bb2791609d
UpdateTestChecks has a make_analyzer_generalizer to replace pointer addressess from the debug output of LAA with a pattern, which is an acceptable solution when there is one RUN line. However, when there are multiple RUN lines with a common pattern, UTC fails to recognize common output due to mismatched pointer addresses. Instead of hacking UTC scrub the output before comparing the outputs from the different RUN lines, fix the issue once and for all by making LAA not output unstable pointer addresses in the first place. The removal of the now-dead make_analyzer_generalizer is left as a non-trivial exercise for a follow-up.
230 lines
8.2 KiB
LLVM
230 lines
8.2 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 4
|
|
; RUN: opt -passes='print<access-info>' -disable-output %s 2>&1 | FileCheck %s
|
|
|
|
target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
|
|
|
|
declare void @llvm.assume(i1)
|
|
|
|
declare void @use(ptr noundef)
|
|
|
|
; %offset is known positive via assume, so we should be able to detect the
|
|
; forward dependence.
|
|
define void @offset_i8_known_positive_via_assume_forward_dep_1(ptr %A, i64 %offset, i64 %N) {
|
|
; CHECK-LABEL: 'offset_i8_known_positive_via_assume_forward_dep_1'
|
|
; CHECK-NEXT: loop:
|
|
; CHECK-NEXT: Memory dependences are safe
|
|
; CHECK-NEXT: Dependences:
|
|
; CHECK-NEXT: Forward:
|
|
; CHECK-NEXT: %l = load i8, ptr %gep.off, align 4 ->
|
|
; CHECK-NEXT: store i8 %add, ptr %gep, align 4
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: Run-time memory checks:
|
|
; CHECK-NEXT: Grouped accesses:
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
|
|
; CHECK-NEXT: SCEV assumptions:
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: Expressions re-written:
|
|
;
|
|
entry:
|
|
%c = icmp sgt i64 %offset, 0
|
|
call void @llvm.assume(i1 %c)
|
|
%off = getelementptr inbounds i8, ptr %A, i64 %offset
|
|
call void @use(ptr noundef %off)
|
|
br label %loop
|
|
|
|
loop:
|
|
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
|
|
%gep.off = getelementptr inbounds i8, ptr %off, i64 %iv
|
|
%l = load i8 , ptr %gep.off, align 4
|
|
%add = add nsw i8 %l, 5
|
|
%gep = getelementptr inbounds i8, ptr %A, i64 %iv
|
|
store i8 %add, ptr %gep, align 4
|
|
%iv.next = add nuw nsw i64 %iv, 1
|
|
%exitcond.not = icmp eq i64 %iv.next, %N
|
|
br i1 %exitcond.not, label %exit, label %loop
|
|
|
|
exit:
|
|
ret void
|
|
}
|
|
|
|
define void @offset_i32_known_positive_via_assume_forward_dep_1(ptr %A, i64 %offset, i64 %N) {
|
|
; CHECK-LABEL: 'offset_i32_known_positive_via_assume_forward_dep_1'
|
|
; CHECK-NEXT: loop:
|
|
; CHECK-NEXT: Memory dependences are safe with run-time checks
|
|
; CHECK-NEXT: Dependences:
|
|
; CHECK-NEXT: Run-time memory checks:
|
|
; CHECK-NEXT: Check 0:
|
|
; CHECK-NEXT: Comparing group GRP0:
|
|
; CHECK-NEXT: %gep = getelementptr inbounds i32, ptr %A, i64 %iv
