[Tests][LV][AArch64] Pre-commit tests for changing loop interleaving count computation for loops that need to run scalar iterations #79640
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
…unt computation for loops that require a scalar epilogue run
|
@llvm/pr-subscribers-llvm-transforms Author: Nilanjana Basu (nilanjana87) ChangesThis patch contains a set of pre-commit tests for changing the loop interleaving count computation in a subsequent patch in order to address loops that need to execute at least a single scalar iteration in the epilogue. Full diff: https://github.com/llvm/llvm-project/pull/79640.diff 3 Files Affected:
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/interleave_count_for_estimated_tc.ll b/llvm/test/Transforms/LoopVectorize/AArch64/interleave_count_for_estimated_tc.ll
index 5552f9dd70c954e..d557ad1ead2563d 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/interleave_count_for_estimated_tc.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/interleave_count_for_estimated_tc.ll
@@ -125,6 +125,30 @@ for.end:
ret void
}
+; This has the same trip count as loop_with_profile_tc_64 but since the resulting interleaved group
+; in this case may access memory out-of-bounds, it requires a scalar epilogue iteration for
+; correctness, making at most 63 iterations available for interleaving.
+; TODO: When the auto-vectorizer chooses VF 16, it should choose IC 1 to leave a smaller scalar remainder
+; than IC 2
+; CHECK: remark: <unknown>:0:0: vectorized loop (vectorization width: 16, interleaved count: 2)
+define void @loop_with_profile_tc_64_scalar_epilogue_reqd(ptr noalias %p, ptr noalias %q, i64 %n) {
+entry:
+ br label %for.body
+
+for.body:
+ %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
+ %gep.src = getelementptr inbounds [3 x i8], ptr %p, i64 %i, i64 0
+ %l = load i8, ptr %gep.src, align 1
+ %gep.dst = getelementptr inbounds i8, ptr %q, i64 %i
+ store i8 %l, ptr %gep.dst, align 1
+ %i.next = add nuw nsw i64 %i, 1
+ %cond = icmp eq i64 %i.next, %n
+ br i1 %cond, label %for.end, label %for.body, !prof !4
+
+for.end:
+ ret void
+}
+
; For a loop with a profile-guided estimated TC of 100, when the auto-vectorizer chooses VF 16,
; it should choose conservatively IC 2 so that the vector loop runs twice at least
; CHECK: remark: <unknown>:0:0: vectorized loop (vectorization width: 16, interleaved count: 2)
@@ -173,6 +197,30 @@ for.end:
ret void
}
+; This has the same trip count as loop_with_profile_tc_128 but since the resulting interleaved group
+; in this case may access memory out-of-bounds, it requires a scalar epilogue iteration for
+; correctness, making at most 127 iterations available for interleaving.
+; TODO: When the auto-vectorizer chooses VF 16, it should choose IC 2 to leave a smaller scalar remainder
+; than IC 4
+; CHECK: remark: <unknown>:0:0: vectorized loop (vectorization width: 16, interleaved count: 4)
+define void @loop_with_profile_tc_128_scalar_epilogue_reqd(ptr noalias %p, ptr noalias %q, i64 %n) {
+entry:
+ br label %for.body
+
+for.body:
+ %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
+ %gep.src = getelementptr inbounds [3 x i8], ptr %p, i64 %i, i64 0
+ %l = load i8, ptr %gep.src, align 1
+ %gep.dst = getelementptr inbounds i8, ptr %q, i64 %i
+ store i8 %l, ptr %gep.dst, align 1
+ %i.next = add nuw nsw i64 %i, 1
+ %cond = icmp eq i64 %i.next, %n
+ br i1 %cond, label %for.end, label %for.body, !prof !6
+
+for.end:
+ ret void
+}
+
; For a loop with a profile-guided estimated TC of 129, when the auto-vectorizer chooses VF 16,
; it should choose conservatively IC 4 so that the vector loop runs twice at least
; CHECK: remark: <unknown>:0:0: vectorized loop (vectorization width: 16, interleaved count: 4)
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/interleave_count_for_known_tc.ll b/llvm/test/Transforms/LoopVectorize/AArch64/interleave_count_for_known_tc.ll
index 0569bfb2ae4e027..0a90798a3437110 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/interleave_count_for_known_tc.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/interleave_count_for_known_tc.ll
@@ -29,6 +29,30 @@ for.end:
ret void
}
+; This has the same trip count as loop_with_tc_32 but since the resulting interleaved group
+; in this case may access memory out-of-bounds, it requires a scalar epilogue iteration for
+; correctness, making at most 31 iterations available for interleaving.
