[llvm] [LV] Pre-committing tests for changing loop interleaving count computation (PR #70272)

via llvm-commits llvm-commits at lists.llvm.org
Wed Oct 25 17:25:03 PDT 2023 

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-transforms

Author: Nilanjana Basu (nilanjana87)

<details>
<summary>Changes</summary>

Added tests for evaluating changes to loop interleaving count computation and for removing loop interleaving threshold.

---
Full diff: https://github.com/llvm/llvm-project/pull/70272.diff


2 Files Affected:

- (added) llvm/test/Transforms/LoopVectorize/AArch64/interleave_count.ll (+106) 
- (modified) llvm/test/Transforms/LoopVectorize/X86/unroll-small-loops.ll (+190-4) 


``````````diff
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/interleave_count.ll b/llvm/test/Transforms/LoopVectorize/AArch64/interleave_count.ll
new file mode 100644
index 000000000000000..74fd2f44fce0109
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/interleave_count.ll
@@ -0,0 +1,106 @@
+; RUN: opt < %s -force-vector-width=64 -O3 -S -pass-remarks=loop-vectorize 2>&1 | FileCheck %s
+
+target triple = "aarch64-linux-gnu"
+
+%pair = type { i8, i8 }
+
+; For a loop with known trip count of 128, when we force VF 64, it should use
+; IC 2, since there is no remainder loop needed when the vector loop runs.
+; CHECK: remark: <unknown>:0:0: vectorized loop (vectorization width: 64, interleaved count: 2)
+define void @loop_with_tc_128(ptr %p, ptr %q) {
+entry:
+  br label %for.body
+
+for.body:
+  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
+  %tmp0 = getelementptr %pair, ptr %p, i64 %i, i32 0
+  %tmp1 = load i8, ptr %tmp0, align 1
+  %tmp2 = getelementptr %pair, ptr %p, i64 %i, i32 1
+  %tmp3 = load i8, ptr %tmp2, align 1
+  %add = add i8 %tmp1, %tmp3
+  %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
+}
+
+; For a loop with known trip count of 129, when we force VF 64, it should use
+; IC 1, since there may be a remainder loop that needs to run after the vector loop.
+; CHECK: remark: <unknown>:0:0: vectorized loop (vectorization width: 64, interleaved count: 2)
+define void @loop_with_tc_129(ptr %p, ptr %q) {
+entry:
+  br label %for.body
+
+for.body:
+  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
+  %tmp0 = getelementptr %pair, ptr %p, i64 %i, i32 0
+  %tmp1 = load i8, ptr %tmp0, align 1
+  %tmp2 = getelementptr %pair, ptr %p, i64 %i, i32 1
+  %tmp3 = load i8, ptr %tmp2, align 1
+  %add = add i8 %tmp1, %tmp3
+  %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, 129
+  br i1 %cond, label %for.end, label %for.body
+
+for.end:
+  ret void
+}
+
+; For a loop with unknown trip count but a profile showing an approx TC estimate of 128, 
+; when we force VF 64, it should use IC 2, since chances are high that the remainder loop
+; won't need to run
+; CHECK: remark: <unknown>:0:0: vectorized loop (vectorization width: 64, interleaved count: 2)
+define void @loop_with_profile_tc_128(ptr %p, ptr %q, i64 %n) {
+entry:
+  br label %for.body
+
+for.body:
+  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
+  %tmp0 = getelementptr %pair, ptr %p, i64 %i, i32 0
+  %tmp1 = load i8, ptr %tmp0, align 1
+  %tmp2 = getelementptr %pair, ptr %p, i64 %i, i32 1
+  %tmp3 = load i8, ptr %tmp2, align 1
+  %add = add i8 %tmp1, %tmp3
+  %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
+}
+
+; For a loop with unknown trip count but a profile showing an approx TC estimate of 129, 
+; when we force VF 64, it should use IC 1, since chances are high that the remainder loop
+; will need to run
+; CHECK: remark: <unknown>:0:0: vectorized loop (vectorization width: 64, interleaved count: 2)
+define void @loop_with_profile_tc_129(ptr %p, ptr %q, i64 %n) {
+entry:
+  br label %for.body
+
+for.body:
+  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
+  %tmp0 = getelementptr %pair, ptr %p, i64 %i, i32 0