|
|
; CHECK-NEXT: Against group GRP1:
|
|
; CHECK-NEXT: %gep.off = getelementptr inbounds i32, ptr %off, i64 %iv
|
|
; CHECK-NEXT: Grouped accesses:
|
|
; CHECK-NEXT: Group GRP0:
|
|
; CHECK-NEXT: (Low: %A High: (-3 + (4 * %N) + %A))
|
|
; CHECK-NEXT: Member: {%A,+,4}<nuw><%loop>
|
|
; CHECK-NEXT: Group GRP1:
|
|
; CHECK-NEXT: (Low: ((4 * %offset)<nsw> + %A) High: (-3 + (4 * %offset)<nsw> + (4 * %N) + %A))
|
|
; CHECK-NEXT: Member: {((4 * %offset)<nsw> + %A),+,4}<nw><%loop>
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
|
|
; CHECK-NEXT: SCEV assumptions:
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: Expressions re-written:
|
|
;
|
|
entry:
|
|
%c = icmp sgt i64 %offset, 0
|
|
call void @llvm.assume(i1 %c)
|
|
%off = getelementptr inbounds i32, ptr %A, i64 %offset
|
|
call void @use(ptr noundef %off)
|
|
|
|
br label %loop
|
|
|
|
loop:
|
|
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
|
|
%gep.off = getelementptr inbounds i32, ptr %off, i64 %iv
|
|
%l = load i8 , ptr %gep.off, align 4
|
|
%add = add nsw i8 %l, 5
|
|
%gep = getelementptr inbounds i32, ptr %A, i64 %iv
|
|
store i8 %add, ptr %gep, align 4
|
|
%iv.next = add nuw nsw i64 %iv, 1
|
|
%exitcond.not = icmp eq i64 %iv.next, %N
|
|
br i1 %exitcond.not, label %exit, label %loop
|
|
|
|
exit:
|
|
ret void
|
|
}
|
|
|
|
; %offset is known positive via assume, so we should be able to detect the
|
|
; forward dependence.
|
|
define void @offset_known_positive_via_assume_forward_dep_2(ptr %A, i64 %offset, i64 %N) {
|
|
; CHECK-LABEL: 'offset_known_positive_via_assume_forward_dep_2'
|
|
; CHECK-NEXT: loop:
|
|
; CHECK-NEXT: Memory dependences are safe
|
|
; CHECK-NEXT: Dependences:
|
|
; CHECK-NEXT: Forward:
|
|
; CHECK-NEXT: %l = load i32, ptr %gep.off, align 4 ->
|
|
; CHECK-NEXT: store i32 %add, ptr %gep, align 4
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: Run-time memory checks:
|
|
; CHECK-NEXT: Grouped accesses:
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
|
|
; CHECK-NEXT: SCEV assumptions:
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: Expressions re-written:
|
|
;
|
|
entry:
|
|
%c = icmp sgt i64 %offset, 0
|
|
call void @llvm.assume(i1 %c)
|
|
%c.2 = icmp slt i64 %offset, 20
|
|
call void @llvm.assume(i1 %c.2)
|
|
%off = getelementptr inbounds i32, ptr %A, i64 %offset
|
|
br label %loop
|
|
|
|
loop:
|
|
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
|
|
%gep.off = getelementptr inbounds i32, ptr %off, i64 %iv
|
|
%l = load i32, ptr %gep.off, align 4
|
|
%add = add nsw i32 %l, 5
|
|
%gep = getelementptr inbounds i32, ptr %A, i64 %iv
|
|
store i32 %add, ptr %gep, align 4
|
|
%iv.next = add nuw nsw i64 %iv, 1
|
|
%exitcond.not = icmp eq i64 %iv.next, %N
|
|
br i1 %exitcond.not, label %exit, label %loop
|
|
|
|
exit:
|
|
ret void
|
|
}
|
|
|
|
; The range of %offset is known via assumes, but it may be positive or negative.
