February 2019 - Wine-devel - list.winehq.org

[PATCH vkd3d 8/8] vkd3d-shader: Handle VKD3DSPR_INCONTROLPOINT in vkd3d_dxbc_compiler_get_register_id().
by Józef Kucia 20 Feb '19

20 Feb '19

From: Józef Kucia <jkucia(a)codeweavers.com> Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> --- libs/vkd3d-shader/spirv.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-) diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index 52c22fee33cc..2b52334dc5e4 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -2604,21 +2604,22 @@ static uint32_t vkd3d_dxbc_compiler_get_register_id(struct vkd3d_dxbc_compiler * switch (reg->type) { - case VKD3DSPR_TEMP: - case VKD3DSPR_IDXTEMP: - case VKD3DSPR_INPUT: - case VKD3DSPR_OUTPUT: case VKD3DSPR_COLOROUT: + case VKD3DSPR_CONSTBUFFER: case VKD3DSPR_DEPTHOUT: case VKD3DSPR_DEPTHOUTGE: case VKD3DSPR_DEPTHOUTLE: - case VKD3DSPR_CONSTBUFFER: + case VKD3DSPR_IDXTEMP: case VKD3DSPR_IMMCONSTBUFFER: - case VKD3DSPR_SAMPLER: - case VKD3DSPR_THREADID: + case VKD3DSPR_INCONTROLPOINT: + case VKD3DSPR_INPUT: case VKD3DSPR_LOCALTHREADID: case VKD3DSPR_LOCALTHREADINDEX: + case VKD3DSPR_OUTPUT: + case VKD3DSPR_SAMPLER: + case VKD3DSPR_TEMP: case VKD3DSPR_THREADGROUPID: + case VKD3DSPR_THREADID: if (vkd3d_dxbc_compiler_get_register_info(compiler, reg, &register_info)) { vkd3d_dxbc_compiler_emit_dereference_register(compiler, reg, &register_info); -- 2.19.2

2 1

[PATCH vkd3d 7/8] tests: Add another hull shader to NOP tessellation shaders test.
by Józef Kucia 20 Feb '19

20 Feb '19

From: Józef Kucia <jkucia(a)codeweavers.com> Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> --- tests/d3d12.c | 78 ++++++++++++++++++++++++++++++++++++++++++--------- 1 file changed, 65 insertions(+), 13 deletions(-) diff --git a/tests/d3d12.c b/tests/d3d12.c index 5a4e90b76b35..4e42000905dd 100644 --- a/tests/d3d12.c +++ b/tests/d3d12.c @@ -21838,6 +21838,7 @@ static void test_nop_tessellation_shaders(void) struct test_context context; ID3D12CommandQueue *queue; struct vec4 tess_factors; + unsigned int i; HRESULT hr; static const DWORD hs_cb_code[] = @@ -21912,6 +21913,16 @@ static void test_nop_tessellation_shaders(void) output.inside = 1.0f; } + [domain("tri")] + [outputcontrolpoints(3)] + [partitioning("integer")] + [outputtopology("triangle_cw")] + [patchconstantfunc("patch_constant")] + data hs_main(InputPatch<data, 3> input, uint i : SV_OutputControlPointID) + { + return input[i]; + } + [domain("tri")] void ds_main(patch_constant_data input, float3 tess_coord : SV_DomainLocation, @@ -21923,6 +21934,27 @@ static void test_nop_tessellation_shaders(void) + tess_coord.z * patch[2].position; } #endif + static const DWORD hs_code[] = + { + 0x43425844, 0x0e9a8861, 0x39351e76, 0x0e10883f, 0x6054b5a1, 0x00000001, 0x0000020c, 0x00000004, + 0x00000030, 0x00000064, 0x00000098, 0x0000012c, 0x4e475349, 0x0000002c, 0x00000001, 0x00000008, + 0x00000020, 0x00000000, 0x00000001, 0x00000003, 0x00000000, 0x00000f0f, 0x505f5653, 0x7469736f, + 0x006e6f69, 0x4e47534f, 0x0000002c, 0x00000001, 0x00000008, 0x00000020, 0x00000000, 0x00000001, + 0x00000003, 0x00000000, 0x0000000f, 0x505f5653, 0x7469736f, 0x006e6f69, 0x47534350, 0x0000008c, + 0x00000004, 0x00000008, 0x00000068, 0x00000000, 0x0000000d, 0x00000003, 0x00000000, 0x00000e01, + 0x00000068, 0x00000001, 0x0000000d, 0x00000003, 0x00000001, 0x00000e01, 0x00000068, 0x00000002, + 0x0000000d, 0x00000003, 0x00000002, 0x00000e01, 0x00000076, 0x00000000, 0x0000000e, 0x00000003, + 0x00000003, 0x00000e01, 0x545f5653, 0x46737365, 0x6f746361, 0x56530072, 0x736e495f, 0x54656469, + 0x46737365, 0x6f746361, 0xabab0072, 0x58454853, 0x000000d8, 0x00030050, 0x00000036, 0x01000071, + 0x01001893, 0x01001894, 0x01001095, 0x01000896, 0x01001897, 0x0100086a, 0x01000073, 0x02000099, + 0x00000003, 0x0200005f, 