mirror of
https://github.com/RPCS3/llvm.git
synced 2024-12-13 23:18:58 +00:00
74615f5548
1) PR25154. This is basically a repeat of PR18102, which was fixed in r200201, and broken again by r234430. The latter changed which of the store nodes was merged into from the first to the last. Thus, we now also need to prefer merging a later store at a given address into the target node, instead of an earlier one. 2) While investigating that, I also realized I'd introduced a bug in r236850. There, I removed a check for alignment -- not realizing that nothing except the alignment check was ensuring that none of the stores were overlapping! This is a really bogus way to ensure there's no aliased stores. A better solution to both of these issues is likely to always use the code added in the 'if (UseAA)' branches which rearrange the chain based on a more principled analysis. I'll look into whether that can be used always, but in the interest of getting things back to working, I think a minimal change makes sense. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251816 91177308-0d34-0410-b5e6-96231b3b80d8
58 lines
2.3 KiB
LLVM
58 lines
2.3 KiB
LLVM
; RUN: llc < %s | FileCheck %s
|
|
|
|
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
|
|
target triple = "x86_64-unknown-linux-gnu"
|
|
|
|
%structTy = type { i8, i32, i32 }
|
|
|
|
@e = common global %structTy zeroinitializer, align 4
|
|
|
|
;; Ensure that MergeConsecutiveStores doesn't incorrectly reorder
|
|
;; store operations. The first test stores in increasing address
|
|
;; order, the second in decreasing -- but in both cases should have
|
|
;; the same result in memory in the end.
|
|
|
|
; CHECK-LABEL: redundant_stores_merging:
|
|
; CHECK: movl $123, e+8(%rip)
|
|
; CHECK: movabsq $1958505086977, %rax
|
|
; CHECK: movq %rax, e+4(%rip)
|
|
define void @redundant_stores_merging() {
|
|
entry:
|
|
store i32 1, i32* getelementptr inbounds (%structTy, %structTy* @e, i64 0, i32 1), align 4
|
|
store i32 123, i32* getelementptr inbounds (%structTy, %structTy* @e, i64 0, i32 2), align 4
|
|
store i32 456, i32* getelementptr inbounds (%structTy, %structTy* @e, i64 0, i32 2), align 4
|
|
ret void
|
|
}
|
|
|
|
;; This variant tests PR25154.
|
|
; CHECK-LABEL: redundant_stores_merging_reverse:
|
|
; CHECK: movl $123, e+8(%rip)
|
|
; CHECK: movabsq $1958505086977, %rax
|
|
; CHECK: movq %rax, e+4(%rip)
|
|
define void @redundant_stores_merging_reverse() {
|
|
entry:
|
|
store i32 123, i32* getelementptr inbounds (%structTy, %structTy* @e, i64 0, i32 2), align 4
|
|
store i32 456, i32* getelementptr inbounds (%structTy, %structTy* @e, i64 0, i32 2), align 4
|
|
store i32 1, i32* getelementptr inbounds (%structTy, %structTy* @e, i64 0, i32 1), align 4
|
|
ret void
|
|
}
|
|
|
|
@b = common global [8 x i8] zeroinitializer, align 2
|
|
|
|
;; The 2-byte store to offset 3 overlaps the 2-byte store to offset 2;
|
|
;; these must not be reordered in MergeConsecutiveStores such that the
|
|
;; store to 3 comes first (e.g. by merging the stores to 0 and 2 into
|
|
;; a movl, after the store to 3).
|
|
|
|
;; CHECK-LABEL: overlapping_stores_merging:
|
|
;; CHECK: movw $0, b+2(%rip)
|
|
;; CHECK: movw $2, b+3(%rip)
|
|
;; CHECK: movw $1, b(%rip)
|
|
define void @overlapping_stores_merging() {
|
|
entry:
|
|
store i16 0, i16* bitcast (i8* getelementptr inbounds ([8 x i8], [8 x i8]* @b, i64 0, i64 2) to i16*), align 2
|
|
store i16 2, i16* bitcast (i8* getelementptr inbounds ([8 x i8], [8 x i8]* @b, i64 0, i64 3) to i16*), align 1
|
|
store i16 1, i16* bitcast (i8* getelementptr inbounds ([8 x i8], [8 x i8]* @b, i64 0, i64 0) to i16*), align 2
|
|
ret void
|
|
}
|