[GlobalISel] Combine (X == 0) & (Y == 0) -> (X | Y) == 0 #69017
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-backend-aarch64 @llvm/pr-subscribers-llvm-globalisel Author: Dávid Ferenc Szabó (dfszabo) ChangesAlso combine (X != 0) | (X != 0) -> (X | Y) != 0 Patch is 30.87 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/69017.diff 5 Files Affected:
diff --git a/llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h b/llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h
index d8f19c19ee106b6..6be43395232364d 100644
--- a/llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h
+++ b/llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h
@@ -805,6 +805,10 @@ class CombinerHelper {
/// (X ^ Y) != X -> Y != 0
bool matchRedundantBinOpInEquality(MachineInstr &MI, BuildFnTy &MatchInfo);
+ /// Transform: (X == 0 & Y == 0) -> (X | Y) == 0
+ /// (X != 0 | Y != 0) -> (X | Y) != 0
+ bool matchDoubleICmpZeroAndOr(MachineInstr &MI, BuildFnTy &MatchInfo);
+
/// Match shifts greater or equal to the bitwidth of the operation.
bool matchShiftsTooBig(MachineInstr &MI);
diff --git a/llvm/include/llvm/Target/GlobalISel/Combine.td b/llvm/include/llvm/Target/GlobalISel/Combine.td
index 7e0691e1ee95048..d68dfcdd7246a19 100644
--- a/llvm/include/llvm/Target/GlobalISel/Combine.td
+++ b/llvm/include/llvm/Target/GlobalISel/Combine.td
@@ -903,6 +903,12 @@ def redundant_binop_in_equality : GICombineRule<
[{ return Helper.matchRedundantBinOpInEquality(*${root}, ${info}); }]),
(apply [{ Helper.applyBuildFn(*${root}, ${info}); }])>;
+def double_icmp_zero_and_or_combine : GICombineRule<
+ (defs root:$root, build_fn_matchinfo:$info),
+ (match (wip_match_opcode G_AND, G_OR):$root,
+ [{ return Helper.matchDoubleICmpZeroAndOr(*${root}, ${info}); }]),
+ (apply [{ Helper.applyBuildFn(*${root}, ${info}); }])>;
+
def and_or_disjoint_mask : GICombineRule<
(defs root:$root, build_fn_matchinfo:$info),
(match (wip_match_opcode G_AND):$root,
@@ -1265,7 +1271,7 @@ def all_combines : GICombineGroup<[trivial_combines, insert_vec_elt_combines,
intdiv_combines, mulh_combines, redundant_neg_operands,
and_or_disjoint_mask, fma_combines, fold_binop_into_select,
sub_add_reg, select_to_minmax, redundant_binop_in_equality,
- fsub_to_fneg, commute_constant_to_rhs]>;
+ double_icmp_zero_and_or_combine, fsub_to_fneg, commute_constant_to_rhs]>;
// A combine group used to for prelegalizer combiners at -O0. The combines in
// this group have been selected based on experiments to balance code size and
diff --git a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
index 9efb70f28fee3ee..aeaf807889baa83 100644
--- a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
@@ -6066,6 +6066,47 @@ bool CombinerHelper::matchRedundantBinOpInEquality(MachineInstr &MI,
return CmpInst::isEquality(Pred) && Y.isValid();
}
+bool CombinerHelper::matchDoubleICmpZeroAndOr(
+ MachineInstr &MI, std::function<void(MachineIRBuilder &)> &MatchInfo) {
+ const unsigned Opcode = MI.getOpcode();
+ assert(Opcode == TargetOpcode::G_OR || Opcode == TargetOpcode::G_AND);
+
+ Register Dst = MI.getOperand(0).getReg();
+ CmpInst::Predicate LHSPred, RHSPred;
+ int64_t LHSImm, RHSImm;
+ Register LHSCmpSrc, RHSCmpSrc;
+ if (!mi_match(
+ Dst, MRI,
+ m_BinOp(Opcode,
+ m_GICmp(m_Pred(LHSPred), m_Reg(LHSCmpSrc), m_ICst(LHSImm)),
+ m_GICmp(m_Pred(RHSPred), m_Reg(RHSCmpSrc), m_ICst(RHSImm)))))
