- Wine-Devel - list.winehq.org

[PATCH vkd3d 4/8] tests: Add a test for patch constant inputs in tessellation control shader join phases.
by Henri Verbeet March 22, 2021

March 22, 2021

Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> --- tests/d3d12.c | 233 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 233 insertions(+) diff --git a/tests/d3d12.c b/tests/d3d12.c index 3c8f6d90..dfdbefa5 100644 --- a/tests/d3d12.c +++ b/tests/d3d12.c @@ -33585,6 +33585,238 @@ static void test_hull_shader_relative_addressing(void) destroy_test_context(&context); } +static void test_hull_shader_patch_constant_inputs(void) +{ + D3D12_GRAPHICS_PIPELINE_STATE_DESC pso_desc; + ID3D12GraphicsCommandList *command_list; + D3D12_STREAM_OUTPUT_BUFFER_VIEW sobv; + D3D12_INPUT_LAYOUT_DESC input_layout; + ID3D12Resource *vb, *so_buffer; + struct test_context_desc desc; + D3D12_VERTEX_BUFFER_VIEW vbv; + struct resource_readback rb; + struct test_context context; + ID3D12CommandQueue *queue; + HRESULT hr; + +#if 0 + struct patch_constant_data + { + float2 p[3] : P; + float2 c : C; + float factor[3] : SV_TessFactor; + float inside : SV_InsideTessFactor; + }; + + struct point_data + { + float2 p[2] : P; + }; + + struct patch_constant_data patch_constant(InputPatch<struct point_data, 3> input) + { + struct patch_constant_data output; + + output.factor[0] = 1.0f; + output.factor[1] = 1.0f; + output.factor[2] = 1.0f; + output.inside = 1.0f; + + output.p[0] = input[0].p[0] + input[0].p[1]; + output.p[1] = input[1].p[0] + input[1].p[1]; + output.p[2] = input[2].p[0] + input[2].p[1]; + output.c = (output.p[0] + output.p[1] + output.p[2]) / 3; + + return output; + } + + struct point_data vs_main(struct point_data input) + { + return input; + } + + [domain("tri")] + [outputcontrolpoints(3)] + [outputtopology("triangle_cw")] + [partitioning("fractional_odd")] + [patchconstantfunc("patch_constant")] + void hs_main(InputPatch<struct point_data, 3> input) + { + } + + [domain("tri")] + float4 ds_main(struct patch_constant_data input, float3 tess_coord : SV_DomainLocation, + OutputPatch<struct point_data, 3> patch) : SV_POSITION + { + return float4(input.c, input.p[0] * tess_coord.x + input.p[1] * tess_coord.y + input.p[2] * tess_coord.z); + } +#endif + static const DWORD vs_code[] = + { + 0x43425844, 0x2b21ed94, 0x44ae97f2, 0x654a3d25, 0x437c7e3e, 0x00000001, 0x00000124, 0x00000003, + 0x0000002c, 0x00000070, 0x000000b4, 0x4e475349, 0x0000003c, 0x00000002, 0x00000008, 0x00000038, + 0x00000000, 0x00000000, 0x00000003, 0x00000000, 0x00000303, 0x00000038, 0x00000001, 0x00000000, + 0x00000003, 0x00000001, 0x00000303, 0xabab0050, 0x4e47534f, 0x0000003c, 0x00000002, 0x00000008, + 0x00000038, 0x00000000, 0x00000000, 0x00000003, 0x00000000, 0x00000c03, 0x00000038, 0x00000001, + 0x00000000, 0x00000003, 0x00000001, 0x00000c03, 0xabab0050, 0x58454853, 0x00000068, 0x00010050, + 0x0000001a, 0x0100086a, 0x0300005f, 0x00101032, 0x00000000, 0x0300005f, 0x00101032, 0x00000001, + 0x03000065, 0x00102032, 0x00000000, 0x03000065, 0x00102032, 0x00000001, 0x05000036, 0x00102032, + 0x00000000, 0x00101046, 0x00000000, 0x05000036, 0x00102032, 0x00000001, 0x00101046, 0x00000001, + 0x0100003e, + }; + static const DWORD hs_code[] = + { + 0x43425844, 0xaad1f130, 0xc203f4bd, 0x61d030b1, 0x0444e4b3, 0x00000001, 0x0000054c, 0x00000004, + 