llvm-project/llvm/test/CodeGen/X86/statepoint-vreg-details.ll

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

354 lines
25 KiB
LLVM
Raw Normal View History

; This file contains some of the same basic tests as statepoint-vreg.ll, but
; focuses on examining the intermediate representation. It's separate so that
; the main file is easy to update with update_llc_test_checks.py
; This run is to demonstrate what MIR SSA looks like.
; RUN: llc -max-registers-for-gc-values=4 -stop-after finalize-isel < %s | FileCheck --check-prefix=CHECK-VREG %s
; This run is to demonstrate register allocator work.
; RUN: llc -max-registers-for-gc-values=4 -stop-after virtregrewriter < %s | FileCheck --check-prefix=CHECK-PREG %s
target datalayout = "e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-pc-linux-gnu"
declare dso_local i1 @return_i1()
declare dso_local void @func()
declare dso_local void @consume(i32 addrspace(1)*)
declare dso_local void @consume2(i32 addrspace(1)*, i32 addrspace(1)*)
declare dso_local void @consume5(i32 addrspace(1)*, i32 addrspace(1)*, i32 addrspace(1)*, i32 addrspace(1)*, i32 addrspace(1)*)
declare dso_local void @use1(i32 addrspace(1)*, i8 addrspace(1)*)
declare dso_local void @bar(i8 addrspace(1)*, i8 addrspace(1)*)
; test most simple relocate
define i1 @test_relocate(i32 addrspace(1)* %a) gc "statepoint-example" {
; CHECK-VREG-LABEL: name: test_relocate
; CHECK-VREG: %0:gr64 = COPY $rdi
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-VREG: %1:gr64 = STATEPOINT 0, 0, 0, @return_i1, 2, 0, 2, 0, 2, 0, 2, 1, %0(tied-def 0), 2, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp, implicit-def $al
; CHECK-VREG: %2:gr8 = COPY $al
; CHECK-VREG: $rdi = COPY %1
; CHECK-VREG: CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
; CHECK-PREG-LABEL: name: test_relocate
; CHECK-PREG: renamable $rbx = COPY $rdi
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-PREG: renamable $rbx = STATEPOINT 0, 0, 0, @return_i1, 2, 0, 2, 0, 2, 0, 2, 1, killed renamable $rbx(tied-def 0), 2, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp, implicit-def $al
; CHECK-PREG: renamable $bpl = COPY killed $al
; CHECK-PREG: $rdi = COPY killed renamable $rbx
; CHECK-PREG: CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
entry:
%safepoint_token = tail call token (i64, i32, i1 ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_i1f(i64 0, i32 0, i1 ()* @return_i1, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)* %a)]
%rel1 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 0, i32 0)
%res1 = call zeroext i1 @llvm.experimental.gc.result.i1(token %safepoint_token)
call void @consume(i32 addrspace(1)* %rel1)
ret i1 %res1
}
; test pointer variables intermixed with pointer constants
define void @test_mixed(i32 addrspace(1)* %a, i32 addrspace(1)* %b, i32 addrspace(1)* %c) gc "statepoint-example" {
; CHECK-VREG-LABEL: name: test_mixed
; CHECK-VREG: %2:gr64 = COPY $rdx
; CHECK-VREG: %1:gr64 = COPY $rsi
; CHECK-VREG: %0:gr64 = COPY $rdi
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-VREG: %3:gr64, %4:gr64, %5:gr64 = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, 2, 4, %2(tied-def 0), 2, 0, %1(tied-def 1), %0(tied-def 2), 2, 0, 2, 4, 0, 0, 1, 1, 2, 2, 3, 3, csr_64, implicit-def $rsp, implicit-def $ssp
; CHECK-VREG: %6:gr32 = MOV32r0 implicit-def dead $eflags
; CHECK-VREG: %7:gr64 = SUBREG_TO_REG 0, killed %6, %subreg.sub_32bit
