[LV] Align legacy cost model to vplan-based model for gather/scatter w/ uniform addr. #155739
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 patch check if the addr is uniform in legacy cost model to align vplan-based cost model after llvm#150371. This patch fixes llvm-test-suite assertion due to cost model misaligned after llvm#149955 under RISCV. I've tested this patch (on top of llvm#149955) on the llvm-test-suite locally with crached options `rva23u64`, `rva23u64_zvl1024b` and build successfully.
ElvisWang123
requested review from
Mel-Chen,
arcbbb,
david-arm,
fhahn and
lukel97
|
@llvm/pr-subscribers-vectorizers @llvm/pr-subscribers-backend-risc-v Author: Elvis Wang (ElvisWang123) ChangesThis patch check if the addr is uniform in legacy cost model to align vplan-based cost model after #150371. This patch fixes llvm-test-suite assertion due to cost model misaligned after #149955 under RISCV. I've tested this patch (on top of #149955) on the llvm-test-suite locally with crashed options Since this fix will change LV, I think would be better to create a PR to fix this. Full diff: https://github.com/llvm/llvm-project/pull/155739.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 6317bc3c20e25..c1cdd59dee39d 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -5256,8 +5256,11 @@ LoopVectorizationCostModel::getGatherScatterCost(Instruction *I,
Type *ValTy = getLoadStoreType(I);
auto *VectorTy = cast<VectorType>(toVectorTy(ValTy, VF));
const Align Alignment = getLoadStoreAlignment(I);
- const Value *Ptr = getLoadStorePointerOperand(I);
- Type *PtrTy = toVectorTy(Ptr->getType(), VF);
+ Value *Ptr = getLoadStorePointerOperand(I);
+ Type *PtrTy = Ptr->getType();
+
+ if (!Legal->isUniform(Ptr, VF))
+ PtrTy = toVectorTy(PtrTy, VF);
return TTI.getAddressComputationCost(PtrTy, nullptr, nullptr, CostKind) +
TTI.getGatherScatterOpCost(I->getOpcode(), VectorTy, Ptr,
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/gather-scatter-cost.ll b/llvm/test/Transforms/LoopVectorize/RISCV/gather-scatter-cost.ll
new file mode 100644
index 0000000000000..6313fe9c1a7f8
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/gather-scatter-cost.ll
@@ -0,0 +1,190 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+rva23u64 -S | FileCheck %s -check-prefixes=CHECK,RVA23
+; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+rva23u64,+zvl1024b -S | FileCheck %s -check-prefixes=CHECK,RVA23ZVL1024B
+
+define i32 @getNeighborBoxes(ptr %boxes, i32 %iBox, ptr %nbrBoxes) {
+; CHECK-LABEL: @getNeighborBoxes(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[IBOX:%.*]] to i64
+; CHECK-NEXT: [[TMP1:%.*]] = add nsw i64 [[TMP0]], 1
+; CHECK-NEXT: [[SMAX2:%.*]] = call i64 @llvm.smax.i64(i64 [[TMP1]], i64 0)
+; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[SMAX2]] to i32
+; CHECK-NEXT: [[TMP3:%.*]] = add nuw i32 [[TMP2]], 1
+; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
+; CHECK: vector.scevcheck:
+; CHECK-NEXT: [[TMP4:%.*]] = sext i32 [[IBOX]] to i64
+; CHECK-NEXT: [[TMP5:%.*]] = add nsw i64 [[TMP4]], 1
+; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[TMP5]], i64 0)
+; CHECK-NEXT: [[TMP6:%.*]] = trunc i64 [[SMAX]] to i32
+; CHECK-NEXT: [[TMP7:%.*]] = icmp slt i32 [[TMP6]], 0
+; CHECK-NEXT: [[TMP8:%.*]] = icmp ugt i64 [[SMAX]], 4294967295
+; CHECK-NEXT: [[TMP9:%.*]] = or i1 [[TMP7]], [[TMP8]]
+; CHECK-NEXT: br i1 [[TMP9]], label [[SCALAR_PH]], label [[VECTOR_MEMCHECK:%.*]]
+; CHECK: vector.memcheck:
+; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[NBRBOXES:%.*]], i64 4
+; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[BOXES:%.*]], i64 4
+; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[NBRBOXES]], [[SCEVGEP1]]
+; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[BOXES]], [[SCEVGEP]]
+; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
+; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
+; CHECK: vector.ph:
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 4 x ptr> poison, ptr [[BOXES]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x ptr> [[BROADCAST_SPLATINSERT]], <vscale x 4 x ptr> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <vscale x 4 x ptr> poison, ptr [[NBRBOXES]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT4:%.*]] = shufflevector <vscale x 4 x ptr> [[BROADCAST_SPLATINSERT3]], <vscale x 4 x ptr> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
+; CHECK: vector.body:
+; CHECK-NEXT: [[AVL:%.*]] = phi i32 [ [[TMP3]], [[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP10:%.*]] = call i32 @llvm.experimental.get.vector.length.i32(i32 [[AVL]], i32 4, i1 true)
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT5:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[TMP10]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT6:%.*]] = shufflevector <vscale x 4 x i32> [[BROADCAST_SPLATINSERT5]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK-NEXT: [[TMP11:%.*]] = call <vscale x 4 x i32> @llvm.stepvector.nxv4i32()
+; CHECK-NEXT: [[TMP12:%.*]] = icmp ult <vscale x 4 x i32> [[TMP11]], [[BROADCAST_SPLAT6]]
+; CHECK-NEXT: [[TMP13:%.*]] = select <vscale x 4 x i1> [[TMP12]], <vscale x 4 x i1> splat (i1 true), <vscale x 4 x i1> zeroinitializer
+; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 4 x i32> @llvm.vp.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> align 4 [[BROADCAST_SPLAT]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP10]]), !alias.scope [[META0:![0-9]+]]
+; CHECK-NEXT: [[TMP14:%.*]] = select <vscale x 4 x i1> [[TMP12]], <vscale x 4 x i1> zeroinitializer, <vscale x 4 x i1> zeroinitializer
+; CHECK-NEXT: [[TMP15:%.*]] = or <vscale x 4 x i1> [[TMP13]], [[TMP14]]
+; CHECK-NEXT: [[PREDPHI:%.*]] = select <vscale x 4 x i1> [[TMP14]], <vscale x 4 x i32> zeroinitializer, <vscale x 4 x i32> [[WIDE_MASKED_GATHER]]
+; CHECK-NEXT: call void @llvm.vp.scatter.nxv4i32.nxv4p0(<vscale x 4 x i32> [[PREDPHI]], <vscale x 4 x ptr> align 4 [[BROADCAST_SPLAT4]], <vscale x 4 x i1> [[TMP15]], i32 [[TMP10]]), !alias.scope [[META3:![0-9]+]], !noalias [[META0]]
+; CHECK-NEXT: [[AVL_NEXT]] = sub nuw i32 [[AVL]], [[TMP10]]
+; CHECK-NEXT: [[TMP16:%.*]] = icmp eq i32 [[AVL_NEXT]], 0
+; CHECK-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
+; CHECK: middle.block:
+; CHECK-NEXT: br label [[EXIT:%.*]]
+; CHECK: scalar.ph:
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[VECTOR_MEMCHECK]] ]
+; CHECK-NEXT: br label [[LOOP:%.*]]
+; CHECK: loop:
+; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_EXIT:%.*]] ]
+; CHECK-NEXT: br i1 false, label [[EXIT_CRIT_EDGE:%.*]], label [[IF:%.*]]
+; CHECK: exit_crit_edge:
+; CHECK-NEXT: br label [[LOOP_EXIT]]
+; CHECK: if:
+; CHECK-NEXT: [[TMP17:%.*]] = load i32, ptr [[BOXES]], align 4
+; CHECK-NEXT: br label [[LOOP_EXIT]]
+; CHECK: loop.exit:
+; CHECK-NEXT: [[IBOX_0_I:%.*]] = phi i32 [ [[TMP17]], [[IF]] ], [ 0, [[EXIT_CRIT_EDGE]] ]
+; CHECK-NEXT: store i32 [[IBOX_0_I]], ptr [[NBRBOXES]], align 4
