182 lines
6.7 KiB
LLVM
182 lines
6.7 KiB
LLVM
|
; RUN: opt < %s -loop-vectorize -force-vector-width=4 -dce -instcombine -S | FileCheck %s
|
||
|
|
|
||
|
|
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
|
||
|
|
|
||
|
|
; tests skipping iterations within a VF through break/continue/gotos.
|
||
|
|
|
||
|
|
; The main difficulty in vectorizing these loops in test1,test2 and test3 is
|
||
|
|
; safely speculating that the widened load of A[i] should not fault if the
|
||
|
|
; scalarized loop does not fault. For example, the
|
||
|
|
; original load in the scalar loop may not fault, but the last iteration of the
|
||
|
|
; vectorized load can fault (if it crosses a page boudary for example).
|
||
|
|
; This last vector iteration is where *one* of the
|
||
|
|
; scalar iterations lead to the early exit.
|
||
|
|
|
||
|
|
; int test(int *A, int Length) {
|
||
|
|
; for (int i = 0; i < Length; i++) {
|
||
|
|
; if (A[i] > 10.0) goto end;
|
||
|
|
; A[i] = 0;
|
||
|
|
; }
|
||
|
|
; end:
|
||
|
|
; return 0;
|
||
|
|
; }
|
||
|
|
; CHECK-LABEL: test1(
|
||
|
|
; CHECK-NOT: <4 x i32>
|
||
|
|
define i32 @test1(i32* nocapture %A, i32 %Length) {
|
||
|
|
entry:
|
||
|
|
%cmp8 = icmp sgt i32 %Length, 0
|
||
|
|
br i1 %cmp8, label %for.body.preheader, label %end
|
||
|
|
|
||
|
|
for.body.preheader: ; preds = %entry
|
||
|
|
br label %for.body
|
||
|
|
|
||
|
|
for.body: ; preds = %for.body.preheader, %if.else
|
||
|
|
%indvars.iv = phi i64 [ %indvars.iv.next, %if.else ], [ 0, %for.body.preheader ]
|
||
|
|
%arrayidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
|
||
|
|
%0 = load i32, i32* %arrayidx, align 4, !tbaa !15
|
||
|
|
%cmp1 = icmp sgt i32 %0, 10
|
||
|
|
br i1 %cmp1, label %end.loopexit, label %if.else
|
||
|
|
|
||
|
|
if.else: ; preds = %for.body
|
||
|
|
store i32 0, i32* %arrayidx, align 4, !tbaa !15
|
||
|
|
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
|
||
|
|
%1 = trunc i64 %indvars.iv.next to i32
|
||
|
|
%cmp = icmp slt i32 %1, %Length
|
||
|
|
br i1 %cmp, label %for.body, label %end.loopexit
|
||
|
|
|
||
|
|
end.loopexit: ; preds = %if.else, %for.body
|
||
|
|
br label %end
|
||
|
|
|
||
|
|
end: ; preds = %end.loopexit, %entry
|
||
|
|
ret i32 0
|
||
|
|
}
|
||
|
|
|
||
|
|
; We don't use anything from within the loop at the early exit path
|
||
|
|
; so we do not need to know which iteration caused the early exit path.
|
||
|
|
; bool test2(int *A, int Length, int K) {
|
||
|
|
; for (int i = 0; i < Length; i++) {
|
||
|
|
; if (A[i] == K) return true;
|
||
|
|
; }
|
||
|
|
; return false;
|
||
|
|
; }
|
||
|
|
; TODO: Today we do not vectorize this, but we could teach the vectorizer, once
|
||
|
|
; the hard part of proving/speculating A[i:VF - 1] loads does not fault is handled by the
|
||
|
|
; compiler/hardware.
|
||
|
|
|
||
|
|
; CHECK-LABEL: test2(
|
||
|
|
; CHECK-NOT: <4 x i32>
|
||
|
|
define i32 @test2(i32* nocapture %A, i32 %Length, i32 %K) {
|
||
|
|
entry:
|
||
|
|
%cmp8 = icmp sgt i32 %Length, 0
|
||
|
|
br i1 %cmp8, label %for.body.preheader, label %end
|
||
|
|
|
||
|
|
for.body.preheader: ; preds = %entry
|
||
|
|
br label %for.body
|
||
|
|
|
||
|
|
for.body: ; preds = %for.body.preheader, %if.else
|
||
|
|
%indvars.iv = phi i64 [ %indvars.iv.next, %if.else ], [ 0, %for.body.preheader ]
|
||
|
|
%arrayidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
|
||
|
|
%ld = load i32, i32* %arrayidx, align 4
|
||
|
|
%cmp1 = icmp eq i32 %ld, %K
|
||
|
|
br i1 %cmp1, label %end.loopexit, label %if.else
|
||
|
|
|
||
|
|
if.else: ; preds = %for.body
|
||
|
|
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
|
||
|
|
%trunc = trunc i64 %indvars.iv.next to i32
|
||
|
|
%cmp = icmp slt i32 %trunc, %Length
|
||
|
|
br i1 %cmp, label %for.body, label %end.loopexit
|
||
|
|
|
||
|
|
end.loopexit: ; preds = %if.else, %for.body
|
||
|
|
%result.lcssa = phi i32 [ 1, %for.body ], [ 0, %if.else ]
|
||
|
|
br label %end
|
||
|
|
|
||
|
|
end: ; preds = %end.loopexit, %entry
|
||
|
|
%result = phi i32 [ %result.lcssa, %end.loopexit ], [ 0, %entry ]
|
||
|
|
ret i32 %result
|
||
|
|
}
|
||
|
|
|
||
|
|
; We use the IV in the early exit
|
||
|
|
; so we need to know which iteration caused the early exit path.