; CHECK-VREG: $rdi = COPY %5
; CHECK-VREG: $rsi = COPY %7
; CHECK-VREG: $rdx = COPY %4
; CHECK-VREG: $rcx = COPY %7
; CHECK-VREG: $r8 = COPY %3
; CHECK-VREG: CALL64pcrel32 @consume5, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit $rsi, implicit $rdx, implicit $rcx, implicit $r8, implicit-def $rsp, implicit-def $ssp
; CHECK-PREG-LABEL: name: test_mixed
; CHECK-PREG: renamable $r14 = COPY $rdx
; CHECK-PREG: renamable $r15 = COPY $rsi
; CHECK-PREG: renamable $rbx = COPY $rdi
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-PREG: renamable $r14, renamable $r15, renamable $rbx = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, 2, 4, killed renamable $r14(tied-def 0), 2, 0, killed renamable $r15(tied-def 1), killed renamable $rbx(tied-def 2), 2, 0, 2, 4, 0, 0, 1, 1, 2, 2, 3, 3, csr_64, implicit-def $rsp, implicit-def $ssp
; CHECK-PREG: $rdi = COPY killed renamable $rbx
; CHECK-PREG: dead $esi = MOV32r0 implicit-def dead $eflags, implicit-def $rsi
; CHECK-PREG: $rdx = COPY killed renamable $r15
; CHECK-PREG: dead $ecx = MOV32r0 implicit-def dead $eflags, implicit-def $rcx
; CHECK-PREG: $r8 = COPY killed renamable $r14
; CHECK-PREG: CALL64pcrel32 @consume5, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit $rsi, implicit $rdx, implicit killed $rcx, implicit killed $r8, implicit-def $rsp, implicit-def $ssp
entry:
%safepoint_token = tail call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)* %a, i32 addrspace(1)* null, i32 addrspace(1)* %b, i32 addrspace(1)* null, i32 addrspace(1)* %c)]
%rel1 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 0, i32 0)
%rel2 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 1, i32 1)
%rel3 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 2, i32 2)
%rel4 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 3, i32 3)
%rel5 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 4, i32 4)
call void @consume5(i32 addrspace(1)* %rel1, i32 addrspace(1)* %rel2, i32 addrspace(1)* %rel3, i32 addrspace(1)* %rel4, i32 addrspace(1)* %rel5)
ret void
}
; same as above, but for alloca
define i32 addrspace(1)* @test_alloca(i32 addrspace(1)* %ptr) gc "statepoint-example" {
; CHECK-VREG-LABEL: name: test_alloca
; CHECK-VREG: %0:gr64 = COPY $rdi
; CHECK-VREG: MOV64mr %stack.0.alloca, 1, $noreg, 0, $noreg, %0 :: (store 8 into %ir.alloca)
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-VREG: %1:gr64 = STATEPOINT 0, 0, 0, @return_i1, 2, 0, 2, 0, 2, 0, 2, 1, %0(tied-def 0), 2, 1, 0, %stack.0.alloca, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp, implicit-def $al :: (volatile load store 8 on %stack.0.alloca)
; CHECK-VREG: %2:gr8 = COPY $al
; CHECK-VREG: %3:gr64 = MOV64rm %stack.0.alloca, 1, $noreg, 0, $noreg :: (dereferenceable load 8 from %ir.alloca)
; CHECK-VREG: $rdi = COPY %1
; CHECK-VREG: CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
; CHECK-PREG-LABEL: name: test_alloca
; CHECK-PREG: renamable $rbx = COPY $rdi
; CHECK-PREG: MOV64mr %stack.0.alloca, 1, $noreg, 0, $noreg, renamable $rbx :: (store 8 into %ir.alloca)
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-PREG: renamable $rbx = STATEPOINT 0, 0, 0, @return_i1, 2, 0, 2, 0, 2, 0, 2, 1, killed renamable $rbx(tied-def 0), 2, 1, 0, %stack.0.alloca, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp, implicit-def dead $al :: (volatile load store 8 on %stack.0.alloca)
