maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/Hexagon/swp-epilog-phi5.ll

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

185 lines
7.7 KiB
LLVM
Raw Normal View History

[ModuloSchedule] Peel out prologs and epilogs, generate actual code 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
2019-10-02 20:46:44 +08:00
; RUN: llc -march=hexagon < %s -pipeliner-experimental-cg=true | FileCheck %s
[Hexagon] Add more lit tests llvm-svn: 327271
2018-03-12 22:01:28 +08:00
; Test that we use the correct name in an epilog phi for a phi value
; that is defined for the last time in the kernel. Previously, we
; used the value from kernel loop definition, but we really need
; to use the value from the Phi in the kernel instead.
; In this test case, the second loop is pipelined, block b5.
[LSR] Attempt to increase the accuracy of LSR's setup cost In some loops, we end up generating loop induction variables that look like: {(-1 * (zext i16 (%i0 * %i1) to i32))<nsw>,+,1} As opposed to the simpler: {(zext i16 (%i0 * %i1) to i32),+,-1} i.e we count up from -limit to 0, not the simpler counting down from limit to 0. This is because the scores, as LSR calculates them, are the same and the second is filtered in place of the first. We end up with a redundant SUB from 0 in the code. This patch tries to make the calculation of the setup cost a little more thoroughly, recursing into the scev members to better approximate the setup required. The cost function for comparing LSR costs is: return std::tie(C1.NumRegs, C1.AddRecCost, C1.NumIVMuls, C1.NumBaseAdds, C1.ScaleCost, C1.ImmCost, C1.SetupCost) < std::tie(C2.NumRegs, C2.AddRecCost, C2.NumIVMuls, C2.NumBaseAdds, C2.ScaleCost, C2.ImmCost, C2.SetupCost); So this will only alter results if none of the other variables turn out to be different. Differential Revision: https://reviews.llvm.org/D58770 llvm-svn: 355597
2019-03-07 21:44:40 +08:00
; CHECK: loop1
[Hexagon] Add more lit tests llvm-svn: 327271
2018-03-12 22:01:28 +08:00
; CHECK: [[REG0:r([0-9]+)]] += mpyi
[PatternMatch] add special-case uaddo matching for increment-by-one (2nd try) This is the most important uaddo problem mentioned in PR31754: https://bugs.llvm.org/show_bug.cgi?id=31754 ...but that was overcome in x86 codegen with D57637. That patch also corrects the inc vs. add regressions seen with the previous attempt at this. Still, we want to make this matcher complete, so we can potentially canonicalize the pattern even if it's an 'add 1' operation. Pattern matching, however, shouldn't assume that we have canonicalized IR, so we match 4 commuted variants of uaddo. There's also a test with a crazy type to show that the existing CGP transform based on this matcher is not limited by target legality checks. I'm not sure if the Hexagon diff means the test is no longer testing what it intended to test, but that should be solvable in a follow-up. Differential Revision: https://reviews.llvm.org/D57516 llvm-svn: 352998
2019-02-04 00:16:48 +08:00
; CHECK: [[REG2:r([0-9]+)]] = add([[REG1:r([0-9]+)]],add([[REG0]],#8
[LSR] Attempt to increase the accuracy of LSR's setup cost In some loops, we end up generating loop induction variables that look like: {(-1 * (zext i16 (%i0 * %i1) to i32))<nsw>,+,1} As opposed to the simpler: {(zext i16 (%i0 * %i1) to i32),+,-1} i.e we count up from -limit to 0, not the simpler counting down from limit to 0. This is because the scores, as LSR calculates them, are the same and the second is filtered in place of the first. We end up with a redundant SUB from 0 in the code. This patch tries to make the calculation of the setup cost a little more thoroughly, recursing into the scev members to better approximate the setup required. The cost function for comparing LSR costs is: return std::tie(C1.NumRegs, C1.AddRecCost, C1.NumIVMuls, C1.NumBaseAdds, C1.ScaleCost, C1.ImmCost, C1.SetupCost) < std::tie(C2.NumRegs, C2.AddRecCost, C2.NumIVMuls, C2.NumBaseAdds, C2.ScaleCost, C2.ImmCost, C2.SetupCost); So this will only alter results if none of the other variables turn out to be different. Differential Revision: https://reviews.llvm.org/D58770 llvm-svn: 355597
