, v4 -; GFX12-NEXT: v_min_num_f32_e32 v3, v3, v2 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: flat_atomic_cmpswap_b32 v3, v[0:1], v[3:4] th:TH_ATOMIC_RETURN +; GFX12-NEXT: flat_atomic_min_num_f32 v0, v[0:1], v2 th:TH_ATOMIC_RETURN ; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 -; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: s_cbranch_execnz .LBB0_1 -; GFX12-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: v_mov_b32_e32 v0, v3 ; GFX12-NEXT: s_setpc_b64 s[30:31] ; ; GFX940-LABEL: flat_agent_atomic_fmin_ret_f32: @@ -72,55 +54,21 @@ define float @flat_agent_atomic_fmin_ret_f32(ptr %ptr, float %val) #0 { ; GFX11-LABEL: flat_agent_atomic_fmin_ret_f32: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: flat_load_b32 v3, v[0:1] -; GFX11-NEXT: v_max_f32_e32 v2, v2, v2 -; GFX11-NEXT: s_mov_b32 s0, 0 -; GFX11-NEXT: .LBB0_1: ; %atomicrmw.start -; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_mov_b32_e32 v4, v3 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_max_f32_e32 v3, v4, v4 -; GFX11-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: flat_atomic_cmpswap_b32 v3, v[0:1], v[3:4] glc +; GFX11-NEXT: flat_atomic_min_f32 v0, v[0:1], v2 glc ; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 -; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: s_cbranch_execnz .LBB0_1 -; GFX11-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: v_mov_b32_e32 v0, v3 ; GFX11-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: flat_agent_atomic_fmin_ret_f32: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: flat_load_dword v3, v[0:1] -; GFX10-NEXT: v_max_f32_e32 v2, v2, v2 -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: .LBB0_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_mov_b32_e32 v4, v3 -; GFX10-NEXT: v_max_f32_e32 v3, v4, v4 -; GFX10-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap v3, v[0:1], v[3:4] glc +; GFX10-NEXT: flat_atomic_fmin v0, v[0:1], v2 glc ; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB0_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: v_mov_b32_e32 v0, v3 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_ret_f32: @@ -198,25 +146,9 @@ define float @flat_agent_atomic_fmin_ret_f32(ptr %ptr, float %val) #0 { ; GFX7-LABEL: flat_agent_atomic_fmin_ret_f32: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: flat_load_dword v3, v[0:1] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: v_mul_f32_e32 v2, 1.0, v2 -; GFX7-NEXT: .LBB0_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mov_b32_e32 v4, v3 -; GFX7-NEXT: v_mul_f32_e32 v3, 1.0, v4 -; GFX7-NEXT: v_min_f32_e32 v3, v3, v2 -; GFX7-NEXT: flat_atomic_cmpswap v3, v[0:1], v[3:4] glc +; GFX7-NEXT: flat_atomic_fmin v0, v[0:1], v2 glc ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB0_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v0, v3 ; GFX7-NEXT: s_setpc_b64 s[30:31] %result = atomicrmw fmin ptr %ptr, float %val syncscope("agent") seq_cst ret float %result @@ -230,28 +162,10 @@ define float @flat_agent_atomic_fmin_ret_f32__offset12b_pos(ptr %ptr, float %val ; GFX12-NEXT: s_wait_samplecnt 0x0 ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 -; GFX12-NEXT: flat_load_b32 v3, v[0:1] offset:2044 -; GFX12-NEXT: v_max_num_f32_e32 v2, v2, v2 -; GFX12-NEXT: s_mov_b32 s0, 0 -; GFX12-NEXT: .LBB1_1: ; %atomicrmw.start -; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX12-NEXT: v_mov_b32_e32 v4, v3 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-NEXT: v_max_num_f32_e32 v3, v4, v4 -; GFX12-NEXT: v_min_num_f32_e32 v3, v3, v2 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: flat_atomic_cmpswap_b32 v3, v[0:1], v[3:4] offset:2044 th:TH_ATOMIC_RETURN +; GFX12-NEXT: flat_atomic_min_num_f32 v0, v[0:1], v2 offset:2044 th:TH_ATOMIC_RETURN ; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 -; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: s_cbranch_execnz .LBB1_1 -; GFX12-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: v_mov_b32_e32 v0, v3 ; GFX12-NEXT: s_setpc_b64 s[30:31] ; ; GFX940-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_pos: @@ -282,56 +196,23 @@ define float @flat_agent_atomic_fmin_ret_f32__offset12b_pos(ptr %ptr, float %val ; GFX11-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_pos: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: flat_load_b32 v3, v[0:1] offset:2044 -; GFX11-NEXT: v_max_f32_e32 v2, v2, v2 -; GFX11-NEXT: s_mov_b32 s0, 0 -; GFX11-NEXT: .LBB1_1: ; %atomicrmw.start -; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_mov_b32_e32 v4, v3 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_max_f32_e32 v3, v4, v4 -; GFX11-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: flat_atomic_cmpswap_b32 v3, v[0:1], v[3:4] offset:2044 glc +; GFX11-NEXT: flat_atomic_min_f32 v0, v[0:1], v2 offset:2044 glc ; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 -; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: s_cbranch_execnz .LBB1_1 -; GFX11-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: v_mov_b32_e32 v0, v3 ; GFX11-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_pos: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_co_u32 v3, vcc_lo, 0x7fc, v0 -; GFX10-NEXT: v_add_co_ci_u32_e32 v4, vcc_lo, 0, v1, vcc_lo -; GFX10-NEXT: v_max_f32_e32 v1, v2, v2 -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: flat_load_dword v0, v[3:4] -; GFX10-NEXT: .LBB1_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_mov_b32_e32 v6, v0 -; GFX10-NEXT: v_max_f32_e32 v0, v6, v6 -; GFX10-NEXT: v_min_f32_e32 v5, v0, v1 +; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, 0x7fc, v0 +; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap v0, v[3:4], v[5:6] glc +; GFX10-NEXT: flat_atomic_fmin v0, v[0:1], v2 glc ; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v0, v6 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB1_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_pos: @@ -410,26 +291,11 @@ define float @flat_agent_atomic_fmin_ret_f32__offset12b_pos(ptr %ptr, float %val ; GFX7-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_pos: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_add_i32_e32 v3, vcc, 0x7fc, v0 -; GFX7-NEXT: v_addc_u32_e32 v4, vcc, 0, v1, vcc -; GFX7-NEXT: flat_load_dword v0, v[3:4] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: v_mul_f32_e32 v1, 1.0, v2 -; GFX7-NEXT: .LBB1_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mov_b32_e32 v6, v0 -; GFX7-NEXT: v_mul_f32_e32 v0, 1.0, v6 -; GFX7-NEXT: v_min_f32_e32 v5, v0, v1 -; GFX7-NEXT: flat_atomic_cmpswap v0, v[3:4], v[5:6] glc +; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0x7fc, v0 +; GFX7-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc +; GFX7-NEXT: flat_atomic_fmin v0, v[0:1], v2 glc ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v0, v6 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB1_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %gep = getelementptr float, ptr %ptr, i64 511 %result = atomicrmw fmin ptr %gep, float %val syncscope("agent") seq_cst @@ -444,28 +310,10 @@ define float @flat_agent_atomic_fmin_ret_f32__offset12b_neg(ptr %ptr, float %val ; GFX12-NEXT: s_wait_samplecnt 0x0 ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 -; GFX12-NEXT: flat_load_b32 v3, v[0:1] offset:-2048 -; GFX12-NEXT: v_max_num_f32_e32 v2, v2, v2 -; GFX12-NEXT: s_mov_b32 s0, 0 -; GFX12-NEXT: .LBB2_1: ; %atomicrmw.start -; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX12-NEXT: v_mov_b32_e32 v4, v3 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-NEXT: v_max_num_f32_e32 v3, v4, v4 -; GFX12-NEXT: v_min_num_f32_e32 v3, v3, v2 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: flat_atomic_cmpswap_b32 v3, v[0:1], v[3:4] offset:-2048 th:TH_ATOMIC_RETURN +; GFX12-NEXT: flat_atomic_min_num_f32 v0, v[0:1], v2 offset:-2048 th:TH_ATOMIC_RETURN ; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 -; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: s_cbranch_execnz .LBB2_1 -; GFX12-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: v_mov_b32_e32 v0, v3 ; GFX12-NEXT: s_setpc_b64 s[30:31] ; ; GFX940-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_neg: @@ -503,61 +351,25 @@ define float @flat_agent_atomic_fmin_ret_f32__offset12b_neg(ptr %ptr, float %val ; GFX11-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_neg: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_mov_b32_e32 v3, v0 -; GFX11-NEXT: s_mov_b32 s0, 0 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_add_co_u32 v4, vcc_lo, 0xfffff800, v3 -; GFX11-NEXT: v_add_co_ci_u32_e32 v5, vcc_lo, -1, v1, vcc_lo -; GFX11-NEXT: v_add_co_u32 v3, vcc_lo, 0xfffff800, v3 -; GFX11-NEXT: flat_load_b32 v0, v[4:5] -; GFX11-NEXT: v_add_co_ci_u32_e32 v4, vcc_lo, -1, v1, vcc_lo -; GFX11-NEXT: v_max_f32_e32 v1, v2, v2 -; GFX11-NEXT: .LBB2_1: ; %atomicrmw.start -; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_mov_b32_e32 v6, v0 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_max_f32_e32 v0, v6, v6 -; GFX11-NEXT: v_min_f32_e32 v5, v0, v1 +; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, 0xfffff800, v0 +; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, -1, v1, vcc_lo ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: flat_atomic_cmpswap_b32 v0, v[3:4], v[5:6] glc +; GFX11-NEXT: flat_atomic_min_f32 v0, v[0:1], v2 glc ; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v0, v6 -; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: s_cbranch_execnz .LBB2_1 -; GFX11-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX11-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_neg: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_co_u32 