InstCombine: extend select-equiv to support vectors #111966
Merged
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-llvm-analysis @llvm/pr-subscribers-llvm-transforms Author: Ramkumar Ramachandra (artagnon) ChangesfoldSelectEquivalence currently doesn't support comparisons on vector types due to correctness concerns. Note that the only concern is lane-crossing; ShuffleVector is the only possible lane-crossing instruction, and ShuffleVectorInst::{isIdentity,isShuffle} are the exact properties that should not be broken for valid vector-replacements. Put in the checks, and lift the limitation. -- 8< -- Full diff: https://github.com/llvm/llvm-project/pull/111966.diff 4 Files Affected:
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
index 3f780285efe423..47dfe356cdf771 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -1288,6 +1288,13 @@ bool InstCombinerImpl::replaceInInstruction(Value *V, Value *Old, Value *New,
!isSafeToSpeculativelyExecuteWithVariableReplaced(I))
return false;
+ // ShuffleVector is the only possible lane-crossing instruction. isIdentity()
+ // or isSelect() must be satisfied for length-preserving non-lane-crossing
+ // shuffles.
+ auto *Shuffle = dyn_cast<ShuffleVectorInst>(I);
+ if (Shuffle && !Shuffle->isIdentity() && !Shuffle->isSelect())
+ return false;
+
bool Changed = false;
for (Use &U : I->operands()) {
if (U == Old) {
@@ -1366,9 +1373,8 @@ Instruction *InstCombinerImpl::foldSelectValueEquivalence(SelectInst &Sel,
// with different operands, which should not cause side-effects or trigger
// undefined behavior). Only do this if CmpRHS is a constant, as
// profitability is not clear for other cases.
- // FIXME: Support vectors.
if (OldOp == CmpLHS && match(NewOp, m_ImmConstant()) &&
- !match(OldOp, m_Constant()) && !Cmp.getType()->isVectorTy() &&
+ !match(OldOp, m_Constant()) &&
isGuaranteedNotToBeUndef(NewOp, SQ.AC, &Sel, &DT))
if (replaceInInstruction(TrueVal, OldOp, NewOp))
return &Sel;
diff --git a/llvm/test/Transforms/InstCombine/and-or-icmps.ll b/llvm/test/Transforms/InstCombine/and-or-icmps.ll
index ad28ad980de5b4..eb4723c86542de 100644
--- a/llvm/test/Transforms/InstCombine/and-or-icmps.ll
+++ b/llvm/test/Transforms/InstCombine/and-or-icmps.ll
@@ -983,7 +983,7 @@ define <2 x i1> @substitute_constant_or_ne_slt_swap_vec_poison(<2 x i8> %x, <2 x
define <2 x i1> @substitute_constant_or_ne_slt_swap_vec_logical(<2 x i8> %x, <2 x i8> %y) {
; CHECK-LABEL: @substitute_constant_or_ne_slt_swap_vec_logical(
; CHECK-NEXT: [[C1:%.*]] = icmp ne <2 x i8> [[X:%.*]], <i8 42, i8 poison>
-; CHECK-NEXT: [[C2:%.*]] = icmp slt <2 x i8> [[Y:%.*]], [[X]]
+; CHECK-NEXT: [[C2:%.*]] = icmp slt <2 x i8> [[Y:%.*]], <i8 42, i8 poison>
; CHECK-NEXT: [[R:%.*]] = select <2 x i1> [[C1]], <2 x i1> <i1 true, i1 true>, <2 x i1> [[C2]]
; CHECK-NEXT: ret <2 x i1> [[R]]
;
diff --git a/llvm/test/Transforms/InstCombine/select-binop-cmp.ll b/llvm/test/Transforms/InstCombine/select-binop-cmp.ll
index 647287ef5ebad1..cd8c29ba4cd819 100644
--- a/llvm/test/Transforms/InstCombine/select-binop-cmp.ll
+++ b/llvm/test/Transforms/InstCombine/select-binop-cmp.ll
@@ -552,12 +552,12 @@ define i32 @select_xor_icmp_bad_6(i32 %x, i32 %y, i32 %z) {
ret i32 %C
}
-; Value equivalence substitution is all-or-nothing, so needs a scalar compare.
+; Value equivalence substitution is valid.