0x00000030, 0x00000074, 0x000000b8, 0x000001b0, 0x4e475349, 0x0000003c, 0x00000002, 0x00000008, + 0x00000038, 0x00000000, 0x00000000, 0x00000003, 0x00000000, 0x00000303, 0x00000038, 0x00000001, + 0x00000000, 0x00000003, 0x00000001, 0x00000303, 0xabab0050, 0x4e47534f, 0x0000003c, 0x00000002, + 0x00000008, 0x00000038, 0x00000000, 0x00000000, 0x00000003, 0x00000000, 0x00000c03, 0x00000038, + 0x00000001, 0x00000000, 0x00000003, 0x00000001, 0x00000c03, 0xabab0050, 0x47534350, 0x000000f0, + 0x00000008, 0x00000008, 0x000000c8, 0x00000000, 0x00000000, 0x00000003, 0x00000000, 0x00000c03, + 0x000000ca, 0x00000000, 0x00000000, 0x00000003, 0x00000000, 0x0000030c, 0x000000c8, 0x00000001, + 0x00000000, 0x00000003, 0x00000001, 0x00000c03, 0x000000cc, 0x00000000, 0x0000000e, 0x00000003, + 0x00000001, 0x00000b04, 0x000000c8, 0x00000002, 0x00000000, 0x00000003, 0x00000002, 0x00000c03, + 0x000000e0, 0x00000000, 0x0000000d, 0x00000003, 0x00000003, 0x00000e01, 0x000000e0, 0x00000001, + 0x0000000d, 0x00000003, 0x00000004, 0x00000e01, 0x000000e0, 0x00000002, 0x0000000d, 0x00000003, + 0x00000005, 0x00000e01, 0x00430050, 0x495f5653, 0x6469736e, 0x73655465, 0x63614673, 0x00726f74, + 0x545f5653, 0x46737365, 0x6f746361, 0xabab0072, 0x58454853, 0x00000394, 0x00030050, 0x000000e5, + 0x01000071, 0x01001893, 0x01001894, 0x01001095, 0x01001896, 0x01001897, 0x0100086a, 0x01000073, + 0x02000099, 0x00000003, 0x0200005f, 0x00017000, 0x0400005f, 0x00219012, 0x00000003, 0x00000000, + 0x0400005f, 0x00219012, 0x00000003, 0x00000001, 0x03000065, 0x00102012, 0x00000000, 0x03000065, + 0x00102012, 0x00000001, 0x03000065, 0x00102012, 0x00000002, 0x02000068, 0x00000001, 0x0400005b, + 0x00102012, 0x00000000, 0x00000003, 0x04000036, 0x00100012, 0x00000000, 0x0001700a, 0x0b000000, + 0x00100022, 0x00000000, 0x00a1900a, 0x0010000a, 0x00000000, 0x00000000, 0x00a1900a, 0x0010000a, + 0x00000000, 0x00000001, 0x06000036, 0x00902012, 0x0010000a, 0x00000000, 0x0010001a, 0x00000000, + 0x0100003e, 0x01000073, 0x02000099, 0x00000003, 0x0200005f, 0x00017000, 0x0400005f, 0x00219022, + 0x00000003, 0x00000000, 0x0400005f, 0x00219022, 0x00000003, 0x00000001, 0x03000065, 0x00102022, + 0x00000000, 0x03000065, 0x00102022, 0x00000001, 0x03000065, 0x00102022, 0x00000002, 0x02000068, + 0x00000001, 0x0400005b, 0x00102022, 0x00000000, 0x00000003, 0x04000036, 0x00100012, 0x00000000, + 0x0001700a, 0x0b000000, 0x00100022, 0x00000000, 0x00a1901a, 0x0010000a, 0x00000000, 0x00000000, + 0x00a1901a, 0x0010000a, 0x00000000, 0x00000001, 0x06000036, 0x00902022, 0x0010000a, 0x00000000, + 0x0010001a, 0x00000000, 0x0100003e, 0x01000073, 0x02000099, 0x00000003, 0x0200005f, 0x00017000, + 0x04000067, 0x00102012, 0x00000003, 0x00000011, 0x04000067, 0x00102012, 0x00000004, 0x00000012, + 0x04000067, 0x00102012, 0x00000005, 0x00000013, 0x02000068, 0x00000001, 0x0400005b, 0x00102012, + 0x00000003, 0x00000003, 0x04000036, 0x00100012, 0x00000000, 0x0001700a, 0x07000036, 0x00d02012, + 0x00000003, 0x0010000a, 0x00000000, 0x00004001, 0x3f800000, 0x0100003e, 0x01000073, 0x04000067, + 0x00102042, 0x00000001, 0x00000014, 0x05000036, 0x00102042, 0x00000001, 0x00004001, 0x3f800000, + 0x0100003e, 0x01000074, 0x0300005f, 0x0011b012, 0x00000000, 0x0300005f, 0x0011b012, 0x00000001, + 0x0300005f, 0x0011b012, 0x00000002, 0x03000065, 0x00102042, 0x00000000, 0x02000068, 0x00000001, + 