RegAlloc: Fix failure on undef use when all registers are reserved #119647
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This was referenced Dec 12, 2024
|
@llvm/pr-subscribers-backend-amdgpu Author: Matt Arsenault (arsenm) ChangesGreedy and fast would hit different assertions on undef uses if all Full diff: https://github.com/llvm/llvm-project/pull/119647.diff 3 Files Affected:
diff --git a/llvm/lib/CodeGen/RegAllocFast.cpp b/llvm/lib/CodeGen/RegAllocFast.cpp
index 8323a050bcbc4a..87c0636fc45188 100644
--- a/llvm/lib/CodeGen/RegAllocFast.cpp
+++ b/llvm/lib/CodeGen/RegAllocFast.cpp
@@ -982,16 +982,23 @@ void RegAllocFastImpl::allocVirtRegUndef(MachineOperand &MO) {
if (!shouldAllocateRegister(VirtReg))
return;
- LiveRegMap::const_iterator LRI = findLiveVirtReg(VirtReg);
+ LiveRegMap::iterator LRI = findLiveVirtReg(VirtReg);
MCPhysReg PhysReg;
if (LRI != LiveVirtRegs.end() && LRI->PhysReg) {
PhysReg = LRI->PhysReg;
} else {
const TargetRegisterClass &RC = *MRI->getRegClass(VirtReg);
ArrayRef<MCPhysReg> AllocationOrder = RegClassInfo.getOrder(&RC);
- // FIXME: This can happen, and should fall back to a reserved entry in RC.
- assert(!AllocationOrder.empty() && "Allocation order must not be empty");
- PhysReg = AllocationOrder[0];
+ if (AllocationOrder.empty()) {
+ // All registers in the class were reserved.
+ //
+ // It might be OK to take any entry from the class as this is an undef
+ // use, but accepting this would give different behavior than greedy and
+ // basic.
+ PhysReg = getErrorAssignment(*LRI, *MO.getParent(), RC);
+ LRI->Error = true;
+ } else
+ PhysReg = AllocationOrder[0];
}
unsigned SubRegIdx = MO.getSubReg();
diff --git a/llvm/lib/CodeGen/RegAllocGreedy.cpp b/llvm/lib/CodeGen/RegAllocGreedy.cpp
index cb29218e966e06..af48e916feab45 100644
--- a/llvm/lib/CodeGen/RegAllocGreedy.cpp
+++ b/llvm/lib/CodeGen/RegAllocGreedy.cpp
@@ -2465,7 +2465,7 @@ MCRegister RAGreedy::selectOrSplitImpl(const LiveInterval &VirtReg,
return 0;
}
- if (Stage < RS_Spill) {
+ if (Stage < RS_Spill && !VirtReg.empty()) {
// Try splitting VirtReg or interferences.
unsigned NewVRegSizeBefore = NewVRegs.size();
Register PhysReg = trySplit(VirtReg, Order, NewVRegs, FixedRegisters);
diff --git a/llvm/test/CodeGen/AMDGPU/ran-out-of-registers-error-all-regs-reserved.ll b/llvm/test/CodeGen/AMDGPU/ran-out-of-registers-error-all-regs-reserved.ll
index 05975920ebeb8d..388a8e804a8896 100644
--- a/llvm/test/CodeGen/AMDGPU/ran-out-of-registers-error-all-regs-reserved.ll
+++ b/llvm/test/CodeGen/AMDGPU/ran-out-of-registers-error-all-regs-reserved.ll
@@ -24,10 +24,10 @@ define <32 x i32> @no_registers_from_class_available_to_allocate_asm_def() #0 {
ret <32 x i32> %ret
}
-; FIXME: Special case in fast RA, asserts. Also asserts in greedy
-; define void @no_registers_from_class_available_to_allocate_undef_asm() #0 {
-; call void asm sideeffect "; use $0", "v"(<32 x i32> poison)
-; ret void
-; }
+; CHECK: error: <unknown>:0:0: no registers from class available to allocate in function 'no_registers_from_class_available_to_allocate_undef_asm'
+define void @no_registers_from_class_available_to_allocate_undef_asm() #0 {
+ call void asm sideeffect "; use $0", "v"(<32 x i32> poison)
+ ret void
+}
attributes #0 = { "amdgpu-waves-per-eu"="10,10" }
|
|
@llvm/pr-subscribers-llvm-regalloc Author: Matt Arsenault (arsenm) ChangesGreedy and fast would hit different assertions on undef uses if all Full diff: https://github.com/llvm/llvm-project/pull/119647.diff 3 Files Affected:
diff --git a/llvm/lib/CodeGen/RegAllocFast.cpp b/llvm/lib/CodeGen/RegAllocFast.cpp
index 8323a050bcbc4a..87c0636fc45188 100644
--- a/llvm/lib/CodeGen/RegAllocFast.cpp
+++ b/llvm/lib/CodeGen/RegAllocFast.cpp
@@ -982,16 +982,23 @@ void RegAllocFastImpl::allocVirtRegUndef(MachineOperand &MO) {
if (!shouldAllocateRegister(VirtReg))
return;
- LiveRegMap::const_iterator LRI = findLiveVirtReg(VirtReg);
+ LiveRegMap::iterator LRI = findLiveVirtReg(VirtReg);
MCPhysReg PhysReg;
if (LRI != LiveVirtRegs.end() && LRI->PhysReg) {
PhysReg = LRI->PhysReg;
} else {
const TargetRegisterClass &RC = *MRI->getRegClass(VirtReg);
ArrayRef<MCPhysReg> AllocationOrder = RegClassInfo.getOrder(&RC);
- // FIXME: This can happen, and should fall back to a reserved entry in RC.
