diff options
| author | Francisco Jerez <currojerez@riseup.net> | 2022-09-27 15:28:48 -0700 |
|---|---|---|
| committer | Marge Bot <emma+marge@anholt.net> | 2023-01-25 22:22:12 +0000 |
| commit | 7b5e9336298b00eeffeec738adbcc0acc7683868 (patch) | |
| tree | b233c495d93756594671be4430bd5a3da6970199 | |
| parent | 51b8abe0ba5ea2cf0ba882faa7e3e36c273db4c5 (diff) | |
intel/fs: Fix src and dst types of LOAD_PAYLOAD ACP entries during copy propagation.
The ACP entries created by copy propagation to track the implied
copies of LOAD_PAYLOAD instructions don't model the behavior of
LOAD_PAYLOAD correctly, since (as of 41868bb6824c6106a55c844) header
moves are implicitly retyped to UD and the destination of non-header
copies implicitly uses the same type as the corresponding source, even
though the ACP entries created for such copies could incorrectly
represent a type conversion, which can lead to mis-optimization of the
program.
According to Marcin, this fixes the func.mesh.ext.workgroup_id.task.q0
crucible test.
Fixes: 41868bb6824c6106a55c844 ("i965/fs: Rework the fs_visitor LOAD_PAYLOAD instruction")
Reported-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Tested-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18980>
| -rw-r--r-- | src/intel/compiler/brw_fs_copy_propagation.cpp | 6 |
1 files changed, 4 insertions, 2 deletions
diff --git a/src/intel/compiler/brw_fs_copy_propagation.cpp b/src/intel/compiler/brw_fs_copy_propagation.cpp index b4b48684898..69ceeb9bbc4 100644 --- a/src/intel/compiler/brw_fs_copy_propagation.cpp +++ b/src/intel/compiler/brw_fs_copy_propagation.cpp @@ -1108,9 +1108,11 @@ fs_visitor::opt_copy_propagation_local(void *copy_prop_ctx, bblock_t *block, if (inst->src[i].file == VGRF || (inst->src[i].file == FIXED_GRF && inst->src[i].is_contiguous())) { + const brw_reg_type t = i < inst->header_size ? + BRW_REGISTER_TYPE_UD : inst->src[i].type; acp_entry *entry = rzalloc(copy_prop_ctx, acp_entry); - entry->dst = byte_offset(inst->dst, offset); - entry->src = inst->src[i]; + entry->dst = byte_offset(retype(inst->dst, t), offset); + entry->src = retype(inst->src[i], t); entry->size_written = size_written; entry->size_read = inst->size_read(i); entry->opcode = inst->opcode; |