41fc9b9845
When a memory-reading or memory-writing instruction is not a LoadInst/StoreInst, the dyn_cast to Ld/St returns nullptr, which is then passed to recordAnalysis. This causes the optimization remark to fall back to the loop header location instead of pointing at the actual problematic instruction. Pass &I (the actual Instruction) instead.
96 lines
3.3 KiB
LLVM
96 lines
3.3 KiB
LLVM
; RUN: opt < %s -passes=loop-vectorize -disable-output -pass-remarks-output=- 2>&1 | FileCheck %s
|
|
|
|
; Verify that when a call instruction in a loop cannot be vectorized,
|
|
; the optimization remark points to the call instruction's debug location
|
|
; (source.c:7:10), not the loop header (source.c:5:3).
|
|
;
|
|
; C source code for the tests (line numbers match debug metadata):
|
|
;
|
|
; 1: int opaque(int *);
|
|
; 2: void opaque_write(int *);
|
|
; 3:
|
|
; 4: void call_reads_memory(int *p, long n) {
|
|
; 5: for (long i = 0; i < n; i++) {
|
|
; 6: int *addr = &p[i];
|
|
; 7: int val = opaque(addr);
|
|
; 8: p[i] = val;
|
|
; 9: }
|
|
; 10: }
|
|
; 11:
|
|
; 12:
|
|
; 13:
|
|
; 14: void call_writes_memory(int *p, long n) {
|
|
; 15: for (long i = 0; i < n; i++) {
|
|
; 16: int *addr = &p[i];
|
|
; 17: opaque_write(addr);
|
|
; 18: }
|
|
; 19:
|
|
; 20: }
|
|
|
|
; The remark should point at the call instruction, not the loop header.
|
|
; CHECK: --- !Analysis
|
|
; CHECK: Pass: loop-vectorize
|
|
; CHECK: Name: CantVectorizeInstruction
|
|
; CHECK-NEXT: DebugLoc: { File: source.c, Line: 7, Column: 10 }
|
|
; CHECK-NEXT: Function: call_reads_memory
|
|
; CHECK-NEXT: Args:
|
|
; CHECK-NEXT: - String: 'loop not vectorized: '
|
|
; CHECK-NEXT: - String: instruction cannot be vectorized
|
|
; CHECK: --- !Analysis
|
|
; CHECK: Pass: loop-vectorize
|
|
; CHECK: Name: CantVectorizeInstruction
|
|
; CHECK-NEXT: DebugLoc: { File: source.c, Line: 17, Column: 5 }
|
|
; CHECK-NEXT: Function: call_writes_memory
|
|
; CHECK-NEXT: Args:
|
|
; CHECK-NEXT: - String: 'loop not vectorized: '
|
|
; CHECK-NEXT: - String: instruction cannot be vectorized
|
|
|
|
define void @call_reads_memory(ptr %p, i64 %n) !dbg !4 {
|
|
entry:
|
|
br label %loop
|
|
|
|
loop:
|
|
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
|
|
%gep = getelementptr inbounds i32, ptr %p, i64 %iv
|
|
%val = call i32 @opaque(ptr %gep), !dbg !5
|
|
store i32 %val, ptr %gep, align 4
|
|
%iv.next = add nuw nsw i64 %iv, 1
|
|
%exitcond = icmp eq i64 %iv.next, %n
|
|
br i1 %exitcond, label %exit, label %loop
|
|
|
|
exit:
|
|
ret void
|
|
}
|
|
|
|
define void @call_writes_memory(ptr %p, i64 %n) !dbg !6 {
|
|
entry:
|
|
br label %loop
|
|
|
|
loop:
|
|
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
|
|
%gep = getelementptr inbounds i32, ptr %p, i64 %iv
|
|
call void @opaque_write(ptr %gep), !dbg !7
|
|
%iv.next = add nuw nsw i64 %iv, 1
|
|
%exitcond = icmp eq i64 %iv.next, %n
|
|
br i1 %exitcond, label %exit, label %loop
|
|
|
|
exit:
|
|
ret void
|
|
}
|
|
|
|
declare i32 @opaque(ptr)
|
|
declare void @opaque_write(ptr)
|
|
|
|
!llvm.dbg.cu = !{!0}
|
|
!llvm.module.flags = !{!2, !3}
|
|
|
|
!0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly)
|
|
!1 = !DIFile(filename: "source.c", directory: "/tmp")
|
|
!2 = !{i32 2, !"Dwarf Version", i32 4}
|
|
!3 = !{i32 2, !"Debug Info Version", i32 3}
|
|
!4 = distinct !DISubprogram(name: "call_reads_memory", scope: !1, file: !1, line: 4, type: !8, scopeLine: 4, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0)
|
|
!5 = !DILocation(line: 7, column: 10, scope: !4)
|
|
!6 = distinct !DISubprogram(name: "call_writes_memory", scope: !1, file: !1, line: 14, type: !8, scopeLine: 14, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0)
|
|
!7 = !DILocation(line: 17, column: 5, scope: !6)
|
|
!8 = !DISubroutineType(types: !{})
|