fb7c0d70a7
Trip count expressions sometimes consist of adding 3 operands, i.e. (Const + A + B). There may be guard info for A + B, and if so, apply it. We can probably more generally apply this, but need to be careful w.r.t compile-time. Alive2 Proof for changes in miniters.ll: https://alive2.llvm.org/ce/z/HFfXOx Fixes https://github.com/llvm/llvm-project/issues/155941. PR: https://github.com/llvm/llvm-project/pull/156013
98 lines
4.0 KiB
LLVM
98 lines
4.0 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --filter-out-after "vector.ph:" --version 5
|
|
; RUN: opt %s -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -S | FileCheck %s
|
|
; RUN: opt %s -passes=loop-vectorize -force-vector-interleave=2 -force-vector-width=4 -S | FileCheck %s -check-prefix=UNROLL
|
|
|
|
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
|
|
|
|
@b = common global [1000 x i32] zeroinitializer, align 16
|
|
@c = common global [1000 x i32] zeroinitializer, align 16
|
|
@a = common global [1000 x i32] zeroinitializer, align 16
|
|
|
|
; Generate min.iters.check to skip the vector loop and jump to scalar.ph directly when loop iteration number is less than VF * UF.
|
|
define void @foo(i64 %N) {
|
|
; CHECK-LABEL: define void @foo(
|
|
; CHECK-SAME: i64 [[N:%.*]]) {
|
|
; CHECK-NEXT: [[ENTRY:.*:]]
|
|
; CHECK-NEXT: [[C:%.*]] = icmp sgt i64 [[N]], 0
|
|
; CHECK-NEXT: br i1 [[C]], label %[[LOOP_PREHEADER:.*]], [[EXIT:label %.*]]
|
|
; CHECK: [[LOOP_PREHEADER]]:
|
|
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
|
|
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], [[SCALAR_PH:label %.*]], label %[[VECTOR_PH:.*]]
|
|
; CHECK: [[VECTOR_PH]]:
|
|
;
|
|
; UNROLL-LABEL: define void @foo(
|
|
; UNROLL-SAME: i64 [[N:%.*]]) {
|
|
; UNROLL-NEXT: [[ENTRY:.*:]]
|
|
; UNROLL-NEXT: [[C:%.*]] = icmp sgt i64 [[N]], 0
|
|
; UNROLL-NEXT: br i1 [[C]], label %[[LOOP_PREHEADER:.*]], [[EXIT:label %.*]]
|
|
; UNROLL: [[LOOP_PREHEADER]]:
|
|
; UNROLL-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 8
|
|
; UNROLL-NEXT: br i1 [[MIN_ITERS_CHECK]], [[SCALAR_PH:label %.*]], label %[[VECTOR_PH:.*]]
|
|
; UNROLL: [[VECTOR_PH]]:
|
|
;
|
|
entry:
|
|
%c = icmp sgt i64 %N, 0
|
|
br i1 %c, label %loop, label %exit
|
|
|
|
loop:
|
|
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
|
|
%arrayidx = getelementptr inbounds [1000 x i32], ptr @b, i64 0, i64 %iv
|
|
%tmp = load i32, ptr %arrayidx, align 4
|
|
%arrayidx1 = getelementptr inbounds [1000 x i32], ptr @c, i64 0, i64 %iv
|
|
%tmp1 = load i32, ptr %arrayidx1, align 4
|
|
%add = add nsw i32 %tmp1, %tmp
|
|
%arrayidx2 = getelementptr inbounds [1000 x i32], ptr @a, i64 0, i64 %iv
|
|
store i32 %add, ptr %arrayidx2, align 4
|
|
%iv.next = add nuw nsw i64 %iv, 1
|
|
%ec = icmp eq i64 %iv.next, %N
|
|
br i1 %ec, label %exit, label %loop
|
|
|
|
exit:
|
|
ret void
|
|
}
|
|
|
|
define void @min_iters_known_via_loop_guards_add(i32 %start, i32 %end, ptr %src) {
|
|
; CHECK-LABEL: define void @min_iters_known_via_loop_guards_add(
|
|
; CHECK-SAME: i32 [[START:%.*]], i32 [[END:%.*]], ptr [[SRC:%.*]]) {
|
|
; CHECK-NEXT: [[ENTRY:.*:]]
|
|
; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[END]], [[START]]
|
|
; CHECK-NEXT: [[PRE:%.*]] = icmp sgt i32 [[SUB]], 100
|
|
; CHECK-NEXT: call void @llvm.assume(i1 [[PRE]])
|
|
; CHECK-NEXT: [[ADD_1:%.*]] = add i32 [[SUB]], 1
|
|
; CHECK-NEXT: [[IV_START:%.*]] = zext i32 [[ADD_1]] to i64
|
|
; CHECK-NEXT: [[TMP0:%.*]] = sub i64 101, [[IV_START]]
|
|
; CHECK-NEXT: br i1 false, [[SCALAR_PH:label %.*]], label %[[VECTOR_PH:.*]]
|
|
; CHECK: [[VECTOR_PH]]:
|
|
;
|
|
; UNROLL-LABEL: define void @min_iters_known_via_loop_guards_add(
|
|
; UNROLL-SAME: i32 [[START:%.*]], i32 [[END:%.*]], ptr [[SRC:%.*]]) {
|
|
; UNROLL-NEXT: [[ENTRY:.*:]]
|
|
; UNROLL-NEXT: [[SUB:%.*]] = sub i32 [[END]], [[START]]
|
|
; UNROLL-NEXT: [[PRE:%.*]] = icmp sgt i32 [[SUB]], 100
|
|
; UNROLL-NEXT: call void @llvm.assume(i1 [[PRE]])
|
|
; UNROLL-NEXT: [[ADD_1:%.*]] = add i32 [[SUB]], 1
|
|
; UNROLL-NEXT: [[IV_START:%.*]] = zext i32 [[ADD_1]] to i64
|
|
; UNROLL-NEXT: [[TMP0:%.*]] = sub i64 101, [[IV_START]]
|
|
; UNROLL-NEXT: br i1 false, [[SCALAR_PH:label %.*]], label %[[VECTOR_PH:.*]]
|
|
; UNROLL: [[VECTOR_PH]]:
|
|
;
|
|
entry:
|
|
%sub = sub i32 %end, %start
|
|
%pre = icmp sgt i32 %sub, 100
|
|
call void @llvm.assume(i1 %pre)
|
|
%add.1 = add i32 %sub, 1
|
|
%iv.start = zext i32 %add.1 to i64
|
|
br label %loop
|
|
|
|
loop:
|
|
%iv = phi i64 [ %iv.start, %entry ], [ %iv.next, %loop ]
|
|
%gep = getelementptr inbounds i64, ptr %src, i64 %iv
|
|
store i64 %iv, ptr %gep
|
|
%iv.next = add i64 %iv, 1
|
|
%ec = icmp eq i64 %iv, 100
|
|
br i1 %ec, label %exit, label %loop
|
|
|
|
exit:
|
|
ret void
|
|
}
|