diff --git a/llvm/test/CodeGen/AMDGPU/or.ll b/llvm/test/CodeGen/AMDGPU/or.ll --- a/llvm/test/CodeGen/AMDGPU/or.ll +++ b/llvm/test/CodeGen/AMDGPU/or.ll @@ -1,15 +1,64 @@ -; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s -; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %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 UTC_ARGS: --version 3 +; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefix=GFX6 %s +; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GFX8 %s +; RUN: llc -amdgpu-scalarize-global-loads=false -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG %s - -; FUNC-LABEL: {{^}}or_v2i32: -; EG: OR_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}} -; EG: OR_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}} - -; SI: v_or_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}} -; SI: v_or_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}} define amdgpu_kernel void @or_v2i32(ptr addrspace(1) %out, ptr addrspace(1) %in) { +; GFX6-LABEL: or_v2i32: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s7, 0xf000 +; GFX6-NEXT: s_mov_b32 s6, -1 +; GFX6-NEXT: s_mov_b32 s10, s6 +; GFX6-NEXT: s_mov_b32 s11, s7 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s8, s2 +; GFX6-NEXT: s_mov_b32 s9, s3 +; GFX6-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0 +; GFX6-NEXT: s_mov_b32 s4, s0 +; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_or_b32_e32 v1, v1, v3 +; GFX6-NEXT: v_or_b32_e32 v0, v0, v2 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: or_v2i32: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_mov_b32 s6, -1 +; GFX8-NEXT: s_mov_b32 s10, s6 +; GFX8-NEXT: s_mov_b32 s11, s7 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s8, s2 +; GFX8-NEXT: s_mov_b32 s9, s3 +; GFX8-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0 +; GFX8-NEXT: s_mov_b32 s4, s0 +; GFX8-NEXT: s_mov_b32 s5, s1 +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: v_or_b32_e32 v1, v1, v3 +; GFX8-NEXT: v_or_b32_e32 v0, v0, v2 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: or_v2i32: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 3, @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_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: OR_INT * T0.Y, T0.Y, T0.W, +; EG-NEXT: OR_INT T0.X, T0.X, T0.Z, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %b_ptr = getelementptr <2 x i32>, ptr addrspace(1) %in, i32 1 %a = load <2 x i32>, ptr addrspace(1) %in %b = load <2 x i32>, ptr addrspace(1) %b_ptr @@ -18,17 +67,71 @@ ret void } -; FUNC-LABEL: {{^}}or_v4i32: -; EG: OR_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}} -; EG: OR_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}} -; EG: OR_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}} -; EG: OR_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}} - -; SI: v_or_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}} -; SI: v_or_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}} -; SI: v_or_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}} -; SI: v_or_b32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}} define amdgpu_kernel void @or_v4i32(ptr addrspace(1) %out, ptr addrspace(1) %in) { +; GFX6-LABEL: or_v4i32: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s7, 0xf000 +; GFX6-NEXT: s_mov_b32 s6, -1 +; GFX6-NEXT: s_mov_b32 s10, s6 +; GFX6-NEXT: s_mov_b32 s11, s7 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s8, s2 +; GFX6-NEXT: s_mov_b32 s9, s3 +; GFX6-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0 +; GFX6-NEXT: buffer_load_dwordx4 v[4:7], off, s[8:11], 0 offset:16 +; GFX6-NEXT: s_mov_b32 s4, s0 +; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_or_b32_e32 v3, v3, v7 +; GFX6-NEXT: v_or_b32_e32 v2, v2, v6 +; GFX6-NEXT: v_or_b32_e32 v1, v1, v5 +; GFX6-NEXT: v_or_b32_e32 v0, v0, v4 +; GFX6-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: or_v4i32: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_mov_b32 s6, -1 +; GFX8-NEXT: s_mov_b32 s10, s6 +; GFX8-NEXT: s_mov_b32 s11, s7 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s8, s2 +; GFX8-NEXT: s_mov_b32 s9, s3 +; GFX8-NEXT: buffer_load_dwordx4 v[0:3], off, s[8:11], 0 +; GFX8-NEXT: buffer_load_dwordx4 v[4:7], off, s[8:11], 0 offset:16 +; GFX8-NEXT: s_mov_b32 s4, s0 +; GFX8-NEXT: s_mov_b32 s5, s1 +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: v_or_b32_e32 v3, v3, v7 +; GFX8-NEXT: v_or_b32_e32 v2, v2, v6 +; GFX8-NEXT: v_or_b32_e32 v1, v1, v5 +; GFX8-NEXT: v_or_b32_e32 v0, v0, v4 +; GFX8-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: or_v4i32: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @10, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 1 @6 +; EG-NEXT: ALU 5, @11, 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 T1.XYZW, T0.X, 16, #1 +; EG-NEXT: VTX_READ_128 T0.XYZW, 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: OR_INT * T0.W, T0.W, T1.W, +; EG-NEXT: OR_INT * T0.Z, T0.Z, T1.Z, +; EG-NEXT: OR_INT * T0.Y, T0.Y, T1.Y, +; EG-NEXT: OR_INT T0.X, T0.X, T1.X, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %b_ptr = getelementptr <4 x i32>, ptr addrspace(1) %in, i32 1 %a = load <4 x i32>, ptr addrspace(1) %in %b = load <4 x i32>, ptr addrspace(1) %b_ptr @@ -37,52 +140,263 @@ ret void } -; FUNC-LABEL: {{^}}scalar_or_i32: -; SI: s_or_b32 define amdgpu_kernel void @scalar_or_i32(ptr addrspace(1) %out, i32 %a, i32 %b) { +; GFX6-LABEL: scalar_or_i32: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s7, 0xf000 +; GFX6-NEXT: s_mov_b32 s6, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s4, s0 +; GFX6-NEXT: s_or_b32 s0, s2, s3 +; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: v_mov_b32_e32 v0, s0 +; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: scalar_or_i32: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_mov_b32 s6, -1 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s4, s0 +; GFX8-NEXT: s_or_b32 s0, s2, s3 +; GFX8-NEXT: s_mov_b32 s5, s1 +; GFX8-NEXT: v_mov_b32_e32 v0, s0 +; GFX8-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: scalar_or_i32: +; EG: ; %bb.0: +; EG-NEXT: ALU 2, @4, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T1.X, T0.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: ALU clause starting at 4: +; EG-NEXT: LSHR T0.X, KC0[2].Y, literal.x, +; EG-NEXT: OR_INT * T1.X, KC0[2].Z, KC0[2].W, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %or = or i32 %a, %b store i32 %or, ptr addrspace(1) %out ret void } -; FUNC-LABEL: {{^}}vector_or_i32: -; SI: v_or_b32_e32 v{{[0-9]}} define amdgpu_kernel void @vector_or_i32(ptr addrspace(1) %out, ptr addrspace(1) %a, i32 %b) { +; GFX6-LABEL: vector_or_i32: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GFX6-NEXT: s_load_dword s12, s[0:1], 0xd +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_mov_b32 s10, s2 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s8, s6 +; GFX6-NEXT: s_mov_b32 s9, s7 +; GFX6-NEXT: s_mov_b32 s11, s3 +; GFX6-NEXT: buffer_load_dword v0, off, s[8:11], 0 +; GFX6-NEXT: s_mov_b32 s0, s4 +; GFX6-NEXT: s_mov_b32 s1, s5 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_or_b32_e32 v0, s12, v0 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: vector_or_i32: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; GFX8-NEXT: s_load_dword s12, s[0:1], 0x34 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: s_mov_b32 s10, s2 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s8, s6 +; GFX8-NEXT: s_mov_b32 s9, s7 +; GFX8-NEXT: s_mov_b32 s11, s3 +; GFX8-NEXT: buffer_load_dword v0, off, s[8:11], 0 +; GFX8-NEXT: s_mov_b32 s0, s4 +; GFX8-NEXT: s_mov_b32 s1, s5 +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: v_or_b32_e32 v0, s12, v0 +; GFX8-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: vector_or_i32: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 2, @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: OR_INT T0.X, T0.X, KC0[2].W, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %loada = load i32, ptr addrspace(1) %a %or = or i32 %loada, %b store i32 %or, ptr addrspace(1) %out ret void } -; FUNC-LABEL: {{^}}scalar_or_literal_i32: -; SI: s_or_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x1869f define amdgpu_kernel void @scalar_or_literal_i32(ptr addrspace(1) %out, i32 %a) { +; GFX6-LABEL: scalar_or_literal_i32: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dword s4, s[0:1], 0xb +; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_or_b32 s4, s4, 0x1869f +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: scalar_or_literal_i32: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dword s4, s[0:1], 0x2c +; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_or_b32 s4, s4, 0x1869f +; GFX8-NEXT: v_mov_b32_e32 v0, s4 +; GFX8-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: scalar_or_literal_i32: +; EG: ; %bb.0: +; EG-NEXT: ALU 2, @4, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T1.X, T0.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: ALU clause starting at 4: +; EG-NEXT: LSHR T0.X, KC0[2].Y, literal.x, +; EG-NEXT: OR_INT * T1.X, KC0[2].Z, literal.y, +; EG-NEXT: 2(2.802597e-45), 99999(1.401284e-40) %or = or i32 %a, 99999 store i32 %or, ptr addrspace(1) %out, align 4 ret void } -; FUNC-LABEL: {{^}}scalar_or_literal_i64: -; SI: s_load_dwordx2 s[[[LO:[0-9]+]]:[[HI:[0-9]+]]], s{{\[[0-9]+:[0-9]+\]}}, {{0x13|0x4c}} -; SI-DAG: s_or_b32 s[[RES_HI:[0-9]+]], s[[HI]], 0xf237b -; SI-DAG: s_or_b32 s[[RES_LO:[0-9]+]], s[[LO]], 0x3039 -; SI-DAG: v_mov_b32_e32 v{{[0-9]+}}, s[[RES_LO]] -; SI-DAG: v_mov_b32_e32 v{{[0-9]+}}, s[[RES_HI]] define amdgpu_kernel void @scalar_or_literal_i64(ptr addrspace(1) %out, [8 x i32], i64 %a) { +; GFX6-LABEL: scalar_or_literal_i64: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x13 +; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_or_b32 s5, s5, 0xf237b +; GFX6-NEXT: s_or_b32 s4, s4, 0x3039 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: scalar_or_literal_i64: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x4c +; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_or_b32 s5, s5, 0xf237b +; GFX8-NEXT: s_or_b32 s4, s4, 0x3039 +; GFX8-NEXT: v_mov_b32_e32 v0, s4 +; GFX8-NEXT: v_mov_b32_e32 v1, s5 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: scalar_or_literal_i64: +; EG: ; %bb.0: +; EG-NEXT: ALU 4, @4, 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: ALU clause starting at 4: +; EG-NEXT: OR_INT * T0.Y, KC0[5].X, literal.x, +; EG-NEXT: 992123(1.390260e-39), 0(0.000000e+00) +; EG-NEXT: OR_INT T0.X, KC0[4].W, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: 