Correctly report liveness of variables passed as function parameters #267
Conversation
|
|
| const uint32_t result_index = p_node->word_offset + 2; | ||
| for (uint32_t j = 0, parameter_count = p_node->word_count - 4; j < parameter_count; j++) { | ||
| const uint32_t ptr_index = p_node->word_offset + 4 + j; | ||
| SpvReflectPrvAccessedVariable* access_ptr = &p_func->accessed_variables[p_func->accessed_variable_count]; |
There was a problem hiding this comment.
Does access_ptr indicate that it's accessing a pointer or is the _ptr suffix to indicate that this is variable is a pointer? If it's the latter, then it should be p_access instead.
There was a problem hiding this comment.
This line was copied and pasted from the SpvOpLoad..SpvOpImageTexelPointer case below (see line 1204). I can change it on this line, but perhaps it's better to fix both lines at once in a separate PR?
There was a problem hiding this comment.
Ahh, apologies I didn't see that. You're right, that should be handled in a separate PR, not critical.
daniel-story
force-pushed
the
function-parameter-liveness
branch
from
a22e23a
to
16ac489
Compare
| @@ -101,7 +101,7 @@ all_descriptor_bindings: | |||
| accessed: 1 | |||
| uav_counter_id: 4294967295 | |||
| uav_counter_binding: | |||
| ByteAddressBuffer offsets: [4, 5, 11, 13] | |||
| ByteAddressBuffer offsets: [13, 5, 11, 4] | |||
There was a problem hiding this comment.
little concerned why this was flipped here, but not in user_type/byte_address_buffer_3.spv.yaml
There was a problem hiding this comment.
less concern because the order wasn't important/required
Closes #102