llvm-6502/test/Transforms/GVN/rle-phi-translate.ll

147 lines
3.2 KiB
LLVM
Raw Normal View History

; RUN: opt < %s -gvn -S | FileCheck %s
Implement initial support for PHI translation in memdep. This means that memdep keeps track of how PHIs affect the pointer in dep queries, which allows it to eliminate the load in cases like rle-phi-translate.ll, which basically end up being: BB1: X = load P br BB3 BB2: Y = load Q br BB3 BB3: R = phi [P] [Q] load R turning "load R" into a phi of X/Y. In addition to additional exposed opportunities, this makes memdep safe in many cases that it wasn't before (which is required for load PRE) and also makes it substantially more efficient. For example, consider: bb1: // has many predecessors. P = some_operator() load P In this example, previously memdep would scan all the predecessors of BB1 to see if they had something that would mustalias P. In some cases (e.g. test/Transforms/GVN/rle-must-alias.ll) it would actually find them and end up eliminating something. In many other cases though, it would scan and not find anything useful. MemDep now stops at a block if the pointer is defined in that block and cannot be phi translated to predecessors. This causes it to miss the (rare) cases like rle-must-alias.ll, but makes it faster by not scanning tons of stuff that is unlikely to be useful. For example, this speeds up GVN as a whole from 3.928s to 2.448s (60%)!. IMO, scalar GVN should be enhanced to simplify the rle-must-alias pointer base anyway, which would allow the loads to be eliminated. In the future, this should be enhanced to phi translate through geps and bitcasts as well (as indicated by FIXMEs) making memdep even more powerful. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61022 91177308-0d34-0410-b5e6-96231b3b80d8
2008-12-15 03:35:32 +00:00
target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
target triple = "i386-apple-darwin7"
define i32 @test1(i32* %b, i32* %c) nounwind {
; CHECK: @test1
Implement initial support for PHI translation in memdep. This means that memdep keeps track of how PHIs affect the pointer in dep queries, which allows it to eliminate the load in cases like rle-phi-translate.ll, which basically end up being: BB1: X = load P br BB3 BB2: Y = load Q br BB3 BB3: R = phi [P] [Q] load R turning "load R" into a phi of X/Y. In addition to additional exposed opportunities, this makes memdep safe in many cases that it wasn't before (which is required for load PRE) and also makes it substantially more efficient. For example, consider: bb1: // has many predecessors. P = some_operator() load P In this example, previously memdep would scan all the predecessors of BB1 to see if they had something that would mustalias P. In some cases (e.g. test/Transforms/GVN/rle-must-alias.ll) it would actually find them and end up eliminating something. In many other cases though, it would scan and not find anything useful. MemDep now stops at a block if the pointer is defined in that block and cannot be phi translated to predecessors. This causes it to miss the (rare) cases like rle-must-alias.ll, but makes it faster by not scanning tons of stuff that is unlikely to be useful. For example, this speeds up GVN as a whole from 3.928s to 2.448s (60%)!. IMO, scalar GVN should be enhanced to simplify the rle-must-alias pointer base anyway, which would allow the loads to be eliminated. In the future, this should be enhanced to phi translate through geps and bitcasts as well (as indicated by FIXMEs) making memdep even more powerful. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61022 91177308-0d34-0410-b5e6-96231b3b80d8
2008-12-15 03:35:32 +00:00
entry:
%g = alloca i32
%t1 = icmp eq i32* %b, null
Implement initial support for PHI translation in memdep. This means that memdep keeps track of how PHIs affect the pointer in dep queries, which allows it to eliminate the load in cases like rle-phi-translate.ll, which basically end up being: BB1: X = load P br BB3 BB2: Y = load Q br BB3 BB3: R = phi [P] [Q] load R turning "load R" into a phi of X/Y. In addition to additional exposed opportunities, this makes memdep safe in many cases that it wasn't before (which is required for load PRE) and also makes it substantially more efficient. For example, consider: bb1: // has many predecessors. P = some_operator() load P In this example, previously memdep would scan all the predecessors