From: Giovanni Mascellani Subject: Re: [PATCH vkd3d v3 4/7] vkd3d-shader/hlsl: Lower matrix casts. Message-Id: <53429bd9-de36-7b10-fe6d-07561f268365@codeweavers.com> Date: Fri, 3 Jun 2022 09:08:59 +0200 In-Reply-To: References: <20220602140609.3419134-1-gmascellani@codeweavers.com> <20220602140609.3419134-4-gmascellani@codeweavers.com> Hi, Il 03/06/22 07:03, Zebediah Figura ha scritto: > With this patch we're able to successfully generate IR for the tests > added in patch 1. Unfortunately, the IR is not quite valid. It generates > SM4_RT_OUTPUT registers that are not actually declared (neither in the > output signature nor with dcl_output instructions). This seems to > confuse the DXBC->SPIRV translator, generating the following message and > producing SPIR-V that does not consistently pass or fail tests: > > fixme:vkd3d_dxbc_compiler_get_register_info: Unrecognized register > (register 0x8, 4294967295). Yes, I am aware of this problem, but it's a different bug that is merely exposed by the fact that now code generation for the affected tests is not blocked by a matrix cast. I have a couple of patches that should fix the problem, but I am not completely sure they are correct and given that freeze is imminent I would like to have the matrix patches in before that. If what bothers you is that the test is flaky, then would it be ok for you if I just commented it until the register declaration logic is fixed? Giovanni.