[llvm] [VPlan] Fix crash with EVL tail folding intrinsic with no corresponding VP (PR #121542)

[Luke Lau via llvm-commits]

https://github.com/lukel97 created https://github.com/llvm/llvm-project/pull/121542

This fixes a crash when building SPEC CPU 2017 with EVL tail folding when widening @llvm.log10 intrinsics.

@llvm.log10 and some other intrinsics don't have a corresponding VP intrinsic, so this fixes the crash by removing the assert and bailing instead.


>From bc50f809bbaf7afcbdfd5225a7528c0f281b3524 Mon Sep 17 00:00:00 2001
From: Luke Lau <luke at igalia.com>
Date: Fri, 3 Jan 2025 12:19:45 +0800
Subject: [PATCH] [VPlan] Fix crash with EVL tail folding intrinsic with no
 corresponding VP

This fixes a crash when building SPEC CPU 2017 with EVL tail folding when widening @llvm.log10 intrinsics.

@llvm.log10 and some other intrinsics don't have a corresponding VP intrinsic, so this fixes the crash by removing the assert and bailing instead.
---
 .../Transforms/Vectorize/VPlanTransforms.cpp  |  5 +-
 ...ize-force-tail-with-evl-call-intrinsics.ll | 56 +++++++++++++++++++
 2 files changed, 59 insertions(+), 2 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 8ac2bd5160c267..4a9329269af8b8 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -1498,10 +1498,11 @@ static VPRecipeBase *createEVLRecipe(VPValue *HeaderMask,
               auto *CastR = cast<VPWidenCastRecipe>(CR);
               VPID = VPIntrinsic::getForOpcode(CastR->getOpcode());
             }
-            assert(VPID != Intrinsic::not_intrinsic && "Expected VP intrinsic");
+            if (VPID == Intrinsic::not_intrinsic)
+              return nullptr;
             assert(VPIntrinsic::getMaskParamPos(VPID) &&
                    VPIntrinsic::getVectorLengthParamPos(VPID) &&
-                   "Expected VP intrinsic");
+                   "Expected VP intrinsic to have mask and EVL");
 
             SmallVector<VPValue *> Ops(CR->operands());
             Ops.push_back(&AllOneMask);
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-call-intrinsics.ll b/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-call-intrinsics.ll
index 11cf832c8abbf2..f07aaecfa84678 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-call-intrinsics.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-call-intrinsics.ll
@@ -989,6 +989,62 @@ exit:
   ret void
 }
 
+; There's no @llvm.vp.log10, so don't transform it.
+define void @log10(ptr %a, ptr %b, i64 %N) {
+; IF-EVL-LABEL: define void @log10(
+; IF-EVL-SAME: ptr [[A:%.*]], ptr [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
+; IF-EVL-NEXT:  [[ENTRY:.*]]:
+; IF-EVL-NEXT:    br label %[[LOOP:.*]]
+; IF-EVL:       [[LOOP]]:
+; IF-EVL-NEXT:    [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], %[[LOOP]] ], [ 0, %[[ENTRY]] ]
+; IF-EVL-NEXT:    [[GEP:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[IV]]
+; IF-EVL-NEXT:    [[TMP0:%.*]] = load float, ptr [[GEP]], align 4
+; IF-EVL-NEXT:    [[COND:%.*]] = tail call float @llvm.log10.f32(float [[TMP0]])
+; IF-EVL-NEXT:    [[GEP9:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]]
+; IF-EVL-NEXT:    store float [[COND]], ptr [[GEP9]], align 4
+; IF-EVL-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
+; IF-EVL-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
+; IF-EVL-NEXT:    br i1 [[EXITCOND_NOT]], label %[[EXIT:.*]], label %[[LOOP]]
+; IF-EVL:       [[EXIT]]:
+; IF-EVL-NEXT:    ret void
+;
+; NO-VP-LABEL: define void @log10(
+; NO-VP-SAME: ptr [[A:%.*]], ptr [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
+; NO-VP-NEXT:  [[ENTRY:.*]]:
+; NO-VP-NEXT:    br label %[[LOOP:.*]]
+; NO-VP:       [[LOOP]]:
+; NO-VP-NEXT:    [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], %[[LOOP]] ], [ 0, %[[ENTRY]] ]
+; NO-VP-NEXT:    [[GEP:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[IV]]
+; NO-VP-NEXT:    [[TMP0:%.*]] = load float, ptr [[GEP]], align 4
+; NO-VP-NEXT:    [[COND:%.*]] = tail call float @llvm.log10.f32(float [[TMP0]])
+; NO-VP-NEXT:    [[GEP9:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]]
+; NO-VP-NEXT:    store float [[COND]], ptr [[GEP9]], align 4
+; NO-VP-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
+; NO-VP-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
+; NO-VP-NEXT:    br i1 [[EXITCOND_NOT]], label %[[EXIT:.*]], label %[[LOOP]]
+; NO-VP:       [[EXIT]]:
+; NO-VP-NEXT:    ret void
+;
+
+entry:
+  br label %loop
+
+loop:
+  %iv = phi i64 [ %iv.next, %loop ], [ 0, %entry ]
+  %gep = getelementptr inbounds float, ptr %b, i64 %iv
+  %0 = load float, ptr %gep, align 4
+  %cond = tail call float @llvm.log10.f32(float %0)
+  %gep9 = getelementptr inbounds float, ptr %a, i64 %iv
+  store float %cond, ptr %gep9, align 4
+  %iv.next = add nuw nsw i64 %iv, 1
+  %exitcond.not = icmp eq i64 %iv.next, %N
+  br i1 %exitcond.not, label %exit, label %loop
+
+exit:
+  ret void
+}
+
+
 declare i32 @llvm.smax.i32(i32, i32)
 declare i32 @llvm.smin.i32(i32, i32)
 declare i32 @llvm.umax.i32(i32, i32)