llvm-project/llvm/test/CodeGen/X86/win-alloca-expander.ll

157 lines
3.5 KiB
LLVM
Raw Normal View History

; RUN: llc < %s -mtriple=i686-pc-win32 | FileCheck %s
; RUN: llc < %s -mtriple=i686-pc-win32 -O0
%struct.S = type { [1024 x i8] }
%struct.T = type { [3000 x i8] }
%struct.U = type { [10000 x i8] }
define void @basics() {
; CHECK-LABEL: basics:
entry:
br label %bb1
; Allocation move sizes should have been removed.
; CHECK-NOT: movl $1024
; CHECK-NOT: movl $3000
bb1:
%p0 = alloca %struct.S
; The allocation is small enough not to require stack probing, but the %esp
; offset after the prologue is not known, so the stack must be touched before
; the pointer is adjusted.
; CHECK: pushl %eax
; CHECK: subl $1020, %esp
%saved_stack = tail call i8* @llvm.stacksave()
%p1 = alloca %struct.S
; We know the %esp offset from above, so there is no need to touch the stack
; before adjusting it.
; CHECK: subl $1024, %esp
%p2 = alloca %struct.T
; The offset is now 2048 bytes, so allocating a T must touch the stack again.
; CHECK: pushl %eax
; CHECK: subl $2996, %esp
call void @f(%struct.S* %p0)
; CHECK: calll
%p3 = alloca %struct.T
; The call above touched the stack, so there is room for a T object.
; CHECK: subl $3000, %esp
%p4 = alloca %struct.U
; The U object is large enough to require stack probing.
; CHECK: movl $10000, %eax
; CHECK: calll __chkstk
%p5 = alloca %struct.T
; The stack probing above touched the tip of the stack, so there's room for a T.
; CHECK: subl $3000, %esp
call void @llvm.stackrestore(i8* %saved_stack)
%p6 = alloca %struct.S
; The stack restore means we lose track of the stack pointer and must probe.
; CHECK: pushl %eax
; CHECK: subl $1020, %esp
; Use the pointers so they're not optimized away.
call void @f(%struct.S* %p1)
call void @g(%struct.T* %p2)
call void @g(%struct.T* %p3)
call void @h(%struct.U* %p4)
call void @g(%struct.T* %p5)
ret void
}
define void @loop() {
; CHECK-LABEL: loop:
entry:
br label %bb1
bb1:
%p1 = alloca %struct.S
; The entry offset is unknown; touch-and-sub.
; CHECK: pushl %eax
; CHECK: subl $1020, %esp
br label %loop1
loop1:
%i1 = phi i32 [ 10, %bb1 ], [ %dec1, %loop1 ]
%p2 = alloca %struct.S
; We know the incoming offset from bb1, but from the back-edge, we assume the
; worst, and therefore touch-and-sub to allocate.
; CHECK: pushl %eax
; CHECK: subl $1020, %esp
%dec1 = sub i32 %i1, 1
%cmp1 = icmp sgt i32 %i1, 0
br i1 %cmp1, label %loop1, label %end
; CHECK: decl
; CHECK: jg
end:
call void @f(%struct.S* %p1)
call void @f(%struct.S* %p2)
ret void
}
define void @probe_size_attribute() "stack-probe-size"="512" {
; CHECK-LABEL: probe_size_attribute:
entry:
br label %bb1
bb1:
%p0 = alloca %struct.S
; The allocation would be small enough not to require probing, if it wasn't
; for the stack-probe-size attribute.
; CHECK: movl $1024, %eax
; CHECK: calll __chkstk
call void @f(%struct.S* %p0)
ret void
}
define void @cfg(i1 %x, i1 %y) {
; Test that the blocks are analyzed in the correct order.
