From: Sanjay Patel Date: Sun, 15 Jul 2018 16:27:07 +0000 (+0000) Subject: [DAGCombiner] extend(ifpositive(X)) -> shift-right (not X) X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=08378e6ecfb30066def16d16c18d9c147dce9f4a;p=llvm [DAGCombiner] extend(ifpositive(X)) -> shift-right (not X) This is almost the same as an existing IR canonicalization in instcombine, so I'm assuming this is a good early generic DAG combine too. The motivation comes from reduced bit-hacking for select-of-constants in IR after rL331486. We want to restore that functionality in the DAG as noted in the commit comments for that change and the llvm-dev discussion here: http://lists.llvm.org/pipermail/llvm-dev/2018-July/124433.html The PPC and AArch tests show that those targets are already doing something similar. x86 will be neutral in the minimal case and generally better when this pattern is extended with other ops as shown in the signbit-shift.ll tests. Note the asymmetry: we don't include the (extend (ifneg X)) transform because it already exists in SimplifySelectCC(), and that is verified in the later unchanged tests in the signbit-shift.ll files. Without the 'not' op, the general transform to use a shift is always a win because that's a single instruction. Alive proofs: https://rise4fun.com/Alive/ysli Name: if pos, get -1 %c = icmp sgt i16 %x, -1 %r = sext i1 %c to i16 => %n = xor i16 %x, -1 %r = ashr i16 %n, 15 Name: if pos, get 1 %c = icmp sgt i16 %x, -1 %r = zext i1 %c to i16 => %n = xor i16 %x, -1 %r = lshr i16 %n, 15 Differential Revision: https://reviews.llvm.org/D48970 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@337130 91177308-0d34-0410-b5e6-96231b3b80d8 --- diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp index 5de44428445..2825e45c541 100644 --- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp +++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp @@ -8099,6 +8099,37 @@ static SDValue tryToFoldExtOfLoad(SelectionDAG &DAG, DAGCombiner &Combiner, return SDValue(N, 0); // Return N so it doesn't get rechecked! } +static SDValue foldExtendedSignBitTest(SDNode *N, SelectionDAG &DAG, + bool LegalOperations) { + assert((N->getOpcode() == ISD::SIGN_EXTEND || + N->getOpcode() == ISD::ZERO_EXTEND) && "Expected sext or zext"); + + SDValue SetCC = N->getOperand(0); + if (LegalOperations || SetCC.getOpcode() != ISD::SETCC || + !SetCC.hasOneUse() || SetCC.getValueType() != MVT::i1) + return SDValue(); + + SDValue X = SetCC.getOperand(0); + SDValue Ones = SetCC.getOperand(1); + ISD::CondCode CC = cast(SetCC.getOperand(2))->get(); + EVT VT = N->getValueType(0); + EVT