v3, vcc_lo, 0xfffff800, v0 -; GFX10-NEXT: v_add_co_ci_u32_e32 v4, vcc_lo, -1, v1, vcc_lo -; GFX10-NEXT: v_max_f32_e32 v1, v2, v2 -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: flat_load_dword v0, v[3:4] -; GFX10-NEXT: .LBB2_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_mov_b32_e32 v6, v0 -; GFX10-NEXT: v_max_f32_e32 v0, v6, v6 -; GFX10-NEXT: v_min_f32_e32 v5, v0, v1 +; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, 0xfffff800, v0 +; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, -1, v1, vcc_lo ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap v0, v[3:4], v[5:6] glc +; GFX10-NEXT: flat_atomic_fmin v0, v[0:1], v2 glc ; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v0, v6 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB2_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_neg: @@ -642,26 +454,11 @@ define float @flat_agent_atomic_fmin_ret_f32__offset12b_neg(ptr %ptr, float %val ; GFX7-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_neg: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_add_i32_e32 v3, vcc, 0xfffff800, v0 -; GFX7-NEXT: v_addc_u32_e32 v4, vcc, -1, v1, vcc -; GFX7-NEXT: flat_load_dword v0, v[3:4] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: v_mul_f32_e32 v1, 1.0, v2 -; GFX7-NEXT: .LBB2_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mov_b32_e32 v6, v0 -; GFX7-NEXT: v_mul_f32_e32 v0, 1.0, v6 -; GFX7-NEXT: v_min_f32_e32 v5, v0, v1 -; GFX7-NEXT: flat_atomic_cmpswap v0, v[3:4], v[5:6] glc +; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0xfffff800, v0 +; GFX7-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GFX7-NEXT: flat_atomic_fmin v0, v[0:1], v2 glc ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v0, v6 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB2_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %gep = getelementptr float, ptr %ptr, i64 -512 %result = atomicrmw fmin ptr %gep, float %val syncscope("agent") seq_cst @@ -676,27 +473,10 @@ define void @flat_agent_atomic_fmin_noret_f32(ptr %ptr, float %val) #0 { ; GFX12-NEXT: s_wait_samplecnt 0x0 ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 -; GFX12-NEXT: flat_load_b32 v3, v[0:1] -; GFX12-NEXT: v_max_num_f32_e32 v4, v2, v2 -; GFX12-NEXT: s_mov_b32 s0, 0 -; GFX12-NEXT: .LBB3_1: ; %atomicrmw.start -; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX12-NEXT: v_max_num_f32_e32 v2, v3, v3 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-NEXT: v_min_num_f32_e32 v2, v2, v4 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] th:TH_ATOMIC_RETURN -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: flat_atomic_min_num_f32 v[0:1], v2 +; GFX12-NEXT: s_wait_storecnt_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX12-NEXT: v_mov_b32_e32 v3, v2 -; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: s_cbranch_execnz .LBB3_1 -; GFX12-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX12-NEXT: s_setpc_b64 s[30:31] ; ; GFX940-LABEL: flat_agent_atomic_fmin_noret_f32: @@ -726,53 +506,23 @@ define void @flat_agent_atomic_fmin_noret_f32(ptr %ptr, float %val) #0 { ; GFX11-LABEL: flat_agent_atomic_fmin_noret_f32: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: flat_load_b32 v3, v[0:1] -; GFX11-NEXT: v_max_f32_e32 v4, v2, v2 -; GFX11-NEXT: s_mov_b32 s0, 0 -; GFX11-NEXT: .LBB3_1: ; %atomicrmw.start -; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_max_f32_e32 v2, v3, v3 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_min_f32_e32 v2, v2, v4 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] glc -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX11-NEXT: flat_atomic_min_f32 v[0:1], v2 +; GFX11-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX11-NEXT: v_mov_b32_e32 v3, v2 -; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: s_cbranch_execnz .LBB3_1 -; GFX11-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX11-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: flat_agent_atomic_fmin_noret_f32: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: flat_load_dword v3, v[0:1] -; GFX10-NEXT: v_max_f32_e32 v4, v2, v2 -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: .LBB3_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_max_f32_e32 v2, v3, v3 -; GFX10-NEXT: v_min_f32_e32 v2, v2, v4 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap v2, v[0:1], v[2:3] glc -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX10-NEXT: flat_atomic_fmin v[0:1], v2 +; GFX10-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v2 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB3_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_noret_f32: @@ -847,24 +597,9 @@ define void @flat_agent_atomic_fmin_noret_f32(ptr %ptr, float %val) #0 { ; GFX7-LABEL: flat_agent_atomic_fmin_noret_f32: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: flat_load_dword v3, v[0:1] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: v_mul_f32_e32 v4, 1.0, v2 -; GFX7-NEXT: .LBB3_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mul_f32_e32 v2, 1.0, v3 -; GFX7-NEXT: v_min_f32_e32 v2, v2, v4 -; GFX7-NEXT: flat_atomic_cmpswap v2, v[0:1], v[2:3] glc +; GFX7-NEXT: flat_atomic_fmin v[0:1], v2 ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v2 -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB3_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %unused = atomicrmw fmin ptr %ptr, float %val syncscope("agent") seq_cst ret void @@ -878,27 +613,10 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_pos(ptr %ptr, float %va ; GFX12-NEXT: s_wait_samplecnt 0x0 ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 -; GFX12-NEXT: flat_load_b32 v3, v[0:1] offset:2044 -; GFX12-NEXT: v_max_num_f32_e32 v4, v2, v2 -; GFX12-NEXT: s_mov_b32 s0, 0 -; GFX12-NEXT: .LBB4_1: ; %atomicrmw.start -; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX12-NEXT: v_max_num_f32_e32 v2, v3, v3 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-NEXT: v_min_num_f32_e32 v2, v2, v4 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:2044 th:TH_ATOMIC_RETURN -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: flat_atomic_min_num_f32 v[0:1], v2 offset:2044 +; GFX12-NEXT: s_wait_storecnt_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX12-NEXT: v_mov_b32_e32 v3, v2 -; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: s_cbranch_execnz .LBB4_1 -; GFX12-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX12-NEXT: s_setpc_b64 s[30:31] ; ; GFX940-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_pos: @@ -928,28 +646,12 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_pos(ptr %ptr, float %va ; GFX11-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_pos: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: flat_load_b32 v3, v[0:1] offset:2044 -; GFX11-NEXT: v_max_f32_e32 v4, v2, v2 -; GFX11-NEXT: s_mov_b32 s0, 0 -; GFX11-NEXT: .LBB4_1: ; %atomicrmw.start -; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_max_f32_e32 v2, v3, v3 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_min_f32_e32 v2, v2, v4 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:2044 glc -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX11-NEXT: flat_atomic_min_f32 v[0:1], v2 offset:2044 +; GFX11-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX11-NEXT: v_mov_b32_e32 v3, v2 -; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: s_cbranch_execnz .LBB4_1 -; GFX11-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX11-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_pos: @@ -957,26 +659,12 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_pos(ptr %ptr, float %va ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, 0x7fc, v0 ; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo -; GFX10-NEXT: v_max_f32_e32 v4, v2, v2 -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: flat_load_dword v3, v[0:1] -; GFX10-NEXT: .LBB4_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_max_f32_e32 v2, v3, v3 -; GFX10-NEXT: v_min_f32_e32 v2, v2, v4 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap v2, v[0:1], v[2:3] glc -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX10-NEXT: flat_atomic_fmin v[0:1], v2 +; GFX10-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v2 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB4_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_pos: @@ -1055,24 +743,9 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_pos(ptr %ptr, float %va ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0x7fc, v0 ; GFX7-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc -; GFX7-NEXT: flat_load_dword v3, v[0:1] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: v_mul_f32_e32 v4, 1.0, v2 -; GFX7-NEXT: .LBB4_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mul_f32_e32 v2, 1.0, v3 -; GFX7-NEXT: v_min_f32_e32 v2, v2, v4 -; GFX7-NEXT: flat_atomic_cmpswap v2, v[0:1], v[2:3] glc +; GFX7-NEXT: flat_atomic_fmin v[0:1], v2 ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v2 -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB4_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %gep = getelementptr float, ptr %ptr, i64 511 %unused = atomicrmw fmin ptr %gep, float %val syncscope("agent") seq_cst @@ -1087,27 +760,10 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_neg(ptr %ptr, float %va ; GFX12-NEXT: s_wait_samplecnt 0x0 ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 -; GFX12-NEXT: flat_load_b32 v3, v[0:1] offset:-2048 -; GFX12-NEXT: v_max_num_f32_e32 v4, v2, v2 -; GFX12-NEXT: s_mov_b32 s0, 0 -; GFX12-NEXT: .LBB5_1: ; %atomicrmw.start -; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX12-NEXT: v_max_num_f32_e32 