llvm/test/Transforms/LoopReroll/nonconst_lb.ll
Chandler Carruth eca46e623a [LPM] Factor all of the loop analysis usage updates into a common helper
routine.

We were getting this wrong in small ways and generally being very
inconsistent about it across loop passes. Instead, let's have a common
place where we do this. One minor downside is that this will require
some analyses like SCEV in more places than they are strictly needed.
However, this seems benign as these analyses are complete no-ops, and
without this consistency we can in many cases end up with the legacy
pass manager scheduling deciding to split up a loop pass pipeline in
order to run the function analysis half-way through. It is very, very
annoying to fix these without just being very pedantic across the board.

The only loop passes I've not updated here are ones that use
AU.setPreservesAll() such as IVUsers (an analysis) and the pass printer.
They seemed less relevant.

With this patch, almost all of the problems in PR24804 around loop pass
pipelines are fixed. The one remaining issue is that we run simplify-cfg
and instcombine in the middle of the loop pass pipeline. We've recently
added some loop variants of these passes that would seem substantially
cleaner to use, but this at least gets us much closer to the previous
state. Notably, the seven loop pass managers is down to three.

I've not updated the loop passes using LoopAccessAnalysis because that
analysis hasn't been fully wired into LoopSimplify/LCSSA, and it isn't
clear that those transforms want to support those forms anyways. They
all run late anyways, so this is harmless. Similarly, LSR is left alone
because it already carefully manages its forms and doesn't need to get
fused into a single loop pass manager with a bunch of other loop passes.

LoopReroll didn't use loop simplified form previously, and I've updated
the test case to match the trivially different output.

Finally, I've also factored all the pass initialization for the passes
that use this technique as well, so that should be done regularly and
reliably.

Thanks to James for the help reviewing and thinking about this stuff,
and Ben for help thinking about it as well!

Differential Revision: http://reviews.llvm.org/D17435

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@261316 91177308-0d34-0410-b5e6-96231b3b80d8
2016-02-19 10:45:18 +00:00