+  %tmp1 = load i8, ptr %tmp0, align 1
+  %tmp2 = getelementptr %pair, ptr %p, i64 %i, i32 1
+  %tmp3 = load i8, ptr %tmp2, align 1
+  %add = add i8 %tmp1, %tmp3
+  %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 !1
+
+for.end:
+  ret void
+}
+
+!0 = !{!"branch_weights", i32 1, i32 127}
+!1 = !{!"branch_weights", i32 1, i32 128}
diff --git a/llvm/test/Transforms/LoopVectorize/X86/unroll-small-loops.ll b/llvm/test/Transforms/LoopVectorize/X86/unroll-small-loops.ll
index 290be569bc12524..6b71cfef06be149 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/unroll-small-loops.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/unroll-small-loops.ll
@@ -8,20 +8,20 @@ target triple = "x86_64-apple-macosx10.8.0"
 
 ; We don't unroll this loop because it has a small constant trip count.
 ;
-; CHECK-VECTOR-LABEL: @foo(
+; CHECK-VECTOR-LABEL: @foo_trip_count_8(
 ; CHECK-VECTOR: load <4 x i32>
 ; CHECK-VECTOR-NOT: load <4 x i32>
 ; CHECK-VECTOR: store <4 x i32>
 ; CHECK-VECTOR-NOT: store <4 x i32>
 ; CHECK-VECTOR: ret
 ;
-; CHECK-SCALAR-LABEL: @foo(
+; CHECK-SCALAR-LABEL: @foo_trip_count_8(
 ; CHECK-SCALAR: load i32, ptr
 ; CHECK-SCALAR-NOT: load i32, ptr
 ; CHECK-SCALAR: store i32
 ; CHECK-SCALAR-NOT: store i32
 ; CHECK-SCALAR: ret
-define i32 @foo(ptr nocapture %A) nounwind uwtable ssp {
+define i32 @foo_trip_count_8(ptr nocapture %A) nounwind uwtable ssp {
   br label %1
 
 ; <label>:1                                       ; preds = %1, %0
@@ -32,7 +32,193 @@ define i32 @foo(ptr nocapture %A) nounwind uwtable ssp {
   store i32 %4, ptr %2, align 4
   %indvars.iv.next = add i64 %indvars.iv, 1
   %lftr.wideiv = trunc i64 %indvars.iv.next to i32
-  %exitcond = icmp eq i32 %lftr.wideiv, 100
+  %exitcond = icmp eq i32 %lftr.wideiv, 8
+  br i1 %exitcond, label %5, label %1
+
+; <label>:5                                       ; preds = %1
+  ret i32 undef
+}
+
+; CHECK-VECTOR-LABEL: @foo_trip_count_16(
+; CHECK-VECTOR: load <4 x i32>
+; CHECK-VECTOR-NOT: load <4 x i32>
+; CHECK-VECTOR: store <4 x i32>
+; CHECK-VECTOR-NOT: store <4 x i32>
+; CHECK-VECTOR: ret
+;
+; CHECK-SCALAR-LABEL: @foo_trip_count_16(
+; CHECK-SCALAR: load i32, ptr
+; CHECK-SCALAR-NOT: load i32, ptr
+; CHECK-SCALAR: store i32
+; CHECK-SCALAR-NOT: store i32
+; CHECK-SCALAR: ret
+define i32 @foo_trip_count_16(ptr nocapture %A) nounwind uwtable ssp {
+  br label %1
+
+; <label>:1                                       ; preds = %1, %0
+  %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ]
+  %2 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv
+  %3 = load i32, ptr %2, align 4
+  %4 = add nsw i32 %3, 6
+  store i32 %4, ptr %2, align 4
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 16
+  br i1 %exitcond, label %5, label %1
+
+; <label>:5                                       ; preds = %1
+  ret i32 undef
+}
+
+; CHECK-VECTOR-LABEL: @foo_trip_count_17(
+; CHECK-VECTOR: load <4 x i32>
+; CHECK-VECTOR-NOT: load <4 x i32>
+; CHECK-VECTOR: store <4 x i32>
+; CHECK-VECTOR-NOT: store <4 x i32>
+; CHECK-VECTOR: ret
+;
+; CHECK-SCALAR-LABEL: @foo_trip_count_17(
+; CHECK-SCALAR: load i32, ptr
+; CHECK-SCALAR-NOT: load i32, ptr
+; CHECK-SCALAR: store i32
+; CHECK-SCALAR-NOT: store i32
+; CHECK-SCALAR: ret
+define i32 @foo_trip_count_17(ptr nocapture %A) nounwind uwtable ssp {
+  br label %1
+
+; <label>:1                                       ; preds = %1, %0
+  %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ]
+  %2 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv
+  %3 = load i32, ptr %2, align 4
+  %4 = add nsw i32 %3, 6
+  store i32 %4, ptr %2, align 4
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 17
+  br i1 %exitcond, label %5, label %1
+
+; <label>:5                                       ; preds = %1
+  ret i32 undef
+}
+
+; CHECK-VECTOR-LABEL: @foo_trip_count_24(