0x00017000, 0x04000067, 0x00102012, 0x00000000, 0x00000011, 0x04000067, + 0x00102012, 0x00000001, 0x00000012, 0x04000067, 0x00102012, 0x00000002, 0x00000013, 0x02000068, + 0x00000001, 0x0400005b, 0x00102012, 0x00000000, 0x00000003, 0x04000036, 0x00100012, 0x00000000, + 0x0001700a, 0x06000036, 0x00902012, 0x0010000a, 0x00000000, 0x00004001, 0x3f800000, 0x0100003e, + 0x01000073, 0x04000067, 0x00102012, 0x00000003, 0x00000014, 0x05000036, 0x00102012, 0x00000003, + 0x00004001, 0x3f800000, 0x0100003e, + }; + static const D3D12_SHADER_BYTECODE hs = {hs_code, sizeof(hs_code)}; static const DWORD ds_code[] = { 0x43425844, 0x8ed11021, 0x414dff74, 0x426849eb, 0x312f4860, 0x00000001, 0x000001e0, 0x00000004, @@ -21942,6 +21974,7 @@ static void test_nop_tessellation_shaders(void) 0x00000000, 0x0001caa6, 0x00219e46, 0x00000002, 0x00000000, 0x00100e46, 0x00000000, 0x0100003e, }; static const D3D12_SHADER_BYTECODE ds = {ds_code, sizeof(ds_code)}; + static const D3D12_SHADER_BYTECODE *hull_shaders[] = {&hs_cb, &hs}; memset(&desc, 0, sizeof(desc)); desc.no_root_signature = true; @@ -21964,23 +21997,42 @@ static void test_nop_tessellation_shaders(void) hr = ID3D12Device_CreateGraphicsPipelineState(context.device, &pso_desc, &IID_ID3D12PipelineState, (void **)&context.pipeline_state); ok(hr == S_OK, "Failed to create state, hr %#x.\n", hr); + ID3D12PipelineState_Release(context.pipeline_state); - ID3D12GraphicsCommandList_ClearRenderTargetView(command_list, context.rtv, white, 0, NULL); + for (i = 0; i < ARRAY_SIZE(hull_shaders); ++i) + { + vkd3d_test_set_context("Test %u", i); - ID3D12GraphicsCommandList_OMSetRenderTargets(command_list, 1, &context.rtv, FALSE, NULL); - 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_RSSetViewports(command_list, 1, &context.viewport); - ID3D12GraphicsCommandList_RSSetScissorRects(command_list, 1, &context.scissor_rect); - tess_factors.x = tess_factors.y = tess_factors.z = tess_factors.w = 1.0f; - ID3D12GraphicsCommandList_SetGraphicsRoot32BitConstants(command_list, 0, 4, &tess_factors.x, 0); - ID3D12GraphicsCommandList_DrawInstanced(command_list, 3, 1, 0, 0); + pso_desc.HS = *hull_shaders[i]; + hr = ID3D12Device_CreateGraphicsPipelineState(context.device, &pso_desc, + &IID_ID3D12PipelineState, (void **)&context.pipeline_state); + ok(hr == S_OK, "Failed to create state, hr %#x.\n", hr); - transition_resource_state(command_list, context.render_target, - D3D12_RESOURCE_STATE_RENDER_TARGET, D3D12_RESOURCE_STATE_COPY_SOURCE); + ID3D12GraphicsCommandList_ClearRenderTargetView(command_list, context.rtv, white, 0, NULL); - check_sub_resource_uint(context.render_target, 0, queue, command_list, 0xff00ff00, 0); + ID3D12GraphicsCommandList_OMSetRenderTargets(command_list, 1, &context.rtv, FALSE, NULL); + 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_RSSetViewports(command_list, 1, &context.viewport); + ID3D12GraphicsCommandList_RSSetScissorRects(command_list, 1, &context.scissor_rect); + tess_factors.x = tess_factors.y = tess_factors.z = tess_factors.w = 1.0f; + ID3D12GraphicsCommandList_SetGraphicsRoot32BitConstants(command_list, 0, 4, &tess_factors.x, 0); + ID3D12GraphicsCommandList_DrawInstanced(command_list, 3, 1, 0, 0); + + transition_resource_state(command_list, context.render_target, + D3D12_RESOURCE_STATE_RENDER_TARGET, D3D12_RESOURCE_STATE_COPY_SOURCE); + + check_sub_resource_uint(context.render_target, 0, queue, command_list, 0xff00ff00, 0); + + reset_command_list(command_list, context.allocator); + transition_resource_state(command_list, context.render_target, + D3D12_RESOURCE_STATE_COPY_SOURCE, D3D12_RESOURCE_STATE_RENDER_TARGET); + + ID3D12PipelineState_Release(context.pipeline_state); + context.pipeline_state = NULL; + } + vkd3d_test_set_context(NULL); destroy_test_context(&context); } -- 2.19.2