+ return false;
+
+ // G_OR and G_AND cannot handle pointers
+ auto LHSSrcTy = MRI.getType(LHSCmpSrc);
+ if (LHSSrcTy.isPointer() || MRI.getType(RHSCmpSrc).isPointer())
+ return false;
+
+ const bool IsAnd = Opcode == TargetOpcode::G_AND;
+ if ((IsAnd && LHSPred != CmpInst::ICMP_EQ) ||
+ (!IsAnd && LHSPred != CmpInst::ICMP_NE) || LHSPred != RHSPred)
+ return false;
+
+ if (LHSImm != 0 || RHSImm != 0)
+ return false;
+
+ MatchInfo = [=](MachineIRBuilder &B) {
+ const Register OrDst = MRI.createGenericVirtualRegister(LHSSrcTy);
+ auto Zero = B.buildConstant(LHSSrcTy, 0);
+ B.buildOr(OrDst, LHSCmpSrc, RHSCmpSrc);
+ B.buildICmp(IsAnd ? CmpInst::Predicate::ICMP_EQ
+ : CmpInst::Predicate::ICMP_NE,
+ Dst, OrDst, Zero);
+ };
+
+ return true;
+}
+
bool CombinerHelper::matchShiftsTooBig(MachineInstr &MI) {
Register ShiftReg = MI.getOperand(2).getReg();
LLT ResTy = MRI.getType(MI.getOperand(0).getReg());
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/combine-2-icmps-of-0-and-or.mir b/llvm/test/CodeGen/AArch64/GlobalISel/combine-2-icmps-of-0-and-or.mir
new file mode 100644
index 000000000000000..d2bf7f5ba51d746
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/combine-2-icmps-of-0-and-or.mir
@@ -0,0 +1,731 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple aarch64 -run-pass=aarch64-prelegalizer-combiner --aarch64prelegalizercombinerhelper-only-enable-rule="double_icmp_zero_and_or_combine" -verify-machineinstrs %s -o - | FileCheck %s
+# REQUIRES: asserts
+
+
+---
+name: valid_and_eq_0_eq_0_s32
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $w0, $w1
+
+ ; CHECK-LABEL: name: valid_and_eq_0_eq_0_s32
+ ; CHECK: liveins: $w0, $w1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s32) = COPY $w0
+ ; CHECK-NEXT: %y:_(s32) = COPY $w1
+ ; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR %x, %y
+ ; CHECK-NEXT: %and:_(s1) = G_ICMP intpred(eq), [[OR]](s32), %zero
+ ; CHECK-NEXT: %zext:_(s32) = G_ZEXT %and(s1)
+ ; CHECK-NEXT: $w0 = COPY %zext(s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0
+ %x:_(s32) = COPY $w0
+ %y:_(s32) = COPY $w1
+ %zero:_(s32) = G_CONSTANT i32 0
+ %cmp1:_(s1) = G_ICMP intpred(eq), %x:_(s32), %zero:_
+ %cmp2:_(s1) = G_ICMP intpred(eq), %y:_(s32), %zero:_
+ %and:_(s1) = G_AND %cmp1, %cmp2
+ %zext:_(s32) = G_ZEXT %and:_(s1)
+ $w0 = COPY %zext
+ RET_ReallyLR implicit $w0
+
+...
+---
+name: invalid_and_eq_1_eq_0_s32
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $w0, $w1
+
+ ; CHECK-LABEL: name: invalid_and_eq_1_eq_0_s32
+ ; CHECK: liveins: $w0, $w1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s32) = COPY $w0
+ ; CHECK-NEXT: %y:_(s32) = COPY $w1
+ ; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: %one:_(s32) = G_CONSTANT i32 1
+ ; CHECK-NEXT: %cmp1:_(s1) = G_ICMP intpred(eq), %x(s32), %one
+ ; CHECK-NEXT: %cmp2:_(s1) = G_ICMP intpred(eq), %y(s32), %zero
+ ; CHECK-NEXT: %and:_(s1) = G_AND %cmp1, %cmp2
+ ; CHECK-NEXT: %zext:_(s32) = G_ZEXT %and(s1)
+ ; CHECK-NEXT: $w0 = COPY %zext(s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0
+ %x:_(s32) = COPY $w0
+ %y:_(s32) = COPY $w1
+ %zero:_(s32) = G_CONSTANT i32 0
+ %one:_(s32) = G_CONSTANT i32 1
+ %cmp1:_(s1) = G_ICMP intpred(eq), %x:_(s32), %one:_
+ %cmp2:_(s1) = G_ICMP intpred(eq), %y:_(s32), %zero:_
+ %and:_(s1) = G_AND %cmp1, %cmp2
+ %zext:_(s32) = G_ZEXT %and:_(s1)
+ $w0 = COPY %zext
+ RET_ReallyLR implicit $w0
+
+...