; CHECK-LABEL: cfg:
entry:
Codegen: Make chains from trellis-shaped CFGs Lay out trellis-shaped CFGs optimally. A trellis of the shape below: A B |\ /| | \ / | | X | | / \ | |/ \| C D would be laid out A; B->C ; D by the current layout algorithm. Now we identify trellises and lay them out either A->C; B->D or A->D; B->C. This scales with an increasing number of predecessors. A trellis is a a group of 2 or more predecessor blocks that all have the same successors. because of this we can tail duplicate to extend existing trellises. As an example consider the following CFG: B D F H / \ / \ / \ / \ A---C---E---G---Ret Where A,C,E,G are all small (Currently 2 instructions). The CFG preserving layout is then A,B,C,D,E,F,G,H,Ret. The current code will copy C into B, E into D and G into F and yield the layout A,C,B(C),E,D(E),F(G),G,H,ret define void @straight_test(i32 %tag) { entry: br label %test1 test1: ; A %tagbit1 = and i32 %tag, 1 %tagbit1eq0 = icmp eq i32 %tagbit1, 0 br i1 %tagbit1eq0, label %test2, label %optional1 optional1: ; B call void @a() br label %test2 test2: ; C %tagbit2 = and i32 %tag, 2 %tagbit2eq0 = icmp eq i32 %tagbit2, 0 br i1 %tagbit2eq0, label %test3, label %optional2 optional2: ; D call void @b() br label %test3 test3: ; E %tagbit3 = and i32 %tag, 4 %tagbit3eq0 = icmp eq i32 %tagbit3, 0 br i1 %tagbit3eq0, label %test4, label %optional3 optional3: ; F call void @c() br label %test4 test4: ; G %tagbit4 = and i32 %tag, 8 %tagbit4eq0 = icmp eq i32 %tagbit4, 0 br i1 %tagbit4eq0, label %exit, label %optional4 optional4: ; H call void @d() br label %exit exit: ret void } here is the layout after D27742: straight_test: # @straight_test ; ... Prologue elided ; BB#0: # %entry ; A (merged with test1) ; ... More prologue elided mr 30, 3 andi. 3, 30, 1 bc 12, 1, .LBB0_2 ; BB#1: # %test2 ; C rlwinm. 3, 30, 0, 30, 30 beq 0, .LBB0_3 b .LBB0_4 .LBB0_2: # %optional1 ; B (copy of C) bl a nop rlwinm. 3, 30, 0, 30, 30 bne 0, .LBB0_4 .LBB0_3: # %test3 ; E rlwinm. 3, 30, 0, 29, 29 beq 0, .LBB0_5 b .LBB0_6 .LBB0_4: # %optional2 ; D (copy of E) bl b nop rlwinm. 3, 30, 0, 29, 29 bne 0, .LBB0_6 .LBB0_5: # %test4 ; G rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 b .LBB0_7 .LBB0_6: # %optional3 ; F (copy of G) bl c nop rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 .LBB0_7: # %optional4 ; H bl d nop .LBB0_8: # %exit ; Ret ld 30, 96(1) # 8-byte Folded Reload addi 1, 1, 112 ld 0, 16(1) mtlr 0 blr The tail-duplication has produced some benefit, but it has also produced a trellis which is not laid out optimally. With this patch, we improve the layouts of such trellises, and decrease the cost calculation for tail-duplication accordingly. This patch produces the layout A,C,E,G,B,D,F,H,Ret. This layout does have back edges, which is a negative, but it has a bigger compensating positive, which is that it handles the case where there are long strings of skipped blocks much better than the original layout. Both layouts handle runs of executed blocks equally well. Branch prediction also improves if there is any correlation between subsequent optional blocks. Here is the resulting concrete layout: straight_test: # @straight_test ; BB#0: # %entry ; A (merged with test1) mr 30, 3 andi. 3, 30, 1 bc 12, 1, .LBB0_4 ; BB#1: # %test2 ; C rlwinm. 3, 30, 0, 30, 30 bne 0, .LBB0_5 .LBB0_2: # %test3 ; E rlwinm. 3, 30, 0, 29, 29 bne 0, .LBB0_6 .LBB0_3: # %test4 ; G rlwinm. 3, 30, 0, 28, 28 bne 0, .LBB0_7 b .LBB0_8 .LBB0_4: # %optional1 ; B (Copy of C) bl a nop rlwinm. 3, 30, 0, 30, 30 beq 0, .LBB0_2 .LBB0_5: # %optional2 ; D (Copy of E) bl b nop rlwinm. 3, 30, 0, 29, 29 beq 0, .LBB0_3 .LBB0_6: # %optional3 ; F (Copy of G) bl c nop rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 .LBB0_7: # %optional4 ; H bl d nop .LBB0_8: # %exit Differential Revision: https://reviews.llvm.org/D28522 llvm-svn: 295223