2 1

[PATCH vkd3d 6/8] vkd3d-shader: Add basic support for dcl_index_range.
by Józef Kucia 20 Feb '19

20 Feb '19

From: Józef Kucia <jkucia(a)codeweavers.com> We can address SPIR-V arrays dynamically. Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> --- libs/vkd3d-shader/dxbc.c | 4 +- libs/vkd3d-shader/spirv.c | 74 ++++++++++++++++++++++-- libs/vkd3d-shader/trace.c | 4 +- libs/vkd3d-shader/vkd3d_shader_private.h | 4 +- 4 files changed, 76 insertions(+), 10 deletions(-) diff --git a/libs/vkd3d-shader/dxbc.c b/libs/vkd3d-shader/dxbc.c index 49116681dbab..7cc11e7586ac 100644 --- a/libs/vkd3d-shader/dxbc.c +++ b/libs/vkd3d-shader/dxbc.c @@ -619,8 +619,8 @@ static void shader_sm4_read_dcl_index_range(struct vkd3d_shader_instruction *ins struct vkd3d_sm4_data *priv) { shader_sm4_read_dst_param(priv, &tokens, &tokens[token_count], VKD3D_DATA_OPAQUE, - &ins->declaration.index_range.first_register); - ins->declaration.index_range.last_register = *tokens; + &ins->declaration.index_range.dst); + ins->declaration.index_range.register_count = *tokens; } static void shader_sm4_read_dcl_output_topology(struct vkd3d_shader_instruction *ins, diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index afed8a27050a..52c22fee33cc 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -2568,14 +2568,17 @@ static void vkd3d_dxbc_compiler_emit_dereference_register(struct vkd3d_dxbc_comp } else if (register_info->is_aggregate) { - if (reg->idx[0].rel_addr || reg->idx[1].rel_addr) + if (reg->idx[1].rel_addr) FIXME("Relative addressing not implemented.\n"); - indexes[index_count++] = vkd3d_dxbc_compiler_get_constant_uint(compiler, register_info->member_idx); + if (reg->idx[0].rel_addr) + indexes[index_count++] = vkd3d_dxbc_compiler_emit_register_addressing(compiler, &reg->idx[0]); + else + indexes[index_count++] = vkd3d_dxbc_compiler_get_constant_uint(compiler, register_info->member_idx); } else { - if (reg->idx[1].rel_addr) + if (reg->idx[1].rel_addr || (reg->idx[1].offset == ~0u && reg->idx[0].rel_addr)) FIXME("Relative addressing not implemented.\n"); /* Handle arrayed registers, e.g. v[3][0]. */ @@ -4754,6 +4757,64 @@ static void vkd3d_dxbc_compiler_emit_dcl_output_siv(struct vkd3d_dxbc_compiler * vkd3d_dxbc_compiler_emit_output(compiler, dst, sysval); } +static bool vkd3d_dxbc_compiler_check_index_range(struct vkd3d_dxbc_compiler *compiler, + const struct vkd3d_shader_index_range *range) +{ + const struct vkd3d_shader_register *reg = &range->dst.reg; + struct vkd3d_shader_register_info reg_info; + struct vkd3d_shader_register current_reg; + struct vkd3d_symbol reg_symbol; + unsigned int i; + uint32_t id; + + current_reg = *reg; + vkd3d_symbol_make_register(&reg_symbol, &current_reg); + if (!vkd3d_dxbc_compiler_get_register_info(compiler, &current_reg, &reg_info)) + { + ERR("Failed to get register info.\n"); + return false; + } + + /* FIXME: We should check if it's an array. */ + if (!reg_info.is_aggregate) + { + FIXME("Unhandled register %#x.\n", reg->type); + return false; + } + id = reg_info.id; + + for (i = reg->idx[0].offset; i < reg->idx[0].offset + range->register_count; ++i) + { + current_reg.idx[0].offset = i; + vkd3d_symbol_make_register(&reg_symbol, &current_reg); + + if (range->dst.write_mask != reg_info.write_mask + || vkd3d_write_mask_component_count(reg_info.write_mask) != 1) + { + FIXME("Unhandled index range write mask %#x (%#x).\n", + range->dst.write_mask, reg_info.write_mask); + return false; + } + + if (reg_info.id != id) + { + FIXME("Unhandled index range %#x, %u.\n", reg->type, i); + return false; + } + } + + return true; +} + +static void vkd3d_dxbc_compiler_emit_dcl_index_range(struct vkd3d_dxbc_compiler *compiler, + const struct vkd3d_shader_instruction *instruction) +{ + const struct vkd3d_shader_index_range *range = &instruction->declaration.index_range; + + if (!vkd3d_dxbc_compiler_check_index_range(compiler, range)) + FIXME("Ignoring dcl_index_range %#x %u.\n", range->dst.reg.type, range->register_count); +} + static void vkd3d_dxbc_compiler_emit_dcl_stream(struct vkd3d_dxbc_compiler *compiler, const struct vkd3d_shader_instruction *instruction) { @@ -7388,6 +7449,9 @@ int vkd3d_dxbc_compiler_handle_instruction(struct vkd3d_dxbc_compiler *compiler, case VKD3DSIH_DCL_OUTPUT_SIV: vkd3d_dxbc_compiler_emit_dcl_output_siv(compiler, instruction); break; + case VKD3DSIH_DCL_INDEX_RANGE: + vkd3d_dxbc_compiler_emit_dcl_index_range(compiler, instruction); + break; case VKD3DSIH_DCL_STREAM: vkd3d_dxbc_compiler_emit_dcl_stream(compiler, instruction); break; @@ -7730,7 +7794,6 @@ static void vkd3d_dxbc_compiler_emit_shader_epilogue_function(struct vkd3d_dxbc_ vkd3d_spirv_build_op_function(builder, void_id, function_id, SpvFunctionControlMaskNone, function_type_id); - vkd3d_spirv_build_op_name(builder, function_id, "epilogue"); for (i = 0; i < ARRAY_SIZE(compiler->private_output_variable); ++i) { @@ -7789,7 +7852,10 @@ int vkd3d_dxbc_compiler_generate_spirv(struct vkd3d_dxbc_compiler *compiler, } if (compiler->epilogue_function_id) + { + vkd3d_spirv_build_op_name(builder, compiler->epilogue_function_id, "epilogue"); vkd3d_dxbc_compiler_emit_shader_epilogue_function(compiler); + } if (compiler->options & VKD3D_SHADER_STRIP_DEBUG) vkd3d_spirv_stream_clear(&builder->debug_stream); diff --git a/libs/vkd3d-shader/trace.c b/libs/vkd3d-shader/trace.c index 6375e714c1fc..bb0e6ad5a3d0 100644 --- a/libs/vkd3d-shader/trace.c +++ b/libs/vkd3d-shader/trace.c @@ -1389,8 +1389,8 @@ static void shader_dump_instruction(struct vkd3d_string_buffer *buffer, case VKD3DSIH_DCL_INDEX_RANGE: shader_addline(buffer, "%s ", shader_opcode_names[ins->handler_idx]); - shader_dump_dst_param(buffer, &ins->declaration.index_range.first_register, shader_version); - shader_addline(buffer, " %u", ins->declaration.index_range.last_register); + shader_dump_dst_param(buffer, &ins->declaration.index_range.dst, shader_version); + shader_addline(buffer, " %u", ins->declaration.index_range.register_count); break; case VKD3DSIH_DCL_INDEXABLE_TEMP: diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index 6b095a9cbba6..7ffcdd9c5e77 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -584,8 +584,8 @@ struct vkd3d_shader_src_param struct vkd3d_shader_index_range { - struct vkd3d_shader_dst_param first_register; - unsigned int last_register; + struct vkd3d_shader_dst_param dst; + unsigned int register_count; }; enum vkd3d_decl_usage -- 2.19.2

2 1

[PATCH vkd3d 5/8] tests: Add test for quad tessellation.
by Józef Kucia 20 Feb '19