0x07000000, 0x00100012, 0x00000000, 0x0011b00a, 0x00000000, 0x0011b00a, 0x00000001, 0x07000000, + 0x00100012, 0x00000000, 0x0010000a, 0x00000000, 0x0011b00a, 0x00000002, 0x07000038, 0x00102042, + 0x00000000, 0x0010000a, 0x00000000, 0x00004001, 0x3eaaaaab, 0x0100003e, 0x01000074, 0x0300005f, + 0x0011b022, 0x00000000, 0x0300005f, 0x0011b022, 0x00000001, 0x0300005f, 0x0011b022, 0x00000002, + 0x03000065, 0x00102082, 0x00000000, 0x02000068, 0x00000001, 0x07000000, 0x00100012, 0x00000000, + 0x0011b01a, 0x00000000, 0x0011b01a, 0x00000001, 0x07000000, 0x00100012, 0x00000000, 0x0010000a, + 0x00000000, 0x0011b01a, 0x00000002, 0x07000038, 0x00102082, 0x00000000, 0x0010000a, 0x00000000, + 0x00004001, 0x3eaaaaab, 0x0100003e, + }; + static const DWORD ds_code[] = + { + 0x43425844, 0xd5cd44e8, 0x94050d92, 0xd93d3a7c, 0x5c86220a, 0x00000001, 0x0000027c, 0x00000004, + 0x00000030, 0x00000074, 0x0000016c, 0x000001a0, 0x4e475349, 0x0000003c, 0x00000002, 0x00000008, + 0x00000038, 0x00000000, 0x00000000, 0x00000003, 0x00000000, 0x00000003, 0x00000038, 0x00000001, + 0x00000000, 0x00000003, 0x00000001, 0x00000003, 0xabab0050, 0x47534350, 0x000000f0, 0x00000008, + 0x00000008, 0x000000c8, 0x00000000, 0x00000000, 0x00000003, 0x00000000, 0x00000303, 0x000000ca, + 0x00000000, 0x00000000, 0x00000003, 0x00000000, 0x00000c0c, 0x000000c8, 0x00000001, 0x00000000, + 0x00000003, 0x00000001, 0x00000303, 0x000000cc, 0x00000000, 0x0000000e, 0x00000003, 0x00000001, + 0x00000004, 0x000000c8, 0x00000002, 0x00000000, 0x00000003, 0x00000002, 0x00000303, 0x000000e0, + 0x00000000, 0x0000000d, 0x00000003, 0x00000003, 0x00000001, 0x000000e0, 0x00000001, 0x0000000d, + 0x00000003, 0x00000004, 0x00000001, 0x000000e0, 0x00000002, 0x0000000d, 0x00000003, 0x00000005, + 0x00000001, 0x00430050, 0x495f5653, 0x6469736e, 0x73655465, 0x63614673, 0x00726f74, 0x545f5653, + 0x46737365, 0x6f746361, 0xabab0072, 0x4e47534f, 0x0000002c, 0x00000001, 0x00000008, 0x00000020, + 0x00000000, 0x00000001, 0x00000003, 0x00000000, 0x0000000f, 0x505f5653, 0x5449534f, 0x004e4f49, + 0x58454853, 0x000000d4, 0x00040050, 0x00000035, 0x01001893, 0x01001095, 0x0100086a, 0x0300005f, + 0x0011b032, 0x00000000, 0x0300005f, 0x0011b0c2, 0x00000000, 0x0300005f, 0x0011b032, 0x00000001, + 0x0300005f, 0x0011b032, 0x00000002, 0x0200005f, 0x0001c072, 0x04000067, 0x001020f2, 0x00000000, + 0x00000001, 0x02000068, 0x00000001, 0x06000038, 0x00100032, 0x00000000, 0x0011b046, 0x00000001, + 0x0001c556, 0x08000032, 0x00100032, 0x00000000, 0x0011b046, 0x00000000, 0x0001c006, 0x00100046, + 0x00000000, 0x08000032, 0x001020c2, 0x00000000, 0x0011b406, 0x00000002, 0x0001caa6, 0x00100406, + 0x00000000, 0x05000036, 0x00102032, 0x00000000, 0x0011bae6, 0x00000000, 0x0100003e, + }; + static const D3D12_INPUT_ELEMENT_DESC layout_desc[] = + { + {"P", 0, DXGI_FORMAT_R32G32_FLOAT, 0, 0, D3D12_INPUT_CLASSIFICATION_PER_VERTEX_DATA, 0}, + {"P", 1, DXGI_FORMAT_R32G32_FLOAT, 0, 8, D3D12_INPUT_CLASSIFICATION_PER_VERTEX_DATA, 0}, + }; + static const D3D12_SO_DECLARATION_ENTRY so_declaration[] = + { + {0, "SV_POSITION", 0, 0, 4, 0}, + }; + static const struct + { + struct vec2 p[2]; + } + vertices[] = + { + {{{ 1.0f, 2.0f}, { 64.0f, 128.0f}}}, + {{{ 4.0f, 8.0f}, { 256.0f, 512.0f}}}, + {{{16.0f, 32.0f}, {1024.0f, 2048.0f}}}, + }; + static const struct triangle expected_triangle = + {{ + {455.0f, 910.0f, 