2017-02-16 03:49:14 +08:00
br i1 %x, label %bb1, label %bb3
bb1:
%p1 = alloca %struct.S
; CHECK: pushl %eax
; CHECK: subl $1020, %esp
Codegen: Make chains from trellis-shaped CFGs Lay out trellis-shaped CFGs optimally. A trellis of the shape below: A B |\ /| | \ / | | X | | / \ | |/ \| C D would be laid out A; B->C ; D by the current layout algorithm. Now we identify trellises and lay them out either A->C; B->D or A->D; B->C. This scales with an increasing number of predecessors. A trellis is a a group of 2 or more predecessor blocks that all have the same successors. because of this we can tail duplicate to extend existing trellises. As an example consider the following CFG: B D F H / \ / \ / \ / \ A---C---E---G---Ret Where A,C,E,G are all small (Currently 2 instructions). The CFG preserving layout is then A,B,C,D,E,F,G,H,Ret. The current code will copy C into B, E into D and G into F and yield the layout A,C,B(C),E,D(E),F(G),G,H,ret define void @straight_test(i32 %tag) { entry: br label %test1 test1: ; A %tagbit1 = and i32 %tag, 1 %tagbit1eq0 = icmp eq i32 %tagbit1, 0 br i1 %tagbit1eq0, label %test2, label %optional1 optional1: ; B call void @a() br label %test2 test2: ; C %tagbit2 = and i32 %tag, 2 %tagbit2eq0 = icmp eq i32 %tagbit2, 0 br i1 %tagbit2eq0, label %test3, label %optional2 optional2: ; D call void @b() br label %test3 test3: ; E %tagbit3 = and i32 %tag, 4 %tagbit3eq0 = icmp eq i32 %tagbit3, 0 br i1 %tagbit3eq0, label %test4, label %optional3 optional3: ; F call void @c() br label %test4 test4: ; G %tagbit4 = and i32 %tag, 8 %tagbit4eq0 = icmp eq i32 %tagbit4, 0 br i1 %tagbit4eq0, label %exit, label %optional4 optional4: ; H call void @d() br label %exit exit: ret void } here is the layout after D27742: straight_test: # @straight_test ; ... Prologue elided ; BB#0: # %entry ; A (merged with test1) ; ... More prologue elided mr 30, 3 andi. 3, 30, 1 bc 12, 1, .LBB0_2 ; BB#1: # %test2 ; C rlwinm. 3, 30, 0, 30, 30 beq 0, .LBB0_3 b .LBB0_4 .LBB0_2: # %optional1 ; B (copy of C) bl a nop rlwinm. 3, 30, 0, 30, 30 bne 0, .LBB0_4 .LBB0_3: # %test3 ; E rlwinm. 3, 30, 0, 29, 29 beq 0, .LBB0_5 b .LBB0_6 .LBB0_4: # %optional2 ; D (copy of E) bl b nop rlwinm. 3, 30, 0, 29, 29 bne 0, .LBB0_6 .LBB0_5: # %test4 ; G rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 b .LBB0_7 .LBB0_6: # %optional3 ; F (copy of G) bl c nop rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 .LBB0_7: # %optional4 ; H bl d nop .LBB0_8: # %exit ; Ret ld 30, 96(1) # 8-byte Folded Reload addi 1, 1, 112 ld 0, 16(1) mtlr 0 blr The tail-duplication has produced some benefit, but it has also produced a trellis which is not laid out optimally. With this patch, we improve the layouts of such trellises, and decrease the cost calculation for tail-duplication accordingly. This patch produces the layout A,C,E,G,B,D,F,H,Ret. This layout does have back edges, which is a negative, but it has a bigger compensating positive, which is that it handles the case where there are long strings of skipped blocks much better than the original layout. Both layouts handle runs of executed blocks equally well. Branch prediction also improves if there is any correlation between subsequent optional blocks. Here is the resulting concrete layout: straight_test: # @straight_test ; BB#0: # %entry ; A (merged with test1) mr 30, 3 andi. 3, 30, 1 bc 12, 1, .LBB0_4 ; BB#1: # %test2 ; C rlwinm. 