20 Feb '19

From: Józef Kucia <jkucia(a)codeweavers.com> Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> --- tests/d3d12.c | 488 +++++++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 485 insertions(+), 3 deletions(-) diff --git a/tests/d3d12.c b/tests/d3d12.c index e83ea48fdb6a..5a4e90b76b35 100644 --- a/tests/d3d12.c +++ b/tests/d3d12.c @@ -658,10 +658,10 @@ static ID3D12RootSignature *create_cb_root_signature_(unsigned int line, } #define create_32bit_constants_root_signature(a, b, c, e) \ - create_32bit_constants_root_signature_(__LINE__, a, b, c, e) + create_32bit_constants_root_signature_(__LINE__, a, b, c, e, 0) static ID3D12RootSignature *create_32bit_constants_root_signature_(unsigned int line, ID3D12Device *device, unsigned int reg_idx, unsigned int element_count, - D3D12_SHADER_VISIBILITY shader_visibility) + D3D12_SHADER_VISIBILITY shader_visibility, D3D12_ROOT_SIGNATURE_FLAGS flags) { D3D12_ROOT_SIGNATURE_DESC root_signature_desc; ID3D12RootSignature *root_signature = NULL; @@ -677,7 +677,7 @@ static ID3D12RootSignature *create_32bit_constants_root_signature_(unsigned int memset(&root_signature_desc, 0, sizeof(root_signature_desc)); root_signature_desc.NumParameters = 1; root_signature_desc.pParameters = &root_parameter; - root_signature_desc.Flags = 0; + root_signature_desc.Flags = flags; hr = create_root_signature(device, &root_signature_desc, &root_signature); ok_(line)(SUCCEEDED(hr), "Failed to create root signature, hr %#x.\n", hr); @@ -21985,6 +21985,487 @@ static void test_nop_tessellation_shaders(void) destroy_test_context(&context); } +struct triangle +{ + struct vec4 v[3]; +}; + +#define check_triangles(a, b, c, d, e) check_triangles_(__LINE__, a, b, c, d, e) +static void check_triangles_(unsigned int line, ID3D12Resource *buffer, + ID3D12CommandQueue *queue, ID3D12GraphicsCommandList *command_list, + const struct triangle *triangles, unsigned int triangle_count) +{ + const struct triangle *current, *expected; + struct resource_readback rb; + unsigned int i, j, offset; + bool all_match = true; + + get_buffer_readback_with_command_list(buffer, DXGI_FORMAT_UNKNOWN, &rb, queue, command_list); + + for (i = 0; i < triangle_count; ++i) + { + current = get_readback_data(&rb, i, 0, 0, sizeof(*current)); + expected = &triangles[i]; + + offset = ~0u; + for (j = 0; j < ARRAY_SIZE(expected->v); ++j) + { + if (compare_vec4(&current->v[0], &expected->v[j], 0)) + { + offset = j; + break; + } + } + + if (offset == ~0u) + { + all_match = false; + break; + } + + for (j = 0; j < ARRAY_SIZE(expected->v); ++j) + { + if (!compare_vec4(&current->v[j], &expected->v[(j + offset) % 3], 0)) + { + all_match = false; + break; + } + } + if (!all_match) + break; + } + + ok_(line)(all_match, "Triangle %u vertices {%.8e, %.8e, %.8e, %.8e}, " + "{%.8e, %.8e, %.8e, %.8e}, {%.8e, %.8e, %.8e, %.8e} " + "do not match {%.8e, %.8e, %.8e, %.8e}, {%.8e, %.8e, %.8e, %.8e}, " + "{%.8e, %.8e, %.8e, %.8e}.\n", i, + current->v[0].x, current->v[0].y, current->v[0].z, current->v[0].w, + current->v[1].x, current->v[1].y, current->v[1].z, current->v[1].w, + current->v[2].x, current->v[2].y, current->v[2].z, current->v[2].w, + expected->v[0].x, expected->v[0].y, expected->v[0].z, expected->v[0].w, + expected->v[1].x, expected->v[1].y, expected->v[1].z, expected->v[1].w, + expected->v[2].x, expected->v[2].y, expected->v[2].z, expected->v[2].w); + + release_resource_readback(&rb); +} + +static void test_quad_tessellation(void) +{ + static const DWORD vs_code[] = + { +#if 0 + void main(float4 in_position : POSITION, out float4 out_position : SV_POSITION) + { + out_position = in_position; + } +#endif + 0x43425844, 0xa7a2f22d, 0x83ff2560, 0xe61638bd, 0x87e3ce90, 0x00000001, 0x000000d8, 0x00000003, + 0x0000002c, 0x00000060, 0x00000094, 0x4e475349, 0x0000002c, 0x00000001, 0x00000008, 0x00000020, + 0x00000000, 0x00000000, 0x00000003, 0x00000000, 0x00000f0f, 0x49534f50, 0x4e4f4954, 0xababab00, + 0x4e47534f, 0x0000002c, 0x00000001, 0x00000008, 0x00000020, 0x00000000, 0x00000001, 0x00000003, + 0x00000000, 0x0000000f, 0x505f5653, 0x5449534f, 0x004e4f49, 0x52444853, 0x0000003c, 0x00010040, + 0x0000000f, 0x0300005f, 0x001010f2, 0x00000000, 0x04000067, 0x001020f2, 0x00000000, 0x00000001, + 0x05000036, 0x001020f2, 0x00000000, 0x00101e46, 0x00000000, 0x0100003e, + }; + static const D3D12_SHADER_BYTECODE vs = {vs_code, sizeof(vs_code)}; +#if 0 + struct point_data + { + float4 position : SV_POSITION; + }; + + struct patch_constant_data + { + float edges[4] : SV_TessFactor; + float inside[2] : SV_InsideTessFactor; + }; + + float4 tess_factors; + float2 inside_tess_factors; + + patch_constant_data patch_constant(InputPatch<point_data, 4> input) + { + patch_constant_data