; CHECK-PREG: renamable $r14 = MOV64rm %stack.0.alloca, 1, $noreg, 0, $noreg :: (dereferenceable load 8 from %ir.alloca)
; CHECK-PREG: $rdi = COPY killed renamable $rbx
; CHECK-PREG: CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
entry:
%alloca = alloca i32 addrspace(1)*, align 8
store i32 addrspace(1)* %ptr, i32 addrspace(1)** %alloca
%safepoint_token = call token (i64, i32, i1 ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_i1f(i64 0, i32 0, i1 ()* @return_i1, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)** %alloca, i32 addrspace(1)* %ptr)]
%rel1 = load i32 addrspace(1)*, i32 addrspace(1)** %alloca
%rel2 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 1, i32 1)
call void @consume(i32 addrspace(1)* %rel2)
ret i32 addrspace(1)* %rel1
}
; test base != derived
define void @test_base_derived(i32 addrspace(1)* %base, i32 addrspace(1)* %derived) gc "statepoint-example" {
; CHECK-VREG-LABEL: name: test_base_derived
; CHECK-VREG: %1:gr64 = COPY $rsi
; CHECK-VREG: %0:gr64 = COPY $rdi
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-VREG: %2:gr64, %3:gr64 = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, 2, 2, %1(tied-def 0), %0(tied-def 1), 2, 0, 2, 1, 1, 0, csr_64, implicit-def $rsp, implicit-def $ssp
; CHECK-VREG: $rdi = COPY %2
; CHECK-VREG: CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
; CHECK-PREG-LABEL: name: test_base_derived
; CHECK-PREG: renamable $rbx = COPY $rsi
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-PREG: renamable $rbx, dead renamable $r14 = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, 2, 2, killed renamable $rbx(tied-def 0), killed renamable $r14(tied-def 1), 2, 0, 2, 1, 1, 0, csr_64, implicit-def $rsp, implicit-def $ssp
; CHECK-PREG: $rdi = COPY killed renamable $rbx
; CHECK-PREG: CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
%safepoint_token = tail call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)* %base, i32 addrspace(1)* %derived)]
%reloc = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 0, i32 1)
call void @consume(i32 addrspace(1)* %reloc)
ret void
}
; deopt GC pointer not present in GC args must be spilled
define void @test_deopt_gcpointer(i32 addrspace(1)* %a, i32 addrspace(1)* %b) gc "statepoint-example" {
; CHECK-VREG-LABEL: name: test_deopt_gcpointer
; CHECK-VREG: %1:gr64 = COPY $rsi
; CHECK-VREG: %0:gr64 = COPY $rdi
; CHECK-VREG: MOV64mr %stack.0, 1, $noreg, 0, $noreg, %0 :: (store 8 into %stack.0)
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-VREG: %2:gr64 = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 1, 1, 8, %stack.0, 0, 2, 1, %1(tied-def 0), 2, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp :: (volatile load store 8 on %stack.0)
; CHECK-VREG: $rdi = COPY %2
; CHECK-VREG: CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
; CHECK-VREG: RET 0
; CHECK-PREG-LABEL: name: test_deopt_gcpointer
; CHECK-PREG: renamable $rbx = COPY $rsi
; CHECK-PREG: MOV64mr %stack.0, 1, $noreg, 0, $noreg, killed renamable $rdi :: (store 8 into %stack.0)
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-PREG: renamable $rbx = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 1, 1, 8, %stack.0, 0, 2, 1, killed renamable $rbx(tied-def 0), 2, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp :: (volatile load store 8 on %stack.0)
; CHECK-PREG: $rdi = COPY killed renamable $rbx
; CHECK-PREG: CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
