InstCombine: Basic insertelement support for SimplifyDemandedFPClass #174100
+197
−0
Conversation
This file contains hidden or 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
arsenm
added
floating-point
Graphite App
Contributor
Author
|
Warning This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
This stack of pull requests is managed by Graphite. Learn more about stacking. |
arsenm
requested review from
RKSimon,
andykaylor,
dtcxzyw,
jayfoad,
nikic and
spavloff
Member
|
@llvm/pr-subscribers-llvm-transforms Author: Matt Arsenault (arsenm) ChangesEventually this should pull up the known elements logic from Full diff: https://github.com/llvm/llvm-project/pull/174100.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
index 39b95c45cb6e1..87b8664760b09 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
@@ -2303,6 +2303,17 @@ Value *InstCombinerImpl::SimplifyDemandedUseFPClass(Value *V,
return I;
break;
}
+ case Instruction::InsertElement: {
+ KnownFPClass KnownInserted, KnownVec;
+ if (SimplifyDemandedFPClass(I, 1, DemandedMask, KnownInserted, Depth + 1) ||
+ SimplifyDemandedFPClass(I, 0, DemandedMask, KnownVec, Depth + 1))
+ return I;
+ break;
+
+ // TODO: Use demanded elements logic from computeKnownFPClass
+ Known = KnownVec | KnownInserted;
+ break;
+ }
default:
Known = computeKnownFPClass(I, DemandedMask, CxtI, Depth + 1);
break;
diff --git a/llvm/test/Transforms/InstCombine/simplify-demanded-fpclass-insertelement.ll b/llvm/test/Transforms/InstCombine/simplify-demanded-fpclass-insertelement.ll
new file mode 100644
index 0000000000000..36d8766d967aa
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/simplify-demanded-fpclass-insertelement.ll
@@ -0,0 +1,187 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt -S -passes=instcombine < %s | FileCheck %s
+
+declare nofpclass(inf zero sub norm) <4 x half> @returns_nan()
+declare nofpclass(nan zero sub norm) <4 x half> @returns_inf()
+declare nofpclass(inf zero sub norm) half @returns_nan_f16()
+declare nofpclass(inf zero sub norm) <vscale x 4 x half> @returns_nan_nxv4f16()
+
+define nofpclass(inf zero sub norm) <4 x half> @ret_only_nan__insert_unknown_unknown(<4 x half> %vec, half %elt, i32 %idx) {
+; CHECK-LABEL: define nofpclass(inf zero sub norm) <4 x half> @ret_only_nan__insert_unknown_unknown(
+; CHECK-SAME: <4 x half> [[VEC:%.*]], half [[ELT:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: [[INSERT:%.*]] = insertelement <4 x half> [[VEC]], half [[ELT]], i32 [[IDX]]
+; CHECK-NEXT: ret <4 x half> [[INSERT]]
+;
+ %insert = insertelement <4 x half> %vec, half %elt, i32 %idx
+ ret <4 x half> %insert
+}
+
+define nofpclass(qnan inf zero sub norm) <4 x half> @ret_only_snan__insert_unknown_unknown(<4 x half> %vec, half %elt, i32 %idx) {
+; CHECK-LABEL: define nofpclass(qnan inf zero sub norm) <4 x half> @ret_only_snan__insert_unknown_unknown(
+; CHECK-SAME: <4 x half> [[VEC:%.*]], half [[ELT:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: [[INSERT:%.*]] = insertelement <4 x half> [[VEC]], half [[ELT]], i32 [[IDX]]
+; CHECK-NEXT: ret <4 x half> [[INSERT]]
+;
+ %insert = insertelement <4 x half> %vec, half %elt, i32 %idx
+ ret <4 x half> %insert
+}
+