XVT = X.getValueType(); + // setge X, C is canonicalized to setgt, so we do not need to match that + // pattern. The setlt sibling is folded in SimplifySelectCC() becaus it does + // not require the 'not' op. + if (CC == ISD::SETGT && isAllOnesConstant(Ones) && VT == XVT) { + // Invert and smear/shift the sign bit: + // sext i1 (setgt iN X, -1) --> sra (not X), (N - 1) + // zext i1 (setgt iN X, -1) --> srl (not X), (N - 1) + SDLoc DL(N); + SDValue NotX = DAG.getNOT(DL, X, VT); + SDValue ShiftAmount = DAG.getConstant(VT.getSizeInBits() - 1, DL, VT); + auto ShiftOpcode = N->getOpcode() == ISD::SIGN_EXTEND ? ISD::SRA : ISD::SRL; + return DAG.getNode(ShiftOpcode, DL, VT, NotX, ShiftAmount); + } + return SDValue(); +} + SDValue DAGCombiner::visitSIGN_EXTEND(SDNode *N) { SDValue N0 = N->getOperand(0); EVT VT = N->getValueType(0); @@ -8224,6 +8255,9 @@ SDValue DAGCombiner::visitSIGN_EXTEND(SDNode *N) { } } + if (SDValue V = foldExtendedSignBitTest(N, DAG, LegalOperations)) + return V; + if (N0.getOpcode() == ISD::SETCC) { SDValue N00 = N0.getOperand(0); SDValue N01 = N0.getOperand(1); @@ -8510,6 +8544,9 @@ SDValue DAGCombiner::visitZERO_EXTEND(SDNode *N) { DAG, *this, TLI, VT, LegalOperations, N, N0, ISD::ZEXTLOAD)) return foldedExt; + if (SDValue V = foldExtendedSignBitTest(N, DAG, LegalOperations)) + return V; + if (N0.getOpcode() == ISD::SETCC) { // Only do this before legalize for now. if (!LegalOperations && VT.isVector() && diff --git a/test/CodeGen/AArch64/signbit-shift.ll b/test/CodeGen/AArch64/signbit-shift.ll index f9300adeaaf..4d6dff4961b 100644 --- a/test/CodeGen/AArch64/signbit-shift.ll +++ b/test/CodeGen/AArch64/signbit-shift.ll @@ -6,8 +6,8 @@ define i32 @zext_ifpos(i32 %x) { ; CHECK-LABEL: zext_ifpos: ; CHECK: // %bb.0: -; CHECK-NEXT: lsr w8, w0, #31 -; CHECK-NEXT: eor w0, w8, #0x1 +; CHECK-NEXT: mvn w8, w0 +; CHECK-NEXT: lsr w0, w8, #31 ; CHECK-NEXT: ret %c = icmp sgt i32 %x, -1 %e = zext i1 %c to i32 @@ -17,8 +17,8 @@ define i32 @zext_ifpos(i32 %x) { define i32 @add_zext_ifpos(i32 %x) { ; CHECK-LABEL: add_zext_ifpos: ; CHECK: // %bb.0: -; CHECK-NEXT: lsr w8, w0, #31 -; CHECK-NEXT: eor w8, w8, #0x1 +; CHECK-NEXT: mvn w8, w0 +; CHECK-NEXT: lsr w8, w8, #31 ; CHECK-NEXT: add w0, w8, #41 // =41 ; CHECK-NEXT: ret %c = icmp sgt i32 %x, -1 @@ -42,8 +42,8 @@ define i32 @sel_ifpos_tval_bigger(i32 %x) { define i32 @sext_ifpos(i32 %x) { ; CHECK-LABEL: sext_ifpos: ; CHECK: // %bb.0: -; CHECK-NEXT: mov w8, #-1 -; CHECK-NEXT: eor w0, w8, w0, asr #31 +; CHECK-NEXT: mvn w8, w0 +; CHECK-NEXT: asr w0, w8, #31 ; CHECK-NEXT: ret %c = icmp sgt i32 %x, -1 %e = sext i1 %c to i32 @@ -53,10 +53,9 @@ define i32 @sext_ifpos(i32 %x) { define i32 @add_sext_ifpos(i32 %x) { ; CHECK-LABEL: add_sext_ifpos: ; CHECK: // %bb.0: -; CHECK-NEXT: lsr w8, w0, #31 -; CHECK-NEXT: eor w8, w8, #0x1 +; CHECK-NEXT: mvn w8, w0 ; CHECK-NEXT: mov w9, #42 -; CHECK-NEXT: sub w0, w9, w8 +; CHECK-NEXT: sub w0, w9, w8, lsr #31 ; CHECK-NEXT: ret %c = icmp sgt i32 %x, -1 %e = sext i1 %c to i32 diff --git a/test/CodeGen/AMDGPU/commute-compares.ll b/test/CodeGen/AMDGPU/commute-compares.ll index e12a968a987..20d1ea5ac54 100644 --- a/test/CodeGen/AMDGPU/commute-compares.ll +++ b/test/CodeGen/AMDGPU/commute-compares.ll @@ -114,7 +114,7 @@ define amdgpu_kernel void @commute_ule_64_i32(i32 addrspace(1)* %out, i32 addrsp } ; GCN-LABEL: {{^}}commute_sgt_neg1_i32: -; GCN: v_cmp_lt_i32_e32 vcc, -1, v{{[0-9]+}} +; GCN: v_ashrrev_i32_e32 v2, 31, v2 define amdgpu_kernel void @commute_sgt_neg1_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #1 { %tid = call i32 @llvm.amdgcn.workitem.id.x() #0 %gep.in = getelementptr i32, i32 addrspace(1)* %in, i32 %tid diff --git a/test/CodeGen/PowerPC/signbit-shift.ll b/test/CodeGen/PowerPC/signbit-shift.ll index a64e71bbf74..6d7897d83e5 100644 --- a/test/CodeGen/PowerPC/signbit-shift.ll +++ b/test/CodeGen/PowerPC/signbit-shift.ll @@ -17,8 +17,8 @@ define i32 @zext_ifpos(i32 %x) { define i32 @add_zext_ifpos(i32 %x) { ; CHECK-LABEL: add_zext_ifpos: ; CHECK: # %bb.0: +; CHECK-NEXT: nor 3, 3, 3 ; CHECK-NEXT: srwi 3, 3, 31 -; CHECK-NEXT: xori 3, 3, 1 ; CHECK-NEXT: addi 3, 3, 41 ; CHECK-NEXT: blr %c = icmp sgt i32 %x, -1 @@ -54,8 +54,8 @@ define i32 @sext_ifpos(i32 %x) { define i32 @add_sext_ifpos(i32 %x) { ; CHECK-LABEL: add_sext_ifpos: ; CHECK: # %bb.0: -; CHECK-NEXT: srawi 3, 3, 31 ; CHECK-NEXT: nor 3, 3, 3 +; CHECK-NEXT: srawi 3, 3, 31 ; CHECK-NEXT: addi 3, 3, 42 ; CHECK-NEXT: blr %c = icmp sgt i32 %x, -1 diff --git a/test/CodeGen/PowerPC/testComparesigesll.ll b/test/CodeGen/PowerPC/testComparesigesll.ll index 30efe3da3e9..0926d9eb212 100644 --- a/test/CodeGen/PowerPC/testComparesigesll.ll +++ b/test/CodeGen/PowerPC/testComparesigesll.ll @@ -99,9 +99,9 @@ define void @test_igesll_z_store(i64 %a) { ; CHECK-LABEL: test_igesll_z_store: ; CHECK: # %bb.0: # %entry ; CHECK-NEXT: addis r4, r2, .LC0@toc@ha -; CHECK-NEXT: rldicl r3, r3, 1, 63 +; CHECK-NEXT: not r3, r3 ; CHECK-NEXT: ld r4, .LC0@toc@l(r4) -; CHECK-NEXT: xori r3, r3, 1 +; CHECK-NEXT: rldicl r3, r3, 1, 63 ; CHECK-NEXT: std r3, 0(r4) ; CHECK-NEXT: blr entry: @@ -115,9 +115,9 @@ define void @test_igesll_sext_z_store(i64 %a) { ; CHECK-LABEL: test_igesll_sext_z_store: ; CHECK: # %bb.0: # %entry ; CHECK-NEXT: addis r4, r2, .LC0@toc@ha -; CHECK-NEXT: sradi r3, r3, 63 -; CHECK-NEXT: ld r4, .LC0@toc@l(r4) ; CHECK-NEXT: not r3, r3 +; CHECK-NEXT: ld r4, .LC0@toc@l(r4) +; CHECK-NEXT: sradi r3, r3, 63 ; CHECK-NEXT: std r3, ; CHECK-NEXT: blr entry: diff --git a/test/CodeGen/PowerPC/testComparesllgesll.ll b/test/CodeGen/PowerPC/testComparesllgesll.ll index 6fb53977a55..f2096c28ee2 100644 --- a/test/CodeGen/PowerPC/testComparesllgesll.ll +++ b/test/CodeGen/PowerPC/testComparesllgesll.ll @@ -39,8 +39,8 @@ entry: define i64 @test_llgesll_z(i64 %a) { ; CHECK-LABEL: test_llgesll_z: ; CHECK: # %bb.0: # %entry +; CHECK-NEXT: not r3, r3 ; CHECK-NEXT: rldicl r3, r3, 1, 63 -; CHECK-NEXT: xori r3, r3, 1 ; CHECK-NEXT: blr entry: %cmp = icmp sgt i64 %a, -1 @@ -51,8 +51,8 @@ entry: define i64 @test_llgesll_sext_z(i64 %a) { ; CHECK-LABEL: test_llgesll_sext_z: ; CHECK: # %bb.0: # %entry -; CHECK-NEXT: sradi r3, r3, 63 ; CHECK-NEXT: not r3, r3 +; CHECK-NEXT: sradi r3, r3, 63 ; CHECK-NEXT: blr entry: %cmp = icmp sgt i64 %a, -1 @@ -99,9 +99,9 @@ define void @test_llgesll_z_store(i64 %a) { ; CHECK-LABEL: test_llgesll_z_store: ; CHECK: # %bb.0: # %entry ; CHECK-NEXT: addis r4, r2, .LC0@toc@ha -; CHECK-NEXT: rldicl r3, r3, 1, 63 +; CHECK-NEXT: not r3, r3 ; CHECK-NEXT: ld r4, .LC0@toc@l(r4) -; CHECK-NEXT: xori r3, r3, 1 +; CHECK-NEXT: rldicl r3, r3, 1, 63 ; CHECK-NEXT: std r3, 0(r4) ; CHECK-NEXT: blr entry: @@ -115,9 +115,9 @@ define void @test_llgesll_sext_z_store(i64 %a) { ; CHECK-LABEL: test_llgesll_sext_z_store: ; CHECK: # %bb.0: # %entry ; CHECK-NEXT: addis r4, r2, .LC0@toc@ha -; CHECK-NEXT: sradi r3, r3, 63 -; CHECK-NEXT: ld r4, .LC0@toc@l(r4) ; CHECK-NEXT: not r3, r3 +; CHECK-NEXT: ld r4, .LC0@toc@l(r4) +; CHECK-NEXT: sradi r3, r3, 63 ; CHECK-NEXT: std r3, 0(r4) ; CHECK-NEXT: blr entry: diff --git a/test/CodeGen/X86/select.ll b/test/CodeGen/X86/select.ll index 4fc88d56a67..7d518ab8751 100644 --- a/test/CodeGen/X86/select.ll +++ b/test/CodeGen/X86/select.ll @@ -297,11 +297,10 @@ define x86_fp80 @test7(i32 %tmp8) nounwind { ; ; MCU-LABEL: test7: ; MCU: # %bb.0: -; MCU-NEXT: xorl %ecx, %ecx -; MCU-NEXT: testl %eax, %eax -; MCU-NEXT: setns %cl -; MCU-NEXT: shll $4, %ecx -; MCU-NEXT: fldt {{\.LCPI.*}}(%ecx) +; MCU-NEXT: notl %eax +; MCU-NEXT: shrl $27, %eax +; MCU-NEXT: andl $-16, %eax +; MCU-NEXT: fldt {{\.LCPI.