diff --git a/llvm/test/CodeGen/AMDGPU/sdiv.ll b/llvm/test/CodeGen/AMDGPU/sdiv.ll index a67e83035a67..03784dcba51f 100644 --- a/llvm/test/CodeGen/AMDGPU/sdiv.ll +++ b/llvm/test/CodeGen/AMDGPU/sdiv.ll @@ -1,7 +1,8 @@ -; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s -; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s -; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s -; RUN: llc -amdgpu-scalarize-global-loads=false -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc < %s -amdgpu-scalarize-global-loads=false -march=amdgcn | FileCheck %s -check-prefixes=FUNC,SI,GCN +; RUN: llc < %s -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga -mattr=-flat-for-global | FileCheck %s -check-prefixes=FUNC,SI,TONGA +; RUN: llc < %s -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global | FileCheck %s -check-prefixes=FUNC,SI,GFX9 +; RUN: llc < %s -amdgpu-scalarize-global-loads=false -march=r600 -mcpu=redwood | FileCheck %s -check-prefixes=FUNC,EG ; The code generated by sdiv is long and complex and may frequently change. ; The goal of this test is to make sure the ISel doesn't fail. @@ -12,9 +13,198 @@ ; This was fixed by adding an additional pattern in R600Instructions.td to ; match this pattern with a CNDGE_INT. -; FUNC-LABEL: {{^}}sdiv_i32: -; EG: CF_END define amdgpu_kernel void @sdiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { +; GCN-LABEL: sdiv_i32: +; GCN: ; %bb.0: +; GCN-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: s_mov_b32 s10, s6 +; GCN-NEXT: s_mov_b32 s11, s7 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b32 s8, s2 +; GCN-NEXT: s_mov_b32 s9, s3 +; GCN-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: s_mov_b32 s4, s0 +; GCN-NEXT: s_mov_b32 s5, s1 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_ashrrev_i32_e32 v2, 31, v0 +; GCN-NEXT: v_ashrrev_i32_e32 v3, 31, v1 +; GCN-NEXT: v_xor_b32_e32 v4, v2, v3 +; GCN-NEXT: v_add_i32_e32 v0, vcc, v2, v0 +; GCN-NEXT: v_add_i32_e32 v1, vcc, v3, v1 +; GCN-NEXT: v_xor_b32_e32 v0, v0, v2 +; GCN-NEXT: v_xor_b32_e32 v1, v1, v3 +; GCN-NEXT: v_cvt_f32_u32_e32 v2, v1 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v2 +; GCN-NEXT: v_mul_f32_e32 v2, 0x4f800000, v2 +; GCN-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-NEXT: v_mul_hi_u32 v3, v2, v1 +; GCN-NEXT: v_mul_lo_u32 v5, v2, v1 +; GCN-NEXT: v_sub_i32_e32 v6, vcc, 0, v5 +; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v3 +; GCN-NEXT: v_cndmask_b32_e64 v3, v5, v6, s[0:1] +; GCN-NEXT: v_mul_hi_u32 v3, v3, v2 +; GCN-NEXT: v_add_i32_e32 v5, vcc, v3, v2 +; GCN-NEXT: v_subrev_i32_e32 v2, vcc, v3, v2 +; GCN-NEXT: v_cndmask_b32_e64 v2, v2, v5, s[0:1] +; GCN-NEXT: v_mul_hi_u32 v2, v2, v0 +; GCN-NEXT: v_mul_lo_u32 v3, v2, v1 +; GCN-NEXT: v_add_i32_e32 v5, vcc, 1, v2 +; GCN-NEXT: v_add_i32_e32 v6, vcc, -1, v2 +; GCN-NEXT: v_subrev_i32_e32 v7, vcc, v3, v0 +; GCN-NEXT: v_cmp_ge_u32_e32 vcc, v0, v3 +; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], v7, v1 +; GCN-NEXT: s_and_b64 s[0:1], s[0:1], vcc +; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v5, s[0:1] +; GCN-NEXT: v_cndmask_b32_e32 v0, v6, v0, vcc +; GCN-NEXT: v_xor_b32_e32 v0, v0, v4 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, v0, v4 +; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; TONGA-LABEL: sdiv_i32: +; TONGA: ; %bb.0: +; TONGA-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x24 +; TONGA-NEXT: s_mov_b32 s7, 0xf000 +; TONGA-NEXT: s_mov_b32 s6, -1 +; TONGA-NEXT: s_mov_b32 s2, s6 +; TONGA-NEXT: s_mov_b32 s3, s7 +; TONGA-NEXT: s_waitcnt lgkmcnt(0) +; TONGA-NEXT: s_mov_b32 s0, s10 +; TONGA-NEXT: s_mov_b32 s1, s11 +; TONGA-NEXT: buffer_load_dwordx2 v[0:1], off, s[0:3], 0 +; TONGA-NEXT: s_mov_b32 s4, s8 +; TONGA-NEXT: s_mov_b32 s5, s9 +; TONGA-NEXT: s_waitcnt vmcnt(0) +; TONGA-NEXT: v_ashrrev_i32_e32 v2, 31, v1 +; TONGA-NEXT: v_add_u32_e32 v1, vcc, v2, v1 +; TONGA-NEXT: v_xor_b32_e32 v1, v1, v2 +; TONGA-NEXT: v_cvt_f32_u32_e32 v3, v1 +; TONGA-NEXT: v_ashrrev_i32_e32 v6, 31, v0 +; TONGA-NEXT: v_add_u32_e32 v0, vcc, v6, v0 +; TONGA-NEXT: v_xor_b32_e32 v0, v0, v6 +; TONGA-NEXT: v_rcp_iflag_f32_e32 v3, v3 +; TONGA-NEXT: v_xor_b32_e32 v2, v6, v2 +; TONGA-NEXT: v_mul_f32_e32 v3, 0x4f800000, v3 +; TONGA-NEXT: v_cvt_u32_f32_e32 v3, v3 +; TONGA-NEXT: v_mul_lo_u32 v4, v3, v1 +; TONGA-NEXT: v_mul_hi_u32 v5, v3, v1 +; TONGA-NEXT: v_sub_u32_e32 v7, vcc, 0, v4 +; TONGA-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v5 +; TONGA-NEXT: v_cndmask_b32_e64 v4, v4, v7, s[0:1] +; TONGA-NEXT: v_mul_hi_u32 v4, v4, v3 +; TONGA-NEXT: v_add_u32_e32 v5, vcc, v4, v3 +; TONGA-NEXT: v_subrev_u32_e32 v3, vcc, v4, v3 +; TONGA-NEXT: v_cndmask_b32_e64 v3, v3, v5, s[0:1] +; TONGA-NEXT: v_mul_hi_u32 v3, v3, v0 +; TONGA-NEXT: v_mul_lo_u32 v4, v3, v1 +; TONGA-NEXT: v_add_u32_e32 v5, vcc, 1, v3 +; TONGA-NEXT: v_add_u32_e32 v6, vcc, -1, v3 +; TONGA-NEXT: v_subrev_u32_e32 v7, vcc, v4, v0 +; TONGA-NEXT: v_cmp_ge_u32_e32 vcc, v0, v4 +; TONGA-NEXT: v_cmp_ge_u32_e64 s[0:1], v7, v1 +; TONGA-NEXT: s_and_b64 s[0:1], s[0:1], vcc +; TONGA-NEXT: v_cndmask_b32_e64 v0, v3, v5, s[0:1] +; TONGA-NEXT: v_cndmask_b32_e32 v0, v6, v0, vcc +; TONGA-NEXT: v_xor_b32_e32 v0, v0, v2 +; TONGA-NEXT: v_sub_u32_e32 v0, vcc, v0, v2 +; TONGA-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; TONGA-NEXT: s_endpgm +; +; GFX9-LABEL: sdiv_i32: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; GFX9-NEXT: s_mov_b32 s7, 0xf000 +; GFX9-NEXT: s_mov_b32 s6, -1 +; GFX9-NEXT: s_mov_b32 s10, s6 +; GFX9-NEXT: s_mov_b32 s11, s7 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s8, s2 +; GFX9-NEXT: s_mov_b32 s9, s3 +; GFX9-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0 +; GFX9-NEXT: s_mov_b32 s4, s0 +; GFX9-NEXT: s_mov_b32 s5, s1 +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_ashrrev_i32_e32 v2, 31, v1 +; GFX9-NEXT: v_add_u32_e32 v1, v1, v2 +; GFX9-NEXT: v_xor_b32_e32 v1, v1, v2 +; GFX9-NEXT: v_cvt_f32_u32_e32 v3, v1 +; GFX9-NEXT: v_rcp_iflag_f32_e32 v3, v3 +; GFX9-NEXT: v_mul_f32_e32 v3, 0x4f800000, v3 +; GFX9-NEXT: v_cvt_u32_f32_e32 v3, v3 +; GFX9-NEXT: v_mul_lo_u32 v4, v3, v1 +; GFX9-NEXT: v_mul_hi_u32 v5, v3, v1 +; GFX9-NEXT: v_sub_u32_e32 v6, 0, v4 +; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v5 +; GFX9-NEXT: v_cndmask_b32_e32 v4, v4, v6, vcc +; GFX9-NEXT: v_mul_hi_u32 v4, v4, v3 +; GFX9-NEXT: v_ashrrev_i32_e32 v5, 31, v0 +; GFX9-NEXT: v_add_u32_e32 v0, v0, v5 +; GFX9-NEXT: v_xor_b32_e32 v0, v0, v5 +; GFX9-NEXT: v_add_u32_e32 v6, v3, v4 +; GFX9-NEXT: v_sub_u32_e32 v3, v3, v4 +; GFX9-NEXT: v_cndmask_b32_e32 v3, v3, v6, vcc +; GFX9-NEXT: v_mul_hi_u32 v3, v3, v0 +; GFX9-NEXT: v_xor_b32_e32 v2, v5, v2 +; GFX9-NEXT: v_mul_lo_u32 v4, v3, v1 +; GFX9-NEXT: v_add_u32_e32 v5, 1, v3 +; GFX9-NEXT: v_add_u32_e32 v6, -1, v3 +; GFX9-NEXT: v_sub_u32_e32 v7, v0, v4 +; GFX9-NEXT: v_cmp_ge_u32_e32 vcc, v0, v4 +; GFX9-NEXT: v_cmp_ge_u32_e64 s[0:1], v7, v1 +; GFX9-NEXT: s_and_b64 s[0:1], s[0:1], vcc +; GFX9-NEXT: v_cndmask_b32_e64 v0, v3, v5, s[0:1] +; GFX9-NEXT: v_cndmask_b32_e32 v0, v6, v0, vcc +; GFX9-NEXT: v_xor_b32_e32 v0, v0, v2 +; GFX9-NEXT: v_sub_u32_e32 v0, v0, v2 +; GFX9-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX9-NEXT: s_endpgm +; +; EG-LABEL: sdiv_i32: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 30, @9, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 6: +; EG-NEXT: VTX_READ_64 T0.XY, T0.X, 0, #1 +; EG-NEXT: ALU clause starting at 8: +; EG-NEXT: MOV * T0.X, KC0[2].Z, +; EG-NEXT: ALU clause starting at 9: +; EG-NEXT: SETGT_INT * T0.W, 0.0, T0.Y, +; EG-NEXT: ADD_INT * T1.W, T0.Y, PV.W, +; EG-NEXT: XOR_INT * T1.W, PV.W, T0.W, +; EG-NEXT: RECIP_UINT * T0.Y, PV.W, +; EG-NEXT: MULLO_INT * T0.Z, PS, T1.W, +; EG-NEXT: SUB_INT T2.W, 0.0, PS, +; EG-NEXT: MULHI * T1.X, T0.Y, T1.W, +; EG-NEXT: CNDE_INT T2.W, PS, PV.W, T0.Z, +; EG-NEXT: SETGT_INT * T3.W, 0.0, T0.X, +; EG-NEXT: MULHI * T0.Z, PV.W, T0.Y, +; EG-NEXT: ADD_INT T1.Z, T0.X, T3.W, +; EG-NEXT: ADD_INT T2.W, T0.Y, PS, +; EG-NEXT: SUB_INT * T4.W, T0.Y, PS, +; EG-NEXT: CNDE_INT T2.W, T1.X, PV.W, PS, +; EG-NEXT: XOR_INT * T4.W, PV.Z, T3.W, +; EG-NEXT: MULHI * T0.X, PV.W, PS, +; EG-NEXT: MULLO_INT * T0.Y, PS, T1.W, +; EG-NEXT: SUB_INT * T2.W, T4.W, PS, +; EG-NEXT: SETGE_UINT T1.W, PV.W, T1.W, +; EG-NEXT: SETGE_UINT * T2.W, T4.W, T0.Y, +; EG-NEXT: AND_INT T1.W, PV.W, PS, +; EG-NEXT: ADD_INT * T4.W, T0.X, 1, +; EG-NEXT: CNDE_INT T1.W, PV.W, T0.X, PS, +; EG-NEXT: ADD_INT * T4.W, T0.X, literal.x, +; EG-NEXT: -1(nan), 0(0.000000e+00) +; EG-NEXT: CNDE_INT T1.W, T2.W, PS, PV.W, +; EG-NEXT: XOR_INT * T0.W, T3.W, T0.W, +; EG-NEXT: XOR_INT * T1.W, PV.W, PS, +; EG-NEXT: SUB_INT T0.X, PV.W, T0.W, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %den_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1 %num = load i32, i32 addrspace(1) * %in %den = load i32, i32 addrspace(1) * %den_ptr @@ -23,8 +213,91 @@ define amdgpu_kernel void @sdiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %i ret void } -; FUNC-LABEL: {{^}}sdiv_i32_4: define amdgpu_kernel void @sdiv_i32_4(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { +; GCN-LABEL: sdiv_i32_4: +; GCN: ; %bb.0: +; GCN-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: s_mov_b32 s10, s6 +; GCN-NEXT: s_mov_b32 s11, s7 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b32 s8, s2 +; GCN-NEXT: s_mov_b32 s9, s3 +; GCN-NEXT: buffer_load_dword v0, off, s[8:11], 0 +; GCN-NEXT: s_mov_b32 s4, s0 +; GCN-NEXT: s_mov_b32 s5, s1 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-NEXT: v_lshrrev_b32_e32 v1, 30, v1 +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v1 +; GCN-NEXT: v_ashrrev_i32_e32 v0, 2, v0 +; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; TONGA-LABEL: sdiv_i32_4: +; TONGA: ; %bb.0: +; TONGA-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; TONGA-NEXT: s_mov_b32 s3, 0xf000 +; TONGA-NEXT: s_mov_b32 s2, -1 +; TONGA-NEXT: s_waitcnt lgkmcnt(0) +; TONGA-NEXT: s_mov_b32 s0, s4 +; TONGA-NEXT: s_mov_b32 s1, s5 +; TONGA-NEXT: s_mov_b32 s4, s6 +; TONGA-NEXT: s_mov_b32 s5, s7 +; TONGA-NEXT: s_mov_b32 s6, s2 +; TONGA-NEXT: s_mov_b32 s7, s3 +; TONGA-NEXT: buffer_load_dword v0, off, s[4:7], 0 +; TONGA-NEXT: s_waitcnt vmcnt(0) +; TONGA-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; TONGA-NEXT: v_lshrrev_b32_e32 v1, 30, v1 +; TONGA-NEXT: v_add_u32_e32 v0, vcc, v0, v1 +; TONGA-NEXT: v_ashrrev_i32_e32 v0, 2, v0 +; TONGA-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; TONGA-NEXT: s_endpgm +; +; GFX9-LABEL: sdiv_i32_4: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; GFX9-NEXT: s_mov_b32 s3, 0xf000 +; GFX9-NEXT: s_mov_b32 s2, -1 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s0, s4 +; GFX9-NEXT: s_mov_b32 s1, s5 +; GFX9-NEXT: s_mov_b32 s4, s6 +; GFX9-NEXT: s_mov_b32 s5, s7 +; GFX9-NEXT: s_mov_b32 s6, s2 +; GFX9-NEXT: s_mov_b32 s7, s3 +; GFX9-NEXT: buffer_load_dword v0, off, s[4:7], 0 +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GFX9-NEXT: v_lshrrev_b32_e32 v1, 30, v1 +; GFX9-NEXT: v_add_u32_e32 v0, v0, v1 +; GFX9-NEXT: v_ashrrev_i32_e32 v0, 2, v0 +; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX9-NEXT: s_endpgm +; +; EG-LABEL: sdiv_i32_4: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 7, @9, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 6: +; EG-NEXT: VTX_READ_32 T0.X, T0.X, 0, #1 +; EG-NEXT: ALU clause starting at 8: +; EG-NEXT: MOV * T0.X, KC0[2].Z, +; EG-NEXT: ALU clause starting at 9: +; EG-NEXT: ASHR * T0.W, T0.X, literal.x, +; EG-NEXT: 31(4.344025e-44), 0(0.000000e+00) +; EG-NEXT: LSHR * T0.W, PV.W, literal.x, +; EG-NEXT: 30(4.203895e-44), 0(0.000000e+00) +; EG-NEXT: ADD_INT * T0.W, T0.X, PV.W, +; EG-NEXT: ASHR T0.X, PV.W, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %num = load i32, i32 addrspace(1) * %in %result = sdiv i32 %num, 4 store i32 %result, i32 addrspace(1)* %out @@ -34,17 +307,98 @@ define amdgpu_kernel void @sdiv_i32_4(i32 addrspace(1)* %out, i32 addrspace(1)* ; Multiply by a weird constant to make sure setIntDivIsCheap is ; working. -; FUNC-LABEL: {{^}}slow_sdiv_i32_3435: -; SI-DAG: buffer_load_dword [[VAL:v[0-9]+]], -; SI-DAG: s_mov_b32 [[MAGIC:s[0-9]+]], 0x98a1930b -; SI: v_mul_hi_i32 [[TMP:v[0-9]+]], [[VAL]], [[MAGIC]] -; SI: v_add_{{[iu]}}32 -; SI: v_lshrrev_b32 -; SI: v_ashrrev_i32 -; SI: v_add_{{[iu]}}32 -; SI: buffer_store_dword -; SI: s_endpgm define amdgpu_kernel void @slow_sdiv_i32_3435(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { +; GCN-LABEL: slow_sdiv_i32_3435: +; GCN: ; %bb.0: +; GCN-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: s_mov_b32 s10, s6 +; GCN-NEXT: s_mov_b32 s11, s7 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b32 s8, s2 +; GCN-NEXT: s_mov_b32 s9, s3 +; GCN-NEXT: buffer_load_dword v0, off, s[8:11], 0 +; GCN-NEXT: s_mov_b32 s2, 0x98a1930b +; GCN-NEXT: s_mov_b32 s4, s0 +; GCN-NEXT: s_mov_b32 s5, s1 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_mul_hi_i32 v1, v0, s2 +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v1 +; GCN-NEXT: v_lshrrev_b32_e32 v1, 31, v0 +; GCN-NEXT: v_ashrrev_i32_e32 v0, 11, v0 +; GCN-NEXT: v_add_i32_e32 v0, vcc, v1, v0 +; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; TONGA-LABEL: slow_sdiv_i32_3435: +; TONGA: ; %bb.0: +; TONGA-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; TONGA-NEXT: s_mov_b32 s3, 0xf000 +; TONGA-NEXT: s_mov_b32 s2, -1 +; TONGA-NEXT: s_mov_b32 s10, s2 +; TONGA-NEXT: s_mov_b32 s11, s3 +; TONGA-NEXT: s_waitcnt lgkmcnt(0) +; TONGA-NEXT: s_mov_b32 s8, s6 +; TONGA-NEXT: s_mov_b32 s9, s7 +; TONGA-NEXT: buffer_load_dword v0, off, s[8:11], 0 +; TONGA-NEXT: s_mov_b32 s0, 0x98a1930b +; TONGA-NEXT: s_mov_b32 s1, s5 +; TONGA-NEXT: s_waitcnt vmcnt(0) +; TONGA-NEXT: v_mul_hi_i32 v1, v0, s0 +; TONGA-NEXT: s_mov_b32 s0, s4 +; TONGA-NEXT: v_add_u32_e32 v0, vcc, v0, v1 +; TONGA-NEXT: v_lshrrev_b32_e32 v1, 31, v0 +; TONGA-NEXT: v_ashrrev_i32_e32 v0, 11, v0 +; TONGA-NEXT: v_add_u32_e32 v0, vcc, v1, v0 +; TONGA-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; TONGA-NEXT: s_endpgm +; +; GFX9-LABEL: slow_sdiv_i32_3435: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; GFX9-NEXT: s_mov_b32 s3, 0xf000 +; GFX9-NEXT: s_mov_b32 s2, -1 +; GFX9-NEXT: s_mov_b32 s10, s2 +; GFX9-NEXT: s_mov_b32 s11, s3 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s8, s6 +; GFX9-NEXT: s_mov_b32 s9, s7 +; GFX9-NEXT: buffer_load_dword v0, off, s[8:11], 0 +; GFX9-NEXT: s_mov_b32 s0, 0x98a1930b +; GFX9-NEXT: s_mov_b32 s1, s5 +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_mul_hi_i32 v1, v0, s0 +; GFX9-NEXT: s_mov_b32 s0, s4 +; GFX9-NEXT: v_add_u32_e32 v0, v1, v0 +; GFX9-NEXT: v_lshrrev_b32_e32 v1, 31, v0 +; GFX9-NEXT: v_ashrrev_i32_e32 v0, 11, v0 +; GFX9-NEXT: v_add_u32_e32 v0, v0, v1 +; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX9-NEXT: s_endpgm +; +; EG-LABEL: slow_sdiv_i32_3435: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 8, @9, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 6: +; EG-NEXT: VTX_READ_32 T0.X, T0.X, 0, #1 +; EG-NEXT: ALU clause starting at 8: +; EG-NEXT: MOV * T0.X, KC0[2].Z, +; EG-NEXT: ALU clause starting at 9: +; EG-NEXT: MULHI_INT * T0.Y, T0.X, literal.x, +; EG-NEXT: -1734241525(-4.176600e-24), 0(0.000000e+00) +; EG-NEXT: ADD_INT * T0.W, PS, T0.X, +; EG-NEXT: ASHR T1.W, PV.W, literal.x, +; EG-NEXT: LSHR * T0.W, PV.W, literal.y, +; EG-NEXT: 11(1.541428e-44), 31(4.344025e-44) +; EG-NEXT: ADD_INT T0.X, PV.W, PS, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %num = load i32, i32 addrspace(1) * %in %result = sdiv i32 %num, 3435 store i32 %result, i32 addrspace(1)* %out @@ -52,6 +406,326 @@ define amdgpu_kernel void @slow_sdiv_i32_3435(i32 addrspace(1)* %out, i32 addrsp } define amdgpu_kernel void @sdiv_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) { +; GCN-LABEL: sdiv_v2i32: +; GCN: ; %bb.0: +; GCN-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-NEXT: s_mov_b32 s11, 0xf000 +; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_mov_b32 s6, s10 +; GCN-NEXT: s_mov_b32 s7, s11 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b32 s4, s2 +; GCN-NEXT: s_mov_b32 s5, s3 +; GCN-NEXT: buffer_load_dwordx4 v[0:3], off, s[4:7], 0 +; GCN-NEXT: s_mov_b32 s2, 0x4f800000 +; GCN-NEXT: s_mov_b32 s8, s0 +; GCN-NEXT: s_mov_b32 s9, s1 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_ashrrev_i32_e32 v4, 31, v0 +; GCN-NEXT: v_ashrrev_i32_e32 v5, 31, v2 +; GCN-NEXT: v_ashrrev_i32_e32 v6, 31, v1 +; GCN-NEXT: v_ashrrev_i32_e32 v7, 31, v3 +; GCN-NEXT: v_xor_b32_e32 v8, v4, v5 +; GCN-NEXT: v_add_i32_e32 v0, vcc, v4, v0 +; GCN-NEXT: v_add_i32_e32 v2, vcc, v5, v2 +; GCN-NEXT: v_xor_b32_e32 v9, v6, v7 +; GCN-NEXT: v_add_i32_e32 v1, vcc, v6, v1 +; GCN-NEXT: v_add_i32_e32 v3, vcc, v7, v3 +; GCN-NEXT: v_xor_b32_e32 v0, v0, v4 +; GCN-NEXT: v_xor_b32_e32 v2, v2, v5 +; GCN-NEXT: v_xor_b32_e32 v1, v1, v6 +; GCN-NEXT: v_xor_b32_e32 v3, v3, v7 +; GCN-NEXT: v_cvt_f32_u32_e32 v4, v2 +; GCN-NEXT: v_cvt_f32_u32_e32 v5, v3 +; GCN-NEXT: v_rcp_iflag_f32_e32 v4, v4 +; GCN-NEXT: v_rcp_iflag_f32_e32 v5, v5 +; GCN-NEXT: v_mul_f32_e32 v4, s2, v4 +; GCN-NEXT: v_mul_f32_e32 v5, s2, v5 +; GCN-NEXT: v_cvt_u32_f32_e32 v4, v4 +; GCN-NEXT: v_cvt_u32_f32_e32 v5, v5 +; GCN-NEXT: v_mul_hi_u32 v6, v4, v2 +; GCN-NEXT: v_mul_lo_u32 v7, v4, v2 +; GCN-NEXT: v_mul_hi_u32 v10, v5, v3 +; GCN-NEXT: v_mul_lo_u32 v11, v5, v3 +; GCN-NEXT: v_sub_i32_e32 v12, vcc, 0, v7 +; GCN-NEXT: v_sub_i32_e32 v13, vcc, 0, v11 +; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v6 +; GCN-NEXT: v_cndmask_b32_e64 v6, v7, v12, s[0:1] +; GCN-NEXT: v_cmp_eq_u32_e64 s[2:3], 0, v10 +; GCN-NEXT: v_cndmask_b32_e64 v7, v11, v13, s[2:3] +; GCN-NEXT: v_mul_hi_u32 v6, v6, v4 +; GCN-NEXT: v_mul_hi_u32 v7, v7, v5 +; GCN-NEXT: v_add_i32_e32 v10, vcc, v6, v4 +; GCN-NEXT: v_subrev_i32_e32 v4, vcc, v6, v4 +; GCN-NEXT: v_add_i32_e32 v6, vcc, v7, v5 +; GCN-NEXT: v_subrev_i32_e32 v5, vcc, v7, v5 +; GCN-NEXT: v_cndmask_b32_e64 v4, v4, v10, s[0:1] +; GCN-NEXT: v_cndmask_b32_e64 v5, v5, v6, s[2:3] +; GCN-NEXT: v_mul_hi_u32 v4, v4, v0 +; GCN-NEXT: v_mul_hi_u32 v5, v5, v1 +; GCN-NEXT: v_mul_lo_u32 v6, v4, v2 +; GCN-NEXT: v_add_i32_e32 v7, vcc, 1, v4 +; GCN-NEXT: v_add_i32_e32 v10, vcc, -1, v4 +; GCN-NEXT: v_mul_lo_u32 v11, v5, v3 +; GCN-NEXT: v_add_i32_e32 v12, vcc, 1, v5 +; GCN-NEXT: v_add_i32_e32 v13, vcc, -1, v5 +; GCN-NEXT: v_subrev_i32_e32 v14, vcc, v6, v0 +; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], v0, v6 +; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v11, v1 +; GCN-NEXT: v_cmp_ge_u32_e32 vcc, v1, v11 +; GCN-NEXT: v_cmp_ge_u32_e64 s[2:3], v14, v2 +; GCN-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v3 +; GCN-NEXT: s_and_b64 s[2:3], s[2:3], s[0:1] +; GCN-NEXT: v_cndmask_b32_e64 v0, v4, v7, s[2:3] +; GCN-NEXT: s_and_b64 s[2:3], s[4:5], vcc +; GCN-NEXT: v_cndmask_b32_e64 v1, v5, v12, s[2:3] +; GCN-NEXT: v_cndmask_b32_e64 v0, v10, v0, s[0:1] +; GCN-NEXT: v_cndmask_b32_e32 v1, v13, v1, vcc +; GCN-NEXT: v_xor_b32_e32 v0, v0, v8 +; GCN-NEXT: v_xor_b32_e32 v1, v1, v9 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, v0, v8 +; GCN-NEXT: v_sub_i32_e32 v1, vcc, v1, v9 +; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: s_endpgm +; +; TONGA-LABEL: sdiv_v2i32: +; TONGA: ; %bb.0: +; TONGA-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; TONGA-NEXT: s_mov_b32 s11, 0xf000 +; TONGA-NEXT: s_mov_b32 s10, -1 +; TONGA-NEXT: s_mov_b32 s4, 0x4f800000 +; TONGA-NEXT: s_waitcnt lgkmcnt(0) +; TONGA-NEXT: s_mov_b32 s8, s0 +; TONGA-NEXT: s_mov_b32 s9, s1 +; TONGA-NEXT: s_mov_b32 s0, s2 +; TONGA-NEXT: s_mov_b32 s1, s3 +; TONGA-NEXT: s_mov_b32 s2, s10 +; TONGA-NEXT: s_mov_b32 s3, s11 +; TONGA-NEXT: buffer_load_dwordx4 v[0:3], off, s[0:3], 0 +; TONGA-NEXT: s_waitcnt vmcnt(0) +; TONGA-NEXT: v_ashrrev_i32_e32 v5, 31, v2 +; TONGA-NEXT: v_ashrrev_i32_e32 v7, 31, v3 +; TONGA-NEXT: v_add_u32_e32 v2, vcc, v5, v2 +; TONGA-NEXT: v_ashrrev_i32_e32 v4, 31, v0 +; TONGA-NEXT: v_add_u32_e32 v3, vcc, v7, v3 +; TONGA-NEXT: v_xor_b32_e32 v2, v2, v5 +; TONGA-NEXT: v_ashrrev_i32_e32 v6, 31, v1 +; TONGA-NEXT: v_xor_b32_e32 v8, v4, v5 +; TONGA-NEXT: v_cvt_f32_u32_e32 v5, v2 +; TONGA-NEXT: v_xor_b32_e32 v3, v3, v7 +; TONGA-NEXT: v_xor_b32_e32 v9, v6, v7 +; TONGA-NEXT: v_cvt_f32_u32_e32 v7, v3 +; TONGA-NEXT: v_rcp_iflag_f32_e32 v5, v5 +; TONGA-NEXT: v_add_u32_e32 v0, vcc, v4, v0 +; TONGA-NEXT: v_xor_b32_e32 v0, v0, v4 +; TONGA-NEXT: v_rcp_iflag_f32_e32 v7, v7 +; TONGA-NEXT: v_mul_f32_e32 v4, s4, v5 +; TONGA-NEXT: v_cvt_u32_f32_e32 v4, v4 +; TONGA-NEXT: v_add_u32_e32 v1, vcc, v6, v1 +; TONGA-NEXT: v_mul_f32_e32 v5, s4, v7 +; TONGA-NEXT: v_cvt_u32_f32_e32 v5, v5 +; TONGA-NEXT: v_xor_b32_e32 v1, v1, v6 +; TONGA-NEXT: v_mul_hi_u32 v6, v4, v2 +; TONGA-NEXT: v_mul_lo_u32 v7, v4, v2 +; TONGA-NEXT: v_mul_hi_u32 v10, v5, v3 +; TONGA-NEXT: v_mul_lo_u32 v11, v5, v3 +; TONGA-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v6 +; TONGA-NEXT: v_sub_u32_e32 v12, vcc, 0, v7 +; TONGA-NEXT: v_cndmask_b32_e64 v6, v7, v12, s[0:1] +; TONGA-NEXT: v_sub_u32_e32 v13, vcc, 0, v11 +; TONGA-NEXT: v_cmp_eq_u32_e64 s[2:3], 0, v10 +; TONGA-NEXT: v_cndmask_b32_e64 v7, v11, v13, s[2:3] +; TONGA-NEXT: v_mul_hi_u32 v6, v6, v4 +; TONGA-NEXT: v_mul_hi_u32 v7, v7, v5 +; TONGA-NEXT: v_add_u32_e32 v10, vcc, v6, v4 +; TONGA-NEXT: v_subrev_u32_e32 v4, vcc, v6, v4 +; TONGA-NEXT: v_cndmask_b32_e64 v4, v4, v10, s[0:1] +; TONGA-NEXT: v_add_u32_e32 v6, vcc, v7, v5 +; TONGA-NEXT: v_subrev_u32_e32 v5, vcc, v7, v5 +; TONGA-NEXT: v_mul_hi_u32 v4, v4, v0 +; TONGA-NEXT: v_cndmask_b32_e64 v5, v5, v6, s[2:3] +; TONGA-NEXT: v_mul_hi_u32 v5, v5, v1 +; TONGA-NEXT: v_mul_lo_u32 v6, v4, v2 +; TONGA-NEXT: v_add_u32_e32 v7, vcc, 1, v4 +; TONGA-NEXT: v_mul_lo_u32 v11, v5, v3 +; TONGA-NEXT: v_add_u32_e32 v10, vcc, -1, v4 +; TONGA-NEXT: v_subrev_u32_e32 v14, vcc, v6, v0 +; TONGA-NEXT: v_cmp_ge_u32_e64 s[0:1], v0, v6 +; TONGA-NEXT: v_cmp_ge_u32_e64 s[2:3], v14, v2 +; TONGA-NEXT: v_subrev_u32_e32 v0, vcc, v11, v1 +; TONGA-NEXT: v_add_u32_e32 v12, vcc, 1, v5 +; TONGA-NEXT: v_add_u32_e32 v13, vcc, -1, v5 +; TONGA-NEXT: v_cmp_ge_u32_e32 vcc, v1, v11 +; TONGA-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v3 +; TONGA-NEXT: s_and_b64 s[2:3], s[2:3], s[0:1] +; TONGA-NEXT: v_cndmask_b32_e64 v0, v4, v7, s[2:3] +; TONGA-NEXT: s_and_b64 s[2:3], s[4:5], vcc +; TONGA-NEXT: v_cndmask_b32_e64 v1, v5, v12, s[2:3] +; TONGA-NEXT: v_cndmask_b32_e64 v0, v10, v0, s[0:1] +; TONGA-NEXT: v_cndmask_b32_e32 v1, v13, v1, vcc +; TONGA-NEXT: v_xor_b32_e32 v0, v0, v8 +; TONGA-NEXT: v_xor_b32_e32 v1, v1, v9 +; TONGA-NEXT: v_sub_u32_e32 v0, vcc, v0, v8 +; TONGA-NEXT: v_sub_u32_e32 v1, vcc, v1, v9 +; TONGA-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; TONGA-NEXT: s_endpgm +; +; GFX9-LABEL: sdiv_v2i32: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; GFX9-NEXT: s_mov_b32 s11, 0xf000 +; GFX9-NEXT: s_mov_b32 s10, -1 +; GFX9-NEXT: s_mov_b32 s4, 0x4f800000 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s8, s0 +; GFX9-NEXT: s_mov_b32 s9, s1 +; GFX9-NEXT: s_mov_b32 s0, s2 +; GFX9-NEXT: s_mov_b32 s1, s3 +; GFX9-NEXT: s_mov_b32 s2, s10 +; GFX9-NEXT: s_mov_b32 s3, s11 +; GFX9-NEXT: buffer_load_dwordx4 v[0:3], off, s[0:3], 0 +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_ashrrev_i32_e32 v5, 31, v2 +; GFX9-NEXT: v_ashrrev_i32_e32 v6, 31, v3 +; GFX9-NEXT: v_add_u32_e32 v2, v2, v5 +; GFX9-NEXT: v_add_u32_e32 v3, v3, v6 +; GFX9-NEXT: v_xor_b32_e32 v2, v2, v5 +; GFX9-NEXT: v_cvt_f32_u32_e32 v7, v2 +; GFX9-NEXT: v_xor_b32_e32 v3, v3, v6 +; GFX9-NEXT: v_cvt_f32_u32_e32 v8, v3 +; GFX9-NEXT: v_ashrrev_i32_e32 v4, 31, v0 +; GFX9-NEXT: v_rcp_iflag_f32_e32 v7, v7 +; GFX9-NEXT: v_add_u32_e32 v0, v0, v4 +; GFX9-NEXT: v_rcp_iflag_f32_e32 v8, v8 +; GFX9-NEXT: v_xor_b32_e32 v5, v4, v5 +; GFX9-NEXT: v_mul_f32_e32 v7, s4, v7 +; GFX9-NEXT: v_cvt_u32_f32_e32 v7, v7 +; GFX9-NEXT: v_mul_f32_e32 v8, s4, v8 +; GFX9-NEXT: v_cvt_u32_f32_e32 v8, v8 +; GFX9-NEXT: v_xor_b32_e32 v0, v0, v4 +; GFX9-NEXT: v_mul_lo_u32 v4, v7, v2 +; GFX9-NEXT: v_mul_hi_u32 v11, v7, v2 +; GFX9-NEXT: v_mul_lo_u32 v10, v8, v3 +; GFX9-NEXT: v_mul_hi_u32 v12, v8, v3 +; GFX9-NEXT: v_sub_u32_e32 v13, 0, v4 +; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v11 +; GFX9-NEXT: v_cndmask_b32_e32 v4, v4, v13, vcc +; GFX9-NEXT: v_sub_u32_e32 v14, 0, v10 +; GFX9-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v12 +; GFX9-NEXT: v_cndmask_b32_e64 