+define nofpclass(snan inf zero sub norm) <4 x half> @ret_only_qnan__insert_unknown_unknown(<4 x half> %vec, half %elt, i32 %idx) {
+; CHECK-LABEL: define nofpclass(snan inf zero sub norm) <4 x half> @ret_only_qnan__insert_unknown_unknown(
+; CHECK-SAME: <4 x half> [[VEC:%.*]], half [[ELT:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: [[INSERT:%.*]] = insertelement <4 x half> [[VEC]], half [[ELT]], i32 [[IDX]]
+; CHECK-NEXT: ret <4 x half> [[INSERT]]
+;
+ %insert = insertelement <4 x half> %vec, half %elt, i32 %idx
+ ret <4 x half> %insert
+}
+
+define nofpclass(nan zero sub norm) <4 x half> @ret_only_inf__insert_unknown_unknown(<4 x half> %vec, half %elt, i32 %idx) {
+; CHECK-LABEL: define nofpclass(nan zero sub norm) <4 x half> @ret_only_inf__insert_unknown_unknown(
+; CHECK-SAME: <4 x half> [[VEC:%.*]], half [[ELT:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: [[INSERT:%.*]] = insertelement <4 x half> [[VEC]], half [[ELT]], i32 [[IDX]]
+; CHECK-NEXT: ret <4 x half> [[INSERT]]
+;
+ %insert = insertelement <4 x half> %vec, half %elt, i32 %idx
+ ret <4 x half> %insert
+}
+
+define nofpclass(nan ninf zero sub norm) <4 x half> @ret_only_pinf__insert_unknown_unknown(<4 x half> %vec, half %elt, i32 %idx) {
+; CHECK-LABEL: define nofpclass(nan ninf zero sub norm) <4 x half> @ret_only_pinf__insert_unknown_unknown(
+; CHECK-SAME: <4 x half> [[VEC:%.*]], half [[ELT:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: ret <4 x half> splat (half 0xH7C00)
+;
+ %insert = insertelement <4 x half> %vec, half %elt, i32 %idx
+ ret <4 x half> %insert
+}
+
+define nofpclass(nan pinf zero sub norm) <4 x half> @ret_only_ninf__insert_unknown_unknown(<4 x half> %vec, half %elt, i32 %idx) {
+; CHECK-LABEL: define nofpclass(nan pinf zero sub norm) <4 x half> @ret_only_ninf__insert_unknown_unknown(
+; CHECK-SAME: <4 x half> [[VEC:%.*]], half [[ELT:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: ret <4 x half> splat (half 0xHFC00)
+;
+ %insert = insertelement <4 x half> %vec, half %elt, i32 %idx
+ ret <4 x half> %insert
+}
+
+define nofpclass(nan inf sub norm) <4 x half> @ret_only_zero__insert_unknown_unknown(<4 x half> %vec, half %elt, i32 %idx) {
+; CHECK-LABEL: define nofpclass(nan inf sub norm) <4 x half> @ret_only_zero__insert_unknown_unknown(
+; CHECK-SAME: <4 x half> [[VEC:%.*]], half [[ELT:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: [[INSERT:%.*]] = insertelement <4 x half> [[VEC]], half [[ELT]], i32 [[IDX]]
+; CHECK-NEXT: ret <4 x half> [[INSERT]]
+;
+ %insert = insertelement <4 x half> %vec, half %elt, i32 %idx
+ ret <4 x half> %insert
+}
+
+define nofpclass(nan inf nzero sub norm) <4 x half> @ret_only_pzero__insert_unknown_unknown(<4 x half> %vec, half %elt, i32 %idx) {
+; CHECK-LABEL: define nofpclass(nan inf nzero sub norm) <4 x half> @ret_only_pzero__insert_unknown_unknown(
+; CHECK-SAME: <4 x half> [[VEC:%.*]], half [[ELT:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: ret <4 x half> zeroinitializer
+;
+ %insert = insertelement <4 x half> %vec, half %elt, i32 %idx
+ ret <4 x half> %insert
+}
+
+define nofpclass(nan inf pzero sub norm) <4 x half> @ret_only_nzero__insert_unknown_unknown(<4 x half> %vec, half %elt, i32 %idx) {
