https://github.com/MitchBriles updated https://github.com/llvm/llvm-project/pull/240472
>From ea8ea6081afb1851929426f0128510c20f04cf66 Mon Sep 17 00:00:00 2001 From: Mitch <[email protected]> Date: Fri, 23 Oct 2026 10:10:23 -0600 Subject: [PATCH] [SelectionDAG] Don't CSE freeze of undef Each freeze of undef may have its own arbitrary value. This also folds a bitcast of a one-use frozen undef into a frozen undef. Assisted-By: Claude Opus 5.5 <[email protected]> --- llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp | 4 + .../lib/CodeGen/SelectionDAG/SelectionDAG.cpp | 8 +- llvm/test/CodeGen/AArch64/freeze.ll | 73 ++++++---- llvm/test/CodeGen/AMDGPU/permute_i8.ll | 6 +- ...-alloca-vector-dynamic-idx-bitcasts-llc.ll | 134 +++--------------- llvm/test/CodeGen/Hexagon/pr196966.ll | 41 ++---- llvm/test/CodeGen/SystemZ/pr60413.ll | 32 ----- llvm/test/CodeGen/X86/freeze-vector.ll | 39 ++--- llvm/test/CodeGen/X86/freeze.ll | 3 - .../unittests/CodeGen/SelectionDAGCSETest.cpp | 20 +++ 10 files changed, 138 insertions(+), 222 deletions(-) diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp index 0bd1ccb51aec60..057fcda5260c20 100644 --- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp +++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp @@ -18480,6 +18480,10 @@ SDValue DAGCombiner::visitBITCAST(SDNode *N) { if (SDValue R = propagateUnaryUndef(DAG, N0, VT)) return R; + // A bitcast of an otherwise unused frozen undef is itself a frozen undef. + if (ISD::isFreezeUndef(N0) && N0.hasOneUse()) + return DAG.getFreeze(DAG.getUNDEF(VT)); + // If the input is a BUILD_VECTOR with all constant elements, fold this now. // Only do this before legalize types, unless both types are integer and the // scalar type is legal. Only do this before legalize ops, since the target diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp index c07494f5a05026..5798df03a61722 100644 --- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp +++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp @@ -1060,6 +1060,9 @@ static bool doNotCSE(SDNode *N) { case ISD::HANDLENODE: case ISD::EH_LABEL: return true; // Never CSE these nodes. + case ISD::FREEZE: + // Each freeze of undef is a distinct value. + return N->getOperand(0).isUndef(); } // Check that remaining values produced are not flags. @@ -7603,7 +7606,8 @@ SDValue SelectionDAG::getNode(unsigned Opcode, const SDLoc &DL, EVT VT, SDNode *N; SDVTList VTs = getVTList(VT); SDValue Ops[] = {N1}; - if (VT != MVT::Glue) { // Don't CSE glue producing nodes + // Don't CSE glue producing nodes or freezes of undef. + if (VT != MVT::Glue && !