Files
clang-p2996/llvm/test/Transforms/LoopStrengthReduce/X86/canonical-2.ll
Wei Mi 8848c1e3c7 [LSR] Call canonicalize after we generate a new Formula in GenerateTruncates. Fix PR33077.
The testcase in PR33077 generates a LSR Use Formula with two SCEVAddRecExprs for the same
loop. Such uncommon formula will become non-canonical after GenerateTruncates adds sign
extension to the ScaledReg of the Formula, and it will break the assertion that every
Formula to be inserted is canonical.

The fix is to call canonicalize for the raw Formula generated by GenerateTruncates
before inserting it.

llvm-svn: 303361
2017-05-18 17:21:22 +00:00

37 lines
1.3 KiB
LLVM

; REQUIRES: asserts
; RUN: opt -mtriple=x86_64-unknown-linux-gnu -loop-reduce -S < %s
; PR33077. Check the LSR Use formula to be inserted is already canonicalized and
; will not trigger assertion.
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
; Function Attrs: uwtable
define void @foo() {
cHeapLvb.exit:
br label %not_zero48.us
not_zero48.us: ; preds = %not_zero48.us, %cHeapLvb.exit
%indvars.iv.us = phi i64 [ %indvars.iv.next.us.7, %not_zero48.us ], [ undef, %cHeapLvb.exit ]
%0 = phi i32 [ %13, %not_zero48.us ], [ undef, %cHeapLvb.exit ]
%indvars.iv.next.us = add nuw nsw i64 %indvars.iv.us, 1
%1 = add i32 %0, 2
%2 = getelementptr inbounds i32, i32 addrspace(1)* undef, i64 %indvars.iv.next.us
%3 = load i32, i32 addrspace(1)* %2, align 4
%4 = add i32 %0, 3
%5 = load i32, i32 addrspace(1)* undef, align 4
%6 = sub i32 undef, %5
%factor.us.2 = shl i32 %6, 1
%7 = add i32 %factor.us.2, %1
%8 = load i32, i32 addrspace(1)* undef, align 4
%9 = sub i32 %7, %8
%factor.us.3 = shl i32 %9, 1
%10 = add i32 %factor.us.3, %4
%11 = load i32, i32 addrspace(1)* undef, align 4
%12 = sub i32 %10, %11
%factor.us.4 = shl i32 %12, 1
%13 = add i32 %0, 8
%indvars.iv.next.us.7 = add nsw i64 %indvars.iv.us, 8
br label %not_zero48.us
}