260.0f, 520.0f}, + {455.0f, 910.0f, 1040.0f, 2080.0f}, + {455.0f, 910.0f, 65.0f, 130.0f}, + }}; + unsigned int strides[] = {sizeof(struct vec4)}; + + memset(&desc, 0, sizeof(desc)); + desc.root_signature_flags = D3D12_ROOT_SIGNATURE_FLAG_ALLOW_INPUT_ASSEMBLER_INPUT_LAYOUT + | D3D12_ROOT_SIGNATURE_FLAG_ALLOW_STREAM_OUTPUT; + desc.no_pipeline = true; + if (!init_test_context(&context, &desc)) + return; + command_list = context.list; + queue = context.queue; + + input_layout.pInputElementDescs = layout_desc; + input_layout.NumElements = ARRAY_SIZE(layout_desc); + init_pipeline_state_desc(&pso_desc, context.root_signature, DXGI_FORMAT_UNKNOWN, NULL, NULL, &input_layout); + pso_desc.VS = shader_bytecode(vs_code, sizeof(vs_code)); + pso_desc.HS = shader_bytecode(hs_code, sizeof(hs_code)); + pso_desc.DS = shader_bytecode(ds_code, sizeof(ds_code)); + memset(&pso_desc.PS, 0, sizeof(pso_desc.PS)); + pso_desc.PrimitiveTopologyType = D3D12_PRIMITIVE_TOPOLOGY_TYPE_PATCH; + pso_desc.StreamOutput.NumEntries = ARRAY_SIZE(so_declaration); + pso_desc.StreamOutput.pSODeclaration = so_declaration; + pso_desc.StreamOutput.pBufferStrides = strides; + pso_desc.StreamOutput.NumStrides = ARRAY_SIZE(strides); + pso_desc.StreamOutput.RasterizedStream = D3D12_SO_NO_RASTERIZED_STREAM; + hr = ID3D12Device_CreateGraphicsPipelineState(context.device, &pso_desc, + &IID_ID3D12PipelineState, (void **)&context.pipeline_state); + ok(hr == S_OK, "Got unexpected hr %#x.\n", hr); + + vb = create_upload_buffer(context.device, sizeof(vertices), vertices); + vbv.BufferLocation = ID3D12Resource_GetGPUVirtualAddress(vb); + vbv.StrideInBytes = sizeof(*vertices); + vbv.SizeInBytes = sizeof(vertices); + + so_buffer = create_default_buffer(context.device, 4096, + D3D12_RESOURCE_FLAG_NONE, D3D12_RESOURCE_STATE_STREAM_OUT); + sobv.BufferLocation = ID3D12Resource_GetGPUVirtualAddress(so_buffer); + sobv.SizeInBytes = 1024; + sobv.BufferFilledSizeLocation = sobv.BufferLocation + sobv.SizeInBytes; + + ID3D12GraphicsCommandList_SetGraphicsRootSignature(command_list, context.root_signature); + ID3D12GraphicsCommandList_SetPipelineState(command_list, context.pipeline_state); + ID3D12GraphicsCommandList_IASetPrimitiveTopology(command_list, D3D_PRIMITIVE_TOPOLOGY_3_CONTROL_POINT_PATCHLIST); + ID3D12GraphicsCommandList_IASetVertexBuffers(command_list, 0, 1, &vbv); + ID3D12GraphicsCommandList_RSSetViewports(command_list, 1, &context.viewport); + ID3D12GraphicsCommandList_RSSetScissorRects(command_list, 1, &context.scissor_rect); + ID3D12GraphicsCommandList_SOSetTargets(command_list, 0, 1, &sobv); + ID3D12GraphicsCommandList_DrawInstanced(command_list, 3, 1, 0, 0); + + transition_resource_state(command_list, so_buffer, + D3D12_RESOURCE_STATE_STREAM_OUT, D3D12_RESOURCE_STATE_COPY_SOURCE); + get_buffer_readback_with_command_list(so_buffer, DXGI_FORMAT_UNKNOWN, &rb, queue, command_list); + todo check_triangles(&rb, &expected_triangle, 1); + release_resource_readback(&rb); + + ID3D12Resource_Release(so_buffer); + ID3D12Resource_Release(vb); + destroy_test_context(&context); +} + START_TEST(d3d12) { parse_args(argc, argv); @@ -33752,4 +33984,5 @@ START_TEST(d3d12) run_test(test_register_space); run_test(test_sampler_register_space); run_test(test_hull_shader_relative_addressing); + run_test(test_hull_shader_patch_constant_inputs); } -- 2.20.1

