RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 03:02:36 +01:00
Code Issues Projects Releases Wiki Activity

[AMDGPU] Disallow dpp combining for dpp instructions without Src2 operand (when Src2 is required)

Browse Source 
Differential revision: https://reviews.llvm.org/D69430
This commit is contained in:
vpykhtin 2019-10-25 20:42:52 +03:00
parent 7245746772
commit 7d2c3bcc82
2 changed files with 18 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
lib/Target/AMDGPU/GCNDPPCombine.cpp
View File

@ -235,7 +235,8 @@ MachineInstr *GCNDPPCombine::createDPPInst(MachineInstr &OrigMI,
}
if (auto *Src2 = TII->getNamedOperand(OrigMI, AMDGPU::OpName::src2)) {
if (!TII->isOperandLegal(*DPPInst.getInstr(), NumOperands, Src2)) {
if (!TII->getNamedOperand(*DPPInst.getInstr(), AMDGPU::OpName::src2) ||
!TII->isOperandLegal(*DPPInst.getInstr(), NumOperands, Src2)) {
LLVM_DEBUG(dbgs() << " failed: src2 is illegal\n");
Fail = true;
break;

16
test/CodeGen/AMDGPU/dpp_combine.mir
View File

@ -794,3 +794,19 @@ body: |
%4:vgpr_32 = V_ADD_I32_e32 %2.sub0, undef %3:vgpr_32, implicit-def $vcc, implicit $exec
%5:vgpr_32, dead %6:sreg_64_xexec = V_ADDC_U32_e64 1, %2.sub1, undef $vcc, 0, implicit $exec
...
# GCN-LABEL: name: dont_combine_cndmask_with_src2
# GCN: %5:vgpr_32 = V_CNDMASK_B32_e64 0, %3, 0, %1, %4, implicit $exec
name: dont_combine_cndmask_with_src2
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0, $vgpr1
%0:vgpr_32 = COPY $vgpr0
%1:vgpr_32 = COPY $vgpr1
%2:vgpr_32 = IMPLICIT_DEF
%3:vgpr_32 = V_MOV_B32_dpp %2, %0, 1, 15, 15, 1, implicit $exec
%4:sreg_64_xexec = IMPLICIT_DEF
%5:vgpr_32 = V_CNDMASK_B32_e64 0, %3, 0, %1, %4, implicit $exec
...