v10, v10, v14, s[0:1] +; GFX9-NEXT: v_mul_hi_u32 v4, v4, v7 +; GFX9-NEXT: v_mul_hi_u32 v10, v10, v8 +; GFX9-NEXT: v_ashrrev_i32_e32 v9, 31, v1 +; GFX9-NEXT: v_add_u32_e32 v1, v1, v9 +; GFX9-NEXT: v_xor_b32_e32 v6, v9, v6 +; GFX9-NEXT: v_xor_b32_e32 v1, v1, v9 +; GFX9-NEXT: v_add_u32_e32 v9, v7, v4 +; GFX9-NEXT: v_sub_u32_e32 v4, v7, v4 +; GFX9-NEXT: v_cndmask_b32_e32 v4, v4, v9, vcc +; GFX9-NEXT: v_add_u32_e32 v7, v8, v10 +; GFX9-NEXT: v_sub_u32_e32 v8, v8, v10 +; GFX9-NEXT: v_mul_hi_u32 v4, v4, v0 +; GFX9-NEXT: v_cndmask_b32_e64 v7, v8, v7, s[0:1] +; GFX9-NEXT: v_mul_hi_u32 v7, v7, v1 +; GFX9-NEXT: v_mul_lo_u32 v8, v4, v2 +; GFX9-NEXT: v_add_u32_e32 v9, 1, v4 +; GFX9-NEXT: v_mul_lo_u32 v11, v7, v3 +; GFX9-NEXT: v_add_u32_e32 v12, 1, v7 +; GFX9-NEXT: v_sub_u32_e32 v14, v0, v8 +; GFX9-NEXT: v_cmp_ge_u32_e32 vcc, v0, v8 +; GFX9-NEXT: v_cmp_ge_u32_e64 s[2:3], v14, v2 +; GFX9-NEXT: v_sub_u32_e32 v0, v1, v11 +; GFX9-NEXT: v_cmp_ge_u32_e64 s[0:1], v1, v11 +; GFX9-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v3 +; GFX9-NEXT: s_and_b64 s[2:3], s[2:3], vcc +; GFX9-NEXT: v_cndmask_b32_e64 v0, v4, v9, s[2:3] +; GFX9-NEXT: s_and_b64 s[2:3], s[4:5], s[0:1] +; GFX9-NEXT: v_add_u32_e32 v10, -1, v4 +; GFX9-NEXT: v_add_u32_e32 v13, -1, v7 +; GFX9-NEXT: v_cndmask_b32_e64 v1, v7, v12, s[2:3] +; GFX9-NEXT: v_cndmask_b32_e32 v0, v10, v0, vcc +; GFX9-NEXT: v_cndmask_b32_e64 v1, v13, v1, s[0:1] +; GFX9-NEXT: v_xor_b32_e32 v0, v0, v5 +; GFX9-NEXT: v_xor_b32_e32 v1, v1, v6 +; GFX9-NEXT: v_sub_u32_e32 v0, v0, v5 +; GFX9-NEXT: v_sub_u32_e32 v1, v1, v6 +; GFX9-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GFX9-NEXT: s_endpgm +; +; EG-LABEL: sdiv_v2i32: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @10, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 1 @6 +; EG-NEXT: ALU 59, @11, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 6: +; EG-NEXT: VTX_READ_64 T1.XY, T0.X, 8, #1 +; EG-NEXT: VTX_READ_64 T0.XY, T0.X, 0, #1 +; EG-NEXT: ALU clause starting at 10: +; EG-NEXT: MOV * T0.X, KC0[2].Z, +; EG-NEXT: ALU clause starting at 11: +; EG-NEXT: SETGT_INT * T0.W, 0.0, T1.Y, +; EG-NEXT: ADD_INT * T1.W, T1.Y, PV.W, +; EG-NEXT: XOR_INT T1.W, PV.W, T0.W, +; EG-NEXT: SETGT_INT * T2.W, 0.0, T1.X, +; EG-NEXT: ADD_INT T3.W, T1.X, PS, +; EG-NEXT: RECIP_UINT * T0.Z, PV.W, +; EG-NEXT: XOR_INT T3.W, PV.W, T2.W, BS:VEC_021/SCL_122 +; EG-NEXT: MULLO_INT * T1.X, PS, T1.W, +; EG-NEXT: RECIP_UINT * T1.Y, PV.W, +; EG-NEXT: MULLO_INT * T1.Z, PS, T3.W, +; EG-NEXT: SUB_INT T4.W, 0.0, PS, +; EG-NEXT: MULHI * T2.X, T1.Y, T3.W, +; EG-NEXT: CNDE_INT T1.Z, PS, PV.W, T1.Z, BS:VEC_021/SCL_122 +; EG-NEXT: SUB_INT T4.W, 0.0, T1.X, +; EG-NEXT: MULHI * T2.Y, T0.Z, T1.W, +; EG-NEXT: CNDE_INT T2.Z, PS, PV.W, T1.X, +; EG-NEXT: SETGT_INT T4.W, 0.0, T0.X, +; EG-NEXT: MULHI * T1.X, PV.Z, T1.Y, +; EG-NEXT: SETGT_INT T3.X, 0.0, T0.Y, +; EG-NEXT: ADD_INT T3.Y, T0.X, PV.W, +; EG-NEXT: ADD_INT T1.Z, T1.Y, PS, +; EG-NEXT: SUB_INT T5.W, T1.Y, PS, +; EG-NEXT: MULHI * T0.X, PV.Z, T0.Z, +; EG-NEXT: CNDE_INT T1.X, T2.X, PV.Z, PV.W, +; EG-NEXT: XOR_INT T1.Y, PV.Y, T4.W, +; EG-NEXT: ADD_INT T1.Z, T0.Y, PV.X, +; EG-NEXT: ADD_INT T5.W, T0.Z, PS, +; EG-NEXT: SUB_INT * T6.W, T0.Z, PS, +; EG-NEXT: CNDE_INT T0.Z, T2.Y, PV.W, PS, +; EG-NEXT: XOR_INT T5.W, PV.Z, T3.X, +; EG-NEXT: MULHI * T0.X, PV.X, PV.Y, +; EG-NEXT: MULHI * T0.Y, PV.Z, PV.W, +; EG-NEXT: MULLO_INT * T0.Z, PS, T1.W, +; EG-NEXT: SUB_INT T6.W, T5.W, PS, +; EG-NEXT: MULLO_INT * T1.X, T0.X, T3.W, +; EG-NEXT: SUB_INT T1.Z, T1.Y, PS, +; EG-NEXT: SETGE_UINT T1.W, PV.W, T1.W, +; EG-NEXT: SETGE_UINT * T5.W, T5.W, T0.Z, +; EG-NEXT: AND_INT T2.Y, PV.W, PS, +; EG-NEXT: ADD_INT T0.Z, T0.Y, 1, +; EG-NEXT: SETGE_UINT T1.W, PV.Z, T3.W, +; EG-NEXT: SETGE_UINT * T3.W, T1.Y, T1.X, +; EG-NEXT: AND_INT T1.Y, PV.W, PS, +; EG-NEXT: ADD_INT T1.Z, T0.X, 1, +; EG-NEXT: CNDE_INT T1.W, PV.Y, T0.Y, PV.Z, +; EG-NEXT: ADD_INT * T6.W, T0.Y, literal.x, +; EG-NEXT: -1(nan), 0(0.000000e+00) +; EG-NEXT: CNDE_INT T0.Y, T5.W, PS, PV.W, +; EG-NEXT: XOR_INT T0.Z, T3.X, T0.W, +; EG-NEXT: CNDE_INT T0.W, PV.Y, T0.X, PV.Z, +; EG-NEXT: ADD_INT * T1.W, T0.X, literal.x, +; EG-NEXT: -1(nan), 0(0.000000e+00) +; EG-NEXT: CNDE_INT T1.Z, T3.W, PS, PV.W, +; EG-NEXT: XOR_INT T0.W, T4.W, T2.W, BS:VEC_120/SCL_212 +; EG-NEXT: XOR_INT * T1.W, PV.Y, PV.Z, +; EG-NEXT: SUB_INT T0.Y, PS, T0.Z, +; EG-NEXT: XOR_INT * T1.W, PV.Z, PV.W, +; EG-NEXT: SUB_INT T0.X, PV.W, T0.W, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %den_ptr = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %in, i32 1 %num = load <2 x i32>, <2 x i32> addrspace(1) * %in %den = load <2 x i32>, <2 x i32> addrspace(1) * %den_ptr @@ -61,6 +735,108 @@ define amdgpu_kernel void @sdiv_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> ad } define amdgpu_kernel void @sdiv_v2i32_4(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) { +; GCN-LABEL: sdiv_v2i32_4: +; GCN: ; %bb.0: +; GCN-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: s_mov_b32 s10, s6 +; GCN-NEXT: s_mov_b32 s11, s7 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b32 s8, s2 +; GCN-NEXT: s_mov_b32 s9, s3 +; GCN-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: s_mov_b32 s4, s0 +; GCN-NEXT: s_mov_b32 s5, s1 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_ashrrev_i32_e32 v2, 31, v0 +; GCN-NEXT: v_ashrrev_i32_e32 v3, 31, v1 +; GCN-NEXT: v_lshrrev_b32_e32 v2, 30, v2 +; GCN-NEXT: v_lshrrev_b32_e32 v3, 30, v3 +; GCN-NEXT: v_add_i32_e32 v0, vcc, v2, v0 +; GCN-NEXT: v_add_i32_e32 v1, vcc, v3, v1 +; GCN-NEXT: v_ashrrev_i32_e32 v0, 2, v0 +; GCN-NEXT: v_ashrrev_i32_e32 v1, 2, v1 +; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; TONGA-LABEL: sdiv_v2i32_4: +; TONGA: ; %bb.0: +; TONGA-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; TONGA-NEXT: s_mov_b32 s3, 0xf000 +; TONGA-NEXT: s_mov_b32 s2, -1 +; TONGA-NEXT: s_waitcnt lgkmcnt(0) +; TONGA-NEXT: s_mov_b32 s0, s4 +; TONGA-NEXT: s_mov_b32 s1, s5 +; TONGA-NEXT: s_mov_b32 s4, s6 +; TONGA-NEXT: s_mov_b32 s5, s7 +; TONGA-NEXT: s_mov_b32 s6, s2 +; TONGA-NEXT: s_mov_b32 s7, s3 +; TONGA-NEXT: buffer_load_dwordx2 v[0:1], off, s[4:7], 0 +; TONGA-NEXT: s_waitcnt vmcnt(0) +; TONGA-NEXT: v_ashrrev_i32_e32 v2, 31, v0 +; TONGA-NEXT: v_ashrrev_i32_e32 v3, 31, v1 +; TONGA-NEXT: v_lshrrev_b32_e32 v2, 30, v2 +; TONGA-NEXT: v_lshrrev_b32_e32 v3, 30, v3 +; TONGA-NEXT: v_add_u32_e32 v0, vcc, v2, v0 +; TONGA-NEXT: v_add_u32_e32 v1, vcc, v3, v1 +; TONGA-NEXT: v_ashrrev_i32_e32 v0, 2, v0 +; TONGA-NEXT: v_ashrrev_i32_e32 v1, 2, v1 +; TONGA-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; TONGA-NEXT: s_endpgm +; +; GFX9-LABEL: sdiv_v2i32_4: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; GFX9-NEXT: s_mov_b32 s3, 0xf000 +; GFX9-NEXT: s_mov_b32 s2, -1 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s0, s4 +; GFX9-NEXT: s_mov_b32 s1, s5 +; GFX9-NEXT: s_mov_b32 s4, s6 +; GFX9-NEXT: s_mov_b32 s5, s7 +; GFX9-NEXT: s_mov_b32 s6, s2 +; GFX9-NEXT: s_mov_b32 s7, s3 +; GFX9-NEXT: buffer_load_dwordx2 v[0:1], off, s[4:7], 0 +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_ashrrev_i32_e32 v2, 31, v0 +; GFX9-NEXT: v_ashrrev_i32_e32 v3, 31, v1 +; GFX9-NEXT: v_lshrrev_b32_e32 v2, 30, v2 +; GFX9-NEXT: v_lshrrev_b32_e32 v3, 30, v3 +; GFX9-NEXT: v_add_u32_e32 v0, v0, v2 +; GFX9-NEXT: v_add_u32_e32 v1, v1, v3 +; GFX9-NEXT: v_ashrrev_i32_e32 v0, 2, v0 +; GFX9-NEXT: v_ashrrev_i32_e32 v1, 2, v1 +; GFX9-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX9-NEXT: s_endpgm +; +; EG-LABEL: sdiv_v2i32_4: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 13, @9, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 6: +; EG-NEXT: VTX_READ_64 T0.XY, T0.X, 0, #1 +; EG-NEXT: ALU clause starting at 8: +; EG-NEXT: MOV * T0.X, KC0[2].Z, +; EG-NEXT: ALU clause starting at 9: +; EG-NEXT: ASHR * T0.W, T0.Y, literal.x, +; EG-NEXT: 31(4.344025e-44), 0(0.000000e+00) +; EG-NEXT: LSHR T0.W, PV.W, literal.x, +; EG-NEXT: ASHR * T1.W, T0.X, literal.y, +; EG-NEXT: 30(4.203895e-44), 31(4.344025e-44) +; EG-NEXT: LSHR T1.W, PS, literal.x, +; EG-NEXT: ADD_INT * T0.W, T0.Y, PV.W, +; EG-NEXT: 30(4.203895e-44), 0(0.000000e+00) +; EG-NEXT: ASHR T0.Y, PS, literal.x, +; EG-NEXT: ADD_INT * T0.W, T0.X, PV.W, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; EG-NEXT: ASHR T0.X, PV.W, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %num = load <2 x i32>, <2 x i32> addrspace(1) * %in %result = sdiv <2 x i32> %num, store <2 x i32> %result, <2 x i32> addrspace(1)* %out @@ -68,6 +844,586 @@ define amdgpu_kernel void @sdiv_v2i32_4(<2 x i32> addrspace(1)* %out, <2 x i32> } define amdgpu_kernel void @sdiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) { +; GCN-LABEL: sdiv_v4i32: +; GCN: ; %bb.0: +; GCN-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x9 +; GCN-NEXT: s_mov_b32 s19, 0xf000 +; GCN-NEXT: s_mov_b32 s18, -1 +; GCN-NEXT: s_mov_b32 s2, s18 +; GCN-NEXT: s_mov_b32 s3, s19 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b32 s0, s10 +; GCN-NEXT: s_mov_b32 s1, s11 +; GCN-NEXT: buffer_load_dwordx4 v[0:3], off, s[0:3], 0 +; GCN-NEXT: buffer_load_dwordx4 v[4:7], off, s[0:3], 0 offset:16 +; GCN-NEXT: s_mov_b32 s6, 0x4f800000 +; GCN-NEXT: s_waitcnt vmcnt(1) +; GCN-NEXT: v_ashrrev_i32_e32 v8, 31, v0 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_ashrrev_i32_e32 v9, 31, v4 +; GCN-NEXT: v_ashrrev_i32_e32 v10, 31, v1 +; GCN-NEXT: v_ashrrev_i32_e32 v11, 31, v5 +; GCN-NEXT: v_ashrrev_i32_e32 v12, 31, v2 +; GCN-NEXT: v_ashrrev_i32_e32 v13, 31, v6 +; GCN-NEXT: v_ashrrev_i32_e32 v14, 31, v3 +; GCN-NEXT: v_ashrrev_i32_e32 v15, 31, v7 +; GCN-NEXT: v_xor_b32_e32 v16, v8, v9 +; GCN-NEXT: v_add_i32_e32 v0, vcc, v8, v0 +; GCN-NEXT: v_add_i32_e32 v1, vcc, v10, v1 +; GCN-NEXT: v_add_i32_e32 v2, vcc, v12, v2 +; GCN-NEXT: v_add_i32_e32 v3, vcc, v14, v3 +; GCN-NEXT: v_add_i32_e32 v4, vcc, v9, v4 +; GCN-NEXT: v_add_i32_e32 v5, vcc, v11, v5 +; GCN-NEXT: v_add_i32_e32 v6, vcc, v13, v6 +; GCN-NEXT: v_add_i32_e32 v7, vcc, v15, v7 +; GCN-NEXT: v_xor_b32_e32 v17, v10, v11 +; GCN-NEXT: v_xor_b32_e32 v18, v12, v13 +; GCN-NEXT: v_xor_b32_e32 v19, v14, v15 +; GCN-NEXT: v_xor_b32_e32 v0, v0, v8 +; GCN-NEXT: v_xor_b32_e32 v4, v4, v9 +; GCN-NEXT: v_xor_b32_e32 v1, v1, v10 +; GCN-NEXT: v_xor_b32_e32 v5, v5, v11 +; GCN-NEXT: v_xor_b32_e32 v2, v2, v12 +; GCN-NEXT: v_xor_b32_e32 v6, v6, v13 +; GCN-NEXT: v_xor_b32_e32 v3, v3, v14 +; GCN-NEXT: v_xor_b32_e32 v7, v7, v15 +; GCN-NEXT: v_cvt_f32_u32_e32 v8, v4 +; GCN-NEXT: v_cvt_f32_u32_e32 v9, v5 +; GCN-NEXT: v_cvt_f32_u32_e32 v10, v6 +; GCN-NEXT: v_rcp_iflag_f32_e32 v8, v8 +; GCN-NEXT: v_rcp_iflag_f32_e32 v9, v9 +; GCN-NEXT: v_rcp_iflag_f32_e32 v10, v10 +; GCN-NEXT: v_mul_f32_e32 v8, s6, v8 +; GCN-NEXT: v_mul_f32_e32 v9, s6, v9 +; GCN-NEXT: v_mul_f32_e32 v10, s6, v10 +; GCN-NEXT: v_cvt_u32_f32_e32 v8, v8 +; GCN-NEXT: v_cvt_u32_f32_e32 v9, v9 +; GCN-NEXT: v_cvt_u32_f32_e32 v10, v10 +; GCN-NEXT: v_mul_hi_u32 v11, v8, v4 +; GCN-NEXT: v_mul_lo_u32 v12, v8, v4 +; GCN-NEXT: v_mul_hi_u32 v13, v9, v5 +; GCN-NEXT: v_mul_lo_u32 v14, v9, v5 +; GCN-NEXT: v_sub_i32_e32 v15, vcc, 0, v12 +; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v11 +; GCN-NEXT: v_mul_hi_u32 v11, v10, v6 +; GCN-NEXT: v_cndmask_b32_e64 v12, v12, v15, s[0:1] +; GCN-NEXT: v_sub_i32_e32 v15, vcc, 0, v14 +; GCN-NEXT: v_cmp_eq_u32_e64 s[2:3], 0, v13 +; GCN-NEXT: v_mul_lo_u32 v13, v10, v6 +; GCN-NEXT: v_cndmask_b32_e64 v14, v14, v15, s[2:3] +; GCN-NEXT: v_sub_i32_e32 v15, vcc, 0, v13 +; GCN-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v11 +; GCN-NEXT: v_cvt_f32_u32_e32 v11, v7 +; GCN-NEXT: v_rcp_iflag_f32_e32 v11, v11 +; GCN-NEXT: v_mul_f32_e32 v11, s6, v11 +; GCN-NEXT: v_cvt_u32_f32_e32 v11, v11 +; GCN-NEXT: v_cndmask_b32_e64 v13, v13, v15, s[4:5] +; GCN-NEXT: v_mul_hi_u32 v15, v11, v7 +; GCN-NEXT: v_mul_lo_u32 v20, v11, v7 +; GCN-NEXT: v_sub_i32_e32 v21, vcc, 0, v20 +; GCN-NEXT: v_cmp_eq_u32_e64 s[6:7], 0, v15 +; GCN-NEXT: v_cndmask_b32_e64 v15, v20, v21, s[6:7] +; GCN-NEXT: v_mul_hi_u32 v12, v12, v8 +; GCN-NEXT: v_add_i32_e32 v20, vcc, v12, v8 +; GCN-NEXT: v_subrev_i32_e32 v8, vcc, v12, v8 +; GCN-NEXT: v_mul_hi_u32 v12, v14, v9 +; GCN-NEXT: v_add_i32_e32 v14, vcc, v12, v9 +; GCN-NEXT: v_subrev_i32_e32 v9, vcc, v12, v9 +; GCN-NEXT: v_mul_hi_u32 v12, v13, v10 +; GCN-NEXT: v_add_i32_e32 v13, vcc, v12, v10 +; GCN-NEXT: v_subrev_i32_e32 v10, vcc, v12, v10 +; GCN-NEXT: v_mul_hi_u32 v12, v15, v11 +; GCN-NEXT: v_add_i32_e32 v15, vcc, v12, v11 +; GCN-NEXT: v_subrev_i32_e32 v11, vcc, v12, v11 +; GCN-NEXT: s_mov_b32 s16, s8 +; GCN-NEXT: s_mov_b32 s17, s9 +; GCN-NEXT: v_cndmask_b32_e64 v8, v8, v20, s[0:1] +; GCN-NEXT: v_cndmask_b32_e64 v9, v9, v14, s[2:3] +; GCN-NEXT: v_cndmask_b32_e64 v10, v10, v13, s[4:5] +; GCN-NEXT: v_cndmask_b32_e64 v11, v11, v15, s[6:7] +; GCN-NEXT: v_mul_hi_u32 v8, v8, v0 +; GCN-NEXT: v_mul_hi_u32 v9, v9, v1 +; GCN-NEXT: v_mul_hi_u32 v10, v10, v2 +; GCN-NEXT: v_mul_hi_u32 v11, v11, v3 +; GCN-NEXT: v_mul_lo_u32 v12, v8, v4 +; GCN-NEXT: v_add_i32_e32 v13, vcc, 1, v8 +; GCN-NEXT: v_add_i32_e32 v14, vcc, -1, v8 +; GCN-NEXT: v_mul_lo_u32 v15, v9, v5 +; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], v0, v12 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, v0, v12 +; GCN-NEXT: v_add_i32_e32 v12, vcc, 1, v9 +; GCN-NEXT: v_cmp_ge_u32_e64 s[2:3], v0, v4 +; GCN-NEXT: v_add_i32_e32 v0, vcc, -1, v9 +; GCN-NEXT: v_mul_lo_u32 v4, v10, v6 +; GCN-NEXT: v_cmp_ge_u32_e64 s[4:5], v1, v15 +; GCN-NEXT: v_sub_i32_e32 v1, vcc, v1, v15 +; GCN-NEXT: v_add_i32_e32 v15, vcc, 1, v10 +; GCN-NEXT: v_cmp_ge_u32_e64 s[6:7], v1, v5 +; GCN-NEXT: v_add_i32_e32 v1, vcc, -1, v10 +; GCN-NEXT: v_mul_lo_u32 v5, v11, v7 +; GCN-NEXT: v_cmp_ge_u32_e64 s[8:9], v2, v4 +; GCN-NEXT: v_sub_i32_e32 v2, vcc, v2, v4 +; GCN-NEXT: v_add_i32_e32 v4, vcc, -1, v11 +; GCN-NEXT: v_cmp_ge_u32_e64 s[10:11], v3, v5 +; GCN-NEXT: v_sub_i32_e32 v3, vcc, v3, v5 +; GCN-NEXT: v_add_i32_e32 v5, vcc, 1, v11 +; GCN-NEXT: v_cmp_ge_u32_e32 vcc, v2, v6 +; GCN-NEXT: v_cmp_ge_u32_e64 s[12:13], v3, v7 +; GCN-NEXT: s_and_b64 s[2:3], s[2:3], s[0:1] +; GCN-NEXT: v_cndmask_b32_e64 v2, v8, v13, s[2:3] +; GCN-NEXT: s_and_b64 s[2:3], s[6:7], s[4:5] +; GCN-NEXT: v_cndmask_b32_e64 v3, v9, v12, s[2:3] +; GCN-NEXT: s_and_b64 vcc, vcc, s[8:9] +; GCN-NEXT: v_cndmask_b32_e32 v6, v10, v15, vcc +; GCN-NEXT: s_and_b64 vcc, s[12:13], s[10:11] +; GCN-NEXT: v_cndmask_b32_e32 v5, v11, v5, vcc +; GCN-NEXT: v_cndmask_b32_e64 v2, v14, v2, s[0:1] +; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v3, s[4:5] +; GCN-NEXT: v_cndmask_b32_e64 v1, v1, v6, s[8:9] +; GCN-NEXT: v_cndmask_b32_e64 v3, v4, v5, s[10:11] +; GCN-NEXT: v_xor_b32_e32 v2, v2, v16 +; GCN-NEXT: v_xor_b32_e32 v4, v0, v17 +; GCN-NEXT: v_xor_b32_e32 v5, v1, v18 +; GCN-NEXT: v_xor_b32_e32 v3, v3, v19 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, v2, v16 +; GCN-NEXT: v_sub_i32_e32 v1, vcc, v4, v17 +; GCN-NEXT: v_sub_i32_e32 v2, vcc, v5, v18 +; GCN-NEXT: v_sub_i32_e32 v3, vcc, v3, v19 +; GCN-NEXT: buffer_store_dwordx4 v[0:3], off, s[16:19], 0 +; GCN-NEXT: s_endpgm +; +; TONGA-LABEL: sdiv_v4i32: +; TONGA: ; %bb.0: +; TONGA-NEXT: s_load_dwordx4 s[12:15], s[0:1], 0x24 +; TONGA-NEXT: s_mov_b32 s11, 0xf000 +; TONGA-NEXT: s_mov_b32 s10, -1 +; TONGA-NEXT: s_mov_b32 s2, s10 +; TONGA-NEXT: s_mov_b32 s3, s11 +; TONGA-NEXT: s_waitcnt lgkmcnt(0) +; TONGA-NEXT: s_mov_b32 s0, s14 +; TONGA-NEXT: s_mov_b32 s1, s15 +; TONGA-NEXT: buffer_load_dwordx4 v[4:7], off, s[0:3], 0 offset:16 +; TONGA-NEXT: buffer_load_dwordx4 v[0:3], off, s[0:3], 0 +; TONGA-NEXT: s_mov_b32 s14, 0x4f800000 +; TONGA-NEXT: s_mov_b32 s8, s12 +; TONGA-NEXT: s_mov_b32 s9, s13 +; TONGA-NEXT: s_waitcnt vmcnt(1) +; TONGA-NEXT: v_ashrrev_i32_e32 v9, 31, v4 +; TONGA-NEXT: v_add_u32_e32 v4, vcc, v9, v4 +; TONGA-NEXT: s_waitcnt vmcnt(0) +; TONGA-NEXT: v_ashrrev_i32_e32 v8, 31, v0 +; TONGA-NEXT: v_xor_b32_e32 v4, v4, v9 +; TONGA-NEXT: v_xor_b32_e32 v15, v8, v9 +; TONGA-NEXT: v_cvt_f32_u32_e32 v9, v4 +; TONGA-NEXT: v_ashrrev_i32_e32 v11, 31, v5 +; TONGA-NEXT: v_add_u32_e32 v5, vcc, v11, v5 +; TONGA-NEXT: v_add_u32_e32 v0, vcc, v8, v0 +; TONGA-NEXT: v_rcp_iflag_f32_e32 v9, v9 +; TONGA-NEXT: v_xor_b32_e32 v5, v5, v11 +; TONGA-NEXT: v_xor_b32_e32 v0, v0, v8 +; TONGA-NEXT: v_cvt_f32_u32_e32 v8, v5 +; TONGA-NEXT: v_mul_f32_e32 v9, s14, v9 +; TONGA-NEXT: v_cvt_u32_f32_e32 v9, v9 +; TONGA-NEXT: v_ashrrev_i32_e32 v10, 31, v1 +; TONGA-NEXT: v_rcp_iflag_f32_e32 v8, v8 +; TONGA-NEXT: v_add_u32_e32 v1, vcc, v10, v1 +; TONGA-NEXT: v_xor_b32_e32 v16, v10, v11 +; TONGA-NEXT: v_xor_b32_e32 v1, v1, v10 +; TONGA-NEXT: v_mul_f32_e32 v8, s14, v8 +; TONGA-NEXT: v_mul_hi_u32 v11, v9, v4 +; TONGA-NEXT: v_mul_lo_u32 v10, v9, v4 +; TONGA-NEXT: v_cvt_u32_f32_e32 v8, v8 +; TONGA-NEXT: v_ashrrev_i32_e32 v12, 31, v2 +; TONGA-NEXT: v_ashrrev_i32_e32 v13, 31, v6 +; TONGA-NEXT: v_add_u32_e32 v2, vcc, v12, v2 +; TONGA-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v11 +; TONGA-NEXT: v_xor_b32_e32 v17, v12, v13 +; TONGA-NEXT: v_xor_b32_e32 v2, v2, v12 +; TONGA-NEXT: v_sub_u32_e32 v12, vcc, 0, v10 +; TONGA-NEXT: v_cndmask_b32_e64 v10, v10, v12, s[0:1] +; TONGA-NEXT: v_mul_hi_u32 v12, v8, v5 +; TONGA-NEXT: v_add_u32_e32 v6, vcc, v13, v6 +; TONGA-NEXT: v_xor_b32_e32 v6, v6, v13 +; TONGA-NEXT: v_mul_lo_u32 v11, v8, v5 +; TONGA-NEXT: v_cmp_eq_u32_e64 s[2:3], 0, v12 +; TONGA-NEXT: v_cvt_f32_u32_e32 v12, v6 +; TONGA-NEXT: v_mul_hi_u32 v10, v10, v9 +; TONGA-NEXT: v_sub_u32_e32 v13, vcc, 0, v11 +; TONGA-NEXT: v_cndmask_b32_e64 v11, v11, v13, s[2:3] +; TONGA-NEXT: v_rcp_iflag_f32_e32 v12, v12 +; TONGA-NEXT: v_ashrrev_i32_e32 v14, 31, v7 +; TONGA-NEXT: v_add_u32_e32 v7, vcc, v14, v7 +; TONGA-NEXT: v_xor_b32_e32 v7, v7, v14 +; TONGA-NEXT: v_mul_f32_e32 v12, s14, v12 +; TONGA-NEXT: v_cvt_u32_f32_e32 v12, v12 +; TONGA-NEXT: v_mul_hi_u32 v18, v12, v6 +; TONGA-NEXT: v_mul_lo_u32 v13, v12, v6 +; TONGA-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v18 +; TONGA-NEXT: v_add_u32_e32 v18, vcc, v10, v9 +; TONGA-NEXT: v_subrev_u32_e32 v9, vcc, v10, v9 +; TONGA-NEXT: v_mul_hi_u32 v10, v11, v8 +; TONGA-NEXT: v_cndmask_b32_e64 v9, v9, v18, s[0:1] +; TONGA-NEXT: v_mul_hi_u32 v9, v9, v0 +; TONGA-NEXT: v_sub_u32_e32 v19, vcc, 0, v13 +; TONGA-NEXT: v_add_u32_e32 v11, vcc, v10, v8 +; TONGA-NEXT: v_subrev_u32_e32 v8, vcc, v10, v8 +; TONGA-NEXT: v_cndmask_b32_e64 v13, v13, v19, s[4:5] +; TONGA-NEXT: v_cndmask_b32_e64 v8, v8, v11, s[2:3] +; TONGA-NEXT: v_mul_hi_u32 v10, v13, v12 +; TONGA-NEXT: v_mul_lo_u32 v11, v9, v4 +; TONGA-NEXT: v_mul_hi_u32 v8, v8, v1 +; TONGA-NEXT: v_add_u32_e32 v13, vcc, v10, v12 +; TONGA-NEXT: v_subrev_u32_e32 v10, vcc, v10, v12 +; TONGA-NEXT: v_cmp_ge_u32_e64 s[0:1], v0, v11 +; TONGA-NEXT: v_sub_u32_e32 v0, vcc, v0, v11 +; TONGA-NEXT: v_cmp_ge_u32_e64 s[2:3], v0, v4 +; TONGA-NEXT: v_cndmask_b32_e64 v10, v10, v13, s[4:5] +; TONGA-NEXT: v_mul_lo_u32 v0, v8, v5 +; TONGA-NEXT: v_mul_hi_u32 v4, v10, v2 +; TONGA-NEXT: v_add_u32_e32 v12, vcc, -1, v9 +; TONGA-NEXT: v_add_u32_e32 v10, vcc, -1, v8 +; TONGA-NEXT: v_cmp_ge_u32_e64 s[4:5], v1, v0 +; TONGA-NEXT: v_sub_u32_e32 v0, vcc, v1, v0 +; TONGA-NEXT: v_cmp_ge_u32_e64 s[6:7], v0, v5 +; TONGA-NEXT: v_mul_lo_u32 v5, v4, v6 +; TONGA-NEXT: v_add_u32_e32 v1, vcc, 1, v9 +; TONGA-NEXT: v_add_u32_e32 v0, vcc, 1, v8 +; TONGA-NEXT: s_and_b64 vcc, s[2:3], s[0:1] +; TONGA-NEXT: v_cndmask_b32_e32 v1, v9, v1, vcc +; TONGA-NEXT: v_sub_u32_e32 v9, vcc, v2, v5 +; TONGA-NEXT: s_and_b64 vcc, s[6:7], s[4:5] +; TONGA-NEXT: v_cvt_f32_u32_e32 v11, v7 +; TONGA-NEXT: v_cndmask_b32_e32 v0, v8, v0, vcc +; TONGA-NEXT: v_cndmask_b32_e64 v1, v12, v1, s[0:1] +; TONGA-NEXT: v_cndmask_b32_e64 v0, v10, v0, s[4:5] +; TONGA-NEXT: v_xor_b32_e32 v1, v1, v15 +; TONGA-NEXT: v_xor_b32_e32 v8, v0, v16 +; TONGA-NEXT: v_sub_u32_e32 v0, vcc, v1, v15 +; TONGA-NEXT: v_sub_u32_e32 v1, vcc, v8, v16 +; TONGA-NEXT: v_rcp_iflag_f32_e32 v8, v11 +; TONGA-NEXT: v_cmp_ge_u32_e64 s[0:1], v9, v6 +; TONGA-NEXT: v_cmp_ge_u32_e64 s[2:3], v2, v5 +; TONGA-NEXT: v_ashrrev_i32_e32 v10, 31, v3 +; TONGA-NEXT: v_mul_f32_e32 v8, s14, v8 +; TONGA-NEXT: v_cvt_u32_f32_e32 v8, v8 +; TONGA-NEXT: v_add_u32_e32 v3, vcc, v10, v3 +; TONGA-NEXT: v_xor_b32_e32 v3, v3, v10 +; TONGA-NEXT: v_add_u32_e32 v6, vcc, -1, v4 +; TONGA-NEXT: v_mul_lo_u32 v5, v8, v7 +; TONGA-NEXT: v_mul_hi_u32 v9, v8, v7 +; TONGA-NEXT: v_add_u32_e32 v2, vcc, 1, v4 +; TONGA-NEXT: v_sub_u32_e32 v11, vcc, 0, v5 +; TONGA-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v9 +; TONGA-NEXT: v_cndmask_b32_e64 v5, v5, v11, s[4:5] +; TONGA-NEXT: v_mul_hi_u32 v5, v5, v8 +; TONGA-NEXT: v_add_u32_e32 v9, vcc, v5, v8 +; TONGA-NEXT: v_subrev_u32_e32 v5, vcc, v5, v8 +; TONGA-NEXT: v_cndmask_b32_e64 v5, v5, v9, s[4:5] +; TONGA-NEXT: v_mul_hi_u32 v5, v5, v3 +; TONGA-NEXT: s_and_b64 vcc, s[0:1], s[2:3] +; TONGA-NEXT: v_cndmask_b32_e32 v2, v4, v2, vcc +; TONGA-NEXT: v_cndmask_b32_e64 v2, v6, v2, s[2:3] +; TONGA-NEXT: v_mul_lo_u32 v4, v5, v7 +; TONGA-NEXT: v_xor_b32_e32 v2, v2, v17 +; TONGA-NEXT: v_sub_u32_e32 v2, vcc, v2, v17 +; TONGA-NEXT: v_xor_b32_e32 v6, v10, v14 +; TONGA-NEXT: v_sub_u32_e32 v8, vcc, v3, v4 +; TONGA-NEXT: v_cmp_ge_u32_e64 s[0:1], v8, v7 +; TONGA-NEXT: v_cmp_ge_u32_e64 s[2:3], v3, v4 +; TONGA-NEXT: v_add_u32_e32 v7, vcc, -1, v5 +; TONGA-NEXT: v_add_u32_e32 v3, vcc, 1, v5 +; TONGA-NEXT: s_and_b64 vcc, s[0:1], s[2:3] +; TONGA-NEXT: v_cndmask_b32_e32 v3, v5, v3, vcc +; TONGA-NEXT: v_cndmask_b32_e64 v3, v7, v3, s[2:3] +; TONGA-NEXT: v_xor_b32_e32 v3, v3, v6 +; TONGA-NEXT: v_sub_u32_e32 v3, vcc, v3, v6 +; TONGA-NEXT: buffer_store_dwordx4 v[0:3], off, s[8:11], 0 +; TONGA-NEXT: s_endpgm +; +; GFX9-LABEL: sdiv_v4i32: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x24 +; GFX9-NEXT: s_mov_b32 s15, 0xf000 +; GFX9-NEXT: s_mov_b32 s14, -1 +; GFX9-NEXT: s_mov_b32 s2, s14 +; GFX9-NEXT: s_mov_b32 s3, s15 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s0, s10 +; GFX9-NEXT: s_mov_b32 s1, s11 +; GFX9-NEXT: buffer_load_dwordx4 v[4:7], off, s[0:3], 0 offset:16 +; GFX9-NEXT: buffer_load_dwordx4 v[0:3], off, s[0:3], 0 +; GFX9-NEXT: s_mov_b32 s4, 0x4f800000 +; GFX9-NEXT: s_mov_b32 s12, s8 +; GFX9-NEXT: s_mov_b32 s13, s9 +; GFX9-NEXT: s_waitcnt vmcnt(1) +; GFX9-NEXT: v_ashrrev_i32_e32 v9, 31, v4 +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_ashrrev_i32_e32 v8, 31, v0 +; GFX9-NEXT: v_add_u32_e32 v4, v4, v9 +; GFX9-NEXT: v_add_u32_e32 v0, v0, v8 +; GFX9-NEXT: v_xor_b32_e32 v4, v4, v9 +; GFX9-NEXT: v_xor_b32_e32 v16, v8, v9 +; GFX9-NEXT: v_xor_b32_e32 v0, v0, v8 +; GFX9-NEXT: v_cvt_f32_u32_e32 v8, v4 +; GFX9-NEXT: v_ashrrev_i32_e32 v11, 31, v5 +; GFX9-NEXT: v_add_u32_e32 v5, v5, v11 +; GFX9-NEXT: v_xor_b32_e32 v5, v5, v11 +; GFX9-NEXT: v_cvt_f32_u32_e32 v9, v5 +; GFX9-NEXT: v_rcp_iflag_f32_e32 v8, v8 +; GFX9-NEXT: v_ashrrev_i32_e32 v13, 31, v6 +; GFX9-NEXT: v_ashrrev_i32_e32 v10, 31, v1 +; GFX9-NEXT: v_add_u32_e32 v6, v6, v13 +; GFX9-NEXT: v_add_u32_e32 v1, v1, v10 +; GFX9-NEXT: v_rcp_iflag_f32_e32 v9, v9 +; GFX9-NEXT: v_mul_f32_e32 v8, s4, v8 +; GFX9-NEXT: v_xor_b32_e32 v6, v6, v13 +; GFX9-NEXT: v_xor_b32_e32 v17, v10, v11 +; GFX9-NEXT: v_xor_b32_e32 v1, v1, v10 +; GFX9-NEXT: v_cvt_f32_u32_e32 v10, v6 +; GFX9-NEXT: v_cvt_u32_f32_e32 v8, v8 +; GFX9-NEXT: v_ashrrev_i32_e32 v12, 31, v2 +; GFX9-NEXT: v_add_u32_e32 v2, v2, v12 +; GFX9-NEXT: v_mul_f32_e32 v9, s4, v9 +; GFX9-NEXT: v_rcp_iflag_f32_e32 v10, v10 +; GFX9-NEXT: v_xor_b32_e32 v18, v12, v13 +; GFX9-NEXT: v_xor_b32_e32 v2, v2, v12 +; GFX9-NEXT: v_cvt_u32_f32_e32 v9, v9 +; GFX9-NEXT: v_mul_hi_u32 v12, v8, v4 +; GFX9-NEXT: v_mul_lo_u32 v11, v8, v4 +; GFX9-NEXT: v_mul_f32_e32 v10, s4, v10 +; GFX9-NEXT: v_mul_lo_u32 v13, v9, v5 +; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v12 +; GFX9-NEXT: v_mul_hi_u32 v12, v9, v5 +; GFX9-NEXT: v_cvt_u32_f32_e32 v10, v10 +; GFX9-NEXT: v_sub_u32_e32 v19, 0, v11 +; GFX9-NEXT: v_cndmask_b32_e32 v11, v11, v19, vcc +; GFX9-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v12 +; GFX9-NEXT: v_sub_u32_e32 v19, 0, v13 +; GFX9-NEXT: v_cndmask_b32_e64 v13, v13, v19, s[0:1] +; GFX9-NEXT: v_mul_hi_u32 v19, v10, v6 +; GFX9-NEXT: v_ashrrev_i32_e32 v15, 31, v7 +; GFX9-NEXT: v_add_u32_e32 v7, v7, v15 +; GFX9-NEXT: v_xor_b32_e32 v7, v7, v15 +; GFX9-NEXT: v_cmp_eq_u32_e64 s[2:3], 0, v19 +; GFX9-NEXT: v_cvt_f32_u32_e32 v19, v7 +; GFX9-NEXT: v_mul_hi_u32 v11, v11, v8 +; GFX9-NEXT: v_mul_lo_u32 v12, v10, v6 +; GFX9-NEXT: v_ashrrev_i32_e32 