[InstSimplify] Add basic constant folding for llvm.sincos
#114527
Merged
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
This calls into the existing constant folding for `llvm.sin` and `llvm.cos`, which currently does not fold for any non-finite values, so most tests are negative tests.
|
@llvm/pr-subscribers-llvm-analysis @llvm/pr-subscribers-llvm-transforms Author: Benjamin Maxwell (MacDue) ChangesThis calls into the existing constant folding for Full diff: https://github.com/llvm/llvm-project/pull/114527.diff 2 Files Affected:
diff --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
index c5a2c2f52f8dc2..ae327340aeec66 100644
--- a/llvm/lib/Analysis/ConstantFolding.cpp
+++ b/llvm/lib/Analysis/ConstantFolding.cpp
@@ -1567,6 +1567,7 @@ bool llvm::canConstantFoldCallTo(const CallBase *Call, const Function *F) {
case Intrinsic::sqrt:
case Intrinsic::sin:
case Intrinsic::cos:
+ case Intrinsic::sincos:
case Intrinsic::pow:
case Intrinsic::powi:
case Intrinsic::ldexp:
@@ -3466,6 +3467,44 @@ ConstantFoldStructCall(StringRef Name, Intrinsic::ID IntrinsicID,
return nullptr;
return ConstantStruct::get(StTy, Result0, Result1);
}
+ case Intrinsic::sincos: {
+ Type *Ty = StTy->getContainedType(0);
+ Type *TyScalar = Ty->getScalarType();
+
+ auto ConstantFoldScalarSincosCall =
+ [&](Constant *Op) -> std::pair<Constant *, Constant *> {
+ Constant *SinResult =
+ ConstantFoldScalarCall(Name, Intrinsic::sin, TyScalar, Op, TLI, Call);
+ if (!SinResult)
+ return {};
+ Constant *CosResult =
+ ConstantFoldScalarCall(Name, Intrinsic::cos, TyScalar, Op, TLI, Call);
+ if (!CosResult)
+ return {};
+ return std::make_pair(SinResult, CosResult);
+ };
+
+ if (auto *FVTy = dyn_cast<FixedVectorType>(Ty)) {
+ SmallVector<Constant *, 4> SinResults(FVTy->getNumElements());
+ SmallVector<Constant *, 4> CosResults(FVTy->getNumElements());
+
+ for (unsigned I = 0, E = FVTy->getNumElements(); I != E; ++I) {
+ Constant *Lane = Operands[0]->getAggregateElement(I);
+ std::tie(SinResults[I], CosResults[I]) =
+ ConstantFoldScalarSincosCall(Lane);
+ if (!SinResults[I])
+ return nullptr;
+ }
+
+ return ConstantStruct::get(StTy, ConstantVector::get(SinResults),
+ ConstantVector::get(CosResults));
+ }
+
+ auto [SinResult, CosResult] = ConstantFoldScalarSincosCall(Operands[0]);
+ if (!SinResult)
+ return nullptr;
+ return ConstantStruct::get(StTy, SinResult, CosResult);
+ }
default:
// TODO: Constant folding of vector intrinsics that fall through here does
// not work (e.g. overflow intrinsics)
diff --git a/llvm/test/Transforms/InstSimplify/sincos.ll b/llvm/test/Transforms/InstSimplify/sincos.ll
new file mode 100644
index 00000000000000..c6dc1357b4197a
--- /dev/null
+++ b/llvm/test/Transforms/InstSimplify/sincos.ll
@@ -0,0 +1,125 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 3
+; RUN: opt -S -passes=instsimplify %s | FileCheck %s
+
+define { float, float } @sincos_zero() {
+; CHECK-LABEL: define { float, float } @sincos_zero() {
+; CHECK-NEXT: ret { float, float } { float 0.000000e+00, float 1.000000e+00 }
+;
+ %ret = call { float, float } @llvm.sincos.f32(float zeroinitializer)
+ ret { float, float } %ret
+}
+
+define { float, float } @sincos_neg_zero() {
+; CHECK-LABEL: define { float, float } @sincos_neg_zero() {
+; CHECK-NEXT: ret { float, float } { float -0.000000e+00, float 1.000000e+00 }
+;
+ %ret = call { float, float } @llvm.sincos.f32(float -0.0)
+ ret { float, float } %ret
+}
+
+define { <2 x float>, <2 x float> } @sincos_zero_vector() {
+; CHECK-LABEL: define { <2 x float>, <2 x float> } @sincos_zero_vector() {
+; CHECK-NEXT: ret { <2 x float>, <2 x float> } { <2 x float> zeroinitializer, <2 x float> <float 1.000000e+00, float 1.000000e+00> }
+;
+ %ret = call { <2 x float>, <2 x float> } @llvm.sincos.v2f32(<2 x float> zeroinitializer)
+ ret { <2 x float>, <2 x float> } %ret
+}
+
+define { float, float } @sincos_poison() {
+; CHECK-LABEL: define { float, float } @sincos_poison() {
+; CHECK-NEXT: [[RET:%.*]] = call { float, float } @llvm.sincos.f32(float poison)
+; CHECK-NEXT: ret { float, float } [[RET]]
+;
+ %ret = call { float, float } @llvm.sincos.f32(float poison)
+ ret { float, float } %ret
+}
+
+define { <2 x float>, <2 x float> } @sincos_poison_vector() {
+; CHECK-LABEL: define { <2 x float>, <2 x float> } @sincos_poison_vector() {