output; + + output.edges[0] = tess_factors.x; + output.edges[1] = tess_factors.y; + output.edges[2] = tess_factors.z; + output.edges[3] = tess_factors.w; + output.inside[0] = inside_tess_factors.x; + output.inside[1] = inside_tess_factors.y; + + return output; + } + + [domain("quad")] + [outputcontrolpoints(4)] + [outputtopology("triangle_ccw")] + [partitioning("integer")] + [patchconstantfunc("patch_constant")] + point_data hs_main(InputPatch<point_data, 4> input, + uint i : SV_OutputControlPointID) + { + return input[i]; + } + + [domain("quad")] + point_data ds_main(patch_constant_data input, + float2 tess_coord : SV_DomainLocation, + const OutputPatch<point_data, 4> patch) + { + point_data output; + + float4 a = lerp(patch[0].position, patch[1].position, tess_coord.x); + float4 b = lerp(patch[2].position, patch[3].position, tess_coord.x); + output.position = lerp(a, b, tess_coord.y); + + return output; + } +#endif + static const DWORD hs_quad_ccw_code[] = + { + 0x43425844, 0xdf8df700, 0x58b08fb1, 0xbd23d2c3, 0xcf884094, 0x00000001, 0x000002b8, 0x00000004, + 0x00000030, 0x00000064, 0x00000098, 0x0000015c, 0x4e475349, 0x0000002c, 0x00000001, 0x00000008, + 0x00000020, 0x00000000, 0x00000001, 0x00000003, 0x00000000, 0x00000f0f, 0x505f5653, 0x5449534f, + 0x004e4f49, 0x4e47534f, 0x0000002c, 0x00000001, 0x00000008, 0x00000020, 0x00000000, 0x00000000, + 0x00000003, 0x00000000, 0x0000000f, 0x505f5653, 0x5449534f, 0x004e4f49, 0x47534350, 0x000000bc, + 0x00000006, 0x00000008, 0x00000098, 0x00000000, 0x0000000b, 0x00000003, 0x00000000, 0x00000e01, + 0x00000098, 0x00000001, 0x0000000b, 0x00000003, 0x00000001, 0x00000e01, 0x00000098, 0x00000002, + 0x0000000b, 0x00000003, 0x00000002, 0x00000e01, 0x00000098, 0x00000003, 0x0000000b, 0x00000003, + 0x00000003, 0x00000e01, 0x000000a6, 0x00000000, 0x0000000c, 0x00000003, 0x00000004, 0x00000e01, + 0x000000a6, 0x00000001, 0x0000000c, 0x00000003, 0x00000005, 0x00000e01, 0x545f5653, 0x46737365, + 0x6f746361, 0x56530072, 0x736e495f, 0x54656469, 0x46737365, 0x6f746361, 0xabab0072, 0x58454853, + 0x00000154, 0x00030050, 0x00000055, 0x01000071, 0x01002093, 0x01002094, 0x01001895, 0x01000896, + 0x01002097, 0x0100086a, 0x04000059, 0x00208e46, 0x00000000, 0x00000002, 0x01000073, 0x04000067, + 0x00102012, 0x00000000, 0x0000000b, 0x06000036, 0x00102012, 0x00000000, 0x0020800a, 0x00000000, + 0x00000000, 0x0100003e, 0x01000073, 0x04000067, 0x00102012, 0x00000001, 0x0000000c, 0x06000036, + 0x00102012, 0x00000001, 0x0020801a, 0x00000000, 0x00000000, 0x0100003e, 0x01000073, 0x04000067, + 0x00102012, 0x00000002, 0x0000000d, 0x06000036, 0x00102012, 0x00000002, 0x0020802a, 0x00000000, + 0x00000000, 0x0100003e, 0x01000073, 0x04000067, 0x00102012, 0x00000003, 0x0000000e, 0x06000036, + 0x00102012, 0x00000003, 0x0020803a, 0x00000000, 0x00000000, 0x0100003e, 0x01000073, 0x04000067, + 0x00102012, 0x00000004, 0x0000000f, 0x06000036, 0x00102012, 0x00000004, 0x0020800a, 0x00000000, + 0x00000001, 0x0100003e, 0x01000073, 0x04000067, 0x00102012, 0x00000005, 0x00000010, 0x06000036, + 0x00102012, 0x00000005, 0x0020801a, 0x00000000, 0x00000001, 0x0100003e, + }; + static const D3D12_SHADER_BYTECODE hs_quad_ccw = {hs_quad_ccw_code, sizeof(hs_quad_ccw_code)}; + static const DWORD ds_code[] = + { + 0x43425844, 0xeb6b7631, 0x07f5469e, 0xed0cbf4a, 0x7158b3a6, 0x00000001, 0x00000284, 0x00000004, + 0x00000030, 0x00000064, 0x00000128, 0x0000015c, 0x4e475349, 0x0000002c, 0x00000001, 0x00000008, + 0x00000020, 0x00000000, 0x00000000, 0x00000003, 0x00000000, 0x00000f0f, 0x505f5653, 0x5449534f, + 0x004e4f49, 0x47534350, 0x000000bc, 0x00000006, 0x00000008, 0x00000098, 0x00000000, 0x0000000b, + 0x00000003, 0x00000000, 0x00000001, 0x00000098, 0x00000001, 0x0000000b, 0x00000003, 0x00000001, + 0x00000001, 0x00000098, 0x00000002, 0x0000000b, 0x00000003, 0x00000002, 0x00000001, 0x00000098, + 0x00000003, 0x0000000b, 0x00000003, 0x00000003, 0x00000001, 0x000000a6, 0x00000000, 0x0000000c, + 0x00000003, 0x00000004, 0x00000001, 0x000000a6, 0x00000001, 0x0000000c, 0x00000003, 0x00000005, + 0x00000001, 0x545f5653, 0x46737365, 0x6f746361, 0x56530072, 0x736e495f, 0x54656469, 0x46737365, + 0x6f746361, 0xabab0072, 0x4e47534f, 0x0000002c, 0x00000001, 0x00000008, 0x00000020, 0x00000000, + 0x00000001, 0x00000003, 0x00000000, 0x0000000f, 0x505f5653, 0x5449534f, 0x004e4f49, 0x58454853, + 0x00000120, 0x00040050, 0x00000048, 0x01002093, 0x01001895, 0x0100086a, 0x0200005f, 0x0001c032, + 0x0400005f, 0x002190f2, 0x00000004, 0x00000000, 0x04000067, 0x001020f2, 0x00000000, 0x00000001, + 0x02000068, 0x00000002, 0x0a000000, 0x001000f2, 0x00000000, 0x80219e46, 0x00000041, 0x00000002, + 0x00000000, 0x00219e46, 0x00000003, 0x00000000, 0x09000032, 0x001000f2, 0x00000000, 0x0001c006, + 0x00100e46, 0x00000000, 0x00219e46, 0x00000002, 0x00000000, 0x0a000000, 0x001000f2, 