*}}(%eax) ; MCU-NEXT: retl %tmp9 = icmp sgt i32 %tmp8, -1 %retval = select i1 %tmp9, x86_fp80 0xK4005B400000000000000, x86_fp80 0xK40078700000000000000 diff --git a/test/CodeGen/X86/selectcc-to-shiftand.ll b/test/CodeGen/X86/selectcc-to-shiftand.ll index 833ac1dfe2d..2c7fb197e11 100644 --- a/test/CodeGen/X86/selectcc-to-shiftand.ll +++ b/test/CodeGen/X86/selectcc-to-shiftand.ll @@ -93,18 +93,18 @@ define i32 @not_pos_sel_same_variable(i32 %a) { define i32 @pos_sel_constants(i32 %a) { ; CHECK-NOBMI-LABEL: pos_sel_constants: ; CHECK-NOBMI: # %bb.0: -; CHECK-NOBMI-NEXT: xorl %eax, %eax -; CHECK-NOBMI-NEXT: testl %edi, %edi -; CHECK-NOBMI-NEXT: setns %al -; CHECK-NOBMI-NEXT: leal (%rax,%rax,4), %eax +; CHECK-NOBMI-NEXT: # kill: def $edi killed $edi def $rdi +; CHECK-NOBMI-NEXT: notl %edi +; CHECK-NOBMI-NEXT: shrl $31, %edi +; CHECK-NOBMI-NEXT: leal (%rdi,%rdi,4), %eax ; CHECK-NOBMI-NEXT: retq ; ; CHECK-BMI-LABEL: pos_sel_constants: ; CHECK-BMI: # %bb.0: -; CHECK-BMI-NEXT: xorl %eax, %eax -; CHECK-BMI-NEXT: testl %edi, %edi -; CHECK-BMI-NEXT: setns %al -; CHECK-BMI-NEXT: leal (%rax,%rax,4), %eax +; CHECK-BMI-NEXT: # kill: def $edi killed $edi def $rdi +; CHECK-BMI-NEXT: notl %edi +; CHECK-BMI-NEXT: shrl $31, %edi +; CHECK-BMI-NEXT: leal (%rdi,%rdi,4), %eax ; CHECK-BMI-NEXT: retq %tmp.1 = icmp sgt i32 %a, -1 %retval = select i1 %tmp.1, i32 5, i32 0 @@ -116,18 +116,18 @@ define i32 @pos_sel_constants(i32 %a) { define i32 @pos_sel_special_constant(i32 %a) { ; CHECK-NOBMI-LABEL: pos_sel_special_constant: ; CHECK-NOBMI: # %bb.0: -; CHECK-NOBMI-NEXT: xorl %eax, %eax -; CHECK-NOBMI-NEXT: testl %edi, %edi -; CHECK-NOBMI-NEXT: setns %al -; CHECK-NOBMI-NEXT: shll $9, %eax +; CHECK-NOBMI-NEXT: notl %edi +; CHECK-NOBMI-NEXT: shrl $22, %edi +; CHECK-NOBMI-NEXT: andl $512, %edi # imm = 0x200 +; CHECK-NOBMI-NEXT: movl %edi, %eax ; CHECK-NOBMI-NEXT: retq ; ; CHECK-BMI-LABEL: pos_sel_special_constant: ; CHECK-BMI: # %bb.0: -; CHECK-BMI-NEXT: xorl %eax, %eax -; CHECK-BMI-NEXT: testl %edi, %edi -; CHECK-BMI-NEXT: setns %al -; CHECK-BMI-NEXT: shll $9, %eax +; CHECK-BMI-NEXT: notl %edi +; CHECK-BMI-NEXT: shrl $22, %edi +; CHECK-BMI-NEXT: andl $512, %edi # imm = 0x200 +; CHECK-BMI-NEXT: movl %edi, %eax ; CHECK-BMI-NEXT: retq %tmp.1 = icmp sgt i32 %a, -1 %retval = select i1 %tmp.1, i32 512, i32 0 diff --git a/test/CodeGen/X86/signbit-shift.ll b/test/CodeGen/X86/signbit-shift.ll index d974187126d..743664fd506 100644 --- a/test/CodeGen/X86/signbit-shift.ll +++ b/test/CodeGen/X86/signbit-shift.ll @@ -6,9 +6,9 @@ define i32 @zext_ifpos(i32 %x) { ; CHECK-LABEL: zext_ifpos: ; CHECK: # %bb.0: -; CHECK-NEXT: xorl %eax, %eax -; CHECK-NEXT: testl %edi, %edi -; CHECK-NEXT: setns %al +; CHECK-NEXT: notl %edi +; CHECK-NEXT: shrl $31, %edi +; CHECK-NEXT: movl %edi, %eax ; CHECK-NEXT: retq %c = icmp sgt i32 %x, -1 %e = zext i1 %c to i32 @@ -18,10 +18,10 @@ define i32 @zext_ifpos(i32 %x) { define