1 0

[PATCH vkd3d 3/8] vkd3d-shader: Use the (potentially) modified write mask as dcl_mask in vkd3d_dxbc_compiler_emit_input().
by Henri Verbeet March 22, 2021

March 22, 2021

Like outputs, when inputs are coalesced, the write mask is modified by needs_private_io_variable(). Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> --- libs/vkd3d-shader/spirv.c | 28 +++++++++------------------- 1 file changed, 9 insertions(+), 19 deletions(-) diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index d9bfa091..a639f7e8 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -4187,11 +4187,11 @@ static uint32_t vkd3d_dxbc_compiler_emit_input(struct vkd3d_dxbc_compiler *compi const struct vkd3d_shader_dst_param *dst, enum vkd3d_shader_input_sysval_semantic sysval, enum vkd3d_shader_interpolation_mode interpolation_mode) { - unsigned int component_idx, component_count, input_component_count; struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; const struct vkd3d_shader_signature_element *signature_element; const struct vkd3d_shader_signature *shader_signature; const struct vkd3d_shader_register *reg = &dst->reg; + unsigned int component_idx, input_component_count; enum vkd3d_shader_component_type component_type; uint32_t type_id, ptr_type_id, float_type_id; const struct vkd3d_spirv_builtin *builtin; @@ -4238,7 +4238,6 @@ static uint32_t vkd3d_dxbc_compiler_emit_input(struct vkd3d_dxbc_compiler *compi write_mask = signature_element->mask; - component_count = vkd3d_write_mask_component_count(dst->write_mask); if (builtin) { component_type = builtin->component_type; @@ -4252,21 +4251,11 @@ static uint32_t vkd3d_dxbc_compiler_emit_input(struct vkd3d_dxbc_compiler *compi component_idx = vkd3d_write_mask_get_component_idx(signature_element->mask); } - if ((use_private_var = builtin && builtin->fixup_pfn)) - { - component_count = VKD3D_VEC4_SIZE; - write_mask = VKD3DSP_WRITEMASK_ALL; - } - else if (needs_private_io_variable(shader_signature, reg_idx, builtin, &input_component_count, &write_mask)) - { + if (!(use_private_var = builtin && builtin->fixup_pfn) + && needs_private_io_variable(shader_signature, reg_idx, builtin, &input_component_count, &write_mask)) use_private_var = true; - component_count = VKD3D_VEC4_SIZE; - write_mask = VKD3DSP_WRITEMASK_ALL; - } else - { component_idx = vkd3d_write_mask_get_component_idx(write_mask); - } storage_class = SpvStorageClassInput; @@ -4293,7 +4282,7 @@ static uint32_t vkd3d_dxbc_compiler_emit_input(struct vkd3d_dxbc_compiler *compi } } - if (!symbol || ~symbol->info.reg.dcl_mask & signature_element->mask) + if (!symbol || ~symbol->info.reg.dcl_mask & write_mask) { if (builtin) { @@ -4326,19 +4315,20 @@ static uint32_t vkd3d_dxbc_compiler_emit_input(struct vkd3d_dxbc_compiler *compi { storage_class = SpvStorageClassPrivate; var_id = vkd3d_dxbc_compiler_emit_array_variable(compiler, &builder->global_stream, - storage_class, VKD3D_SHADER_COMPONENT_FLOAT, component_count, array_size); + storage_class, VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_VEC4_SIZE, array_size); } vkd3d_symbol_set_register_info(&reg_symbol, var_id, storage_class, - use_private_var ? VKD3D_SHADER_COMPONENT_FLOAT : component_type, write_mask); - reg_symbol.info.reg.dcl_mask |= signature_element->mask; + use_private_var ? VKD3D_SHADER_COMPONENT_FLOAT : component_type, + use_private_var ? VKD3DSP_WRITEMASK_ALL : write_mask); + reg_symbol.info.reg.dcl_mask |= write_mask; vkd3d_dxbc_compiler_put_symbol(compiler, &reg_symbol); vkd3d_dxbc_compiler_emit_register_debug_name(builder, var_id, reg); } else { - symbol->info.reg.dcl_mask |= signature_element->mask; + symbol->info.reg.dcl_mask |= write_mask; } if (use_private_var) -- 2.20.1

1 0

[PATCH vkd3d 2/8] vkd3d-shader: Use the (potentially) modified write mask as dcl_mask in vkd3d_dxbc_compiler_emit_output().
by Henri Verbeet March 22, 2021

March 22, 2021

When outputs are coalesced, the write mask is modified by needs_private_io_variable(). We want the "dcl_mask" field to reflect that as well. Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> --- libs/vkd3d-shader/spirv.c | 24 ++++++++---------------- 1 file changed, 8 insertions(+), 16 deletions(-) diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index 4598ffa5..d9bfa091 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -4702,20 +4702,14 @@ static void vkd3d_dxbc_compiler_emit_output(struct vkd3d_dxbc_compiler *compiler storage_class = SpvStorageClassOutput; - if ((use_private_variable = builtin && builtin->spirv_array_size)) - write_mask = VKD3DSP_WRITEMASK_ALL; - else if (get_shader_output_swizzle(compiler, signature_element->register_index) != VKD3D_SHADER_NO_SWIZZLE + if (!(use_private_variable = builtin && builtin->spirv_array_size) + && (get_shader_output_swizzle(compiler, signature_element->register_index) != VKD3D_SHADER_NO_SWIZZLE || needs_private_io_variable(shader_signature, signature_element->register_index, builtin, &output_component_count, &write_mask) - || is_patch_constant) - { + || is_patch_constant)) use_private_variable = true; - write_mask = VKD3DSP_WRITEMASK_ALL; - } else - { component_idx = vkd3d_write_mask_get_component_idx(write_mask); - } vkd3d_symbol_make_register(&reg_symbol, reg); @@ -4725,16 +4719,13 @@ static void vkd3d_dxbc_compiler_emit_output(struct vkd3d_dxbc_compiler *compiler id = symbol->id; } - if (!symbol || ~symbol->info.reg.dcl_mask & signature_element->mask) + if (!symbol || ~symbol->info.reg.dcl_mask & write_mask) { if (compiler->output_info[signature_idx].id) { id = compiler->output_info[signature_idx].id; if (compiler->output_info[signature_idx].array_element_mask) - { use_private_variable = true; - write_mask = VKD3DSP_WRITEMASK_ALL; - } } else if (builtin) { @@ -4795,7 +4786,8 @@ static void vkd3d_dxbc_compiler_emit_output(struct vkd3d_dxbc_compiler *compiler storage_class, VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_VEC4_SIZE); vkd3d_symbol_set_register_info(&reg_symbol, var_id, storage_class, - use_private_variable ? VKD3D_SHADER_COMPONENT_FLOAT : component_type, write_mask); + use_private_variable ? VKD3D_SHADER_COMPONENT_FLOAT : component_type, + use_private_variable ? VKD3DSP_WRITEMASK_ALL : write_mask); reg_symbol.info.reg.is_aggregate = use_private_variable ? is_patch_constant : array_size; if (!use_private_variable && is_control_point_phase(phase)) { @@ -4806,7 +4798,7 @@ static void vkd3d_dxbc_compiler_emit_output(struct vkd3d_dxbc_compiler *compiler { reg_symbol.info.reg.member_idx = reg->idx[0].offset; } - reg_symbol.info.reg.dcl_mask = signature_element->mask; + reg_symbol.info.reg.dcl_mask = write_mask; vkd3d_dxbc_compiler_put_symbol(compiler, &reg_symbol); @@ -4815,7 +4807,7 @@ static void vkd3d_dxbc_compiler_emit_output(struct vkd3d_dxbc_compiler *compiler } else { - symbol->info.reg.dcl_mask |= signature_element->mask; + symbol->info.reg.dcl_mask |= write_mask; var_id = symbol->id; } -- 2.20.1