-define <2 x i8> @select_xor_icmp_vec_bad(<2 x i8> %x, <2 x i8> %y, <2 x i8> %z) {
-; CHECK-LABEL: @select_xor_icmp_vec_bad(
+define <2 x i8> @select_xor_icmp_vec_equivalence(<2 x i8> %x, <2 x i8> %y, <2 x i8> %z) {
+; CHECK-LABEL: @select_xor_icmp_vec_equivalence(
; CHECK-NEXT: [[A:%.*]] = icmp eq <2 x i8> [[X:%.*]], <i8 5, i8 3>
-; CHECK-NEXT: [[B:%.*]] = xor <2 x i8> [[X]], [[Z:%.*]]
+; CHECK-NEXT: [[B:%.*]] = xor <2 x i8> [[Z:%.*]], <i8 5, i8 3>
; CHECK-NEXT: [[C:%.*]] = select <2 x i1> [[A]], <2 x i8> [[B]], <2 x i8> [[Y:%.*]]
; CHECK-NEXT: ret <2 x i8> [[C]]
;
@@ -567,7 +567,7 @@ define <2 x i8> @select_xor_icmp_vec_bad(<2 x i8> %x, <2 x i8> %y, <2 x i8> %z)
ret <2 x i8> %C
}
-; Value equivalence substitution is all-or-nothing, so needs a scalar compare.
+; Value equivalence substitution is invalid due to lane-crossing shufflevector.
define <2 x i32> @vec_select_no_equivalence(<2 x i32> %x) {
; CHECK-LABEL: @vec_select_no_equivalence(
diff --git a/llvm/test/Transforms/InstCombine/select-value-equivalence.ll b/llvm/test/Transforms/InstCombine/select-value-equivalence.ll
new file mode 100644
index 00000000000000..da8970a7630f28
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/select-value-equivalence.ll
@@ -0,0 +1,267 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -passes=instcombine -S %s | FileCheck %s
+
+define <2 x i8> @select_icmp_insertelement_eq(<2 x i8> %x, <2 x i8> %y, i8 %i) {
+; CHECK-LABEL: define <2 x i8> @select_icmp_insertelement_eq(
+; CHECK-SAME: <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]], i8 [[I:%.*]]) {
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i8> [[Y]], <i8 2, i8 2>
+; CHECK-NEXT: [[INSERT:%.*]] = insertelement <2 x i8> <i8 2, i8 2>, i8 0, i8 [[I]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[CMP]], <2 x i8> [[INSERT]], <2 x i8> [[X]]
+; CHECK-NEXT: ret <2 x i8> [[RETVAL]]
+;
+ %cmp = icmp eq <2 x i8> %y, <i8 2, i8 2>
+ %insert = insertelement <2 x i8> %y, i8 0, i8 %i
+ %retval = select <2 x i1> %cmp, <2 x i8> %insert, <2 x i8> %x
+ ret <2 x i8> %retval
+}
+
+define <2 x i8> @select_icmp_insertelement_ne(<2 x i8> %x, <2 x i8> %y, i8 %i) {
+; CHECK-LABEL: define <2 x i8> @select_icmp_insertelement_ne(
+; CHECK-SAME: <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]], i8 [[I:%.*]]) {
+; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq <2 x i8> [[Y]], <i8 2, i8 2>
+; CHECK-NEXT: [[INSERT:%.*]] = insertelement <2 x i8> <i8 2, i8 2>, i8 0, i8 [[I]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[CMP_NOT]], <2 x i8> [[INSERT]], <2 x i8> [[X]]
+; CHECK-NEXT: ret <2 x i8> [[RETVAL]]
+;
+ %cmp = icmp ne <2 x i8> %y, <i8 2, i8 2>
+ %insert = insertelement <2 x i8> %y, i8 0, i8 %i
+ %retval = select <2 x i1> %cmp, <2 x i8> %x, <2 x i8> %insert
+ ret <2 x i8> %retval
+}
+
+define <2 x i8> @select_icmp_shufflevector_identity(<2 x i8> %x, <2 x i8> %y) {
+; CHECK-LABEL: define <2 x i8> @select_icmp_shufflevector_identity(
+; CHECK-SAME: <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]]) {
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i8> [[Y]], <i8 2, i8 2>
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[CMP]], <2 x i8> <i8 2, i8 2>, <2 x i8> [[X]]
+; CHECK-NEXT: ret <2 x i8> [[RETVAL]]
+;
+ %cmp = icmp eq <2 x i8> %y, <i8 2, i8 2>
+ %shuffle = shufflevector <2 x i8> %y, <2 x i8> poison, <2 x i32> <i32 0, i32 1>
+ %retval = select <2 x i1> %cmp, <2 x i8> %shuffle, <2 x i8> %x