v14, 31, v3 +; GFX9-NEXT: v_rcp_iflag_f32_e32 v19, v19 +; GFX9-NEXT: v_add_u32_e32 v3, v3, v14 +; GFX9-NEXT: v_sub_u32_e32 v20, 0, v12 +; GFX9-NEXT: v_cndmask_b32_e64 v12, v12, v20, s[2:3] +; GFX9-NEXT: v_mul_f32_e32 v19, s4, v19 +; GFX9-NEXT: v_cvt_u32_f32_e32 v19, v19 +; GFX9-NEXT: v_xor_b32_e32 v3, v3, v14 +; GFX9-NEXT: v_mul_hi_u32 v21, v19, v7 +; GFX9-NEXT: v_mul_lo_u32 v20, v19, v7 +; GFX9-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v21 +; GFX9-NEXT: v_add_u32_e32 v21, v8, v11 +; GFX9-NEXT: v_sub_u32_e32 v8, v8, v11 +; GFX9-NEXT: v_mul_hi_u32 v11, v13, v9 +; GFX9-NEXT: v_cndmask_b32_e32 v8, v8, v21, vcc +; GFX9-NEXT: v_mul_hi_u32 v8, v8, v0 +; GFX9-NEXT: v_sub_u32_e32 v22, 0, v20 +; GFX9-NEXT: v_add_u32_e32 v13, v9, v11 +; GFX9-NEXT: v_sub_u32_e32 v9, v9, v11 +; GFX9-NEXT: v_mul_hi_u32 v11, v12, v10 +; GFX9-NEXT: v_cndmask_b32_e64 v9, v9, v13, s[0:1] +; GFX9-NEXT: v_mul_hi_u32 v9, v9, v1 +; GFX9-NEXT: v_cndmask_b32_e64 v20, v20, v22, s[4:5] +; GFX9-NEXT: v_add_u32_e32 v12, v10, v11 +; GFX9-NEXT: v_sub_u32_e32 v10, v10, v11 +; GFX9-NEXT: v_cndmask_b32_e64 v10, v10, v12, s[2:3] +; GFX9-NEXT: v_mul_lo_u32 v12, v8, v4 +; GFX9-NEXT: v_mul_hi_u32 v11, v20, v19 +; GFX9-NEXT: v_mul_hi_u32 v10, v10, v2 +; GFX9-NEXT: v_add_u32_e32 v13, 1, v8 +; GFX9-NEXT: v_cmp_ge_u32_e32 vcc, v0, v12 +; GFX9-NEXT: v_sub_u32_e32 v0, v0, v12 +; GFX9-NEXT: v_mul_lo_u32 v12, v9, v5 +; GFX9-NEXT: v_add_u32_e32 v20, v19, v11 +; GFX9-NEXT: v_sub_u32_e32 v11, v19, v11 +; GFX9-NEXT: v_cndmask_b32_e64 v11, v11, v20, s[4:5] +; GFX9-NEXT: v_cmp_ge_u32_e64 s[2:3], v1, v12 +; GFX9-NEXT: v_sub_u32_e32 v1, v1, v12 +; GFX9-NEXT: v_mul_lo_u32 v12, v10, v6 +; GFX9-NEXT: v_mul_hi_u32 v11, v11, v3 +; GFX9-NEXT: v_cmp_ge_u32_e64 s[0:1], v0, v4 +; GFX9-NEXT: v_cmp_ge_u32_e64 s[4:5], v1, v5 +; GFX9-NEXT: v_cmp_ge_u32_e64 s[6:7], v2, v12 +; GFX9-NEXT: v_sub_u32_e32 v2, v2, v12 +; GFX9-NEXT: s_and_b64 s[0:1], s[0:1], vcc +; GFX9-NEXT: v_cmp_ge_u32_e64 s[8:9], v2, v6 +; GFX9-NEXT: v_cndmask_b32_e64 v2, v8, v13, s[0:1] +; GFX9-NEXT: v_add_u32_e32 v0, 1, v9 +; GFX9-NEXT: s_and_b64 s[0:1], s[4:5], s[2:3] +; GFX9-NEXT: v_cndmask_b32_e64 v0, v9, v0, s[0:1] +; GFX9-NEXT: v_add_u32_e32 v1, 1, v10 +; GFX9-NEXT: s_and_b64 s[0:1], s[8:9], s[6:7] +; GFX9-NEXT: v_mul_lo_u32 v12, v11, v7 +; GFX9-NEXT: v_add_u32_e32 v19, -1, v8 +; GFX9-NEXT: v_cndmask_b32_e64 v1, v10, v1, s[0:1] +; GFX9-NEXT: v_add_u32_e32 v5, -1, v10 +; GFX9-NEXT: v_cndmask_b32_e32 v2, v19, v2, vcc +; GFX9-NEXT: v_cndmask_b32_e64 v1, v5, v1, s[6:7] +; GFX9-NEXT: v_add_u32_e32 v4, -1, v9 +; GFX9-NEXT: v_cndmask_b32_e64 v0, v4, v0, s[2:3] +; GFX9-NEXT: v_xor_b32_e32 v2, v2, v16 +; GFX9-NEXT: v_xor_b32_e32 v5, v1, v18 +; GFX9-NEXT: v_xor_b32_e32 v4, v0, v17 +; GFX9-NEXT: v_sub_u32_e32 v0, v2, v16 +; GFX9-NEXT: v_sub_u32_e32 v2, v5, v18 +; GFX9-NEXT: v_sub_u32_e32 v5, v3, v12 +; GFX9-NEXT: v_cmp_ge_u32_e32 vcc, v5, v7 +; GFX9-NEXT: v_cmp_ge_u32_e64 s[0:1], v3, v12 +; GFX9-NEXT: v_add_u32_e32 v3, 1, v11 +; GFX9-NEXT: s_and_b64 vcc, vcc, s[0:1] +; GFX9-NEXT: v_add_u32_e32 v5, -1, v11 +; GFX9-NEXT: v_cndmask_b32_e32 v3, v11, v3, vcc +; GFX9-NEXT: v_sub_u32_e32 v1, v4, v17 +; GFX9-NEXT: v_xor_b32_e32 v4, v14, v15 +; GFX9-NEXT: v_cndmask_b32_e64 v3, v5, v3, s[0:1] +; GFX9-NEXT: v_xor_b32_e32 v3, v3, v4 +; GFX9-NEXT: v_sub_u32_e32 v3, v3, v4 +; GFX9-NEXT: buffer_store_dwordx4 v[0:3], off, s[12:15], 0 +; GFX9-NEXT: s_endpgm +; +; EG-LABEL: sdiv_v4i32: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @12, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @8 +; EG-NEXT: ALU 2, @13, KC0[], KC1[] +; EG-NEXT: TEX 0 @10 +; EG-NEXT: ALU 114, @16, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T3.XYZW, T0.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 8: +; EG-NEXT: VTX_READ_128 T1.XYZW, T0.X, 16, #1 +; EG-NEXT: Fetch clause starting at 10: +; EG-NEXT: VTX_READ_128 T3.XYZW, T0.X, 0, #1 +; EG-NEXT: ALU clause starting at 12: +; EG-NEXT: MOV * T0.X, KC0[2].Z, +; EG-NEXT: ALU clause starting at 13: +; EG-NEXT: SETGT_INT * T0.W, 0.0, T1.Z, +; EG-NEXT: ADD_INT * T2.W, T1.Z, PV.W, +; EG-NEXT: XOR_INT * T2.W, PV.W, T0.W, +; EG-NEXT: ALU clause starting at 16: +; EG-NEXT: RECIP_UINT * T0.X, T2.W, +; EG-NEXT: MULLO_INT * T0.Y, PS, T2.W, +; EG-NEXT: SUB_INT T4.W, 0.0, PS, +; EG-NEXT: MULHI * T0.Z, T0.X, T2.W, +; EG-NEXT: CNDE_INT T4.W, PS, PV.W, T0.Y, +; EG-NEXT: SETGT_INT * T5.W, 0.0, T3.Z, +; EG-NEXT: MULHI * T0.Y, PV.W, T0.X, +; EG-NEXT: SETGT_INT T2.Y, 0.0, T1.W, +; EG-NEXT: ADD_INT T1.Z, T3.Z, T5.W, BS:VEC_021/SCL_122 +; EG-NEXT: ADD_INT T4.W, T0.X, PS, +; EG-NEXT: SUB_INT * T6.W, T0.X, PS, +; EG-NEXT: CNDE_INT T0.Z, T0.Z, PV.W, PS, +; EG-NEXT: XOR_INT T4.W, PV.Z, T5.W, +; EG-NEXT: ADD_INT * T1.W, T1.W, PV.Y, +; EG-NEXT: XOR_INT T1.W, PS, T2.Y, +; EG-NEXT: MULHI * T0.X, PV.Z, PV.W, +; EG-NEXT: SETGT_INT T6.W, 0.0, T1.Y, +; EG-NEXT: RECIP_UINT * T0.Y, PV.W, +; EG-NEXT: ADD_INT T7.W, T1.Y, PV.W, +; EG-NEXT: MULLO_INT * T0.Z, PS, T1.W, +; EG-NEXT: XOR_INT T1.Z, PV.W, T6.W, BS:VEC_021/SCL_122 +; EG-NEXT: SUB_INT T7.W, 0.0, PS, +; EG-NEXT: MULHI * T1.Y, T0.Y, T1.W, +; EG-NEXT: CNDE_INT T7.W, PS, PV.W, T0.Z, +; EG-NEXT: RECIP_UINT * T0.Z, PV.Z, +; EG-NEXT: SETGT_INT T8.W, 0.0, T3.W, +; EG-NEXT: MULHI * T2.X, PV.W, T0.Y, +; EG-NEXT: ADD_INT T4.Y, T3.W, PV.W, +; EG-NEXT: ADD_INT T2.Z, T0.Y, PS, +; EG-NEXT: SUB_INT T3.W, T0.Y, PS, +; EG-NEXT: MULLO_INT * T0.Y, T0.Z, T1.Z, +; EG-NEXT: CNDE_INT T2.X, T1.Y, PV.Z, PV.W, +; EG-NEXT: XOR_INT T1.Y, PV.Y, T8.W, +; EG-NEXT: SETGT_INT T2.Z, 0.0, T1.X, +; EG-NEXT: SUB_INT T3.W, 0.0, PS, +; EG-NEXT: MULHI * T3.Z, T0.Z, T1.Z, +; EG-NEXT: CNDE_INT T4.Z, PS, PV.W, T0.Y, +; EG-NEXT: ADD_INT T3.W, T1.X, PV.Z, +; EG-NEXT: MULHI * T0.Y, PV.X, PV.Y, +; EG-NEXT: XOR_INT T3.W, PV.W, T2.Z, BS:VEC_021/SCL_122 +; EG-NEXT: MULHI * T1.X, PV.Z, T0.Z, +; EG-NEXT: RECIP_UINT * T2.X, PV.W, +; EG-NEXT: MULLO_INT * T4.X, PS, T3.W, +; EG-NEXT: SETGT_INT T4.Z, 0.0, T3.Y, +; EG-NEXT: SUB_INT T7.W, 0.0, PS, +; EG-NEXT: MULHI * T4.Y, T2.X, T3.W, +; EG-NEXT: CNDE_INT T4.X, PS, PV.W, T4.X, +; EG-NEXT: ADD_INT T3.Y, T3.Y, PV.Z, +; EG-NEXT: ADD_INT T5.Z, T0.Z, T1.X, +; EG-NEXT: SUB_INT T7.W, T0.Z, T1.X, +; EG-NEXT: MULLO_INT * T0.Z, T0.Y, T1.W, +; EG-NEXT: CNDE_INT T5.Y, T3.Z, PV.Z, PV.W, +; EG-NEXT: XOR_INT T3.Z, PV.Y, T4.Z, +; EG-NEXT: SUB_INT T7.W, T1.Y, PS, +; EG-NEXT: MULHI * T1.X, PV.X, T2.X, +; EG-NEXT: SETGE_UINT T5.Z, PV.W, T1.W, +; EG-NEXT: SETGE_UINT T1.W, T1.Y, T0.Z, +; EG-NEXT: MULHI * T0.Z, PV.Y, PV.Z, +; EG-NEXT: AND_INT T1.Y, PV.Z, PV.W, +; EG-NEXT: ADD_INT T5.Z, T0.Y, 1, +; EG-NEXT: SETGT_INT T7.W, 0.0, T3.X, +; EG-NEXT: MULLO_INT * T3.Y, PS, T1.Z, +; EG-NEXT: SUB_INT T4.X, T3.Z, PS, +; EG-NEXT: ADD_INT T5.Y, T3.X, PV.W, +; EG-NEXT: ADD_INT T6.Z, T2.X, T1.X, BS:VEC_120/SCL_212 +; EG-NEXT: SUB_INT * T9.W, T2.X, T1.X, BS:VEC_120/SCL_212 +; EG-NEXT: MULLO_INT * T1.X, T0.X, T2.W, +; EG-NEXT: CNDE_INT T2.X, T4.Y, T6.Z, T9.W, +; EG-NEXT: XOR_INT T4.Y, T5.Y, T7.W, BS:VEC_201 +; EG-NEXT: SUB_INT T6.Z, T4.W, PS, BS:VEC_120/SCL_212 +; EG-NEXT: SETGE_UINT T9.W, T4.X, T1.Z, BS:VEC_102/SCL_221 +; EG-NEXT: SETGE_UINT * T10.W, T3.Z, T3.Y, +; EG-NEXT: AND_INT T3.X, PV.W, PS, +; EG-NEXT: ADD_INT T3.Y, T0.Z, 1, +; EG-NEXT: SETGE_UINT T1.Z, PV.Z, T2.W, +; EG-NEXT: SETGE_UINT T2.W, T4.W, T1.X, +; EG-NEXT: MULHI * T1.X, PV.X, PV.Y, +; EG-NEXT: AND_INT T2.X, PV.Z, PV.W, +; EG-NEXT: ADD_INT T5.Y, T0.X, 1, +; EG-NEXT: CNDE_INT T1.Z, PV.X, T0.Z, PV.Y, +; EG-NEXT: ADD_INT T4.W, T0.Z, literal.x, +; EG-NEXT: MULLO_INT * T0.Z, PS, T3.W, +; EG-NEXT: -1(nan), 0(0.000000e+00) +; EG-NEXT: CNDE_INT T3.X, T10.W, PV.W, PV.Z, +; EG-NEXT: CNDE_INT T3.Y, PV.X, T0.X, PV.Y, +; EG-NEXT: CNDE_INT T1.Z, T1.Y, T0.Y, T5.Z, +; EG-NEXT: ADD_INT T4.W, T0.Y, literal.x, BS:VEC_120/SCL_212 +; EG-NEXT: SUB_INT * T9.W, T4.Y, PS, +; EG-NEXT: -1(nan), 0(0.000000e+00) +; EG-NEXT: ADD_INT T0.X, T0.X, literal.x, +; EG-NEXT: SETGE_UINT T0.Y, PS, T3.W, +; EG-NEXT: SETGE_UINT T0.Z, T4.Y, T0.Z, +; EG-NEXT: CNDE_INT T1.W, T1.W, PV.W, PV.Z, +; EG-NEXT: XOR_INT * T3.W, T8.W, T2.Y, +; EG-NEXT: -1(nan), 0(0.000000e+00) +; EG-NEXT: XOR_INT T2.X, PV.W, PS, +; EG-NEXT: AND_INT T0.Y, PV.Y, PV.Z, +; EG-NEXT: ADD_INT T1.Z, T1.X, 1, +; EG-NEXT: CNDE_INT T1.W, T2.W, PV.X, T3.Y, +; EG-NEXT: XOR_INT * T0.W, T5.W, T0.W, +; EG-NEXT: XOR_INT T0.X, T4.Z, T6.W, BS:VEC_021/SCL_122 +; EG-NEXT: XOR_INT T1.Y, PV.W, PS, +; EG-NEXT: CNDE_INT T1.Z, PV.Y, T1.X, PV.Z, +; EG-NEXT: ADD_INT T1.W, T1.X, literal.x, +; EG-NEXT: SUB_INT * T3.W, PV.X, T3.W, +; EG-NEXT: -1(nan), 0(0.000000e+00) +; EG-NEXT: CNDE_INT T0.Y, T0.Z, PV.W, PV.Z, +; EG-NEXT: SUB_INT T3.Z, PV.Y, T0.W, +; EG-NEXT: XOR_INT T0.W, T7.W, T2.Z, +; EG-NEXT: XOR_INT * T1.W, T3.X, PV.X, +; EG-NEXT: SUB_INT T3.Y, PS, T0.X, +; EG-NEXT: XOR_INT * T1.W, PV.Y, PV.W, +; EG-NEXT: SUB_INT T3.X, PV.W, T0.W, +; EG-NEXT: LSHR * T0.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %den_ptr = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %in, i32 1 %num = load <4 x i32>, <4 x i32> addrspace(1) * %in %den = load <4 x i32>, <4 x i32> addrspace(1) * %den_ptr @@ -77,17 +1433,284 @@ define amdgpu_kernel void @sdiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad } define amdgpu_kernel void @sdiv_v4i32_4(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) { +; GCN-LABEL: sdiv_v4i32_4: +; GCN: ; %bb.0: +; GCN-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: s_mov_b32 s10, s6 +; GCN-NEXT: s_mov_b32 s11, s7 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b32 s8, s2 +; GCN-NEXT: s_mov_b32 s9, s3 +; GCN-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0 +; GCN-NEXT: s_mov_b32 s4, s0 +; GCN-NEXT: s_mov_b32 s5, s1 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_ashrrev_i32_e32 v4, 31, v0 +; GCN-NEXT: v_ashrrev_i32_e32 v5, 31, v1 +; GCN-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-NEXT: v_ashrrev_i32_e32 v7, 31, v3 +; GCN-NEXT: v_lshrrev_b32_e32 v4, 30, v4 +; GCN-NEXT: v_lshrrev_b32_e32 v5, 30, v5 +; GCN-NEXT: v_lshrrev_b32_e32 v6, 30, v6 +; GCN-NEXT: v_lshrrev_b32_e32 v7, 30, v7 +; GCN-NEXT: v_add_i32_e32 v0, vcc, v4, v0 +; GCN-NEXT: v_add_i32_e32 v1, vcc, v5, v1 +; GCN-NEXT: v_add_i32_e32 v2, vcc, v6, v2 +; GCN-NEXT: v_add_i32_e32 v3, vcc, v7, v3 +; GCN-NEXT: v_ashrrev_i32_e32 v0, 2, v0 +; GCN-NEXT: v_ashrrev_i32_e32 v1, 2, v1 +; GCN-NEXT: v_ashrrev_i32_e32 v2, 2, v2 +; GCN-NEXT: v_ashrrev_i32_e32 v3, 2, v3 +; GCN-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; TONGA-LABEL: sdiv_v4i32_4: +; TONGA: ; %bb.0: +; TONGA-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; TONGA-NEXT: s_mov_b32 s3, 0xf000 +; TONGA-NEXT: s_mov_b32 s2, -1 +; TONGA-NEXT: s_waitcnt lgkmcnt(0) +; TONGA-NEXT: s_mov_b32 s0, s4 +; TONGA-NEXT: s_mov_b32 s1, s5 +; TONGA-NEXT: s_mov_b32 s4, s6 +; TONGA-NEXT: s_mov_b32 s5, s7 +; TONGA-NEXT: s_mov_b32 s6, s2 +; TONGA-NEXT: s_mov_b32 s7, s3 +; TONGA-NEXT: buffer_load_dwordx4 v[0:3], off, s[4:7], 0 +; TONGA-NEXT: s_waitcnt vmcnt(0) +; TONGA-NEXT: v_ashrrev_i32_e32 v4, 31, v0 +; TONGA-NEXT: v_ashrrev_i32_e32 v5, 31, v1 +; TONGA-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; TONGA-NEXT: v_ashrrev_i32_e32 v7, 31, v3 +; TONGA-NEXT: v_lshrrev_b32_e32 v4, 30, v4 +; TONGA-NEXT: v_lshrrev_b32_e32 v5, 30, v5 +; TONGA-NEXT: v_lshrrev_b32_e32 v6, 30, v6 +; TONGA-NEXT: v_lshrrev_b32_e32 v7, 30, v7 +; TONGA-NEXT: v_add_u32_e32 v0, vcc, v4, v0 +; TONGA-NEXT: v_add_u32_e32 v1, vcc, v5, v1 +; TONGA-NEXT: v_add_u32_e32 v2, vcc, v6, v2 +; TONGA-NEXT: v_add_u32_e32 v3, vcc, v7, v3 +; TONGA-NEXT: v_ashrrev_i32_e32 v0, 2, v0 +; TONGA-NEXT: v_ashrrev_i32_e32 v1, 2, v1 +; TONGA-NEXT: v_ashrrev_i32_e32 v2, 2, v2 +; TONGA-NEXT: v_ashrrev_i32_e32 v3, 2, v3 +; TONGA-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0 +; TONGA-NEXT: s_endpgm +; +; GFX9-LABEL: sdiv_v4i32_4: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; GFX9-NEXT: s_mov_b32 s3, 0xf000 +; GFX9-NEXT: s_mov_b32 s2, -1 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s0, s4 +; GFX9-NEXT: s_mov_b32 s1, s5 +; GFX9-NEXT: s_mov_b32 s4, s6 +; GFX9-NEXT: s_mov_b32 s5, s7 +; GFX9-NEXT: s_mov_b32 s6, s2 +; GFX9-NEXT: s_mov_b32 s7, s3 +; GFX9-NEXT: buffer_load_dwordx4 v[0:3], off, s[4:7], 0 +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_ashrrev_i32_e32 v4, 31, v0 +; GFX9-NEXT: v_ashrrev_i32_e32 v5, 31, v1 +; GFX9-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GFX9-NEXT: v_ashrrev_i32_e32 v7, 31, v3 +; GFX9-NEXT: v_lshrrev_b32_e32 v4, 30, v4 +; GFX9-NEXT: v_lshrrev_b32_e32 v5, 30, v5 +; GFX9-NEXT: v_lshrrev_b32_e32 v6, 30, v6 +; GFX9-NEXT: v_lshrrev_b32_e32 v7, 30, v7 +; GFX9-NEXT: v_add_u32_e32 v0, v0, v4 +; GFX9-NEXT: v_add_u32_e32 v1, v1, v5 +; GFX9-NEXT: v_add_u32_e32 v2, v2, v6 +; GFX9-NEXT: v_add_u32_e32 v3, v3, v7 +; GFX9-NEXT: v_ashrrev_i32_e32 v0, 2, v0 +; GFX9-NEXT: v_ashrrev_i32_e32 v1, 2, v1 +; GFX9-NEXT: v_ashrrev_i32_e32 v2, 2, v2 +; GFX9-NEXT: v_ashrrev_i32_e32 v3, 2, v3 +; GFX9-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0 +; GFX9-NEXT: s_endpgm +; +; EG-LABEL: sdiv_v4i32_4: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 24, @9, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T1.XYZW, T0.