3, 30, 0, 30, 30 bne 0, .LBB0_5 .LBB0_2: # %test3 ; E rlwinm. 3, 30, 0, 29, 29 bne 0, .LBB0_6 .LBB0_3: # %test4 ; G rlwinm. 3, 30, 0, 28, 28 bne 0, .LBB0_7 b .LBB0_8 .LBB0_4: # %optional1 ; B (Copy of C) bl a nop rlwinm. 3, 30, 0, 30, 30 beq 0, .LBB0_2 .LBB0_5: # %optional2 ; D (Copy of E) bl b nop rlwinm. 3, 30, 0, 29, 29 beq 0, .LBB0_3 .LBB0_6: # %optional3 ; F (Copy of G) bl c nop rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 .LBB0_7: # %optional4 ; H bl d nop .LBB0_8: # %exit Differential Revision: https://reviews.llvm.org/D28522 llvm-svn: 295223
2017-02-16 03:49:14 +08:00
br label %bb4
bb2:
Codegen: Make chains from trellis-shaped CFGs Lay out trellis-shaped CFGs optimally. A trellis of the shape below: A B |\ /| | \ / | | X | | / \ | |/ \| C D would be laid out A; B->C ; D by the current layout algorithm. Now we identify trellises and lay them out either A->C; B->D or A->D; B->C. This scales with an increasing number of predecessors. A trellis is a a group of 2 or more predecessor blocks that all have the same successors. because of this we can tail duplicate to extend existing trellises. As an example consider the following CFG: B D F H / \ / \ / \ / \ A---C---E---G---Ret Where A,C,E,G are all small (Currently 2 instructions). The CFG preserving layout is then A,B,C,D,E,F,G,H,Ret. The current code will copy C into B, E into D and G into F and yield the layout A,C,B(C),E,D(E),F(G),G,H,ret define void @straight_test(i32 %tag) { entry: br label %test1 test1: ; A %tagbit1 = and i32 %tag, 1 %tagbit1eq0 = icmp eq i32 %tagbit1, 0 br i1 %tagbit1eq0, label %test2, label %optional1 optional1: ; B call void @a() br label %test2 test2: ; C %tagbit2 = and i32 %tag, 2 %tagbit2eq0 = icmp eq i32 %tagbit2, 0 br i1 %tagbit2eq0, label %test3, label %optional2 optional2: ; D call void @b() br label %test3 test3: ; E %tagbit3 = and i32 %tag, 4 %tagbit3eq0 = icmp eq i32 %tagbit3, 0 br i1 %tagbit3eq0, label %test4, label %optional3 optional3: ; F call void @c() br label %test4 test4: ; G %tagbit4 = and i32 %tag, 8 %tagbit4eq0 = icmp eq i32 %tagbit4, 0 br i1 %tagbit4eq0, label %exit, label %optional4 optional4: ; H call void @d() br label %exit exit: ret void } here is the layout after D27742: straight_test: # @straight_test ; ... Prologue elided ; BB#0: # %entry ; A (merged with test1) ; ... More prologue elided mr 30, 3 andi. 3, 30, 1 bc 12, 1, .LBB0_2 ; BB#1: # %test2 ; C rlwinm. 3, 30, 0, 30, 30 beq 0, .LBB0_3 b .LBB0_4 .LBB0_2: # %optional1 ; B (copy of C) bl a nop rlwinm. 3, 30, 0, 30, 30 bne 0, .LBB0_4 .LBB0_3: # %test3 ; E rlwinm. 3, 30, 0, 29, 29 beq 0, .LBB0_5 b .LBB0_6 .LBB0_4: # %optional2 ; D (copy of E) bl b nop rlwinm. 