v2, v3, v3 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-NEXT: v_min_num_f32_e32 v2, v2, v4 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:-2048 th:TH_ATOMIC_RETURN -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: flat_atomic_min_num_f32 v[0:1], v2 offset:-2048 +; GFX12-NEXT: s_wait_storecnt_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX12-NEXT: v_mov_b32_e32 v3, v2 -; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: s_cbranch_execnz .LBB5_1 -; GFX12-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX12-NEXT: s_setpc_b64 s[30:31] ; ; GFX940-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_neg: @@ -1143,32 +799,14 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_neg(ptr %ptr, float %va ; GFX11-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_neg: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_add_co_u32 v3, vcc_lo, 0xfffff800, v0 -; GFX11-NEXT: v_add_co_ci_u32_e32 v4, vcc_lo, -1, v1, vcc_lo ; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, 0xfffff800, v0 ; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, -1, v1, vcc_lo -; GFX11-NEXT: flat_load_b32 v3, v[3:4] -; GFX11-NEXT: v_max_f32_e32 v4, v2, v2 -; GFX11-NEXT: s_mov_b32 s0, 0 -; GFX11-NEXT: .LBB5_1: ; %atomicrmw.start -; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_max_f32_e32 v2, v3, v3 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_min_f32_e32 v2, v2, v4 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] glc -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX11-NEXT: flat_atomic_min_f32 v[0:1], v2 +; GFX11-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX11-NEXT: v_mov_b32_e32 v3, v2 -; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: s_cbranch_execnz .LBB5_1 -; GFX11-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX11-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_neg: @@ -1176,26 +814,12 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_neg(ptr %ptr, float %va ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, 0xfffff800, v0 ; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, -1, v1, vcc_lo -; GFX10-NEXT: v_max_f32_e32 v4, v2, v2 -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: flat_load_dword v3, v[0:1] -; GFX10-NEXT: .LBB5_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_max_f32_e32 v2, v3, v3 -; GFX10-NEXT: v_min_f32_e32 v2, v2, v4 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap v2, v[0:1], v[2:3] glc -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX10-NEXT: flat_atomic_fmin v[0:1], v2 +; GFX10-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v2 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB5_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_neg: @@ -1282,24 +906,9 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_neg(ptr %ptr, float %va ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0xfffff800, v0 ; GFX7-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc -; GFX7-NEXT: flat_load_dword v3, v[0:1] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: v_mul_f32_e32 v4, 1.0, v2 -; GFX7-NEXT: .LBB5_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mul_f32_e32 v2, 1.0, v3 -; GFX7-NEXT: v_min_f32_e32 v2, v2, v4 -; GFX7-NEXT: flat_atomic_cmpswap v2, v[0:1], v[2:3] glc +; GFX7-NEXT: flat_atomic_fmin v[0:1], v2 ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v2 -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB5_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %gep = getelementptr float, ptr %ptr, i64 -512 %unused = atomicrmw fmin ptr %gep, float %val syncscope("agent") seq_cst @@ -1745,28 +1354,10 @@ define float @flat_agent_atomic_fmin_ret_f32__ftz(ptr %ptr, float %val) #1 { ; GFX12-NEXT: s_wait_samplecnt 0x0 ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 -; GFX12-NEXT: flat_load_b32 v3, v[0:1] -; GFX12-NEXT: v_max_num_f32_e32 v2, v2, v2 -; GFX12-NEXT: s_mov_b32 s0, 0 -; GFX12-NEXT: .LBB8_1: ; %atomicrmw.start -; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX12-NEXT: v_mov_b32_e32 v4, v3 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-NEXT: v_max_num_f32_e32 v3, v4, v4 -; GFX12-NEXT: v_min_num_f32_e32 v3, v3, v2 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: flat_atomic_cmpswap_b32 v3, v[0:1], v[3:4] th:TH_ATOMIC_RETURN +; GFX12-NEXT: flat_atomic_min_num_f32 v0, v[0:1], v2 th:TH_ATOMIC_RETURN ; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 -; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: s_cbranch_execnz .LBB8_1 -; GFX12-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: v_mov_b32_e32 v0, v3 ; GFX12-NEXT: s_setpc_b64 s[30:31] ; ; GFX940-LABEL: flat_agent_atomic_fmin_ret_f32__ftz: @@ -1797,55 +1388,21 @@ define float @flat_agent_atomic_fmin_ret_f32__ftz(ptr %ptr, float %val) #1 { ; GFX11-LABEL: flat_agent_atomic_fmin_ret_f32__ftz: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: flat_load_b32 v3, v[0:1] -; GFX11-NEXT: v_max_f32_e32 v2, v2, v2 -; GFX11-NEXT: s_mov_b32 s0, 0 -; GFX11-NEXT: .LBB8_1: ; %atomicrmw.start -; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_mov_b32_e32 v4, v3 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_max_f32_e32 v3, v4, v4 -; GFX11-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: flat_atomic_cmpswap_b32 v3, v[0:1], v[3:4] glc +; GFX11-NEXT: flat_atomic_min_f32 v0, v[0:1], v2 glc ; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 -; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: s_cbranch_execnz .LBB8_1 -; GFX11-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: v_mov_b32_e32 v0, v3 ; GFX11-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: flat_agent_atomic_fmin_ret_f32__ftz: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: flat_load_dword v3, v[0:1] -; GFX10-NEXT: v_max_f32_e32 v2, v2, v2 -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: .LBB8_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_mov_b32_e32 v4, v3 -; GFX10-NEXT: v_max_f32_e32 v3, v4, v4 -; GFX10-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap v3, v[0:1], v[3:4] glc +; GFX10-NEXT: flat_atomic_fmin v0, v[0:1], v2 glc ; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB8_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: v_mov_b32_e32 v0, v3 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_ret_f32__ftz: @@ -1923,25 +1480,9 @@ define float @flat_agent_atomic_fmin_ret_f32__ftz(ptr %ptr, float %val) #1 { ; GFX7-LABEL: flat_agent_atomic_fmin_ret_f32__ftz: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: flat_load_dword v3, v[0:1] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: v_mul_f32_e32 v2, 1.0, v2 -; GFX7-NEXT: .LBB8_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mov_b32_e32 v4, v3 -; GFX7-NEXT: v_mul_f32_e32 v3, 1.0, v4 -; GFX7-NEXT: v_min_f32_e32 v3, v3, v2 -; GFX7-NEXT: flat_atomic_cmpswap v3, v[0:1], v[3:4] glc +; GFX7-NEXT: flat_atomic_fmin v0, v[0:1], v2 glc ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB8_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v0, v3 ; GFX7-NEXT: s_setpc_b64 s[30:31] %result = atomicrmw fmin ptr %ptr, float %val syncscope("agent") seq_cst ret float %result @@ -1955,28 +1496,10 @@ define float @flat_agent_atomic_fmin_ret_f32__offset12b_pos__ftz(ptr %ptr, float ; GFX12-NEXT: s_wait_samplecnt 0x0 ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 -; GFX12-NEXT: flat_load_b32 v3, v[0:1] offset:2044 -; GFX12-NEXT: v_max_num_f32_e32 v2, v2, v2 -; GFX12-NEXT: s_mov_b32 s0, 0 -; GFX12-NEXT: .LBB9_1: ; %atomicrmw.start -; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX12-NEXT: v_mov_b32_e32 v4, v3 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-NEXT: v_max_num_f32_e32 v3, v4, v4 -; GFX12-NEXT: v_min_num_f32_e32 v3, v3, v2 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: flat_atomic_cmpswap_b32 v3, v[0:1], v[3:4] offset:2044 th:TH_ATOMIC_RETURN +; GFX12-NEXT: flat_atomic_min_num_f32 v0, v[0:1], v2 offset:2044 th:TH_ATOMIC_RETURN ; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 -; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: s_cbranch_execnz .LBB9_1 -; GFX12-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: v_mov_b32_e32 v0, v3 ; GFX12-NEXT: s_setpc_b64 s[30:31] ; ; GFX940-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_pos__ftz: @@ -2007,56 +1530,23 @@ define float @flat_agent_atomic_fmin_ret_f32__offset12b_pos__ftz(ptr %ptr, float ; GFX11-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_pos__ftz: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: flat_load_b32 v3, v[0:1] offset:2044 -; GFX11-NEXT: v_max_f32_e32 v2, v2, v2 -; GFX11-NEXT: s_mov_b32 s0, 0 -; GFX11-NEXT: .LBB9_1: ; %atomicrmw.start -; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_mov_b32_e32 v4, v3 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_max_f32_e32 v3, v4, v4 -; GFX11-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: flat_atomic_cmpswap_b32 v3, v[0:1], v[3:4] offset:2044 glc +; GFX11-NEXT: flat_atomic_min_f32 v0, v[0:1], v2 offset:2044 glc ; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 -; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: s_cbranch_execnz .LBB9_1 -; GFX11-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: v_mov_b32_e32 v0, v3 ; GFX11-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_pos__ftz: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_co_u32 v3, vcc_lo, 0x7fc, v0 -; GFX10-NEXT: v_add_co_ci_u32_e32 v4, vcc_lo, 0, v1, vcc_lo -; GFX10-NEXT: v_max_f32_e32 v1, v2, v2 -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: flat_load_dword v0, v[3:4] -; GFX10-NEXT: .LBB9_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_mov_b32_e32 v6, v0 -; GFX10-NEXT: v_max_f32_e32 v0, v6, v6 -; GFX10-NEXT: v_min_f32_e32 v5, v0, v1 +; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, 0x7fc, v0 +; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap v0, v[3:4], v[5:6] glc +; GFX10-NEXT: flat_atomic_fmin v0, v[0:1], v2 glc ; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v0, v6 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB9_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_pos__ftz: @@ -2135,26 +1625,11 @@ define float @flat_agent_atomic_fmin_ret_f32__offset12b_pos__ftz(ptr %ptr, float ; GFX7-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_pos__ftz: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_add_i32_e32 v3, vcc, 0x7fc, v0 -; GFX7-NEXT: v_addc_u32_e32 v4, vcc, 0, v1, vcc -; GFX7-NEXT: flat_load_dword v0, v[3:4] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: v_mul_f32_e32 v1, 1.0, v2 -; GFX7-NEXT: .LBB9_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mov_b32_e32 v6, v0 -; GFX7-NEXT: v_mul_f32_e32 v0, 1.0, v6 -; GFX7-NEXT: v_min_f32_e32 v5, v0, v1 -; GFX7-NEXT: flat_atomic_cmpswap v0, v[3:4], v[5:6] glc +; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0x7fc, v0 +; GFX7-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc +; GFX7-NEXT: flat_atomic_fmin v0, v[0:1], v2 glc ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v0, v6 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB9_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %gep = getelementptr float, ptr %ptr, i64 511 %result = atomicrmw fmin ptr %gep, float %val syncscope("agent") seq_cst @@ -2169,28 +1644,10 @@ define float @flat_agent_atomic_fmin_ret_f32__offset12b_neg__ftz(ptr %ptr, float ; GFX12-NEXT: s_wait_samplecnt 0x0 ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 -; GFX12-NEXT: flat_load_b32 v3, v[0:1] offset:-2048 -; GFX12-NEXT: v_max_num_f32_e32 v2, v2, v2 -; GFX12-NEXT: s_mov_b32 s0, 0 -; GFX12-NEXT: .LBB10_1: ; %atomicrmw.start -; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX12-NEXT: v_mov_b32_e32 v4, v3 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-NEXT: v_max_num_f32_e32 v3, v4, v4 -; GFX12-NEXT: v_min_num_f32_e32 v3, v3, v2 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: flat_atomic_cmpswap_b32 v3, v[0:1], v[3:4] offset:-2048 th:TH_ATOMIC_RETURN +; GFX12-NEXT: flat_atomic_min_num_f32 v0, v[0:1], v2 offset:-2048 th:TH_ATOMIC_RETURN ; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 -; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: s_cbranch_execnz .LBB10_1 -; GFX12-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: v_mov_b32_e32 v0, v3 ; GFX12-NEXT: s_setpc_b64 s[30:31] ; ; GFX940-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_neg__ftz: @@ -2228,61 +1685,25 @@ define float @flat_agent_atomic_fmin_ret_f32__offset12b_neg__ftz(ptr %ptr, float ; GFX11-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_neg__ftz: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_mov_b32_e32 v3, v0 -; GFX11-NEXT: s_mov_b32 s0, 0 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_add_co_u32 v4, vcc_lo, 0xfffff800, v3 -; GFX11-NEXT: v_add_co_ci_u32_e32 v5, vcc_lo, -1, v1, vcc_lo -; GFX11-NEXT: v_add_co_u32 v3, vcc_lo, 0xfffff800, v3 -; GFX11-NEXT: flat_load_b32 v0, v[4:5] -; GFX11-NEXT: v_add_co_ci_u32_e32 v4, vcc_lo, -1, v1, vcc_lo -; GFX11-NEXT: v_max_f32_e32 v1, v2, v2 -; GFX11-NEXT: .LBB10_1: ; %atomicrmw.start -; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_mov_b32_e32 v6, v0 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_max_f32_e32 v0, v6, v6 -; GFX11-NEXT: v_min_f32_e32 v5, v0, v1 +; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, 0xfffff800, v0 +; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, -1, v1, vcc_lo ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: flat_atomic_cmpswap_b32 v0, v[3:4], v[5:6] glc +; GFX11-NEXT: flat_atomic_min_f32 v0, v[0:1], v2 glc ; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v0, v6 -; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: s_cbranch_execnz .LBB10_1 -; GFX11-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX11-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_neg__ftz: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_co_u32 v3, vcc_lo, 0xfffff800, v0 -; GFX10-NEXT: v_add_co_ci_u32_e32 v4, vcc_lo, -1, v1, vcc_lo -; GFX10-NEXT: v_max_f32_e32 v1, v2, v2 -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: flat_load_dword v0, v[3:4] -; GFX10-NEXT: .LBB10_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_mov_b32_e32 v6, v0 -; GFX10-NEXT: v_max_f32_e32 v0, v6, v6 -; GFX10-NEXT: v_min_f32_e32 v5, v0, v1 +; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, 0xfffff800, v0 +; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, -1, v1, vcc_lo ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap v0, v[3:4], v[5:6] glc +; GFX10-NEXT: flat_atomic_fmin v0, v[0:1], v2 glc ; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v0, v6 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB10_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_neg__ftz: @@ -2367,26 +1788,11 @@ define float @flat_agent_atomic_fmin_ret_f32__offset12b_neg__ftz(ptr %ptr, float ; GFX7-LABEL: flat_agent_atomic_fmin_ret_f32__offset12b_neg__ftz: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_add_i32_e32 v3, vcc, 0xfffff800, v0 -; GFX7-NEXT: v_addc_u32_e32 v4, vcc, -1, v1, vcc -; GFX7-NEXT: flat_load_dword v0, v[3:4] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: v_mul_f32_e32 v1, 1.0, v2 -; GFX7-NEXT: .LBB10_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mov_b32_e32 v6, v0 -; GFX7-NEXT: v_mul_f32_e32 v0, 1.0, v6 -; GFX7-NEXT: v_min_f32_e32 v5, v0, v1 -; GFX7-NEXT: flat_atomic_cmpswap v0, v[3:4], v[5:6] glc +; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0xfffff800, v0 +; GFX7-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GFX7-NEXT: flat_atomic_fmin v0, v[0:1], v2 glc ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v0, v6 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB10_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %gep = getelementptr float, ptr %ptr, i64 -512 %result = atomicrmw fmin ptr %gep, float %val syncscope("agent") seq_cst @@ -2401,27 +1807,10 @@ define void @flat_agent_atomic_fmin_noret_f32__ftz(ptr %ptr, float %val) #1 { ; GFX12-NEXT: s_wait_samplecnt 0x0 ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 -; GFX12-NEXT: flat_load_b32 v3, v[0:1] -; GFX12-NEXT: v_max_num_f32_e32 v4, v2, v2 -; GFX12-NEXT: s_mov_b32 s0, 0 -; GFX12-NEXT: .LBB11_1: ; %atomicrmw.start -; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX12-NEXT: v_max_num_f32_e32 v2, v3, v3 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-NEXT: v_min_num_f32_e32 v2, v2, v4 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] th:TH_ATOMIC_RETURN -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: flat_atomic_min_num_f32 v[0:1], v2 +; GFX12-NEXT: s_wait_storecnt_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX12-NEXT: v_mov_b32_e32 v3, v2 -; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: s_cbranch_execnz .LBB11_1 -; GFX12-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX12-NEXT: s_setpc_b64 s[30:31] ; ; GFX940-LABEL: flat_agent_atomic_fmin_noret_f32__ftz: @@ -2451,53 +1840,23 @@ define void @flat_agent_atomic_fmin_noret_f32__ftz(ptr %ptr, float %val) #1 { ; GFX11-LABEL: flat_agent_atomic_fmin_noret_f32__ftz: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: flat_load_b32 v3, v[0:1] -; GFX11-NEXT: v_max_f32_e32 v4, v2, v2 -; GFX11-NEXT: s_mov_b32 s0, 0 -; GFX11-NEXT: .LBB11_1: ; %atomicrmw.start -; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_max_f32_e32 v2, v3, v3 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_min_f32_e32 v2, v2, v4 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] glc -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX11-NEXT: flat_atomic_min_f32 v[0:1], v2 +; GFX11-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX11-NEXT: v_mov_b32_e32 v3, v2 -; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: s_cbranch_execnz .LBB11_1 -; GFX11-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX11-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: flat_agent_atomic_fmin_noret_f32__ftz: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: flat_load_dword v3, v[0:1] -; GFX10-NEXT: v_max_f32_e32 v4, v2, v2 -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: .LBB11_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_max_f32_e32 v2, v3, v3 -; GFX10-NEXT: v_min_f32_e32 v2, v2, v4 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap v2, v[0:1], v[2:3] glc -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX10-NEXT: flat_atomic_fmin v[0:1], v2 +; GFX10-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v2 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB11_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_noret_f32__ftz: @@ -2572,24 +1931,9 @@ define void @flat_agent_atomic_fmin_noret_f32__ftz(ptr %ptr, float %val) #1 { ; GFX7-LABEL: flat_agent_atomic_fmin_noret_f32__ftz: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: flat_load_dword v3, v[0:1] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: v_mul_f32_e32 v4, 1.0, v2 -; GFX7-NEXT: .LBB11_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mul_f32_e32 v2, 1.0, v3 -; GFX7-NEXT: v_min_f32_e32 v2, v2, v4 -; GFX7-NEXT: flat_atomic_cmpswap v2, v[0:1], v[2:3] glc +; GFX7-NEXT: flat_atomic_fmin v[0:1], v2 ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v2 -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB11_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %unused = atomicrmw