de2fed6152
The unrolling code was previously inserting new cloned blocks at the end of the function. The result of this with typical loop structures is that the new iterations are placed far from the initial iteration. With unrolling, the general assumption is that the a) the loop is reasonable hot, and b) the first Count-1 copies of the loop are rarely (if ever) loop exiting. As such, placing Count-1 copies out of line is a fairly poor code placement choice. We'd much rather fall through into the hot (non-exiting) path. For code with branch profiles, later layout would fix this, but this may have a positive impact on non-PGO compiled code. However, the real motivation for this change isn't performance. Its readability and human understanding. Having to jump around long distances in an IR file to trace an unrolled loop structure is error prone and tedious.
258 lines
10 KiB
LLVM
258 lines
10 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
|
|
; RUN: opt -loop-unroll -S %s -verify-loop-info -verify-dom-info -verify-loop-lcssa | FileCheck %s
|
|
|
|
%struct.spam = type { double, double, double, double, double, double, double }
|
|
|
|
define void @test2(i32* %arg, i64* %out) {
|
|
; CHECK-LABEL: @test2(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: br label [[FOR_HEADER:%.*]]
|
|
; CHECK: for.header:
|
|
; CHECK-NEXT: store i32 0, i32* [[ARG:%.*]], align 4
|
|
; CHECK-NEXT: br label [[FOR_LATCH:%.*]]
|
|
; CHECK: for.latch:
|
|
; CHECK-NEXT: store volatile i64 0, i64* [[OUT:%.*]], align 4
|
|
; CHECK-NEXT: [[PTR_1:%.*]] = getelementptr inbounds i32, i32* [[ARG]], i64 1
|
|
; CHECK-NEXT: store i32 0, i32* [[PTR_1]], align 4
|
|
; CHECK-NEXT: br label [[FOR_LATCH_1:%.*]]
|
|
; CHECK: for.latch.1:
|
|
; CHECK-NEXT: store volatile i64 1, i64* [[OUT]], align 4
|
|
; CHECK-NEXT: [[PTR_2:%.*]] = getelementptr inbounds i32, i32* [[ARG]], i64 2
|
|
; CHECK-NEXT: store i32 0, i32* [[PTR_2]], align 4
|
|
; CHECK-NEXT: br label [[FOR_LATCH_2:%.*]]
|
|
; CHECK: for.latch.2:
|
|
; CHECK-NEXT: store volatile i64 2, i64* [[OUT]], align 4
|
|
; CHECK-NEXT: [[PTR_3:%.*]] = getelementptr inbounds i32, i32* [[ARG]], i64 3
|
|
; CHECK-NEXT: store i32 0, i32* [[PTR_3]], align 4
|
|
; CHECK-NEXT: br i1 true, label [[IF_END_LOOPEXIT:%.*]], label [[FOR_LATCH_3:%.*]]
|
|
; CHECK: for.latch.3:
|
|
; CHECK-NEXT: store volatile i64 3, i64* [[OUT]], align 4
|
|
; CHECK-NEXT: unreachable
|
|
; CHECK: if.end.loopexit:
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
|
|
entry:
|
|
br label %for.header
|
|
|
|
for.header: ; preds = %for.latch, %entry
|
|
%indvars.iv800 = phi i64 [ 0, %entry ], [ %indvars.iv.next801, %for.latch ]
|
|
%ptr = getelementptr inbounds i32, i32* %arg, i64 %indvars.iv800
|
|