+; CHECK-NEXT: [[IV_NEXT]] = add i32 [[IV]], 1
+; CHECK-NEXT: [[EXITCOND:%.*]] = icmp sgt i32 [[IV]], [[IBOX]]
+; CHECK-NEXT: br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP9:![0-9]+]]
+; CHECK: exit:
+; CHECK-NEXT: ret i32 0
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.exit ]
+ br i1 false, label %exit_crit_edge, label %if
+
+exit_crit_edge:
+ br label %loop.exit
+
+if:
+ %0 = load i32, ptr %boxes, align 4
+ br label %loop.exit
+
+loop.exit:
+ %iBox.0.i = phi i32 [ %0, %if ], [ 0, %exit_crit_edge ]
+ store i32 %iBox.0.i, ptr %nbrBoxes, align 4
+ %iv.next = add i32 %iv, 1
+ %exitcond = icmp sgt i32 %iv, %iBox
+ br i1 %exitcond, label %exit, label %loop
+
+exit:
+ ret i32 0
+}
+
+define void @_Z9BM_MemCmpILi7E12LessThanZero5FirstEvRN9benchmark5StateE() {
+; RVA23-LABEL: @_Z9BM_MemCmpILi7E12LessThanZero5FirstEvRN9benchmark5StateE(
+; RVA23-NEXT: entry:
+; RVA23-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; RVA23: vector.ph:
+; RVA23-NEXT: [[TMP0:%.*]] = call <vscale x 16 x i64> @llvm.stepvector.nxv16i64()
+; RVA23-NEXT: [[TMP1:%.*]] = mul <vscale x 16 x i64> [[TMP0]], splat (i64 1)
+; RVA23-NEXT: [[INDUCTION:%.*]] = add <vscale x 16 x i64> zeroinitializer, [[TMP1]]
+; RVA23-NEXT: br label [[VECTOR_BODY:%.*]]
+; RVA23: vector.body:
+; RVA23-NEXT: [[VEC_IND:%.*]] = phi <vscale x 16 x i64> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
+; RVA23-NEXT: [[AVL:%.*]] = phi i64 [ 586, [[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], [[VECTOR_BODY]] ]
+; RVA23-NEXT: [[TMP2:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 16, i1 true)
+; RVA23-NEXT: [[TMP3:%.*]] = zext i32 [[TMP2]] to i64
+; RVA23-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 16 x i64> poison, i64 [[TMP3]], i64 0
+; RVA23-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 16 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 16 x i64> poison, <vscale x 16 x i32> zeroinitializer
+; RVA23-NEXT: [[TMP4:%.*]] = mul <vscale x 16 x i64> [[VEC_IND]], splat (i64 7)
+; RVA23-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr null, <vscale x 16 x i64> [[TMP4]]
+; RVA23-NEXT: call void @llvm.vp.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> zeroinitializer, <vscale x 16 x ptr> align 1 [[TMP5]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP2]])
+; RVA23-NEXT: [[TMP6:%.*]] = zext i32 [[TMP2]] to i64
+; RVA23-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP6]]
+; RVA23-NEXT: [[VEC_IND_NEXT]] = add <vscale x 16 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; RVA23-NEXT: [[TMP7:%.*]] = icmp eq i64 [[AVL_NEXT]], 0
+; RVA23-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
+; RVA23: middle.block:
+; RVA23-NEXT: br label [[EXIT:%.*]]
+; RVA23: scalar.ph:
+; RVA23-NEXT: br label [[LOOP:%.*]]
+; RVA23: loop:
+; RVA23-NEXT: [[IV:%.*]] = phi i64 [ 0, [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
+; RVA23-NEXT: [[TMP8:%.*]] = mul i64 [[IV]], 7
+; RVA23-NEXT: [[ADD_PTR:%.*]] = getelementptr i8, ptr null, i64 [[TMP8]]
+; RVA23-NEXT: store i8 0, ptr [[ADD_PTR]], align 1
+; RVA23-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
+; RVA23-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[IV]], 585
+; RVA23-NEXT: br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP11:![0-9]+]]
+; RVA23: exit:
+; RVA23-NEXT: ret void
+;
+; RVA23ZVL1024B-LABEL: @_Z9BM_MemCmpILi7E12LessThanZero5FirstEvRN9benchmark5StateE(
+; RVA23ZVL1024B-NEXT: entry:
+; RVA23ZVL1024B-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; RVA23ZVL1024B: vector.ph:
+; RVA23ZVL1024B-NEXT: [[TMP0:%.*]] = call <vscale x 2 x i64> @llvm.stepvector.nxv2i64()
+; RVA23ZVL1024B-NEXT: [[TMP1:%.*]] = mul <vscale x 2 x i64> [[TMP0]], splat (i64 1)
+; RVA23ZVL1024B-NEXT: [[INDUCTION:%.*]] = add <vscale x 2 x i64> zeroinitializer, [[TMP1]]
+; RVA23ZVL1024B-NEXT: br label [[VECTOR_BODY:%.*]]
+; RVA23ZVL1024B: vector.body:
+; RVA23ZVL1024B-NEXT: [[VEC_IND:%.*]] = phi <vscale x 2 x i64> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
+; RVA23ZVL1024B-NEXT: [[AVL:%.*]] = phi i64 [ 586, [[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], [[VECTOR_BODY]] ]
+; RVA23ZVL1024B-NEXT: [[TMP2:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 2, i1 true)
+; RVA23ZVL1024B-NEXT: [[TMP3:%.*]] = zext i32 [[TMP2]] to i64
+; RVA23ZVL1024B-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP3]], i64 0
+; RVA23ZVL1024B-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
+; RVA23ZVL1024B-NEXT: [[TMP4:%.*]] = mul <vscale x 2 x i64> [[VEC_IND]], splat (i64 7)
+; RVA23ZVL1024B-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr null, <vscale x 2 x i64> [[TMP4]]
+; RVA23ZVL1024B-NEXT: call void @llvm.vp.scatter.nxv2i8.nxv2p0(<vscale x 2 x i8> zeroinitializer, <vscale x 2 x ptr> align 1 [[TMP5]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP2]])
+; RVA23ZVL1024B-NEXT: [[TMP6:%.*]] = zext i32 [[TMP2]] to i64
+; RVA23ZVL1024B-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP6]]
+; RVA23ZVL1024B-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; RVA23ZVL1024B-NEXT: [[TMP7:%.*]] = icmp eq i64 [[AVL_NEXT]], 0
+; RVA23ZVL1024B-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
+; RVA23ZVL1024B: middle.block:
+; RVA23ZVL1024B-NEXT: br label [[EXIT:%.*]]
+; RVA23ZVL1024B: scalar.ph:
+; RVA23ZVL1024B-NEXT: br label [[LOOP:%.*]]
+; RVA23ZVL1024B: loop:
+; RVA23ZVL1024B-NEXT: [[IV:%.*]] = phi i64 [ 0, [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
+; RVA23ZVL1024B-NEXT: [[TMP8:%.*]] = mul i64 [[IV]], 7
+; RVA23ZVL1024B-NEXT: [[ADD_PTR:%.*]] = getelementptr i8, ptr null, i64 [[TMP8]]
+; RVA23ZVL1024B-NEXT: store i8 0, ptr [[ADD_PTR]], align 1
+; RVA23ZVL1024B-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
+; RVA23ZVL1024B-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[IV]], 585
+; RVA23ZVL1024B-NEXT: br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP11:![0-9]+]]
+; RVA23ZVL1024B: exit:
+; RVA23ZVL1024B-NEXT: ret void
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+ %0 = mul i64 %iv, 7
+ %add.ptr = getelementptr i8, ptr null, i64 %0
+ store i8 0, ptr %add.ptr, align 1
+ %iv.next = add i64 %iv, 1
+ %exitcond = icmp eq i64 %iv, 585
+ br i1 %exitcond, label %exit, label %loop
+
+exit:
+ ret void
+}
|
Mel-Chen
reviewed
Aug 28, 2025
fhahn
reviewed
Aug 28, 2025
fhahn
reviewed
Aug 29, 2025
| ret i32 0 | ||
| } | ||
|
|
||
| define void @_Z9BM_MemCmpILi7E12LessThanZero5FirstEvRN9benchmark5StateE() { |
There was a problem hiding this comment.
Could you adjust the names to be more descriptive, instead of using the ones from the original source function? Also above
There was a problem hiding this comment.
Updated, thanks!
|
|
||
| loop: | ||
| %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.exit ] | ||
| br i1 false, label %exit_crit_edge, label %if |
There was a problem hiding this comment.
is a known condition required here?