of BB1 to see if they had something that would mustalias P. In some cases (e.g. test/Transforms/GVN/rle-must-alias.ll) it would actually find them and end up eliminating something. In many other cases though, it would scan and not find anything useful. MemDep now stops at a block if the pointer is defined in that block and cannot be phi translated to predecessors. This causes it to miss the (rare) cases like rle-must-alias.ll, but makes it faster by not scanning tons of stuff that is unlikely to be useful. For example, this speeds up GVN as a whole from 3.928s to 2.448s (60%)!. IMO, scalar GVN should be enhanced to simplify the rle-must-alias pointer base anyway, which would allow the loads to be eliminated. In the future, this should be enhanced to phi translate through geps and bitcasts as well (as indicated by FIXMEs) making memdep even more powerful. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61022 91177308-0d34-0410-b5e6-96231b3b80d8
2008-12-15 03:35:32 +00:00
br i1 %t1, label %bb, label %bb1
bb:
%t2 = load i32* %c, align 4
%t3 = add i32 %t2, 1
Implement initial support for PHI translation in memdep. This means that memdep keeps track of how PHIs affect the pointer in dep queries, which allows it to eliminate the load in cases like rle-phi-translate.ll, which basically end up being: BB1: X = load P br BB3 BB2: Y = load Q br BB3 BB3: R = phi [P] [Q] load R turning "load R" into a phi of X/Y. In addition to additional exposed opportunities, this makes memdep safe in many cases that it wasn't before (which is required for load PRE) and also makes it substantially more efficient. For example, consider: bb1: // has many predecessors. P = some_operator() load P In this example, previously memdep would scan all the predecessors of BB1 to see if they had something that would mustalias P. In some cases (e.g. test/Transforms/GVN/rle-must-alias.ll) it would actually find them and end up eliminating something. In many other cases though, it would scan and not find anything useful. MemDep now stops at a block if the pointer is defined in that block and cannot be phi translated to predecessors. This causes it to miss the (rare) cases like rle-must-alias.ll, but makes it faster by not scanning tons of stuff that is unlikely to be useful. For example, this speeds up GVN as a whole from 3.928s to 2.448s (60%)!. IMO, scalar GVN should be enhanced to simplify the rle-must-alias pointer base anyway, which would allow the loads to be eliminated. In the future, this should be enhanced to phi translate through geps and bitcasts as well (as indicated by FIXMEs) making memdep even more powerful. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61022 91177308-0d34-0410-b5e6-96231b3b80d8
2008-12-15 03:35:32 +00:00
store i32 %t3, i32* %g, align 4
br label %bb2
bb1: ; preds = %entry
%t5 = load i32* %b, align 4
%t6 = add i32 %t5, 1
Implement initial support for PHI translation in memdep. This means that memdep keeps track of how PHIs affect the pointer in dep queries, which allows it to eliminate the load in cases like rle-phi-translate.ll, which basically end up being: BB1: X = load P br BB3 BB2: Y = load Q br BB3 BB3: R = phi [P] [Q] load R turning "load R" into a phi of X/Y. In addition to additional exposed opportunities, this makes memdep safe in many cases that it wasn't before (which is required for load PRE) and also makes it substantially more efficient. For example, consider: bb1: // has many predecessors. P = some_operator() load P In this example, previously memdep would scan all the predecessors of BB1 to see if they had something that would mustalias P. In some cases (e.g. test/Transforms/GVN/rle-must-alias.ll) it would actually find them and end up eliminating something. In many other cases though, it would scan and not find anything useful. MemDep now stops at a block if the pointer is defined in that block and cannot be phi translated to predecessors. This causes it to miss the (rare) cases like rle-must-alias.ll, but makes it faster by not scanning tons of stuff that is unlikely to be useful. For example, this speeds up GVN as a whole from 3.928s to 2.448s (60%)!. IMO, scalar GVN should be enhanced to simplify the rle-must-alias pointer base anyway, which would allow the loads to be eliminated. In the future, this should be enhanced to phi translate through geps and bitcasts as well (as indicated by FIXMEs) making memdep even more powerful. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61022 91177308-0d34-0410-b5e6-96231b3b80d8