(Opcode == ISD::FREEZE && N1.isUndef())) { SDNodeKey ID(Opcode, VTs, Ops); FoldingSetInsertToken InsertToken; if (SDNode *E = lookupNode(ID, DL, InsertToken)) { @@ -12495,7 +12499,7 @@ SDNode *SelectionDAG::UpdateNodeOperands(SDNode *N, SDValue Op) { updateDivergence(N); // If this gets put into a CSE map, add it. - if (InsertToken) + if (InsertToken && !doNotCSE(N)) CSEMap.insert(N, InsertToken); return N; } diff --git a/llvm/test/CodeGen/AArch64/freeze.ll b/llvm/test/CodeGen/AArch64/freeze.ll index 3476dc0c815737..a660c52c567ede 100644 --- a/llvm/test/CodeGen/AArch64/freeze.ll +++ b/llvm/test/CodeGen/AArch64/freeze.ll @@ -118,11 +118,17 @@ define <16 x i8> @freeze_v16i8() { } define <32 x i8> @freeze_v32i8() { -; CHECK-LABEL: freeze_v32i8: -; CHECK: // %bb.0: -; CHECK-NEXT: add v0.16b, v0.16b, v0.16b -; CHECK-NEXT: mov v1.16b, v0.16b -; CHECK-NEXT: ret +; CHECK-SD-LABEL: freeze_v32i8: +; CHECK-SD: // %bb.0: +; CHECK-SD-NEXT: add v0.16b, v0.16b, v0.16b +; CHECK-SD-NEXT: add v1.16b, v0.16b, v0.16b +; CHECK-SD-NEXT: ret +; +; CHECK-GI-LABEL: freeze_v32i8: +; CHECK-GI: // %bb.0: +; CHECK-GI-NEXT: add v0.16b, v0.16b, v0.16b +; CHECK-GI-NEXT: mov v1.16b, v0.16b +; CHECK-GI-NEXT: ret %y1 = freeze <32 x i8> undef %t1 = add <32 x i8> %y1, %y1 ret <32 x i8> %t1 @@ -177,11 +183,17 @@ define <8 x i16> @freeze_v8i16() { } define <16 x i16> @freeze_v16i16() { -; CHECK-LABEL: freeze_v16i16: -; CHECK: // %bb.0: -; CHECK-NEXT: add v0.8h, v0.8h, v0.8h -; CHECK-NEXT: mov v1.16b, v0.16b -; CHECK-NEXT: ret +; CHECK-SD-LABEL: freeze_v16i16: +; CHECK-SD: // %bb.0: +; CHECK-SD-NEXT: add v0.8h, v0.8h, v0.8h +; CHECK-SD-NEXT: add v1.8h, v0.8h, v0.8h +; CHECK-SD-NEXT: ret +; +; CHECK-GI-LABEL: freeze_v16i16: +; CHECK-GI: // %bb.0: +; CHECK-GI-NEXT: add v0.8h, v0.8h, v0.8h +; CHECK-GI-NEXT: mov v1.16b, v0.16b +; CHECK-GI-NEXT: ret %y1 = freeze <16 x i16> undef %t1 = add <16 x i16> %y1, %y1 ret <16 x i16> %t1 @@ -218,11 +230,17 @@ define <4 x i32> @freeze_v4i32() { } define <8 x i32> @freeze_v8i32() { -; CHECK-LABEL: freeze_v8i32: -; CHECK: // %bb.0: -; CHECK-NEXT: add v0.4s, v0.4s, v0.4s -; CHECK-NEXT: mov v1.16b, v0.16b -; CHECK-NEXT: ret +; CHECK-SD-LABEL: freeze_v8i32: +; CHECK-SD: // %bb.0: +; CHECK-SD-NEXT: add v0.4s, v0.4s, v0.4s +; CHECK-SD-NEXT: add v1.4s, v0.4s, v0.4s +; CHECK-SD-NEXT: ret +; +; CHECK-GI-LABEL: freeze_v8i32: +; CHECK-GI: // %bb.0: +; CHECK-GI-NEXT: add v0.4s, v0.4s, v0.4s +; CHECK-GI-NEXT: mov v1.16b, v0.16b +; CHECK-GI-NEXT: ret %y1 = freeze <8 x i32> undef %t1 = add <8 x i32> %y1, %y1 ret <8 x i32> %t1 @@ -243,7 +261,8 @@ define <3 x i64> @freeze_v3i64() { ; CHECK-SD: // %bb.0: ; CHECK-SD-NEXT: add v0.2d, v0.2d, v0.2d ; CHECK-SD-NEXT: mov d1, v0.d[1] -; CHECK-SD-NEXT: fmov d2, d0 +; CHECK-SD-NEXT: add d2, d0, d0 +; CHECK-SD-NEXT: // kill: def $d0 killed $d0 killed $q0 ; CHECK-SD-NEXT: ret ; ; CHECK-GI-LABEL: freeze_v3i64: @@ -260,11 +279,17 @@ define <3 x i64> @freeze_v3i64() { } define <4 x i64> @freeze_v4i64() { -; CHECK-LABEL: freeze_v4i64: -; CHECK: // %bb.0: -; CHECK-NEXT: add v0.2d, v0.2d, v0.2d -; CHECK-NEXT: mov v1.16b, v0.16b -; CHECK-NEXT: ret +; CHECK-SD-LABEL: freeze_v4i64: +; CHECK-SD: // %bb.0: +; CHECK-SD-NEXT: add v0.2d, v0.2d, v0.2d +; CHECK-SD-NEXT: add v1.2d, v0.2d, v0.2d +; CHECK-SD-NEXT: ret +; +; CHECK-GI-LABEL: freeze_v4i64: +; CHECK-GI: // %bb.0: +; CHECK-GI-NEXT: add v0.2d, v0.2d, v0.2d +; CHECK-GI-NEXT: mov v1.16b, v0.16b +; CHECK-GI-NEXT: ret %y1 = freeze <4 x i64> undef %t1 = add <4 x i64> %y1, %y1 ret <4 x i64> %t1 @@ -318,9 +343,9 @@ define <4 x ptr> @freeze_v4p0() { ; CHECK-SD-LABEL: freeze_v4p0: ; CHECK-SD: // %bb.0: ; CHECK-SD-NEXT: mov w8, #4 // =0x4 -; CHECK-SD-NEXT: dup v0.2d, x8 -; CHECK-SD-NEXT: add v0.2d, v0.2d, v0.2d -; CHECK-SD-NEXT: mov v1.16b, v0.16b +; CHECK-SD-NEXT: dup v1.2d, x8 +; CHECK-SD-NEXT: add v0.2d, v0.2d, v1.2d +; CHECK-SD-NEXT: add v1.2d, v0.2d, v1.2d ; CHECK-SD-NEXT: ret ; ; CHECK-GI-LABEL: freeze_v4p0: diff --git a/llvm/test/CodeGen/AMDGPU/permute_i8.ll b/llvm/test/CodeGen/AMDGPU/permute_i8.ll index 90e655d7bf8b56..0e67c1c517b684 100644 --- a/llvm/test/CodeGen/AMDGPU/permute_i8.ll +++ b/llvm/test/CodeGen/AMDGPU/permute_i8.ll @@ -3769,6 +3769,7 @@ define hidden void @any_extend_to_perm_wide_src(ptr addrspace(1) %in0, ptr addrs ; GFX10-NEXT: ; kill: killed $vgpr0 killed $vgpr1 ; GFX10-NEXT: ; kill: killed $vgpr2 killed $vgpr3 ; GFX10-NEXT: s_lshl_b32 s4, s4, 8 +; GFX10-NEXT: s_lshl_b32 s5, s4, 8 ; GFX10-NEXT: s_waitcnt vmcnt(2) ; GFX10-NEXT: v_lshrrev_b64 v[0:1], 24, v[6:7] ; GFX10-NEXT: s_waitcnt vmcnt(1) @@ -3778,7 +3779,7 @@ define hidden void @any_extend_to_perm_wide_src(ptr addrspace(1) %in0, ptr addrs ; GFX10-NEXT: v_or_b32_sdwa v3, v8, s4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD ; GFX10-NEXT: v_perm_b32 v0, v1, v0, 0xc0c0004 ; GFX10-NEXT: v_perm_b32 v1, s4, v2, 0xc0c0004 -; GFX10-NEXT: v_or_b32_sdwa v2, v9, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD +; GFX10-NEXT: v_or_b32_sdwa v2, v9, s5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD ; GFX10-NEXT: v_lshl_or_b32 v0, v0, 16, v1 ; GFX10-NEXT: v_or_b32_sdwa v1, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 ; GFX10-NEXT: global_store_dwordx2 v[4:5], v[0:1], off @@ -3795,6 +3796,7 @@ define hidden void @any_extend_to_perm_wide_src(ptr addrspace(1) %in0, ptr addrs ; GFX9-NEXT: s_mov_b32 s4, 0xc0c0004 ; GFX9-NEXT: v_mov_b32_e32 v14, 0xc0c0004 ; GFX9-NEXT: s_lshl_b32 s5, s4, 8 +; GFX9-NEXT: s_lshl_b32 s6, s4, 8 ; GFX9-NEXT: s_waitcnt vmcnt(2) ; GFX9-NEXT: v_lshrrev_b64 v[0:1], 8, v[6:7] ; GFX9-NEXT: s_waitcnt vmcnt(1) @@ -3802,7 +3804,7 @@ define hidden void @any_extend_to_perm_wide_src(ptr addrspace(1) %in0, ptr addrs ; GFX9-NEXT: v_lshrrev_b64 v[2:3], 16, v[8:9] ; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: v_or_b32_sdwa v3, v12, s5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX9-NEXT: v_or_b32_sdwa v6, v13, s5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD +; GFX9-NEXT: v_or_b32_sdwa v6, v13, s6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD ; GFX9-NEXT: v_perm_b32 v1, v2, v1, s4 ; GFX9-NEXT: v_perm_b32 v0, s4, v0, v14 ; GFX9-NEXT: v_lshl_or_b32 v0, v1, 16, v0 diff --git a/llvm/test/CodeGen/AMDGPU/promote-alloca-vector-dynamic-idx-bitcasts-llc.ll b/llvm/test/CodeGen/AMDGPU/promote-alloca-vector-dynamic-idx-bitcasts-llc.ll index 10287c131acdbb..c73d31c368846b 100644 --- a/llvm/test/CodeGen/AMDGPU/promote-alloca-vector-dynamic-idx-bitcasts-llc.ll +++ b/llvm/test/CodeGen/AMDGPU/promote-alloca-vector-dynamic-idx-bitcasts-llc.ll @@ -6,140 +6,46 @@ define amdgpu_kernel void @test_bitcast_llc_v128i8_v16i8(ptr addrspace(1) %out, i32 %idx) { ; GFX9-LABEL: test_bitcast_llc_v128i8_v16i8: ; GFX9: ; %bb.0: ; %entry -; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0 ; GFX9-NEXT: s_load_dword s2, s[4:5], 0x8 -; GFX9-NEXT: v_mov_b32_e32 v0, 0x6050400 -; GFX9-NEXT: v_mov_b32_e32 v36, 0 +; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0 +; GFX9-NEXT: v_mov_b32_e32 v4, 0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_perm_b32 v0, s0, s0, v0 -; GFX9-NEXT: v_lshlrev_b32_e32 v1, 16, v0 -; GFX9-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD ; GFX9-NEXT: s_lshl_b32 s2, s2, 2 -; GFX9-NEXT: v_mov_b32_e32 v1, v0 -; GFX9-NEXT: v_mov_b32_e32 v2, v0 -; GFX9-NEXT: v_mov_b32_e32 v3, v0 -; GFX9-NEXT: v_mov_b32_e32 v4, v0 -; GFX9-NEXT: v_mov_b32_e32 v5, v0 -; GFX9-NEXT: v_mov_b32_e32 v6, v0 -; GFX9-NEXT: v_mov_b32_e32 v7, v0 -; GFX9-NEXT: v_mov_b32_e32 v8, v0 -; GFX9-NEXT: v_mov_b32_e32 v9, v0 -; GFX9-NEXT: v_mov_b32_e32 v10, v0 -; GFX9-NEXT: v_mov_b32_e32 v11, v0 -; GFX9-NEXT: v_mov_b32_e32 v12, v0 -; GFX9-NEXT: v_mov_b32_e32 v13, v0 -; GFX9-NEXT: v_mov_b32_e32 v14, v0 -; GFX9-NEXT: v_mov_b32_e32 v15, v0 -; GFX9-NEXT: v_mov_b32_e32 v16, v0 -; GFX9-NEXT: v_mov_b32_e32 v17, v0 -; GFX9-NEXT: v_mov_b32_e32 v18, v0 -; GFX9-NEXT: v_mov_b32_e32 v19, v0 -; GFX9-NEXT: v_mov_b32_e32 v20, v0 -; GFX9-NEXT: v_mov_b32_e32 v21, v0 -; GFX9-NEXT: v_mov_b32_e32 v22, v0 -; GFX9-NEXT: v_mov_b32_e32 v23, v0 -; GFX9-NEXT: v_mov_b32_e32 v24, v0 -; GFX9-NEXT: v_mov_b32_e32 v25, v0 -; GFX9-NEXT: v_mov_b32_e32 v26, v0 -; GFX9-NEXT: v_mov_b32_e32 v27, v0 -; GFX9-NEXT: v_mov_b32_e32 v28, v0 -; GFX9-NEXT: v_mov_b32_e32 v29, v0 -; GFX9-NEXT: v_mov_b32_e32 v30, v0 -; GFX9-NEXT: v_mov_b32_e32 v31, v0 ; GFX9-NEXT: s_set_gpr_idx_on s2, gpr_idx(SRC0) -; GFX9-NEXT: v_mov_b32_e32 v35, v3 -; GFX9-NEXT: v_mov_b32_e32 v34, v2 -; GFX9-NEXT: v_mov_b32_e32 v33, v1 -; GFX9-NEXT: v_mov_b32_e32 v32, v0 +; GFX9-NEXT: v_mov_b32_e32 v3, v3 +; GFX9-NEXT: v_mov_b32_e32 v2, v2 +; GFX9-NEXT: v_mov_b32_e32 v1, v1 +; GFX9-NEXT: v_mov_b32_e32 v0, v0 ; GFX9-NEXT: s_set_gpr_idx_off -; GFX9-NEXT: global_store_dwordx4 v36, v[32:35], s[0:1] +; GFX9-NEXT: global_store_dwordx4 v4, v[0:3], s[0:1] ; GFX9-NEXT: s_endpgm ; ; GFX11-LABEL: test_bitcast_llc_v128i8_v16i8: ; GFX11: ; %bb.0: ; %entry -; GFX11-NEXT: v_perm_b32 v0, s0, s0, 0x6050400 ; GFX11-NEXT: s_clause 0x1 -; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x0 ; GFX11-NEXT: s_load_b32 s2, s[4:5], 0x8 -; GFX11-NEXT: v_mov_b32_e32 v35, 0 -; GFX11-NEXT: v_and_b32_e32 v1, 0xffff, v0 -; GFX11-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_or_b32_e32 v0, v1, v0 -; GFX11-NEXT: v_mov_b32_e32 v1, v0 -; GFX11-NEXT: v_mov_b32_e32 v2, v0 -; GFX11-NEXT: v_mov_b32_e32 v3, v0 -; GFX11-NEXT: v_mov_b32_e32 v31, v0 -; GFX11-NEXT: v_mov_b32_e32 v4, v0 +; GFX11-NEXT: s_load_b64 s[0:1], s[4:5], 0x0 +; GFX11-NEXT: v_mov_b32_e32 v4, 0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_lshl_b32 