+---
+name: invalid_and_eq_0_eq_1_s32
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $w0, $w1
+
+ ; CHECK-LABEL: name: invalid_and_eq_0_eq_1_s32
+ ; CHECK: liveins: $w0, $w1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s32) = COPY $w0
+ ; CHECK-NEXT: %y:_(s32) = COPY $w1
+ ; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: %one:_(s32) = G_CONSTANT i32 1
+ ; CHECK-NEXT: %cmp1:_(s1) = G_ICMP intpred(eq), %x(s32), %zero
+ ; CHECK-NEXT: %cmp2:_(s1) = G_ICMP intpred(eq), %y(s32), %one
+ ; CHECK-NEXT: %and:_(s1) = G_AND %cmp1, %cmp2
+ ; CHECK-NEXT: %zext:_(s32) = G_ZEXT %and(s1)
+ ; CHECK-NEXT: $w0 = COPY %zext(s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0
+ %x:_(s32) = COPY $w0
+ %y:_(s32) = COPY $w1
+ %zero:_(s32) = G_CONSTANT i32 0
+ %one:_(s32) = G_CONSTANT i32 1
+ %cmp1:_(s1) = G_ICMP intpred(eq), %x:_(s32), %zero:_
+ %cmp2:_(s1) = G_ICMP intpred(eq), %y:_(s32), %one:_
+ %and:_(s1) = G_AND %cmp1, %cmp2
+ %zext:_(s32) = G_ZEXT %and:_(s1)
+ $w0 = COPY %zext
+ RET_ReallyLR implicit $w0
+
+...
+---
+name: invalid_and_ne_0_eq_0_s32
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $w0, $w1
+
+ ; CHECK-LABEL: name: invalid_and_ne_0_eq_0_s32
+ ; CHECK: liveins: $w0, $w1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s32) = COPY $w0
+ ; CHECK-NEXT: %y:_(s32) = COPY $w1
+ ; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: %cmp1:_(s1) = G_ICMP intpred(ne), %x(s32), %zero
+ ; CHECK-NEXT: %cmp2:_(s1) = G_ICMP intpred(eq), %y(s32), %zero
+ ; CHECK-NEXT: %and:_(s1) = G_AND %cmp1, %cmp2
+ ; CHECK-NEXT: %zext:_(s32) = G_ZEXT %and(s1)
+ ; CHECK-NEXT: $w0 = COPY %zext(s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0
+ %x:_(s32) = COPY $w0
+ %y:_(s32) = COPY $w1
+ %zero:_(s32) = G_CONSTANT i32 0
+ %cmp1:_(s1) = G_ICMP intpred(ne), %x:_(s32), %zero:_
+ %cmp2:_(s1) = G_ICMP intpred(eq), %y:_(s32), %zero:_
+ %and:_(s1) = G_AND %cmp1, %cmp2
+ %zext:_(s32) = G_ZEXT %and:_(s1)
+ $w0 = COPY %zext
+ RET_ReallyLR implicit $w0
+
+...
