PreISelIntrinsicLowering: Lower llvm.exp/llvm.exp2 to a loop if scalable vec arg #117568
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
|
@llvm/pr-subscribers-llvm-analysis @llvm/pr-subscribers-llvm-transforms Author: Stephen Long (steplong) ChangesFull diff: https://github.com/llvm/llvm-project/pull/117568.diff 2 Files Affected:
diff --git a/llvm/lib/CodeGen/PreISelIntrinsicLowering.cpp b/llvm/lib/CodeGen/PreISelIntrinsicLowering.cpp
index 4a3d1673c2a7c1..74f54e43a8386f 100644
--- a/llvm/lib/CodeGen/PreISelIntrinsicLowering.cpp
+++ b/llvm/lib/CodeGen/PreISelIntrinsicLowering.cpp
@@ -335,6 +335,59 @@ bool PreISelIntrinsicLowering::expandMemIntrinsicUses(Function &F) const {
return Changed;
}
+static bool lowerExpIntrinsicToLoop(Module &M, Function &F, CallInst *CI) {
+ ScalableVectorType *ScalableTy =
+ dyn_cast<ScalableVectorType>(F.getArg(0)->getType());
+ if (!ScalableTy) {
+ return false;
+ }
+
+ BasicBlock *PreLoopBB = CI->getParent();
+ BasicBlock *PostLoopBB = nullptr;
+ Function *ParentFunc = PreLoopBB->getParent();
+ LLVMContext &Ctx = PreLoopBB->getContext();
+
+ PostLoopBB = PreLoopBB->splitBasicBlock(CI);
+ BasicBlock *LoopBB = BasicBlock::Create(Ctx, "", ParentFunc, PostLoopBB);
+ PreLoopBB->getTerminator()->setSuccessor(0, LoopBB);
+
+ // loop preheader
+ IRBuilder<> PreLoopBuilder(PreLoopBB->getTerminator());
+ Value *VScale = PreLoopBuilder.CreateVScale(
+ ConstantInt::get(PreLoopBuilder.getInt64Ty(), 1));
+ Value *N = ConstantInt::get(PreLoopBuilder.getInt64Ty(),
+ ScalableTy->getMinNumElements());
+ Value *LoopEnd = PreLoopBuilder.CreateMul(VScale, N);
+
+ // loop body
+ IRBuilder<> LoopBuilder(LoopBB);
+ Type *Int64Ty = LoopBuilder.getInt64Ty();
+
+ PHINode *LoopIndex = LoopBuilder.CreatePHI(Int64Ty, 2);
+ LoopIndex->addIncoming(ConstantInt::get(Int64Ty, 0U), PreLoopBB);
+ PHINode *Vec = LoopBuilder.CreatePHI(ScalableTy, 2);
+ Vec->addIncoming(CI->getArgOperand(0), PreLoopBB);
+
+ Value *Elem = LoopBuilder.CreateExtractElement(Vec, LoopIndex);
+ Function *Exp = Intrinsic::getOrInsertDeclaration(
+ &M, Intrinsic::exp, ScalableTy->getElementType());
+ Value *Res = LoopBuilder.CreateCall(Exp, Elem);
+ Value *NewVec = LoopBuilder.CreateInsertElement(Vec, Res, LoopIndex);
+ Vec->addIncoming(NewVec, LoopBB);
+
+ Value *One = ConstantInt::get(Int64Ty, 1U);
+ Value *NextLoopIndex = LoopBuilder.CreateAdd(LoopIndex, One);
+ LoopIndex->addIncoming(NextLoopIndex, LoopBB);
+
+ Value *ExitCond =
+ LoopBuilder.CreateICmp(CmpInst::ICMP_EQ, NextLoopIndex, LoopEnd);
+ LoopBuilder.CreateCondBr(ExitCond, PostLoopBB, LoopBB);
+
+ CI->replaceAllUsesWith(NewVec);
+ CI->eraseFromParent();
+ return true;
+}
+
bool PreISelIntrinsicLowering::lowerIntrinsics(Module &M) const {
bool Changed = false;
for (Function &F : M) {
@@ -453,6 +506,10 @@ bool PreISelIntrinsicLowering::lowerIntrinsics(Module &M) const {
case Intrinsic::objc_sync_exit:
Changed |= lowerObjCCall(F, "objc_sync_exit");
break;