i32 @add_zext_ifpos(i32 %x) { ; CHECK-LABEL: add_zext_ifpos: ; CHECK: # %bb.0: -; CHECK-NEXT: xorl %eax, %eax -; CHECK-NEXT: testl %edi, %edi -; CHECK-NEXT: setns %al -; CHECK-NEXT: addl $41, %eax +; CHECK-NEXT: # kill: def $edi killed $edi def $rdi +; CHECK-NEXT: notl %edi +; CHECK-NEXT: shrl $31, %edi +; CHECK-NEXT: leal 41(%rdi), %eax ; CHECK-NEXT: retq %c = icmp sgt i32 %x, -1 %e = zext i1 %c to i32 @@ -32,10 +32,10 @@ define i32 @add_zext_ifpos(i32 %x) { define i32 @sel_ifpos_tval_bigger(i32 %x) { ; CHECK-LABEL: sel_ifpos_tval_bigger: ; CHECK: # %bb.0: -; CHECK-NEXT: xorl %eax, %eax -; CHECK-NEXT: testl %edi, %edi -; CHECK-NEXT: setns %al -; CHECK-NEXT: addl $41, %eax +; CHECK-NEXT: # kill: def $edi killed $edi def $rdi +; CHECK-NEXT: notl %edi +; CHECK-NEXT: shrl $31, %edi +; CHECK-NEXT: leal 41(%rdi), %eax ; CHECK-NEXT: retq %c = icmp sgt i32 %x, -1 %r = select i1 %c, i32 42, i32 41 @@ -45,10 +45,9 @@ define i32 @sel_ifpos_tval_bigger(i32 %x) { define i32 @sext_ifpos(i32 %x) { ; CHECK-LABEL: sext_ifpos: ; CHECK: # %bb.0: -; CHECK-NEXT: xorl %eax, %eax -; CHECK-NEXT: testl %edi, %edi -; CHECK-NEXT: setns %al -; CHECK-NEXT: negl %eax +; CHECK-NEXT: notl %edi +; CHECK-NEXT: sarl $31, %edi +; CHECK-NEXT: movl %edi, %eax ; CHECK-NEXT: retq %c = icmp sgt i32 %x, -1 %e = sext i1 %c to i32 @@ -58,11 +57,10 @@ define i32 @sext_ifpos(i32 %x) { define i32 @add_sext_ifpos(i32 %x) { ; CHECK-LABEL: add_sext_ifpos: ; CHECK: # %bb.0: -; CHECK-NEXT: xorl %ecx, %ecx -; CHECK-NEXT: testl %edi, %edi -; CHECK-NEXT: setns %cl +; CHECK-NEXT: notl %edi +; CHECK-NEXT: shrl $31, %edi ; CHECK-NEXT: movl $42, %eax -; CHECK-NEXT: subl %ecx, %eax +; CHECK-NEXT: subl %edi, %eax ; CHECK-NEXT: retq %c = icmp sgt i32 %x, -1 %e = sext i1 %c to i32 @@ -73,11 +71,10 @@ define i32 @add_sext_ifpos(i32 %x) { define i32 @sel_ifpos_fval_bigger(i32 %x) { ; CHECK-LABEL: sel_ifpos_fval_bigger: ; CHECK: # %bb.0: -; CHECK-NEXT: xorl %ecx, %ecx -; CHECK-NEXT: testl %edi, %edi -; CHECK-NEXT: setns %cl +; CHECK-NEXT: notl %edi +; CHECK-NEXT: shrl $31, %edi ; CHECK-NEXT: movl $42, %eax -; CHECK-NEXT: subl %ecx, %eax +; CHECK-NEXT: subl %edi, %eax ; CHECK-NEXT: retq %c = icmp sgt i32 %x, -1 %r = select i1 %c, i32 41, i32 42 diff --git a/test/CodeGen/XCore/ashr.ll b/test/CodeGen/XCore/ashr.ll index 78cb1440cc0..f4247dc2a08 100644 --- a/test/CodeGen/XCore/ashr.ll +++ b/test/CodeGen/XCore/ashr.ll @@ -72,5 +72,6 @@ define i32 @f5(i32 %a) nounwind { ret i32 %2 } ; CHECK-LABEL: f5: -; CHECK-NEXT: ashr r0, r0, 32 -; CHECK-NEXT: eq r0, r0, 0 +; CHECK-NEXT: not r0, r0 +; CHECK-NEXT: mkmsk r1, 5 +; CHECK-NEXT: shr r0, r0, r1