+; CHECK-LABEL: define nofpclass(nan inf pzero sub norm) <4 x half> @ret_only_nzero__insert_unknown_unknown(
+; CHECK-SAME: <4 x half> [[VEC:%.*]], half [[ELT:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: ret <4 x half> splat (half 0xH8000)
+;
+ %insert = insertelement <4 x half> %vec, half %elt, i32 %idx
+ ret <4 x half> %insert
+}
+
+; Remove select
+define nofpclass(nan) <4 x half> @ret_nonan__insert__select_unknown_nan__unknown(i1 %cond, <4 x half> %unknown.vec, half %elt, i32 %idx) {
+; CHECK-LABEL: define nofpclass(nan) <4 x half> @ret_nonan__insert__select_unknown_nan__unknown(
+; CHECK-SAME: i1 [[COND:%.*]], <4 x half> [[UNKNOWN_VEC:%.*]], half [[ELT:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: [[NAN_VEC:%.*]] = call <4 x half> @returns_nan()
+; CHECK-NEXT: [[INSERT:%.*]] = insertelement <4 x half> [[UNKNOWN_VEC]], half [[ELT]], i32 [[IDX]]
+; CHECK-NEXT: ret <4 x half> [[INSERT]]
+;
+ %nan.vec = call <4 x half> @returns_nan()
+ %select.vec = select i1 %cond, <4 x half> %unknown.vec, <4 x half> %nan.vec
+ %insert = insertelement <4 x half> %select.vec, half %elt, i32 %idx
+ ret <4 x half> %insert
+}
+
+; Remove select
+define nofpclass(nan) <4 x half> @ret_nonan__insert__unknown__select_unknown_nan(i1 %cond, <4 x half> %unknown.vec, half %unknown.elt, i32 %idx) {
+; CHECK-LABEL: define nofpclass(nan) <4 x half> @ret_nonan__insert__unknown__select_unknown_nan(
+; CHECK-SAME: i1 [[COND:%.*]], <4 x half> [[UNKNOWN_VEC:%.*]], half [[UNKNOWN_ELT:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: [[NAN_ELT:%.*]] = call half @returns_nan_f16()
+; CHECK-NEXT: [[INSERT:%.*]] = insertelement <4 x half> [[UNKNOWN_VEC]], half [[UNKNOWN_ELT]], i32 [[IDX]]
+; CHECK-NEXT: ret <4 x half> [[INSERT]]
+;
+ %nan.elt = call half @returns_nan_f16()
+ %select.elt = select i1 %cond, half %unknown.elt, half %nan.elt
+ %insert = insertelement <4 x half> %unknown.vec, half %select.elt, i32 %idx
+ ret <4 x half> %insert
+}
+
+; Remove selects
+define nofpclass(nan) <4 x half> @ret_nonan__insert__select_unknown_nan__select_unknown_nan(i1 %cond, <4 x half> %unknown.vec, half %unknown.elt, i32 %idx) {
+; CHECK-LABEL: define nofpclass(nan) <4 x half> @ret_nonan__insert__select_unknown_nan__select_unknown_nan(
+; CHECK-SAME: i1 [[COND:%.*]], <4 x half> [[UNKNOWN_VEC:%.*]], half [[UNKNOWN_ELT:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: [[NAN_VEC:%.*]] = call <4 x half> @returns_nan()
+; CHECK-NEXT: [[NAN_ELT:%.*]] = call half @returns_nan_f16()
+; CHECK-NEXT: [[INSERT:%.*]] = insertelement <4 x half> [[UNKNOWN_VEC]], half [[UNKNOWN_ELT]], i32 [[IDX]]
+; CHECK-NEXT: ret <4 x half> [[INSERT]]
+;
+ %nan.vec = call <4 x half> @returns_nan()
+ %nan.elt = call half @returns_nan_f16()
+ %select.elt = select i1 %cond, half %unknown.elt, half %nan.elt
+ %select.vec = select i1 %cond, <4 x half> %unknown.vec, <4 x half> %nan.vec
+ %insert = insertelement <4 x half> %select.vec, half %select.elt, i32 %idx
+ ret <4 x half> %insert
+}
+
+; Remove selects
+define nofpclass(nan) <vscale x 4 x half> @ret_nonan__insert__select_unknown_nan__select_unknown_nan__scalable(i1 %cond, <vscale x 4 x half> %unknown.vec, half %unknown.elt, i32 %idx) {
+; CHECK-LABEL: define nofpclass(nan) <vscale x 4 x half> @ret_nonan__insert__select_unknown_nan__select_unknown_nan__scalable(