|
|
define void @offset_may_be_negative_via_assume_unknown_dep(ptr %A, i64 %offset, i64 %N) {
|
|
; CHECK-LABEL: 'offset_may_be_negative_via_assume_unknown_dep'
|
|
; CHECK-NEXT: loop:
|
|
; CHECK-NEXT: Memory dependences are safe with run-time checks
|
|
; CHECK-NEXT: Dependences:
|
|
; CHECK-NEXT: Run-time memory checks:
|
|
; CHECK-NEXT: Check 0:
|
|
; CHECK-NEXT: Comparing group GRP0:
|
|
; CHECK-NEXT: %gep.mul.2 = getelementptr inbounds i32, ptr %A, i64 %iv
|
|
; CHECK-NEXT: Against group GRP1:
|
|
; CHECK-NEXT: %gep = getelementptr inbounds i32, ptr %off, i64 %iv
|
|
; CHECK-NEXT: Grouped accesses:
|
|
; CHECK-NEXT: Group GRP0:
|
|
; CHECK-NEXT: (Low: %A High: ((4 * %N) + %A))
|
|
; CHECK-NEXT: Member: {%A,+,4}<nuw><%loop>
|
|
; CHECK-NEXT: Group GRP1:
|
|
; CHECK-NEXT: (Low: ((4 * %offset)<nsw> + %A) High: ((4 * %offset)<nsw> + (4 * %N) + %A))
|
|
; CHECK-NEXT: Member: {((4 * %offset)<nsw> + %A),+,4}<nw><%loop>
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
|
|
; CHECK-NEXT: SCEV assumptions:
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: Expressions re-written:
|
|
;
|
|
entry:
|
|
%c = icmp sgt i64 %offset, -4
|
|
call void @llvm.assume(i1 %c)
|
|
%c.2 = icmp slt i64 %offset, 20
|
|
call void @llvm.assume(i1 %c.2)
|
|
%off = getelementptr inbounds i32, ptr %A, i64 %offset
|
|
br label %loop
|
|
|
|
loop:
|
|
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
|
|
%gep = getelementptr inbounds i32, ptr %off, i64 %iv
|
|
%l = load i32, ptr %gep, align 4
|
|
%add = add nsw i32 %l, 5
|
|
%gep.mul.2 = getelementptr inbounds i32, ptr %A, i64 %iv
|
|
store i32 %add, ptr %gep.mul.2, align 4
|
|
%iv.next = add nuw nsw i64 %iv, 1
|
|
%exitcond.not = icmp eq i64 %iv.next, %N
|
|
br i1 %exitcond.not, label %exit, label %loop
|
|
|
|
exit:
|
|
ret void
|
|
}
|
|
|
|
define void @offset_no_assumes(ptr %A, i64 %offset, i64 %N) {
|
|
; CHECK-LABEL: 'offset_no_assumes'
|
|
; CHECK-NEXT: loop:
|
|
; CHECK-NEXT: Memory dependences are safe with run-time checks
|
|
; CHECK-NEXT: Dependences:
|
|
; CHECK-NEXT: Run-time memory checks:
|
|
; CHECK-NEXT: Check 0:
|
|
; CHECK-NEXT: Comparing group GRP0:
|
|
; CHECK-NEXT: %gep = getelementptr inbounds i32, ptr %A, i64 %iv
|
|
; CHECK-NEXT: Against group GRP1:
|
|
; CHECK-NEXT: %gep.off = getelementptr inbounds i32, ptr %off, i64 %iv
|
|
; CHECK-NEXT: Grouped accesses:
|
|
; CHECK-NEXT: Group GRP0:
|
|
; CHECK-NEXT: (Low: %A High: ((4 * %N) + %A))
|
|
; CHECK-NEXT: Member: {%A,+,4}<nuw><%loop>
|
|
; CHECK-NEXT: Group GRP1:
|
|
; CHECK-NEXT: (Low: ((4 * %offset)<nsw> + %A) High: ((4 * %offset)<nsw> + (4 * %N) + %A))
|
|
; CHECK-NEXT: Member: {((4 * %offset)<nsw> + %A),+,4}<nw><%loop>
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
|
|
; CHECK-NEXT: SCEV assumptions:
|
|
; CHECK-EMPTY:
|
|
; CHECK-NEXT: Expressions re-written:
|
|
;
|
|
entry:
|
|
%off = getelementptr inbounds i32, ptr %A, i64 %offset
|
|
br label %loop
|
|
|
|
loop:
|
|
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
|
|
%gep.off = getelementptr inbounds i32, ptr %off, i64 %iv
|
|
%l = load i32, ptr %gep.off, align 4
|
|
%add = add nsw i32 %l, 5
|
|
%gep = getelementptr inbounds i32, ptr %A, i64 %iv
|
|
store i32 %add, ptr %gep, align 4
|
|
%iv.next = add nuw nsw i64 %iv, 1
|
|
%exitcond.not = icmp eq i64 %iv.next, %N
|
|
br i1 %exitcond.not, label %exit, label %loop
|
|
|
|
exit:
|
|
ret void
|
|
}
|