+; CHECK-NEXT: [[RET:%.*]] = call { <2 x float>, <2 x float> } @llvm.sincos.v2f32(<2 x float> poison)
+; CHECK-NEXT: ret { <2 x float>, <2 x float> } [[RET]]
+;
+ %ret = call { <2 x float>, <2 x float> } @llvm.sincos.v2f32(<2 x float> poison)
+ ret { <2 x float>, <2 x float> } %ret
+}
+
+define { <vscale x 2 x float>, <vscale x 2 x float> } @sincos_poison_scalable_vector() {
+; CHECK-LABEL: define { <vscale x 2 x float>, <vscale x 2 x float> } @sincos_poison_scalable_vector() {
+; CHECK-NEXT: [[RET:%.*]] = call { <vscale x 2 x float>, <vscale x 2 x float> } @llvm.sincos.nxv2f32(<vscale x 2 x float> poison)
+; CHECK-NEXT: ret { <vscale x 2 x float>, <vscale x 2 x float> } [[RET]]
+;
+ %ret = call { <vscale x 2 x float>, <vscale x 2 x float> } @llvm.sincos.nxv2f32(<vscale x 2 x float> poison)
+ ret { <vscale x 2 x float>, <vscale x 2 x float> } %ret
+}
+
+define { float, float } @sincos_undef() {
+; CHECK-LABEL: define { float, float } @sincos_undef() {
+; CHECK-NEXT: [[RET:%.*]] = call { float, float } @llvm.sincos.f32(float undef)
+; CHECK-NEXT: ret { float, float } [[RET]]
+;
+ %ret = call { float, float } @llvm.sincos.f32(float undef)
+ ret { float, float } %ret
+}
+
+define { <2 x float>, <2 x float> } @sincos_undef_vector() {
+; CHECK-LABEL: define { <2 x float>, <2 x float> } @sincos_undef_vector() {
+; CHECK-NEXT: [[RET:%.*]] = call { <2 x float>, <2 x float> } @llvm.sincos.v2f32(<2 x float> undef)
+; CHECK-NEXT: ret { <2 x float>, <2 x float> } [[RET]]
+;
+ %ret = call { <2 x float>, <2 x float> } @llvm.sincos.v2f32(<2 x float> undef)
+ ret { <2 x float>, <2 x float> } %ret
+}
+
+define { <vscale x 2 x float>, <vscale x 2 x float> } @sincos_undef_scalable_vector() {
+; CHECK-LABEL: define { <vscale x 2 x float>, <vscale x 2 x float> } @sincos_undef_scalable_vector() {
+; CHECK-NEXT: [[RET:%.*]] = call { <vscale x 2 x float>, <vscale x 2 x float> } @llvm.sincos.nxv2f32(<vscale x 2 x float> undef)
+; CHECK-NEXT: ret { <vscale x 2 x float>, <vscale x 2 x float> } [[RET]]
+;
+ %ret = call { <vscale x 2 x float>, <vscale x 2 x float> } @llvm.sincos.nxv2f32(<vscale x 2 x float> undef)
+ ret { <vscale x 2 x float>, <vscale x 2 x float> } %ret
+}
+
+define { <vscale x 2 x float>, <vscale x 2 x float> } @sincos_zero_scalable_vector() {
+; CHECK-LABEL: define { <vscale x 2 x float>, <vscale x 2 x float> } @sincos_zero_scalable_vector() {
+; CHECK-NEXT: [[RET:%.*]] = call { <vscale x 2 x float>, <vscale x 2 x float> } @llvm.sincos.nxv2f32(<vscale x 2 x float> zeroinitializer)
+; CHECK-NEXT: ret { <vscale x 2 x float>, <vscale x 2 x float> } [[RET]]
+;
+ %ret = call { <vscale x 2 x float>, <vscale x 2 x float> } @llvm.sincos.nxv2f32(<vscale x 2 x float> zeroinitializer)
+ ret { <vscale x 2 x float>, <vscale x 2 x float> } %ret
+}
+
+define { float, float } @sincos_inf() {
+; CHECK-LABEL: define { float, float } @sincos_inf() {
+; CHECK-NEXT: [[RET:%.*]] = call { float, float } @llvm.sincos.f32(float 0x7FF0000000000000)
+; CHECK-NEXT: ret { float, float } [[RET]]
+;
+ %ret = call { float, float } @llvm.sincos.f32(float 0x7FF0000000000000)
+ ret { float, float } %ret
+}
+
+define { float, float } @sincos_neginf() {
+; CHECK-LABEL: define { float, float } @sincos_neginf() {
+; CHECK-NEXT: [[RET:%.*]] = call { float, float } @llvm.sincos.f32(float 0xFFF0000000000000)
+; CHECK-NEXT: ret { float, float } [[RET]]
+;
+ %ret = call { float, float } @llvm.sincos.f32(float 0xFFF0000000000000)
+ ret { float, float } %ret
+}
+
+define { float, float } @sincos_qnam() {
+; CHECK-LABEL: define { float, float } @sincos_qnam() {
+; CHECK-NEXT: [[RET:%.*]] = call { float, float } @llvm.sincos.f32(float 0x7FF8000000000000)
+; CHECK-NEXT: ret { float, float } [[RET]]
+;
+ %ret = call { float, float } @llvm.sincos.f32(float 0x7FF8000000000000)
+ ret { float, float } %ret
+}
+
+define { float, float } @sincos_snam() {
+; CHECK-LABEL: define { float, float } @sincos_snam() {
+; CHECK-NEXT: [[RET:%.*]] = call { float, float } @llvm.sincos.f32(float 0x7FF0000020000000)
+; CHECK-NEXT: ret { float, float } [[RET]]
+;
+ %ret = call { float, float } @llvm.sincos.f32(float bitcast (i32 2139095041 to float))
+ ret { float, float } %ret
+}
|
|
Could you add something about |
Sure, I don't believe constant folding requires |
|
I thought about exploiting the flag. |
You mean constant folding differently based on if |