+---
+name: invalid_and_eq_0_ne_0_s32
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $w0, $w1
+
+ ; CHECK-LABEL: name: invalid_and_eq_0_ne_0_s32
+ ; CHECK: liveins: $w0, $w1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s32) = COPY $w0
+ ; CHECK-NEXT: %y:_(s32) = COPY $w1
+ ; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: %cmp1:_(s1) = G_ICMP intpred(eq), %x(s32), %zero
+ ; CHECK-NEXT: %cmp2:_(s1) = G_ICMP intpred(ne), %y(s32), %zero
+ ; CHECK-NEXT: %and:_(s1) = G_AND %cmp1, %cmp2
+ ; CHECK-NEXT: %zext:_(s32) = G_ZEXT %and(s1)
+ ; CHECK-NEXT: $w0 = COPY %zext(s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0
+ %x:_(s32) = COPY $w0
+ %y:_(s32) = COPY $w1
+ %zero:_(s32) = G_CONSTANT i32 0
+ %cmp1:_(s1) = G_ICMP intpred(eq), %x:_(s32), %zero:_
+ %cmp2:_(s1) = G_ICMP intpred(ne), %y:_(s32), %zero:_
+ %and:_(s1) = G_AND %cmp1, %cmp2
+ %zext:_(s32) = G_ZEXT %and:_(s1)
+ $w0 = COPY %zext
+ RET_ReallyLR implicit $w0
+
+...
+---
+name: invalid_and_ne_0_ne_0_s32
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $w0, $w1
+
+ ; CHECK-LABEL: name: invalid_and_ne_0_ne_0_s32
+ ; CHECK: liveins: $w0, $w1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s32) = COPY $w0
+ ; CHECK-NEXT: %y:_(s32) = COPY $w1
+ ; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: %cmp1:_(s1) = G_ICMP intpred(ne), %x(s32), %zero
+ ; CHECK-NEXT: %cmp2:_(s1) = G_ICMP intpred(ne), %y(s32), %zero
+ ; CHECK-NEXT: %and:_(s1) = G_AND %cmp1, %cmp2
+ ; CHECK-NEXT: %zext:_(s32) = G_ZEXT %and(s1)
+ ; CHECK-NEXT: $w0 = COPY %zext(s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0
+ %x:_(s32) = COPY $w0
+ %y:_(s32) = COPY $w1
+ %zero:_(s32) = G_CONSTANT i32 0
+ %cmp1:_(s1) = G_ICMP intpred(ne), %x:_(s32), %zero:_
+ %cmp2:_(s1) = G_ICMP intpred(ne), %y:_(s32), %zero:_
+ %and:_(s1) = G_AND %cmp1, %cmp2
+ %zext:_(s32) = G_ZEXT %and:_(s1)
+ $w0 = COPY %zext
+ RET_ReallyLR implicit $w0
+
+...
+---
+name: valid_or_ne_0_ne_0_s32
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $w0, $w1
+
+ ; CHECK-LABEL: name: valid_or_ne_0_ne_0_s32
+ ; CHECK: liveins: $w0, $w1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s32) = COPY $w0
+ ; CHECK-NEXT: %y:_(s32) = COPY $w1
+ ; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR %x, %y
+ ; CHECK-NEXT: %or:_(s1) = G_ICMP intpred(ne), [[OR]](s32), %zero
+ ; CHECK-NEXT: %zext:_(s32) = G_ZEXT %or(s1)
+ ; CHECK-NEXT: $w0 = COPY %zext(s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0
+ %x:_(s32) = COPY $w0
+ %y:_(s32) = COPY $w1
+ %zero:_(s32) = G_CONSTANT i32 0
+ %cmp1:_(s1) = G_ICMP intpred(ne), %x:_(s32), %zero:_
+ %cmp2:_(s1) = G_ICMP intpred(ne), %y:_(s32), %zero:_
+ %or:_(s1) = G_OR %cmp1, %cmp2
+ %zext:_(s32) = G_ZEXT %or:_(s1)
+ $w0 = COPY %zext
+ RET_ReallyLR implicit $w0
+
+...