+ ret <2 x i8> %retval
+}
+
+define <4 x i8> @select_icmp_shufflevector_select(<4 x i8> %x, <4 x i8> %y, <4 x i8> %z) {
+; CHECK-LABEL: define <4 x i8> @select_icmp_shufflevector_select(
+; CHECK-SAME: <4 x i8> [[X:%.*]], <4 x i8> [[Y:%.*]], <4 x i8> [[Z:%.*]]) {
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq <4 x i8> [[Y]], <i8 2, i8 2, i8 2, i8 2>
+; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i8> [[Z]], <4 x i8> <i8 poison, i8 2, i8 poison, i8 2>, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+; CHECK-NEXT: [[RETVAL:%.*]] = select <4 x i1> [[CMP]], <4 x i8> [[SHUFFLE]], <4 x i8> [[X]]
+; CHECK-NEXT: ret <4 x i8> [[RETVAL]]
+;
+ %cmp = icmp eq <4 x i8> %y, <i8 2, i8 2, i8 2, i8 2>
+ %shuffle = shufflevector <4 x i8> %y, <4 x i8> %z, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+ %retval = select <4 x i1> %cmp, <4 x i8> %shuffle, <4 x i8> %x
+ ret <4 x i8> %retval
+}
+
+define <2 x i8> @select_icmp_shufflevector_lanecrossing(<2 x i8> %x, <2 x i8> %y) {
+; CHECK-LABEL: define <2 x i8> @select_icmp_shufflevector_lanecrossing(
+; CHECK-SAME: <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]]) {
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i8> [[Y]], <i8 2, i8 2>
+; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <2 x i8> [[Y]], <2 x i8> poison, <2 x i32> <i32 1, i32 0>
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[CMP]], <2 x i8> [[SHUFFLE]], <2 x i8> [[X]]
+; CHECK-NEXT: ret <2 x i8> [[RETVAL]]
+;
+ %cmp = icmp eq <2 x i8> %y, <i8 2, i8 2>
+ %shuffle = shufflevector <2 x i8> %y, <2 x i8> poison, <2 x i32> <i32 1, i32 0>
+ %retval = select <2 x i1> %cmp, <2 x i8> %shuffle, <2 x i8> %x
+ ret <2 x i8> %retval
+}
+
+define i8 @select_icmp_udiv(i8 %x, i8 %y) {
+; CHECK-LABEL: define i8 @select_icmp_udiv(
+; CHECK-SAME: i8 [[X:%.*]], i8 [[Y:%.*]]) {
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i8 [[Y]], 2
+; CHECK-NEXT: [[UDIV:%.*]] = udiv i8 [[X]], [[Y]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select i1 [[CMP]], i8 [[UDIV]], i8 [[X]]
+; CHECK-NEXT: ret i8 [[RETVAL]]
+;
+ %cmp = icmp eq i8 %y, 2
+ %udiv = udiv i8 %x, %y
+ %retval = select i1 %cmp, i8 %udiv, i8 %x
+ ret i8 %retval
+}
+
+define i8 @select_icmp_urem(i8 %x, i8 %y) {
+; CHECK-LABEL: define i8 @select_icmp_urem(
+; CHECK-SAME: i8 [[X:%.*]], i8 [[Y:%.*]]) {
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i8 [[Y]], 2
+; CHECK-NEXT: [[UREM:%.*]] = urem i8 [[X]], [[Y]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select i1 [[CMP]], i8 [[UREM]], i8 [[X]]
+; CHECK-NEXT: ret i8 [[RETVAL]]
+;
+ %cmp = icmp eq i8 %y, 2
+ %urem = urem i8 %x, %y
+ %retval = select i1 %cmp, i8 %urem, i8 %x
+ ret i8 %retval
+}
+
+define <2 x i8> @select_icmp_udiv_vec(<2 x i8> %x, <2 x i8> %y) {
+; CHECK-LABEL: define <2 x i8> @select_icmp_udiv_vec(
+; CHECK-SAME: <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]]) {
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i8> [[Y]], <i8 2, i8 2>
+; CHECK-NEXT: [[UDIV:%.*]] = udiv <2 x i8> [[X]], [[Y]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[CMP]], <2 x i8> [[UDIV]], <2 x i8> [[X]]
+; CHECK-NEXT: ret <2 x i8> [[RETVAL]]
+;
+ %cmp = icmp eq <2 x i8> %y, <i8 2, i8 2>
+ %udiv = udiv <2 x i8> %x, %y
+ %retval = select <2 x i1> %cmp, <2 x i8> %udiv, <2 x i8> %x
+ ret <2 x i8> %retval
+}
+
+define <2 x i8> @select_icmp_urem_vec(<2 x i8> %x, <2 x i8> %y) {