0x00000001, + 0x80219e46, 0x00000041, 0x00000000, 0x00000000, 0x00219e46, 0x00000001, 0x00000000, 0x09000032, + 0x001000f2, 0x00000001, 0x0001c006, 0x00100e46, 0x00000001, 0x00219e46, 0x00000000, 0x00000000, + 0x08000000, 0x001000f2, 0x00000000, 0x00100e46, 0x00000000, 0x80100e46, 0x00000041, 0x00000001, + 0x08000032, 0x001020f2, 0x00000000, 0x0001c556, 0x00100e46, 0x00000000, 0x00100e46, 0x00000001, + 0x0100003e, + }; + static const D3D12_SHADER_BYTECODE ds = {ds_code, sizeof(ds_code)}; +#if 0 + ... + [outputtopology("triangle_cw")] + ... +#endif + static const DWORD hs_quad_cw_code[] = + { + 0x43425844, 0x1ab30cc8, 0x94174771, 0x61f4cdd0, 0xa287f62c, 0x00000001, 0x000002b8, 0x00000004, + 0x00000030, 0x00000064, 0x00000098, 0x0000015c, 0x4e475349, 0x0000002c, 0x00000001, 0x00000008, + 0x00000020, 0x00000000, 0x00000001, 0x00000003, 0x00000000, 0x00000f0f, 0x505f5653, 0x5449534f, + 0x004e4f49, 0x4e47534f, 0x0000002c, 0x00000001, 0x00000008, 0x00000020, 0x00000000, 0x00000000, + 0x00000003, 0x00000000, 0x0000000f, 0x505f5653, 0x5449534f, 0x004e4f49, 0x47534350, 0x000000bc, + 0x00000006, 0x00000008, 0x00000098, 0x00000000, 0x0000000b, 0x00000003, 0x00000000, 0x00000e01, + 0x00000098, 0x00000001, 0x0000000b, 0x00000003, 0x00000001, 0x00000e01, 0x00000098, 0x00000002, + 0x0000000b, 0x00000003, 0x00000002, 0x00000e01, 0x00000098, 0x00000003, 0x0000000b, 0x00000003, + 0x00000003, 0x00000e01, 0x000000a6, 0x00000000, 0x0000000c, 0x00000003, 0x00000004, 0x00000e01, + 0x000000a6, 0x00000001, 0x0000000c, 0x00000003, 0x00000005, 0x00000e01, 0x545f5653, 0x46737365, + 0x6f746361, 0x56530072, 0x736e495f, 0x54656469, 0x46737365, 0x6f746361, 0xabab0072, 0x58454853, + 0x00000154, 0x00030050, 0x00000055, 0x01000071, 0x01002093, 0x01002094, 0x01001895, 0x01000896, + 0x01001897, 0x0100086a, 0x04000059, 0x00208e46, 0x00000000, 0x00000002, 0x01000073, 0x04000067, + 0x00102012, 0x00000000, 0x0000000b, 0x06000036, 0x00102012, 0x00000000, 0x0020800a, 0x00000000, + 0x00000000, 0x0100003e, 0x01000073, 0x04000067, 0x00102012, 0x00000001, 0x0000000c, 0x06000036, + 0x00102012, 0x00000001, 0x0020801a, 0x00000000, 0x00000000, 0x0100003e, 0x01000073, 0x04000067, + 0x00102012, 0x00000002, 0x0000000d, 0x06000036, 0x00102012, 0x00000002, 0x0020802a, 0x00000000, + 0x00000000, 0x0100003e, 0x01000073, 0x04000067, 0x00102012, 0x00000003, 0x0000000e, 0x06000036, + 0x00102012, 0x00000003, 0x0020803a, 0x00000000, 0x00000000, 0x0100003e, 0x01000073, 0x04000067, + 0x00102012, 0x00000004, 0x0000000f, 0x06000036, 0x00102012, 0x00000004, 0x0020800a, 0x00000000, + 0x00000001, 0x0100003e, 0x01000073, 0x04000067, 0x00102012, 0x00000005, 0x00000010, 0x06000036, + 0x00102012, 0x00000005, 0x0020801a, 0x00000000, 0x00000001, 0x0100003e, + }; + static const D3D12_SHADER_BYTECODE hs_quad_cw = {hs_quad_cw_code, sizeof(hs_quad_cw_code)}; + static const struct vec4 quad[] = + { + {-1.0f, -1.0f, 0.0f, 1.0f}, + {-1.0f, 1.0f, 0.0f, 1.0f}, + { 1.0f, -1.0f, 0.0f, 1.0f}, + { 1.0f, 1.0f, 0.0f, 1.0f}, + }; + static const D3D12_INPUT_ELEMENT_DESC layout_desc[] = + { + {"POSITION", 0, DXGI_FORMAT_R32G32B32A32_FLOAT, 0, 0, D3D12_INPUT_CLASSIFICATION_PER_VERTEX_DATA, 0}, + }; + static const D3D12_SO_DECLARATION_ENTRY so_declaration[] = + { + {0, "SV_POSITION", 0, 0, 4, 0}, + }; + unsigned int strides[] = {16}; + static const float white[] = {1.0f, 1.0f, 1.0f, 1.0f}; + static const BYTE zero_data[2048]; + static const struct triangle expected_quad_ccw[] = + { + {{{-1.0f, -1.0f, 0.0f, 1.0f}, + { 1.0f, -1.0f, 0.0f, 1.0f}, + {-1.0f, 1.0f, 0.0f, 1.0f}}}, + {{{-1.0f, 1.0f, 0.0f, 1.0f}, + { 1.0f, -1.0f, 0.0f, 1.0f}, + { 1.0f, 1.0f, 0.0f, 1.0f}}}, + {{{ 0.0f, 0.0f, 0.0f, 0.0f}, + { 0.0f, 0.0f, 0.0f, 0.0f}, + { 0.0f, 0.0f, 0.0f, 0.0f}}}, + }; + static const struct triangle expected_quad_cw[] = + { + {{{-1.0f, -1.0f, 0.0f, 1.0f}, + {-1.0f, 1.0f, 0.0f, 1.0f}, + { 1.0f, -1.0f, 0.0f, 1.0f}}}, + {{{-1.0f, 1.0f, 0.0f, 1.0f}, + { 1.0f, 1.0f, 0.0f, 1.0f}, + { 1.0f, -1.0f, 0.0f, 1.0f}}}, + {{{ 0.0f, 0.0f, 0.0f, 0.0f}, + { 0.0f, 0.0f, 0.0f, 0.0f}, + { 0.0f, 0.0f, 0.0f, 0.0f}}}, + }; + struct + { + float tess_factors[4]; + float inside_tess_factors[2]; + uint32_t padding[2]; + } constant; + + ID3D12Resource *vb, *so_buffer, *upload_buffer, *readback_buffer; + D3D12_QUERY_DATA_SO_STATISTICS *so_statistics; + D3D12_GRAPHICS_PIPELINE_STATE_DESC pso_desc; + ID3D12GraphicsCommandList *command_list; + D3D12_QUERY_HEAP_DESC query_heap_desc; + D3D12_STREAM_OUTPUT_BUFFER_VIEW sobv; + D3D12_INPUT_LAYOUT_DESC input_layout; + struct test_context_desc desc; + D3D12_VERTEX_BUFFER_VIEW vbv; + struct resource_readback rb; + ID3D12QueryHeap *query_heap; + struct test_context context; + ID3D12CommandQueue *queue; + ID3D12Device *device; + unsigned int i; + HRESULT hr; + + memset(&desc, 0, sizeof(desc)); + desc.no_root_signature = true; + if (!init_test_context(&context, &desc)) + return; + device = context.device; + command_list = context.list; + queue = context.queue; + + query_heap_desc.Type = D3D12_QUERY_HEAP_TYPE_SO_STATISTICS; + query_heap_desc.Count = 2; + query_heap_desc.NodeMask = 0; + hr = ID3D12Device_CreateQueryHeap(device, &query_heap_desc, &IID_ID3D12QueryHeap, (void **)&query_heap); + if (hr == E_NOTIMPL) + { + skip("Stream output is not supported.