3, 30, 0, 29, 29 bne 0, .LBB0_6 .LBB0_5: # %test4 ; G rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 b .LBB0_7 .LBB0_6: # %optional3 ; F (copy of G) bl c nop rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 .LBB0_7: # %optional4 ; H bl d nop .LBB0_8: # %exit ; Ret ld 30, 96(1) # 8-byte Folded Reload addi 1, 1, 112 ld 0, 16(1) mtlr 0 blr The tail-duplication has produced some benefit, but it has also produced a trellis which is not laid out optimally. With this patch, we improve the layouts of such trellises, and decrease the cost calculation for tail-duplication accordingly. This patch produces the layout A,C,E,G,B,D,F,H,Ret. This layout does have back edges, which is a negative, but it has a bigger compensating positive, which is that it handles the case where there are long strings of skipped blocks much better than the original layout. Both layouts handle runs of executed blocks equally well. Branch prediction also improves if there is any correlation between subsequent optional blocks. Here is the resulting concrete layout: straight_test: # @straight_test ; BB#0: # %entry ; A (merged with test1) mr 30, 3 andi. 3, 30, 1 bc 12, 1, .LBB0_4 ; BB#1: # %test2 ; C rlwinm. 3, 30, 0, 30, 30 bne 0, .LBB0_5 .LBB0_2: # %test3 ; E rlwinm. 3, 30, 0, 29, 29 bne 0, .LBB0_6 .LBB0_3: # %test4 ; G rlwinm. 3, 30, 0, 28, 28 bne 0, .LBB0_7 b .LBB0_8 .LBB0_4: # %optional1 ; B (Copy of C) bl a nop rlwinm. 3, 30, 0, 30, 30 beq 0, .LBB0_2 .LBB0_5: # %optional2 ; D (Copy of E) bl b nop rlwinm. 3, 30, 0, 29, 29 beq 0, .LBB0_3 .LBB0_6: # %optional3 ; F (Copy of G) bl c nop rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 .LBB0_7: # %optional4 ; H bl d nop .LBB0_8: # %exit Differential Revision: https://reviews.llvm.org/D28522 llvm-svn: 295223
2017-02-16 03:49:14 +08:00
%p5 = alloca %struct.T
; CHECK: pushl %eax
; CHECK: subl $2996, %esp
Codegen: Make chains from trellis-shaped CFGs Lay out trellis-shaped CFGs optimally. A trellis of the shape below: A B |\ /| | \ / | | X | | / \ | |/ \| C D would be laid out A; B->C ; D by the current layout algorithm. Now we identify trellises and lay them out either A->C; B->D or A->D; B->C. This scales with an increasing number of predecessors. A trellis is a a group of 2 or more predecessor blocks that all have the same successors. because of this we can tail duplicate to extend existing trellises. As an example consider the following CFG: B D F H / \ / \ / \ / \ A---C---E---G---Ret Where A,C,E,G are all small (Currently 2 instructions). The CFG preserving layout is then A,B,C,D,E,F,G,H,Ret. The current code will copy C into B, E into D and G into F and yield the layout A,C,B(C),E,D(E),F(G),G,H,ret define void @straight_test(i32 %tag) { entry: br label %test1 test1: ; A %tagbit1 = and i32 %tag, 1 %tagbit1eq0 = icmp eq i32 %tagbit1, 0 br i1 %tagbit1eq0, label %test2, label %optional1 optional1: ; B call void @a() br label %test2 test2: ; C %tagbit2 = and i32 %tag, 2 %tagbit2eq0 = icmp eq i32 %tagbit2, 0 br i1 %tagbit2eq0, label %test3, label %optional2 optional2: ; D call void @b() br label %test3 test3: ; E %tagbit3 = and i32 %tag, 4 %tagbit3eq0 = icmp eq i32 %tagbit3, 0 br i1 %tagbit3eq0, label %test4, label %optional3 optional3: ; F call void @c() br label %test4 test4: ; G %tagbit4 = and i32 %tag, 8 %tagbit4eq0 = icmp eq i32 %tagbit4, 0 br i1 %tagbit4eq0, label %exit, label %optional4 optional4: ; H call void @d() br label %exit exit: ret void } here is the layout after D27742: straight_test: # @straight_test ; ... Prologue elided ; BB#0: # %entry ; A (merged with test1) ; ... More prologue elided mr 30, 3 andi. 3, 30, 1 bc 12, 1, .LBB0_2 ; BB#1: # %test2 ; C rlwinm. 