[PATCH 0/5] MR700: tests: Do some housekeeping for Vulkan implementations nobody loves.
From: Giovanni Mascellani <gmascellani(a)codeweavers.com> MoltenVK doesn't support the Component decoration on tessellation shaders. --- tests/d3d12.c | 1 + 1 file changed, 1 insertion(+) diff --git a/tests/d3d12.c b/tests/d3d12.c index 49b40a67a..3e75fada4 100644 --- a/tests/d3d12.c +++ b/tests/d3d12.c @@ -38139,6 +38139,7 @@ static void test_hull_shader_punned_array(void) transition_resource_state(command_list, context.render_target, D3D12_RESOURCE_STATE_RENDER_TARGET, D3D12_RESOURCE_STATE_COPY_SOURCE); + bug_if(is_mvk_device(context.device)) check_sub_resource_uint(context.render_target, 0, queue, command_list, 0xcc996633, 0); destroy_test_context(&context); -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/700
From: Giovanni Mascellani <gmascellani(a)codeweavers.com> --- tests/d3d12.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/tests/d3d12.c b/tests/d3d12.c index 3e75fada4..ce907c966 100644 --- a/tests/d3d12.c +++ b/tests/d3d12.c @@ -32383,7 +32383,10 @@ static void test_resource_allocation_info(void) ok(infos1[i].SizeInBytes == sizes[i], "Got unexpected size %"PRIu64".\n", infos1[i].SizeInBytes); if (!i) + { + vkd3d_test_pop_context(); continue; + } ok(infos1[i].Offset - infos1[i - 1].Offset >= infos1[i - 1].SizeInBytes, "Got unexpected prev size %"PRIu64", prev offset %"PRIu64", offset %"PRIu64".\n", -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/700
From: Giovanni Mascellani <gmascellani(a)codeweavers.com> Apparently MoltenVK doesn't set the SampleMask built-in to the fragment coverage mask. --- tests/d3d12.c | 1 + 1 file changed, 1 insertion(+) diff --git a/tests/d3d12.c b/tests/d3d12.c index ce907c966..7655c4e92 100644 --- a/tests/d3d12.c +++ b/tests/d3d12.c @@ -33454,6 +33454,7 @@ static void test_coverage(void) reset_command_list(command_list, context.allocator); transition_resource_state(command_list, texture, D3D12_RESOURCE_STATE_UNORDERED_ACCESS, D3D12_RESOURCE_STATE_COPY_SOURCE); + bug_if(is_mvk_device(context.device)) check_sub_resource_uint(texture, 0, queue, command_list, tests[i].sample_mask & 0xf, 0); reset_command_list(command_list, context.allocator); -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/700
From: Giovanni Mascellani <gmascellani(a)codeweavers.com> Otherwise the test fails on Intel, NVIDIA and llvmpipe. --- tests/hlsl/nointerpolation.shader_test | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tests/hlsl/nointerpolation.shader_test b/tests/hlsl/nointerpolation.shader_test index 94584ffe2..d2b7d8a46 100644 --- a/tests/hlsl/nointerpolation.shader_test +++ b/tests/hlsl/nointerpolation.shader_test @@ -144,7 +144,7 @@ shader model >= 6.0 [test] draw triangle list 3 -probe (320,240) rgba (0.250390589, 0.499088615, 0.250520796, 1.0) 1 +probe (320,240) rgba (0.250390589, 0.499088615, 0.250520796, 1.0) 4 [require] shader model >= 4.0 -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/700
From: Giovanni Mascellani <gmascellani(a)codeweavers.com> There is no guarantee that a resource is cleared upon creation. This fixes the test on NVIDIA and llvmpipe. --- tests/hlsl_d3d12.c | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/tests/hlsl_d3d12.c b/tests/hlsl_d3d12.c index b2273ffbc..b13242562 100644 --- a/tests/hlsl_d3d12.c +++ b/tests/hlsl_d3d12.c @@ -516,10 +516,16 @@ static void test_thread_id(void) for (i = 0; i < 3; ++i) { + const UINT clear_value[4] = {0}; + textures[i] = create_default_texture3d(device, 16, 8, 8, 1, DXGI_FORMAT_R32G32B32A32_UINT, D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS, D3D12_RESOURCE_STATE_UNORDERED_ACCESS); ID3D12Device_CreateUnorderedAccessView(device, textures[i], NULL, NULL, get_cpu_descriptor_handle(&context, heap, i)); + ID3D12GraphicsCommandList_ClearUnorderedAccessViewUint(command_list, + get_gpu_descriptor_handle(&context, heap, i), + get_cpu_descriptor_handle(&context, heap, i), + textures[i], clear_value, 0, NULL); } cs_code = compile_shader(cs_source, "cs_5_0"); -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/700
This merge request was approved by Giovanni Mascellani. -- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/700
``` Subject: [PATCH 5/5] tests: Clear the UAV before testing.
There is no guarantee that a resource is cleared upon creation. This fixes the test on NVIDIA and llvmpipe. ```
Does this need an UAV barrier between the clear and the dispatch? -- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/700#note_64225
On Mon Mar 11 13:57:16 2024 +0000, Henri Verbeet wrote:
``` Subject: [PATCH 5/5] tests: Clear the UAV before testing.
There is no guarantee that a resource is cleared upon creation. This fixes the test on NVIDIA and llvmpipe. ```
Does this need an UAV barrier between the clear and the dispatch? Good point. I couldn't find any definitive answer, but from https://learn.microsoft.com/en-us/windows/win32/api/d3d12/ns-d3d12-d3d12_res... it seems that barriers must always used between pairs of write operations (even if the second operation doesn't read, or at least doesn't apparently read; the page rather gives the negative version of this: you're allowed to skip the barrier between two read-only operations). I couldn't find any formal definition of what is a "write operation", so it's not clear whether a clear operation is a write operation. For the avoidance of doubt I'll pretend it is a fix the MR.
-- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/700#note_64235
participants (3)
-
Giovanni Mascellani -
Giovanni Mascellani (＠giomasce)
-
Henri Verbeet (＠hverbeet)