[PATCH] D87032: [AMDGPU] Remove the dead spill slots while spilling FP/BP to memory
Venkata Ramanaiah Nalamothu via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Wed Sep 2 08:23:22 PDT 2020
RamNalamothu created this revision.
RamNalamothu added reviewers: kerbowa, arsenm, scott.linder.
Herald added subscribers: llvm-commits, hiraditya, t-tye, tpr, dstuttard, yaxunl, nhaehnle, jvesely, kzhuravl.
Herald added a project: LLVM.
RamNalamothu requested review of this revision.
Herald added a subscriber: wdng.
During the PEI pass, the dead TargetStackID::SGPRSpill spill slots
are not being removed while spilling the FP/BP to memory.
Fixes: SWDEV-250393
Repository:
rG LLVM Github Monorepo
https://reviews.llvm.org/D87032
Files:
llvm/lib/Target/AMDGPU/SIFrameLowering.cpp
llvm/test/CodeGen/AMDGPU/callee-frame-setup.ll
llvm/test/CodeGen/AMDGPU/stack-realign.ll
Index: llvm/test/CodeGen/AMDGPU/stack-realign.ll
===================================================================
--- llvm/test/CodeGen/AMDGPU/stack-realign.ll
+++ llvm/test/CodeGen/AMDGPU/stack-realign.ll
@@ -291,12 +291,12 @@
; GCN-LABEL: spill_bp_to_memory_scratch_reg_needed_mubuf_offset
; GCN: s_or_saveexec_b64 s[4:5], -1
; GCN: v_mov_b32_e32 v0, s33
-; GCN-NOT: v_mov_b32_e32 v0, 0x1088
-; GCN-NEXT: v_mov_b32_e32 v1, 0x1088
+; GCN-NOT: v_mov_b32_e32 v0, 0x1084
+; GCN-NEXT: v_mov_b32_e32 v1, 0x1084
; GCN-NEXT: buffer_store_dword v0, v1, s[0:3], s32 offen
; GCN: v_mov_b32_e32 v0, s34
-; GCN-NOT: v_mov_b32_e32 v0, 0x1090
-; GCN-NEXT: v_mov_b32_e32 v1, 0x1090
+; GCN-NOT: v_mov_b32_e32 v0, 0x1088
+; GCN-NEXT: v_mov_b32_e32 v1, 0x1088
; GCN-NEXT: buffer_store_dword v0, v1, s[0:3], s32 offen
%local_val = alloca i32, align 128, addrspace(5)
store volatile i32 %b, i32 addrspace(5)* %local_val, align 128
Index: llvm/test/CodeGen/AMDGPU/callee-frame-setup.ll
===================================================================
--- llvm/test/CodeGen/AMDGPU/callee-frame-setup.ll
+++ llvm/test/CodeGen/AMDGPU/callee-frame-setup.ll
@@ -529,8 +529,8 @@
; GCN-LABEL: {{^}}spill_fp_to_memory_scratch_reg_needed_mubuf_offset
; GCN: s_or_saveexec_b64 s[4:5], -1
; GCN: v_mov_b32_e32 v0, s33
-; GCN-NOT: v_mov_b32_e32 v0, 0x100c
-; GCN-NEXT: v_mov_b32_e32 v1, 0x100c
+; GCN-NOT: v_mov_b32_e32 v0, 0x1008
+; GCN-NEXT: v_mov_b32_e32 v1, 0x1008
; GCN-NEXT: buffer_store_dword v0, v1, s[0:3], s32 offen
define void @spill_fp_to_memory_scratch_reg_needed_mubuf_offset([4096 x i8] addrspace(5)* byval align 4 %arg) #3 {
%alloca = alloca i32, addrspace(5)
Index: llvm/lib/Target/AMDGPU/SIFrameLowering.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/SIFrameLowering.cpp
+++ llvm/lib/Target/AMDGPU/SIFrameLowering.cpp
@@ -112,15 +112,19 @@
// 3: There's no free lane to spill, and no free register to save FP/BP,
// so we're forced to spill another VGPR to use for the spill.
FrameIndex = NewFI;
+
+ LLVM_DEBUG(
+ auto Spill = MFI->getSGPRToVGPRSpills(NewFI).front();
+ dbgs() << (IsFP ? "FP" : "BP") << " requires fallback spill to "
+ << printReg(Spill.VGPR, TRI) << ':' << Spill.Lane << '\n';);
} else {
+ // Remove dead <NewFI> index
+ MF.getFrameInfo().RemoveStackObject(NewFI);
// 4: If all else fails, spill the FP/BP to memory.
FrameIndex = FrameInfo.CreateSpillStackObject(4, Align(4));
+ LLVM_DEBUG(dbgs() << "Reserved FI " << FrameIndex << " for spilling "
+ << (IsFP ? "FP" : "BP") << "\n");
}
-
- LLVM_DEBUG(auto Spill = MFI->getSGPRToVGPRSpills(NewFI).front();
- dbgs() << (IsFP ? "FP" : "BP") << " requires fallback spill to "
- << printReg(Spill.VGPR, TRI) << ':' << Spill.Lane
- << '\n';);
} else {
LLVM_DEBUG(dbgs() << "Saving " << (IsFP ? "FP" : "BP") << " with copy to "
<< printReg(TempSGPR, TRI) << '\n');
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D87032.289450.patch
Type: text/x-patch
Size: 3104 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200902/88d8f289/attachment.bin>
More information about the llvm-commits
mailing list