\n"); + destroy_test_context(&context); + return; + } + ok(hr == S_OK, "Failed to create query heap, hr %#x.\n", hr); + + context.root_signature = create_32bit_constants_root_signature_(__LINE__, + device, 0, 6, D3D12_SHADER_VISIBILITY_HULL, + D3D12_ROOT_SIGNATURE_FLAG_ALLOW_INPUT_ASSEMBLER_INPUT_LAYOUT + | D3D12_ROOT_SIGNATURE_FLAG_ALLOW_STREAM_OUTPUT); + + input_layout.pInputElementDescs = layout_desc; + input_layout.NumElements = ARRAY_SIZE(layout_desc); + + init_pipeline_state_desc(&pso_desc, context.root_signature, + context.render_target_desc.Format, NULL, NULL, &input_layout); + pso_desc.VS = vs; + pso_desc.HS = hs_quad_cw; + pso_desc.DS = ds; + 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 = 0; + pso_desc.PrimitiveTopologyType = D3D12_PRIMITIVE_TOPOLOGY_TYPE_PATCH; + + vb = create_upload_buffer(context.device, sizeof(quad), quad); + + vbv.BufferLocation = ID3D12Resource_GetGPUVirtualAddress(vb); + vbv.StrideInBytes = sizeof(*quad); + vbv.SizeInBytes = sizeof(quad); + + upload_buffer = create_upload_buffer(device, sizeof(zero_data), &zero_data); + + so_buffer = create_default_buffer(device, sizeof(zero_data), + D3D12_RESOURCE_FLAG_NONE, D3D12_RESOURCE_STATE_COPY_DEST); + + ID3D12GraphicsCommandList_CopyBufferRegion(command_list, so_buffer, 0, + upload_buffer, 0, sizeof(zero_data)); + transition_resource_state(command_list, so_buffer, + D3D12_RESOURCE_STATE_COPY_DEST, D3D12_RESOURCE_STATE_STREAM_OUT); + + sobv.BufferLocation = ID3D12Resource_GetGPUVirtualAddress(so_buffer); + sobv.SizeInBytes = 1024; + sobv.BufferFilledSizeLocation = sobv.BufferLocation + sobv.SizeInBytes; + + for (i = 0; i < ARRAY_SIZE(constant.tess_factors); ++i) + constant.tess_factors[i] = 1.0f; + for (i = 0; i < ARRAY_SIZE(constant.inside_tess_factors); ++i) + constant.inside_tess_factors[i] = 1.0f; + + pso_desc.HS = hs_quad_ccw; + hr = ID3D12Device_CreateGraphicsPipelineState(context.device, &pso_desc, + &IID_ID3D12PipelineState, (void **)&context.pipeline_state); + ok(hr == S_OK, "Failed to create state, hr %#x.\n", hr); + + ID3D12GraphicsCommandList_ClearRenderTargetView(command_list, context.rtv, white, 0, NULL); + + ID3D12GraphicsCommandList_OMSetRenderTargets(command_list, 1, &context.rtv, FALSE, NULL); + ID3D12GraphicsCommandList_SetGraphicsRootSignature(command_list, context.root_signature); + ID3D12GraphicsCommandList_SetPipelineState(command_list, context.pipeline_state); + ID3D12GraphicsCommandList_IASetPrimitiveTopology(command_list, D3D_PRIMITIVE_TOPOLOGY_4_CONTROL_POINT_PATCHLIST); + ID3D12GraphicsCommandList_RSSetViewports(command_list, 1, &context.viewport); + ID3D12GraphicsCommandList_RSSetScissorRects(command_list, 1, &context.scissor_rect); + ID3D12GraphicsCommandList_SetGraphicsRoot32BitConstants(command_list, 0, 6, &constant, 0); + ID3D12GraphicsCommandList_IASetVertexBuffers(command_list, 0, 1, &vbv); + ID3D12GraphicsCommandList_SOSetTargets(command_list, 0, 1, &sobv); + ID3D12GraphicsCommandList_DrawInstanced(command_list, 4, 1, 0, 0); + + transition_resource_state(command_list, context.render_target, + D3D12_RESOURCE_STATE_RENDER_TARGET, D3D12_RESOURCE_STATE_COPY_SOURCE); + check_sub_resource_uint(context.render_target, 0, queue, command_list, 0xffffffff, 0); + + reset_command_list(command_list, context.allocator); + transition_resource_state(command_list, context.render_target, + D3D12_RESOURCE_STATE_COPY_SOURCE, D3D12_RESOURCE_STATE_RENDER_TARGET); + + transition_resource_state(command_list, so_buffer, + D3D12_RESOURCE_STATE_STREAM_OUT, D3D12_RESOURCE_STATE_COPY_SOURCE); + check_triangles(so_buffer, queue, command_list, expected_quad_ccw, ARRAY_SIZE(expected_quad_ccw)); + + reset_command_list(command_list, context.allocator); + transition_resource_state(command_list, so_buffer, + D3D12_RESOURCE_STATE_COPY_SOURCE, D3D12_RESOURCE_STATE_COPY_DEST); + ID3D12GraphicsCommandList_CopyBufferRegion(command_list, so_buffer, 0, + upload_buffer, 0, sizeof(zero_data)); + transition_resource_state(command_list, so_buffer, + D3D12_RESOURCE_STATE_COPY_DEST, D3D12_RESOURCE_STATE_STREAM_OUT); + + ID3D12PipelineState_Release(context.pipeline_state); + pso_desc.HS = hs_quad_cw; + hr = ID3D12Device_CreateGraphicsPipelineState(context.device, &pso_desc, + &IID_ID3D12PipelineState, (void **)&context.pipeline_state); + ok(hr == S_OK, "Failed to create state, hr %#x.