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 6: +; EG-NEXT: VTX_READ_128 T0.XYZW, T0.X, 0, #1 +; EG-NEXT: ALU clause starting at 8: +; EG-NEXT: MOV * T0.X, KC0[2].Z, +; EG-NEXT: ALU clause starting at 9: +; EG-NEXT: ASHR T1.W, T0.W, literal.x, +; EG-NEXT: ASHR * T2.W, T0.Z, literal.x, +; EG-NEXT: 31(4.344025e-44), 0(0.000000e+00) +; EG-NEXT: LSHR * T1.W, PV.W, literal.x, +; EG-NEXT: 30(4.203895e-44), 0(0.000000e+00) +; EG-NEXT: ADD_INT T1.Z, T0.W, PV.W, +; EG-NEXT: LSHR T0.W, T2.W, literal.x, BS:VEC_120/SCL_212 +; EG-NEXT: ASHR * T1.W, T0.Y, literal.y, +; EG-NEXT: 30(4.203895e-44), 31(4.344025e-44) +; EG-NEXT: LSHR T1.Y, PS, literal.x, +; EG-NEXT: ASHR T2.Z, T0.X, literal.y, +; EG-NEXT: ADD_INT T0.W, T0.Z, PV.W, +; EG-NEXT: ASHR * T1.W, PV.Z, literal.z, +; EG-NEXT: 30(4.203895e-44), 31(4.344025e-44) +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; EG-NEXT: ASHR T1.Z, PV.W, literal.x, +; EG-NEXT: LSHR T0.W, PV.Z, literal.y, +; EG-NEXT: ADD_INT * T2.W, T0.Y, PV.Y, +; EG-NEXT: 2(2.802597e-45), 30(4.203895e-44) +; EG-NEXT: ASHR T1.Y, PS, literal.x, +; EG-NEXT: ADD_INT * T0.W, T0.X, PV.W, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; EG-NEXT: ASHR T1.X, PV.W, literal.x, +; EG-NEXT: LSHR * T0.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %num = load <4 x i32>, <4 x i32> addrspace(1) * %in %result = sdiv <4 x i32> %num, store <4 x i32> %result, <4 x i32> addrspace(1)* %out ret void } -; FUNC-LABEL: {{^}}v_sdiv_i8: -; SI: v_rcp_iflag_f32 -; SI: v_bfe_i32 [[BFE:v[0-9]+]], v{{[0-9]+}}, 0, 8 -; SI: buffer_store_dword [[BFE]] define amdgpu_kernel void @v_sdiv_i8(i32 addrspace(1)* %out, i8 addrspace(1)* %in) { +; GCN-LABEL: v_sdiv_i8: +; GCN: ; %bb.0: +; GCN-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: s_mov_b32 s10, s6 +; GCN-NEXT: s_mov_b32 s11, s7 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b32 s8, s2 +; GCN-NEXT: s_mov_b32 s9, s3 +; GCN-NEXT: buffer_load_sbyte v0, off, s[8:11], 0 +; GCN-NEXT: buffer_load_sbyte v1, off, s[8:11], 0 offset:1 +; GCN-NEXT: s_mov_b32 s4, s0 +; GCN-NEXT: s_mov_b32 s5, s1 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_xor_b32_e32 v2, v0, v1 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GCN-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GCN-NEXT: v_ashrrev_i32_e32 v2, 30, v2 +; GCN-NEXT: v_rcp_iflag_f32_e32 v3, v1 +; GCN-NEXT: v_or_b32_e32 v2, 1, v2 +; GCN-NEXT: v_mul_f32_e32 v3, v0, v3 +; GCN-NEXT: v_trunc_f32_e32 v3, v3 +; GCN-NEXT: v_mad_f32 v0, -v3, v1, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v3, v3 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v3 +; GCN-NEXT: v_bfe_i32 v0, v0, 0, 8 +; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; TONGA-LABEL: v_sdiv_i8: +; TONGA: ; %bb.0: +; TONGA-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; TONGA-NEXT: s_mov_b32 s3, 0xf000 +; TONGA-NEXT: s_mov_b32 s2, -1 +; TONGA-NEXT: s_waitcnt lgkmcnt(0) +; TONGA-NEXT: s_mov_b32 s0, s4 +; TONGA-NEXT: s_mov_b32 s1, s5 +; TONGA-NEXT: s_mov_b32 s4, s6 +; TONGA-NEXT: s_mov_b32 s5, s7 +; TONGA-NEXT: s_mov_b32 s6, s2 +; TONGA-NEXT: s_mov_b32 s7, s3 +; TONGA-NEXT: buffer_load_sbyte v0, off, s[4:7], 0 offset:1 +; TONGA-NEXT: buffer_load_sbyte v2, off, s[4:7], 0 +; TONGA-NEXT: s_waitcnt vmcnt(1) +; TONGA-NEXT: v_cvt_f32_i32_e32 v1, v0 +; TONGA-NEXT: s_waitcnt vmcnt(0) +; TONGA-NEXT: v_cvt_f32_i32_e32 v3, v2 +; TONGA-NEXT: v_xor_b32_e32 v0, v2, v0 +; TONGA-NEXT: v_ashrrev_i32_e32 v0, 30, v0 +; TONGA-NEXT: v_rcp_iflag_f32_e32 v4, v1 +; TONGA-NEXT: v_or_b32_e32 v0, 1, v0 +; TONGA-NEXT: v_mul_f32_e32 v2, v3, v4 +; TONGA-NEXT: v_trunc_f32_e32 v2, v2 +; TONGA-NEXT: v_mad_f32 v3, -v2, v1, v3 +; TONGA-NEXT: v_cvt_i32_f32_e32 v2, v2 +; TONGA-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v1| +; TONGA-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc +; TONGA-NEXT: v_add_u32_e32 v0, vcc, v0, v2 +; TONGA-NEXT: v_bfe_i32 v0, v0, 0, 8 +; TONGA-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; TONGA-NEXT: s_endpgm +; +; GFX9-LABEL: v_sdiv_i8: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; GFX9-NEXT: s_mov_b32 s3, 0xf000 +; GFX9-NEXT: s_mov_b32 s2, -1 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s0, s4 +; GFX9-NEXT: s_mov_b32 s1, s5 +; GFX9-NEXT: s_mov_b32 s4, s6 +; GFX9-NEXT: s_mov_b32 s5, s7 +; GFX9-NEXT: s_mov_b32 s6, s2 +; GFX9-NEXT: s_mov_b32 s7, s3 +; GFX9-NEXT: buffer_load_sbyte v0, off, s[4:7], 0 offset:1 +; GFX9-NEXT: buffer_load_sbyte v2, off, s[4:7], 0 +; GFX9-NEXT: s_waitcnt vmcnt(1) +; GFX9-NEXT: v_cvt_f32_i32_e32 v1, v0 +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_cvt_f32_i32_e32 v3, v2 +; GFX9-NEXT: v_xor_b32_e32 v0, v2, v0 +; GFX9-NEXT: v_ashrrev_i32_e32 v0, 30, v0 +; GFX9-NEXT: v_rcp_iflag_f32_e32 v4, v1 +; GFX9-NEXT: v_or_b32_e32 v0, 1, v0 +; GFX9-NEXT: v_mul_f32_e32 v2, v3, v4 +; GFX9-NEXT: v_trunc_f32_e32 v2, v2 +; GFX9-NEXT: v_cvt_i32_f32_e32 v4, v2 +; GFX9-NEXT: v_mad_f32 v2, -v2, v1, v3 +; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v2|, |v1| +; GFX9-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc +; GFX9-NEXT: v_add_u32_e32 v0, v4, v0 +; GFX9-NEXT: v_bfe_i32 v0, v0, 0, 8 +; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX9-NEXT: s_endpgm +; +; EG-LABEL: v_sdiv_i8: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @10, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 1 @6 +; EG-NEXT: ALU 21, @11, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 6: +; EG-NEXT: VTX_READ_8 T1.X, T0.X, 1, #1 +; EG-NEXT: VTX_READ_8 T0.X, T0.X, 0, #1 +; EG-NEXT: ALU clause starting at 10: +; EG-NEXT: MOV * T0.X, KC0[2].Z, +; EG-NEXT: ALU clause starting at 11: +; EG-NEXT: BFE_INT * T0.W, T1.X, 0.0, literal.x, +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: INT_TO_FLT * T0.Y, PV.W, +; EG-NEXT: BFE_INT T1.W, T0.X, 0.0, literal.x, +; EG-NEXT: RECIP_IEEE * T0.X, PS, +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: INT_TO_FLT * T0.Z, PV.W, +; EG-NEXT: MUL_IEEE * T2.W, PS, T0.X, +; EG-NEXT: TRUNC T2.W, PV.W, +; EG-NEXT: XOR_INT * T0.W, T1.W, T0.W, +; EG-NEXT: ASHR T0.W, PS, literal.x, +; EG-NEXT: MULADD_IEEE * T1.W, -PV.W, T0.Y, T0.Z, +; EG-NEXT: 30(4.203895e-44), 0(0.000000e+00) +; EG-NEXT: TRUNC T0.Z, T2.W, +; EG-NEXT: SETGE T1.W, |PS|, |T0.Y|, +; EG-NEXT: OR_INT * T0.W, PV.W, 1, +; EG-NEXT: CNDE T0.W, PV.W, 0.0, PS, +; EG-NEXT: FLT_TO_INT * T1.W, PV.Z, +; EG-NEXT: ADD_INT * T0.W, PS, PV.W, +; EG-NEXT: BFE_INT T0.X, PV.W, 0.0, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: 8(1.121039e-44), 2(2.802597e-45) %den_ptr = getelementptr i8, i8 addrspace(1)* %in, i8 1 %num = load i8, i8 addrspace(1) * %in %den = load i8, i8 addrspace(1) * %den_ptr @@ -97,11 +1720,184 @@ define amdgpu_kernel void @v_sdiv_i8(i32 addrspace(1)* %out, i8 addrspace(1)* %i ret void } -; FUNC-LABEL: {{^}}v_sdiv_i23: -; SI: v_rcp_iflag_f32 -; SI: v_bfe_i32 [[BFE:v[0-9]+]], v{{[0-9]+}}, 0, 23 -; SI: buffer_store_dword [[BFE]] define amdgpu_kernel void @v_sdiv_i23(i32 addrspace(1)* %out, i23 addrspace(1)* %in) { +; GCN-LABEL: v_sdiv_i23: +; GCN: ; %bb.0: +; GCN-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: s_mov_b32 s10, s6 +; GCN-NEXT: s_mov_b32 s11, s7 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b32 s4, s0 +; GCN-NEXT: s_mov_b32 s5, s1 +; GCN-NEXT: s_mov_b32 s8, s2 +; GCN-NEXT: s_mov_b32 s9, s3 +; GCN-NEXT: buffer_load_ubyte v1, off, s[8:11], 0 offset:2 +; GCN-NEXT: buffer_load_ubyte v3, off, s[8:11], 0 offset:6 +; GCN-NEXT: buffer_load_ushort v0, off, s[8:11], 0 +; GCN-NEXT: buffer_load_ushort v2, off, s[8:11], 0 offset:4 +; GCN-NEXT: s_waitcnt vmcnt(3) +; GCN-NEXT: v_lshlrev_b32_e32 v1, 16, v1 +; GCN-NEXT: s_waitcnt vmcnt(2) +; GCN-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GCN-NEXT: s_waitcnt vmcnt(1) +; GCN-NEXT: v_or_b32_e32 v0, v0, v1 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_or_b32_e32 v1, v2, v3 +; GCN-NEXT: v_bfe_i32 v0, v0, 0, 23 +; GCN-NEXT: v_bfe_i32 v1, v1, 0, 23 +; GCN-NEXT: v_xor_b32_e32 v2, v0, v1 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GCN-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GCN-NEXT: v_ashrrev_i32_e32 v2, 30, v2 +; GCN-NEXT: v_rcp_iflag_f32_e32 v3, v1 +; GCN-NEXT: v_or_b32_e32 v2, 1, v2 +; GCN-NEXT: v_mul_f32_e32 v3, v0, v3 +; GCN-NEXT: v_trunc_f32_e32 v3, v3 +; GCN-NEXT: v_mad_f32 v0, -v3, v1, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v3, v3 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v3 +; GCN-NEXT: v_bfe_i32 v0, v0, 0, 23 +; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; TONGA-LABEL: v_sdiv_i23: +; TONGA: ; %bb.0: +; TONGA-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; TONGA-NEXT: s_mov_b32 s3, 0xf000 +; TONGA-NEXT: s_mov_b32 s2, -1 +; TONGA-NEXT: s_waitcnt lgkmcnt(0) +; TONGA-NEXT: s_mov_b32 s0, s4 +; TONGA-NEXT: s_mov_b32 s1, s5 +; TONGA-NEXT: s_mov_b32 s4, s6 +; TONGA-NEXT: s_mov_b32 s5, s7 +; TONGA-NEXT: s_mov_b32 s6, s2 +; TONGA-NEXT: s_mov_b32 s7, s3 +; TONGA-NEXT: buffer_load_ubyte v0, off, s[4:7], 0 offset:2 +; TONGA-NEXT: buffer_load_ushort v1, off, s[4:7], 0 offset:4 +; TONGA-NEXT: buffer_load_ubyte v2, off, s[4:7], 0 offset:6 +; TONGA-NEXT: buffer_load_ushort v3, off, s[4:7], 0 +; TONGA-NEXT: s_waitcnt vmcnt(3) +; TONGA-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; TONGA-NEXT: s_waitcnt vmcnt(1) +; TONGA-NEXT: v_lshlrev_b32_e32 v2, 16, v2 +; TONGA-NEXT: v_or_b32_e32 v1, v1, v2 +; TONGA-NEXT: v_bfe_i32 v1, v1, 0, 23 +; TONGA-NEXT: v_cvt_f32_i32_e32 v2, v1 +; TONGA-NEXT: s_waitcnt vmcnt(0) +; TONGA-NEXT: v_or_b32_e32 v0, v3, v0 +; TONGA-NEXT: v_bfe_i32 v0, v0, 0, 23 +; TONGA-NEXT: v_cvt_f32_i32_e32 v3, v0 +; TONGA-NEXT: v_rcp_iflag_f32_e32 v4, v2 +; TONGA-NEXT: v_xor_b32_e32 v0, v0, v1 +; TONGA-NEXT: v_ashrrev_i32_e32 v0, 30, v0 +; TONGA-NEXT: v_or_b32_e32 v0, 1, v0 +; TONGA-NEXT: v_mul_f32_e32 v1, v3, v4 +; TONGA-NEXT: v_trunc_f32_e32 v1, v1 +; TONGA-NEXT: v_mad_f32 v3, -v1, v2, v3 +; TONGA-NEXT: v_cvt_i32_f32_e32 v1, v1 +; TONGA-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v2| +; TONGA-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc +; TONGA-NEXT: v_add_u32_e32 v0, vcc, v0, v1 +; TONGA-NEXT: v_bfe_i32 v0, v0, 0, 23 +; TONGA-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; TONGA-NEXT: s_endpgm +; +; GFX9-LABEL: v_sdiv_i23: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; GFX9-NEXT: s_mov_b32 s3, 0xf000 +; GFX9-NEXT: s_mov_b32 s2, -1 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s0, s4 +; GFX9-NEXT: s_mov_b32 s1, s5 +; GFX9-NEXT: s_mov_b32 s4, s6 +; GFX9-NEXT: s_mov_b32 s5, s7 +; GFX9-NEXT: s_mov_b32 s6, s2 +; GFX9-NEXT: s_mov_b32 s7, s3 +; GFX9-NEXT: buffer_load_ubyte v0, off, s[4:7], 0 offset:2 +; GFX9-NEXT: buffer_load_ushort v1, off, s[4:7], 0 offset:4 +; GFX9-NEXT: buffer_load_ubyte v2, off, s[4:7], 0 offset:6 +; GFX9-NEXT: buffer_load_ushort v3, off, s[4:7], 0 +; GFX9-NEXT: s_waitcnt vmcnt(3) +; GFX9-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX9-NEXT: s_waitcnt vmcnt(1) +; GFX9-NEXT: v_lshlrev_b32_e32 v2, 16, v2 +; GFX9-NEXT: v_or_b32_e32 v1, v1, v2 +; GFX9-NEXT: v_bfe_i32 v1, v1, 0, 23 +; GFX9-NEXT: v_cvt_f32_i32_e32 v2, v1 +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_or_b32_e32 v0, v3, v0 +; GFX9-NEXT: v_bfe_i32 v0, v0, 0, 23 +; GFX9-NEXT: v_cvt_f32_i32_e32 v3, v0 +; GFX9-NEXT: v_rcp_iflag_f32_e32 v4, v2 +; GFX9-NEXT: v_xor_b32_e32 v0, v0, v1 +; GFX9-NEXT: v_ashrrev_i32_e32 v0, 30, v0 +; GFX9-NEXT: v_or_b32_e32 v0, 1, v0 +; GFX9-NEXT: v_mul_f32_e32 v1, v3, v4 +; GFX9-NEXT: v_trunc_f32_e32 v1, v1 +; GFX9-NEXT: v_cvt_i32_f32_e32 v4, v1 +; GFX9-NEXT: v_mad_f32 v1, -v1, v2, v3 +; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, |v2| +; GFX9-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc +; GFX9-NEXT: v_add_u32_e32 v0, v4, v0 +; GFX9-NEXT: v_bfe_i32 v0, v0, 0, 23 +; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX9-NEXT: s_endpgm +; +; EG-LABEL: v_sdiv_i23: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @14, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 3 @6 +; EG-NEXT: ALU 33, @15, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 6: +; EG-NEXT: VTX_READ_8 T1.X, T0.X, 6, #1 +; EG-NEXT: VTX_READ_16 T2.X, T0.X, 0, #1 +; EG-NEXT: VTX_READ_8 T3.X, T0.X, 2, #1 +; EG-NEXT: VTX_READ_16 T0.X, T0.X, 4, #1 +; EG-NEXT: ALU clause starting at 14: +; EG-NEXT: MOV * T0.X, KC0[2].Z, +; EG-NEXT: ALU clause starting at 15: +; EG-NEXT: LSHL * T0.W, T1.X, literal.x, +; EG-NEXT: 16(2.242078e-44), 0(0.000000e+00) +; EG-NEXT: OR_INT T0.W, T0.X, PV.W, +; EG-NEXT: LSHL * T1.W, T3.X, literal.x, +; EG-NEXT: 16(2.242078e-44), 0(0.000000e+00) +; EG-NEXT: LSHL * T0.W, PV.W, literal.x, +; EG-NEXT: 9(1.261169e-44), 0(0.000000e+00) +; EG-NEXT: ASHR T0.W, PV.W, literal.x, +; EG-NEXT: OR_INT * T1.W, T2.X, T1.W, +; EG-NEXT: 9(1.261169e-44), 0(0.000000e+00) +; EG-NEXT: LSHL T1.W, PS, literal.x, +; EG-NEXT: INT_TO_FLT * T0.X, PV.W, +; EG-NEXT: 9(1.261169e-44), 0(0.000000e+00) +; EG-NEXT: ASHR T1.W, PV.W, literal.x, +; EG-NEXT: RECIP_IEEE * T0.Y, PS, +; EG-NEXT: 9(1.261169e-44), 0(0.000000e+00) +; EG-NEXT: INT_TO_FLT * T0.Z, PV.W, +; EG-NEXT: MUL_IEEE * T2.W, PS, T0.Y, +; EG-NEXT: TRUNC T2.W, PV.W, +; EG-NEXT: XOR_INT * T0.W, T1.W, T0.W, +; EG-NEXT: ASHR T0.W, PS, literal.x, +; EG-NEXT: MULADD_IEEE * T1.W, -PV.W, T0.X, T0.Z, +; EG-NEXT: 30(4.203895e-44), 0(0.000000e+00) +; EG-NEXT: TRUNC T0.Z, T2.W, +; EG-NEXT: SETGE T1.W, |PS|, |T0.X|, +; EG-NEXT: OR_INT * T0.W, PV.W, 1, +; EG-NEXT: CNDE T0.W, PV.W, 0.0, PS, +; EG-NEXT: FLT_TO_INT * T1.W, PV.Z, +; EG-NEXT: ADD_INT * T0.W, PS, PV.W, +; EG-NEXT: LSHL * T0.W, PV.W, literal.x, +; EG-NEXT: 9(1.261169e-44), 0(0.000000e+00) +; EG-NEXT: ASHR T0.X, PV.W, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: 9(1.261169e-44), 2(2.802597e-45) %den_ptr = getelementptr i23, i23 addrspace(1)* %in, i23 1 %num = load i23, i23 addrspace(1) * %in %den = load i23, i23 addrspace(1) * %den_ptr @@ -111,11 +1907,187 @@ define amdgpu_kernel void @v_sdiv_i23(i32 addrspace(1)* %out, i23 addrspace(1)* ret void } -; FUNC-LABEL: {{^}}v_sdiv_i24: -; SI: v_rcp_iflag_f32 -; SI: v_bfe_i32 [[BFE:v[0-9]+]], v{{[0-9]+}}, 0, 24 -; SI: buffer_store_dword [[BFE]] define amdgpu_kernel void @v_sdiv_i24(i32 addrspace(1)* %out, i24 addrspace(1)* %in) { +; GCN-LABEL: v_sdiv_i24: +; GCN: ; %bb.0: +; GCN-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: s_mov_b32 s10, s6 +; GCN-NEXT: s_mov_b32 s11, s7 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b32 s4, s0 +; GCN-NEXT: s_mov_b32 s5, s1 +; GCN-NEXT: s_mov_b32 s8, s2 +; GCN-NEXT: s_mov_b32 s9, s3 +; GCN-NEXT: buffer_load_sbyte v1, off, s[8:11], 0 offset:2 +; GCN-NEXT: buffer_load_sbyte v3, off, s[8:11], 0 offset:6 +; GCN-NEXT: buffer_load_ushort v0, off, s[8:11], 0 +; GCN-NEXT: buffer_load_ushort v2, off, s[8:11], 0 offset:4 +; GCN-NEXT: s_waitcnt vmcnt(3) +; GCN-NEXT: v_lshlrev_b32_e32 v1, 16, v1 +; GCN-NEXT: s_waitcnt vmcnt(2) +; GCN-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GCN-NEXT: s_waitcnt vmcnt(1) +; GCN-NEXT: v_or_b32_e32 v0, v0, v1 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_or_b32_e32 v1, v2, v3 +; GCN-NEXT: v_xor_b32_e32 v2, v0, v1 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GCN-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GCN-NEXT: v_ashrrev_i32_e32 v2, 30, v2 +; GCN-NEXT: v_rcp_iflag_f32_e32 v3, v1 +; GCN-NEXT: v_or_b32_e32 v2, 1, v2 +; GCN-NEXT: v_mul_f32_e32 v3, v0, v3 +; GCN-NEXT: v_trunc_f32_e32 v3, v3 +; GCN-NEXT: v_mad_f32 v0, -v3, v1, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v3, v3 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v3 +; GCN-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; TONGA-LABEL: v_sdiv_i24: +; TONGA: ; %bb.0: +; TONGA-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; TONGA-NEXT: s_mov_b32 s3, 0xf000 +; TONGA-NEXT: s_mov_b32 s2, -1 +; TONGA-NEXT: s_waitcnt lgkmcnt(0) +; TONGA-NEXT: s_mov_b32 s0, s4 +; TONGA-NEXT: s_mov_b32 s1, s5 +; TONGA-NEXT: s_mov_b32 s4, s6 +; TONGA-NEXT: s_mov_b32 s5, s7 +; TONGA-NEXT: s_mov_b32 s6, s2 +; TONGA-NEXT: s_mov_b32 s7, s3 +; TONGA-NEXT: buffer_load_sbyte v0, off, s[4:7], 0 offset:2 +; TONGA-NEXT: buffer_load_ushort v1, off, s[4:7], 0 offset:4 +; TONGA-NEXT: buffer_load_sbyte v2, off, s[4:7], 0 offset:6 +; TONGA-NEXT: buffer_load_ushort v3, off, s[4:7], 0 +; TONGA-NEXT: s_waitcnt vmcnt(3) +; TONGA-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; TONGA-NEXT: s_waitcnt vmcnt(1) +; TONGA-NEXT: v_lshlrev_b32_e32 v2, 16, v2 +; TONGA-NEXT: v_or_b32_e32 v1, v1, v2 +; TONGA-NEXT: v_cvt_f32_i32_e32 v2, v1 +; TONGA-NEXT: s_waitcnt vmcnt(0) +; TONGA-NEXT: v_or_b32_e32 v0, v3, v0 +; TONGA-NEXT: v_cvt_f32_i32_e32 v3, v0 +; TONGA-NEXT: v_xor_b32_e32 v0, v0, v1 +; TONGA-NEXT: v_rcp_iflag_f32_e32 v4, v2 +; TONGA-NEXT: v_ashrrev_i32_e32 v0, 30, v0 +; TONGA-NEXT: v_or_b32_e32 v0, 1, v0 +; TONGA-NEXT: v_mul_f32_e32 v1, v3, v4 +; TONGA-NEXT: v_trunc_f32_e32 v1, v1 +; TONGA-NEXT: v_mad_f32 v3, -v1, v2, v3 +; TONGA-NEXT: v_cvt_i32_f32_e32 v1, v1 +; TONGA-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v2| +; TONGA-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc +; TONGA-NEXT: v_add_u32_e32 v0, vcc, v0, v1 +; TONGA-NEXT: v_bfe_i32 v0, v0, 0, 24 +; TONGA-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; TONGA-NEXT: s_endpgm +; +; GFX9-LABEL: v_sdiv_i24: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; GFX9-NEXT: s_mov_b32 s3, 0xf000 +; GFX9-NEXT: s_mov_b32 s2, -1 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s0, s4 +; GFX9-NEXT: s_mov_b32 s1, s5 +; GFX9-NEXT: s_mov_b32 s4, s6 +; GFX9-NEXT: s_mov_b32 s5, s7 +; GFX9-NEXT: s_mov_b32 s6, s2 +; GFX9-NEXT: s_mov_b32 s7, s3 +; GFX9-NEXT: buffer_load_ushort v0, off, s[4:7], 0 +; GFX9-NEXT: buffer_load_sbyte v1, off, s[4:7], 0 offset:2 +; GFX9-NEXT: buffer_load_ushort v2, off, s[4:7], 0 offset:4 +; GFX9-NEXT: buffer_load_sbyte v3, off, s[4:7], 0 offset:6 +; GFX9-NEXT: s_waitcnt vmcnt(2) +; GFX9-NEXT: v_lshlrev_b32_e32 v1, 16, v1 +; GFX9-NEXT: v_or_b32_e32 v0, v0, v1 +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX9-NEXT: v_or_b32_e32 v2, v2, v3 +; GFX9-NEXT: v_cvt_f32_i32_e32 v3, v2 +; GFX9-NEXT: v_cvt_f32_i32_e32 v1, v0 +; GFX9-NEXT: v_xor_b32_e32 v0, v0, v2 +; GFX9-NEXT: v_ashrrev_i32_e32 v0, 30, v0 +; GFX9-NEXT: v_rcp_iflag_f32_e32 v4, v3 +; GFX9-NEXT: v_or_b32_e32 v0, 1, v0 +; GFX9-NEXT: v_mul_f32_e32 v2, v1, v4 +; GFX9-NEXT: v_trunc_f32_e32 v2, v2 +; GFX9-NEXT: v_cvt_i32_f32_e32 v4, v2 +; GFX9-NEXT: v_mad_f32 v1, -v2, v3, v1 +; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, |v3| +; GFX9-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc +; GFX9-NEXT: v_add_u32_e32 v0, v4, v0 +; GFX9-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX9-NEXT: s_endpgm +; +; EG-LABEL: v_sdiv_i24: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @14, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 3 @6 +; EG-NEXT: ALU 43, @15, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 6: +; EG-NEXT: VTX_READ_8 T1.X, T0.X, 6, #1 +; EG-NEXT: VTX_READ_16 T2.X, T0.X, 0, #1 +; EG-NEXT: VTX_READ_8 T3.X, T0.X, 2, #1 +; EG-NEXT: VTX_READ_16 T0.X, T0.X, 4, #1 +; EG-NEXT: ALU clause starting at 14: +; EG-NEXT: MOV * T0.X, KC0[2].Z, +; EG-NEXT: ALU clause starting at 15: +; EG-NEXT: BFE_INT * T0.W, T1.X, 0.0, literal.x, +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: LSHL * T0.W, PV.W, literal.x, +; EG-NEXT: 16(2.242078e-44), 0(0.000000e+00) +; EG-NEXT: OR_INT * T0.W, T0.X, PV.W, +; EG-NEXT: SETGT_INT * T1.W, 0.0, PV.W, +; EG-NEXT: ADD_INT * T0.W, T0.W, PV.W, +; EG-NEXT: XOR_INT * T0.W, PV.W, T1.W, +; EG-NEXT: RECIP_UINT * T0.X, PV.W, +; EG-NEXT: BFE_INT T2.W, T3.X, 0.0, literal.x, +; EG-NEXT: MULLO_INT * T0.Y, PS, T0.W, +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: LSHL T0.Z, PV.W, literal.x, +; EG-NEXT: SUB_INT T2.W, 0.0, PS, +; EG-NEXT: MULHI * T1.X, T0.X, T0.W, +; EG-NEXT: 16(2.242078e-44), 0(0.000000e+00) +; EG-NEXT: CNDE_INT T2.W, PS, PV.W, T0.Y, +; EG-NEXT: OR_INT * T3.W, T2.X, PV.Z, +; EG-NEXT: SETGT_INT T4.W, 0.0, PS, +; EG-NEXT: MULHI * T0.Y, PV.W, T0.X, +; EG-NEXT: ADD_INT T0.Z, T3.W, PV.W, +; EG-NEXT: ADD_INT T2.W, T0.X, PS, +; EG-NEXT: SUB_INT * T3.W, T0.X, PS, +; EG-NEXT: CNDE_INT T2.W, T1.X, PV.W, PS, +; EG-NEXT: XOR_INT * T3.W, PV.Z, T4.W, +; EG-NEXT: MULHI * T0.X, PV.W, PS, +; EG-NEXT: MULLO_INT * T0.Y, PS, T0.W, +; EG-NEXT: SUB_INT * T2.W, T3.W, PS, +; EG-NEXT: SETGE_UINT T0.W, PV.W, T0.W, +; EG-NEXT: SETGE_UINT * T2.W, T3.W, T0.Y, +; EG-NEXT: AND_INT T0.W, PV.W, PS, +; EG-NEXT: ADD_INT * T3.W, T0.X, 1, +; EG-NEXT: CNDE_INT T0.W, PV.W, T0.X, PS, +; EG-NEXT: ADD_INT * T3.W, T0.X, literal.x, +; EG-NEXT: -1(nan), 0(0.000000e+00) +; EG-NEXT: CNDE_INT T0.W, T2.W, PS, PV.W, +; EG-NEXT: XOR_INT * T1.W, T4.W, T1.W, +; EG-NEXT: XOR_INT * T0.W, PV.W, PS, +; EG-NEXT: SUB_INT * T0.W, PV.W, T1.W, +; EG-NEXT: LSHL * T0.W, PV.W, literal.x, +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: ASHR T0.X, PV.W, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: 8(1.121039e-44), 2(2.802597e-45) %den_ptr = getelementptr i24, i24 addrspace(1)* %in, i24 1 %num = load i24, i24 addrspace(1) * %in %den = load i24, i24 addrspace(1) * %den_ptr @@ -125,9 +2097,220 @@ define amdgpu_kernel void @v_sdiv_i24(i32 addrspace(1)* %out, i24 addrspace(1)* ret void } -; FUNC-LABEL: {{^}}v_sdiv_i25: -; SI-NOT: v_rcp_f32 define amdgpu_kernel void @v_sdiv_i25(i32 addrspace(1)* %out, i25 addrspace(1)* %in) { +; GCN-LABEL: v_sdiv_i25: +; GCN: ; %bb.0: +; GCN-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: s_mov_b32 s10, s6 +; GCN-NEXT: s_mov_b32 s11, s7 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b32 s8, s2 +; GCN-NEXT: s_mov_b32 s9, s3 +; GCN-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: s_mov_b32 s4, s0 +; GCN-NEXT: s_mov_b32 s5, s1 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_bfe_i32 v2, v0, 0, 25 +; GCN-NEXT: v_bfe_i32 v3, v1, 0, 25 +; GCN-NEXT: v_bfe_i32 v0, v0, 24, 1 +; GCN-NEXT: v_bfe_i32 v1, v1, 24, 1 +; GCN-NEXT: v_xor_b32_e32 v4, v0, v1 +; GCN-NEXT: v_add_i32_e32 v2, vcc, v0, v2 +; GCN-NEXT: v_add_i32_e32 v3, vcc, v1, v3 +; GCN-NEXT: v_xor_b32_e32 v0, v2, v0 +; GCN-NEXT: v_xor_b32_e32 v1, v3, v1 +; GCN-NEXT: v_cvt_f32_u32_e32 v2, v1 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v2 +; GCN-NEXT: v_mul_f32_e32 v2, 0x4f800000, v2 +; GCN-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-NEXT: v_mul_hi_u32 v3, v2, v1 +; GCN-NEXT: v_mul_lo_u32 v5, v2, v1 +; GCN-NEXT: v_sub_i32_e32 v6, vcc, 0, v5 +; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v3 +; GCN-NEXT: v_cndmask_b32_e64 v3, v5, v6, s[0:1] +; GCN-NEXT: v_mul_hi_u32 v3, v3, v2 +; GCN-NEXT: v_add_i32_e32 v5, vcc, v3, v2 +; GCN-NEXT: v_subrev_i32_e32 v2, vcc, v3, v2 +; GCN-NEXT: v_cndmask_b32_e64 v2, v2, v5, s[0:1] +; GCN-NEXT: v_mul_hi_u32 v2, v2, v0 +; GCN-NEXT: v_mul_lo_u32 v3, v2, v1 +; GCN-NEXT: v_add_i32_e32 v5, vcc, 1, v2 +; GCN-NEXT: v_add_i32_e32 v6, vcc, -1, v2 +; GCN-NEXT: v_subrev_i32_e32 v7, vcc, v3, v0 +; GCN-NEXT: v_cmp_ge_u32_e32 vcc, v0, v3 +; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], v7, v1 +; GCN-NEXT: s_and_b64 s[0:1], s[0:1], vcc +; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v5, s[0:1] +; GCN-NEXT: v_cndmask_b32_e32 v0, v6, v0, vcc +; GCN-NEXT: v_xor_b32_e32 v0, v0, v4 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, v0, v4 +; GCN-NEXT: v_bfe_i32 v0, v0, 0, 25 +; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; TONGA-LABEL: v_sdiv_i25: +; TONGA: ; %bb.0: +; TONGA-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x24 +; TONGA-NEXT: s_mov_b32 s7, 0xf000 +; TONGA-NEXT: s_mov_b32 s6, -1 +; TONGA-NEXT: s_mov_b32 s2, s6 +; TONGA-NEXT: s_mov_b32 s3, s7 +; TONGA-NEXT: s_waitcnt lgkmcnt(0) +; TONGA-NEXT: s_mov_b32 s0, s10 +; TONGA-NEXT: s_mov_b32 s1, s11 +; TONGA-NEXT: buffer_load_dwordx2 v[0:1], off, s[0:3], 0 +; TONGA-NEXT: s_mov_b32 s4, s8 +; TONGA-NEXT: s_mov_b32 s5, s9 +; TONGA-NEXT: s_waitcnt vmcnt(0) +; TONGA-NEXT: v_bfe_i32 v2, v1, 0, 25 +; TONGA-NEXT: v_bfe_i32 v1, v1, 24, 1 +; TONGA-NEXT: v_add_u32_e32 v2, vcc, v1, v2 +; TONGA-NEXT: v_xor_b32_e32 v2, v2, v1 +; TONGA-NEXT: v_cvt_f32_u32_e32 v3, v2 +; TONGA-NEXT: v_bfe_i32 v4, v0, 0, 25 +; TONGA-NEXT: v_bfe_i32 v0, v0, 24, 1 +; TONGA-NEXT: v_add_u32_e32 v4, vcc, v0, v4 +; TONGA-NEXT: v_rcp_iflag_f32_e32 v3, v3 +; TONGA-NEXT: v_xor_b32_e32 v4, v4, v0 +; TONGA-NEXT: v_xor_b32_e32 v0, v0, v1 +; TONGA-NEXT: v_mul_f32_e32 v3, 0x4f800000, v3 +; TONGA-NEXT: v_cvt_u32_f32_e32 v3, v3 +; TONGA-NEXT: v_mul_lo_u32 v5, v3, v2 +; TONGA-NEXT: v_mul_hi_u32 v6, v3, v2 +; TONGA-NEXT: v_sub_u32_e32 v7, vcc, 0, v5 +; TONGA-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v6 +; TONGA-NEXT: v_cndmask_b32_e64 v5, v5, v7, s[0:1] +; TONGA-NEXT: v_mul_hi_u32 v5, v5, v3 +; TONGA-NEXT: v_add_u32_e32 v6, vcc, v5, v3 +; TONGA-NEXT: v_subrev_u32_e32 v3, vcc, v5, v3 +; TONGA-NEXT: v_cndmask_b32_e64 v3, v3, v6, s[0:1] +; TONGA-NEXT: v_mul_hi_u32 v3, v3, v4 +; TONGA-NEXT: v_mul_lo_u32 v1, v3, v2 +; TONGA-NEXT: v_add_u32_e32 v5, vcc, 1, v3 +; TONGA-NEXT: v_add_u32_e32 v6, vcc, -1, v3 +; TONGA-NEXT: v_subrev_u32_e32 v7, vcc, v1, v4 +; TONGA-NEXT: v_cmp_ge_u32_e32 vcc, v4, v1 +; TONGA-NEXT: v_cmp_ge_u32_e64 s[0:1], v7, v2 +; TONGA-NEXT: s_and_b64 s[0:1], s[0:1], vcc +; TONGA-NEXT: v_cndmask_b32_e64 v1, v3, v5, s[0:1] +; TONGA-NEXT: v_cndmask_b32_e32 v1, v6, v1, vcc +; TONGA-NEXT: v_xor_b32_e32 v1, v1, v0 +; TONGA-NEXT: v_sub_u32_e32 v0, vcc, v1, v0 +; TONGA-NEXT: v_bfe_i32 v0, v0, 0, 25 +; TONGA-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; TONGA-NEXT: s_endpgm +; +; GFX9-LABEL: v_sdiv_i25: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; GFX9-NEXT: s_mov_b32 s7, 0xf000 +; GFX9-NEXT: s_mov_b32 s6, -1 +; GFX9-NEXT: s_mov_b32 s10, s6 +; GFX9-NEXT: s_mov_b32 s11, s7 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s8, s2 +; GFX9-NEXT: s_mov_b32 s9, s3 +; GFX9-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0 +; GFX9-NEXT: s_mov_b32 s4, s0 +; GFX9-NEXT: s_mov_b32 s5, s1 +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_bfe_i32 v2, v1, 0, 25 +; GFX9-NEXT: v_bfe_i32 v1, v1, 24, 1 +; GFX9-NEXT: v_add_u32_e32 v2, v2, v1 +; GFX9-NEXT: v_xor_b32_e32 v2, v2, v1 +; GFX9-NEXT: v_cvt_f32_u32_e32 v3, v2 +; GFX9-NEXT: v_bfe_i32 v6, v0, 0, 25 +; GFX9-NEXT: v_bfe_i32 v0, v0, 24, 1 +; GFX9-NEXT: v_rcp_iflag_f32_e32 v3, v3 +; GFX9-NEXT: v_mul_f32_e32 v3, 0x4f800000, v3 +; GFX9-NEXT: v_cvt_u32_f32_e32 v3, v3 +; GFX9-NEXT: v_mul_lo_u32 v4, v3, v2 +; GFX9-NEXT: v_mul_hi_u32 v5, v3, v2 +; GFX9-NEXT: v_sub_u32_e32 v7, 0, v4 +; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v5 +; GFX9-NEXT: v_cndmask_b32_e32 v4, v4, v7, vcc +; GFX9-NEXT: v_mul_hi_u32 v4, v4, v3 +; GFX9-NEXT: v_add_u32_e32 v5, v6, v0 +; GFX9-NEXT: v_xor_b32_e32 v5, v5, v0 +; GFX9-NEXT: v_xor_b32_e32 v0, v0, v1 +; GFX9-NEXT: v_add_u32_e32 v6, v3, v4 +; GFX9-NEXT: v_sub_u32_e32 v3, v3, v4 +; GFX9-NEXT: v_cndmask_b32_e32 v3, v3, v6, vcc +; GFX9-NEXT: v_mul_hi_u32 v3, v3, v5 +; GFX9-NEXT: v_mul_lo_u32 v4, v3, v2 +; GFX9-NEXT: v_add_u32_e32 v1, 1, v3 +; GFX9-NEXT: v_add_u32_e32 v6, -1, v3 +; GFX9-NEXT: v_sub_u32_e32 v7, v5, v4 +; GFX9-NEXT: v_cmp_ge_u32_e32 vcc, v5, v4 +; GFX9-NEXT: v_cmp_ge_u32_e64 s[0:1], v7, v2 +; GFX9-NEXT: s_and_b64 s[0:1], s[0:1], vcc +; GFX9-NEXT: v_cndmask_b32_e64 v1, v3, v1, s[0:1] +; GFX9-NEXT: v_cndmask_b32_e32 v1, v6, v1, vcc +; GFX9-NEXT: v_xor_b32_e32 v1, v1, v0 +; GFX9-NEXT: v_sub_u32_e32 v0, v1, v0 +; GFX9-NEXT: v_bfe_i32 v0, v0, 0, 25 +; GFX9-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX9-NEXT: s_endpgm +; +; EG-LABEL: v_sdiv_i25: +; EG: ; %bb.0: +; EG-NEXT: ALU 1, @10, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 1 @6 +; EG-NEXT: ALU 41, @12, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 6: +; EG-NEXT: VTX_READ_32 T0.X, T0.X, 4, #1 +; EG-NEXT: VTX_READ_32 T1.X, T1.X, 0, #1 +; EG-NEXT: ALU clause starting at 10: +; EG-NEXT: MOV * T0.X, KC0[2].Z, +; EG-NEXT: MOV * T1.X, PV.X, +; EG-NEXT: ALU clause starting at 12: +; EG-NEXT: LSHL * T0.W, T0.X, literal.x, +; EG-NEXT: 7(9.809089e-45), 0(0.000000e+00) +; EG-NEXT: ASHR * T0.W, PV.W, literal.x, +; EG-NEXT: 7(9.809089e-45), 0(0.000000e+00) +; EG-NEXT: SETGT_INT * T1.W, 0.0, PV.W, +; EG-NEXT: ADD_INT * T0.W, T0.W, PV.W, +; EG-NEXT: XOR_INT * T0.W, PV.W, T1.W, +; EG-NEXT: RECIP_UINT * T0.X, PV.W, +; EG-NEXT: MULLO_INT * T0.Y, PS, T0.W, +; EG-NEXT: LSHL T0.Z, T1.X, literal.x, +; EG-NEXT: SUB_INT T2.W, 0.0, PS, +; EG-NEXT: MULHI * T1.X, T0.X, T0.W, +; EG-NEXT: 7(9.809089e-45), 0(0.000000e+00) +; EG-NEXT: CNDE_INT T2.W, PS, PV.W, T0.Y, +; EG-NEXT: ASHR * T3.W, PV.Z, literal.x, +; EG-NEXT: 7(9.809089e-45), 0(0.000000e+00) +; EG-NEXT: SETGT_INT T4.W, 0.0, PS, +; EG-NEXT: MULHI * T0.Y, PV.W, T0.X, +; EG-NEXT: ADD_INT T0.Z, T3.W, PV.W, +; EG-NEXT: ADD_INT T2.W, T0.X, PS, +; EG-NEXT: SUB_INT * T3.W, T0.X, PS, +; EG-NEXT: CNDE_INT T2.W, T1.X, PV.W, PS, +; EG-NEXT: XOR_INT * T3.W, PV.Z, T4.W, +; EG-NEXT: MULHI * T0.X, PV.W, PS, +; EG-NEXT: MULLO_INT * T0.Y, PS, T0.W, +; EG-NEXT: SUB_INT * T2.W, T3.W, PS, +; EG-NEXT: SETGE_UINT T0.W, PV.W, T0.W, +; EG-NEXT: SETGE_UINT * T2.W, T3.W, T0.Y, +; EG-NEXT: AND_INT T0.W, PV.W, PS, +; EG-NEXT: ADD_INT * T3.W, T0.X, 1, +; EG-NEXT: CNDE_INT T0.W, PV.W, T0.X, PS, +; EG-NEXT: ADD_INT * T3.W, T0.X, literal.x, +; EG-NEXT: -1(nan), 0(0.000000e+00) +; EG-NEXT: CNDE_INT T0.W, T2.W, PS, PV.W, +; EG-NEXT: XOR_INT * T1.W, T4.W, T1.W, +; EG-NEXT: XOR_INT * T0.W, PV.W, PS, +; EG-NEXT: SUB_INT * T0.W, PV.W, T1.W, +; EG-NEXT: LSHL * T0.W, PV.W, literal.x, +; EG-NEXT: 7(9.809089e-45), 0(0.000000e+00) +; EG-NEXT: ASHR T0.X, PV.W, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: 7(9.809089e-45), 2(2.802597e-45) %den_ptr = getelementptr i25, i25 addrspace(1)* %in, i25 1 %num = load i25, i25 addrspace(1) * %in %den = load i25, i25 addrspace(1) * %den_ptr @@ -158,13 +2341,146 @@ define amdgpu_kernel void @v_sdiv_i25(i32 addrspace(1)* %out, i25 addrspace(1)* ; ret void ; } -; FUNC-LABEL: @scalarize_mulhs_4xi32 -; SI: v_mul_hi_i32 -; SI: v_mul_hi_i32 -; SI: v_mul_hi_i32 -; SI: v_mul_hi_i32 - define amdgpu_kernel void @scalarize_mulhs_4xi32(<4 x i32> addrspace(1)* nocapture readonly %in, <4 x i32> addrspace(1)* nocapture %out) { +; GCN-LABEL: scalarize_mulhs_4xi32: +; GCN: ; %bb.0: +; GCN-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b32 s4, s0 +; GCN-NEXT: s_mov_b32 s5, s1 +; GCN-NEXT: buffer_load_dwordx4 v[0:3], off, s[4:7], 0 +; GCN-NEXT: s_mov_b32 s0, 0x1389c755 +; GCN-NEXT: s_mov_b32 s4, s2 +; GCN-NEXT: s_mov_b32 s5, s3 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_mul_hi_i32 v0, v0, s0 +; GCN-NEXT: v_mul_hi_i32 v1, v1, s0 +; GCN-NEXT: v_mul_hi_i32 v2, v2, s0 +; GCN-NEXT: v_mul_hi_i32 v3, v3, s0 +; GCN-NEXT: v_lshrrev_b32_e32 v4, 31, v0 +; GCN-NEXT: v_ashrrev_i32_e32 v0, 12, v0 +; GCN-NEXT: v_lshrrev_b32_e32 v5, 31, v1 +; GCN-NEXT: v_ashrrev_i32_e32 v1, 12, v1 +; GCN-NEXT: v_lshrrev_b32_e32 v6, 31, v2 +; GCN-NEXT: v_ashrrev_i32_e32 v2, 12, v2 +; GCN-NEXT: v_lshrrev_b32_e32 v7, 31, v3 +; GCN-NEXT: v_ashrrev_i32_e32 v3, 12, v3 +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v4 +; GCN-NEXT: v_add_i32_e32 v1, vcc, v1, v5 +; GCN-NEXT: v_add_i32_e32 v2, vcc, v2, v6 +; GCN-NEXT: v_add_i32_e32 v3, vcc, v3, v7 +; GCN-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; TONGA-LABEL: scalarize_mulhs_4xi32: +; TONGA: ; %bb.0: +; TONGA-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; TONGA-NEXT: s_mov_b32 s3, 0xf000 +; TONGA-NEXT: s_mov_b32 s2, -1 +; TONGA-NEXT: s_waitcnt lgkmcnt(0) +; TONGA-NEXT: s_mov_b32 s0, s4 +; TONGA-NEXT: s_mov_b32 s1, s5 +; TONGA-NEXT: buffer_load_dwordx4 v[0:3], off, s[0:3], 0 +; TONGA-NEXT: s_mov_b32 s0, 0x1389c755 +; TONGA-NEXT: s_mov_b32 s4, s6 +; TONGA-NEXT: s_mov_b32 s5, s7 +; TONGA-NEXT: s_mov_b32 s6, s2 +; TONGA-NEXT: s_mov_b32 s7, s3 +; TONGA-NEXT: s_waitcnt vmcnt(0) +; TONGA-NEXT: v_mul_hi_i32 v0, v0, s0 +; TONGA-NEXT: v_mul_hi_i32 v1, v1, s0 +; TONGA-NEXT: v_mul_hi_i32 v2, v2, s0 +; TONGA-NEXT: v_mul_hi_i32 v3, v3, s0 +; TONGA-NEXT: v_lshrrev_b32_e32 v4, 31, v0 +; TONGA-NEXT: v_ashrrev_i32_e32 v0, 12, v0 +; TONGA-NEXT: v_lshrrev_b32_e32 v5, 31, v1 +; TONGA-NEXT: v_ashrrev_i32_e32 v1, 12, v1 +; TONGA-NEXT: v_lshrrev_b32_e32 v6, 31, v2 +; TONGA-NEXT: v_ashrrev_i32_e32 v2, 12, v2 +; TONGA-NEXT: v_lshrrev_b32_e32 v7, 31, v3 +; TONGA-NEXT: v_ashrrev_i32_e32 v3, 12, v3 +; TONGA-NEXT: v_add_u32_e32 v0, vcc, v0, v4 +; TONGA-NEXT: v_add_u32_e32 v1, vcc, v1, v5 +; TONGA-NEXT: v_add_u32_e32 v2, vcc, v2, v6 +; TONGA-NEXT: v_add_u32_e32 v3, vcc, v3, v7 +; TONGA-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 +; TONGA-NEXT: s_endpgm +; +; GFX9-LABEL: scalarize_mulhs_4xi32: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; GFX9-NEXT: s_mov_b32 s3, 0xf000 +; GFX9-NEXT: s_mov_b32 s2, -1 +; GFX9-NEXT: s_waitcnt lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s0, s4 +; GFX9-NEXT: s_mov_b32 s1, s5 +; GFX9-NEXT: buffer_load_dwordx4 v[0:3], off, s[0:3], 0 +; GFX9-NEXT: s_mov_b32 s0, 0x1389c755 +; GFX9-NEXT: s_mov_b32 s4, s6 +; GFX9-NEXT: s_mov_b32 s5, s7 +; GFX9-NEXT: s_mov_b32 s6, s2 +; GFX9-NEXT: s_mov_b32 s7, s3 +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: v_mul_hi_i32 v0, v0, s0 +; GFX9-NEXT: v_mul_hi_i32 v1, v1, s0 +; GFX9-NEXT: v_mul_hi_i32 v2, v2, s0 +; GFX9-NEXT: v_mul_hi_i32 v3, v3, s0 +; GFX9-NEXT: v_lshrrev_b32_e32 v4, 31, v0 +; GFX9-NEXT: v_ashrrev_i32_e32 v0, 12, v0 +; GFX9-NEXT: v_lshrrev_b32_e32 v5, 31, v1 +; GFX9-NEXT: v_ashrrev_i32_e32 v1, 12, v1 +; GFX9-NEXT: v_lshrrev_b32_e32 v6, 31, v2 +; GFX9-NEXT: v_ashrrev_i32_e32 v2, 12, v2 +; GFX9-NEXT: v_lshrrev_b32_e32 v7, 31, v3 +; GFX9-NEXT: v_ashrrev_i32_e32 v3, 12, v3 +; GFX9-NEXT: v_add_u32_e32 v0, v0, v4 +; GFX9-NEXT: v_add_u32_e32 v1, v1, v5 +; GFX9-NEXT: v_add_u32_e32 v2, v2, v6 +; GFX9-NEXT: v_add_u32_e32 v3, v3, v7 +; GFX9-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 +; GFX9-NEXT: s_endpgm +; +; EG-LABEL: scalarize_mulhs_4xi32: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 25, @9, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XYZW, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 6: +; EG-NEXT: VTX_READ_128 T0.XYZW, T0.X, 0, #1 +; EG-NEXT: ALU clause starting at 8: +; EG-NEXT: MOV * T0.X, KC0[2].Y, +; EG-NEXT: ALU clause starting at 9: +; EG-NEXT: MULHI_INT * T0.W, T0.W, literal.x, +; EG-NEXT: 327796565(3.478022e-27), 0(0.000000e+00) +; EG-NEXT: ASHR T1.Z, PS, literal.x, +; EG-NEXT: LSHR T0.W, PS, literal.y, +; EG-NEXT: MULHI_INT * T0.Z, T0.Z, literal.z, +; EG-NEXT: 12(1.681558e-44), 31(4.344025e-44) +; EG-NEXT: 327796565(3.478022e-27), 0(0.000000e+00) +; EG-NEXT: ASHR T1.Y, PS, literal.x, +; EG-NEXT: LSHR T0.Z, PS, literal.y, +; EG-NEXT: ADD_INT T0.W, PV.Z, PV.W, +; EG-NEXT: MULHI_INT * T0.Y, T0.Y, literal.z, +; EG-NEXT: 12(1.681558e-44), 31(4.344025e-44) +; EG-NEXT: 327796565(3.478022e-27), 0(0.000000e+00) +; EG-NEXT: ASHR T2.Y, PS, literal.x, +; EG-NEXT: ADD_INT T0.Z, PV.Y, PV.Z, +; EG-NEXT: LSHR T1.W, PS, literal.y, +; EG-NEXT: MULHI_INT * T0.X, T0.X, literal.z, +; EG-NEXT: 12(1.681558e-44), 31(4.344025e-44) +; EG-NEXT: 327796565(3.478022e-27), 0(0.000000e+00) +; EG-NEXT: ADD_INT T0.Y, PV.Y, PV.W, +; EG-NEXT: ASHR T1.W, PS, literal.x, +; EG-NEXT: LSHR * T2.W, PS, literal.y, +; EG-NEXT: 12(1.681558e-44), 31(4.344025e-44) +; EG-NEXT: ADD_INT T0.X, PV.W, PS, +; EG-NEXT: LSHR * T1.X, KC0[2].Z, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %1 = load <4 x i32>, <4 x i32> addrspace(1)* %in, align 16 %2 = sdiv <4 x i32> %1, store <4 x i32> %2, <4 x i32> addrspace(1)* %out, align 16