3, 30, 0, 30, 30 beq 0, .LBB0_3 b .LBB0_4 .LBB0_2: # %optional1 ; B (copy of C) bl a nop rlwinm. 3, 30, 0, 30, 30 bne 0, .LBB0_4 .LBB0_3: # %test3 ; E rlwinm. 3, 30, 0, 29, 29 beq 0, .LBB0_5 b .LBB0_6 .LBB0_4: # %optional2 ; D (copy of E) bl b nop rlwinm. 3, 30, 0, 29, 29 bne 0, .LBB0_6 .LBB0_5: # %test4 ; G rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 b .LBB0_7 .LBB0_6: # %optional3 ; F (copy of G) bl c nop rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 .LBB0_7: # %optional4 ; H bl d nop .LBB0_8: # %exit ; Ret ld 30, 96(1) # 8-byte Folded Reload addi 1, 1, 112 ld 0, 16(1) mtlr 0 blr The tail-duplication has produced some benefit, but it has also produced a trellis which is not laid out optimally. With this patch, we improve the layouts of such trellises, and decrease the cost calculation for tail-duplication accordingly. This patch produces the layout A,C,E,G,B,D,F,H,Ret. This layout does have back edges, which is a negative, but it has a bigger compensating positive, which is that it handles the case where there are long strings of skipped blocks much better than the original layout. Both layouts handle runs of executed blocks equally well. Branch prediction also improves if there is any correlation between subsequent optional blocks. Here is the resulting concrete layout: straight_test: # @straight_test ; BB#0: # %entry ; A (merged with test1) mr 30, 3 andi. 3, 30, 1 bc 12, 1, .LBB0_4 ; BB#1: # %test2 ; C rlwinm. 3, 30, 0, 30, 30 bne 0, .LBB0_5 .LBB0_2: # %test3 ; E rlwinm. 3, 30, 0, 29, 29 bne 0, .LBB0_6 .LBB0_3: # %test4 ; G rlwinm. 3, 30, 0, 28, 28 bne 0, .LBB0_7 b .LBB0_8 .LBB0_4: # %optional1 ; B (Copy of C) bl a nop rlwinm. 3, 30, 0, 30, 30 beq 0, .LBB0_2 .LBB0_5: # %optional2 ; D (Copy of E) bl b nop rlwinm. 3, 30, 0, 29, 29 beq 0, .LBB0_3 .LBB0_6: # %optional3 ; F (Copy of G) bl c nop rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 .LBB0_7: # %optional4 ; H bl d nop .LBB0_8: # %exit Differential Revision: https://reviews.llvm.org/D28522 llvm-svn: 295223
2017-02-16 03:49:14 +08:00
call void @g(%struct.T* %p5)
ret void
bb3:
Codegen: Make chains from trellis-shaped CFGs Lay out trellis-shaped CFGs optimally. A trellis of the shape below: A B |\ /| | \ / | | X | | / \ | |/ \| C D would be laid out A; B->C ; D by the current layout algorithm. Now we identify trellises and lay them out either A->C; B->D or A->D; B->C. This scales with an increasing number of predecessors. A trellis is a a group of 2 or more predecessor blocks that all have the same successors. because of this we can tail duplicate to extend existing trellises. As an example consider the following CFG: B D F H / \ / \ / \ / \ A---C---E---G---Ret Where A,C,E,G are all small (Currently 2 instructions). The CFG preserving layout is then A,B,C,D,E,F,G,H,Ret. The current code will copy C into B, E into D and G into F and yield the layout A,C,B(C),E,D(E),F(G),G,H,ret define void @straight_test(i32 %tag) { entry: br label %test1 test1: ; A %tagbit1 = and i32 %tag, 1 %tagbit1eq0 = icmp eq i32 %tagbit1, 0 br i1 %tagbit1eq0, label %test2, label %optional1 optional1: ; B call void @a() br label %test2 test2: ; C %tagbit2 = and i32 %tag, 2 %tagbit2eq0 = icmp eq i32 %tagbit2, 0 br i1 %tagbit2eq0, label %test3, label %optional2 optional2: ; D call void @b() br label %test3 test3: ; E %tagbit3 = and i32 %tag, 4 %tagbit3eq0 = icmp