fmin ptr %ptr, float %val syncscope("agent") seq_cst ret void @@ -2603,27 +1947,10 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_pos__ftz(ptr %ptr, floa ; GFX12-NEXT: s_wait_samplecnt 0x0 ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 -; GFX12-NEXT: flat_load_b32 v3, v[0:1] offset:2044 -; GFX12-NEXT: v_max_num_f32_e32 v4, v2, v2 -; GFX12-NEXT: s_mov_b32 s0, 0 -; GFX12-NEXT: .LBB12_1: ; %atomicrmw.start -; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX12-NEXT: v_max_num_f32_e32 v2, v3, v3 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-NEXT: v_min_num_f32_e32 v2, v2, v4 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:2044 th:TH_ATOMIC_RETURN -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: flat_atomic_min_num_f32 v[0:1], v2 offset:2044 +; GFX12-NEXT: s_wait_storecnt_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX12-NEXT: v_mov_b32_e32 v3, v2 -; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: s_cbranch_execnz .LBB12_1 -; GFX12-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX12-NEXT: s_setpc_b64 s[30:31] ; ; GFX940-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_pos__ftz: @@ -2653,28 +1980,12 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_pos__ftz(ptr %ptr, floa ; GFX11-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_pos__ftz: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: flat_load_b32 v3, v[0:1] offset:2044 -; GFX11-NEXT: v_max_f32_e32 v4, v2, v2 -; GFX11-NEXT: s_mov_b32 s0, 0 -; GFX11-NEXT: .LBB12_1: ; %atomicrmw.start -; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_max_f32_e32 v2, v3, v3 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_min_f32_e32 v2, v2, v4 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:2044 glc -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX11-NEXT: flat_atomic_min_f32 v[0:1], v2 offset:2044 +; GFX11-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX11-NEXT: v_mov_b32_e32 v3, v2 -; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: s_cbranch_execnz .LBB12_1 -; GFX11-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX11-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_pos__ftz: @@ -2682,26 +1993,12 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_pos__ftz(ptr %ptr, floa ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, 0x7fc, v0 ; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo -; GFX10-NEXT: v_max_f32_e32 v4, v2, v2 -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: flat_load_dword v3, v[0:1] -; GFX10-NEXT: .LBB12_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_max_f32_e32 v2, v3, v3 -; GFX10-NEXT: v_min_f32_e32 v2, v2, v4 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap v2, v[0:1], v[2:3] glc -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX10-NEXT: flat_atomic_fmin v[0:1], v2 +; GFX10-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v2 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB12_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_pos__ftz: @@ -2780,24 +2077,9 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_pos__ftz(ptr %ptr, floa ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0x7fc, v0 ; GFX7-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc -; GFX7-NEXT: flat_load_dword v3, v[0:1] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: v_mul_f32_e32 v4, 1.0, v2 -; GFX7-NEXT: .LBB12_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mul_f32_e32 v2, 1.0, v3 -; GFX7-NEXT: v_min_f32_e32 v2, v2, v4 -; GFX7-NEXT: flat_atomic_cmpswap v2, v[0:1], v[2:3] glc +; GFX7-NEXT: flat_atomic_fmin v[0:1], v2 ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v2 -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB12_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %gep = getelementptr float, ptr %ptr, i64 511 %unused = atomicrmw fmin ptr %gep, float %val syncscope("agent") seq_cst @@ -2812,27 +2094,10 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_neg__ftz(ptr %ptr, floa ; GFX12-NEXT: s_wait_samplecnt 0x0 ; GFX12-NEXT: s_wait_bvhcnt 0x0 ; GFX12-NEXT: s_wait_kmcnt 0x0 -; GFX12-NEXT: flat_load_b32 v3, v[0:1] offset:-2048 -; GFX12-NEXT: v_max_num_f32_e32 v4, v2, v2 -; GFX12-NEXT: s_mov_b32 s0, 0 -; GFX12-NEXT: .LBB13_1: ; %atomicrmw.start -; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX12-NEXT: v_max_num_f32_e32 v2, v3, v3 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-NEXT: v_min_num_f32_e32 v2, v2, v4 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] offset:-2048 th:TH_ATOMIC_RETURN -; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: flat_atomic_min_num_f32 v[0:1], v2 offset:-2048 +; GFX12-NEXT: s_wait_storecnt_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_DEV -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX12-NEXT: v_mov_b32_e32 v3, v2 -; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX12-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX12-NEXT: s_cbranch_execnz .LBB13_1 -; GFX12-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX12-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX12-NEXT: s_setpc_b64 s[30:31] ; ; GFX940-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_neg__ftz: @@ -2868,32 +2133,14 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_neg__ftz(ptr %ptr, floa ; GFX11-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_neg__ftz: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_add_co_u32 v3, vcc_lo, 0xfffff800, v0 -; GFX11-NEXT: v_add_co_ci_u32_e32 v4, vcc_lo, -1, v1, vcc_lo ; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, 0xfffff800, v0 ; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, -1, v1, vcc_lo -; GFX11-NEXT: flat_load_b32 v3, v[3:4] -; GFX11-NEXT: v_max_f32_e32 v4, v2, v2 -; GFX11-NEXT: s_mov_b32 s0, 0 -; GFX11-NEXT: .LBB13_1: ; %atomicrmw.start -; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_max_f32_e32 v2, v3, v3 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_min_f32_e32 v2, v2, v4 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: flat_atomic_cmpswap_b32 v2, v[0:1], v[2:3] glc -; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX11-NEXT: flat_atomic_min_f32 v[0:1], v2 +; GFX11-NEXT: s_waitcnt lgkmcnt(0) +; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX11-NEXT: buffer_gl1_inv ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX11-NEXT: v_mov_b32_e32 v3, v2 -; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) -; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 -; GFX11-NEXT: s_cbranch_execnz .LBB13_1 -; GFX11-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0 ; GFX11-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_neg__ftz: @@ -2901,26 +2148,12 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_neg__ftz(ptr %ptr, floa ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, 0xfffff800, v0 ; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, -1, v1, vcc_lo -; GFX10-NEXT: v_max_f32_e32 v4, v2, v2 -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: flat_load_dword v3, v[0:1] -; GFX10-NEXT: .LBB13_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_max_f32_e32 v2, v3, v3 -; GFX10-NEXT: v_min_f32_e32 v2, v2, v4 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap v2, v[0:1], v[2:3] glc -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX10-NEXT: flat_atomic_fmin v[0:1], v2 +; GFX10-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v2 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB13_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_noret_f32__offset12b_neg__ftz: @@ -3007,24 +2240,9 @@ define void @flat_agent_atomic_fmin_noret_f32__offset12b_neg__ftz(ptr %ptr, floa ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0xfffff800, v0 ; GFX7-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc -; GFX7-NEXT: flat_load_dword v3, v[0:1] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: v_mul_f32_e32 v4, 1.0, v2 -; GFX7-NEXT: .LBB13_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mul_f32_e32 v2, 1.0, v3 -; GFX7-NEXT: v_min_f32_e32 v2, v2, v4 -; GFX7-NEXT: flat_atomic_cmpswap v2, v[0:1], v[2:3] glc +; GFX7-NEXT: flat_atomic_fmin v[0:1], v2 ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v2 -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB13_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %gep = getelementptr float, ptr %ptr, i64 -512 %unused = atomicrmw fmin ptr %gep, float %val syncscope("agent") seq_cst @@ -3497,27 +2715,10 @@ define double @flat_agent_atomic_fmin_ret_f64(ptr %ptr, double %val) #0 { ; GFX940-LABEL: flat_agent_atomic_fmin_ret_f64: ; GFX940: ; %bb.0: ; GFX940-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX940-NEXT: flat_load_dwordx2 v[4:5], v[0:1] -; GFX940-NEXT: s_mov_b64 s[0:1], 0 -; GFX940-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3] -; GFX940-NEXT: .LBB16_1: ; %atomicrmw.start -; GFX940-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX940-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX940-NEXT: v_mov_b64_e32 v[6:7], v[4:5] -; GFX940-NEXT: v_max_f64 v[4:5], v[6:7], v[6:7] -; GFX940-NEXT: v_min_f64 v[4:5], v[4:5], v[2:3] ; GFX940-NEXT: buffer_wbl2 sc1 -; GFX940-NEXT: flat_atomic_cmpswap_x2 v[4:5], v[0:1], v[4:7] sc0 +; GFX940-NEXT: flat_atomic_min_f64 v[0:1], v[0:1], v[2:3] sc0 ; GFX940-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX940-NEXT: buffer_inv sc1 -; GFX940-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[6:7] -; GFX940-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX940-NEXT: s_andn2_b64 exec, exec, s[0:1] -; GFX940-NEXT: s_cbranch_execnz .LBB16_1 -; GFX940-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX940-NEXT: s_or_b64 exec, exec, s[0:1] -; GFX940-NEXT: v_mov_b32_e32 v0, v4 -; GFX940-NEXT: v_mov_b32_e32 v1, v5 ; GFX940-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: flat_agent_atomic_fmin_ret_f64: @@ -3551,54 +2752,19 @@ define double @flat_agent_atomic_fmin_ret_f64(ptr %ptr, double %val) #0 { ; GFX10-LABEL: flat_agent_atomic_fmin_ret_f64: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: flat_load_dwordx2 v[4:5], v[0:1] -; GFX10-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3] -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: .LBB16_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_mov_b32_e32 v7, v5 -; GFX10-NEXT: v_mov_b32_e32 v6, v4 -; GFX10-NEXT: v_max_f64 v[4:5], v[6:7], v[6:7] -; GFX10-NEXT: v_min_f64 v[4:5], v[4:5], v[2:3] ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap_x2 v[4:5], v[0:1], v[4:7] glc +; GFX10-NEXT: flat_atomic_fmin_x2 v[0:1], v[0:1], v[2:3] glc ; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[4:5], v[6:7] -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB16_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: v_mov_b32_e32 v0, v4 -; GFX10-NEXT: v_mov_b32_e32 v1, v5 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_ret_f64: ; GFX90A: ; %bb.0: ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX90A-NEXT: flat_load_dwordx2 v[4:5], v[0:1] -; GFX90A-NEXT: s_mov_b64 s[4:5], 0 -; GFX90A-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3] -; GFX90A-NEXT: .LBB16_1: ; %atomicrmw.start -; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX90A-NEXT: v_pk_mov_b32 v[6:7], v[4:5], v[4:5] op_sel:[0,1] -; GFX90A-NEXT: v_max_f64 v[4:5], v[6:7], v[6:7] -; GFX90A-NEXT: v_min_f64 v[4:5], v[4:5], v[2:3] -; GFX90A-NEXT: flat_atomic_cmpswap_x2 v[4:5], v[0:1], v[4:7] glc +; GFX90A-NEXT: flat_atomic_min_f64 v[0:1], v[0:1], v[2:3] glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 -; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[6:7] -; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX90A-NEXT: s_cbranch_execnz .LBB16_1 -; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX90A-NEXT: s_or_b64 exec, exec, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v0, v4 -; GFX90A-NEXT: v_mov_b32_e32 v1, v5 ; GFX90A-NEXT: s_setpc_b64 s[30:31] ; ; GFX908-LABEL: flat_agent_atomic_fmin_ret_f64: @@ -3659,30 +2825,9 @@ define double @flat_agent_atomic_fmin_ret_f64(ptr %ptr, double %val) #0 { ; GFX7-LABEL: flat_agent_atomic_fmin_ret_f64: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_add_i32_e32 v5, vcc, 4, v0 -; GFX7-NEXT: v_addc_u32_e32 v6, vcc, 0, v1, vcc -; GFX7-NEXT: flat_load_dword v4, v[0:1] -; GFX7-NEXT: flat_load_dword v5, v[5:6] -; GFX7-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: .LBB16_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mov_b32_e32 v7, v5 -; GFX7-NEXT: v_mov_b32_e32 v6, v4 -; GFX7-NEXT: v_max_f64 v[4:5], v[6:7], v[6:7] -; GFX7-NEXT: v_min_f64 v[4:5], v[4:5], v[2:3] -; GFX7-NEXT: flat_atomic_cmpswap_x2 v[4:5], v[0:1], v[4:7] glc +; GFX7-NEXT: flat_atomic_fmin_x2 v[0:1], v[0:1], v[2:3] glc ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[6:7] -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB16_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v0, v4 -; GFX7-NEXT: v_mov_b32_e32 v1, v5 ; GFX7-NEXT: s_setpc_b64 s[30:31] %result = atomicrmw fmin ptr %ptr, double %val syncscope("agent") seq_cst ret double %result @@ -3723,27 +2868,10 @@ define double @flat_agent_atomic_fmin_ret_f64__offset12b_pos(ptr %ptr, double %v ; GFX940-LABEL: flat_agent_atomic_fmin_ret_f64__offset12b_pos: ; GFX940: ; %bb.0: ; GFX940-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX940-NEXT: flat_load_dwordx2 v[4:5], v[0:1] offset:2040 -; GFX940-NEXT: s_mov_b64 s[0:1], 0 -; GFX940-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3] -; GFX940-NEXT: .LBB17_1: ; %atomicrmw.start -; GFX940-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX940-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX940-NEXT: v_mov_b64_e32 v[6:7], v[4:5] -; GFX940-NEXT: v_max_f64 v[4:5], v[6:7], v[6:7] -; GFX940-NEXT: v_min_f64 v[4:5], v[4:5], v[2:3] ; GFX940-NEXT: buffer_wbl2 sc1 -; GFX940-NEXT: flat_atomic_cmpswap_x2 v[4:5], v[0:1], v[4:7] offset:2040 sc0 +; GFX940-NEXT: flat_atomic_min_f64 v[0:1], v[0:1], v[2:3] offset:2040 sc0 ; GFX940-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX940-NEXT: buffer_inv sc1 -; GFX940-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[6:7] -; GFX940-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX940-NEXT: s_andn2_b64 exec, exec, s[0:1] -; GFX940-NEXT: s_cbranch_execnz .LBB17_1 -; GFX940-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX940-NEXT: s_or_b64 exec, exec, s[0:1] -; GFX940-NEXT: v_mov_b32_e32 v0, v4 -; GFX940-NEXT: v_mov_b32_e32 v1, v5 ; GFX940-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: flat_agent_atomic_fmin_ret_f64__offset12b_pos: @@ -3777,54 +2905,21 @@ define double @flat_agent_atomic_fmin_ret_f64__offset12b_pos(ptr %ptr, double %v ; GFX10-LABEL: flat_agent_atomic_fmin_ret_f64__offset12b_pos: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_co_u32 v4, vcc_lo, 0x7f8, v0 -; GFX10-NEXT: v_add_co_ci_u32_e32 v5, vcc_lo, 0, v1, vcc_lo -; GFX10-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3] -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: flat_load_dwordx2 v[0:1], v[4:5] -; GFX10-NEXT: .LBB17_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_mov_b32_e32 v9, v1 -; GFX10-NEXT: v_mov_b32_e32 v8, v0 -; GFX10-NEXT: v_max_f64 v[0:1], v[8:9], v[8:9] -; GFX10-NEXT: v_min_f64 v[6:7], v[0:1], v[2:3] +; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, 0x7f8, v0 +; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap_x2 v[0:1], v[4:5], v[6:9] glc +; GFX10-NEXT: flat_atomic_fmin_x2 v[0:1], v[0:1], v[2:3] glc ; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[8:9] -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB17_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_ret_f64__offset12b_pos: ; GFX90A: ; %bb.0: ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX90A-NEXT: flat_load_dwordx2 v[4:5], v[0:1] offset:2040 -; GFX90A-NEXT: s_mov_b64 s[4:5], 0 -; GFX90A-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3] -; GFX90A-NEXT: .LBB17_1: ; %atomicrmw.start -; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX90A-NEXT: v_pk_mov_b32 v[6:7], v[4:5], v[4:5] op_sel:[0,1] -; GFX90A-NEXT: v_max_f64 v[4:5], v[6:7], v[6:7] -; GFX90A-NEXT: v_min_f64 v[4:5], v[4:5], v[2:3] -; GFX90A-NEXT: flat_atomic_cmpswap_x2 v[4:5], v[0:1], v[4:7] offset:2040 glc +; GFX90A-NEXT: flat_atomic_min_f64 v[0:1], v[0:1], v[2:3] offset:2040 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 -; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[6:7] -; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX90A-NEXT: s_cbranch_execnz .LBB17_1 -; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX90A-NEXT: s_or_b64 exec, exec, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v0, v4 -; GFX90A-NEXT: v_mov_b32_e32 v1, v5 ; GFX90A-NEXT: s_setpc_b64 s[30:31] ; ; GFX908-LABEL: flat_agent_atomic_fmin_ret_f64__offset12b_pos: @@ -3885,30 +2980,11 @@ define double @flat_agent_atomic_fmin_ret_f64__offset12b_pos(ptr %ptr, double %v ; GFX7-LABEL: flat_agent_atomic_fmin_ret_f64__offset12b_pos: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_add_i32_e32 v4, vcc, 0x7f8, v0 -; GFX7-NEXT: v_addc_u32_e32 v5, vcc, 0, v1, vcc -; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0x7fc, v0 +; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0x7f8, v0 ; GFX7-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc -; GFX7-NEXT: flat_load_dword v1, v[0:1] -; GFX7-NEXT: flat_load_dword v0, v[4:5] -; GFX7-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: .LBB17_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mov_b32_e32 v9, v1 -; GFX7-NEXT: v_mov_b32_e32 v8, v0 -; GFX7-NEXT: v_max_f64 v[0:1], v[8:9], v[8:9] -; GFX7-NEXT: v_min_f64 v[6:7], v[0:1], v[2:3] -; GFX7-NEXT: flat_atomic_cmpswap_x2 v[0:1], v[4:5], v[6:9] glc +; GFX7-NEXT: flat_atomic_fmin_x2 v[0:1], v[0:1], v[2:3] glc ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[8:9] -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB17_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %gep = getelementptr double, ptr %ptr, i64 255 %result = atomicrmw fmin ptr %gep, double %val syncscope("agent") seq_cst @@ -3950,33 +3026,13 @@ define double @flat_agent_atomic_fmin_ret_f64__offset12b_neg(ptr %ptr, double %v ; GFX940-LABEL: flat_agent_atomic_fmin_ret_f64__offset12b_neg: ; GFX940: ; %bb.0: ; GFX940-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX940-NEXT: v_mov_b32_e32 v4, v0 -; GFX940-NEXT: v_mov_b32_e32 v5, v1 -; GFX940-NEXT: v_add_co_u32_e32 v0, vcc, 0xfffff800, v4 -; GFX940-NEXT: s_movk_i32 s0, 0xf800 -; GFX940-NEXT: s_nop 0 -; GFX940-NEXT: v_addc_co_u32_e32 v1, vcc, -1, v5, vcc -; GFX940-NEXT: flat_load_dwordx2 v[0:1], v[0:1] -; GFX940-NEXT: s_mov_b32 s1, -1 -; GFX940-NEXT: v_lshl_add_u64 v[4:5], v[4:5], 0, s[0:1] -; GFX940-NEXT: s_mov_b64 s[0:1], 0 -; GFX940-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3] -; GFX940-NEXT: .LBB18_1: ; %atomicrmw.start -; GFX940-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX940-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX940-NEXT: v_mov_b64_e32 v[8:9], v[0:1] -; GFX940-NEXT: v_max_f64 v[0:1], v[8:9], v[8:9] -; GFX940-NEXT: v_min_f64 v[6:7], v[0:1], v[2:3] +; GFX940-NEXT: v_add_co_u32_e32 v0, vcc, 0xfffff800, v0 +; GFX940-NEXT: s_nop 1 +; GFX940-NEXT: v_addc_co_u32_e32 v1, vcc, -1, v1, vcc ; GFX940-NEXT: buffer_wbl2 sc1 -; GFX940-NEXT: flat_atomic_cmpswap_x2 v[0:1], v[4:5], v[6:9] sc0 +; GFX940-NEXT: flat_atomic_min_f64 