+; TODO: When the auto-vectorizer chooses VF 16, it should choose IC 1 to leave a smaller scalar remainder
+; than IC 2
+; CHECK: remark: <unknown>:0:0: vectorized loop (vectorization width: 16, interleaved count: 2)
+define void @loop_with_tc_32_scalar_epilogue_reqd(ptr noalias %p, ptr noalias %q) {
+entry:
+ br label %for.body
+
+for.body:
+ %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
+ %gep.src = getelementptr inbounds [3 x i8], ptr %p, i64 %i, i64 0
+ %l = load i8, ptr %gep.src, align 1
+ %gep.dst = getelementptr inbounds i8, ptr %q, i64 %i
+ store i8 %l, ptr %gep.dst, align 1
+ %i.next = add nuw nsw i64 %i, 1
+ %cond = icmp eq i64 %i.next, 32
+ br i1 %cond, label %for.end, label %for.body
+
+for.end:
+ ret void
+}
+
; For this loop with known TC of 33, when the auto-vectorizer chooses VF 16, it should choose
; IC 2 since there is a small remainder loop TC that needs to run after the vector loop.
; CHECK: remark: <unknown>:0:0: vectorized loop (vectorization width: 16, interleaved count: 2)
@@ -197,6 +221,28 @@ for.end:
ret void
}
+; This has the same trip count as loop_with_tc_128 but since the resulting interleaved group
+; in this case may access memory out-of-bounds, it requires a scalar epilogue iteration for
+; correctness, making at most 31 iterations available for interleaving.
+; CHECK: remark: <unknown>:0:0: vectorized loop (vectorization width: 16, interleaved count: 8)
+define void @loop_with_tc_128_scalar_epilogue_reqd(ptr noalias %p, ptr noalias %q) {
+entry:
+ br label %for.body
+
+for.body:
+ %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
+ %gep.src = getelementptr inbounds [3 x i8], ptr %p, i64 %i, i64 0
+ %l = load i8, ptr %gep.src, align 1
+ %gep.dst = getelementptr inbounds i8, ptr %q, i64 %i
+ store i8 %l, ptr %gep.dst, align 1
+ %i.next = add nuw nsw i64 %i, 1
+ %cond = icmp eq i64 %i.next, 128
+ br i1 %cond, label %for.end, label %for.body
+
+for.end:
+ ret void
+}
+
; For this loop with known TC of 129, when the auto-vectorizer chooses VF 16, it should choose
; IC 8 since there is a small remainder loop that needs to run after the vector loop
; CHECK: remark: <unknown>:0:0: vectorized loop (vectorization width: 16, interleaved count: 8)
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/interleave_count_for_loops_needing_scalar_epilogue.ll b/llvm/test/Transforms/LoopVectorize/AArch64/interleave_count_for_loops_needing_scalar_epilogue.ll
new file mode 100644
index 000000000000000..b1b9e6d1e20a953
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/interleave_count_for_loops_needing_scalar_epilogue.ll
@@ -0,0 +1,132 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt < %s -force-target-max-vector-interleave=8 -p loop-vectorize -S 2>&1 | FileCheck %s
+; RUN: opt < %s -force-target-max-vector-interleave=8 -p loop-vectorize -pass-remarks=loop-vectorize -disable-output -S 2>&1 | FileCheck %s -check-prefix=CHECK-REMARKS
+
+target triple = "aarch64-linux-gnu"
+
+%pair = type { i8, i8 }
+
+; For this loop with known TC of 128, when the auto-vectorizer chooses VF 16, it should choose
+; IC 8 since there is no remainder loop run needed after the vector loop runs
+; CHECK-REMARKS: remark: <unknown>:0:0: vectorized loop (vectorization width: 16, interleaved count: 8)
+define void @loop_with_tc_128(ptr noalias %p, ptr noalias %q) {
+; CHECK-LABEL: define void @loop_with_tc_128(
+; CHECK-SAME: ptr noalias [[P:%.*]], ptr noalias [[Q:%.*]]) {
+; CHECK-NEXT: entry:
+; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+entry:
+ br label %for.body
+
+for.body:
+ %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
+ %gep.src = getelementptr %pair, ptr %p, i64 %i, i32 0
+ %load.src = load i8, ptr %gep.src, align 1
+ %gep.dst = getelementptr %pair, ptr %p, i64 %i, i32 1
+ %load.dst = load i8, ptr %gep.dst, align 1
+ %add = add i8 %load.src, %load.dst
+ %qi = getelementptr i8, ptr %q, i64 %i
+ store i8 %add, ptr %qi, align 1
+ %i.next = add nuw nsw i64 %i, 1
+ %cond = icmp eq i64 %i.next, 128
+ br i1 %cond, label %for.end, label %for.body
+
+for.end:
+ ret void
+}
+
+; This function has the same trip count as loop_with_tc_128 but since the resulting interleaved group
+; in this case may access memory out-of-bounds, it requires a scalar epilogue iteration for
+; correctness, making at most 127 iterations available for interleaving.
+; TODO: The entry block should branch into the vector loop, instead of the scalar epilogue.
+; TODO: When the auto-vectorizer chooses VF 16, it should choose IC 2, to have a smaller scalar remainder
+; than when using IC 4.