2019-03-07 21:44:40 +08:00
; CHECK: endloop1
[Hexagon] Add more lit tests llvm-svn: 327271
2018-03-12 22:01:28 +08:00
%s.0 = type { %s.1*, %s.4*, %s.7*, i8*, i8, i32, %s.8*, i32, i32, i32, i8, i8, i32, i32, double, i8, i8, i8, i8, i8, i8, i8, i8, i32, i8, i8, i8, i32, i32, i32, i32, i32, i32, i8**, i32, i32, i32, i32, i32, [64 x i32]*, [4 x %s.9*], [4 x %s.10*], [4 x %s.10*], i32, %s.23*, i8, i8, [16 x i8], [16 x i8], [16 x i8], i32, i8, i8, i8, i8, i16, i16, i8, i8, i8, %s.11*, i32, i32, i32, i32, i8*, i32, [4 x %s.23*], i32, i32, i32, [10 x i32], i32, i32, i32, i32, i32, %s.12*, %s.13*, %s.14*, %s.15*, %s.16*, %s.17*, %s.18*, %s.19*, %s.20*, %s.21*, %s.22* }
%s.1 = type { void (%s.2*)*, void (%s.2*, i32)*, void (%s.2*)*, void (%s.2*, i8*)*, void (%s.2*)*, i32, %s.3, i32, i32, i8**, i32, i8**, i32, i32 }
%s.2 = type { %s.1*, %s.4*, %s.7*, i8*, i8, i32 }
%s.3 = type { [8 x i32], [48 x i8] }
%s.4 = type { i8* (%s.2*, i32, i32)*, i8* (%s.2*, i32, i32)*, i8** (%s.2*, i32, i32, i32)*, [64 x i16]** (%s.2*, i32, i32, i32)*, %s.5* (%s.2*, i32, i8, i32, i32, i32)*, %s.6* (%s.2*, i32, i8, i32, i32, i32)*, {}*, i8** (%s.2*, %s.5*, i32, i32, i8)*, [64 x i16]** (%s.2*, %s.6*, i32, i32, i8)*, void (%s.2*, i32)*, {}*, i32, i32 }
%s.5 = type opaque
%s.6 = type opaque
%s.7 = type { {}*, i32, i32, i32, i32 }
%s.8 = type { i8*, i32, {}*, i8 (%s.0*)*, void (%s.0*, i32)*, i8 (%s.0*, i32)*, {}* }
%s.9 = type { [64 x i16], i8 }
%s.10 = type { [17 x i8], [256 x i8], i8 }
%s.11 = type { %s.11*, i8, i32, i32, i8* }
%s.12 = type { {}*, {}*, i8 }
%s.13 = type { void (%s.0*, i8)*, void (%s.0*, i8**, i32*, i32)* }
%s.14 = type { {}*, i32 (%s.0*)*, {}*, i32 (%s.0*, i8***)*, %s.6** }
%s.15 = type { void (%s.0*, i8)*, void (%s.0*, i8***, i32*, i32, i8**, i32*, i32)* }
%s.16 = type { i32 (%s.0*)*, {}*, {}*, {}*, i8, i8 }
%s.17 = type { {}*, i32 (%s.0*)*, i8 (%s.0*)*, i8, i8, i32, i32 }
%s.18 = type { {}*, i8 (%s.0*, [64 x i16]**)*, i8 }
%s.19 = type { {}*, [5 x void (%s.0*, %s.23*, i16*, i8**, i32)*] }
%s.20 = type { {}*, void (%s.0*, i8***, i32*, i32, i8**, i32*, i32)*, i8 }
%s.21 = type { {}*, void (%s.0*, i8***, i32, i8**, i32)* }
%s.22 = type { void (%s.0*, i8)*, void (%s.0*, i8**, i8**, i32)*, {}*, {}* }
%s.23 = type { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i8, i32, i32, i32, i32, i32, i32, %s.9*, i8* }
; Function Attrs: nounwind optsize
define hidden void @f0(%s.0* nocapture readonly %a0, %s.23* nocapture readonly %a1, i8** nocapture readonly %a2, i8*** nocapture readonly %a3) #0 {
b0:
%v0 = load i8**, i8*** %a3, align 4
%v1 = getelementptr inbounds %s.0, %s.0* %a0, i32 0, i32 62
%v2 = load i32, i32* %v1, align 4
%v3 = icmp sgt i32 %v2, 0
br i1 %v3, label %b1, label %b10
b1: ; preds = %b0
%v4 = getelementptr inbounds %s.23, %s.23* %a1, i32 0, i32 10
br label %b2
b2: ; preds = %b8, %b1
%v5 = phi i32 [ 0, %b1 ], [ %v98, %b8 ]
%v6 = phi i32 [ 0, %b1 ], [ %v99, %b8 ]
%v7 = getelementptr inbounds i8*, i8** %a2, i32 %v6
br label %b3
b3: ; preds = %b7, %b2
%v8 = phi i32 [ 0, %b2 ], [ %v96, %b7 ]
%v9 = phi i32 [ %v5, %b2 ], [ %v16, %b7 ]
%v10 = load i8*, i8** %v7, align 4