Sure why not. Or alternatively at least stating, we ignore |
arsenm
reviewed
Nov 1, 2024
arsenm
reviewed
Nov 4, 2024
| %ret = call { float, float } @llvm.sincos.f32(float -0.0) | ||
| ret { float, float } %ret | ||
| } | ||
|
|
There was a problem hiding this comment.
I've added some normal value tests using matching the result values (and checking they're used consistently). I avoided this initially as the result for most normal values is displayed in hexadecimal (like 0x3FED18F6E0000000), which didn't seem that helpful.
|
Kind ping 🙂 |
arsenm
reviewed
Nov 22, 2024
Comment on lines
+105
to
+112
| define { <vscale x 2 x float>, <vscale x 2 x float> } @sincos_zero_scalable_vector() { | ||
| ; CHECK-LABEL: define { <vscale x 2 x float>, <vscale x 2 x float> } @sincos_zero_scalable_vector() { | ||
| ; CHECK-NEXT: [[RET:%.*]] = call { <vscale x 2 x float>, <vscale x 2 x float> } @llvm.sincos.nxv2f32(<vscale x 2 x float> zeroinitializer) | ||
| ; CHECK-NEXT: ret { <vscale x 2 x float>, <vscale x 2 x float> } [[RET]] | ||
| ; | ||
| %ret = call { <vscale x 2 x float>, <vscale x 2 x float> } @llvm.sincos.nxv2f32(<vscale x 2 x float> zeroinitializer) | ||
| ret { <vscale x 2 x float>, <vscale x 2 x float> } %ret | ||
| } |
There was a problem hiding this comment.
In a follow up, should probably make sure the zero, undef, and splat cases work for scalable vectors
arsenm
approved these changes
Dec 8, 2024
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/180/builds/9685 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/144/builds/13426 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/56/builds/13989 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/60/builds/14660 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/140/builds/12596 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/160/builds/9687 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/3/builds/8828 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/65/builds/9010 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/137/builds/9940 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/185/builds/9816 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/153/builds/16913 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/175/builds/9814 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/133/builds/8057 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/154/builds/8683 Here is the relevant piece of the build log for the reference |
MacDue
added a commit
that referenced
this pull request
Dec 8, 2024
…119149) Reverts #114527 Reverting due to buildbot failures (e.g. https://lab.llvm.org/buildbot/#/builders/180/builds/9685)
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/27/builds/3144 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/16/builds/10292 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/174/builds/9722 Here is the relevant piece of the build log for the reference |
|
Reverted this for now (#119149). I'll look into the buildbot issues and reland tomorrow. |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/46/builds/9157 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/108/builds/6864 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/190/builds/10948 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/11/builds/9396 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/33/builds/7883 Here is the relevant piece of the build log for the reference |
MacDue
added a commit
that referenced
this pull request
Dec 9, 2024
…119192) This calls into the existing constant folding for `llvm.sin` and `llvm.cos`, which currently does not fold for any non-finite values, so most tests are negative tests at the moment. Note: The constant folding does not consider the `afn` fast-math flag and will produce the same result regardless of if the flag is set. This is a reland of #114527 that updates the syntax of one of the tests from: `<float 1.000000e+00, float 1.000000e+00>` to `splat (float 1.000000e+00)`.
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.
This calls into the existing constant folding for
llvm.sinandllvm.cos, which currently does not fold for any non-finite values, so most tests are negative tests at the moment.Note: The constant folding does not consider the
afnfast-math flag and will produce the same result regardless of if the flag is set.