+; CHECK-REMARKS: remark: <unknown>:0:0: vectorized loop (vectorization width: 16, interleaved count: 8)
+define void @loop_with_tc_128_scalar_epilogue_reqd(ptr noalias %p, ptr noalias %q) {
+; CHECK-LABEL: define void @loop_with_tc_128_scalar_epilogue_reqd(
+; CHECK-SAME: ptr noalias [[P:%.*]], ptr noalias [[Q:%.*]]) {
+; CHECK-NEXT: entry:
+; CHECK-NEXT: br i1 true, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+entry:
+ br label %for.body
+
+for.body:
+ %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
+ %gep.src = getelementptr inbounds [3 x i8], ptr %p, i64 %i, i64 0
+ %l = load i8, ptr %gep.src, align 1
+ %gep.dst = getelementptr inbounds i8, ptr %q, i64 %i
+ store i8 %l, ptr %gep.dst, align 1
+ %i.next = add nuw nsw i64 %i, 1
+ %cond = icmp eq i64 %i.next, 128
+ br i1 %cond, label %for.end, label %for.body
+
+for.end:
+ ret void
+}
+
+; For a loop with a profile-guided estimated TC of 128, when the auto-vectorizer chooses VF 16,
+; it should choose conservatively IC 4 so that the vector loop runs twice at least
+; CHECK-REMARKS: remark: <unknown>:0:0: vectorized loop (vectorization width: 16, interleaved count: 4)
+define void @loop_with_profile_tc_128(ptr noalias %p, ptr noalias %q, i64 %n) {
+; CHECK-LABEL: define void @loop_with_profile_tc_128(
+; CHECK-SAME: ptr noalias [[P:%.*]], ptr noalias [[Q:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT: iter.check:
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 8
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]], !prof [[PROF6:![0-9]+]]
+; CHECK: vector.main.loop.iter.check:
+; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[N]], 64
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]], !prof [[PROF6]]
+;
+entry:
+ br label %for.body
+
+for.body:
+ %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
+ %gep.src = getelementptr %pair, ptr %p, i64 %i, i32 0
+ %load.src = load i8, ptr %gep.src, align 1
+ %gep.dst = getelementptr %pair, ptr %p, i64 %i, i32 1
+ %load.dst = load i8, ptr %gep.dst, align 1
+ %add = add i8 %load.src, %load.dst
+ %qi = getelementptr i8, ptr %q, i64 %i
+ store i8 %add, ptr %qi, align 1
+ %i.next = add nuw nsw i64 %i, 1
+ %cond = icmp eq i64 %i.next, %n
+ br i1 %cond, label %for.end, label %for.body, !prof !0
+
+for.end:
+ ret void
+}
+
+; This function has the same trip count as loop_with_profile_tc_128 but since the resulting interleaved group
+; in this case may access memory out-of-bounds, it requires a scalar epilogue iteration for
+; correctness, making at most 127 iterations available for interleaving.
+; TODO: When the auto-vectorizer chooses VF 16, it should choose IC 2, to have a smaller scalar remainder
+; than IC 4.
+; CHECK-REMARKS: remark: <unknown>:0:0: vectorized loop (vectorization width: 16, interleaved count: 4)
+define void @loop_with_profile_tc_128_scalar_epilogue_reqd(ptr noalias %p, ptr noalias %q, i64 %n) {
+; CHECK-LABEL: define void @loop_with_profile_tc_128_scalar_epilogue_reqd(
+; CHECK-SAME: ptr noalias [[P:%.*]], ptr noalias [[Q:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT: iter.check:
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ule i64 [[N]], 8
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]], !prof [[PROF6]]
+; CHECK: vector.main.loop.iter.check:
+; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ule i64 [[N]], 64
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]], !prof [[PROF6]]
+;
+entry:
+ br label %for.body
+
+for.body:
+ %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
+ %gep.src = getelementptr inbounds [3 x i8], ptr %p, i64 %i, i64 0
+ %l = load i8, ptr %gep.src, align 1
+ %gep.dst = getelementptr inbounds i8, ptr %q, i64 %i
+ store i8 %l, ptr %gep.dst, align 1
+ %i.next = add nuw nsw i64 %i, 1
+ %cond = icmp eq i64 %i.next, %n
+ br i1 %cond, label %for.end, label %for.body, !prof !0
+
+for.end:
+ ret void
+}
+
+!0 = !{!"branch_weights", i32 1, i32 127}
|
fhahn
reviewed
Jan 26, 2024
llvm/test/Transforms/LoopVectorize/AArch64/interleave_count_for_estimated_tc.ll
Outdated
Show resolved
Hide resolved
llvm/test/Transforms/LoopVectorize/AArch64/interleave_count_for_estimated_tc.ll
Outdated
Show resolved
Hide resolved
.../test/Transforms/LoopVectorize/AArch64/interleave_count_for_loops_needing_scalar_epilogue.ll
Outdated
Show resolved
Hide resolved
…ent for testing loop ic computation.
nilanjana87
added a commit
to swiftlang/llvm-project
that referenced
this pull request
Mar 5, 2024
…count computation for loops that need to run scalar iterations (llvm#79640) This patch contains a set of pre-commit tests for changing the loop interleaving count computation in a subsequent patch in order to address loops that need to execute at least a single scalar iteration in the epilogue.
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.
This patch contains a set of pre-commit tests for changing the loop interleaving count computation in a subsequent patch (#79651) in order to address loops that need to execute at least a single scalar iteration in the epilogue.