mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-07-23 16:30:00 +00:00
5d59315f13
Summary: Because LSR happens at a late stage where mul of a power of 2 is typically canonicalized to shl, this canonicalization emits code that can be better CSE'ed. Test Plan: Transforms/LoopStrengthReduce/shl.ll shows how this change makes GVN more powerful. Fixes some existing tests due to this change. Reviewers: sanjoy, majnemer, atrick Reviewed By: majnemer, atrick Subscribers: majnemer, llvm-commits Differential Revision: http://reviews.llvm.org/D10448 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@240573 91177308-0d34-0410-b5e6-96231b3b80d8
58 lines
2.0 KiB
LLVM
58 lines
2.0 KiB
LLVM
; RUN: opt -loop-reduce -S < %s | FileCheck %s
|
|
;
|
|
; Test LSR's intelligence regarding phi reuse.
|
|
; Verify that scaled GEPs are not reused. rdar://5064068
|
|
|
|
target triple = "x86-apple-darwin"
|
|
|
|
; Provide legal integer types.
|
|
target datalayout = "n8:16:32:64"
|
|
|
|
|
|
; CHECK-LABEL: @test(
|
|
; multiplies are hoisted out of the loop
|
|
; CHECK: while.body.lr.ph:
|
|
; CHECK: shl i64
|
|
; CHECK: shl i64
|
|
; GEPs are ugly
|
|
; CHECK: while.body:
|
|
; CHECK: phi
|
|
; CHECK: phi
|
|
; CHECK: phi
|
|
; CHECK: phi
|
|
; CHECK-NOT: phi
|
|
; CHECK: bitcast float* {{.*}} to i1*
|
|
; CHECK: bitcast float* {{.*}} to i1*
|
|
; CHECK: getelementptr i1, i1*
|
|
; CHECK: getelementptr i1, i1*
|
|
|
|
define float @test(float* nocapture %A, float* nocapture %B, i32 %N, i32 %IA, i32 %IB) nounwind uwtable readonly ssp {
|
|
entry:
|
|
%cmp1 = icmp sgt i32 %N, 0
|
|
br i1 %cmp1, label %while.body.lr.ph, label %while.end
|
|
|
|
while.body.lr.ph: ; preds = %entry
|
|
%idx.ext = sext i32 %IA to i64
|
|
%idx.ext2 = sext i32 %IB to i64
|
|
br label %while.body
|
|
|
|
while.body: ; preds = %while.body.lr.ph, %while.body
|
|
%A.addr.05 = phi float* [ %A, %while.body.lr.ph ], [ %add.ptr, %while.body ]
|
|
%B.addr.04 = phi float* [ %B, %while.body.lr.ph ], [ %add.ptr3, %while.body ]
|
|
%N.addr.03 = phi i32 [ %N, %while.body.lr.ph ], [ %sub, %while.body ]
|
|
%Sum0.02 = phi float [ 0.000000e+00, %while.body.lr.ph ], [ %add, %while.body ]
|
|
%0 = load float, float* %A.addr.05, align 4
|
|
%1 = load float, float* %B.addr.04, align 4
|
|
%mul = fmul float %0, %1
|
|
%add = fadd float %Sum0.02, %mul
|
|
%add.ptr = getelementptr inbounds float, float* %A.addr.05, i64 %idx.ext
|
|
%add.ptr3 = getelementptr inbounds float, float* %B.addr.04, i64 %idx.ext2
|
|
%sub = add nsw i32 %N.addr.03, -1
|
|
%cmp = icmp sgt i32 %sub, 0
|
|
br i1 %cmp, label %while.body, label %while.end
|
|
|
|
while.end: ; preds = %while.body, %entry
|
|
%Sum0.0.lcssa = phi float [ 0.000000e+00, %entry ], [ %add, %while.body ]
|
|
ret float %Sum0.0.lcssa
|
|
}
|