+; CHECK-VECTOR: load <4 x i32>
+; CHECK-VECTOR-NOT: load <4 x i32>
+; CHECK-VECTOR: store <4 x i32>
+; CHECK-VECTOR-NOT: store <4 x i32>
+; CHECK-VECTOR: ret
+;
+; CHECK-SCALAR-LABEL: @foo_trip_count_24(
+; CHECK-SCALAR: load i32, ptr
+; CHECK-SCALAR-NOT: load i32, ptr
+; CHECK-SCALAR: store i32
+; CHECK-SCALAR-NOT: store i32
+; CHECK-SCALAR: ret
+define i32 @foo_trip_count_24(ptr nocapture %A) nounwind uwtable ssp {
+  br label %1
+
+; <label>:1                                       ; preds = %1, %0
+  %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ]
+  %2 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv
+  %3 = load i32, ptr %2, align 4
+  %4 = add nsw i32 %3, 6
+  store i32 %4, ptr %2, align 4
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 24
+  br i1 %exitcond, label %5, label %1
+
+; <label>:5                                       ; preds = %1
+  ret i32 undef
+}
+
+; CHECK-VECTOR-LABEL: @foo_trip_count_25(
+; CHECK-VECTOR: load <4 x i32>
+; CHECK-VECTOR-NOT: load <4 x i32>
+; CHECK-VECTOR: store <4 x i32>
+; CHECK-VECTOR-NOT: store <4 x i32>
+; CHECK-VECTOR: ret
+;
+; CHECK-SCALAR-LABEL: @foo_trip_count_25(
+; CHECK-SCALAR: load i32, ptr
+; CHECK-SCALAR-NOT: load i32, ptr
+; CHECK-SCALAR: store i32
+; CHECK-SCALAR-NOT: store i32
+; CHECK-SCALAR: ret
+define i32 @foo_trip_count_25(ptr nocapture %A) nounwind uwtable ssp {
+  br label %1
+
+; <label>:1                                       ; preds = %1, %0
+  %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ]
+  %2 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv
+  %3 = load i32, ptr %2, align 4
+  %4 = add nsw i32 %3, 6
+  store i32 %4, ptr %2, align 4
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 25
+  br i1 %exitcond, label %5, label %1
+
+; <label>:5                                       ; preds = %1
+  ret i32 undef
+}
+
+; CHECK-VECTOR-LABEL: @foo_trip_count_33(
+; CHECK-VECTOR: load <4 x i32>
+; CHECK-VECTOR-NOT: load <4 x i32>
+; CHECK-VECTOR: store <4 x i32>
+; CHECK-VECTOR-NOT: store <4 x i32>
+; CHECK-VECTOR: ret
+;
+; CHECK-SCALAR-LABEL: @foo_trip_count_33(
+; CHECK-SCALAR: load i32, ptr
+; CHECK-SCALAR-NOT: load i32, ptr
+; CHECK-SCALAR: store i32
+; CHECK-SCALAR-NOT: store i32
+; CHECK-SCALAR: ret
+define i32 @foo_trip_count_33(ptr nocapture %A) nounwind uwtable ssp {
+  br label %1
+
+; <label>:1                                       ; preds = %1, %0
+  %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ]
+  %2 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv
+  %3 = load i32, ptr %2, align 4
+  %4 = add nsw i32 %3, 6
+  store i32 %4, ptr %2, align 4
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 33
+  br i1 %exitcond, label %5, label %1
+
+; <label>:5                                       ; preds = %1
+  ret i32 undef
+}
+
+; CHECK-VECTOR-LABEL: @foo_trip_count_101(
+; CHECK-VECTOR: load <4 x i32>
+; CHECK-VECTOR-NOT: load <4 x i32>
+; CHECK-VECTOR: store <4 x i32>
+; CHECK-VECTOR-NOT: store <4 x i32>
+; CHECK-VECTOR: ret
+;
+; CHECK-SCALAR-LABEL: @foo_trip_count_101(
+; CHECK-SCALAR: load i32, ptr
+; CHECK-SCALAR-NOT: load i32, ptr
+; CHECK-SCALAR: store i32
+; CHECK-SCALAR-NOT: store i32
+; CHECK-SCALAR: ret
+define i32 @foo_trip_count_101(ptr nocapture %A) nounwind uwtable ssp {
+  br label %1
+
+; <label>:1                                       ; preds = %1, %0
+  %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ]
+  %2 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv
+  %3 = load i32, ptr %2, align 4
+  %4 = add nsw i32 %3, 6
+  store i32 %4, ptr %2, align 4
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 101
   br i1 %exitcond, label %5, label %1
 
 ; <label>:5                                       ; preds = %1

``````````

</details>


https://github.com/llvm/llvm-project/pull/70272

More information about the llvm-commits mailing list