+; CHECK-SAME: i1 [[COND:%.*]], <vscale x 4 x half> [[UNKNOWN_VEC:%.*]], half [[UNKNOWN_ELT:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: [[NAN_VEC:%.*]] = call <vscale x 4 x half> @returns_nan_nxv4f16()
+; CHECK-NEXT: [[NAN_ELT:%.*]] = call half @returns_nan_f16()
+; CHECK-NEXT: [[INSERT:%.*]] = insertelement <vscale x 4 x half> [[UNKNOWN_VEC]], half [[UNKNOWN_ELT]], i32 [[IDX]]
+; CHECK-NEXT: ret <vscale x 4 x half> [[INSERT]]
+;
+ %nan.vec = call <vscale x 4 x half> @returns_nan_nxv4f16()
+ %nan.elt = call half @returns_nan_f16()
+ %select.elt = select i1 %cond, half %unknown.elt, half %nan.elt
+ %select.vec = select i1 %cond, <vscale x 4 x half> %unknown.vec, <vscale x 4 x half> %nan.vec
+ %insert = insertelement <vscale x 4 x half> %select.vec, half %select.elt, i32 %idx
+ ret <vscale x 4 x half> %insert
+}
+
+; We cannot prove the source of the exp is nan due to the inserted
+; element, so it won't be rewritten to fadd
+define nofpclass(snan) <4 x half> @insert_unknown_taints_known_nan_vector_input(half %unknown.elt, i32 %idx) {
+; CHECK-LABEL: define nofpclass(snan) <4 x half> @insert_unknown_taints_known_nan_vector_input(
+; CHECK-SAME: half [[UNKNOWN_ELT:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: [[NAN_VEC:%.*]] = call <4 x half> @returns_nan()
+; CHECK-NEXT: [[INSERT:%.*]] = insertelement <4 x half> [[NAN_VEC]], half [[UNKNOWN_ELT]], i32 [[IDX]]
+; CHECK-NEXT: [[EXP:%.*]] = call <4 x half> @llvm.exp.v4f16(<4 x half> [[INSERT]])
+; CHECK-NEXT: ret <4 x half> [[EXP]]
+;
+ %nan.vec = call <4 x half> @returns_nan()
+ %insert = insertelement <4 x half> %nan.vec, half %unknown.elt, i32 %idx
+ %exp = call <4 x half> @llvm.exp.v4f16(<4 x half> %insert)
+ ret <4 x half> %exp
+}
+
+; Do not rewrite exp to fadd
+define nofpclass(snan) <4 x half> @insert_unknown_vector_taints_known_nan_vector_insert(<4 x half> %unknown.vec, i32 %idx) {
+; CHECK-LABEL: define nofpclass(snan) <4 x half> @insert_unknown_vector_taints_known_nan_vector_insert(
+; CHECK-SAME: <4 x half> [[UNKNOWN_VEC:%.*]], i32 [[IDX:%.*]]) {
+; CHECK-NEXT: [[NAN_ELT:%.*]] = call half @returns_nan_f16()
+; CHECK-NEXT: [[INSERT:%.*]] = insertelement <4 x half> [[UNKNOWN_VEC]], half [[NAN_ELT]], i32 [[IDX]]
+; CHECK-NEXT: [[EXP:%.*]] = call <4 x half> @llvm.exp.v4f16(<4 x half> [[INSERT]])
+; CHECK-NEXT: ret <4 x half> [[EXP]]
+;
+ %nan.elt = call half @returns_nan_f16()
+ %insert = insertelement <4 x half> %unknown.vec, half %nan.elt, i32 %idx
+ %exp = call <4 x half> @llvm.exp.v4f16(<4 x half> %insert)
+ ret <4 x half> %exp
+}
+
|
RKSimon
reviewed
Dec 31, 2025
llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
Outdated
Show resolved
Hide resolved
Eventually this should pull up the known elements logic from computeKnownFPClass.
arsenm
force-pushed
the
users/arsenm/instcombine/simplify-demanded-fp-class-fix-defining-undef-vector-elts
branch
from
a3fec0e to
984582e
Compare
arsenm
force-pushed
the
users/arsenm/instcombine/simplify-demanded-fpclass-insertelement
branch
from
fd1d31d to
7050ca2
Compare
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.
Eventually this should pull up the known elements logic from
computeKnownFPClass.