m0, s2, 2 -; GFX11-NEXT: v_mov_b32_e32 v5, v0 -; GFX11-NEXT: v_mov_b32_e32 v6, v0 -; GFX11-NEXT: v_mov_b32_e32 v7, v0 -; GFX11-NEXT: v_mov_b32_e32 v8, v0 -; GFX11-NEXT: v_mov_b32_e32 v9, v0 -; GFX11-NEXT: v_mov_b32_e32 v10, v0 -; GFX11-NEXT: v_mov_b32_e32 v11, v0 -; GFX11-NEXT: v_mov_b32_e32 v12, v0 -; GFX11-NEXT: v_mov_b32_e32 v13, v0 -; GFX11-NEXT: v_mov_b32_e32 v14, v0 -; GFX11-NEXT: v_mov_b32_e32 v15, v0 -; GFX11-NEXT: v_mov_b32_e32 v16, v0 -; GFX11-NEXT: v_mov_b32_e32 v17, v0 -; GFX11-NEXT: v_mov_b32_e32 v18, v0 -; GFX11-NEXT: v_mov_b32_e32 v19, v0 -; GFX11-NEXT: v_mov_b32_e32 v20, v0 -; GFX11-NEXT: v_mov_b32_e32 v21, v0 -; GFX11-NEXT: v_mov_b32_e32 v22, v0 -; GFX11-NEXT: v_mov_b32_e32 v23, v0 -; GFX11-NEXT: v_mov_b32_e32 v24, v0 -; GFX11-NEXT: v_mov_b32_e32 v25, v0 -; GFX11-NEXT: v_mov_b32_e32 v26, v0 -; GFX11-NEXT: v_mov_b32_e32 v27, v0 -; GFX11-NEXT: v_mov_b32_e32 v28, v0 -; GFX11-NEXT: v_mov_b32_e32 v29, v0 -; GFX11-NEXT: v_mov_b32_e32 v30, v0 -; GFX11-NEXT: v_movrels_b32_e32 v34, v3 -; GFX11-NEXT: v_movrels_b32_e32 v33, v2 -; GFX11-NEXT: v_movrels_b32_e32 v32, v1 -; GFX11-NEXT: v_movrels_b32_e32 v31, v0 -; GFX11-NEXT: global_store_b128 v35, v[31:34], s[0:1] +; GFX11-NEXT: v_movrels_b32_e32 v3, v3 +; GFX11-NEXT: v_movrels_b32_e32 v2, v2 +; GFX11-NEXT: v_movrels_b32_e32 v1, v1 +; GFX11-NEXT: v_movrels_b32_e32 v0, v0 +; GFX11-NEXT: global_store_b128 v4, v[0:3], s[0:1] ; GFX11-NEXT: s_endpgm ; ; GFX12-LABEL: test_bitcast_llc_v128i8_v16i8: ; GFX12: ; %bb.0: ; %entry -; GFX12-NEXT: v_perm_b32 v0, s0, s0, 0x6050400 ; GFX12-NEXT: s_load_b96 s[0:2], s[4:5], 0x0 -; GFX12-NEXT: v_mov_b32_e32 v35, 0 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-NEXT: v_and_b32_e32 v1, 0xffff, v0 -; GFX12-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX12-NEXT: v_or_b32_e32 v0, v1, v0 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v4, v0 -; GFX12-NEXT: v_dual_mov_b32 v2, v0 :: v_dual_mov_b32 v3, v0 -; GFX12-NEXT: v_mov_b32_e32 v6, v0 +; GFX12-NEXT: v_mov_b32_e32 v4, 0 ; GFX12-NEXT: s_wait_kmcnt 0x0 ; GFX12-NEXT: s_lshl_b32 m0, s2, 2 -; GFX12-NEXT: v_mov_b32_e32 v31, v0 -; GFX12-NEXT: v_dual_mov_b32 v5, v0 :: v_dual_mov_b32 v8, v0 -; GFX12-NEXT: v_dual_mov_b32 v7, v0 :: v_dual_mov_b32 v10, v0 -; GFX12-NEXT: v_dual_mov_b32 v9, v0 :: v_dual_mov_b32 v12, v0 -; GFX12-NEXT: v_dual_mov_b32 v11, v0 :: v_dual_mov_b32 v14, v0 -; GFX12-NEXT: v_dual_mov_b32 v13, v0 :: v_dual_mov_b32 v16, v0 -; GFX12-NEXT: v_dual_mov_b32 v15, v0 :: v_dual_mov_b32 v18, v0 -; GFX12-NEXT: v_dual_mov_b32 v17, v0 :: v_dual_mov_b32 v20, v0 -; GFX12-NEXT: v_dual_mov_b32 v19, v0 :: v_dual_mov_b32 v22, v0 -; GFX12-NEXT: v_dual_mov_b32 v21, v0 :: v_dual_mov_b32 v24, v0 -; GFX12-NEXT: v_dual_mov_b32 v23, v0 :: v_dual_mov_b32 v26, v0 -; GFX12-NEXT: v_dual_mov_b32 v25, v0 :: v_dual_mov_b32 v28, v0 -; GFX12-NEXT: v_dual_mov_b32 v27, v0 :: v_dual_mov_b32 v30, v0 -; GFX12-NEXT: v_mov_b32_e32 v29, v0 -; GFX12-NEXT: v_movrels_b32_e32 v34, v3 -; GFX12-NEXT: v_movrels_b32_e32 v33, v2 -; GFX12-NEXT: v_movrels_b32_e32 v32, v1 -; GFX12-NEXT: v_movrels_b32_e32 v31, v0 -; GFX12-NEXT: global_store_b128 v35, v[31:34], s[0:1] +; GFX12-NEXT: v_movrels_b32_e32 v3, v3 +; GFX12-NEXT: v_movrels_b32_e32 v2, v2 +; GFX12-NEXT: v_movrels_b32_e32 v1, v1 +; GFX12-NEXT: v_movrels_b32_e32 v0, v0 +; GFX12-NEXT: global_store_b128 v4, v[0:3], s[0:1] ; GFX12-NEXT: s_endpgm entry: %alloca = freeze <128 x i8> poison diff --git a/llvm/test/CodeGen/Hexagon/pr196966.ll b/llvm/test/CodeGen/Hexagon/pr196966.ll index cba853ef529e4a..8612395feb8e40 100644 --- a/llvm/test/CodeGen/Hexagon/pr196966.ll +++ b/llvm/test/CodeGen/Hexagon/pr196966.ll @@ -34,46 +34,33 @@ define void @pr196966(ptr %base) { ; CHECK-NEXT: r0 = add(r29,#0) ; CHECK-NEXT: } ; CHECK-NEXT: { -; CHECK-NEXT: r0 = p0 -; CHECK-NEXT: r1 = add(r16,#6) -; CHECK-NEXT: r5:4 = memd(r29+#8) -; CHECK-NEXT: r3:2 = memd(r29+#24) -; CHECK-NEXT: } -; CHECK-NEXT: { -; CHECK-NEXT: r0 |= asl(r0,#8) -; CHECK-NEXT: r5 = add(r16,#20) +; CHECK-NEXT: r6 = add(r16,#6) +; CHECK-NEXT: r1:0 = memd(r29+#24) +; CHECK-NEXT: r5:4 = memd(r29+#0) ; CHECK-NEXT: } ; CHECK-NEXT: { +; CHECK-NEXT: r5:4 = extractu(r5:4,#16,#32) ; CHECK-NEXT: p0 = bitsclr(r0,#4) -; CHECK-NEXT: if (p0.new) jump:t .LBB0_3 +; CHECK-NEXT: r3:2 = memd(r29+#8) ; CHECK-NEXT: } -; CHECK-NEXT: // %bb.2: // %cond.store3 ; CHECK-NEXT: { -; CHECK-NEXT: r7:6 = memd(r29+#0) +; CHECK-NEXT: p1 = bitsclr(r0,#32) +; CHECK-NEXT: p2 = !tstbit(r0,#13) +; CHECK-NEXT: r3 = add(r16,#20) +; CHECK-NEXT: if (!p0) memh(r0+#0) = r4 ; CHECK-NEXT: } ; CHECK-NEXT: { -; CHECK-NEXT: r7:6 = extractu(r7:6,#16,#32) +; CHECK-NEXT: p0 = !tstbit(r0,#15) +; CHECK-NEXT: if (!p1) memh(r3+#0) = r2.h +; CHECK-NEXT: if (!p2) memh(r6+#0) = r0.h ; CHECK-NEXT: } ; CHECK-NEXT: { -; CHECK-NEXT: memh(r0+#0) = r6 -; CHECK-NEXT: } -; CHECK-NEXT: .LBB0_3: // %else4 -; CHECK-NEXT: { -; CHECK-NEXT: p0 = bitsclr(r0,#32) -; CHECK-NEXT: p1 = !tstbit(r0,#13) ; CHECK-NEXT: r17:16 = memd(r30+#-8) -; CHECK-NEXT: if (!p0.new) memh(r5+#0) = r4.h -; CHECK-NEXT: } // 8-byte Folded Reload -; CHECK-NEXT: { -; CHECK-NEXT: p2 = !tstbit(r0,#15) -; CHECK-NEXT: if (!p1) memh(r1+#0) = r2.h -; CHECK-NEXT: } -; CHECK-NEXT: { ; CHECK-NEXT: r31:30 = deallocframe(r30):raw -; CHECK-NEXT: if (!p2) memh(r0+#0) = r3.h -; CHECK-NEXT: } +; CHECK-NEXT: } // 8-byte Folded Reload ; CHECK-NEXT: { ; CHECK-NEXT: jumpr r31 +; CHECK-NEXT: if (!p0) memh(r0+#0) = r1.h ; CHECK-NEXT: } entry: %mb = icmp ne ptr %base, null diff --git a/llvm/test/CodeGen/SystemZ/pr60413.ll b/llvm/test/CodeGen/SystemZ/pr60413.ll index 8a6a30318ae583..97cf1050408dd5 100644 --- a/llvm/test/CodeGen/SystemZ/pr60413.ll +++ b/llvm/test/CodeGen/SystemZ/pr60413.ll @@ -74,38 +74,6 @@ define dso_local void @m() local_unnamed_addr #1 { ; CHECK-NEXT: rosbg %r2, %r3, 32, 61, 2 ; CHECK-NEXT: rosbg %r2, %r5, 32, 62, 1 ; CHECK-NEXT: or %r2, %r14 -; CHECK-NEXT: vlgvb %r4, %v0, 1 -; CHECK-NEXT: vlgvb %r3, %v0, 0 -; CHECK-NEXT: risbg %r3, %r3, 48, 176, 15 -; CHECK-NEXT: rosbg %r3, %r4, 49, 49, 14 -; CHECK-NEXT: vlgvb %r4, %v0, 2 -; CHECK-NEXT: rosbg %r3, %r4, 50, 50, 13 -; CHECK-NEXT: vlgvb %r4, %v0, 3 -; CHECK-NEXT: rosbg %r3, %r4, 51, 51, 12 -; CHECK-NEXT: vlgvb %r4, %v0, 4 -; CHECK-NEXT: rosbg %r3, %r4, 52, 52, 11 -; CHECK-NEXT: vlgvb %r4, %v0, 5 -; CHECK-NEXT: rosbg %r3, %r4, 53, 53, 10 -; CHECK-NEXT: vlgvb %r4, %v0, 6 -; CHECK-NEXT: rosbg %r3, %r4, 54, 54, 9 -; CHECK-NEXT: vlgvb %r4, %v0, 7 -; CHECK-NEXT: rosbg %r3, %r4, 55, 55, 8 -; CHECK-NEXT: vlgvb %r4, %v0, 8 -; CHECK-NEXT: rosbg %r3, %r4, 56, 56, 7 -; CHECK-NEXT: vlgvb %r4, %v0, 9 -; CHECK-NEXT: rosbg %r3, %r4, 57, 57, 6 -; CHECK-NEXT: vlgvb %r4, %v0, 10 -; CHECK-NEXT: rosbg %r3, %r4, 58, 58, 5 -; CHECK-NEXT: vlgvb %r4, %v0, 11 -; CHECK-NEXT: rosbg %r3, %r4, 59, 59, 4 -; CHECK-NEXT: vlgvb %r4, %v0, 12 -; CHECK-NEXT: rosbg %r3, %r4, 60, 60, 3 -; CHECK-NEXT: vlgvb %r4, %v0, 13 -; CHECK-NEXT: rosbg %r3, %r4, 61, 61, 2 -; CHECK-NEXT: vlgvb %r4, %v0, 14 -; CHECK-NEXT: rosbg %r3, %r4, 62, 62, 1 -; CHECK-NEXT: vlgvb %r4, %v0, 15 -; CHECK-NEXT: rosbg %r3, %r4, 63, 63, 0 ; CHECK-NEXT: xilf %r3, 4294967295 ; CHECK-NEXT: or %r3, %r2 ; CHECK-NEXT: tmll %r3, 65535 diff --git a/llvm/test/CodeGen/X86/freeze-vector.ll b/llvm/test/CodeGen/X86/freeze-vector.ll index b97d75802d050e..2327dac331088d 100644 --- a/llvm/test/CodeGen/X86/freeze-vector.ll +++ b/llvm/test/CodeGen/X86/freeze-vector.ll @@ -389,17 +389,20 @@ define void @freeze_two_frozen_buildvectors(ptr %origin0, ptr %origin1, ptr %dst ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X86-NEXT: movl {{[0-9]+}}(%esp), %edx ; X86-NEXT: movl {{[0-9]+}}(%esp), %esi +; X86-NEXT: movl (%esi), %esi +; X86-NEXT: andl $15, %esi ; X86-NEXT: movl (%edx), %edx -; X86-NEXT: vmovd %eax, %xmm0 -; X86-NEXT: vpinsrd $1, (%esi), %xmm0, %xmm1 -; X86-NEXT: vpinsrd $2, %eax, %xmm1, %xmm1 -; X86-NEXT: vpinsrd $3, %eax, %xmm1, %xmm1 +; X86-NEXT: andl $15, %edx +; X86-NEXT: vmovd %esi, %xmm0 +; X86-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,0,1,1] +; X86-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; X86-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3],xmm1[4,5,6,7] ; X86-NEXT: vbroadcastss {{.*#+}} xmm2 = [7,7,7,7] -; X86-NEXT: vpand %xmm2, %xmm1, %xmm1 -; X86-NEXT: vmovdqa %xmm1, (%ecx) -; X86-NEXT: vpinsrd $1, %eax, %xmm0, %xmm0 -; X86-NEXT: vpinsrd $2, %edx, %xmm0, %xmm0 -; X86-NEXT: vpinsrd $3, %eax, %xmm0, %xmm0 +; X86-NEXT: vpand %xmm2, %xmm0, %xmm0 +; X86-NEXT: vmovdqa %xmm0, (%ecx) +; X86-NEXT: vmovd %edx, %xmm0 +; X86-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1] +; X86-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7] ; X86-NEXT: vpand %xmm2, %xmm0, %xmm0 ; X86-NEXT: vmovdqa %xmm0, (%eax) ; X86-NEXT: popl %esi @@ -408,16 +411,16 @@ define void @freeze_two_frozen_buildvectors(ptr %origin0, ptr %origin1, ptr %dst ; X64-LABEL: freeze_two_frozen_buildvectors: ; X64: # %bb.0: ; X64-NEXT: movl (%rsi), %eax -; X64-NEXT: vmovd %eax, %xmm0 -; X64-NEXT: vpinsrd $1, (%rdi), %xmm0, %xmm1 -; X64-NEXT: vpinsrd $2, %eax, %xmm1, %xmm1 -; X64-NEXT: vpinsrd $3, %eax, %xmm1, %xmm1 +; X64-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero +; X64-NEXT: vpbroadcastd %xmm0, %xmm0 +; X64-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; X64-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3] ; X64-NEXT: vpbroadcastd {{.