store i32 0, i32* %ptr, align 4
|
|
%indvars.iv.next801 = add nuw nsw i64 %indvars.iv800, 1
|
|
%exitcond802 = icmp eq i64 %indvars.iv.next801, 4
|
|
br i1 %exitcond802, label %if.end.loopexit, label %for.latch
|
|
|
|
for.latch: ; preds = %for.header
|
|
store volatile i64 %indvars.iv800, i64* %out
|
|
br label %for.header
|
|
|
|
if.end.loopexit: ; preds = %for.header
|
|
ret void
|
|
}
|
|
|
|
define double @test_with_lcssa(double %arg1, double* %arg2) {
|
|
; CHECK-LABEL: @test_with_lcssa(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: br label [[LOOP_HEADER:%.*]]
|
|
; CHECK: loop.header:
|
|
; CHECK-NEXT: [[RES:%.*]] = fsub double [[ARG1:%.*]], 3.000000e+00
|
|
; CHECK-NEXT: br label [[LOOP_LATCH:%.*]]
|
|
; CHECK: loop.latch:
|
|
; CHECK-NEXT: [[PTR:%.*]] = getelementptr inbounds double, double* [[ARG2:%.*]], i64 1
|
|
; CHECK-NEXT: [[LV:%.*]] = load double, double* [[PTR]], align 8
|
|
; CHECK-NEXT: [[RES_1:%.*]] = fsub double [[LV]], [[RES]]
|
|
; CHECK-NEXT: br i1 true, label [[LOOP_EXIT:%.*]], label [[LOOP_LATCH_1:%.*]]
|
|
; CHECK: loop.latch.1:
|
|
; CHECK-NEXT: unreachable
|
|
; CHECK: loop.exit:
|
|
; CHECK-NEXT: [[RES_LCSSA:%.*]] = phi double [ [[RES_1]], [[LOOP_LATCH]] ]
|
|
; CHECK-NEXT: ret double [[RES_LCSSA]]
|
|
;
|
|
|
|
entry:
|
|
br label %loop.header
|
|
|
|
loop.header: ; preds = %entry, %loop.latch
|
|
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
|
|
%d1 = phi double [ %arg1, %entry ], [ %lv, %loop.latch ]
|
|
%d2 = phi double [ 3.0, %entry ], [ %res, %loop.latch ]
|
|
%res = fsub double %d1, %d2
|
|
%iv.next = add nuw nsw i64 %iv, 1
|
|
%cond = icmp eq i64 %iv.next, 2
|
|
br i1 %cond, label %loop.exit, label %loop.latch
|
|
|
|
loop.latch: ; preds = %bb366
|
|
%ptr = getelementptr inbounds double, double* %arg2, i64 %iv.next
|
|
%lv = load double, double* %ptr, align 8
|
|
br label %loop.header
|
|
|
|
loop.exit: ; preds = %bb366
|
|
%res.lcssa = phi double [ %res, %loop.header ]
|
|
ret double %res.lcssa
|
|
}
|
|
|
|
; We unroll the outer loop and need to preserve LI for the inner loop.
|
|
define void @test_with_nested_loop(i32* %arg) {
|
|
; CHECK-LABEL: @test_with_nested_loop(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: br label [[OUTER_HEADER:%.*]]
|
|
; CHECK: outer.header:
|
|
; CHECK-NEXT: br label [[INNER_BODY_PREHEADER:%.*]]
|
|
; CHECK: inner.body.preheader:
|
|
; CHECK-NEXT: br label [[INNER_BODY:%.*]]
|
|
; CHECK: inner.body:
|
|
; CHECK-NEXT: [[J_IV:%.*]] = phi i64 [ [[J_IV_NEXT:%.*]], [[INNER_BODY]] ], [ 0, [[INNER_BODY_PREHEADER]] ]
|
|
; CHECK-NEXT: [[PTR:%.*]] = getelementptr inbounds i32, i32* [[ARG:%.*]], i64 [[J_IV]]
|
|
; CHECK-NEXT: store i32 0, i32* [[PTR]], align 4
|
|
; CHECK-NEXT: [[J_IV_NEXT]] = add nuw nsw i64 [[J_IV]], 1