%safepoint_token = tail call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["deopt" (i32 addrspace(1)* %a), "gc-live" (i32 addrspace(1)* %b)]
%rel = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 0, i32 0)
call void @consume(i32 addrspace(1)* %rel)
ret void
}
;; Two gc.relocates of the same input, should require only a single spill/fill
define void @test_gcrelocate_uniqueing(i32 addrspace(1)* %ptr) gc "statepoint-example" {
; CHECK-VREG-LABEL: name: test_gcrelocate_uniqueing
; CHECK-VREG: %0:gr64 = COPY $rdi
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-VREG: %1:gr64 = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 2, %0, 2, 4278124286, 2, 1, %0(tied-def 0), 2, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp
; CHECK-VREG: $rdi = COPY %1
; CHECK-VREG: $rsi = COPY %1
; CHECK-VREG: CALL64pcrel32 @consume2, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit $rsi, implicit-def $rsp, implicit-def $ssp
; CHECK-PREG-LABEL: name: test_gcrelocate_uniqueing
; CHECK-PREG: renamable $rbx = COPY $rdi
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-PREG: renamable $rbx = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 2, killed renamable $rbx, 2, 4278124286, 2, 1, renamable $rbx(tied-def 0), 2, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp
; CHECK-PREG: $rdi = COPY renamable $rbx
; CHECK-PREG: $rsi = COPY killed renamable $rbx
; CHECK-PREG: CALL64pcrel32 @consume2, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit killed $rsi, implicit-def $rsp, implicit-def $ssp
%tok = tail call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["deopt" (i32 addrspace(1)* %ptr, i32 undef), "gc-live" (i32 addrspace(1)* %ptr, i32 addrspace(1)* %ptr)]
%a = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %tok, i32 0, i32 0)
%b = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %tok, i32 1, i32 1)
call void @consume2(i32 addrspace(1)* %a, i32 addrspace(1)* %b)
ret void
}
; Two gc.relocates of a bitcasted pointer should only require a single spill/fill
define void @test_gcptr_uniqueing(i32 addrspace(1)* %ptr) gc "statepoint-example" {
; CHECK-VREG-LABEL: name: test_gcptr_uniqueing
; CHECK-VREG: %0:gr64 = COPY $rdi
; CHECK-VREG: ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-VREG: %1:gr64 = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 2, %0, 2, 4278124286, 2, 1, %0(tied-def 0), 2, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp
; CHECK-VREG: ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
; CHECK-VREG: ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
; CHECK-VREG: $rdi = COPY %1
; CHECK-VREG: $rsi = COPY %1
; CHECK-VREG: CALL64pcrel32 @use1, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit $rsi, implicit-def $rsp, implicit-def $ssp
; CHECK-PREG-LABEL: name: test_gcptr_uniqueing
; CHECK-PREG: renamable $rbx = COPY $rdi
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-PREG: renamable $rbx = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 2, killed renamable $rbx, 2, 4278124286, 2, 1, renamable $rbx(tied-def 0), 2, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp
; CHECK-PREG: $rdi = COPY renamable $rbx
; CHECK-PREG: $rsi = COPY killed renamable $rbx
; CHECK-PREG: CALL64pcrel32 @use1, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit killed $rsi, implicit-def $rsp, implicit-def $ssp
%ptr2 = bitcast i32 addrspace(1)* %ptr to i8 addrspace(1)*
%tok = tail call token (i64, i32, void ()*, i32, i32, ...)
@llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["deopt" (i32 addrspace(1)* %ptr, i32 undef), "gc-live" (i32 addrspace(1)* %ptr, i8 addrspace(1)* %ptr2)]
%a = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %tok, i32 0, i32 0)
%b = call i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %tok, i32 1, i32 1)
call void @use1(i32 addrspace(1)* %a, i8 addrspace(1)* %b)
ret void
}
define i1 @test_cross_bb(i32 addrspace(1)* %a, i1 %external_cond) gc "statepoint-example" {
; CHECK-VREG-LABEL: name: test_cross_bb
; CHECK-VREG: bb.0.entry:
; CHECK-VREG: %1:gr32 = COPY $esi
; CHECK-VREG-NEXT: %0:gr64 = COPY $rdi
; CHECK-VREG-NEXT: %4:gr8 = COPY %1.sub_8bit
; CHECK-VREG-NEXT: ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-VREG-NEXT: %2:gr64 = STATEPOINT 0, 0, 0, @return_i1, 2, 0, 2, 0, 2, 0, 2, 1, %0(tied-def 0), 2, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp, implicit-def $al
; CHECK-VREG-NEXT: ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
; CHECK-VREG-NEXT: %5:gr8 = COPY $al
; CHECK-VREG-NEXT: %3:gr8 = COPY %5
; CHECK-VREG-NEXT: TEST8ri killed %4, 1, implicit-def $eflags
; CHECK-VREG-NEXT: JCC_1 %bb.2, 4, implicit $eflags
; CHECK-VREG-NEXT: JMP_1 %bb.1
; CHECK-VREG: bb.1.left:
; CHECK-VREG-NEXT: ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
; CHECK-VREG-NEXT: $rdi = COPY %2
; CHECK-VREG-NEXT: CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
; CHECK-VREG-NEXT: ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
; CHECK-VREG-NEXT: $al = COPY %3
; CHECK-VREG-NEXT: RET 0, $al
; CHECK-VREG: bb.2.right:
; CHECK-VREG-NEXT: %6:gr8 = MOV8ri 1
; CHECK-VREG-NEXT: $al = COPY %6
; CHECK-VREG-NEXT: RET 0, $al
entry:
%safepoint_token = tail call token (i64, i32, i1 ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_i1f(i64 0, i32 0, i1 ()* @return_i1, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)* %a)]
br i1 %external_cond, label %left, label %right
left:
%call1 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 0, i32 0)
%call2 = call zeroext i1 @llvm.experimental.gc.result.i1(token %safepoint_token)
call void @consume(i32 addrspace(1)* %call1)
ret i1 %call2
right:
ret i1 true
}
; No need to check post-regalloc output as it is the same
define i1 @duplicate_reloc() gc "statepoint-example" {
; CHECK-VREG-LABEL: name: duplicate_reloc
; CHECK-VREG: bb.0.entry:
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-VREG: STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, 2, 1, 2, 0, 2, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp
; CHECK-VREG: STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, 2, 1, 2, 0, 2, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp
; CHECK-VREG: %0:gr8 = MOV8ri 1
; CHECK-VREG: $al = COPY %0
; CHECK-VREG: RET 0, $al
entry:
%safepoint_token = call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)* null, i32 addrspace(1)* null)]
%base = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 0, i32 0)
%derived = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 0, i32 1)
%safepoint_token2 = call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)* %base, i32 addrspace(1)* %derived)]
%base_reloc = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token2, i32 0, i32 0)
%derived_reloc = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token2, i32 0, i32 1)
