[CodeGen][RISCV] Inline stack probes immediately after allocateStack in eliminateCallFramePseudoInstr#195456
Conversation
|
Thank you for submitting a Pull Request (PR) to the LLVM Project! This PR will be automatically labeled and the relevant teams will be notified. If you wish to, you can add reviewers by using the "Reviewers" section on this page. If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers. If you have further questions, they may be answered by the LLVM GitHub User Guide. You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums. |
|
@llvm/pr-subscribers-backend-risc-v Author: Rong "Mantle" Bao (CSharperMantle) ChangesThis PR extends the search range for inline stack probe insertion to all BBs, instead of the current " Fixes #195454. Full diff: https://github.com/llvm/llvm-project/pull/195456.diff 2 Files Affected:
diff --git a/llvm/lib/CodeGen/PrologEpilogInserter.cpp b/llvm/lib/CodeGen/PrologEpilogInserter.cpp
index 68fd54cf00146..5c7dc6f48e274 100644
--- a/llvm/lib/CodeGen/PrologEpilogInserter.cpp
+++ b/llvm/lib/CodeGen/PrologEpilogInserter.cpp
@@ -1181,8 +1181,14 @@ void PEIImpl::insertPrologEpilogCode(MachineFunction &MF) {
// Zero call used registers before restoring callee-saved registers.
insertZeroCallUsedRegs(MF);
- for (MachineBasicBlock *SaveBlock : SaveBlocks)
- TFI.inlineStackProbe(MF, *SaveBlock);
+ {
+ SmallVector<MachineBasicBlock *, 8> Blocks;
+ for (MachineBasicBlock &MBB : MF)
+ Blocks.push_back(&MBB);
+
+ for (MachineBasicBlock *MBB : Blocks)
+ TFI.inlineStackProbe(MF, *MBB);
+ }
// Emit additional code that is required to support segmented stacks, if
// we've been asked for it. This, when linked with a runtime with support
diff --git a/llvm/test/CodeGen/RISCV/stack-probing-dynamic-nonentry.ll b/llvm/test/CodeGen/RISCV/stack-probing-dynamic-nonentry.ll
new file mode 100644
index 0000000000000..411168d445a95
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/stack-probing-dynamic-nonentry.ll
@@ -0,0 +1,115 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=riscv64 -mattr=+m -O2 < %s | FileCheck %s -check-prefix=RV64I
+; RUN: llc -mtriple=riscv32 -mattr=+m -O2 < %s | FileCheck %s -check-prefix=RV32I
+
+; Test that very large outgoing call frames in functions with variable-sized
+; objects get proper stack probing. The outgoing args are large enough to force
+; the PROBED_STACKALLOC path, which must be expanded in a non-entry block.
+
+define void @f(i64 %n) #0 {
+; RV64I-LABEL: f:
+; RV64I: # %bb.0: # %entry
+; RV64I-NEXT: addi sp, sp, -16
+; RV64I-NEXT: .cfi_def_cfa_offset 16
+; RV64I-NEXT: sd zero, 0(sp)
+; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT: sd s0, 0(sp) # 8-byte Folded Spill
+; RV64I-NEXT: .cfi_offset ra, -8
+; RV64I-NEXT: .cfi_offset s0, -16
+; RV64I-NEXT: addi s0, sp, 16
+; RV64I-NEXT: .cfi_def_cfa s0, 0
+; RV64I-NEXT: slli a0, a0, 2
+; RV64I-NEXT: addi a0, a0, 15
+; RV64I-NEXT: andi a0, a0, -16
+; RV64I-NEXT: sub a0, sp, a0
+; RV64I-NEXT: lui a1, 1
+; RV64I-NEXT: .LBB0_1: # %entry