|
|
; CHECK-NEXT: [[INNER_COND:%.*]] = icmp eq i64 [[J_IV_NEXT]], 40000
|
|
; CHECK-NEXT: br i1 [[INNER_COND]], label [[OUTER_LATCH:%.*]], label [[INNER_BODY]]
|
|
; CHECK: outer.latch:
|
|
; CHECK-NEXT: br label [[INNER_BODY_PREHEADER_1:%.*]]
|
|
; CHECK: inner.body.preheader.1:
|
|
; CHECK-NEXT: br label [[INNER_BODY_1:%.*]]
|
|
; CHECK: inner.body.1:
|
|
; CHECK-NEXT: [[J_IV_1:%.*]] = phi i64 [ [[J_IV_NEXT_1:%.*]], [[INNER_BODY_1]] ], [ 0, [[INNER_BODY_PREHEADER_1]] ]
|
|
; CHECK-NEXT: [[IDX_1:%.*]] = add i64 1, [[J_IV_1]]
|
|
; CHECK-NEXT: [[PTR_1:%.*]] = getelementptr inbounds i32, i32* [[ARG]], i64 [[IDX_1]]
|
|
; CHECK-NEXT: store i32 0, i32* [[PTR_1]], align 4
|
|
; CHECK-NEXT: [[J_IV_NEXT_1]] = add nuw nsw i64 [[J_IV_1]], 1
|
|
; CHECK-NEXT: [[INNER_COND_1:%.*]] = icmp eq i64 [[J_IV_NEXT_1]], 40000
|
|
; CHECK-NEXT: br i1 [[INNER_COND_1]], label [[OUTER_LATCH_1:%.*]], label [[INNER_BODY_1]]
|
|
; CHECK: outer.latch.1:
|
|
; CHECK-NEXT: br i1 true, label [[EXIT:%.*]], label [[INNER_BODY_PREHEADER_2:%.*]]
|
|
; CHECK: inner.body.preheader.2:
|
|
; CHECK-NEXT: br label [[INNER_BODY_2:%.*]]
|
|
; CHECK: inner.body.2:
|
|
; CHECK-NEXT: [[J_IV_2:%.*]] = phi i64 [ [[J_IV_NEXT_2:%.*]], [[INNER_BODY_2]] ], [ 0, [[INNER_BODY_PREHEADER_2]] ]
|
|
; CHECK-NEXT: [[IDX_2:%.*]] = add i64 2, [[J_IV_2]]
|
|
; CHECK-NEXT: [[PTR_2:%.*]] = getelementptr inbounds i32, i32* [[ARG]], i64 [[IDX_2]]
|
|
; CHECK-NEXT: store i32 0, i32* [[PTR_2]], align 4
|
|
; CHECK-NEXT: [[J_IV_NEXT_2]] = add nuw nsw i64 [[J_IV_2]], 1
|
|
; CHECK-NEXT: [[INNER_COND_2:%.*]] = icmp eq i64 [[J_IV_NEXT_2]], 40000
|
|
; CHECK-NEXT: br i1 [[INNER_COND_2]], label [[OUTER_LATCH_2:%.*]], label [[INNER_BODY_2]]
|
|
; CHECK: outer.latch.2:
|
|
; CHECK-NEXT: unreachable
|
|
; CHECK: exit:
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
|
|
entry:
|
|
br label %outer.header
|
|
|
|
outer.header: ; preds = %outer.latch, %entry
|
|
%outer.iv = phi i64 [ 0, %entry ], [ %outer.iv.next, %outer.latch ]
|
|
%outer.iv.next = add nuw nsw i64 %outer.iv, 1
|
|
%outer.cond = icmp eq i64 %outer.iv, 2
|
|
br i1 %outer.cond, label %exit, label %inner.body
|
|
|
|
inner.body:
|
|
%j.iv = phi i64 [ 0, %outer.header ], [ %j.iv.next, %inner.body ]
|
|
%idx = add i64 %outer.iv, %j.iv
|
|
%ptr = getelementptr inbounds i32, i32* %arg, i64 %idx
|
|
store i32 0, i32* %ptr, align 4
|
|
%j.iv.next = add nuw nsw i64 %j.iv, 1
|
|
%inner.cond = icmp eq i64 %j.iv.next, 40000
|
|
br i1 %inner.cond, label %outer.latch, label %inner.body
|
|
|
|
outer.latch: ; preds = %inner.body
|
|
br label %outer.header
|
|
|
|
exit: ; preds = %outer.header
|
|
ret void
|
|
}
|
|
|
|
; We unroll the inner loop and need to preserve LI for the outer loop.