+; CHECK-LABEL: define <2 x i8> @select_icmp_urem_vec(
+; CHECK-SAME: <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]]) {
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i8> [[Y]], <i8 2, i8 2>
+; CHECK-NEXT: [[UREM:%.*]] = urem <2 x i8> [[X]], [[Y]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[CMP]], <2 x i8> [[UREM]], <2 x i8> [[X]]
+; CHECK-NEXT: ret <2 x i8> [[RETVAL]]
+;
+ %cmp = icmp eq <2 x i8> %y, <i8 2, i8 2>
+ %urem = urem <2 x i8> %x, %y
+ %retval = select <2 x i1> %cmp, <2 x i8> %urem, <2 x i8> %x
+ ret <2 x i8> %retval
+}
+
+define <2 x double> @select_fcmp_fadd_oeq_not_zero_vec(<2 x double> %x, <2 x double> %y) {
+; CHECK-LABEL: define <2 x double> @select_fcmp_fadd_oeq_not_zero_vec(
+; CHECK-SAME: <2 x double> [[X:%.*]], <2 x double> [[Y:%.*]]) {
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp oeq <2 x double> [[Y]], <double 2.000000e+00, double 2.000000e+00>
+; CHECK-NEXT: [[FADD:%.*]] = fadd <2 x double> [[X]], [[Y]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[FCMP]], <2 x double> [[FADD]], <2 x double> [[X]]
+; CHECK-NEXT: ret <2 x double> [[RETVAL]]
+;
+ %fcmp = fcmp oeq <2 x double> %y, <double 2.0, double 2.0>
+ %fadd = fadd <2 x double> %x, %y
+ %retval = select <2 x i1> %fcmp, <2 x double> %fadd, <2 x double> %x
+ ret <2 x double> %retval
+}
+
+define <2 x double> @select_fcmp_fadd_une_not_zero_vec(<2 x double> %x, <2 x double> %y) {
+; CHECK-LABEL: define <2 x double> @select_fcmp_fadd_une_not_zero_vec(
+; CHECK-SAME: <2 x double> [[X:%.*]], <2 x double> [[Y:%.*]]) {
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp une <2 x double> [[Y]], <double 2.000000e+00, double 2.000000e+00>
+; CHECK-NEXT: [[FADD:%.*]] = fadd <2 x double> [[X]], [[Y]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[FCMP]], <2 x double> [[X]], <2 x double> [[FADD]]
+; CHECK-NEXT: ret <2 x double> [[RETVAL]]
+;
+ %fcmp = fcmp une <2 x double> %y, <double 2.0, double 2.0>
+ %fadd = fadd <2 x double> %x, %y
+ %retval = select <2 x i1> %fcmp, <2 x double> %x, <2 x double> %fadd
+ ret <2 x double> %retval
+}
+
+define <2 x double> @select_fcmp_fadd_ueq_nnan_not_zero_vec(<2 x double> %x, <2 x double> %y) {
+; CHECK-LABEL: define <2 x double> @select_fcmp_fadd_ueq_nnan_not_zero_vec(
+; CHECK-SAME: <2 x double> [[X:%.*]], <2 x double> [[Y:%.*]]) {
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp nnan ueq <2 x double> [[Y]], <double 2.000000e+00, double 2.000000e+00>
+; CHECK-NEXT: [[FADD:%.*]] = fadd <2 x double> [[X]], [[Y]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[FCMP]], <2 x double> [[FADD]], <2 x double> [[X]]
+; CHECK-NEXT: ret <2 x double> [[RETVAL]]
+;
+ %fcmp = fcmp nnan ueq <2 x double> %y, <double 2.0, double 2.0>
+ %fadd = fadd <2 x double> %x, %y
+ %retval = select <2 x i1> %fcmp, <2 x double> %fadd, <2 x double> %x
+ ret <2 x double> %retval
+}
+
+define <2 x double> @select_fcmp_fadd_one_nnan_not_zero_vec(<2 x double> %x, <2 x double> %y) {
+; CHECK-LABEL: define <2 x double> @select_fcmp_fadd_one_nnan_not_zero_vec(
+; CHECK-SAME: <2 x double> [[X:%.*]], <2 x double> [[Y:%.*]]) {
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp nnan one <2 x double> [[Y]], <double 2.000000e+00, double 2.000000e+00>
+; CHECK-NEXT: [[FADD:%.*]] = fadd <2 x double> [[X]], [[Y]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[FCMP]], <2 x double> [[X]], <2 x double> [[FADD]]