eq i32 %tagbit3, 0 br i1 %tagbit3eq0, label %test4, label %optional3 optional3: ; F call void @c() br label %test4 test4: ; G %tagbit4 = and i32 %tag, 8 %tagbit4eq0 = icmp eq i32 %tagbit4, 0 br i1 %tagbit4eq0, label %exit, label %optional4 optional4: ; H call void @d() br label %exit exit: ret void } here is the layout after D27742: straight_test: # @straight_test ; ... Prologue elided ; BB#0: # %entry ; A (merged with test1) ; ... More prologue elided mr 30, 3 andi. 3, 30, 1 bc 12, 1, .LBB0_2 ; BB#1: # %test2 ; C rlwinm. 3, 30, 0, 30, 30 beq 0, .LBB0_3 b .LBB0_4 .LBB0_2: # %optional1 ; B (copy of C) bl a nop rlwinm. 3, 30, 0, 30, 30 bne 0, .LBB0_4 .LBB0_3: # %test3 ; E rlwinm. 3, 30, 0, 29, 29 beq 0, .LBB0_5 b .LBB0_6 .LBB0_4: # %optional2 ; D (copy of E) bl b nop rlwinm. 3, 30, 0, 29, 29 bne 0, .LBB0_6 .LBB0_5: # %test4 ; G rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 b .LBB0_7 .LBB0_6: # %optional3 ; F (copy of G) bl c nop rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 .LBB0_7: # %optional4 ; H bl d nop .LBB0_8: # %exit ; Ret ld 30, 96(1) # 8-byte Folded Reload addi 1, 1, 112 ld 0, 16(1) mtlr 0 blr The tail-duplication has produced some benefit, but it has also produced a trellis which is not laid out optimally. With this patch, we improve the layouts of such trellises, and decrease the cost calculation for tail-duplication accordingly. This patch produces the layout A,C,E,G,B,D,F,H,Ret. This layout does have back edges, which is a negative, but it has a bigger compensating positive, which is that it handles the case where there are long strings of skipped blocks much better than the original layout. Both layouts handle runs of executed blocks equally well. Branch prediction also improves if there is any correlation between subsequent optional blocks. Here is the resulting concrete layout: straight_test: # @straight_test ; BB#0: # %entry ; A (merged with test1) mr 30, 3 andi. 3, 30, 1 bc 12, 1, .LBB0_4 ; BB#1: # %test2 ; C rlwinm. 3, 30, 0, 30, 30 bne 0, .LBB0_5 .LBB0_2: # %test3 ; E rlwinm. 3, 30, 0, 29, 29 bne 0, .LBB0_6 .LBB0_3: # %test4 ; G rlwinm. 3, 30, 0, 28, 28 bne 0, .LBB0_7 b .LBB0_8 .LBB0_4: # %optional1 ; B (Copy of C) bl a nop rlwinm. 3, 30, 0, 30, 30 beq 0, .LBB0_2 .LBB0_5: # %optional2 ; D (Copy of E) bl b nop rlwinm. 3, 30, 0, 29, 29 beq 0, .LBB0_3 .LBB0_6: # %optional3 ; F (Copy of G) bl c nop rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 .LBB0_7: # %optional4 ; H bl d nop .LBB0_8: # %exit Differential Revision: https://reviews.llvm.org/D28522 llvm-svn: 295223
2017-02-16 03:49:14 +08:00
%p2 = alloca %struct.T
; CHECK: pushl %eax
; CHECK: subl $2996, %esp
br label %bb4
bb4:
Codegen: Make chains from trellis-shaped CFGs Lay out trellis-shaped CFGs optimally. A trellis of the shape below: A B |\ /| | \ / | | X | | / \ | |/ \| C D would be laid out A; B->C ; D by the current layout algorithm. Now we identify trellises and lay them out either A->C; B->D or A->D; B->C. This scales with an increasing number of predecessors. A trellis is a a group of 2 or more predecessor blocks that all have the same successors. because of this we can tail duplicate to extend existing trellises. As an example consider the following CFG: B D F H / \ / \ / \ / \ A---C---E---G---Ret Where A,C,E,G are all small (Currently 2 instructions). The CFG preserving layout is then A,B,C,D,E,F,G,H,Ret. The current code