[llvm] 2c1b71a - [VPlan] Compute URem via APInt in materializeVectorTripCount (#203604)

[via llvm-commits]

Author: Florian Hahn
Date: 2026-06-12T19:04:23Z
New Revision: 2c1b71af78df9d7c22e225069971d3276d91d546

URL: https://github.com/llvm/llvm-project/commit/2c1b71af78df9d7c22e225069971d3276d91d546
DIFF: https://github.com/llvm/llvm-project/commit/2c1b71af78df9d7c22e225069971d3276d91d546.diff

LOG: [VPlan] Compute URem via APInt in materializeVectorTripCount (#203604)

materializeVectorTripCount has a shortcut for scalable steps: if the
constant trip count is divisible by the maximum possible runtime step,
the vector trip count equals the trip count directly. This called
APInt::getZExtValue unconditionally, which asserts when the constant
value needs more than 64 bits.

Compute the URem in APInt to fix the crash.

Added: 
    llvm/test/Transforms/LoopVectorize/RISCV/i128-trip-count-evl.ll

Modified: 
    llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 64d6c6fff256b..fe8edab40e846 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -5274,7 +5274,7 @@ void VPlanTransforms::materializeVectorTripCount(
   // and the vector trip count equals TC directly.
   const APInt *TCVal;
   if (!RequiresScalarEpilogue && match(TC, m_APInt(TCVal)) && MaxRuntimeStep &&
-      TCVal->getZExtValue() % *MaxRuntimeStep == 0) {
+      TCVal->urem(*MaxRuntimeStep) == 0) {
     VectorTC.replaceAllUsesWith(TC);
     return;
   }

diff  --git a/llvm/test/Transforms/LoopVectorize/RISCV/i128-trip-count-evl.ll b/llvm/test/Transforms/LoopVectorize/RISCV/i128-trip-count-evl.ll
new file mode 100644
index 0000000000000..4e7fefa61c806
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/i128-trip-count-evl.ll
@@ -0,0 +1,42 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 6
+; RUN: opt -passes=loop-vectorize -mtriple=riscv64 -mattr=+v \
+; RUN:   -force-tail-folding-style=data-with-evl \
+; RUN:   --tail-folding-policy=must-fold-tail -S %s | FileCheck %s
+
+define void @i128_trip_count(ptr %p) {
+; CHECK-LABEL: define void @i128_trip_count(
+; CHECK-SAME: ptr [[P:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    br label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i128 [ 0, %[[VECTOR_PH]] ], [ [[CURRENT_ITERATION_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[AVL:%.*]] = phi i128 [ 18446744073709551617, %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @llvm.experimental.get.vector.length.i128(i128 [[AVL]], i32 8, i1 true)
+; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i16, ptr [[P]], i128 [[INDEX]]
+; CHECK-NEXT:    call void @llvm.vp.store.nxv8i16.p0(<vscale x 8 x i16> splat (i16 1), ptr align 2 [[TMP1]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP0]])
+; CHECK-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP0]] to i128
+; CHECK-NEXT:    [[CURRENT_ITERATION_NEXT]] = add i128 [[TMP2]], [[INDEX]]
+; CHECK-NEXT:    [[AVL_NEXT]] = sub nuw i128 [[AVL]], [[TMP2]]
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp eq i128 [[AVL_NEXT]], 0
+; CHECK-NEXT:    br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK:       [[MIDDLE_BLOCK]]:
+; CHECK-NEXT:    br label %[[EXIT:.*]]
+; CHECK:       [[EXIT]]:
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %loop
+
+loop:
+  %iv = phi i128 [ 0, %entry ], [ %iv.next, %loop ]
+  %p.iv = getelementptr inbounds i16, ptr %p, i128 %iv
+  store i16 1, ptr %p.iv, align 2
+  %iv.next = add i128 %iv, 1
+  %ec = icmp eq i128 %iv.next, 18446744073709551617
+  br i1 %ec, label %exit, label %loop
+
+exit:
+  ret void
+}