|
||
|
|
; int test3(int *A, int Length, int K) {
|
||
|
|
; for (int i = 0; i < Length; i++) {
|
||
|
|
; if (A[i] == K) return i;
|
||
|
|
; }
|
||
|
|
; return -1;
|
||
|
|
; }
|
||
|
|
; TODO: Today we do not vectorize this, but we could teach the vectorizer (once
|
||
|
|
; we handle the speculation safety of the widened load).
|
||
|
|
; CHECK-LABEL: test3(
|
||
|
|
; CHECK-NOT: <4 x i32>
|
||
|
|
define i32 @test3(i32* nocapture %A, i32 %Length, i32 %K) {
|
||
|
|
entry:
|
||
|
|
%cmp8 = icmp sgt i32 %Length, 0
|
||
|
|
br i1 %cmp8, label %for.body.preheader, label %end
|
||
|
|
|
||
|
|
for.body.preheader: ; preds = %entry
|
||
|
|
br label %for.body
|
||
|
|
|
||
|
|
for.body: ; preds = %for.body.preheader, %if.else
|
||
|
|
%indvars.iv = phi i64 [ %indvars.iv.next, %if.else ], [ 0, %for.body.preheader ]
|
||
|
|
%arrayidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
|
||
|
|
%ld = load i32, i32* %arrayidx, align 4
|
||
|
|
%cmp1 = icmp eq i32 %ld, %K
|
||
|
|
br i1 %cmp1, label %end.loopexit, label %if.else
|
||
|
|
|
||
|
|
if.else: ; preds = %for.body
|
||
|
|
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
|
||
|
|
%trunc = trunc i64 %indvars.iv.next to i32
|
||
|
|
%cmp = icmp slt i32 %trunc, %Length
|
||
|
|
br i1 %cmp, label %for.body, label %end.loopexit
|
||
|
|
|
||
|
|
end.loopexit: ; preds = %if.else, %for.body
|
||
|
|
%result.lcssa = phi i64 [ %indvars.iv, %for.body ], [ -1, %if.else ]
|
||
|
|
%res.trunc = trunc i64 %result.lcssa to i32
|
||
|
|
br label %end
|
||
|
|
|
||
|
|
end: ; preds = %end.loopexit, %entry
|
||
|
|
%result = phi i32 [ %res.trunc, %end.loopexit ], [ -1, %entry ]
|
||
|
|
ret i32 %result
|
||
|
|
}
|
||
|
|
|
||
|
|
; bool test4(int *A, int Length, int K, int J) {
|
||
|
|
; for (int i = 0; i < Length; i++) {
|
||
|
|
; if (A[i] == K) continue;
|
||
|
|
; A[i] = J;
|
||
|
|
; }
|
||
|
|
; }
|
||
|
|
; For this test, we vectorize and generate predicated stores to A[i].
|
||
|
|
; CHECK-LABEL: test4(
|
||
|
|
; CHECK: <4 x i32>
|
||
|
|
define void @test4(i32* nocapture %A, i32 %Length, i32 %K, i32 %J) {
|
||
|
|
entry:
|
||
|
|
%cmp8 = icmp sgt i32 %Length, 0
|
||
|
|
br i1 %cmp8, label %for.body.preheader, label %end.loopexit
|
||
|
|
|
||
|
|
for.body.preheader: ; preds = %entry
|
||
|
|
br label %for.body
|
||
|
|
|
||
|
|
for.body: ; preds = %for.body.preheader, %if.else
|
||
|
|
%indvars.iv = phi i64 [ %indvars.iv.next, %latch ], [ 0, %for.body.preheader ]
|
||
|
|
%arrayidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
|
||
|
|
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
|
||
|
|
%trunc = trunc i64 %indvars.iv.next to i32
|
||
|
|
%ld = load i32, i32* %arrayidx, align 4
|
||
|
|
%cmp1 = icmp eq i32 %ld, %K
|
||
|
|
br i1 %cmp1, label %latch, label %if.else
|
||
|
|
|
||
|
|
if.else:
|
||
|
|
store i32 %J, i32* %arrayidx, align 4
|
||
|
|
br label %latch
|
||
|
|
|
||
|
|
latch: ; preds = %for.body
|
||
|
|
%cmp = icmp slt i32 %trunc, %Length
|
||
|
|
br i1 %cmp, label %for.body, label %end.loopexit
|
||
|
|
|
||
|
|
end.loopexit: ; preds = %if.else, %for.body
|
||
|
|
ret void
|
||
|
|
}
|
||
|
|
!15 = !{!16, !16, i64 0}
|
||
|
|
!16 = !{!"int", !17, i64 0}
|
||
|
|
!17 = !{!"omnipotent char", !18, i64 0}
|
||
|
|
!18 = !{!"Simple C/C++ TBAA"}
|