+---
+name: invalid_or_ne_1_ne_0_s32
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $w0, $w1
+
+ ; CHECK-LABEL: name: invalid_or_ne_1_ne_0_s32
+ ; CHECK: liveins: $w0, $w1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s32) = COPY $w0
+ ; CHECK-NEXT: %y:_(s32) = COPY $w1
+ ; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: %one:_(s32) = G_CONSTANT i32 1
+ ; CHECK-NEXT: %cmp1:_(s1) = G_ICMP intpred(ne), %x(s32), %one
+ ; CHECK-NEXT: %cmp2:_(s1) = G_ICMP intpred(ne), %y(s32), %zero
+ ; CHECK-NEXT: %or:_(s1) = G_OR %cmp1, %cmp2
+ ; CHECK-NEXT: %zext:_(s32) = G_ZEXT %or(s1)
+ ; CHECK-NEXT: $w0 = COPY %zext(s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0
+ %x:_(s32) = COPY $w0
+ %y:_(s32) = COPY $w1
+ %zero:_(s32) = G_CONSTANT i32 0
+ %one:_(s32) = G_CONSTANT i32 1
+ %cmp1:_(s1) = G_ICMP intpred(ne), %x:_(s32), %one:_
+ %cmp2:_(s1) = G_ICMP intpred(ne), %y:_(s32), %zero:_
+ %or:_(s1) = G_OR %cmp1, %cmp2
+ %zext:_(s32) = G_ZEXT %or:_(s1)
+ $w0 = COPY %zext
+ RET_ReallyLR implicit $w0
+
+...
+---
+name: invalid_or_ne_0_ne_1_s32
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $w0, $w1
+
+ ; CHECK-LABEL: name: invalid_or_ne_0_ne_1_s32
+ ; CHECK: liveins: $w0, $w1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s32) = COPY $w0
+ ; CHECK-NEXT: %y:_(s32) = COPY $w1
+ ; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: %one:_(s32) = G_CONSTANT i32 1
+ ; CHECK-NEXT: %cmp1:_(s1) = G_ICMP intpred(ne), %x(s32), %zero
+ ; CHECK-NEXT: %cmp2:_(s1) = G_ICMP intpred(ne), %y(s32), %one
+ ; CHECK-NEXT: %or:_(s1) = G_OR %cmp1, %cmp2
+ ; CHECK-NEXT: %zext:_(s32) = G_ZEXT %or(s1)
+ ; CHECK-NEXT: $w0 = COPY %zext(s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0
+ %x:_(s32) = COPY $w0
+ %y:_(s32) = COPY $w1
+ %zero:_(s32) = G_CONSTANT i32 0
+ %one:_(s32) = G_CONSTANT i32 1
+ %cmp1:_(s1) = G_ICMP intpred(ne), %x:_(s32), %zero:_
+ %cmp2:_(s1) = G_ICMP intpred(ne), %y:_(s32), %one:_
+ %or:_(s1) = G_OR %cmp1, %cmp2
+ %zext:_(s32) = G_ZEXT %or:_(s1)
+ $w0 = COPY %zext
+ RET_ReallyLR implicit $w0
+
+...
+---
+name: invalid_or_eq_0_ne_0_s32
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $w0, $w1
+
+ ; CHECK-LABEL: name: invalid_or_eq_0_ne_0_s32
+ ; CHECK: liveins: $w0, $w1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s32) = COPY $w0
+ ; CHECK-NEXT: %y:_(s32) = COPY $w1
+ ; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: %cmp1:_(s1) = G_ICMP intpred(eq), %x(s32), %zero
+ ; CHECK-NEXT: %cmp2:_(s1) = G_ICMP intpred(ne), %y(s32), %zero
+ ; CHECK-NEXT: %or:_(s1) = G_OR %cmp1, %cmp2
+ ; CHECK-NEXT: %zext:_(s32) = G_ZEXT %or(s1)
+ ; CHECK-NEXT: $w0 = COPY %zext(s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0
+ %x:_(s32) = COPY $w0
+ %y:_(s32) = COPY $w1
+ %zero:_(s32) = G_CONSTANT i32 0
+ %cmp1:_(s1) = G_ICMP intpred(eq), %x:_(s32), %zero:_
+ %cmp2:_(s1) = G_ICMP intpred(ne), %y:_(s32), %zero:_
+ %or:_(s1) = G_OR %cmp1, %cmp2
+ %zext:_(s32) = G_ZEXT %or:_(s1)
+ $w0 = COPY %zext
+ RET_ReallyLR implicit $w0
+
+...