There was a problem hiding this comment.
Yes. Otherwise it will using fixed VF and not triggering the assertion.
| %0 = load i32, ptr %boxes, align 4 | ||
| br label %loop.exit | ||
|
|
||
| loop.exit: |
There was a problem hiding this comment.
Suggested change
| loop.exit: | |
| loop.latch: |
this is not the exit of the loop, but the (exiting) latch in LLVM terminology.
There was a problem hiding this comment.
Fixed, thanks!
| %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.exit ] | ||
| br i1 false, label %exit_crit_edge, label %if | ||
|
|
||
| exit_crit_edge: |
There was a problem hiding this comment.
Suggested change
| exit_crit_edge: | |
| loop.then: |
There was a problem hiding this comment.
Updated, thanks!
| exit_crit_edge: | ||
| br label %loop.exit | ||
|
|
||
| if: |
There was a problem hiding this comment.
Updated, thanks
| loop: | ||
| %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] | ||
| %0 = mul i64 %iv, 7 | ||
| %add.ptr = getelementptr i8, ptr null, i64 %0 |
There was a problem hiding this comment.
please avoid null, it makes the function have immediate UB on the store
There was a problem hiding this comment.
Using non-null value, thanks!
Mel-Chen
reviewed
Aug 29, 2025
llvm/test/Transforms/LoopVectorize/RISCV/gather-scatter-cost.ll
Outdated
Show resolved
Hide resolved
fhahn
approved these changes
Sep 1, 2025
|
|
||
| loop: | ||
| %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.latch ] | ||
| br i1 false, label %loop.then, label %loop.else |
There was a problem hiding this comment.
If it also crashes without a constant condition, would be good to update the test
There was a problem hiding this comment.
Using a condition passed by operand also crashes. Updated, thanks!
| br label %loop.latch | ||
|
|
||
| loop.else: | ||
| %0 = load i32, ptr %boxes, align 4 |
There was a problem hiding this comment.
Suggested change
| %0 = load i32, ptr %boxes, align 4 | |
| %0 = load i32, ptr %src, align 4 |
There was a problem hiding this comment.
Updated, thanks!
|
|
||
| loop.latch: | ||
| %store = phi i32 [%0, %loop.else], [0, %loop.then] | ||
| store i32 %store, ptr %nbrBoxes, align 4 |
There was a problem hiding this comment.
Suggested change
| store i32 %store, ptr %nbrBoxes, align 4 | |
| store i32 %store, ptr %dst, align 4 |
There was a problem hiding this comment.
Updated, thanks!
| %store = phi i32 [%0, %loop.else], [0, %loop.then] | ||
| store i32 %store, ptr %nbrBoxes, align 4 | ||
| %iv.next = add i32 %iv, 1 | ||
| %exitcond = icmp sgt i32 %iv, %iBox |
There was a problem hiding this comment.
Suggested change
| %exitcond = icmp sgt i32 %iv, %iBox | |
| %exitcond = icmp sgt i32 %iv, %n |
ElvisWang123
added a commit
that referenced
this pull request
Sep 2, 2025
… TTI. #149955" (#156386) This patch implements the `getAddressComputationCost()` in RISCV TTI which make the gather/scatter with address calculation more expansive that stride cost. Note that the only user of `getAddressComputationCost()` with vector type is in `VPWidenMemoryRecipe::computeCost()`. So this patch make some LV tests changes. I've checked the tests changes in LV and seems those changes can be divided into two groups. * gather/scatter with uniform vector ptr, seems can be optimized to masked.load. * can optimize to stride load/store. ---- After #155739 landed, the assertion (cost mis-aligned) is fixed. I've tested llvm-test-suite w/ rva23u64 and rva23u64_zvl1024b locally and no assertion occurred.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
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 patch check if the addr is uniform in legacy cost model to align vplan-based cost model after #150371.
This patch fixes llvm-test-suite assertion (https://lab.llvm.org/buildbot/#/builders/210/builds/1935) due to cost model misaligned after #149955 under RISCV.
I've tested this patch (on top of #149955) on the llvm-test-suite locally with crashed options
rva23u64,rva23u64_zvl1024band build successfully.Since this fix will change LV, I think would be better to create a PR to fix this.