+ case Intrinsic::exp:
+ Changed |= forEachCall(
+ F, [&](CallInst *CI) { return lowerExpIntrinsicToLoop(M, F, CI); });
+ break;
}
}
return Changed;
diff --git a/llvm/test/Transforms/PreISelIntrinsicLowering/expand-exp.ll b/llvm/test/Transforms/PreISelIntrinsicLowering/expand-exp.ll
new file mode 100644
index 00000000000000..6ad181033f233a
--- /dev/null
+++ b/llvm/test/Transforms/PreISelIntrinsicLowering/expand-exp.ll
@@ -0,0 +1,23 @@
+; RUN: opt -passes=pre-isel-intrinsic-lowering -S < %s | FileCheck %s
+
+define <vscale x 4 x float> @softmax_kernel() {
+; CHECK-LABEL: define <vscale x 4 x float> @softmax_kernel(
+; CHECK-NEXT: [[VSCALE:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[LOOPEND:%.*]] = mul i64 [[VSCALE]], 4
+; CHECK-NEXT: br label %[[LOOPBODY:.*]]
+; CHECK: [[LOOPBODY]]:
+; CHECK-NEXT: [[IDX:%.*]] = phi i64 [ 0, %0 ], [ [[NEW_IDX:%.*]], %[[LOOPBODY]] ]
+; CHECK-NEXT: [[VEC:%.*]] = phi <vscale x 4 x float> [ zeroinitializer, %0 ], [ [[NEW_VEC:.*]], %[[LOOPBODY]] ]
+; CHECK-NEXT: [[ELEM:%.*]] = extractelement <vscale x 4 x float> [[VEC]], i64 [[IDX]]
+; CHECK-NEXT: [[RES:%.*]] = call float @llvm.exp.f32(float [[ELEM]])
+; CHECK-NEXT: [[NEW_VEC:%.*]] = insertelement <vscale x 4 x float> [[VEC]], float [[RES]], i64 [[IDX]]
+; CHECK-NEXT: [[NEW_IDX]] = add i64 [[IDX]], 1
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[NEW_IDX]], [[LOOPEND]]
+; CHECK-NEXT: br i1 [[CMP]], label %[[LOOPEXIT:.*]], label %[[LOOPBODY]]
+; CHECK: [[LOOPEXIT]]:
+; CHECK-NEXT: ret <vscale x 4 x float> [[NEW_VEC]]
+ %1 = call <vscale x 4 x float> @llvm.exp.nxv4f32(<vscale x 4 x float> zeroinitializer)
+ ret <vscale x 4 x float> %1
+}
+
+declare <vscale x 4 x float> @llvm.exp.nxv4f32(<vscale x 4 x float>)
|
arsenm
requested changes
Nov 25, 2024
arsenm
reviewed
Dec 4, 2024
arsenm
reviewed
Dec 4, 2024
llvm/test/Transforms/PreISelIntrinsicLowering/AArch64/expand-exp.ll
Outdated
Show resolved
Hide resolved
|
@arsenm I added the fixed vector test case, but if we decide to only do this for scalable vectors, then I'll remove it. Let me know how we should do the legality check as I'm new to this. |
steplong
changed the title
|
Hi, what can be done to get this in? For the fixed vector case, it looks like it's being handled by SelectionDAG, specifically |
arsenm
reviewed
Jan 23, 2025
…ble vec arg If the argument to the intrinsic call to llvm.exp and llvm.exp2 is a scalable vector, lower it into a loop in PreISelIntrinsicLowering. If it is a fixed vector, let SelectionDAG handle it.
llvmbot
added
the
llvm:analysis
arsenm
reviewed
Jan 24, 2025
arsenm
approved these changes
Jan 24, 2025
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/50/builds/9373 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/64/builds/2086 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/146/builds/2158 Here is the relevant piece of the build log for the reference |
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.
No description provided.