+; CHECK-NEXT: ret <2 x double> [[RETVAL]]
+;
+ %fcmp = fcmp nnan one <2 x double> %y, <double 2.0, double 2.0>
+ %fadd = fadd <2 x double> %x, %y
+ %retval = select <2 x i1> %fcmp, <2 x double> %x, <2 x double> %fadd
+ ret <2 x double> %retval
+}
+
+define <2 x double> @select_fcmp_fadd_ueq_vec(<2 x double> %x, <2 x double> %y) {
+; CHECK-LABEL: define <2 x double> @select_fcmp_fadd_ueq_vec(
+; CHECK-SAME: <2 x double> [[X:%.*]], <2 x double> [[Y:%.*]]) {
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp ueq <2 x double> [[Y]], <double 2.000000e+00, double 2.000000e+00>
+; CHECK-NEXT: [[FADD:%.*]] = fadd <2 x double> [[X]], [[Y]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[FCMP]], <2 x double> [[FADD]], <2 x double> [[X]]
+; CHECK-NEXT: ret <2 x double> [[RETVAL]]
+;
+ %fcmp = fcmp ueq <2 x double> %y, <double 2.0, double 2.0>
+ %fadd = fadd <2 x double> %x, %y
+ %retval = select <2 x i1> %fcmp, <2 x double> %fadd, <2 x double> %x
+ ret <2 x double> %retval
+}
+
+define <2 x double> @select_fcmp_fadd_one_vec(<2 x double> %x, <2 x double> %y) {
+; CHECK-LABEL: define <2 x double> @select_fcmp_fadd_one_vec(
+; CHECK-SAME: <2 x double> [[X:%.*]], <2 x double> [[Y:%.*]]) {
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp one <2 x double> [[Y]], <double 2.000000e+00, double 2.000000e+00>
+; CHECK-NEXT: [[FADD:%.*]] = fadd <2 x double> [[X]], [[Y]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[FCMP]], <2 x double> [[X]], <2 x double> [[FADD]]
+; CHECK-NEXT: ret <2 x double> [[RETVAL]]
+;
+ %fcmp = fcmp one <2 x double> %y, <double 2.0, double 2.0>
+ %fadd = fadd <2 x double> %x, %y
+ %retval = select <2 x i1> %fcmp, <2 x double> %x, <2 x double> %fadd
+ ret <2 x double> %retval
+}
+
+define <2 x double> @select_fcmp_fadd_oeq_zero_vec(<2 x double> %x, <2 x double> %y) {
+; CHECK-LABEL: define <2 x double> @select_fcmp_fadd_oeq_zero_vec(
+; CHECK-SAME: <2 x double> [[X:%.*]], <2 x double> [[Y:%.*]]) {
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp oeq <2 x double> [[Y]], zeroinitializer
+; CHECK-NEXT: [[FADD:%.*]] = fadd <2 x double> [[X]], [[Y]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[FCMP]], <2 x double> [[FADD]], <2 x double> [[X]]
+; CHECK-NEXT: ret <2 x double> [[RETVAL]]
+;
+ %fcmp = fcmp oeq <2 x double> %y, zeroinitializer
+ %fadd = fadd <2 x double> %x, %y
+ %retval = select <2 x i1> %fcmp, <2 x double> %fadd, <2 x double> %x
+ ret <2 x double> %retval
+}
+
+define <2 x double> @select_fcmp_fadd_une_zero_vec(<2 x double> %x, <2 x double> %y) {
+; CHECK-LABEL: define <2 x double> @select_fcmp_fadd_une_zero_vec(
+; CHECK-SAME: <2 x double> [[X:%.*]], <2 x double> [[Y:%.*]]) {
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp une <2 x double> [[Y]], zeroinitializer
+; CHECK-NEXT: [[FADD:%.*]] = fadd <2 x double> [[X]], [[Y]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[FCMP]], <2 x double> [[X]], <2 x double> [[FADD]]
+; CHECK-NEXT: ret <2 x double> [[RETVAL]]
+;
+ %fcmp = fcmp une <2 x double> %y, zeroinitializer
+ %fadd = fadd <2 x double> %x, %y
+ %retval = select <2 x i1> %fcmp, <2 x double> %x, <2 x double> %fadd
+ ret <2 x double> %retval
+}
+
+define <2 x double> @select_fcmp_fdiv_oeq_not_zero_vec(<2 x double> %x, <2 x double> %y) {
+; CHECK-LABEL: define <2 x double> @select_fcmp_fdiv_oeq_not_zero_vec(
+; CHECK-SAME: <2 x double> [[X:%.*]], <2 x double> [[Y:%.*]]) {