%cmp1 = icmp eq i32 addrspace(1)* %base_reloc, null
%cmp2 = icmp eq i32 addrspace(1)* %derived_reloc, null
%cmp = and i1 %cmp1, %cmp2
ret i1 %cmp
}
; Vectors cannot go in VRegs
; No need to check post-regalloc output as it is lowered using old scheme
define <2 x i8 addrspace(1)*> @test_vector(<2 x i8 addrspace(1)*> %obj) gc "statepoint-example" {
; CHECK-VREG-LABEL: name: test_vector
; CHECK-VREG: %0:vr128 = COPY $xmm0
; CHECK-VREG: MOVAPSmr %stack.0, 1, $noreg, 0, $noreg, %0 :: (store 16 into %stack.0)
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-VREG: STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, 2, 1, 1, 16, %stack.0, 0, 2, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp :: (volatile load store 16 on %stack.0)
; CHECK-VREG: %1:vr128 = MOVAPSrm %stack.0, 1, $noreg, 0, $noreg :: (load 16 from %stack.0)
; CHECK-VREG: $xmm0 = COPY %1
; CHECK-VREG: RET 0, $xmm0
entry:
%safepoint_token = call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["gc-live" (<2 x i8 addrspace(1)*> %obj)]
%obj.relocated = call coldcc <2 x i8 addrspace(1)*> @llvm.experimental.gc.relocate.v2p1i8(token %safepoint_token, i32 0, i32 0) ; (%obj, %obj)
ret <2 x i8 addrspace(1)*> %obj.relocated
}
; test limit on amount of vregs
define void @test_limit(i32 addrspace(1)* %a, i32 addrspace(1)* %b, i32 addrspace(1)* %c, i32 addrspace(1)* %d, i32 addrspace(1)* %e) gc "statepoint-example" {
; CHECK-VREG-LABEL: name: test_limit
; CHECK-VREG: %4:gr64 = COPY $r8
; CHECK-VREG: %3:gr64 = COPY $rcx
; CHECK-VREG: %2:gr64 = COPY $rdx
; CHECK-VREG: %1:gr64 = COPY $rsi
; CHECK-VREG: %0:gr64 = COPY $rdi
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
; CHECK-VREG: MOV64mr %stack.0, 1, $noreg, 0, $noreg, %0 :: (store 8 into %stack.0)
; CHECK-VREG: %5:gr64, %6:gr64, %7:gr64, %8:gr64 = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, 2, 5, %4(tied-def 0), %3(tied-def 1), %2(tied-def 2), %1(tied-def 3), 1, 8, %stack.0, 0, 2, 0, 2, 5, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, csr_64, implicit-def $rsp, implicit-def $ssp :: (volatile load store 8 on %stack.0)
; CHECK-VREG: %9:gr64 = MOV64rm %stack.0, 1, $noreg, 0, $noreg :: (load 8 from %stack.0)
; CHECK-VREG: $rdi = COPY %9
; CHECK-VREG: $rsi = COPY %8
; CHECK-VREG: $rdx = COPY %7
; CHECK-VREG: $rcx = COPY %6
; CHECK-VREG: $r8 = COPY %5
; CHECK-VREG: CALL64pcrel32 @consume5, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit $rsi, implicit $rdx, implicit $rcx, implicit $r8, implicit-def $rsp, implicit-def $ssp
; CHECK-VREG: RET 0
entry:
%safepoint_token = tail call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)* %a, i32 addrspace(1)* %b, i32 addrspace(1)* %c, i32 addrspace(1)* %d, i32 addrspace(1)* %e)]
%rel1 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 0, i32 0)
%rel2 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 1, i32 1)
%rel3 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 2, i32 2)
%rel4 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 3, i32 3)
%rel5 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token, i32 4, i32 4)
call void @consume5(i32 addrspace(1)* %rel1, i32 addrspace(1)* %rel2, i32 addrspace(1)* %rel3, i32 addrspace(1)* %rel4, i32 addrspace(1)* %rel5)
ret void
}
; Test that CopyFromReg emitted during ISEL processing of gc.relocate are properly ordered w.r.t. statepoint.
