mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-23 11:13:28 +01:00
25cc728d30
Summary: This extends the PeelingModuloScheduleExpander to generate prolog and epilog code, and correctly stitch uses through the prolog, kernel, epilog DAG. The key concept in this patch is to ensure that all transforms are *local*; only a function of a block and its immediate predecessor and successor. By defining the problem in this way we can inductively rewrite the entire DAG using only local knowledge that is easy to reason about. For example, we assume that all prologs and epilogs are near-perfect clones of the steady-state kernel. This means that if a block has an instruction that is predicated out, we can redirect all users of that instruction to that equivalent instruction in our immediate predecessor. As all blocks are clones, every instruction must have an equivalent in every other block. Similarly we can make the assumption by construction that if a value defined in a block is used outside that block, the only possible user is its immediate successors. We maintain this even for values that are used outside the loop by creating a limited form of LCSSA. This code isn't small, but it isn't complex. Enabled a bunch of testing from Hexagon. There are a couple of tests not enabled yet; I'm about 80% sure there isn't buggy codegen but the tests are checking for patterns that we don't produce. Those still need a bit more investigation. In the meantime we (Google) are happy with the code produced by this on our downstream SMS implementation, and believe it generates correct code. Subscribers: mgorny, hiraditya, jsji, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D68205 llvm-svn: 373462
44 lines
1.6 KiB
LLVM
44 lines
1.6 KiB
LLVM
; RUN: llc -march=hexagon -enable-pipeliner -pipeliner-max-stages=1 < %s -pipeliner-experimental-cg=true | FileCheck %s
|
|
|
|
; Test that we update the offset correctly for loads that are
|
|
; moved past stores. In these cases, we change the dependences
|
|
; to make it easier to move the instructions, and we have to update
|
|
; the register/offsets correctly after the schedule is finalized.
|
|
|
|
@g0 = common global [400 x i32] zeroinitializer, align 8
|
|
@g1 = common global [400 x i32] zeroinitializer, align 8
|
|
|
|
; Function Attrs: nounwind
|
|
define void @f0() #0 {
|
|
b0:
|
|
br label %b2
|
|
|
|
b1: ; preds = %b2
|
|
ret void
|
|
|
|
; CHECK: loop0(.LBB0_[[LOOP:.]],
|
|
; CHECK: .LBB0_[[LOOP]]:
|
|
; CHECK: = memd([[REG1:(r[0-9]+)]]+#8)
|
|
; CHECK: memd([[REG1]]++#8) =
|
|
; CHECK: }{{[ \t]*}}:endloop
|
|
|
|
b2: ; preds = %b2, %b0
|
|
%v0 = phi i32* [ getelementptr inbounds ([400 x i32], [400 x i32]* @g0, i32 0, i32 0), %b0 ], [ %v11, %b2 ]
|
|
%v1 = phi i32* [ getelementptr inbounds ([400 x i32], [400 x i32]* @g1, i32 0, i32 0), %b0 ], [ %v12, %b2 ]
|
|
%v2 = phi i32 [ 0, %b0 ], [ %v9, %b2 ]
|
|
%v3 = bitcast i32* %v0 to <2 x i32>*
|
|
%v4 = load <2 x i32>, <2 x i32>* %v3, align 8
|
|
%v5 = mul <2 x i32> %v4, <i32 7, i32 7>
|
|
%v6 = bitcast i32* %v1 to <2 x i32>*
|
|
%v7 = load <2 x i32>, <2 x i32>* %v6, align 8
|
|
%v8 = add <2 x i32> %v7, %v5
|
|
store <2 x i32> %v8, <2 x i32>* %v6, align 8
|
|
%v9 = add nsw i32 %v2, 2
|
|
%v10 = icmp slt i32 %v2, 398
|
|
%v11 = getelementptr i32, i32* %v0, i32 2
|
|
%v12 = getelementptr i32, i32* %v1, i32 2
|
|
br i1 %v10, label %b2, label %b1
|
|
}
|
|
|
|
attributes #0 = { nounwind "target-cpu"="hexagonv55" }
|