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp oeq <2 x double> [[Y]], <double 2.000000e+00, double 2.000000e+00>
+; CHECK-NEXT: [[DIV:%.*]] = fdiv <2 x double> [[X]], [[Y]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[FCMP]], <2 x double> [[DIV]], <2 x double> [[X]]
+; CHECK-NEXT: ret <2 x double> [[RETVAL]]
+;
+ %fcmp = fcmp oeq <2 x double> %y, <double 2.0, double 2.0>
+ %div = fdiv <2 x double> %x, %y
+ %retval = select <2 x i1> %fcmp, <2 x double> %div, <2 x double> %x
+ ret <2 x double> %retval
+}
+
+define <2 x double> @select_fcmp_fdiv_une_not_zero_vec(<2 x double> %x, <2 x double> %y) {
+; CHECK-LABEL: define <2 x double> @select_fcmp_fdiv_une_not_zero_vec(
+; CHECK-SAME: <2 x double> [[X:%.*]], <2 x double> [[Y:%.*]]) {
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp une <2 x double> [[Y]], <double 2.000000e+00, double 2.000000e+00>
+; CHECK-NEXT: [[DIV:%.*]] = fdiv <2 x double> [[X]], [[Y]]
+; CHECK-NEXT: [[RETVAL:%.*]] = select <2 x i1> [[FCMP]], <2 x double> [[X]], <2 x double> [[DIV]]
+; CHECK-NEXT: ret <2 x double> [[RETVAL]]
+;
+ %fcmp = fcmp une <2 x double> %y, <double 2.0, double 2.0>
+ %div = fdiv <2 x double> %x, %y
+ %retval = select <2 x i1> %fcmp, <2 x double> %x, <2 x double> %div
+ ret <2 x double> %retval
+}
|
nikic
requested changes
Oct 11, 2024
There was a problem hiding this comment.
Shufflevector is not the only lane-crossing operation, see
llvm-project/llvm/lib/Analysis/InstructionSimplify.cpp
Lines 4347 to 4351 in 80c15c4
artagnon
force-pushed
the
ic-select-equiv-vec
branch
from
c1299b9 to
e3c62a2
Compare
Might be nice to make this a helper |
|
Yes, we should move that code plus llvm-project/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp Lines 3635 to 3650 in c13f806 isLanewiseOperation() or so.
|
artagnon
force-pushed
the
ic-select-equiv-vec
branch
from
e3c62a2 to
866830e
Compare
|
Rebased, ready to review now. |
foldSelectEquivalence currently doesn't support comparisons on vector
types due to correctness concerns. Note that the only concern is
lane-crossing; ShuffleVector is the only possible lane-crossing
instruction, and ShuffleVectorInst::{isIdentity,isShuffle} are the exact
properties that should not be broken for valid vector-replacements. Put
in the checks, and lift the limitation.
artagnon
force-pushed
the
ic-select-equiv-vec
branch
from
866830e to
8e764f1
Compare
DanielCChen
pushed a commit
to DanielCChen/llvm-project
that referenced
this pull request
Oct 16, 2024
foldSelectEquivalence currently doesn't support GVN-like replacements on vector types. Put in the checks for potentially lane-crossing operations, and lift the limitation.
bricknerb
pushed a commit
to bricknerb/llvm-project
that referenced
this pull request
Oct 17, 2024
foldSelectEquivalence currently doesn't support GVN-like replacements on vector types. Put in the checks for potentially lane-crossing operations, and lift the limitation.
EricWF
pushed a commit
to efcs/llvm-project
that referenced
this pull request
Oct 22, 2024
foldSelectEquivalence currently doesn't support GVN-like replacements on vector types. Put in the checks for potentially lane-crossing operations, and lift the limitation.
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.
foldSelectEquivalence currently doesn't support GVN-like replacements on vector types. Put in the checks for potentially lane-crossing operations, and lift the limitation.