define i8 addrspace(1)* @test_isel_sched(i8 addrspace(1)* %0, i8 addrspace(1)* %1, i32 %2) gc "statepoint-example" {
;CHECK-VREG-LABEL: name: test_isel_sched
;CHECK-VREG: bb.0.entry:
;CHECK-VREG: %2:gr32 = COPY $edx
;CHECK-VREG: %1:gr64 = COPY $rsi
;CHECK-VREG: %0:gr64 = COPY $rdi
;CHECK-VREG: TEST32rr %2, %2, implicit-def $eflags
;CHECK-VREG: %5:gr64 = CMOV64rr %1, %0, 4, implicit $eflags
;CHECK-VREG: %6:gr32 = MOV32r0 implicit-def dead $eflags
;CHECK-VREG: %7:gr64 = SUBREG_TO_REG 0, killed %6, %subreg.sub_32bit
;CHECK-VREG: $rdi = COPY %7
;CHECK-VREG: $rsi = COPY %5
[Statepoints] Change statepoint machine instr format to better suit VReg lowering. Current Statepoint MI format is this: STATEPOINT <id>, <num patch bytes >, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <gc base/derived pairs...> <gc allocas...> Note that GC pointers are listed in pairs <base,derived>. This causes base pointers to appear many times (at least twice) in instruction, which is bad for us when VReg lowering is ON. The problem is that machine operand tiedness is 1-1 relation, so it might look like this: %vr2 = STATEPOINT ... %vr1, %vr1(tied-def0) Since only one instance of %vr1 is tied, that may lead to incorrect codegen (see PR46917 for more details), so we have to always spill base pointers. This mostly defeats new VReg lowering scheme. This patch changes statepoint instruction format so that every gc pointer appears only once in operand list. That way they all can be tied. Additional set of operands is added to preserve base-derived relation required to build stackmap. New statepoint has following format: STATEPOINT <id>, <num patch bytes>, <num call arguments>, <call target>, [call arguments...], <StackMaps::ConstantOp>, <calling convention>, <StackMaps::ConstantOp>, <statepoint flags>, <StackMaps::ConstantOp>, <num deopt args>, [deopt args...], <StackMaps::ConstantOp>, <num gc pointers>, [gc pointers...], <StackMaps::ConstantOp>, <num gc allocas>, [gc allocas...] <StackMaps::ConstantOp>, <num entries in gc map>, [base/derived indices...] Changes are: - every gc pointer is listed only once in a flat length-prefixed list; - alloca list is prefixed with its length too; - following alloca list is length-prefixed list of base-derived indices of pointers from gc pointer list. Note that indices are logical (number of pointer), not absolute (index of machine operand). Differential Revision: https://reviews.llvm.org/D87154
2020-09-05 01:45:41 +08:00
;CHECK-VREG: %3:gr64, %4:gr64 = STATEPOINT 10, 0, 2, @bar, $rdi, $rsi, 2, 0, 2, 0, 2, 0, 2, 2, %1(tied-def 0), %0(tied-def 1), 2, 0, 2, 2, 0, 0, 1, 1, csr_64, implicit-def $rsp, implicit-def $ssp
;CHECK-VREG: TEST32rr %2, %2, implicit-def $eflags
;CHECK-VREG: %8:gr64 = CMOV64rr %3, %4, 4, implicit $eflags
;CHECK-VREG: $rax = COPY %8
;CHECK-VREG: RET 0, $rax
entry:
%cmp = icmp eq i32 %2, 0
%ptr = select i1 %cmp, i8 addrspace(1)* %0, i8 addrspace(1)* %1
%token = call token (i64, i32, void (i8 addrspace(1)*, i8 addrspace(1)*)*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidp1i8p1i8f(i64 10, i32 0, void (i8 addrspace(1)*, i8 addrspace(1)*)* @bar, i32 2, i32 0, i8 addrspace(1)* null, i8 addrspace(1)* %ptr, i32 0, i32 0) [ "deopt"(), "gc-live"(i8 addrspace(1)* %0, i8 addrspace(1)* %1) ]
%rel0 = call coldcc i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %token, i32 0, i32 0)
%rel1 = call coldcc i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %token, i32 1, i32 1)
%res = select i1 %cmp, i8 addrspace(1)* %rel0, i8 addrspace(1)* %rel1
ret i8 addrspace(1)* %res
}
declare token @llvm.experimental.gc.statepoint.p0f_i1f(i64, i32, i1 ()*, i32, i32, ...)
declare token @llvm.experimental.gc.statepoint.p0f_isVoidf(i64, i32, void ()*, i32, i32, ...)
declare dso_local i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token, i32, i32)
declare dso_local i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token, i32, i32)
declare <2 x i8 addrspace(1)*> @llvm.experimental.gc.relocate.v2p1i8(token, i32, i32)
declare dso_local i1 @llvm.experimental.gc.result.i1(token)
declare token @llvm.experimental.gc.statepoint.p0f_isVoidp1i8p1i8f(i64 immarg, i32 immarg, void (i8 addrspace(1)*, i8 addrspace(1)*)*, i32 immarg, i32 immarg, ...)