diff options
author | Sanjay Patel <spatel@rotateright.com> | 2017-03-04 19:18:09 +0000 |
---|---|---|
committer | Sanjay Patel <spatel@rotateright.com> | 2017-03-04 19:18:09 +0000 |
commit | 59071e49a930064569e5953fb849954c08e8d18a (patch) | |
tree | 0422a861a44c9e4b4814cacc0c3ee7740f1a2579 | |
parent | 6a33d44017b42b5ea309b3c92315ee708a06d142 (diff) |
[DAGCombiner] allow transforming (select Cond, C +/- 1, C) to (add(ext Cond), C)
select Cond, C +/- 1, C --> add(ext Cond), C -- with a target hook.
This is part of the ongoing process to obsolete D24480. The motivation is to
canonicalize to select IR in InstCombine whenever possible, so we need to have a way to
undo that easily in codegen.
PowerPC is an obvious winner for this kind of transform because it has fast and complete
bit-twiddling abilities but generally lousy conditional execution perf (although this might
have changed in recent implementations).
x86 also sees some wins, but the effect is limited because these transforms already mostly
exist in its target-specific combineSelectOfTwoConstants(). The fact that we see any x86
changes just shows that that code is a mess of special-case holes. We may be able to remove
some of that logic now.
My guess is that other targets will want to enable this hook for most cases. The likely
follow-ups would be to add value type and/or the constants themselves as parameters for the
hook. As the tests in select_const.ll show, we can transform any select-of-constants to
math/logic, but the general transform for any 2 constants needs one more instruction
(multiply or 'and').
ARM is one target that I think may not want this for most cases. I see infinite loops there
because it wants to use selects to enable conditionally executed instructions.
Differential Revision: https://reviews.llvm.org/D30537
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@296977 91177308-0d34-0410-b5e6-96231b3b80d8
-rw-r--r-- | include/llvm/Target/TargetLowering.h | 7 | ||||
-rw-r--r-- | lib/CodeGen/SelectionDAG/DAGCombiner.cpp | 21 | ||||
-rw-r--r-- | lib/Target/PowerPC/PPCISelLowering.h | 4 | ||||
-rw-r--r-- | lib/Target/X86/X86ISelLowering.h | 4 | ||||
-rw-r--r-- | test/CodeGen/PowerPC/crbits.ll | 2 | ||||
-rw-r--r-- | test/CodeGen/PowerPC/select_const.ll | 180 | ||||
-rw-r--r-- | test/CodeGen/X86/avx512-insert-extract.ll | 10 | ||||
-rw-r--r-- | test/CodeGen/X86/avx512-select.ll | 5 | ||||
-rw-r--r-- | test/CodeGen/X86/select_const.ll | 19 |
9 files changed, 84 insertions, 168 deletions