v[0:1], v[0:1], v[2:3] sc0 ; GFX940-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX940-NEXT: buffer_inv sc1 -; GFX940-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[8:9] -; GFX940-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX940-NEXT: s_andn2_b64 exec, exec, s[0:1] -; GFX940-NEXT: s_cbranch_execnz .LBB18_1 -; GFX940-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX940-NEXT: s_or_b64 exec, exec, s[0:1] ; GFX940-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: flat_agent_atomic_fmin_ret_f64__offset12b_neg: @@ -4015,56 +3071,23 @@ define double @flat_agent_atomic_fmin_ret_f64__offset12b_neg(ptr %ptr, double %v ; GFX10-LABEL: flat_agent_atomic_fmin_ret_f64__offset12b_neg: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_co_u32 v4, vcc_lo, 0xfffff800, v0 -; GFX10-NEXT: v_add_co_ci_u32_e32 v5, vcc_lo, -1, v1, vcc_lo -; GFX10-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3] -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: flat_load_dwordx2 v[0:1], v[4:5] -; GFX10-NEXT: .LBB18_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_mov_b32_e32 v9, v1 -; GFX10-NEXT: v_mov_b32_e32 v8, v0 -; GFX10-NEXT: v_max_f64 v[0:1], v[8:9], v[8:9] -; GFX10-NEXT: v_min_f64 v[6:7], v[0:1], v[2:3] +; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, 0xfffff800, v0 +; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, -1, v1, vcc_lo ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap_x2 v[0:1], v[4:5], v[6:9] glc +; GFX10-NEXT: flat_atomic_fmin_x2 v[0:1], v[0:1], v[2:3] glc ; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[0:1], v[8:9] -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB18_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_ret_f64__offset12b_neg: ; GFX90A: ; %bb.0: ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX90A-NEXT: v_add_co_u32_e32 v4, vcc, 0xfffff800, v0 -; GFX90A-NEXT: v_addc_co_u32_e32 v5, vcc, -1, v1, vcc ; GFX90A-NEXT: v_add_co_u32_e32 v0, vcc, 0xfffff800, v0 ; GFX90A-NEXT: v_addc_co_u32_e32 v1, vcc, -1, v1, vcc -; GFX90A-NEXT: flat_load_dwordx2 v[0:1], v[0:1] -; GFX90A-NEXT: s_mov_b64 s[4:5], 0 -; GFX90A-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3] -; GFX90A-NEXT: .LBB18_1: ; %atomicrmw.start -; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX90A-NEXT: v_pk_mov_b32 v[8:9], v[0:1], v[0:1] op_sel:[0,1] -; GFX90A-NEXT: v_max_f64 v[0:1], v[8:9], v[8:9] -; GFX90A-NEXT: v_min_f64 v[6:7], v[0:1], v[2:3] -; GFX90A-NEXT: flat_atomic_cmpswap_x2 v[0:1], v[4:5], v[6:9] glc +; GFX90A-NEXT: flat_atomic_min_f64 v[0:1], v[0:1], v[2:3] glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 -; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[8:9] -; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX90A-NEXT: s_cbranch_execnz .LBB18_1 -; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX90A-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_setpc_b64 s[30:31] ; ; GFX908-LABEL: flat_agent_atomic_fmin_ret_f64__offset12b_neg: @@ -4129,30 +3152,11 @@ define double @flat_agent_atomic_fmin_ret_f64__offset12b_neg(ptr %ptr, double %v ; GFX7-LABEL: flat_agent_atomic_fmin_ret_f64__offset12b_neg: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_add_i32_e32 v4, vcc, 0xfffff800, v0 -; GFX7-NEXT: v_addc_u32_e32 v5, vcc, -1, v1, vcc -; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0xfffff804, v0 +; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0xfffff800, v0 ; GFX7-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc -; GFX7-NEXT: flat_load_dword v1, v[0:1] -; GFX7-NEXT: flat_load_dword v0, v[4:5] -; GFX7-NEXT: v_max_f64 v[2:3], v[2:3], v[2:3] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: .LBB18_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_mov_b32_e32 v9, v1 -; GFX7-NEXT: v_mov_b32_e32 v8, v0 -; GFX7-NEXT: v_max_f64 v[0:1], v[8:9], v[8:9] -; GFX7-NEXT: v_min_f64 v[6:7], v[0:1], v[2:3] -; GFX7-NEXT: flat_atomic_cmpswap_x2 v[0:1], v[4:5], v[6:9] glc +; GFX7-NEXT: flat_atomic_fmin_x2 v[0:1], v[0:1], v[2:3] glc ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[8:9] -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB18_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %gep = getelementptr double, ptr %ptr, i64 -256 %result = atomicrmw fmin ptr %gep, double %val syncscope("agent") seq_cst @@ -4193,25 +3197,10 @@ define void @flat_agent_atomic_fmin_noret_f64(ptr %ptr, double %val) #0 { ; GFX940-LABEL: flat_agent_atomic_fmin_noret_f64: ; GFX940: ; %bb.0: ; GFX940-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX940-NEXT: flat_load_dwordx2 v[4:5], v[0:1] -; GFX940-NEXT: s_mov_b64 s[0:1], 0 -; GFX940-NEXT: v_max_f64 v[6:7], v[2:3], v[2:3] -; GFX940-NEXT: .LBB19_1: ; %atomicrmw.start -; GFX940-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX940-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX940-NEXT: v_max_f64 v[2:3], v[4:5], v[4:5] -; GFX940-NEXT: v_min_f64 v[2:3], v[2:3], v[6:7] ; GFX940-NEXT: buffer_wbl2 sc1 -; GFX940-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[0:1], v[2:5] sc0 +; GFX940-NEXT: flat_atomic_min_f64 v[0:1], v[2:3] ; GFX940-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX940-NEXT: buffer_inv sc1 -; GFX940-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] -; GFX940-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX940-NEXT: v_mov_b64_e32 v[4:5], v[2:3] -; GFX940-NEXT: s_andn2_b64 exec, exec, s[0:1] -; GFX940-NEXT: s_cbranch_execnz .LBB19_1 -; GFX940-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX940-NEXT: s_or_b64 exec, exec, s[0:1] ; GFX940-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: flat_agent_atomic_fmin_noret_f64: @@ -4244,50 +3233,20 @@ define void @flat_agent_atomic_fmin_noret_f64(ptr %ptr, double %val) #0 { ; GFX10-LABEL: flat_agent_atomic_fmin_noret_f64: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: flat_load_dwordx2 v[4:5], v[0:1] -; GFX10-NEXT: v_max_f64 v[6:7], v[2:3], v[2:3] -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: .LBB19_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_max_f64 v[2:3], v[4:5], v[4:5] -; GFX10-NEXT: v_min_f64 v[2:3], v[2:3], v[6:7] ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[0:1], v[2:5] glc -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX10-NEXT: flat_atomic_fmin_x2 v[0:1], v[2:3] +; GFX10-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5] -; GFX10-NEXT: v_mov_b32_e32 v5, v3 -; GFX10-NEXT: v_mov_b32_e32 v4, v2 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB19_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_noret_f64: ; GFX90A: ; %bb.0: ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX90A-NEXT: flat_load_dwordx2 v[4:5], v[0:1] -; GFX90A-NEXT: s_mov_b64 s[4:5], 0 -; GFX90A-NEXT: v_max_f64 v[6:7], v[2:3], v[2:3] -; GFX90A-NEXT: .LBB19_1: ; %atomicrmw.start -; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX90A-NEXT: v_max_f64 v[2:3], v[4:5], v[4:5] -; GFX90A-NEXT: v_min_f64 v[2:3], v[2:3], v[6:7] -; GFX90A-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[0:1], v[2:5] glc +; GFX90A-NEXT: flat_atomic_min_f64 v[0:1], v[2:3] ; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 -; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] -; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_pk_mov_b32 v[4:5], v[2:3], v[2:3] op_sel:[0,1] -; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX90A-NEXT: s_cbranch_execnz .LBB19_1 -; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX90A-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_setpc_b64 s[30:31] ; ; GFX908-LABEL: flat_agent_atomic_fmin_noret_f64: @@ -4344,28 +3303,9 @@ define void @flat_agent_atomic_fmin_noret_f64(ptr %ptr, double %val) #0 { ; GFX7-LABEL: flat_agent_atomic_fmin_noret_f64: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_add_i32_e32 v5, vcc, 4, v0 -; GFX7-NEXT: v_addc_u32_e32 v6, vcc, 0, v1, vcc -; GFX7-NEXT: flat_load_dword v4, v[0:1] -; GFX7-NEXT: flat_load_dword v5, v[5:6] -; GFX7-NEXT: v_max_f64 v[6:7], v[2:3], v[2:3] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: .LBB19_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_max_f64 v[2:3], v[4:5], v[4:5] -; GFX7-NEXT: v_min_f64 v[2:3], v[2:3], v[6:7] -; GFX7-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[0:1], v[2:5] glc +; GFX7-NEXT: flat_atomic_fmin_x2 v[0:1], v[2:3] ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] -; GFX7-NEXT: v_mov_b32_e32 v5, v3 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v4, v2 -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB19_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %unused = atomicrmw fmin ptr %ptr, double %val syncscope("agent") seq_cst ret void @@ -4405,25 +3345,10 @@ define void @flat_agent_atomic_fmin_noret_f64__offset12b_pos(ptr %ptr, double %v ; GFX940-LABEL: flat_agent_atomic_fmin_noret_f64__offset12b_pos: ; GFX940: ; %bb.0: ; GFX940-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX940-NEXT: flat_load_dwordx2 v[4:5], v[0:1] offset:2040 -; GFX940-NEXT: s_mov_b64 s[0:1], 0 -; GFX940-NEXT: v_max_f64 v[6:7], v[2:3], v[2:3] -; GFX940-NEXT: .LBB20_1: ; %atomicrmw.start -; GFX940-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX940-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX940-NEXT: v_max_f64 v[2:3], v[4:5], v[4:5] -; GFX940-NEXT: v_min_f64 v[2:3], v[2:3], v[6:7] ; GFX940-NEXT: buffer_wbl2 sc1 -; GFX940-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[0:1], v[2:5] offset:2040 sc0 +; GFX940-NEXT: flat_atomic_min_f64 v[0:1], v[2:3] offset:2040 ; GFX940-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX940-NEXT: buffer_inv sc1 -; GFX940-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] -; GFX940-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX940-NEXT: v_mov_b64_e32 v[4:5], v[2:3] -; GFX940-NEXT: s_andn2_b64 exec, exec, s[0:1] -; GFX940-NEXT: s_cbranch_execnz .LBB20_1 -; GFX940-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX940-NEXT: s_or_b64 exec, exec, s[0:1] ; GFX940-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: flat_agent_atomic_fmin_noret_f64__offset12b_pos: @@ -4458,50 +3383,20 @@ define void @flat_agent_atomic_fmin_noret_f64__offset12b_pos(ptr %ptr, double %v ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, 0x7f8, v0 ; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo -; GFX10-NEXT: v_max_f64 v[6:7], v[2:3], v[2:3] -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: flat_load_dwordx2 v[4:5], v[0:1] -; GFX10-NEXT: .LBB20_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_max_f64 v[2:3], v[4:5], v[4:5] -; GFX10-NEXT: v_min_f64 v[2:3], v[2:3], v[6:7] ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[0:1], v[2:5] glc -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX10-NEXT: flat_atomic_fmin_x2 v[0:1], v[2:3] +; GFX10-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5] -; GFX10-NEXT: v_mov_b32_e32 v5, v3 -; GFX10-NEXT: v_mov_b32_e32 v4, v2 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB20_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_noret_f64__offset12b_pos: ; GFX90A: ; %bb.0: ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX90A-NEXT: flat_load_dwordx2 v[4:5], v[0:1] offset:2040 -; GFX90A-NEXT: s_mov_b64 s[4:5], 0 -; GFX90A-NEXT: v_max_f64 v[6:7], v[2:3], v[2:3] -; GFX90A-NEXT: .LBB20_1: ; %atomicrmw.start -; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX90A-NEXT: v_max_f64 v[2:3], v[4:5], v[4:5] -; GFX90A-NEXT: v_min_f64 v[2:3], v[2:3], v[6:7] -; GFX90A-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[0:1], v[2:5] offset:2040 glc +; GFX90A-NEXT: flat_atomic_min_f64 v[0:1], v[2:3] offset:2040 ; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 -; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] -; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_pk_mov_b32 v[4:5], v[2:3], v[2:3] op_sel:[0,1] -; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX90A-NEXT: s_cbranch_execnz .LBB20_1 -; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX90A-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_setpc_b64 s[30:31] ; ; GFX908-LABEL: flat_agent_atomic_fmin_noret_f64__offset12b_pos: @@ -4560,30 +3455,11 @@ define void @flat_agent_atomic_fmin_noret_f64__offset12b_pos(ptr %ptr, double %v ; GFX7-LABEL: flat_agent_atomic_fmin_noret_f64__offset12b_pos: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_add_i32_e32 v6, vcc, 0x7f8, v0 -; GFX7-NEXT: v_addc_u32_e32 v7, vcc, 0, v1, vcc -; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0x7fc, v0 +; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0x7f8, v0 ; GFX7-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc -; GFX7-NEXT: flat_load_dword v5, v[0:1] -; GFX7-NEXT: flat_load_dword v4, v[6:7] -; GFX7-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: .LBB20_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_max_f64 v[2:3], v[4:5], v[4:5] -; GFX7-NEXT: v_min_f64 v[2:3], v[2:3], v[0:1] -; GFX7-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[6:7], v[2:5] glc +; GFX7-NEXT: flat_atomic_fmin_x2 v[0:1], v[2:3] ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] -; GFX7-NEXT: v_mov_b32_e32 v5, v3 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v4, v2 -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB20_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %gep = getelementptr double, ptr %ptr, i64 255 %unused = atomicrmw fmin ptr %gep, double %val syncscope("agent") seq_cst @@ -4624,31 +3500,13 @@ define void @flat_agent_atomic_fmin_noret_f64__offset12b_neg(ptr %ptr, double %v ; GFX940-LABEL: flat_agent_atomic_fmin_noret_f64__offset12b_neg: ; GFX940: ; %bb.0: ; GFX940-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX940-NEXT: v_add_co_u32_e32 v4, vcc, 0xfffff800, v0 -; GFX940-NEXT: s_movk_i32 s0, 0xf800 -; GFX940-NEXT: s_nop 0 -; GFX940-NEXT: v_addc_co_u32_e32 v5, vcc, -1, v1, vcc -; GFX940-NEXT: flat_load_dwordx2 v[4:5], v[4:5] -; GFX940-NEXT: s_mov_b32 s1, -1 -; GFX940-NEXT: v_lshl_add_u64 v[0:1], v[0:1], 0, s[0:1] -; GFX940-NEXT: s_mov_b64 s[0:1], 0 -; GFX940-NEXT: v_max_f64 v[6:7], v[2:3], v[2:3] -; GFX940-NEXT: .LBB21_1: ; %atomicrmw.start -; GFX940-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX940-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX940-NEXT: v_max_f64 v[2:3], v[4:5], v[4:5] -; GFX940-NEXT: v_min_f64 v[2:3], v[2:3], v[6:7] +; GFX940-NEXT: v_add_co_u32_e32 v0, vcc, 0xfffff800, v0 +; GFX940-NEXT: s_nop 1 +; GFX940-NEXT: v_addc_co_u32_e32 v1, vcc, -1, v1, vcc ; GFX940-NEXT: buffer_wbl2 sc1 -; GFX940-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[0:1], v[2:5] sc0 +; GFX940-NEXT: flat_atomic_min_f64 v[0:1], v[2:3] ; GFX940-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX940-NEXT: buffer_inv sc1 -; GFX940-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] -; GFX940-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX940-NEXT: v_mov_b64_e32 v[4:5], v[2:3] -; GFX940-NEXT: s_andn2_b64 exec, exec, s[0:1] -; GFX940-NEXT: s_cbranch_execnz .LBB21_1 -; GFX940-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX940-NEXT: s_or_b64 exec, exec, s[0:1] ; GFX940-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: flat_agent_atomic_fmin_noret_f64__offset12b_neg: @@ -4687,54 +3545,22 @@ define void @flat_agent_atomic_fmin_noret_f64__offset12b_neg(ptr %ptr, double %v ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, 0xfffff800, v0 ; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, -1, v1, vcc_lo -; GFX10-NEXT: v_max_f64 v[6:7], v[2:3], v[2:3] -; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: flat_load_dwordx2 v[4:5], v[0:1] -; GFX10-NEXT: .LBB21_1: ; %atomicrmw.start -; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_max_f64 v[2:3], v[4:5], v[4:5] -; GFX10-NEXT: v_min_f64 v[2:3], v[2:3], v[6:7] ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[0:1], v[2:5] glc -; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) +; GFX10-NEXT: flat_atomic_fmin_x2 v[0:1], v[2:3] +; GFX10-NEXT: s_waitcnt lgkmcnt(0) +; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX10-NEXT: buffer_gl1_inv ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[2:3], v[4:5] -; GFX10-NEXT: v_mov_b32_e32 v5, v3 -; GFX10-NEXT: v_mov_b32_e32 v4, v2 -; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 -; GFX10-NEXT: s_cbranch_execnz .LBB21_1 -; GFX10-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX90A-LABEL: flat_agent_atomic_fmin_noret_f64__offset12b_neg: ; GFX90A: ; %bb.0: ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX90A-NEXT: v_add_co_u32_e32 v6, vcc, 0xfffff800, v0 -; GFX90A-NEXT: v_addc_co_u32_e32 v7, vcc, -1, v1, vcc ; GFX90A-NEXT: v_add_co_u32_e32 v0, vcc, 0xfffff800, v0 ; GFX90A-NEXT: v_addc_co_u32_e32 v1, vcc, -1, v1, vcc -; GFX90A-NEXT: flat_load_dwordx2 v[4:5], v[0:1] -; GFX90A-NEXT: s_mov_b64 s[4:5], 0 -; GFX90A-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3] -; GFX90A-NEXT: .LBB21_1: ; %atomicrmw.start -; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX90A-NEXT: v_max_f64 v[2:3], v[4:5], v[4:5] -; GFX90A-NEXT: v_min_f64 v[2:3], v[2:3], v[0:1] -; GFX90A-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[6:7], v[2:5] glc +; GFX90A-NEXT: flat_atomic_min_f64 v[0:1], v[2:3] ; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX90A-NEXT: buffer_wbinvl1 -; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] -; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_pk_mov_b32 v[4:5], v[2:3], v[2:3] op_sel:[0,1] -; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX90A-NEXT: s_cbranch_execnz .LBB21_1 -; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX90A-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_setpc_b64 s[30:31] ; ; GFX908-LABEL: flat_agent_atomic_fmin_noret_f64__offset12b_neg: @@ -4798,30 +3624,11 @@ define void @flat_agent_atomic_fmin_noret_f64__offset12b_neg(ptr %ptr, double %v ; GFX7-LABEL: flat_agent_atomic_fmin_noret_f64__offset12b_neg: ; GFX7: ; %bb.0: ; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_add_i32_e32 v6, vcc, 0xfffff800, v0 -; GFX7-NEXT: v_addc_u32_e32 v7, vcc, -1, v1, vcc -; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0xfffff804, v0 +; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0xfffff800, v0 ; GFX7-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc -; GFX7-NEXT: flat_load_dword v5, v[0:1] -; GFX7-NEXT: flat_load_dword v4, v[6:7] -; GFX7-NEXT: v_max_f64 v[0:1], v[2:3], v[2:3] -; GFX7-NEXT: s_mov_b64 s[4:5], 0 -; GFX7-NEXT: .LBB21_1: ; %atomicrmw.start -; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 -; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; GFX7-NEXT: v_max_f64 v[2:3], v[4:5], v[4:5] -; GFX7-NEXT: v_min_f64 v[2:3], v[2:3], v[0:1] -; GFX7-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[6:7], v[2:5] glc +; GFX7-NEXT: flat_atomic_fmin_x2 v[0:1], v[2:3] ; GFX7-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX7-NEXT: buffer_wbinvl1 -; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] -; GFX7-NEXT: v_mov_b32_e32 v5, v3 -; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v4, v2 -; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] -; GFX7-NEXT: s_cbranch_execnz .LBB21_1 -; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end -; GFX7-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_setpc_b64 s[30:31] %gep = getelementptr double, ptr %ptr, i64 -256 %unused = atomicrmw fmin ptr %gep, double %val syncscope("agent") seq_cst |