+---
+name: invalid_or_ne_0_eq_0_s32
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $w0, $w1
+
+ ; CHECK-LABEL: name: invalid_or_ne_0_eq_0_s32
+ ; CHECK: liveins: $w0, $w1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s32) = COPY $w0
+ ; CHECK-NEXT: %y:_(s32) = COPY $w1
+ ; CHECK-NEXT: %zero:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: %cmp1:_(s1) = G_ICMP intpred(ne), %x(s32), %zero
+ ; CHECK-NEXT: %cmp2:_(s1) = G_ICMP intpred(eq), %y(s32), %zero
+ ; CHECK-NEXT: %or:_(s1) = G_OR %cmp1, %cmp2
+ ; CHECK-NEXT: %zext:_(s32) = G_ZEXT %or(s1)
+ ; CHECK-NEXT: $w0 = COPY %zext(s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0
+ %x:_(s32) = COPY $w0
+ %y:_(s32) = COPY $w1
+ %zero:_(s32) = G_CONSTANT i32 0
+ %cmp1:_(s1) = G_ICMP intpred(ne), %x:_(s32), %zero:_
+ %cmp2:_(s1) = G_ICMP intpred(eq), %y:_(s32), %zero:_
+ %or:_(s1) = G_OR %cmp1, %cmp2
+ %zext:_(s32) = G_ZEXT %or:_(s1)
+ $w0 = COPY %zext
+ RET_ReallyLR implicit $w0
+
+...
+
+---
+name: valid_and_eq_0_eq_0_s64
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $x0, $x1
+
+ ; CHECK-LABEL: name: valid_and_eq_0_eq_0_s64
+ ; CHECK: liveins: $x0, $x1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s64) = COPY $x0
+ ; CHECK-NEXT: %y:_(s64) = COPY $x1
+ ; CHECK-NEXT: %zero:_(s64) = G_CONSTANT i64 0
+ ; CHECK-NEXT: [[OR:%[0-9]+]]:_(s64) = G_OR %x, %y
+ ; CHECK-NEXT: %and:_(s1) = G_ICMP intpred(eq), [[OR]](s64), %zero
+ ; CHECK-NEXT: %zext:_(s64) = G_ZEXT %and(s1)
+ ; CHECK-NEXT: $x0 = COPY %zext(s64)
+ ; CHECK-NEXT: RET_ReallyLR implicit $x0
+ %x:_(s64) = COPY $x0
+ %y:_(s64) = COPY $x1
+ %zero:_(s64) = G_CONSTANT i64 0
+ %cmp1:_(s1) = G_ICMP intpred(eq), %x:_(s64), %zero:_
+ %cmp2:_(s1) = G_ICMP intpred(eq), %y:_(s64), %zero:_
+ %and:_(s1) = G_AND %cmp1, %cmp2
+ %zext:_(s64) = G_ZEXT %and:_(s1)
+ $x0 = COPY %zext
+ RET_ReallyLR implicit $x0
+
+...
+---
+name: invalid_and_eq_1_eq_0_s64
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $x0, $x1
+
+ ; CHECK-LABEL: name: invalid_and_eq_1_eq_0_s64
+ ; CHECK: liveins: $x0, $x1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s64) = COPY $x0
+ ; CHECK-NEXT: %y:_(s64) = COPY $x1
+ ; CHECK-NEXT: %zero:_(s64) = G_CONSTANT i64 0
+ ; CHECK-NEXT: %one:_(s64) = G_CONSTANT i64 1
+ ; CHECK-NEXT: %cmp1:_(s1) = G_ICMP intpred(eq), %x(s64), %one
+ ; CHECK-NEXT: %cmp2:_(s1) = G_ICMP intpred(eq), %y(s64), %zero
+ ; CHECK-NEXT: %and:_(s1) = G_AND %cmp1, %cmp2
+ ; CHECK-NEXT: %zext:_(s64) = G_ZEXT %and(s1)
+ ; CHECK-NEXT: $x0 = COPY %zext(s64)
+ ; CHECK-NEXT: RET_ReallyLR implicit $x0
+ %x:_(s64) = COPY $x0
+ %y:_(s64) = COPY $x1
+ %zero:_(s64) = G_CONSTANT i64 0
+ %one:_(s64) = G_CONSTANT i64 1
+ %cmp1:_(s1) = G_ICMP intpred(eq), %x:_(s64), %one:_
+ %cmp2:_(s1) = G_ICMP intpred(eq), %y:_(s64), %zero:_
+ %and:_(s1) = G_AND %cmp1, %cmp2
+ %zext:_(s64) = G_ZEXT %and:_(s1)
+ $x0 = COPY %zext
+ RET_ReallyLR implicit $x0
+
+...
