[llvm] [VPlan] Expand SCEV casts in VPSCEVExpander. (PR #206371)
Florian Hahn via llvm-commits
llvm-commits at lists.llvm.org
Sun Jun 28 13:50:23 PDT 2026
https://github.com/fhahn created https://github.com/llvm/llvm-project/pull/206371
Add support for expanding scTruncate, scZeroExtend and scSignExtend SCEV expressions to Trunc/ZExt/SExt VPInstructions in VPSCEVExpander::tryToExpand.
>From 12660cb1e12440bcf4ed4ab5ca5c9e1d34c85221 Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Sat, 27 Jun 2026 17:07:57 +0100
Subject: [PATCH] [VPlan] Expand SCEV casts in VPSCEVExpander.
Add support for expanding scTruncate, scZeroExtend and scSignExtend
SCEV expressions to Trunc/ZExt/SExt VPInstructions in
VPSCEVExpander::tryToExpand.
---
llvm/lib/Transforms/Vectorize/VPlanUtils.cpp | 23 +++++++++++++++++++
.../AArch64/low_trip_count_predicates.ll | 2 +-
.../LoopVectorize/VPlan/expand-scev.ll | 10 ++++----
.../illegal-parallel-loop-uniform-write.ll | 2 +-
.../LoopVectorize/induction-step.ll | 6 ++---
.../nested-loops-scev-expansion.ll | 4 ++--
.../runtime-checks-difference.ll | 8 +++----
7 files changed, 38 insertions(+), 17 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanUtils.cpp b/llvm/lib/Transforms/Vectorize/VPlanUtils.cpp
index 263480924b796..68adce6105911 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanUtils.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanUtils.cpp
@@ -878,6 +878,29 @@ VPValue *VPSCEVExpander::tryToExpand(const SCEV *S) {
VPIRFlags::getDefaultFlags(Instruction::UDiv),
DL);
}
+ case scTruncate:
+ case scZeroExtend:
+ case scSignExtend: {
+ auto *Cast = cast<SCEVCastExpr>(S);
+ VPValue *Op = tryToExpand(Cast->getOperand());
+ if (!Op)
+ return nullptr;
+ Instruction::CastOps Opcode;
+ switch (S->getSCEVType()) {
+ case scTruncate:
+ Opcode = Instruction::Trunc;
+ break;
+ case scZeroExtend:
+ Opcode = Instruction::ZExt;
+ break;
+ case scSignExtend:
+ Opcode = Instruction::SExt;
+ break;
+ default:
+ llvm_unreachable("not a cast SCEV");
+ }
+ return Builder.createScalarCast(Opcode, Op, S->getType(), DL);
+ }
default:
return nullptr;
}
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/low_trip_count_predicates.ll b/llvm/test/Transforms/LoopVectorize/AArch64/low_trip_count_predicates.ll
index 1231ca7a320c5..f123ed20e205c 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/low_trip_count_predicates.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/low_trip_count_predicates.ll
@@ -377,7 +377,7 @@ define void @overflow_indvar_known_false(ptr nocapture noundef %p, i32 noundef %
; CHECK-NEXT: [[TMP0:%.*]] = zext nneg i32 [[TC]] to i64
; CHECK-NEXT: [[TMP19:%.*]] = add i32 [[TC]], 1
; CHECK-NEXT: [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
-; CHECK-NEXT: [[TMP1:%.*]] = sub i64 1028, [[TMP20]]
+; CHECK-NEXT: [[TMP1:%.*]] = sub nsw i64 1028, [[TMP20]]
; CHECK-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
; CHECK: [[VECTOR_SCEVCHECK]]:
; CHECK-NEXT: [[TMP21:%.*]] = add i32 [[TC]], 1
diff --git a/llvm/test/Transforms/LoopVectorize/VPlan/expand-scev.ll b/llvm/test/Transforms/LoopVectorize/VPlan/expand-scev.ll
index fec00b433bbaf..04bffbb846de3 100644
--- a/llvm/test/Transforms/LoopVectorize/VPlan/expand-scev.ll
+++ b/llvm/test/Transforms/LoopVectorize/VPlan/expand-scev.ll
@@ -146,11 +146,10 @@ exit:
define void @scev_zext_expanded(ptr %dst, i32 %a, i32 %b) {
; CHECK-LABEL: VPlan for loop in 'scev_zext_expanded'
; CHECK: VPlan 'Final VPlan for VF={4},UF={1}' {
-; CHECK-NEXT: Live-in ir<%0> = original trip-count
; CHECK-EMPTY:
; CHECK-NEXT: ir-bb<loop.preheader>:
-; CHECK-NEXT: IR %0 = zext i32 %p to i64
-; CHECK-NEXT: EMIT vp<%min.iters.check> = icmp ult ir<%0>, ir<4>
+; CHECK-NEXT: EMIT-SCALAR vp<[[VP2:%[0-9]+]]> = zext ir<%p> to i64
+; CHECK-NEXT: EMIT vp<%min.iters.check> = icmp ult vp<[[VP2]]>, ir<4>
; CHECK-NEXT: EMIT branch-on-cond vp<%min.iters.check>
; CHECK-NEXT: Successor(s): ir-bb<scalar.ph>, vector.ph
;
@@ -176,11 +175,10 @@ exit:
define void @scev_sext_expanded(ptr %dst, i32 %a, i32 %b) {
; CHECK-LABEL: VPlan for loop in 'scev_sext_expanded'
; CHECK: VPlan 'Final VPlan for VF={4},UF={1}' {
-; CHECK-NEXT: Live-in ir<%0> = original trip-count
; CHECK-EMPTY:
; CHECK-NEXT: ir-bb<loop.preheader>:
-; CHECK-NEXT: IR %0 = sext i32 %p to i64
-; CHECK-NEXT: EMIT vp<%min.iters.check> = icmp ult ir<%0>, ir<4>
+; CHECK-NEXT: EMIT-SCALAR vp<[[VP2:%[0-9]+]]> = sext ir<%p> to i64
+; CHECK-NEXT: EMIT vp<%min.iters.check> = icmp ult vp<[[VP2]]>, ir<4>
; CHECK-NEXT: EMIT branch-on-cond vp<%min.iters.check>
; CHECK-NEXT: Successor(s): ir-bb<scalar.ph>, vector.ph
;
diff --git a/llvm/test/Transforms/LoopVectorize/X86/illegal-parallel-loop-uniform-write.ll b/llvm/test/Transforms/LoopVectorize/X86/illegal-parallel-loop-uniform-write.ll
index 0713e1b9736bf..cbe6ef47c7284 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/illegal-parallel-loop-uniform-write.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/illegal-parallel-loop-uniform-write.ll
@@ -26,7 +26,6 @@ define void @foo(ptr nocapture %a, ptr nocapture %b, i32 %k, i32 %m) #0 {
; CHECK: for.body3.lr.ph.us.preheader:
; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[M]], -1
; CHECK-NEXT: [[TMP1:%.*]] = zext i32 [[K:%.*]] to i64
-; CHECK-NEXT: [[TMP2:%.*]] = zext i32 [[M]] to i64
; CHECK-NEXT: br label [[FOR_BODY3_LR_PH_US:%.*]]
; CHECK: for.end.us:
; CHECK-NEXT: [[ARRAYIDX9_US:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 [[INDVARS_IV33:%.*]]
@@ -57,6 +56,7 @@ define void @foo(ptr nocapture %a, ptr nocapture %b, i32 %k, i32 %m) #0 {
; CHECK-NEXT: [[TMP8:%.*]] = trunc i64 [[INDVARS_IV33]] to i32
; CHECK-NEXT: [[ADD_US]] = add i32 [[TMP8]], [[K]]
; CHECK-NEXT: [[ARRAYIDX7_US]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV33]]
+; CHECK-NEXT: [[TMP2:%.*]] = zext i32 [[M]] to i64
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH]], label [[VECTOR_SCEVCHECK:%.*]]
; CHECK: vector.scevcheck:
diff --git a/llvm/test/Transforms/LoopVectorize/induction-step.ll b/llvm/test/Transforms/LoopVectorize/induction-step.ll
index ea2162b1a7fd6..73f4df69a6754 100644
--- a/llvm/test/Transforms/LoopVectorize/induction-step.ll
+++ b/llvm/test/Transforms/LoopVectorize/induction-step.ll
@@ -106,14 +106,14 @@ define i32 @induction_with_loop_inv(i32 %init, ptr noalias nocapture %A, i32 %N,
; CHECK-LABEL: define i32 @induction_with_loop_inv(
; CHECK-SAME: i32 [[INIT:%.*]], ptr noalias captures(none) [[A:%.*]], i32 [[N:%.*]], i32 [[M:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*]]:
-; CHECK-NEXT: [[TMP3:%.*]] = add i32 [[N]], -1
-; CHECK-NEXT: [[TMP11:%.*]] = zext i32 [[TMP3]] to i64
-; CHECK-NEXT: [[TMP0:%.*]] = add nuw nsw i64 [[TMP11]], 1
