1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 18:54:02 +01:00
llvm-mirror/test/CodeGen/AMDGPU/coalescer-identical-values-undef.mir
Matt Arsenault a8a00a4e2a AMDGPU: Use SGPR_128 instead of SReg_128 for vregs
SGPR_128 only includes the real allocatable SGPRs, and SReg_128 adds
the additional non-allocatable TTMP registers. There's no point in
allocating SReg_128 vregs. This shrinks the size of the classes
regalloc needs to consider, which is usually good.

llvm-svn: 374284
2019-10-10 07:11:33 +00:00

30 lines
668 B
YAML

# RUN: llc -mtriple=amdgcn--amdpal -run-pass=simple-register-coalescing -o - %s | FileCheck %s
# Check that this doesn't crash. Check for some legitimate output.
# CHECK: S_CBRANCH_SCC1
---
name: fred
tracksRegLiveness: true
body: |
bb.0:
successors: %bb.1, %bb.2
liveins: $sgpr4
undef %0.sub2:sgpr_128 = COPY $sgpr4
%3 = IMPLICIT_DEF
S_CBRANCH_SCC1 %bb.2, implicit undef $scc
bb.1:
successors: %bb.2
%0.sub0:sgpr_128 = COPY %0.sub2
%0.sub1:sgpr_128 = COPY %0.sub2
%1:sgpr_128 = COPY %0
%2:sgpr_128 = COPY %0
%0:sgpr_128 = COPY %2
%3:sgpr_128 = COPY %1
bb.2:
$sgpr1 = COPY %3
$sgpr2 = COPY %0.sub2
...