2008-12-15 03:35:32 +00:00
store i32 %t6, i32* %g, align 4
br label %bb2
bb2: ; preds = %bb1, %bb
%c_addr.0 = phi i32* [ %g, %bb1 ], [ %c, %bb ]
%b_addr.0 = phi i32* [ %b, %bb1 ], [ %g, %bb ]
%cv = load i32* %c_addr.0, align 4
%bv = load i32* %b_addr.0, align 4
; CHECK: %bv = phi i32
; CHECK: %cv = phi i32
; CHECK-NOT: load
; CHECK: ret i32
%ret = add i32 %cv, %bv
Implement initial support for PHI translation in memdep. This means that memdep keeps track of how PHIs affect the pointer in dep queries, which allows it to eliminate the load in cases like rle-phi-translate.ll, which basically end up being: BB1: X = load P br BB3 BB2: Y = load Q br BB3 BB3: R = phi [P] [Q] load R turning "load R" into a phi of X/Y. In addition to additional exposed opportunities, this makes memdep safe in many cases that it wasn't before (which is required for load PRE) and also makes it substantially more efficient. For example, consider: bb1: // has many predecessors. P = some_operator() load P In this example, previously memdep would scan all the predecessors of BB1 to see if they had something that would mustalias P. In some cases (e.g. test/Transforms/GVN/rle-must-alias.ll) it would actually find them and end up eliminating something. In many other cases though, it would scan and not find anything useful. MemDep now stops at a block if the pointer is defined in that block and cannot be phi translated to predecessors. This causes it to miss the (rare) cases like rle-must-alias.ll, but makes it faster by not scanning tons of stuff that is unlikely to be useful. For example, this speeds up GVN as a whole from 3.928s to 2.448s (60%)!. IMO, scalar GVN should be enhanced to simplify the rle-must-alias pointer base anyway, which would allow the loads to be eliminated. In the future, this should be enhanced to phi translate through geps and bitcasts as well (as indicated by FIXMEs) making memdep even more powerful. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61022 91177308-0d34-0410-b5e6-96231b3b80d8
2008-12-15 03:35:32 +00:00
ret i32 %ret
}
define i8 @test2(i1 %cond, i32* %b, i32* %c) nounwind {
; CHECK: @test2
entry:
br i1 %cond, label %bb, label %bb1
bb:
%b1 = bitcast i32* %b to i8*
store i8 4, i8* %b1
br label %bb2
bb1:
%c1 = bitcast i32* %c to i8*
store i8 92, i8* %c1
br label %bb2
bb2:
%d = phi i32* [ %c, %bb1 ], [ %b, %bb ]
%d1 = bitcast i32* %d to i8*
%dv = load i8* %d1
; CHECK: %dv = phi i8 [ 92, %bb1 ], [ 4, %bb ]
; CHECK-NOT: load
; CHECK: ret i8 %dv
ret i8 %dv
}
define i32 @test3(i1 %cond, i32* %b, i32* %c) nounwind {
; CHECK: @test3
entry:
br i1 %cond, label %bb, label %bb1
bb:
%b1 = getelementptr i32* %b, i32 17
store i32 4, i32* %b1
br label %bb2
bb1:
%c1 = getelementptr i32* %c, i32 7
store i32 82, i32* %c1
br label %bb2
bb2:
%d = phi i32* [ %c, %bb1 ], [ %b, %bb ]
%i = phi i32 [ 7, %bb1 ], [ 17, %bb ]
%d1 = getelementptr i32* %d, i32 %i
%dv = load i32* %d1
; CHECK: %dv = phi i32 [ 82, %bb1 ], [ 4, %bb ]
; CHECK-NOT: load
; CHECK: ret i32 %dv
ret i32 %dv
}
; PR5313
define i32 @test4(i1 %cond, i32* %b, i32* %c) nounwind {
; CHECK: @test4
entry:
br i1 %cond, label %bb, label %bb1
bb:
store i32 4, i32* %b
br label %bb2
bb1:
%c1 = getelementptr i32* %c, i32 7
store i32 82, i32* %c1
br label %bb2
bb2:
%d = phi i32* [ %c, %bb1 ], [ %b, %bb ]
%i = phi i32 [ 7, %bb1 ], [ 0, %bb ]
%d1 = getelementptr i32* %d, i32 %i
%dv = load i32* %d1
; CHECK: %dv = phi i32 [ 82, %bb1 ], [ 4, %bb ]
; CHECK-NOT: load
; CHECK: ret i32 %dv
ret i32 %dv
}
; void test5(int N, double* G) {
; for (long j = 1; j < 1000; j++)
; G[j] = G[j] + G[j-1];
; }
;
; Should compile into one load in the loop.
define void @test5(i32 %N, double* nocapture %G) nounwind ssp {
; CHECK: @test5
bb.nph:
br label %for.body
for.body:
%indvar = phi i64 [ 0, %bb.nph ], [ %tmp, %for.body ]
%arrayidx6 = getelementptr double* %G, i64 %indvar
%tmp = add i64 %indvar, 1
%arrayidx = getelementptr double* %G, i64 %tmp
%tmp3 = load double* %arrayidx
%tmp7 = load double* %arrayidx6
%add = fadd double %tmp3, %tmp7
store double %add, double* %arrayidx
%exitcond = icmp eq i64 %tmp, 999
br i1 %exitcond, label %for.end, label %for.body
; CHECK: for.body:
; CHECK: phi double
; CHECK: load double
; CHECK-NOT: load double
; CHECK: br i1
for.end:
ret void
}