| Issue |
97290
|
| Summary |
MachineVerifier should catch inconsistencies in load memory, result, and range metadata type.
|
| Labels |
good first issue,
llvm:globalisel
|
| Assignees |
|
| Reporter |
arsenm
|
The MachineVerifier does not catch cases where G_LOAD range metadata is applied to scalar memory types, with vector results and vice versa. The verifier should reject cases like this:
```
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=amdgpu-prelegalizer-combiner -verify-machineinstrs %s -o -
--- |
define void @range_metadata_sext_i8_signed_range_i64_load_as_v2i32() {
ret void
}
define void @range_metadata_sext_i8_signed_range_i64_load_as_v2i32_extractlo() {
ret void
}
define void @range_metadata_sext_i33_signed_range_i64_load_as_v2i32() {
ret void
}
!0 = !{i64 -4294967295, i64 4294967296}
!1 = !{i64 -8589934591, i64 8589934592}
...
---
name: range_metadata_sext_i8_signed_range_i64_load_as_v2i32
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0, $vgpr1
%1:_(s32) = COPY $vgpr0
%2:_(s32) = COPY $vgpr1
%0:_(p1) = G_MERGE_VALUES %1(s32), %2(s32)
%3:_(<2 x s32>) = G_LOAD %0(p1) :: (volatile load (s64), align 4, !range !0, addrspace 1)
%6:_(<2 x s32>) = G_SEXT_INREG %3, 9
$vgpr0_vgpr1 = COPY %6
SI_RETURN implicit $vgpr0_vgpr1
...
---
name: range_metadata_sext_i8_signed_range_i64_load_as_v2i32_extractlo
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0, $vgpr1
%1:_(s32) = COPY $vgpr0
%2:_(s32) = COPY $vgpr1
%0:_(p1) = G_MERGE_VALUES %1(s32), %2(s32)
%3:_(<2 x s32>) = G_LOAD %0(p1) :: (volatile load (s64), align 4, !range !0, addrspace 1)
%zero:_(s32) = G_CONSTANT i32 0
%extract:_(s32) = G_EXTRACT_VECTOR_ELT %3, %zero
%6:_(s32) = G_SEXT_INREG %zero, 9
$vgpr0 = COPY %6
SI_RETURN implicit $vgpr0, implicit $vgpr1
...
---
name: range_metadata_sext_i33_signed_range_i64_load_as_v2i32
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0, $vgpr1
%1:_(s32) = COPY $vgpr0
%2:_(s32) = COPY $vgpr1
%0:_(p1) = G_MERGE_VALUES %1(s32), %2(s32)
%3:_(<2 x s32>) = G_LOAD %0(p1) :: (volatile load (s64), align 4, !range !1, addrspace 1)
$vgpr0_vgpr1 = COPY %3
SI_RETURN implicit $vgpr0_vgpr1
...
```
Here we have an i64 range value applied to a <2 x i32> result, which should be invalid.
%3:_(<2 x s32>) = G_LOAD %0(p1) :: (volatile load (s64), align 4, !range !0, addrspace 1)
We also are not clean about ensuring we have vector element matches between the result and memory type, but I think that's a more difficult issue to fix than verifying the range metadata.
![]()
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