will copy C into B, E into D and G into F and yield the layout A,C,B(C),E,D(E),F(G),G,H,ret define void @straight_test(i32 %tag) { entry: br label %test1 test1: ; A %tagbit1 = and i32 %tag, 1 %tagbit1eq0 = icmp eq i32 %tagbit1, 0 br i1 %tagbit1eq0, label %test2, label %optional1 optional1: ; B call void @a() br label %test2 test2: ; C %tagbit2 = and i32 %tag, 2 %tagbit2eq0 = icmp eq i32 %tagbit2, 0 br i1 %tagbit2eq0, label %test3, label %optional2 optional2: ; D call void @b() br label %test3 test3: ; E %tagbit3 = and i32 %tag, 4 %tagbit3eq0 = icmp eq i32 %tagbit3, 0 br i1 %tagbit3eq0, label %test4, label %optional3 optional3: ; F call void @c() br label %test4 test4: ; G %tagbit4 = and i32 %tag, 8 %tagbit4eq0 = icmp eq i32 %tagbit4, 0 br i1 %tagbit4eq0, label %exit, label %optional4 optional4: ; H call void @d() br label %exit exit: ret void } here is the layout after D27742: straight_test: # @straight_test ; ... Prologue elided ; BB#0: # %entry ; A (merged with test1) ; ... More prologue elided mr 30, 3 andi. 3, 30, 1 bc 12, 1, .LBB0_2 ; BB#1: # %test2 ; C rlwinm. 3, 30, 0, 30, 30 beq 0, .LBB0_3 b .LBB0_4 .LBB0_2: # %optional1 ; B (copy of C) bl a nop rlwinm. 3, 30, 0, 30, 30 bne 0, .LBB0_4 .LBB0_3: # %test3 ; E rlwinm. 3, 30, 0, 29, 29 beq 0, .LBB0_5 b .LBB0_6 .LBB0_4: # %optional2 ; D (copy of E) bl b nop rlwinm. 3, 30, 0, 29, 29 bne 0, .LBB0_6 .LBB0_5: # %test4 ; G rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 b .LBB0_7 .LBB0_6: # %optional3 ; F (copy of G) bl c nop rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 .LBB0_7: # %optional4 ; H bl d nop .LBB0_8: # %exit ; Ret ld 30, 96(1) # 8-byte Folded Reload addi 1, 1, 112 ld 0, 16(1) mtlr 0 blr The tail-duplication has produced some benefit, but it has also produced a trellis which is not laid out optimally. With this patch, we improve the layouts of such trellises, and decrease the cost calculation for tail-duplication accordingly. This patch produces the layout A,C,E,G,B,D,F,H,Ret. This layout does have back edges, which is a negative, but it has a bigger compensating positive, which is that it handles the case where there are long strings of skipped blocks much better than the original layout. Both layouts handle runs of executed blocks equally well. Branch prediction also improves if there is any correlation between subsequent optional blocks. Here is the resulting concrete layout: straight_test: # @straight_test ; BB#0: # %entry ; A (merged with test1) mr 30, 3 andi. 3, 30, 1 bc 12, 1, .LBB0_4 ; BB#1: # %test2 ; C rlwinm. 3, 30, 0, 30, 30 bne 0, .LBB0_5 .LBB0_2: # %test3 ; E rlwinm. 3, 30, 0, 29, 29 bne 0, .LBB0_6 .LBB0_3: # %test4 ; G rlwinm. 3, 30, 0, 28, 28 bne 0, .LBB0_7 b .LBB0_8 .LBB0_4: # %optional1 ; B (Copy of C) bl a nop rlwinm. 3, 30, 0, 30, 30 beq 0, .LBB0_2 .LBB0_5: # %optional2 ; D (Copy of E) bl b nop rlwinm. 3, 30, 0, 29, 29 beq 0, .LBB0_3 .LBB0_6: # %optional3 ; F (Copy of G) bl c nop rlwinm. 3, 30, 0, 28, 28 beq 0, .LBB0_8 .LBB0_7: # %optional4 ; H bl d nop .LBB0_8: # %exit Differential Revision: https://reviews.llvm.org/D28522 llvm-svn: 295223
2017-02-16 03:49:14 +08:00
br i1 %y, label %bb5, label %bb2
bb5:
%p4 = alloca %struct.S
; CHECK: subl $1024, %esp
call void @f(%struct.S* %p4)
ret void
}
declare void @f(%struct.S*)
declare void @g(%struct.T*)
declare void @h(%struct.U*)
declare i8* @llvm.stacksave()
declare void @llvm.stackrestore(i8*)