RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-01-31 20:51:52 +01:00
Code Issues Projects Releases Wiki Activity

AMDGPU: Fix incorrect selection of buffer atomic fadd

Browse Source 
There were additional standalone patterns for these nodes which were
missing the subtarget predicate.
This commit is contained in:
Matt Arsenault 2020-06-05 13:57:29 -04:00
parent f452018857
commit 0ca2dfd8e3
2 changed files with 21 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
lib/Target/AMDGPU/BUFInstructions.td
View File

@ -1422,8 +1422,10 @@ multiclass BufferAtomicPatterns_NO_RTN<SDPatternOperator name, ValueType vt,
>;
}
let SubtargetPredicate = HasAtomicFaddInsts in {
defm : BufferAtomicPatterns_NO_RTN<SIbuffer_atomic_fadd, f32, "BUFFER_ATOMIC_ADD_F32">;
defm : BufferAtomicPatterns_NO_RTN<SIbuffer_atomic_pk_fadd, v2f16, "BUFFER_ATOMIC_PK_ADD_F16">;
}
def : GCNPat<
(SIbuffer_atomic_cmpswap

19
test/CodeGen/AMDGPU/fail-select-buffer-atomic-fadd.ll Normal file
View File

@ -0,0 +1,19 @@
; RUN: not --crash llc -march=amdgcn -mcpu=tahiti -o /dev/null %s 2>&1 | FileCheck -check-prefix=FAIL %s
; RUN: not --crash llc -march=amdgcn -mcpu=hawaii -o /dev/null %s 2>&1 | FileCheck -check-prefix=FAIL %s
; RUN: not --crash llc -march=amdgcn -mcpu=fiji -o /dev/null %s 2>&1 | FileCheck -check-prefix=FAIL %s
; RUN: not --crash llc -march=amdgcn -mcpu=gfx900 -o /dev/null %s 2>&1 | FileCheck -check-prefix=FAIL %s
; RUN: not --crash llc -march=amdgcn -mcpu=gfx1010 -o /dev/null %s 2>&1 | FileCheck -check-prefix=FAIL %s
; Make sure selection of these intrinsics fails on targets that do not
; have the instruction available.
; FIXME: Should also really make sure the v2f16 version fails.
; FAIL: LLVM ERROR: Cannot select: {{.+}}: ch = BUFFER_ATOMIC_FADD
define amdgpu_cs void @atomic_fadd(<4 x i32> inreg %arg0) {
call void @llvm.amdgcn.buffer.atomic.fadd.f32(float 1.0, <4 x i32> %arg0, i32 0, i32 112, i1 false)
ret void
}
declare void @llvm.amdgcn.buffer.atomic.fadd.f32(float, <4 x i32>, i32, i32, i1 immarg) #0
attributes #0 = { nounwind }