*#+}} xmm2 = [7,7,7,7] -; X64-NEXT: vpand %xmm2, %xmm1, %xmm1 -; X64-NEXT: vmovdqa %xmm1, (%rdx) -; X64-NEXT: vpinsrd $1, %eax, %xmm0, %xmm0 -; X64-NEXT: vpinsrd $2, %eax, %xmm0, %xmm0 -; X64-NEXT: vpinsrd $3, %eax, %xmm0, %xmm0 +; X64-NEXT: vpand %xmm2, %xmm0, %xmm0 +; X64-NEXT: vmovdqa %xmm0, (%rdx) +; X64-NEXT: vmovd %eax, %xmm0 +; X64-NEXT: vpbroadcastd %xmm0, %xmm0 +; X64-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2],xmm1[3] ; X64-NEXT: vpand %xmm2, %xmm0, %xmm0 ; X64-NEXT: vmovdqa %xmm0, (%rcx) ; X64-NEXT: retq diff --git a/llvm/test/CodeGen/X86/freeze.ll b/llvm/test/CodeGen/X86/freeze.ll index de5ca17bdb2622..5ea78f884e9803 100644 --- a/llvm/test/CodeGen/X86/freeze.ll +++ b/llvm/test/CodeGen/X86/freeze.ll @@ -89,7 +89,6 @@ define ptr @freeze_ptr() { define i32 @freeze_struct() { ; X86ASM-LABEL: freeze_struct: ; X86ASM: # %bb.0: -; X86ASM-NEXT: # implicit-def: $eax ; X86ASM-NEXT: addl %eax, %eax ; X86ASM-NEXT: retq %y1 = freeze %struct.T undef @@ -102,7 +101,6 @@ define i32 @freeze_struct() { define i32 @freeze_anonstruct() { ; X86ASM-LABEL: freeze_anonstruct: ; X86ASM: # %bb.0: -; X86ASM-NEXT: # implicit-def: $eax ; X86ASM-NEXT: addl %eax, %eax ; X86ASM-NEXT: retq %y1 = freeze {i32, i32} undef @@ -129,7 +127,6 @@ define i32 @freeze_anonstruct2() { define i64 @freeze_array() { ; X86ASM-LABEL: freeze_array: ; X86ASM: # %bb.0: -; X86ASM-NEXT: # implicit-def: $rax ; X86ASM-NEXT: addq %rax, %rax ; X86ASM-NEXT: retq %y1 = freeze [2 x i64] undef diff --git a/llvm/unittests/CodeGen/SelectionDAGCSETest.cpp b/llvm/unittests/CodeGen/SelectionDAGCSETest.cpp index ea073bc6b8c55e..3859a02cd1d62f 100644 --- a/llvm/unittests/CodeGen/SelectionDAGCSETest.cpp +++ b/llvm/unittests/CodeGen/SelectionDAGCSETest.cpp @@ -87,6 +87,26 @@ TEST_F(SelectionDAGCSETest, FPEnvMem) { DAG->getSetFPEnv(getChain(), Loc, Ptr, MVT::i32, MMO).getNode()); } +TEST_F(SelectionDAGCSETest, FreezeUndef) { + SDValue Undef = DAG->getUNDEF(MVT::i32); + EXPECT_NE(DAG->getFreeze(Undef).getNode(), DAG->getFreeze(Undef).getNode()); + + auto Load = [&] { + return DAG->getLoad(MVT::i32, Loc, getChain(), + DAG->CreateStackTemporary(MVT::i32), + MachinePointerInfo()); + }; + SDValue X = Load(), Y = Load(); + SDNode *FX = DAG->getFreeze(X).getNode(); + SDNode *FY = DAG->getFreeze(Y).getNode(); + EXPECT_EQ(FX, DAG->getFreeze(X).getNode()); + + // A freeze whose operand becomes undef must leave the CSE map. + EXPECT_EQ(DAG->UpdateNodeOperands(FX, Undef), FX); + EXPECT_EQ(DAG->UpdateNodeOperands(FY, Undef), FY); + EXPECT_NE(DAG->getFreeze(X).getNode(), FX); +} + TEST_F(SelectionDAGCSETest, LifetimeNode) { SDValue FIPtr = DAG->CreateStackTemporary(MVT::i32); int FI = cast<FrameIndexSDNode>(FIPtr.getNode())->getIndex(); _______________________________________________ llvm-branch-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