%v11 = icmp eq i32 %v8, 0
%v12 = select i1 %v11, i32 -1, i32 1
%v13 = add i32 %v12, %v6
%v14 = getelementptr inbounds i8*, i8** %a2, i32 %v13
%v15 = load i8*, i8** %v14, align 4
%v16 = add nsw i32 %v9, 1
%v17 = getelementptr inbounds i8*, i8** %v0, i32 %v9
%v18 = load i8*, i8** %v17, align 4
%v19 = getelementptr inbounds i8, i8* %v10, i32 1
%v20 = load i8, i8* %v10, align 1
%v21 = zext i8 %v20 to i32
%v22 = mul nsw i32 %v21, 3
%v23 = getelementptr inbounds i8, i8* %v15, i32 1
%v24 = load i8, i8* %v15, align 1
%v25 = zext i8 %v24 to i32
%v26 = add nsw i32 %v22, %v25
%v27 = load i8, i8* %v19, align 1
%v28 = zext i8 %v27 to i32
%v29 = mul nsw i32 %v28, 3
%v30 = load i8, i8* %v23, align 1
%v31 = zext i8 %v30 to i32
%v32 = add nsw i32 %v29, %v31
%v33 = mul nsw i32 %v26, 4
%v34 = add nsw i32 %v33, 8
%v35 = lshr i32 %v34, 4
%v36 = trunc i32 %v35 to i8
%v37 = getelementptr inbounds i8, i8* %v18, i32 1
store i8 %v36, i8* %v18, align 1
%v38 = mul nsw i32 %v26, 3
%v39 = add i32 %v38, 7
%v40 = add i32 %v39, %v32
%v41 = lshr i32 %v40, 4
%v42 = trunc i32 %v41 to i8
store i8 %v42, i8* %v37, align 1
%v43 = load i32, i32* %v4, align 4
%v44 = add i32 %v43, -2
%v45 = getelementptr inbounds i8, i8* %v18, i32 2
%v46 = icmp eq i32 %v44, 0
br i1 %v46, label %b7, label %b4
b4: ; preds = %b3
%v47 = getelementptr inbounds i8, i8* %v15, i32 2
%v48 = getelementptr inbounds i8, i8* %v10, i32 2
%v49 = mul i32 %v43, 2
br label %b5
b5: ; preds = %b5, %b4
%v50 = phi i8* [ %v45, %b4 ], [ %v76, %b5 ]
%v51 = phi i32 [ %v44, %b4 ], [ %v75, %b5 ]
%v52 = phi i32 [ %v26, %b4 ], [ %v53, %b5 ]
%v53 = phi i32 [ %v32, %b4 ], [ %v64, %b5 ]
%v54 = phi i8* [ %v18, %b4 ], [ %v50, %b5 ]
%v55 = phi i8* [ %v47, %b4 ], [ %v61, %b5 ]
%v56 = phi i8* [ %v48, %b4 ], [ %v57, %b5 ]
%v57 = getelementptr inbounds i8, i8* %v56, i32 1
%v58 = load i8, i8* %v56, align 1
%v59 = zext i8 %v58 to i32
%v60 = mul nsw i32 %v59, 3
%v61 = getelementptr inbounds i8, i8* %v55, i32 1
%v62 = load i8, i8* %v55, align 1
%v63 = zext i8 %v62 to i32
%v64 = add nsw i32 %v60, %v63
%v65 = mul nsw i32 %v53, 3
%v66 = add i32 %v52, 8
%v67 = add i32 %v66, %v65
%v68 = lshr i32 %v67, 4
%v69 = trunc i32 %v68 to i8
%v70 = getelementptr inbounds i8, i8* %v54, i32 3
store i8 %v69, i8* %v50, align 1
%v71 = add i32 %v65, 7
%v72 = add i32 %v71, %v64
%v73 = lshr i32 %v72, 4
%v74 = trunc i32 %v73 to i8
store i8 %v74, i8* %v70, align 1
%v75 = add i32 %v51, -1
%v76 = getelementptr inbounds i8, i8* %v50, i32 2
%v77 = icmp eq i32 %v75, 0
br i1 %v77, label %b6, label %b5
b6: ; preds = %b5
%v78 = add i32 %v49, -2
%v79 = getelementptr i8, i8* %v18, i32 %v78
%v80 = add i32 %v49, -4
%v81 = getelementptr i8, i8* %v18, i32 %v80
br label %b7
b7: ; preds = %b6, %b3
%v82 = phi i8* [ %v79, %b6 ], [ %v45, %b3 ]
%v83 = phi i32 [ %v53, %b6 ], [ %v26, %b3 ]
%v84 = phi i32 [ %v64, %b6 ], [ %v32, %b3 ]
%v85 = phi i8* [ %v81, %b6 ], [ %v18, %b3 ]
%v86 = mul nsw i32 %v84, 3
%v87 = add i32 %v83, 8
%v88 = add i32 %v87, %v86
%v89 = lshr i32 %v88, 4
%v90 = trunc i32 %v89 to i8
%v91 = getelementptr inbounds i8, i8* %v85, i32 3
store i8 %v90, i8* %v82, align 1
%v92 = mul nsw i32 %v84, 4
%v93 = add nsw i32 %v92, 7
%v94 = lshr i32 %v93, 4
%v95 = trunc i32 %v94 to i8
store i8 %v95, i8* %v91, align 1
%v96 = add nsw i32 %v8, 1
%v97 = icmp eq i32 %v96, 2
br i1 %v97, label %b8, label %b3
b8: ; preds = %b7
%v98 = add i32 %v5, 2
%v99 = add nsw i32 %v6, 1
%v100 = load i32, i32* %v1, align 4
%v101 = icmp slt i32 %v98, %v100
br i1 %v101, label %b2, label %b9
b9: ; preds = %b8
br label %b10
b10: ; preds = %b9, %b0
ret void
}
attributes #0 = { nounwind optsize "target-cpu"="hexagonv60" }