+; RV64I-NEXT: # =>This Inner Loop Header: Depth=1
+; RV64I-NEXT: sub sp, sp, a1
+; RV64I-NEXT: sd zero, 0(sp)
+; RV64I-NEXT: bltu a0, sp, .LBB0_1
+; RV64I-NEXT: # %bb.2: # %entry
+; RV64I-NEXT: mv sp, a0
+; RV64I-NEXT: lui a1, 5
+; RV64I-NEXT: sub t1, sp, a1
+; RV64I-NEXT: lui t2, 1
+; RV64I-NEXT: .LBB0_3: # %entry
+; RV64I-NEXT: # =>This Inner Loop Header: Depth=1
+; RV64I-NEXT: sub sp, sp, t2
+; RV64I-NEXT: sd zero, 0(sp)
+; RV64I-NEXT: bne sp, t1, .LBB0_3
+; RV64I-NEXT: # %bb.4: # %entry
+; RV64I-NEXT: addi sp, sp, -2048
+; RV64I-NEXT: addi sp, sp, -1424
+; RV64I-NEXT: sd zero, 0(sp)
+; RV64I-NEXT: call g
+; RV64I-NEXT: lui a0, 6
+; RV64I-NEXT: addi a0, a0, -624
+; RV64I-NEXT: add sp, sp, a0
+; RV64I-NEXT: addi sp, s0, -16
+; RV64I-NEXT: .cfi_def_cfa sp, 16
+; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT: ld s0, 0(sp) # 8-byte Folded Reload
+; RV64I-NEXT: .cfi_restore ra
+; RV64I-NEXT: .cfi_restore s0
+; RV64I-NEXT: addi sp, sp, 16
+; RV64I-NEXT: .cfi_def_cfa_offset 0
+; RV64I-NEXT: ret
+;
+; RV32I-LABEL: f:
+; RV32I: # %bb.0: # %entry
+; RV32I-NEXT: addi sp, sp, -16
+; RV32I-NEXT: .cfi_def_cfa_offset 16
+; RV32I-NEXT: sw zero, 0(sp)
+; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
+; RV32I-NEXT: .cfi_offset ra, -4
+; RV32I-NEXT: .cfi_offset s0, -8
+; RV32I-NEXT: addi s0, sp, 16
+; RV32I-NEXT: .cfi_def_cfa s0, 0
+; RV32I-NEXT: slli a0, a0, 2
+; RV32I-NEXT: addi a0, a0, 15
+; RV32I-NEXT: andi a0, a0, -16
+; RV32I-NEXT: sub a0, sp, a0
+; RV32I-NEXT: lui a1, 1
+; RV32I-NEXT: .LBB0_1: # %entry
+; RV32I-NEXT: # =>This Inner Loop Header: Depth=1
+; RV32I-NEXT: sub sp, sp, a1
+; RV32I-NEXT: sw zero, 0(sp)
+; RV32I-NEXT: bltu a0, sp, .LBB0_1
+; RV32I-NEXT: # %bb.2: # %entry
+; RV32I-NEXT: mv sp, a0
+; RV32I-NEXT: lui a1, 5
+; RV32I-NEXT: sub t1, sp, a1
+; RV32I-NEXT: lui t2, 1
+; RV32I-NEXT: .LBB0_3: # %entry
+; RV32I-NEXT: # =>This Inner Loop Header: Depth=1
+; RV32I-NEXT: sub sp, sp, t2
+; RV32I-NEXT: sw zero, 0(sp)
+; RV32I-NEXT: bne sp, t1, .LBB0_3
+; RV32I-NEXT: # %bb.4: # %entry
+; RV32I-NEXT: addi sp, sp, -2048
+; RV32I-NEXT: addi sp, sp, -1456
+; RV32I-NEXT: sw zero, 0(sp)
+; RV32I-NEXT: call g
+; RV32I-NEXT: lui a0, 6
+; RV32I-NEXT: addi a0, a0, -592
+; RV32I-NEXT: add sp, sp, a0
+; RV32I-NEXT: addi sp, s0, -16
+; RV32I-NEXT: .cfi_def_cfa sp, 16
+; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
+; RV32I-NEXT: .cfi_restore ra
+; RV32I-NEXT: .cfi_restore s0
+; RV32I-NEXT: addi sp, sp, 16
+; RV32I-NEXT: .cfi_def_cfa_offset 0
+; RV32I-NEXT: ret
+entry:
+ %v = alloca i32, i64 %n
+ call void @g(ptr %v, [3000 x i64] poison)
+ ret void
+}
+
+declare void @g(ptr, [3000 x i64])
+
+attributes #0 = { "probe-stack"="inline-asm" }
|
c35749c to
d8fa841
Compare
| { | ||
| SmallVector<MachineBasicBlock *> Blocks; | ||
| for (MachineBasicBlock &MBB : MF) | ||
| Blocks.push_back(&MBB); | ||
|
|
||
| for (MachineBasicBlock *MBB : Blocks) | ||
| TFI.inlineStackProbe(MF, *MBB); | ||
| } |
There was a problem hiding this comment.
