[PATCH] D58978: AMDGPU: Preserve undef flag when expanding SI_IF
Matt Arsenault via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Tue Mar 5 09:21:09 PST 2019
arsenm created this revision.
arsenm added reviewers: nhaehnle, rampitec.
Herald added subscribers: t-tye, tpr, dstuttard, yaxunl, wdng, jvesely, kzhuravl.
Fixes undefined value verifier error.
https://reviews.llvm.org/D58978
Files:
lib/Target/AMDGPU/SILowerControlFlow.cpp
test/CodeGen/AMDGPU/si-lower-control-flow.mir
Index: test/CodeGen/AMDGPU/si-lower-control-flow.mir
===================================================================
--- test/CodeGen/AMDGPU/si-lower-control-flow.mir
+++ test/CodeGen/AMDGPU/si-lower-control-flow.mir
@@ -1,23 +1,53 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn-amd-amdhsa -run-pass=si-lower-control-flow -verify-machineinstrs %s -o - | FileCheck -check-prefixes=GCN %s
# Check that assert is not triggered
-# GCN-LABEL: name: si-lower-control-flow{{$}}
-# GCN-CHECK: S_LOAD_DWORD_IMM
-
---- |
-
- define amdgpu_kernel void @si-lower-control-flow() {
- ret void
- }
...
---
name: si-lower-control-flow
body: |
bb.0:
+ ; GCN-LABEL: name: si-lower-control-flow
+ ; GCN: [[COPY:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5
+ ; GCN: [[S_LOAD_DWORD_IMM:%[0-9]+]]:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM [[COPY]], 16, 0
+ ; GCN: [[S_AND_B32_:%[0-9]+]]:sreg_32_xm0 = S_AND_B32 [[S_LOAD_DWORD_IMM]], 255, implicit-def $scc
+ ; GCN: [[S_AND_B32_1:%[0-9]+]]:sreg_32_xm0 = S_AND_B32 65535, [[S_AND_B32_]], implicit-def $scc
+ ; GCN: S_ENDPGM
%0:sgpr_64 = COPY $sgpr4_sgpr5
%1:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %0, 16, 0
%2:sreg_32_xm0 = S_AND_B32 %1, 255, implicit-def $scc
%3:sreg_32_xm0 = S_AND_B32 65535, %2, implicit-def $scc
S_ENDPGM
...
+
+---
+name: preserve_undef_flag_si_if_src
+tracksRegLiveness: true
+body: |
+ ; GCN-LABEL: name: preserve_undef_flag_si_if_src
+ ; GCN: bb.0:
+ ; GCN: successors: %bb.1(0x40000000), %bb.2(0x40000000)
+ ; GCN: [[COPY:%[0-9]+]]:sreg_64 = COPY $exec, implicit-def $exec
+ ; GCN: [[S_AND_B64_:%[0-9]+]]:sreg_64 = S_AND_B64 [[COPY]], undef %1:sreg_64, implicit-def dead $scc
+ ; GCN: [[S_XOR_B64_:%[0-9]+]]:sreg_64 = S_XOR_B64 [[S_AND_B64_]], [[COPY]], implicit-def dead $scc
+ ; GCN: $exec = S_MOV_B64_term killed [[S_AND_B64_]]
+ ; GCN: SI_MASK_BRANCH %bb.2, implicit $exec
+ ; GCN: S_BRANCH %bb.1
+ ; GCN: bb.1:
+ ; GCN: successors: %bb.2(0x80000000)
+ ; GCN: bb.2:
+ ; GCN: S_ENDPGM
+ bb.0:
+ successors: %bb.1, %bb.2
+
+ %1:sreg_64 = SI_IF undef %0:sreg_64, %bb.2, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
+ S_BRANCH %bb.1
+
+ bb.1:
+ successors: %bb.2
+
+ bb.2:
+ S_ENDPGM
+
+...
Index: lib/Target/AMDGPU/SILowerControlFlow.cpp
===================================================================
--- lib/Target/AMDGPU/SILowerControlFlow.cpp
+++ lib/Target/AMDGPU/SILowerControlFlow.cpp
@@ -199,8 +199,8 @@
MachineInstr *And =
BuildMI(MBB, I, DL, TII->get(AMDGPU::S_AND_B64), Tmp)
.addReg(CopyReg)
- //.addReg(AMDGPU::EXEC)
- .addReg(Cond.getReg());
+ .add(Cond);
+
setImpSCCDefDead(*And, true);
MachineInstr *Xor = nullptr;
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D58978.189342.patch
Type: text/x-patch
Size: 2811 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190305/b7fb22ad/attachment.bin>
More information about the llvm-commits
mailing list