+---
+name: invalid_and_eq_0_eq_1_s64
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $x0, $x1
+
+ ; CHECK-LABEL: name: invalid_and_eq_0_eq_1_s64
+ ; CHECK: liveins: $x0, $x1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s64) = COPY $x0
+ ; CHECK-NEXT: %y:_(s64) = COPY $x1
+ ; CHECK-NEXT: %zero:_(s64) = G_CONSTANT i64 0
+ ; CHECK-NEXT: %one:_(s64) = G_CONSTANT i64 1
+ ; CHECK-NEXT: %cmp1:_(s1) = G_ICMP intpred(eq), %x(s64), %zero
+ ; CHECK-NEXT: %cmp2:_(s1) = G_ICMP intpred(eq), %y(s64), %one
+ ; CHECK-NEXT: %and:_(s1) = G_AND %cmp1, %cmp2
+ ; CHECK-NEXT: %zext:_(s64) = G_ZEXT %and(s1)
+ ; CHECK-NEXT: $x0 = COPY %zext(s64)
+ ; CHECK-NEXT: RET_ReallyLR implicit $x0
+ %x:_(s64) = COPY $x0
+ %y:_(s64) = COPY $x1
+ %zero:_(s64) = G_CONSTANT i64 0
+ %one:_(s64) = G_CONSTANT i64 1
+ %cmp1:_(s1) = G_ICMP intpred(eq), %x:_(s64), %zero:_
+ %cmp2:_(s1) = G_ICMP intpred(eq), %y:_(s64), %one:_
+ %and:_(s1) = G_AND %cmp1, %cmp2
+ %zext:_(s64) = G_ZEXT %and:_(s1)
+ $x0 = COPY %zext
+ RET_ReallyLR implicit $x0
+
+...
+---
+name: invalid_and_ne_0_eq_0_s64
+tracksRegLiveness: true
+legalized: true
+body: |
+ bb.0:
+ liveins: $x0, $x1
+
+ ; CHECK-LABEL: name: invalid_and_ne_0_eq_0_s64
+ ; CHECK: liveins: $x0, $x1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: %x:_(s64) = COPY $x0
+ ; CHECK-NEXT: %y:_(s64) = COPY $x1
+ ; CHECK-NEXT: %zero:_(s64) = G_CONSTANT i64 0
+ ; CHECK-NEXT: %cmp1:_(s1) = G_ICMP intpred(ne), %x(s64), %zero
+ ; CHECK-NEXT: %cmp2:_(s1) = G_ICMP intpred(eq), %y(s64), %zero
+ ; CHECK-NEXT: %and:_(s1) = G_AND %cmp1, %cmp2
+ ; CHECK-NEXT: %zext:_(s64) = G_ZEXT %and(s1)
+ ; CHECK-NEXT...
[truncated]
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
arsenm
reviewed
Oct 13, 2023
I don't think we can encode some of the pointer type in this patch checks in that, right? |
|
mark |
Also combine (X != 0) | (Y != 0) -> (X | Y) != 0
dfszabo
force-pushed
the
main-global-isel-two-icmp-zero-or-and-combine
branch
from
0d4cb75
to
6f076a2
Compare
Good idea, but for the reason @aemerson mentioned it does not seem feasible. |
You can mix MIR patterns and C++. You can do most of the matching with MIR patterns, have an "out" pattern as well, and just use a smaller C++ function to verify the more complex matching predicates. Can you try that approach and if it proves impossible or just too annoying, let me know your thoughts? It'd help a lot :) |
I tried this approach and I think I nailed it but the tests are crashing. Any idea? Seems like the BB's parent is nullptr, don't know why though. |
Can you post a separate PR with this version for easier debugging? |
Yes, here you go #71949. I commented also my findings so far. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Also combine (X != 0) | (Y != 0) -> (X | Y) != 0