1 0

[PATCH vkd3d 1/8] vkd3d-shader: Properly handle coalesced outputs in vkd3d_dxbc_compiler_emit_store_shader_output().
by Henri Verbeet March 22, 2021

March 22, 2021

Multiple patch constant or output signature elements may be coalesced into a single output register by needs_private_io_variable() in vkd3d_dxbc_compiler_emit_output(). When this happens, only the original element will have a corresponding variable ID assigned in output_info[]. Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> --- libs/vkd3d-shader/spirv.c | 24 ++++++++++++++++++++---- 1 file changed, 20 insertions(+), 4 deletions(-) diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index a7361aa5..4598ffa5 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -4852,19 +4852,33 @@ static uint32_t vkd3d_dxbc_compiler_get_output_array_index(struct vkd3d_dxbc_com } static void vkd3d_dxbc_compiler_emit_store_shader_output(struct vkd3d_dxbc_compiler *compiler, - const struct vkd3d_shader_signature_element *output, + const struct vkd3d_shader_signature *signature, const struct vkd3d_shader_signature_element *output, const struct vkd3d_shader_output_info *output_info, uint32_t output_index_id, uint32_t val_id, unsigned int write_mask) { unsigned int dst_write_mask, use_mask, uninit_mask, swizzle, mask; struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; uint32_t type_id, zero_id, ptr_type_id, chain_id, object_id; + const struct vkd3d_shader_signature_element *element; unsigned int i, index, array_idx; uint32_t output_id; dst_write_mask = output->mask; - write_mask &= dst_write_mask; use_mask = output->used_mask; + if (!output->sysval_semantic) + { + for (i = 0; i < signature->element_count; ++i) + { + element = &signature->elements[i]; + if (element->register_index != output->register_index) + continue; + if (element->sysval_semantic) + continue; + dst_write_mask |= element->mask; + use_mask |= element->used_mask; + } + } + write_mask &= dst_write_mask; if (!write_mask) return; @@ -4985,13 +4999,15 @@ static void vkd3d_dxbc_compiler_emit_shader_epilogue_function(struct vkd3d_dxbc_ for (i = 0; i < signature->element_count; ++i) { + if (!compiler->output_info[i].id) + continue; + variable_idx = vkd3d_dxbc_compiler_get_output_variable_index(compiler, signature->elements[i].register_index); - if (!param_id[variable_idx]) continue; - vkd3d_dxbc_compiler_emit_store_shader_output(compiler, + vkd3d_dxbc_compiler_emit_store_shader_output(compiler, signature, &signature->elements[i], &compiler->output_info[i], output_index_id, param_id[variable_idx], compiler->private_output_variable_write_mask[variable_idx]); } -- 2.20.1

1 0

[PATCH vkd3d 1/2] configure: Use VULKAN_LIBS when looking for the Vulkan library.
by Matteo Bruni March 22, 2021

March 22, 2021

Signed-off-by: Matteo Bruni <mbruni(a)codeweavers.com> --- v2: Actually preserve the contents of VULKAN_LIBS. configure.ac | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-) diff --git a/configure.ac b/configure.ac index 2005c76..5f2f719 100644 --- a/configure.ac +++ b/configure.ac @@ -105,17 +105,20 @@ AC_CHECK_LIB([dl], [dlopen], AC_ARG_VAR([PTHREAD_LIBS], [linker flags for pthreads]) VKD3D_CHECK_PTHREAD -AC_SUBST([VULKAN_LIBS]) +AC_ARG_VAR([VULKAN_LIBS], [linker flags for the Vulkan library]) VKD3D_CHECK_SONAME([vulkan], [vkGetInstanceProcAddr], - [VULKAN_LIBS="-lvulkan"], + [VULKAN_LIBS="$VULKAN_LIBS -lvulkan"], [VKD3D_CHECK_SONAME([vulkan-1], [vkGetInstanceProcAddr], - [VULKAN_LIBS="-lvulkan-1" + [VULKAN_LIBS="$VULKAN_LIBS -lvulkan-1" AC_DEFINE_UNQUOTED([SONAME_LIBVULKAN], ["$ac_cv_lib_soname_vulkan_1"])], [VKD3D_CHECK_SONAME([MoltenVK], [vkGetInstanceProcAddr], - [VULKAN_LIBS="-lMoltenVK" + [VULKAN_LIBS="$VULKAN_LIBS -lMoltenVK" AC_DEFINE_UNQUOTED([SONAME_LIBVULKAN], ["$ac_cv_lib_soname_MoltenVK"])], - [AC_MSG_ERROR([libvulkan and libMoltenVK not found.])])])]) + [AC_MSG_ERROR([libvulkan and libMoltenVK not found.])], + [$VULKAN_LIBS])], + [$VULKAN_LIBS])], + [$VULKAN_LIBS]) AS_IF([test "x$with_ncurses" != "xno"], [PKG_CHECK_MODULES([NCURSES], [ncurses], -- 2.26.2