12345(1.729903e-41), 2(2.802597e-45) %or = or i64 %a, 4261135838621753 store i64 %or, ptr addrspace(1) %out ret void } -; FUNC-LABEL: {{^}}scalar_or_literal_multi_use_i64: -; SI: s_load_dwordx2 s[[[LO:[0-9]+]]:[[HI:[0-9]+]]], s{{\[[0-9]+:[0-9]+\]}}, {{0x13|0x4c}} -; SI-DAG: s_mov_b32 s[[K_HI:[0-9]+]], 0xf237b -; SI-DAG: s_movk_i32 s[[K_LO:[0-9]+]], 0x3039 -; SI: s_or_b64 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, s[[[K_LO]]:[[K_HI]]] - -; SI: s_add_u32 s{{[0-9]+}}, s{{[0-9]+}}, 0x3039 -; SI: s_addc_u32 s{{[0-9]+}}, s{{[0-9]+}}, 0xf237b define amdgpu_kernel void @scalar_or_literal_multi_use_i64(ptr addrspace(1) %out, [8 x i32], i64 %a, [8 x i32], i64 %b) { +; GFX6-LABEL: scalar_or_literal_multi_use_i64: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9 +; GFX6-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x13 +; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x1d +; GFX6-NEXT: s_movk_i32 s8, 0x3039 +; GFX6-NEXT: s_mov_b32 s9, 0xf237b +; GFX6-NEXT: s_mov_b32 s7, 0xf000 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_or_b64 s[2:3], s[2:3], s[8:9] +; GFX6-NEXT: v_mov_b32_e32 v0, s2 +; GFX6-NEXT: s_mov_b32 s6, -1 +; GFX6-NEXT: v_mov_b32_e32 v1, s3 +; GFX6-NEXT: s_add_u32 s0, s0, 0x3039 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: s_addc_u32 s1, s1, 0xf237b +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s0 +; GFX6-NEXT: v_mov_b32_e32 v1, s1 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: scalar_or_literal_multi_use_i64: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x24 +; GFX8-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x4c +; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x74 +; GFX8-NEXT: s_movk_i32 s8, 0x3039 +; GFX8-NEXT: s_mov_b32 s9, 0xf237b +; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_or_b64 s[2:3], s[2:3], s[8:9] +; GFX8-NEXT: v_mov_b32_e32 v0, s2 +; GFX8-NEXT: s_mov_b32 s6, -1 +; GFX8-NEXT: v_mov_b32_e32 v1, s3 +; GFX8-NEXT: s_add_u32 s0, s0, 0x3039 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX8-NEXT: s_addc_u32 s1, s1, 0xf237b +; GFX8-NEXT: v_mov_b32_e32 v0, s0 +; GFX8-NEXT: v_mov_b32_e32 v1, s1 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: scalar_or_literal_multi_use_i64: +; EG: ; %bb.0: +; EG-NEXT: ALU 12, @6, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T3.XY, T4.X, 0 +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T1.X, T2.X, 0 +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T2.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: ALU clause starting at 6: +; EG-NEXT: ADDC_UINT * T0.W, KC0[7].Y, literal.x, +; EG-NEXT: 12345(1.729903e-41), 0(0.000000e+00) +; EG-NEXT: ADD_INT T0.X, KC0[7].Y, literal.x, +; EG-NEXT: ADD_INT * T0.W, KC0[7].Z, PV.W, +; EG-NEXT: 12345(1.729903e-41), 0(0.000000e+00) +; EG-NEXT: ADD_INT T1.X, PV.W, literal.x, +; EG-NEXT: MOV * T2.X, literal.y, +; EG-NEXT: 992123(1.390260e-39), 0(0.000000e+00) +; EG-NEXT: OR_INT * T3.Y, KC0[5].X, literal.x, +; EG-NEXT: 992123(1.390260e-39), 0(0.000000e+00) +; EG-NEXT: OR_INT T3.X, KC0[4].W, literal.x, +; EG-NEXT: LSHR * T4.X, KC0[2].Y, literal.y, +; EG-NEXT: 12345(1.729903e-41), 2(2.802597e-45) %or = or i64 %a, 4261135838621753 store i64 %or, ptr addrspace(1) %out @@ -91,27 +405,112 @@ ret void } -; FUNC-LABEL: {{^}}scalar_or_inline_imm_i64: -; SI: s_load_dwordx2 s[[[VAL_LO:[0-9]+]]:[[VAL_HI:[0-9]+]]], s{{\[[0-9]+:[0-9]+\]}}, {{0x13|0x4c}} -; SI-NOT: or_b32 -; SI: s_or_b32 s[[VAL_LO]], s[[VAL_LO]], 63 -; SI-NOT: or_b32 -; SI: v_mov_b32_e32 v[[VLO:[0-9]+]], s[[VAL_LO]] -; SI-NOT: or_b32 -; SI: v_mov_b32_e32 v[[VHI:[0-9]+]], s[[VAL_HI]] -; SI-NOT: or_b32 -; SI: buffer_store_dwordx2 v[[[VLO]]:[[VHI]]] define amdgpu_kernel void @scalar_or_inline_imm_i64(ptr addrspace(1) %out, [8 x i32], i64 %a) { +; GFX6-LABEL: scalar_or_inline_imm_i64: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x13 +; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_or_b32 s4, s4, 63 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: scalar_or_inline_imm_i64: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x4c +; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_or_b32 s4, s4, 63 +; GFX8-NEXT: v_mov_b32_e32 v0, s4 +; GFX8-NEXT: v_mov_b32_e32 v1, s5 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: scalar_or_inline_imm_i64: +; EG: ; %bb.0: +; EG-NEXT: ALU 3, @4, 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: ALU clause starting at 4: +; EG-NEXT: MOV * T0.Y, KC0[5].X, +; EG-NEXT: OR_INT T0.X, KC0[4].W, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: 63(8.828180e-44), 2(2.802597e-45) %or = or i64 %a, 63 store i64 %or, ptr addrspace(1) %out ret void } -; FUNC-LABEL: {{^}}scalar_or_inline_imm_multi_use_i64: -; SI-NOT: or_b32 -; SI: s_or_b32 s{{[0-9]+}}, s{{[0-9]+}}, 63 -; SI-NOT: or_b32 define amdgpu_kernel void @scalar_or_inline_imm_multi_use_i64(ptr addrspace(1) %out, i64 %a, i64 %b) { +; GFX6-LABEL: scalar_or_inline_imm_multi_use_i64: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GFX6-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0xd +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s0, s4 +; GFX6-NEXT: s_or_b32 s4, s6, 63 +; GFX6-NEXT: s_mov_b32 s1, s5 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: v_mov_b32_e32 v1, s7 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX6-NEXT: s_add_u32 s0, s8, 63 +; GFX6-NEXT: s_addc_u32 s1, s9, 0 +; GFX6-NEXT: s_waitcnt expcnt(0) +; GFX6-NEXT: v_mov_b32_e32 v0, s0 +; GFX6-NEXT: v_mov_b32_e32 v1, s1 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: scalar_or_inline_imm_multi_use_i64: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; GFX8-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0x34 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s0, s4 +; GFX8-NEXT: s_or_b32 s4, s6, 63 +; GFX8-NEXT: s_mov_b32 s1, s5 +; GFX8-NEXT: v_mov_b32_e32 v0, s4 +; GFX8-NEXT: v_mov_b32_e32 v1, s7 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX8-NEXT: s_add_u32 s0, s8, 63 +; GFX8-NEXT: s_addc_u32 s1, s9, 0 +; GFX8-NEXT: v_mov_b32_e32 v0, s0 +; GFX8-NEXT: v_mov_b32_e32 v1, s1 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: scalar_or_inline_imm_multi_use_i64: +; EG: ; %bb.0: +; EG-NEXT: ALU 9, @6, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T3.XY, T4.X, 0 +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T1.X, T2.X, 0 +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T2.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: ALU clause starting at 6: +; EG-NEXT: ADD_INT T0.X, KC0[3].Y, literal.x, +; EG-NEXT: ADDC_UINT * T0.W, KC0[3].Y, literal.x, +; EG-NEXT: 63(8.828180e-44), 0(0.000000e+00) +; EG-NEXT: ADD_INT T1.X, KC0[3].Z, PV.W, +; EG-NEXT: MOV * T2.X, literal.x, +; EG-NEXT: 0(0.000000e+00), 0(0.000000e+00) +; EG-NEXT: MOV * T3.Y, KC0[3].X, +; EG-NEXT: OR_INT T3.X, KC0[2].W, literal.x, +; EG-NEXT: LSHR * T4.X, KC0[2].Y, literal.y, +; EG-NEXT: 63(8.828180e-44), 2(2.802597e-45) %or = or i64 %a, 63 store i64 %or, ptr addrspace(1) %out %foo = add i64 %b, 63 @@ -119,51 +518,281 @@ ret void } -; FUNC-LABEL: {{^}}scalar_or_neg_inline_imm_i64: -; SI-DAG: s_load_dword [[VAL:s[0-9]+]] -; SI-DAG: s_or_b32 [[VAL]], [[VAL]], -8 -; SI-DAG: v_mov_b32_e32 v[[V_HI:[0-9]+]], -1{{$}} -; SI-DAG: v_mov_b32_e32 v[[V_LO:[0-9]+]], [[VAL]] -; SI: buffer_store_dwordx2 v[[[V_LO]]:[[V_HI]]] define amdgpu_kernel void @scalar_or_neg_inline_imm_i64(ptr addrspace(1) %out, [8 x i32], i64 %a) { +; GFX6-LABEL: scalar_or_neg_inline_imm_i64: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dword s4, s[0:1], 0x13 +; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_mov_b32_e32 v1, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_or_b32 s4, s4, -8 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: scalar_or_neg_inline_imm_i64: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dword s4, s[0:1], 0x4c +; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: v_mov_b32_e32 v1, -1 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_or_b32 s4, s4, -8 +; GFX8-NEXT: v_mov_b32_e32 v0, s4 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: scalar_or_neg_inline_imm_i64: +; EG: ; %bb.0: +; EG-NEXT: ALU 4, @4, 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: ALU clause starting at 4: +; EG-NEXT: OR_INT T0.X, KC0[4].W, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: -8(nan), 2(2.802597e-45) +; EG-NEXT: MOV * T0.Y, literal.x, +; EG-NEXT: -1(nan), 0(0.000000e+00) %or = or i64 %a, -8 store i64 %or, ptr addrspace(1) %out ret void } -; FUNC-LABEL: {{^}}vector_or_literal_i32: -; SI: v_or_b32_e32 v{{[0-9]+}}, 0xffff, v{{[0-9]+}} define amdgpu_kernel void @vector_or_literal_i32(ptr addrspace(1) %out, ptr addrspace(1) %a, ptr addrspace(1) %b) { +; GFX6-LABEL: vector_or_literal_i32: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s7, 0xf000 +; GFX6-NEXT: s_mov_b32 s6, -1 +; GFX6-NEXT: s_mov_b32 s10, s6 +; GFX6-NEXT: s_mov_b32 s11, s7 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s8, s2 +; GFX6-NEXT: s_mov_b32 s9, s3 +; GFX6-NEXT: buffer_load_dword v0, off, s[8:11], 0 +; GFX6-NEXT: s_mov_b32 s4, s0 +; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_or_b32_e32 v0, 0xffff, v0 +; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: vector_or_literal_i32: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_mov_b32 s6, -1 +; GFX8-NEXT: s_mov_b32 s10, s6 +; GFX8-NEXT: s_mov_b32 s11, s7 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s8, s2 +; GFX8-NEXT: s_mov_b32 s9, s3 +; GFX8-NEXT: buffer_load_dword v0, off, s[8:11], 0 +; GFX8-NEXT: s_mov_b32 s4, s0 +; GFX8-NEXT: s_mov_b32 s5, s1 +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: v_or_b32_e32 v0, 0xffff, v0 +; GFX8-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: vector_or_literal_i32: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 2, @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: OR_INT T0.X, T0.X, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: 65535(9.183409e-41), 2(2.802597e-45) %loada = load i32, ptr addrspace(1) %a, align 4 %or = or i32 %loada, 65535 store i32 %or, ptr addrspace(1) %out, align 4 ret void } -; FUNC-LABEL: {{^}}vector_or_inline_immediate_i32: -; SI: v_or_b32_e32 v{{[0-9]+}}, 4, v{{[0-9]+}} define amdgpu_kernel void @vector_or_inline_immediate_i32(ptr addrspace(1) %out, ptr addrspace(1) %a, ptr addrspace(1) %b) { +; GFX6-LABEL: vector_or_inline_immediate_i32: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s7, 0xf000 +; GFX6-NEXT: s_mov_b32 s6, -1 +; GFX6-NEXT: s_mov_b32 s10, s6 +; GFX6-NEXT: s_mov_b32 s11, s7 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s8, s2 +; GFX6-NEXT: s_mov_b32 s9, s3 +; GFX6-NEXT: buffer_load_dword v0, off, s[8:11], 0 +; GFX6-NEXT: s_mov_b32 s4, s0 +; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_or_b32_e32 v0, 4, v0 +; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: vector_or_inline_immediate_i32: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_mov_b32 s6, -1 +; GFX8-NEXT: s_mov_b32 s10, s6 +; GFX8-NEXT: s_mov_b32 s11, s7 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s8, s2 +; GFX8-NEXT: s_mov_b32 s9, s3 +; GFX8-NEXT: buffer_load_dword v0, off, s[8:11], 0 +; GFX8-NEXT: s_mov_b32 s4, s0 +; GFX8-NEXT: s_mov_b32 s5, s1 +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: v_or_b32_e32 v0, 4, v0 +; GFX8-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: vector_or_inline_immediate_i32: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 2, @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: OR_INT T0.X, T0.X, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: 4(5.605194e-45), 2(2.802597e-45) %loada = load i32, ptr addrspace(1) %a, align 4 %or = or i32 %loada, 4 store i32 %or, ptr addrspace(1) %out, align 4 ret void } -; FUNC-LABEL: {{^}}scalar_or_i64: -; EG-DAG: OR_INT * T{{[0-9]\.[XYZW]}}, KC0[2].W, KC0[3].Y -; EG-DAG: OR_INT * T{{[0-9]\.[XYZW]}}, KC0[3].X, KC0[3].Z - -; SI: s_or_b64 define amdgpu_kernel void @scalar_or_i64(ptr addrspace(1) %out, i64 %a, i64 %b) { +; GFX6-LABEL: scalar_or_i64: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GFX6-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0xd +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s0, s4 +; GFX6-NEXT: s_mov_b32 s1, s5 +; GFX6-NEXT: s_or_b64 s[4:5], s[6:7], s[8:9] +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: scalar_or_i64: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; GFX8-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0x34 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s0, s4 +; GFX8-NEXT: s_mov_b32 s1, s5 +; GFX8-NEXT: s_or_b64 s[4:5], s[6:7], s[8:9] +; GFX8-NEXT: v_mov_b32_e32 v0, s4 +; GFX8-NEXT: v_mov_b32_e32 v1, s5 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: scalar_or_i64: +; EG: ; %bb.0: +; EG-NEXT: ALU 3, @4, 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: ALU clause starting at 4: +; EG-NEXT: OR_INT * T0.Y, KC0[3].X, KC0[3].Z, +; EG-NEXT: OR_INT * T0.X, KC0[2].W, KC0[3].Y, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %or = or i64 %a, %b store i64 %or, ptr addrspace(1) %out ret void } -; FUNC-LABEL: {{^}}vector_or_i64: -; SI: v_or_b32_e32 v{{[0-9]}} -; SI: v_or_b32_e32 v{{[0-9]}} define amdgpu_kernel void @vector_or_i64(ptr addrspace(1) %out, ptr addrspace(1) %a, ptr addrspace(1) %b) { +; GFX6-LABEL: vector_or_i64: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GFX6-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0xd +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_mov_b32 s10, s2 +; GFX6-NEXT: s_mov_b32 s11, s3 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s12, s6 +; GFX6-NEXT: s_mov_b32 s13, s7 +; GFX6-NEXT: s_mov_b32 s14, s2 +; GFX6-NEXT: s_mov_b32 s15, s3 +; GFX6-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0 +; GFX6-NEXT: buffer_load_dwordx2 v[2:3], off, s[12:15], 0 +; GFX6-NEXT: s_mov_b32 s0, s4 +; GFX6-NEXT: s_mov_b32 s1, s5 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_or_b32_e32 v0, v2, v0 +; GFX6-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: vector_or_i64: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; GFX8-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0x34 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: s_mov_b32 s10, s2 +; GFX8-NEXT: s_mov_b32 s11, s3 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s12, s6 +; GFX8-NEXT: s_mov_b32 s13, s7 +; GFX8-NEXT: s_mov_b32 s14, s2 +; GFX8-NEXT: s_mov_b32 s15, s3 +; GFX8-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0 +; GFX8-NEXT: buffer_load_dwordx2 v[2:3], off, s[12:15], 0 +; GFX8-NEXT: s_mov_b32 s0, s4 +; GFX8-NEXT: s_mov_b32 s1, s5 +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: v_or_b32_e32 v0, v2, v0 +; GFX8-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: vector_or_i64: +; EG: ; %bb.0: +; EG-NEXT: ALU 1, @10, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 1 @6 +; EG-NEXT: ALU 3, @12, 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, T1.X, 0, #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: MOV * T1.X, KC0[2].W, +; EG-NEXT: ALU clause starting at 12: +; EG-NEXT: OR_INT * T0.Y, T0.Y, T1.Y, +; EG-NEXT: OR_INT T0.X, T0.X, T1.X, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %loada = load i64, ptr addrspace(1) %a, align 8 %loadb = load i64, ptr addrspace(1) %b, align 8 %or = or i64 %loada, %loadb @@ -171,22 +800,127 @@ ret void } -; FUNC-LABEL: {{^}}scalar_vector_or_i64: -; SI: v_or_b32_e32 v{{[0-9]}} -; SI: v_or_b32_e32 v{{[0-9]}} define amdgpu_kernel void @scalar_vector_or_i64(ptr addrspace(1) %out, ptr addrspace(1) %a, i64 %b) { +; GFX6-LABEL: scalar_vector_or_i64: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GFX6-NEXT: s_load_dwordx2 s[12:13], s[0:1], 0xd +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_mov_b32 s10, s2 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s8, s6 +; GFX6-NEXT: s_mov_b32 s9, s7 +; GFX6-NEXT: s_mov_b32 s11, s3 +; GFX6-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0 +; GFX6-NEXT: s_mov_b32 s0, s4 +; GFX6-NEXT: s_mov_b32 s1, s5 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_or_b32_e32 v0, s12, v0 +; GFX6-NEXT: v_or_b32_e32 v1, s13, v1 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: scalar_vector_or_i64: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; GFX8-NEXT: s_load_dwordx2 s[12:13], s[0:1], 0x34 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: s_mov_b32 s10, s2 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s8, s6 +; GFX8-NEXT: s_mov_b32 s9, s7 +; GFX8-NEXT: s_mov_b32 s11, s3 +; GFX8-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0 +; GFX8-NEXT: s_mov_b32 s0, s4 +; GFX8-NEXT: s_mov_b32 s1, s5 +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: v_or_b32_e32 v0, s12, v0 +; GFX8-NEXT: v_or_b32_e32 v1, s13, v1 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: scalar_vector_or_i64: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 3, @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: OR_INT * T0.Y, T0.Y, KC0[3].X, +; EG-NEXT: OR_INT T0.X, T0.X, KC0[2].W, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %loada = load i64, ptr addrspace(1) %a %or = or i64 %loada, %b store i64 %or, ptr addrspace(1) %out ret void } -; FUNC-LABEL: {{^}}vector_or_i64_loadimm: -; SI-DAG: buffer_load_dwordx2 v[[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]], -; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, 0xdf77987f, v[[LO_VREG]] -; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, 0x146f, v[[HI_VREG]] -; SI: s_endpgm define amdgpu_kernel void @vector_or_i64_loadimm(ptr addrspace(1) %out, ptr addrspace(1) %a, ptr addrspace(1) %b) { +; GFX6-LABEL: vector_or_i64_loadimm: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s7, 0xf000 +; GFX6-NEXT: s_mov_b32 s6, -1 +; GFX6-NEXT: s_mov_b32 s10, s6 +; GFX6-NEXT: s_mov_b32 s11, s7 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s8, s2 +; GFX6-NEXT: s_mov_b32 s9, s3 +; GFX6-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0 +; GFX6-NEXT: s_mov_b32 s4, s0 +; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_or_b32_e32 v1, 0x146f, v1 +; GFX6-NEXT: v_or_b32_e32 v0, 0xdf77987f, v0 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: vector_or_i64_loadimm: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_mov_b32 s6, -1 +; GFX8-NEXT: s_mov_b32 s10, s6 +; GFX8-NEXT: s_mov_b32 s11, s7 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s8, s2 +; GFX8-NEXT: s_mov_b32 s9, s3 +; GFX8-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0 +; GFX8-NEXT: s_mov_b32 s4, s0 +; GFX8-NEXT: s_mov_b32 s5, s1 +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: v_or_b32_e32 v1, 0x146f, v1 +; GFX8-NEXT: v_or_b32_e32 v0, 0xdf77987f, v0 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: vector_or_i64_loadimm: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 4, @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: OR_INT * T0.Y, T0.Y, literal.x, +; EG-NEXT: 5231(7.330192e-42), 0(0.000000e+00) +; EG-NEXT: OR_INT T0.X, T0.X, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: -545810305(-1.784115e+19), 2(2.802597e-45) %loada = load i64, ptr addrspace(1) %a, align 8 %or = or i64 %loada, 22470723082367 store i64 %or, ptr addrspace(1) %out @@ -194,63 +928,312 @@ } ; FIXME: The or 0 should really be removed. -; FUNC-LABEL: {{^}}vector_or_i64_imm: -; SI: buffer_load_dwordx2 v[[[LO_VREG:[0-9]+]]:[[HI_VREG:[0-9]+]]], -; SI: v_or_b32_e32 v[[LO_RESULT:[0-9]+]], 8, v[[LO_VREG]] -; SI-NOT: v_or_b32_e32 {{v[0-9]+}}, 0 -; SI: buffer_store_dwordx2 v[[[LO_RESULT]]:[[HI_VREG]]] -; SI: s_endpgm define amdgpu_kernel void @vector_or_i64_imm(ptr addrspace(1) %out, ptr addrspace(1) %a, ptr addrspace(1) %b) { +; GFX6-LABEL: vector_or_i64_imm: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s7, 0xf000 +; GFX6-NEXT: s_mov_b32 s6, -1 +; GFX6-NEXT: s_mov_b32 s10, s6 +; GFX6-NEXT: s_mov_b32 s11, s7 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s8, s2 +; GFX6-NEXT: s_mov_b32 s9, s3 +; GFX6-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0 +; GFX6-NEXT: s_mov_b32 s4, s0 +; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_or_b32_e32 v0, 8, v0 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: vector_or_i64_imm: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_mov_b32 s6, -1 +; GFX8-NEXT: s_mov_b32 s10, s6 +; GFX8-NEXT: s_mov_b32 s11, s7 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s8, s2 +; GFX8-NEXT: s_mov_b32 s9, s3 +; GFX8-NEXT: buffer_load_dwordx2 v[0:1], off, s[8:11], 0 +; GFX8-NEXT: s_mov_b32 s4, s0 +; GFX8-NEXT: s_mov_b32 s5, s1 +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: v_or_b32_e32 v0, 8, v0 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: vector_or_i64_imm: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 2, @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: OR_INT T0.X, T0.X, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: 8(1.121039e-44), 2(2.802597e-45) %loada = load i64, ptr addrspace(1) %a, align 8 %or = or i64 %loada, 8 store i64 %or, ptr addrspace(1) %out ret void } -; FUNC-LABEL: {{^}}vector_or_i64_neg_inline_imm: -; SI-DAG: buffer_load_dword v[[LO_VREG:[0-9]+]] -; SI-DAG: v_or_b32_e32 v[[RES_LO:[0-9]+]], -8, v[[LO_VREG]] -; SI-DAG: v_mov_b32_e32 v[[RES_HI:[0-9]+]], -1{{$}} -; SI: buffer_store_dwordx2 v[[[RES_LO]]:[[RES_HI]]] -; SI: s_endpgm define amdgpu_kernel void @vector_or_i64_neg_inline_imm(ptr addrspace(1) %out, ptr addrspace(1) %a, ptr addrspace(1) %b) { +; GFX6-LABEL: vector_or_i64_neg_inline_imm: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s7, 0xf000 +; GFX6-NEXT: s_mov_b32 s6, -1 +; GFX6-NEXT: s_mov_b32 s10, s6 +; GFX6-NEXT: s_mov_b32 s11, s7 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s8, s2 +; GFX6-NEXT: s_mov_b32 s9, s3 +; GFX6-NEXT: buffer_load_dword v0, off, s[8:11], 0 +; GFX6-NEXT: s_mov_b32 s4, s0 +; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: v_mov_b32_e32 v1, -1 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_or_b32_e32 v0, -8, v0 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: vector_or_i64_neg_inline_imm: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_mov_b32 s6, -1 +; GFX8-NEXT: s_mov_b32 s10, s6 +; GFX8-NEXT: s_mov_b32 s11, s7 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s8, s2 +; GFX8-NEXT: s_mov_b32 s9, s3 +; GFX8-NEXT: buffer_load_dword v0, off, s[8:11], 0 +; GFX8-NEXT: s_mov_b32 s4, s0 +; GFX8-NEXT: s_mov_b32 s5, s1 +; GFX8-NEXT: v_mov_b32_e32 v1, -1 +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: v_or_b32_e32 v0, -8, v0 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: vector_or_i64_neg_inline_imm: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 4, @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_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: OR_INT T0.X, T0.X, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: -8(nan), 2(2.802597e-45) +; EG-NEXT: MOV * T0.Y, literal.x, +; EG-NEXT: -1(nan), 0(0.000000e+00) %loada = load i64, ptr addrspace(1) %a, align 8 %or = or i64 %loada, -8 store i64 %or, ptr addrspace(1) %out ret void } -; FUNC-LABEL: {{^}}vector_or_i64_neg_literal: -; SI-DAG: buffer_load_dword v[[LO_VREG:[0-9]+]] -; SI-DAG: v_mov_b32_e32 v{{[0-9]+}}, -1{{$}} -; SI-DAG: v_or_b32_e32 {{v[0-9]+}}, 0xffffff38, v[[LO_VREG]] -; SI: buffer_store_dwordx2 -; SI: s_endpgm define amdgpu_kernel void @vector_or_i64_neg_literal(ptr addrspace(1) %out, ptr addrspace(1) %a, ptr addrspace(1) %b) { +; GFX6-LABEL: vector_or_i64_neg_literal: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s7, 0xf000 +; GFX6-NEXT: s_mov_b32 s6, -1 +; GFX6-NEXT: s_mov_b32 s10, s6 +; GFX6-NEXT: s_mov_b32 s11, s7 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s8, s2 +; GFX6-NEXT: s_mov_b32 s9, s3 +; GFX6-NEXT: buffer_load_dword v0, off, s[8:11], 0 +; GFX6-NEXT: s_mov_b32 s4, s0 +; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: v_mov_b32_e32 v1, -1 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_or_b32_e32 v0, 0xffffff38, v0 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: vector_or_i64_neg_literal: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_mov_b32 s6, -1 +; GFX8-NEXT: s_mov_b32 s10, s6 +; GFX8-NEXT: s_mov_b32 s11, s7 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s8, s2 +; GFX8-NEXT: s_mov_b32 s9, s3 +; GFX8-NEXT: buffer_load_dword v0, off, s[8:11], 0 +; GFX8-NEXT: s_mov_b32 s4, s0 +; GFX8-NEXT: s_mov_b32 s5, s1 +; GFX8-NEXT: v_mov_b32_e32 v1, -1 +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: v_or_b32_e32 v0, 0xffffff38, v0 +; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: vector_or_i64_neg_literal: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @6 +; EG-NEXT: ALU 4, @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_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: OR_INT T0.X, T0.X, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: -200(nan), 2(2.802597e-45) +; EG-NEXT: MOV * T0.Y, literal.x, +; EG-NEXT: -1(nan), 0(0.000000e+00) %loada = load i64, ptr addrspace(1) %a, align 8 %or = or i64 %loada, -200 store i64 %or, ptr addrspace(1) %out ret void } -; FUNC-LABEL: {{^}}trunc_i64_or_to_i32: -; SI: s_load_dword s[[SREG0:[0-9]+]] -; SI: s_load_dword s[[SREG1:[0-9]+]] -; SI: s_or_b32 s[[SRESULT:[0-9]+]], s[[SREG1]], s[[SREG0]] -; SI: v_mov_b32_e32 [[VRESULT:v[0-9]+]], s[[SRESULT]] -; SI: buffer_store_dword [[VRESULT]], define amdgpu_kernel void @trunc_i64_or_to_i32(ptr addrspace(1) %out, [8 x i32], i64 %a, [8 x i32], i64 %b) { +; GFX6-LABEL: trunc_i64_or_to_i32: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dword s4, s[0:1], 0x13 +; GFX6-NEXT: s_load_dword s5, s[0:1], 0x1d +; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_or_b32 s4, s5, s4 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: trunc_i64_or_to_i32: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dword s4, s[0:1], 0x4c +; GFX8-NEXT: s_load_dword s5, s[0:1], 0x74 +; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_or_b32 s4, s5, s4 +; GFX8-NEXT: v_mov_b32_e32 v0, s4 +; GFX8-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: trunc_i64_or_to_i32: +; EG: ; %bb.0: +; EG-NEXT: ALU 2, @4, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T1.X, T0.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: ALU clause starting at 4: +; EG-NEXT: LSHR * T0.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; EG-NEXT: OR_INT * T1.X, KC0[7].Y, KC0[4].W, %add = or i64 %b, %a %trunc = trunc i64 %add to i32 store i32 %trunc, ptr addrspace(1) %out, align 8 ret void } -; FUNC-LABEL: {{^}}or_i1: -; EG: OR_INT * {{\** *}}T{{[0-9]+\.