; CHECK-NEXT: br label %[[OUTER_HEADER:.*]]
; CHECK: [[OUTER_HEADER]]:
; CHECK-NEXT: [[INDVARS_IV15:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[INDVARS_IV_NEXT16:%.*]], %[[OUTER_LATCH:.*]] ]
; CHECK-NEXT: [[J_012:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[INC5:%.*]], %[[OUTER_LATCH]] ]
; CHECK-NEXT: [[X_011:%.*]] = phi i32 [ [[INIT]], %[[ENTRY]] ], [ [[X_0_LCSSA:%.*]], %[[OUTER_LATCH]] ]
+; CHECK-NEXT: [[TMP3:%.*]] = add i32 [[N]], -1
+; CHECK-NEXT: [[TMP11:%.*]] = zext i32 [[TMP3]] to i64
+; CHECK-NEXT: [[TMP0:%.*]] = add nuw nsw i64 [[TMP11]], 1
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 8
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
diff --git a/llvm/test/Transforms/LoopVectorize/nested-loops-scev-expansion.ll b/llvm/test/Transforms/LoopVectorize/nested-loops-scev-expansion.ll
index 8d3a29b931230..7dc1228346873 100644
--- a/llvm/test/Transforms/LoopVectorize/nested-loops-scev-expansion.ll
+++ b/llvm/test/Transforms/LoopVectorize/nested-loops-scev-expansion.ll
@@ -224,14 +224,14 @@ define void @pr52024(ptr %dst, i16 %N) {
; CHECK: [[LOOP_2_PH]]:
; CHECK-NEXT: [[IV_1_NEXT_LCSSA:%.*]] = phi i16 [ [[IV_1_NEXT]], %[[LOOP_1_LATCH]] ]
; CHECK-NEXT: [[IV_1_NEXT_EXT:%.*]] = sext i16 [[IV_1_NEXT_LCSSA]] to i64
-; CHECK-NEXT: [[TMP0:%.*]] = mul i16 [[IV_1]], 3
; CHECK-NEXT: br label %[[LOOP_2_HEADER:.*]]
; CHECK: [[LOOP_2_HEADER]]:
; CHECK-NEXT: [[IV_1_REM:%.*]] = urem i64 100, [[IV_1_NEXT_EXT]]
; CHECK-NEXT: [[REM_TRUNC:%.*]] = trunc i64 [[IV_1_REM]] to i16
; CHECK-NEXT: [[TMP1:%.*]] = udiv i64 100, [[IV_1_NEXT_EXT]]
; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[TMP1]] to i16
-; CHECK-NEXT: [[TMP3:%.*]] = mul i16 [[TMP0]], [[TMP2]]
+; CHECK-NEXT: [[TMP15:%.*]] = mul i16 [[IV_1]], [[TMP2]]
+; CHECK-NEXT: [[TMP3:%.*]] = mul i16 [[TMP15]], 3
; CHECK-NEXT: [[TMP4:%.*]] = add i16 [[TMP3]], -100
; CHECK-NEXT: br label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
diff --git a/llvm/test/Transforms/LoopVectorize/runtime-checks-difference.ll b/llvm/test/Transforms/LoopVectorize/runtime-checks-difference.ll
index 16afd3044d180..1f8e5b6813882 100644
--- a/llvm/test/Transforms/LoopVectorize/runtime-checks-difference.ll
+++ b/llvm/test/Transforms/LoopVectorize/runtime-checks-difference.ll
@@ -505,11 +505,11 @@ define void @diff_check_via_i32_ptrarith(ptr %origin, ptr %dst, ptr %base, i32 %
; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[OP]], [[N]]
; CHECK-NEXT: br i1 [[CMP]], label %[[LOOP_PH:.*]], [[EXIT:label %.*]]
; CHECK: [[LOOP_PH]]:
-; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[RHS]] to i32
-; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[D]], [[TMP0]]
-; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[TMP1]], -1
; CHECK-NEXT: [[TMP3:%.*]] = trunc i64 [[LHS]] to i32
-; CHECK-NEXT: [[TMP4:%.*]] = sub i32 [[TMP2]], [[TMP3]]
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[RHS]] to i32
+; CHECK-NEXT: [[TMP2:%.*]] = sub i32 [[D]], [[TMP3]]
+; CHECK-NEXT: [[TMP14:%.*]] = add i32 [[TMP2]], [[TMP1]]
+; CHECK-NEXT: [[TMP4:%.*]] = add i32 [[TMP14]], -1
; CHECK-NEXT: [[TMP5:%.*]] = zext i32 [[TMP4]] to i64
; CHECK-NEXT: [[TMP6:%.*]] = add nuw nsw i64 [[TMP5]], 1
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP6]], 4
More information about the llvm-commits
mailing list