|
|
define void @test_with_nested_loop_unroll_inner(i32* %arg) {
|
|
; CHECK-LABEL: @test_with_nested_loop_unroll_inner(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: br label [[OUTER_HEADER:%.*]]
|
|
; CHECK: outer.header:
|
|
; CHECK-NEXT: [[OUTER_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[OUTER_IV_NEXT:%.*]], [[INNER_BODY:%.*]] ]
|
|
; CHECK-NEXT: [[OUTER_IV_NEXT]] = add nuw nsw i64 [[OUTER_IV]], 1
|
|
; CHECK-NEXT: [[OUTER_COND:%.*]] = icmp eq i64 [[OUTER_IV]], 40000
|
|
; CHECK-NEXT: br i1 [[OUTER_COND]], label [[EXIT:%.*]], label [[INNER_BODY_PREHEADER:%.*]]
|
|
; CHECK: inner.body.preheader:
|
|
; CHECK-NEXT: br label [[INNER_BODY]]
|
|
; CHECK: inner.body:
|
|
; CHECK-NEXT: [[PTR:%.*]] = getelementptr inbounds i32, i32* [[ARG:%.*]], i64 [[OUTER_IV]]
|
|
; CHECK-NEXT: store i32 0, i32* [[PTR]], align 4
|
|
; CHECK-NEXT: [[IDX_1:%.*]] = add i64 [[OUTER_IV]], 1
|
|
; CHECK-NEXT: [[PTR_1:%.*]] = getelementptr inbounds i32, i32* [[ARG]], i64 [[IDX_1]]
|
|
; CHECK-NEXT: store i32 0, i32* [[PTR_1]], align 4
|
|
; CHECK-NEXT: br label [[OUTER_HEADER]]
|
|
; CHECK: exit:
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
|
|
entry:
|
|
br label %outer.header
|
|
|
|
outer.header: ; preds = %outer.latch, %entry
|
|
%outer.iv = phi i64 [ 0, %entry ], [ %outer.iv.next, %outer.latch ]
|
|
%outer.iv.next = add nuw nsw i64 %outer.iv, 1
|
|
%outer.cond = icmp eq i64 %outer.iv, 40000
|
|
br i1 %outer.cond, label %exit, label %inner.body
|
|
|
|
inner.body:
|
|
%j.iv = phi i64 [ 0, %outer.header ], [ %j.iv.next, %inner.body ]
|
|
%idx = add i64 %outer.iv, %j.iv
|
|
%ptr = getelementptr inbounds i32, i32* %arg, i64 %idx
|
|
store i32 0, i32* %ptr, align 4
|
|
%j.iv.next = add nuw nsw i64 %j.iv, 1
|
|
%inner.cond = icmp eq i64 %j.iv.next, 2
|
|
br i1 %inner.cond, label %outer.latch, label %inner.body
|
|
|
|
outer.latch: ; preds = %inner.body
|
|
br label %outer.header
|
|
|
|
exit: ; preds = %outer.header
|
|
ret void
|
|
}
|
|
|
|
|
|
|
|
; Check that we do not crash for headers with non-branch instructions, e.g.
|
|
; switch. We do not unroll in those cases.
|
|
define void @test_switchinst_in_header() {
|
|
; CHECK-LABEL: @test_switchinst_in_header(
|
|
; CHECK-NEXT: entry:
|
|
; CHECK-NEXT: br label [[WHILE_HEADER:%.*]]
|
|
; CHECK: while.header:
|
|
; CHECK-NEXT: switch i32 undef, label [[EXIT:%.*]] [
|
|
; CHECK-NEXT: i32 11, label [[WHILE_BODY1:%.*]]
|
|
; CHECK-NEXT: i32 5, label [[WHILE_BODY2:%.*]]
|
|
; CHECK-NEXT: ]
|
|
; CHECK: while.body1:
|
|
; CHECK-NEXT: unreachable
|
|
; CHECK: while.body2:
|
|
; CHECK-NEXT: br label [[WHILE_LATCH:%.*]]
|
|
; CHECK: while.latch:
|
|
; CHECK-NEXT: br label [[WHILE_HEADER]]
|
|
; CHECK: exit:
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
|
|
entry:
|
|
br label %while.header
|
|
|
|
while.header: ; preds = %while.latch, %entry
|
|
switch i32 undef, label %exit [
|
|
i32 11, label %while.body1
|
|
i32 5, label %while.body2
|
|
]
|
|
|
|
while.body1: ; preds = %while.header
|
|
unreachable
|
|
|
|
while.body2: ; preds = %while.header
|
|
br label %while.latch
|
|
|
|
while.latch: ; preds = %while.body2
|
|
br label %while.header
|
|
|
|
exit: ; preds = %while.header
|
|
ret void
|
|
}
|