I don't think this is the best place to fix it, this inlineStackProbe call is intended to expand the PROBED_STACKALLOC added by emitPrologue which is also called only for the SaveBlocks.
Also, this is run by other targets like PowerPC which doesn't need all blocks to be searched here.
I would prefer to add a inlineStackProbe call after allocateStack in RISCVFrameLowering::eliminateCallFramePseudoInstr.
There was a problem hiding this comment.
I don't think this is the best place to fix it, this
inlineStackProbecall is intended to expand the PROBED_STACKALLOC added byemitProloguewhich is also called only for the SaveBlocks. Also, this is run by other targets like PowerPC which doesn't need all blocks to be searched here.
Ooh, sorry, I didn't know about that. This was the only call site to TargetFrameLowering::inlineStackProbe(), so I thought it was the place to fix...
I would prefer to add a
inlineStackProbecall afterallocateStackinRISCVFrameLowering::eliminateCallFramePseudoInstr.
Thanks, I'll fix this in v2.
d8fa841 to
1862e39
Compare
allocateStack in eliminateCallFramePseudoInstr
|
The code looks good to me now, but it's in an out-of-date tree. |
1862e39 to
c9b0f97
Compare
|
@rzinsly: Sure. |
c9b0f97 to
2619f42
Compare
…in eliminateCallFramePseudoInstr This prevents the PROBED_STACKALLOC pseudoinstr being leaked to and discarded in later stages, which could lead to underadjusted sp.
2619f42 to
c8fc1c9
Compare
rzinsly
left a comment
There was a problem hiding this comment.
LGTM
Thank you for finding and fixing this!
|
@CSharperMantle Congratulations on having your first Pull Request (PR) merged into the LLVM Project! Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR. Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues. How to do this, and the rest of the post-merge process, is covered in detail here. If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again. If you don't get any reports, no action is required from you. Your changes are working as expected, well done! |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/141/builds/18334 Here is the relevant piece of the build log for the reference |
|
That failure looks irrelevant. |
…` in `eliminateCallFramePseudoInstr` (llvm#195456) This PR adds a call to `inlineStackProbe` immediately after `allocateStack` in `eliminateCallFramePseudoInstr`. This allows code generation for stack probe pseudoinstructions in non-entry BBs. Fixes llvm#195454.
…` in `eliminateCallFramePseudoInstr` (llvm#195456) This PR adds a call to `inlineStackProbe` immediately after `allocateStack` in `eliminateCallFramePseudoInstr`. This allows code generation for stack probe pseudoinstructions in non-entry BBs. Fixes llvm#195454.
…` in `eliminateCallFramePseudoInstr` (llvm#195456) This PR adds a call to `inlineStackProbe` immediately after `allocateStack` in `eliminateCallFramePseudoInstr`. This allows code generation for stack probe pseudoinstructions in non-entry BBs. Fixes llvm#195454.
…llFramePseudoInstr` (llvm#195456) [ Upstream commit 589faed ] Revert `bltu` in probing loops to `blt` because commit f162be2 isn't applied on release/22.x yet. Link: llvm#192485 ("[RISCV] Use unsigned comparison for stack clash probing loop") --- This PR adds a call to `inlineStackProbe` immediately after `allocateStack` in `eliminateCallFramePseudoInstr`. This allows code generation for stack probe pseudoinstructions in non-entry BBs. Fixes llvm#195454.
…llFramePseudoInstr` (llvm#195456) [ Upstream commit 589faed ] Revert `bltu` in probing loops to `blt` because commit f162be2 isn't applied on release/22.x yet. Link: llvm#192485 ("[RISCV] Use unsigned comparison for stack clash probing loop") --- This PR adds a call to `inlineStackProbe` immediately after `allocateStack` in `eliminateCallFramePseudoInstr`. This allows code generation for stack probe pseudoinstructions in non-entry BBs. Fixes llvm#195454.
This PR adds a call to
inlineStackProbeimmediately afterallocateStackineliminateCallFramePseudoInstr. This allows code generation for stack probe pseudoinstructions in non-entry BBs.Fixes #195454.