153 lines
6.1 KiB
LLVM

; RUN: opt < %s -loop-reroll -S | FileCheck %s
target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:32:64-v128:32:128-a0:0:32-n32-S32"
target triple = "thumbv7-none-linux"
;void foo(int *A, int *B, int m, int n) {
; for (int i = m; i < n; i+=4) {
; A[i+0] = B[i+0] * 4;
; A[i+1] = B[i+1] * 4;
; A[i+2] = B[i+2] * 4;
; A[i+3] = B[i+3] * 4;
; }
;}
define void @foo(i32* nocapture %A, i32* nocapture readonly %B, i32 %m, i32 %n) {
entry:
%cmp34 = icmp slt i32 %m, %n
br i1 %cmp34, label %for.body, label %for.end
for.body: ; preds = %entry, %for.body
%i.035 = phi i32 [ %add18, %for.body ], [ %m, %entry ]
%arrayidx = getelementptr inbounds i32, i32* %B, i32 %i.035
%0 = load i32, i32* %arrayidx, align 4
%mul = shl nsw i32 %0, 2
%arrayidx2 = getelementptr inbounds i32, i32* %A, i32 %i.035
store i32 %mul, i32* %arrayidx2, align 4
%add3 = add nsw i32 %i.035, 1
%arrayidx4 = getelementptr inbounds i32, i32* %B, i32 %add3
%1 = load i32, i32* %arrayidx4, align 4
%mul5 = shl nsw i32 %1, 2
%arrayidx7 = getelementptr inbounds i32, i32* %A, i32 %add3
store i32 %mul5, i32* %arrayidx7, align 4
%add8 = add nsw i32 %i.035, 2
%arrayidx9 = getelementptr inbounds i32, i32* %B, i32 %add8
%2 = load i32, i32* %arrayidx9, align 4
%mul10 = shl nsw i32 %2, 2
%arrayidx12 = getelementptr inbounds i32, i32* %A, i32 %add8
store i32 %mul10, i32* %arrayidx12, align 4
%add13 = add nsw i32 %i.035, 3
%arrayidx14 = getelementptr inbounds i32, i32* %B, i32 %add13
%3 = load i32, i32* %arrayidx14, align 4
%mul15 = shl nsw i32 %3, 2
%arrayidx17 = getelementptr inbounds i32, i32* %A, i32 %add13
store i32 %mul15, i32* %arrayidx17, align 4
%add18 = add nsw i32 %i.035, 4
%cmp = icmp slt i32 %add18, %n
br i1 %cmp, label %for.body, label %for.end
for.end: ; preds = %for.body, %entry
ret void
}
; CHECK-LABEL: @foo
; CHECK: for.body.preheader: ; preds = %entry
; CHECK: %0 = add i32 %n, -1
; CHECK: %1 = sub i32 %0, %m
; CHECK: %2 = lshr i32 %1, 2
; CHECK: %3 = shl i32 %2, 2
; CHECK: %4 = add i32 %m, %3
; CHECK: %5 = add i32 %4, 3
; CHECK: br label %for.body
; CHECK: for.body: ; preds = %for.body, %for.body.preheader
; CHECK: %indvar = phi i32 [ 0, %for.body.preheader ], [ %indvar.next, %for.body ]
; CHECK: %6 = add i32 %m, %indvar
; CHECK: %arrayidx = getelementptr inbounds i32, i32* %B, i32 %6
; CHECK: %7 = load i32, i32* %arrayidx, align 4
; CHECK: %mul = shl nsw i32 %7, 2
; CHECK: %arrayidx2 = getelementptr inbounds i32, i32* %A, i32 %6
; CHECK: store i32 %mul, i32* %arrayidx2, align 4
; CHECK: %indvar.next = add i32 %indvar, 1
; CHECK: %exitcond = icmp eq i32 %6, %5
; CHECK: br i1 %exitcond, label %for.end.loopexit, label %for.body
;void daxpy_ur(int n,float da,float *dx,float *dy)
; {
; int m = n % 4;
; for (int i = m; i < n; i = i + 4)
; {
; dy[i] = dy[i] + da*dx[i];
; dy[i+1] = dy[i+1] + da*dx[i+1];
; dy[i+2] = dy[i+2] + da*dx[i+2];
; dy[i+3] = dy[i+3] + da*dx[i+3];
; }
; }
define void @daxpy_ur(i32 %n, float %da, float* nocapture readonly %dx, float* nocapture %dy) {
entry:
%rem = srem i32 %n, 4
%cmp55 = icmp slt i32 %rem, %n
br i1 %cmp55, label %for.body, label %for.end
for.body: ; preds = %entry, %for.body
%i.056 = phi i32 [ %add27, %for.body ], [ %rem, %entry ]
%arrayidx = getelementptr inbounds float, float* %dy, i32 %i.056
%0 = load float, float* %arrayidx, align 4
%arrayidx1 = getelementptr inbounds float, float* %dx, i32 %i.056
%1 = load float, float* %arrayidx1, align 4
%mul = fmul float %1, %da
%add = fadd float %0, %mul
store float %add, float* %arrayidx, align 4
%add3 = add nsw i32 %i.056, 1
%arrayidx4 = getelementptr inbounds float, float* %dy, i32 %add3
%2 = load float, float* %arrayidx4, align 4
%arrayidx6 = getelementptr inbounds float, float* %dx, i32 %add3
%3 = load float, float* %arrayidx6, align 4
%mul7 = fmul float %3, %da
%add8 = fadd float %2, %mul7
store float %add8, float* %arrayidx4, align 4
%add11 = add nsw i32 %i.056, 2
%arrayidx12 = getelementptr inbounds float, float* %dy, i32 %add11
%4 = load float, float* %arrayidx12, align 4
%arrayidx14 = getelementptr inbounds float, float* %dx, i32 %add11
%5 = load float, float* %arrayidx14, align 4
%mul15 = fmul float %5, %da
%add16 = fadd float %4, %mul15
store float %add16, float* %arrayidx12, align 4
%add19 = add nsw i32 %i.056, 3
%arrayidx20 = getelementptr inbounds float, float* %dy, i32 %add19
%6 = load float, float* %arrayidx20, align 4
%arrayidx22 = getelementptr inbounds float, float* %dx, i32 %add19
%7 = load float, float* %arrayidx22, align 4
%mul23 = fmul float %7, %da
%add24 = fadd float %6, %mul23
store float %add24, float* %arrayidx20, align 4
%add27 = add nsw i32 %i.056, 4
%cmp = icmp slt i32 %add27, %n
br i1 %cmp, label %for.body, label %for.end
for.end: ; preds = %for.body, %entry
ret void
}
; CHECK-LABEL: @daxpy_ur
; CHECK: for.body.preheader:
; CHECK: %0 = add i32 %n, -1
; CHECK: %1 = sub i32 %0, %rem
; CHECK: %2 = lshr i32 %1, 2
; CHECK: %3 = shl i32 %2, 2
; CHECK: %4 = add i32 %rem, %3
; CHECK: %5 = add i32 %4, 3
; CHECK: br label %for.body
; CHECK: for.body:
; CHECK: %indvar = phi i32 [ 0, %for.body.preheader ], [ %indvar.next, %for.body ]
; CHECK: %6 = add i32 %rem, %indvar
; CHECK: %arrayidx = getelementptr inbounds float, float* %dy, i32 %6
; CHECK: %7 = load float, float* %arrayidx, align 4
; CHECK: %arrayidx1 = getelementptr inbounds float, float* %dx, i32 %6
; CHECK: %8 = load float, float* %arrayidx1, align 4
; CHECK: %mul = fmul float %8, %da
; CHECK: %add = fadd float %7, %mul
; CHECK: store float %add, float* %arrayidx, align 4
; CHECK: %indvar.next = add i32 %indvar, 1
; CHECK: %exitcond = icmp eq i32 %6, %5
; CHECK: br i1 %exitcond, label %for.end.loopexit, label %for.body