3 5

RFC: ERROR_SXS_CANT_GEN_ACTCTX trying to run atl80:atl
by Francois Gouget March 22, 2021

March 22, 2021

Both WineTest and the TestBot fail to run the atl80 tests because CreateProcess() fails with ERROR_SXS_CANT_GEN_ACTCTX: https://test.winehq.org/data/41df83c50e1c3cfdd6e8ffb65de7838f8503632c/win10… -> see this line: atl80 load error 14001 On the TestBot a few configurations succeed: https://testbot.winehq.org/JobDetails.pl?Key=87365 * 32-bit wvistau64, w8 * 64-bit wvistau64 Before the e618fb526d06 commit, WineTest was not even trying so the current situation is be somewhat better: https://test.winehq.org/data/tests/atl80:atl.html But does anyone know how to get the atl80 tests to run? -- Francois Gouget <fgouget(a)codeweavers.com>

2 1

RFC: Missing entry points -> skip or failure?
by Francois Gouget March 22, 2021

March 22, 2021

TestLauncher now detects missing entry points. For instance: https://testbot.winehq.org/JobDetails.pl?Key=87361&f201=exe32.report#k201 rtworkq.c:0: Tests skipped: missing entry point (details below) | rtworkq_test.exe - Entry Point Not Found | The procedure entry point CoGetApartmentType could not be located in the dynamic link library ole32.dll. The two lines starting with "|" are extracted from the Windows critical error dialog and clearly show the issue. The remaining question is whether this case should really be a skip or a failure. If a dll is missing the past policy was to just skip the test so I carried on for missing entry points, ordinals, and indirectly missing dlls (when a test links with a dll which links with a missing dll). Reporting missing entry points as failures would make it more obvious that the test needs fixing to use GetProcAddress(). But in some cases, such as rtworkq:rtworkq here, that may be somewhat pointless (it would be unable to perform any of the current tests without CoGetApartmentType()). So what's the general consensus on this? See also: https://bugs.winehq.org/show_bug.cgi?id=48208 -- Francois Gouget <fgouget(a)codeweavers.com>

2 2

winegcc library creation/import broken
by Peter Dons Tychsen March 22, 2021

March 22, 2021

Hi AJ, > Because of the loader changes needed to support PE builtins, importing a > Winelib library directly at the Unix level is currently broken. It may > be reimplemented at some point, but that's not trivial. > You can still link to a library at the Unix level (as opposed to > building an import library and linking to that), but you have to tell > Wine explicitly about the loaded library. This can be done by adding a > LoadLibraryA("libtest.so") at the start of your app. Has anyone tried this recently? Are there any examples of this still working? I tried to insert LoadLibraryA("libtest.so") with success, but this did not change anything. Still the same crash. Thanks, /pedro

2 8

[tools] testbot/LogUtils: Don't include garbage in the .errors file.
by Francois Gouget March 22, 2021