diff --git a/include/llvm/Target/TargetLowering.h b/include/llvm/Target/TargetLowering.h index ab91413e764..b03d35a00b2 100644 --- a/include/llvm/Target/TargetLowering.h +++ b/include/llvm/Target/TargetLowering.h @@ -1388,6 +1388,13 @@ public: Action != TypeSplitVector; } + /// Return true if a select of constants (select Cond, C1, C2) should be + /// transformed into simple math ops with the condition value. For example: + /// select Cond, C1, C1-1 --> add (zext Cond), C1-1 + virtual bool convertSelectOfConstantsToMath() const { + return false; + } + //===--------------------------------------------------------------------===// // TargetLowering Configuration Methods - These methods should be invoked by // the derived class constructor to configure this object for the target. diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp index 8ac41197b62..5b29f18c79b 100644 --- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp +++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp @@ -5725,8 +5725,6 @@ SDValue DAGCombiner::foldSelectOfConstants(SDNode *N) { // transforms in the other direction (create a select from a zext/sext). There // is also a target-independent combine here in DAGCombiner in the other // direction for (select Cond, -1, 0) when the condition is not i1. - // TODO: This could be generalized for any 2 constants that differ by 1: - // add ({s/z}ext Cond), C if (CondVT == MVT::i1 && !LegalOperations) { if (C1->isNullValue() && C2->isOne()) { // select Cond, 0, 1 --> zext (!Cond) @@ -5754,6 +5752,25 @@ SDValue DAGCombiner::foldSelectOfConstants(SDNode *N) { Cond = DAG.getNode(ISD::SIGN_EXTEND, DL, VT, Cond); return Cond; } + + // For any constants that differ by 1, we can transform the select into an + // extend and add. Use a target hook because some targets may prefer to + // transform in the other direction. + if (TLI.convertSelectOfConstantsToMath()) { + if (C1->getAPIntValue() - 1 == C2->getAPIntValue()) { + // select Cond, C1, C1-1 --> add (zext Cond), C1-1 + if (VT != MVT::i1) + Cond = DAG.getNode(ISD::ZERO_EXTEND, DL, VT, Cond); + return DAG.getNode(ISD::ADD, DL, VT, Cond, N2); + } + if (C1->getAPIntValue() + 1 == C2->getAPIntValue()) { + // select Cond, C1, C1+1 --> add (sext Cond), C1+1 + if (VT != MVT::i1) + Cond = DAG.getNode(ISD::SIGN_EXTEND, DL, VT, Cond); + return DAG.getNode(ISD::ADD, DL, VT, Cond, N2); + } + } + return SDValue(); } diff --git a/lib/Target/PowerPC/PPCISelLowering.h b/lib/Target/PowerPC/PPCISelLowering.h index d12cd4ab1f3..caaeaa97b0f 100644 --- a/lib/Target/PowerPC/PPCISelLowering.h +++ b/lib/Target/PowerPC/PPCISelLowering.h @@ -711,6 +711,10 @@ namespace llvm { bool