dd5ba694bd
After clearing the dependencies in copyable data, need to recalculate dependencies for the original ScheduleData, if it can be marked as control dependent. Fixes #153289
57 lines
2.6 KiB
LLVM
57 lines
2.6 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
|
|
; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s
|
|
|
|
define i32 @test(i32 %0, i32 %1) {
|
|
; CHECK-LABEL: define i32 @test(
|
|
; CHECK-SAME: i32 [[TMP0:%.*]], i32 [[TMP1:%.*]]) {
|
|
; CHECK-NEXT: [[ENTRY:.*:]]
|
|
; CHECK-NEXT: [[K:%.*]] = alloca [4 x i32], align 16
|
|
; CHECK-NEXT: [[ADD1:%.*]] = add i32 [[TMP0]], [[TMP1]]
|
|
; CHECK-NEXT: [[SUB2:%.*]] = add i32 [[ADD1]], -1
|
|
; CHECK-NEXT: [[CALL:%.*]] = tail call i32 (ptr, ...) @printf(ptr null, i32 [[ADD1]])
|
|
; CHECK-NEXT: [[ADD2:%.*]] = add i32 [[TMP1]], -1
|
|
; CHECK-NEXT: [[SUB3:%.*]] = add i32 [[ADD2]], [[CALL]]
|
|
; CHECK-NEXT: [[ADD4:%.*]] = add i32 [[SUB3]], [[TMP0]]
|
|
; CHECK-NEXT: store i32 [[ADD4]], ptr [[K]], align 16
|
|
; CHECK-NEXT: [[ARRAYINIT_ELEMENT:%.*]] = getelementptr i8, ptr [[K]], i64 4
|
|
; CHECK-NEXT: store i32 0, ptr [[ARRAYINIT_ELEMENT]], align 4
|
|
; CHECK-NEXT: [[ARRAYINIT_ELEMENT5:%.*]] = getelementptr i8, ptr [[K]], i64 8
|
|
; CHECK-NEXT: [[ADD7:%.*]] = add i32 [[ADD2]], [[SUB2]]
|
|
; CHECK-NEXT: [[SUB8:%.*]] = add i32 [[ADD7]], [[TMP0]]
|
|
; CHECK-NEXT: store i32 [[SUB8]], ptr [[ARRAYINIT_ELEMENT5]], align 8
|
|
; CHECK-NEXT: [[ARRAYINIT_ELEMENT9:%.*]] = getelementptr i8, ptr [[K]], i64 12
|
|
; CHECK-NEXT: [[ADD13:%.*]] = add i32 [[TMP1]], 1
|
|
; CHECK-NEXT: [[ADD10:%.*]] = add i32 [[ADD13]], [[TMP0]]
|
|
; CHECK-NEXT: [[ADD11:%.*]] = add i32 [[ADD10]], [[ADD1]]
|
|
; CHECK-NEXT: [[ADD12:%.*]] = add i32 [[ADD11]], [[TMP0]]
|
|
; CHECK-NEXT: store i32 [[ADD12]], ptr [[ARRAYINIT_ELEMENT9]], align 4
|
|
; CHECK-NEXT: [[CALL15:%.*]] = call i32 (ptr, ...) @printf(ptr null, ptr [[K]])
|
|
; CHECK-NEXT: ret i32 [[CALL15]]
|
|
;
|
|
entry:
|
|
%k = alloca [4 x i32], align 16
|
|
%add1 = add i32 %0, %1
|
|
%sub2 = add i32 %add1, -1
|
|
%call = tail call i32 (ptr, ...) @printf(ptr null, i32 %add1)
|
|
%add2 = add i32 %1, -1
|
|
%sub3 = add i32 %add2, %call
|
|
%add4 = add i32 %sub3, %0
|
|
store i32 %add4, ptr %k, align 16
|
|
%arrayinit.element = getelementptr i8, ptr %k, i64 4
|
|
store i32 0, ptr %arrayinit.element, align 4
|
|
%arrayinit.element5 = getelementptr i8, ptr %k, i64 8
|
|
%add7 = add i32 %add2, %sub2
|
|
%sub8 = add i32 %add7, %0
|
|
store i32 %sub8, ptr %arrayinit.element5, align 8
|
|
%arrayinit.element9 = getelementptr i8, ptr %k, i64 12
|
|
%add13 = add i32 %1, 1
|
|
%add10 = add i32 %add13, %0
|
|
%add11 = add i32 %add10, %add1
|
|
%add12 = add i32 %add11, %0
|
|
store i32 %add12, ptr %arrayinit.element9, align 4
|
|
%call15 = call i32 (ptr, ...) @printf(ptr null, ptr %k)
|
|
ret i32 %call15
|
|
}
|
|
|
|
declare i32 @printf(ptr, ...)
|