- assert(!AllocationOrder.empty() && "Allocation order must not be empty");
- PhysReg = AllocationOrder[0];
+ if (AllocationOrder.empty()) {
+ // All registers in the class were reserved.
+ //
+ // It might be OK to take any entry from the class as this is an undef
+ // use, but accepting this would give different behavior than greedy and
+ // basic.
+ PhysReg = getErrorAssignment(*LRI, *MO.getParent(), RC);
+ LRI->Error = true;
+ } else
+ PhysReg = AllocationOrder[0];
}
unsigned SubRegIdx = MO.getSubReg();
diff --git a/llvm/lib/CodeGen/RegAllocGreedy.cpp b/llvm/lib/CodeGen/RegAllocGreedy.cpp
index cb29218e966e06..af48e916feab45 100644
--- a/llvm/lib/CodeGen/RegAllocGreedy.cpp
+++ b/llvm/lib/CodeGen/RegAllocGreedy.cpp
@@ -2465,7 +2465,7 @@ MCRegister RAGreedy::selectOrSplitImpl(const LiveInterval &VirtReg,
return 0;
}
- if (Stage < RS_Spill) {
+ if (Stage < RS_Spill && !VirtReg.empty()) {
// Try splitting VirtReg or interferences.
unsigned NewVRegSizeBefore = NewVRegs.size();
Register PhysReg = trySplit(VirtReg, Order, NewVRegs, FixedRegisters);
diff --git a/llvm/test/CodeGen/AMDGPU/ran-out-of-registers-error-all-regs-reserved.ll b/llvm/test/CodeGen/AMDGPU/ran-out-of-registers-error-all-regs-reserved.ll
index 05975920ebeb8d..388a8e804a8896 100644
--- a/llvm/test/CodeGen/AMDGPU/ran-out-of-registers-error-all-regs-reserved.ll
+++ b/llvm/test/CodeGen/AMDGPU/ran-out-of-registers-error-all-regs-reserved.ll
@@ -24,10 +24,10 @@ define <32 x i32> @no_registers_from_class_available_to_allocate_asm_def() #0 {
ret <32 x i32> %ret
}
-; FIXME: Special case in fast RA, asserts. Also asserts in greedy
-; define void @no_registers_from_class_available_to_allocate_undef_asm() #0 {
-; call void asm sideeffect "; use $0", "v"(<32 x i32> poison)
-; ret void
-; }
+; CHECK: error: <unknown>:0:0: no registers from class available to allocate in function 'no_registers_from_class_available_to_allocate_undef_asm'
+define void @no_registers_from_class_available_to_allocate_undef_asm() #0 {
+ call void asm sideeffect "; use $0", "v"(<32 x i32> poison)
+ ret void
+}
attributes #0 = { "amdgpu-waves-per-eu"="10,10" }
|
This was referenced Dec 12, 2024
ilovepi
reviewed
Dec 12, 2024
qcolombet
approved these changes
Dec 13, 2024
ilovepi
approved these changes
Dec 13, 2024
arsenm
force-pushed
the
users/arsenm/regalloc-no-fatal-error-no-available-registers
branch
from
8db7a16 to
edfb488
Compare
Base automatically changed from
users/arsenm/regalloc-no-fatal-error-no-available-registers
to
main
December 16, 2024 01:52
Greedy and fast would hit different assertions on undef uses if all registers in a class were reserved.
arsenm
force-pushed
the
users/arsenm/regalloc-fix-undef-uses-all-registers-reserved-in-class
branch
from
b1776bc to
be07817
Compare
arsenm
deleted the
users/arsenm/regalloc-fix-undef-uses-all-registers-reserved-in-class
branch
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Greedy and fast would hit different assertions on undef uses if all
registers in a class were reserved.