shouldConvertConstantLoadToIntImm(const APInt &Imm, Type *Ty) const override; + bool convertSelectOfConstantsToMath() const override { + return true; + } + bool isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const override; bool getTgtMemIntrinsic(IntrinsicInfo &Info, diff --git a/lib/Target/X86/X86ISelLowering.h b/lib/Target/X86/X86ISelLowering.h index 14c5531a51b..d9acc1357ec 100644 --- a/lib/Target/X86/X86ISelLowering.h +++ b/lib/Target/X86/X86ISelLowering.h @@ -991,6 +991,10 @@ namespace llvm { bool shouldConvertConstantLoadToIntImm(const APInt &Imm, Type *Ty) const override; + bool convertSelectOfConstantsToMath() const override { + return true; + } + /// Return true if EXTRACT_SUBVECTOR is cheap for this result type /// with this index. bool isExtractSubvectorCheap(EVT ResVT, unsigned Index) const override; diff --git a/test/CodeGen/PowerPC/crbits.ll b/test/CodeGen/PowerPC/crbits.ll index b894a361d26..a85237195c5 100644 --- a/test/CodeGen/PowerPC/crbits.ll +++ b/test/CodeGen/PowerPC/crbits.ll @@ -142,7 +142,7 @@ entry: ret i32 %cond ; CHECK-LABEL: @exttest7 -; CHECK-DAG: cmplwi {{[0-9]+}}, 3, 5 +; CHECK-DAG: cmpwi {{[0-9]+}}, 3, 5 ; CHECK-DAG: li [[REG1:[0-9]+]], 8 ; CHECK-DAG: li [[REG2:[0-9]+]], 7 ; CHECK: isel 3, [[REG2]], [[REG1]], diff --git a/test/CodeGen/PowerPC/select_const.ll b/test/CodeGen/PowerPC/select_const.ll index 142a482bcc1..29548123be8 100644 --- a/test/CodeGen/PowerPC/select_const.ll +++ b/test/CodeGen/PowerPC/select_const.ll @@ -218,70 +218,29 @@ define i32 @select_neg1_or_0_signext(i1 signext %cond) { ; select Cond, C+1, C --> add (zext Cond), C define i32 @select_Cplus1_C(i1 %cond) { -; ISEL-LABEL: select_Cplus1_C: -; ISEL: # BB#0: -; ISEL-NEXT: andi. 3, 3, 1 -; ISEL-NEXT: li 4, 42 -; ISEL-NEXT: li 3, 41 -; ISEL-NEXT: isel 3, 4, 3, 1 -; ISEL-NEXT: blr -; -; NO_ISEL-LABEL: select_Cplus1_C: -; NO_ISEL: # BB#0: -; NO_ISEL-NEXT: andi. 3, 3, 1 -; NO_ISEL-NEXT: li 4, 42 -; NO_ISEL-NEXT: li 3, 41 -; NO_ISEL-NEXT: bc 12, 1, .LBB12_1 -; NO_ISEL-NEXT: blr -; NO_ISEL-NEXT: .LBB12_1: -; NO_ISEL-NEXT: addi 3, 4, 0 -; NO_ISEL-NEXT: blr +; ALL-LABEL: select_Cplus1_C: +; ALL: # BB#0: +; ALL-NEXT: clrldi 3, 3, 63 +; ALL-NEXT: addi 3, 3, 41 +; ALL-NEXT: blr %sel = select i1 %cond, i32 42, i32 41 ret i32 %sel } define i32 @select_Cplus1_C_zeroext(i1 zeroext %cond) { -; ISEL-LABEL: select_Cplus1_C_zeroext: -; ISEL: # BB#0: -; ISEL-NEXT: andi. 3, 3, 1 -; ISEL-NEXT: li 4, 42 -; ISEL-NEXT: li 3, 41 -; ISEL-NEXT: isel 3, 4, 3, 1 -; ISEL-NEXT: blr -; -; NO_ISEL-LABEL: select_Cplus1_C_zeroext: -; NO_ISEL: # BB#0: -; NO_ISEL-NEXT: andi. 3, 3, 1 -; NO_ISEL-NEXT: li 4, 42 -; NO_ISEL-NEXT: li 3, 41 -; NO_ISEL-NEXT: bc 12, 1, .LBB13_1 -; NO_ISEL-NEXT: blr -; NO_ISEL-NEXT: .LBB13_1: -; NO_ISEL-NEXT: addi 3, 4, 0 -; NO_ISEL-NEXT: blr +; ALL-LABEL: select_Cplus1_C_zeroext: +; ALL: # BB#0: +; ALL-NEXT: addi 3, 3, 41 +; ALL-NEXT: blr %sel = select i1 %cond, i32 42, i32 41 ret i32 %sel } define i32 @select_Cplus1_C_signext(i1 signext %cond) { -; ISEL-LABEL: select_Cplus1_C_signext: -; ISEL: # BB#0: -; ISEL-NEXT: andi. 3, 3, 1 -; ISEL-NEXT: li 4, 42 -; ISEL-NEXT: li 3, 41 -; ISEL-NEXT: isel 3, 4, 3, 1 -; ISEL-NEXT: blr -; -; NO_ISEL-LABEL: select_Cplus1_C_signext: -; NO_ISEL: # BB#0: -; NO_ISEL-NEXT: andi. 3, 3, 1 -; NO_ISEL-NEXT: li 4, 42 -; NO_ISEL-NEXT: li 3, 41 -; NO_ISEL-NEXT: bc 12, 1, .LBB14_1 -; NO_ISEL-NEXT: blr -; NO_ISEL-NEXT: .LBB14_1: -; NO_ISEL-NEXT: addi 3, 4, 0 -; NO_ISEL-NEXT: blr +; ALL-LABEL: select_Cplus1_C_signext: +; ALL: # BB#0: +; ALL-NEXT: subfic 3, 3, 41 +; ALL-NEXT: blr %sel = select i1 %cond, i32 42, i32 41 ret i32 %sel } @@ -289,70 +248,29 @@ define i32 @select_Cplus1_C_signext(i1 signext %cond) { ; select Cond, C, C+1 --> add (sext Cond), C define i32 @select_C_Cplus1(i1 %cond) { -; ISEL-LABEL: select_C_Cplus1: -; ISEL: # BB#0: -; ISEL-NEXT: andi. 3, 3, 1 -; ISEL-NEXT: li 4, 41 -; ISEL-NEXT: li 3, 42 -; ISEL-NEXT: isel 3, 4, 3, 1 -; ISEL-NEXT: blr -; -; NO_ISEL-LABEL: select_C_Cplus1: -; NO_ISEL: # BB#0: -; NO_ISEL-NEXT: andi. 3, 3, 1 -; NO_ISEL-NEXT: li 4, 41 -; NO_ISEL-NEXT: li 3, 42 -; NO_ISEL-NEXT: bc 12, 1, .LBB15_1 -; NO_ISEL-NEXT: blr -; NO_ISEL-NEXT: .LBB15_1: -; NO_ISEL-NEXT: addi 3, 4, 0 -; NO_ISEL-NEXT: blr +; ALL-LABEL: select_C_Cplus1: +; ALL: # BB#0: +; ALL-NEXT: clrldi 3, 3, 63 +; ALL-NEXT: subfic 3, 3, 42 +; ALL-NEXT: blr %sel = select i1 %cond, i32 41, i32 42 ret i32 %sel } define i32 @select_C_Cplus1_zeroext(i1 zeroext %cond) { -; ISEL-LABEL: select_C_Cplus1_zeroext: -; ISEL: # BB#0: -; ISEL-NEXT: andi. 3, 3, 1 -; ISEL-NEXT: li 4, 41 -; ISEL-NEXT: li 3, 42 -; ISEL-NEXT: isel 3, 4, 3, 1 -; ISEL-NEXT: blr -; -; NO_ISEL-LABEL: select_C_Cplus1_zeroext: -; NO_ISEL: # BB#0: -; NO_ISEL-NEXT: andi. 3, 3, 1 -; NO_ISEL-NEXT: li 4, 41 -; NO_ISEL-NEXT: li 3, 42 -; NO_ISEL-NEXT: bc 12, 1, .LBB16_1 -; NO_ISEL-NEXT: blr -; NO_ISEL-NEXT: .LBB16_1: -; NO_ISEL-NEXT: addi 3, 4, 0 -; NO_ISEL-NEXT: blr +; ALL-LABEL: select_C_Cplus1_zeroext: +; ALL: # BB#0: +; ALL-NEXT: subfic 3, 3, 42 +; ALL-NEXT: blr %sel = select i1 %cond, i32 41, i32 42 ret i32 %sel } define i32 @select_C_Cplus1_signext(i1 signext %cond) { -; ISEL-LABEL: select_C_Cplus1_signext: -; ISEL: # BB#0: -; ISEL-NEXT: andi. 3, 3, 1 -; ISEL-NEXT: li 4, 41 -; ISEL-NEXT: li 3, 42 -; ISEL-NEXT: isel 3, 4, 3, 1 -; ISEL-NEXT: blr -; -; NO_ISEL-LABEL: select_C_Cplus1_signext: -; NO_ISEL: # BB#0: -; NO_ISEL-NEXT: andi. 3, 3, 1 -; NO_ISEL-NEXT: li 4, 41 -; NO_ISEL-NEXT: li 3, 42 -; NO_ISEL-NEXT: bc 12, 1, .LBB17_1 -; NO_ISEL-NEXT: blr -; NO_ISEL-NEXT: .LBB17_1: -; NO_ISEL-NEXT: addi 3, 4, 0 -; NO_ISEL-NEXT: blr +; ALL-LABEL: select_C_Cplus1_signext: +; ALL: # BB#0: +; ALL-NEXT: addi 3, 3, 42 +; ALL-NEXT: blr %sel = select i1 %cond, i32 41, i32 42 ret i32 %sel } @@ -583,48 +501,22 @@ define i8 @sel_constants_srem_constant(i1 %cond) { } define i8 @sel_constants_urem_constant(i1 %cond) { -; ISEL-LABEL: sel_constants_urem_constant: -; ISEL: # BB#0: -; ISEL-NEXT: andi. 3, 3, 1 -; ISEL-NEXT: li 4, 2 -; ISEL-NEXT: li 3, 3 -; ISEL-NEXT: isel 3, 4, 3, 1 -; ISEL-NEXT: blr -; -; NO_ISEL-LABEL: sel_constants_urem_constant: -; NO_ISEL: # BB#0: -; NO_ISEL-NEXT: andi. 3, 3, 1 -; NO_ISEL-NEXT: li 4, 2 -; NO_ISEL-NEXT: li 3, 3 -; NO_ISEL-NEXT: bc 12, 1, .LBB27_1 -; NO_ISEL-NEXT: blr -; NO_ISEL-NEXT: .LBB27_1: -; NO_ISEL-NEXT: addi 3, 4, 0 -; NO_ISEL-NEXT: blr +; ALL-LABEL: sel_constants_urem_constant: +; ALL: # BB#0: +; ALL-NEXT: rlwinm 3, 3, 0, 31, 31 +; ALL-NEXT: subfic 3, 3, 3 +; ALL-NEXT: blr %sel = select i1 %cond, i8 -4, i8 23 %bo = urem i8 %sel, 5 ret i8 %bo } define i8 @sel_constants_and_constant(i1 %cond) { -; ISEL-LABEL: sel_constants_and_constant: -; ISEL: # BB#0: -; ISEL-NEXT: andi. 3, 3, 1 -; ISEL-NEXT: li 4, 4 -; ISEL-NEXT: li 3, 5 -; ISEL-NEXT: isel 3, 4, 3, 1 -; ISEL-NEXT: blr -; -; NO_ISEL-LABEL: sel_constants_and_constant: -; NO_ISEL: # BB#0: -; NO_ISEL-NEXT: andi. 3, 3, 1 -; NO_ISEL-NEXT: li 4, 4 -; NO_ISEL-NEXT: li 3, 5 -; NO_ISEL-NEXT: bc 12, 1, .LBB28_1 -; NO_ISEL-NEXT: blr -; NO_ISEL-NEXT: .LBB28_1: -; NO_ISEL-NEXT: addi 3, 4, 0 -; NO_ISEL-NEXT: blr +; ALL-LABEL: sel_constants_and_constant: +; ALL: # BB#0: +; ALL-NEXT: rlwinm 3, 3, 0, 31, 31 +; ALL-NEXT: subfic 3, 3, 5 +; ALL-NEXT: blr %sel = select i1 %cond, i8 -4, i8 23 %bo = and i8 %sel, 5 ret i8 %bo diff --git a/test/CodeGen/X86/avx512-insert-extract.ll b/test/CodeGen/X86/avx512-insert-extract.ll index 9373b4d5007..aee4c6367e2 100644 --- a/test/CodeGen/X86/avx512-insert-extract.ll +++ b/test/CodeGen/X86/avx512-insert-extract.ll @@ -1411,10 +1411,8 @@ define zeroext i8 @test_extractelement_v2i1(<2 x i64> %a, <2 x i64> %b) { ; KNL-NEXT: vpxor %xmm2, %xmm1, %xmm1 ; KNL-NEXT: vpxor %xmm2, %xmm0, %xmm0 ; KNL-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm0 -; KNL-NEXT: vmovq %xmm0, %rax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: sete %al -; KNL-NEXT: addb $3, %al +; KNL-NEXT: vpextrb $0, %xmm0, %eax +; KNL-NEXT: addb $4, %al ; KNL-NEXT: movzbl %al, %eax ; KNL-NEXT: retq ; @@ -1527,9 +1525,7 @@ define zeroext i8 @test_extractelement_v64i1(<64 x i8> %a, <64 x i8> %b) { ; KNL-NEXT: vpcmpgtb %ymm2, %ymm0, %ymm0 ; KNL-NEXT: vextracti128 $1, %ymm0, %xmm0 ; KNL-NEXT: vpextrb $15, %xmm0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: sete %al -; KNL-NEXT: addb $3, %al +; KNL-NEXT: addb $4, %al ; KNL-NEXT: movzbl %al, %eax ; KNL-NEXT: retq ; diff --git a/test/CodeGen/X86/avx512-select.ll b/test/CodeGen/X86/avx512-select.ll index 3f427298c17..a2e5a14f78d 100644 --- a/test/CodeGen/X86/avx512-select.ll +++ b/test/CodeGen/X86/avx512-select.ll @@ -149,10 +149,7 @@ define i8 @select07(i8 %a.0, i8 %b.0, i8 %m) { define i64 @pr30249() { ; CHECK-LABEL: pr30249: ; CHECK: ## BB#0: -; CHECK-NEXT: xorl %ecx, %ecx -; CHECK-NEXT: cmpb $1, %cl -; CHECK-NEXT: movl $1, %eax -; CHECK-NEXT: adcxq %rcx, %rax +; CHECK-NEXT: movl $2, %eax ; CHECK-NEXT: retq %v = select i1 undef , i64 1, i64 2 ret i64 %v diff --git a/test/CodeGen/X86/select_const.ll b/test/CodeGen/X86/select_const.ll index 716ee70d926..a97e7c299e7 100644 --- a/test/CodeGen/X86/select_const.ll +++ b/test/CodeGen/X86/select_const.ll @@ -174,10 +174,9 @@ define i32 @select_Cplus1_C_signext(i1 signext %cond) { define i32 @select_C_Cplus1(i1 %cond) { ; CHECK-LABEL: select_C_Cplus1: ; CHECK: # BB#0: -; CHECK-NEXT: andb $1, %dil -; CHECK-NEXT: cmpb $1, %dil -; CHECK-NEXT: movl $41, %eax -; CHECK-NEXT: adcl $0, %eax +; CHECK-NEXT: andl $1, %edi +; CHECK-NEXT: movl $42, %eax +; CHECK-NEXT: subl %edi, %eax ; CHECK-NEXT: retq %sel = select i1 %cond, i32 41, i32 42 ret i32 %sel @@ -186,9 +185,9 @@ define i32 @select_C_Cplus1(i1 %cond) { define i32 @select_C_Cplus1_zeroext(i1 zeroext %cond) { ; CHECK-LABEL: select_C_Cplus1_zeroext: ; CHECK: # BB#0: -; CHECK-NEXT: cmpb $1, %dil -; CHECK-NEXT: movl $41, %eax -; CHECK-NEXT: adcl $0, %eax +; CHECK-NEXT: movzbl %dil, %ecx +; CHECK-NEXT: movl $42, %eax +; CHECK-NEXT: subl %ecx, %eax ; CHECK-NEXT: retq %sel = select i1 %cond, i32 41, i32 42 ret i32 %sel @@ -198,9 +197,9 @@ define i32 @select_C_Cplus1_signext(i1 signext %cond) { ; CHECK-LABEL: select_C_Cplus1_signext: ; CHECK: # BB#0: ; CHECK-NEXT: andb $1, %dil -; CHECK-NEXT: cmpb $1, %dil -; CHECK-NEXT: movl $41, %eax -; CHECK-NEXT: adcl $0, %eax +; CHECK-NEXT: movzbl %dil, %ecx +; CHECK-NEXT: movl $42, %eax +; CHECK-NEXT: subl %ecx, %eax ; CHECK-NEXT: retq %sel = select i1 %cond, i32 41, i32 42 ret i32 %sel |