[XYZW], PS, PV\.[XYZW]}} - -; SI: s_or_b64 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], vcc define amdgpu_kernel void @or_i1(ptr addrspace(1) %out, ptr addrspace(1) %in0, ptr addrspace(1) %in1) { +; GFX6-LABEL: or_i1: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xd +; GFX6-NEXT: s_mov_b32 s11, 0xf000 +; GFX6-NEXT: s_mov_b32 s10, -1 +; GFX6-NEXT: s_mov_b32 s2, s10 +; GFX6-NEXT: s_mov_b32 s3, s11 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b32 s12, s6 +; GFX6-NEXT: s_mov_b32 s13, s7 +; GFX6-NEXT: s_mov_b32 s14, s10 +; GFX6-NEXT: s_mov_b32 s15, s11 +; GFX6-NEXT: buffer_load_dword v0, off, s[0:3], 0 +; GFX6-NEXT: buffer_load_dword v1, off, s[12:15], 0 +; GFX6-NEXT: s_mov_b32 s8, s4 +; GFX6-NEXT: s_mov_b32 s9, s5 +; GFX6-NEXT: s_waitcnt vmcnt(1) +; GFX6-NEXT: v_cmp_le_f32_e32 vcc, 0, v0 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: v_cmp_le_f32_e64 s[0:1], 0, v1 +; GFX6-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GFX6-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; GFX6-NEXT: buffer_store_dword v0, off, s[8:11], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: or_i1: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x24 +; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34 +; GFX8-NEXT: s_mov_b32 s11, 0xf000 +; GFX8-NEXT: s_mov_b32 s10, -1 +; GFX8-NEXT: s_mov_b32 s2, s10 +; GFX8-NEXT: s_mov_b32 s3, s11 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s12, s6 +; GFX8-NEXT: s_mov_b32 s13, s7 +; GFX8-NEXT: s_mov_b32 s14, s10 +; GFX8-NEXT: s_mov_b32 s15, s11 +; GFX8-NEXT: buffer_load_dword v0, off, s[0:3], 0 +; GFX8-NEXT: buffer_load_dword v1, off, s[12:15], 0 +; GFX8-NEXT: s_mov_b32 s8, s4 +; GFX8-NEXT: s_mov_b32 s9, s5 +; GFX8-NEXT: s_waitcnt vmcnt(1) +; GFX8-NEXT: v_cmp_le_f32_e32 vcc, 0, v0 +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: v_cmp_le_f32_e64 s[0:1], 0, v1 +; GFX8-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GFX8-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] +; GFX8-NEXT: buffer_store_dword v0, off, s[8:11], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: or_i1: +; EG: ; %bb.0: +; EG-NEXT: ALU 0, @12, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @8 +; EG-NEXT: ALU 0, @13, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @10 +; EG-NEXT: ALU 5, @14, 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 8: +; EG-NEXT: VTX_READ_32 T0.X, T0.X, 0, #1 +; EG-NEXT: Fetch clause starting at 10: +; EG-NEXT: VTX_READ_32 T1.X, T1.X, 0, #1 +; EG-NEXT: ALU clause starting at 12: +; EG-NEXT: MOV * T0.X, KC0[2].W, +; EG-NEXT: ALU clause starting at 13: +; EG-NEXT: MOV * T1.X, KC0[2].Z, +; EG-NEXT: ALU clause starting at 14: +; EG-NEXT: SETGE_DX10 T0.W, T0.X, 0.0, +; EG-NEXT: SETGE_DX10 * T1.W, T1.X, 0.0, +; EG-NEXT: OR_INT * T0.W, PS, PV.W, +; EG-NEXT: AND_INT T0.X, PV.W, 1, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %a = load float, ptr addrspace(1) %in0 %b = load float, ptr addrspace(1) %in1 %acmp = fcmp oge float %a, 0.000000e+00 @@ -261,16 +1244,65 @@ ret void } -; FUNC-LABEL: {{^}}s_or_i1: -; SI: s_cmp_eq_u32 -; SI: s_cselect_b64 [[C1:[^,]+]], -1, 0 -; SI: s_cmp_eq_u32 -; SI: s_cselect_b64 [[C2:[^,]+]], -1, 0 -; SI: s_or_b64 s[{{[0-9]+:[0-9]+}}], [[C1]], [[C2]] define amdgpu_kernel void @s_or_i1(ptr addrspace(1) %out, i32 %a, i32 %b, i32 %c, i32 %d) { +; GFX6-LABEL: s_or_i1: +; GFX6: ; %bb.0: +; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0xb +; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_cmp_eq_u32 s4, s5 +; GFX6-NEXT: s_cselect_b64 s[4:5], -1, 0 +; GFX6-NEXT: s_cmp_eq_u32 s6, s7 +; GFX6-NEXT: s_cselect_b64 s[6:7], -1, 0 +; GFX6-NEXT: s_or_b64 s[4:5], s[4:5], s[6:7] +; GFX6-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5] +; GFX6-NEXT: buffer_store_byte v0, off, s[0:3], 0 +; GFX6-NEXT: s_endpgm +; +; GFX8-LABEL: s_or_i1: +; GFX8: ; %bb.0: +; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x2c +; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_cmp_eq_u32 s4, s5 +; GFX8-NEXT: s_cselect_b64 s[4:5], -1, 0 +; GFX8-NEXT: s_cmp_eq_u32 s6, s7 +; GFX8-NEXT: s_cselect_b64 s[6:7], -1, 0 +; GFX8-NEXT: s_or_b64 s[4:5], s[4:5], s[6:7] +; GFX8-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5] +; GFX8-NEXT: buffer_store_byte v0, off, s[0:3], 0 +; GFX8-NEXT: s_endpgm +; +; EG-LABEL: s_or_i1: +; EG: ; %bb.0: +; EG-NEXT: ALU 14, @4, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT MSKOR T0.XW, T1.X +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: ALU clause starting at 4: +; EG-NEXT: SETE_INT T0.W, KC0[3].X, KC0[3].Y, +; EG-NEXT: SETE_INT * T1.W, KC0[2].Z, KC0[2].W, +; EG-NEXT: AND_INT T2.W, KC0[2].Y, literal.x, +; EG-NEXT: OR_INT * T0.W, PS, PV.W, +; EG-NEXT: 3(4.203895e-45), 0(0.000000e+00) +; EG-NEXT: AND_INT T0.W, PS, 1, +; EG-NEXT: LSHL * T1.W, PV.W, literal.x, +; EG-NEXT: 3(4.203895e-45), 0(0.000000e+00) +; EG-NEXT: LSHL T0.X, PV.W, PS, +; EG-NEXT: LSHL * T0.W, literal.x, PS, +; EG-NEXT: 255(3.573311e-43), 0(0.000000e+00) +; EG-NEXT: MOV T0.Y, 0.0, +; EG-NEXT: MOV * T0.Z, 0.0, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) %cmp0 = icmp eq i32 %a, %b %cmp1 = icmp eq i32 %c, %d %or = or i1 %cmp0, %cmp1 store i1 %or, ptr addrspace(1) %out ret void } +