b7315ffc3c
This patch adds initial support for a pointer diff based runtime check scheme for vectorization. This scheme requires fewer computations and checks than the existing full overlap checking, if it is applicable. The main idea is to only check if source and sink of a dependency are far enough apart so the accesses won't overlap in the vector loop. To do so, it is sufficient to compute the difference and compare it to the `VF * UF * AccessSize`. It is sufficient to check `(Sink - Src) <u VF * UF * AccessSize` to rule out a backwards dependence in the vector loop with the given VF and UF. If Src >=u Sink, there is not dependence preventing vectorization, hence the overflow should not matter and using the ULT should be sufficient. Note that the initial version is restricted in multiple ways: 1. Pointers must only either be read or written, by a single instruction (this allows re-constructing source/sink for dependences with the available information) 2. Source and sink pointers must be add-recs, with matching steps 3. The step must be a constant. 3. abs(step) == AccessSize. Most of those restrictions can be relaxed in the future. See https://github.com/llvm/llvm-project/issues/53590. Reviewed By: dmgreen Differential Revision: https://reviews.llvm.org/D119078
68 lines
2.0 KiB
LLVM
68 lines
2.0 KiB
LLVM
; RUN: opt -loop-vectorize -force-vector-width=2 -S %s | FileCheck %s
|
|
|
|
; Test cases to make sure LV & loop versioning can handle loops with
|
|
; multiple exiting branches.
|
|
|
|
; Multiple branches exiting the loop to a unique exit block. The loop should
|
|
; be vectorized with versioning.
|
|
define void @multiple_exits_unique_exit_block(i32* %A, i32* %B, i64 %N) {
|
|
; CHECK-LABEL: @multiple_exits_unique_exit_block
|
|
; CHECK: vector.memcheck:
|
|
; CHECK-LABEL: vector.body:
|
|
; CHECK: %wide.load = load <2 x i32>, <2 x i32>* {{.*}}, align 4
|
|
; CHECK: store <2 x i32> %wide.load, <2 x i32>* {{.*}}, align 4
|
|
; CHECK: br
|
|
;
|
|
entry:
|
|
br label %loop.header
|
|
|
|
loop.header:
|
|
%iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
|
|
%cond.0 = icmp eq i64 %iv, %N
|
|
br i1 %cond.0, label %exit, label %for.body
|
|
|
|
for.body:
|
|
%A.gep = getelementptr inbounds i32, i32* %A, i64 %iv
|
|
%lv = load i32, i32* %A.gep, align 4
|
|
%B.gep = getelementptr inbounds i32, i32* %B, i64 %iv
|
|
store i32 %lv, i32* %B.gep, align 4
|
|
%iv.next = add nuw i64 %iv, 1
|
|
%cond.1 = icmp ult i64 %iv.next, 1000
|
|
br i1 %cond.1, label %loop.header, label %exit
|
|
|
|
exit:
|
|
ret void
|
|
}
|
|
|
|
|
|
; Multiple branches exiting the loop to different blocks. Currently this is not supported.
|
|
define i32 @multiple_exits_multiple_exit_blocks(i32* %A, i32* %B, i64 %N) {
|
|
; CHECK-LABEL: @multiple_exits_multiple_exit_blocks
|
|
; CHECK-NEXT: entry:
|
|
; CHECK: br label %loop.header
|
|
; CHECK-NOT: <2 x i32>
|
|
;
|
|
entry:
|
|
br label %loop.header
|
|
|
|
loop.header:
|
|
%iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
|
|
%cond.0 = icmp eq i64 %iv, %N
|
|
br i1 %cond.0, label %exit.0, label %for.body
|
|
|
|
for.body:
|
|
%A.gep = getelementptr inbounds i32, i32* %A, i64 %iv
|
|
%lv = load i32, i32* %A.gep, align 4
|
|
%B.gep = getelementptr inbounds i32, i32* %B, i64 %iv
|
|
store i32 %lv, i32* %B.gep, align 4
|
|
%iv.next = add nuw i64 %iv, 1
|
|
%cond.1 = icmp ult i64 %iv.next, 1000
|
|
br i1 %cond.1, label %loop.header, label %exit.1
|
|
|
|
exit.0:
|
|
ret i32 1
|
|
|
|
exit.1:
|
|
ret i32 2
|
|
}
|