March 22, 2021

When failure lines are garbled because of a race condition or a missing '\n' they often contain unwanted characters at the start of the line. If included in the .errors file these prevent the line from matching past instances of the failure (if only because of line numbers), thus causing it to be incorrectly reported as new. So only include the parts of the line matched by the error matching regular expression. Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=50491 Signed-off-by: Francois Gouget <fgouget(a)codeweavers.com> --- There are some performance implications for using things like $& and $' but that's only for old Perl versions, older than is used on the TestBot. Furthermore nowadays we only parse reports when a task completes, not every time we present the results on the website. For the following wtbsuite test: WTBS Garbled trace, ok, todo, skip sampler (advpack:advpack). --- testbot/lib/WineTestBot/LogUtils.pm | 8 ++++---- testbot/src/reporttest/report.testwtbs | 8 ++++---- 2 files changed, 8 insertions(+), 8 deletions(-) diff --git a/testbot/lib/WineTestBot/LogUtils.pm b/testbot/lib/WineTestBot/LogUtils.pm index 840a5526a..903afe495 100644 --- a/testbot/lib/WineTestBot/LogUtils.pm +++ b/testbot/lib/WineTestBot/LogUtils.pm @@ -573,7 +573,7 @@ sub ParseWineTestReport($$$) $Line =~ /^([_a-z0-9]+)\.c:\d+:[0-9.]* Test (?:failed|succeeded inside todo block): /) { _CheckUnit($LogInfo, $Cur, $1, "failure"); - _AddReportError($LogInfo, $Cur, $LineNo, $Line); + _AddReportError($LogInfo, $Cur, $LineNo, "$&$'"); $Cur->{LineFailures}++; } elsif (($Cur->{Unit} ne "" and @@ -625,7 +625,7 @@ sub ParseWineTestReport($$$) $Cur->{Pids}->{$Pid || 0} = 1; $Cur->{SummaryFailures}++; } - _AddReportError($LogInfo, $Cur, $LineNo, $Line); + _AddReportError($LogInfo, $Cur, $LineNo, "$&$'"); $Cur->{LineFailures}++; } } @@ -636,7 +636,7 @@ sub ParseWineTestReport($$$) # This also replaces a test summary line. The pid is unknown so use 0. $Cur->{Pids}->{0} = 1; $Cur->{SummaryFailures}++; - _AddReportError($LogInfo, $Cur, $LineNo, $Line); + _AddReportError($LogInfo, $Cur, $LineNo, "$&$'"); $Cur->{LineFailures}++; } } @@ -654,7 +654,7 @@ sub ParseWineTestReport($$$) $Cur->{Pids}->{$Pid || 0} = 1; $Cur->{SummaryFailures}++; } - _AddReportError($LogInfo, $Cur, $LineNo, $Line); + _AddReportError($LogInfo, $Cur, $LineNo, $&); $Cur->{LineFailures}++; } } diff --git a/testbot/src/reporttest/report.testwtbs b/testbot/src/reporttest/report.testwtbs index e1988278c..104c8ea60 100644 --- a/testbot/src/reporttest/report.testwtbs +++ b/testbot/src/reporttest/report.testwtbs @@ -50,11 +50,11 @@ n 0 comboex.c:40: Test failed: A failure in the main process n 0 comboex.c:41: Test succeeded inside todo block: Success in the main process n 0 header.c:1: Test failed: Something wrong n 0 1234:rebar: unhandled exception c0000005 at 0040167C -n 0 garbled0123:status: unhandled exception c0000005 at 0040167C -n 0 status.c:33: unhandled exception c0000005 in child process 0123dead +n 0 0123:status: unhandled exception c0000005 at 0040167C +n 0 status.c:33: unhandled exception c0000005 in child process 0123 n 0 1234:subclass: unhandled exception c0000005 at 0040167C -n 0 Do not cut syslink.c:40: Test failed: Something wrong -n 0 Do not cut syslink.c:41: Test succeeded inside todo block: Something right! +n 0 syslink.c:40: Test failed: Something wrong +n 0 syslink.c:41: Test succeeded inside todo block: Something right! n 0 Unhandled exception: page fault on write access to 0x00000000 in 32-bit code (0x0054376d). n 0 1234:taskdialog: unhandled exception c0000005 at 0040167C n 0 taskdialog.c:471: unhandled exception c0000005 in child process 1234 -- 2.20.1

1 0

[PATCH 1/4] wineboot: Initialize user shared data Qpc flags and frequency.
by Rémi Bernon March 22, 2021

March 22, 2021

Signed-off-by: Rémi Bernon <rbernon(a)codeweavers.com> --- This is probably more an RFC but I don't expect it to be merged right away anyway. It fixes the high overhead (Rtl)QueryPerformanceCounter now have with the new syscall frames. There's a few things I'm still not completely confident with: * The reported Qpc frequency now changes from the default 10MHz. This is also the case on some Windows version (since they introduced QpcBypass but before they introduced the hypervisor page), and applications shouldn't rely on a fixed frequency, but we can expect that some do. Since the hypervisor page, the reported frequency is 10MHz again, and the divider is stored in the new shared page. We could very well implement the same thing, but it required the new shared memory page to be introduced. * The rdtsc(p) frequency calibration done when we cannot get the information from cpuid (on AMD and older Intel CPU) is still a little bit imprecise, and I'm not sure how to make it better while keeping the calibration time small. On my computer I can see a ±100KHz error for a nominal 3.6GHz frequency. I believe the Linux kernel also does some estimation there, but it has access to hardware timers that may help to get a better approximation. programs/wineboot/wineboot.c | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+) diff --git a/programs/wineboot/wineboot.c b/programs/wineboot/wineboot.c index 9427448b612..9e36b3c22dd 100644 --- a/programs/wineboot/wineboot.c +++ b/programs/wineboot/wineboot.c @@ -82,6 +82,8 @@ WINE_DEFAULT_DEBUG_CHANNEL(wineboot); +#define TICKSPERSEC 10000000 + extern BOOL shutdown_close_windows( BOOL force ); extern BOOL shutdown_all_desktops( BOOL force ); extern void kill_processes( BOOL kill_desktop ); @@ -241,12 +243,28 @@ static void initialize_xstate_features(struct _KUSER_SHARED_DATA *data) TRACE("XSAVE feature 2 %#x, %#x, %#x, %#x.\n", regs[0], regs[1], regs[2], regs[3]); } +static void initialize_qpc_features(struct _KUSER_SHARED_DATA *data) +{ + data->QpcBypassEnabled = 0; + data->QpcFrequency = TICKSPERSEC; + data->QpcShift = 0; + data->QpcBias = 0; +} + #else static void initialize_xstate_features(struct _KUSER_SHARED_DATA *data) { } +static void initialize_qpc_features(struct _KUSER_SHARED_DATA *data) +{ + data->QpcBypassEnabled = 0; + data->QpcFrequency = TICKSPERSEC; + data->QpcShift = 0; + data->QpcBias = 0; +} + #endif static void create_user_shared_data(void) @@ -336,6 +354,7 @@ static void create_user_shared_data(void) data->ActiveGroupCount = 1; initialize_xstate_features( data ); + initialize_qpc_features( data ); UnmapViewOfFile( data ); } -- 2.30.2

1 3