\n", hr); + + ID3D12GraphicsCommandList_ClearRenderTargetView(command_list, context.rtv, white, 0, NULL); + + ID3D12GraphicsCommandList_OMSetRenderTargets(command_list, 1, &context.rtv, FALSE, NULL); + ID3D12GraphicsCommandList_SetGraphicsRootSignature(command_list, context.root_signature); + ID3D12GraphicsCommandList_SetPipelineState(command_list, context.pipeline_state); + ID3D12GraphicsCommandList_IASetPrimitiveTopology(command_list, D3D_PRIMITIVE_TOPOLOGY_4_CONTROL_POINT_PATCHLIST); + ID3D12GraphicsCommandList_RSSetViewports(command_list, 1, &context.viewport); + ID3D12GraphicsCommandList_RSSetScissorRects(command_list, 1, &context.scissor_rect); + ID3D12GraphicsCommandList_SetGraphicsRoot32BitConstants(command_list, 0, 6, &constant, 0); + ID3D12GraphicsCommandList_IASetVertexBuffers(command_list, 0, 1, &vbv); + ID3D12GraphicsCommandList_SOSetTargets(command_list, 0, 1, &sobv); + ID3D12GraphicsCommandList_DrawInstanced(command_list, 4, 1, 0, 0); + + transition_resource_state(command_list, context.render_target, + D3D12_RESOURCE_STATE_RENDER_TARGET, D3D12_RESOURCE_STATE_COPY_SOURCE); + check_sub_resource_uint(context.render_target, 0, queue, command_list, 0xff00ff00, 0); + + reset_command_list(command_list, context.allocator); + transition_resource_state(command_list, context.render_target, + D3D12_RESOURCE_STATE_COPY_SOURCE, D3D12_RESOURCE_STATE_RENDER_TARGET); + + transition_resource_state(command_list, so_buffer, + D3D12_RESOURCE_STATE_STREAM_OUT, D3D12_RESOURCE_STATE_COPY_SOURCE); + check_triangles(so_buffer, queue, command_list, expected_quad_cw, ARRAY_SIZE(expected_quad_cw)); + + reset_command_list(command_list, context.allocator); + transition_resource_state(command_list, so_buffer, + D3D12_RESOURCE_STATE_COPY_SOURCE, D3D12_RESOURCE_STATE_COPY_DEST); + ID3D12GraphicsCommandList_CopyBufferRegion(command_list, so_buffer, 0, + upload_buffer, 0, sizeof(zero_data)); + transition_resource_state(command_list, so_buffer, + D3D12_RESOURCE_STATE_COPY_DEST, D3D12_RESOURCE_STATE_STREAM_OUT); + + ID3D12GraphicsCommandList_OMSetRenderTargets(command_list, 1, &context.rtv, FALSE, NULL); + ID3D12GraphicsCommandList_SetGraphicsRootSignature(command_list, context.root_signature); + ID3D12GraphicsCommandList_SetPipelineState(command_list, context.pipeline_state); + ID3D12GraphicsCommandList_IASetPrimitiveTopology(command_list, D3D_PRIMITIVE_TOPOLOGY_4_CONTROL_POINT_PATCHLIST); + ID3D12GraphicsCommandList_RSSetViewports(command_list, 1, &context.viewport); + ID3D12GraphicsCommandList_RSSetScissorRects(command_list, 1, &context.scissor_rect); + ID3D12GraphicsCommandList_IASetVertexBuffers(command_list, 0, 1, &vbv); + ID3D12GraphicsCommandList_SOSetTargets(command_list, 0, 1, &sobv); + + ID3D12GraphicsCommandList_BeginQuery(command_list, query_heap, D3D12_QUERY_TYPE_SO_STATISTICS_STREAM0, 0); + + for (i = 0; i < ARRAY_SIZE(constant.tess_factors); ++i) + constant.tess_factors[i] = 2.0f; + ID3D12GraphicsCommandList_SetGraphicsRoot32BitConstants(command_list, 0, 6, &constant, 0); + ID3D12GraphicsCommandList_DrawInstanced(command_list, 4, 1, 0, 0); + + constant.tess_factors[0] = 0.0f; /* A patch is discarded. */ + ID3D12GraphicsCommandList_SetGraphicsRoot32BitConstants(command_list, 0, 6, &constant, 0); + ID3D12GraphicsCommandList_DrawInstanced(command_list, 4, 1, 0, 0); + + ID3D12GraphicsCommandList_EndQuery(command_list, query_heap, D3D12_QUERY_TYPE_SO_STATISTICS_STREAM0, 0); + + ID3D12GraphicsCommandList_BeginQuery(command_list, query_heap, D3D12_QUERY_TYPE_SO_STATISTICS_STREAM0, 1); + + constant.tess_factors[0] = 5.0f; + ID3D12GraphicsCommandList_SetGraphicsRoot32BitConstants(command_list, 0, 6, &constant, 0); + ID3D12GraphicsCommandList_DrawInstanced(command_list, 4, 1, 0, 0); + + ID3D12GraphicsCommandList_EndQuery(command_list, query_heap, D3D12_QUERY_TYPE_SO_STATISTICS_STREAM0, 1); + + readback_buffer = create_readback_buffer(device, 2 * sizeof(*so_statistics)); + ID3D12GraphicsCommandList_ResolveQueryData(command_list, + query_heap, D3D12_QUERY_TYPE_SO_STATISTICS_STREAM0, 0, 2, readback_buffer, 0); + + get_buffer_readback_with_command_list(readback_buffer, DXGI_FORMAT_UNKNOWN, &rb, queue, command_list); + so_statistics = get_readback_data(&rb, 0, 0, 0, sizeof(*so_statistics)); + ok(so_statistics[0].NumPrimitivesWritten == 8, "Got unexpected primitives written %u.\n", + (unsigned int)so_statistics[0].NumPrimitivesWritten); + ok(so_statistics[0].PrimitivesStorageNeeded == 8, "Got unexpected primitives storage needed %u.\n", + (unsigned int)so_statistics[0].PrimitivesStorageNeeded); + ok(so_statistics[1].NumPrimitivesWritten == 11, "Got unexpected primitives written %u.\n", + (unsigned int)so_statistics[1].NumPrimitivesWritten); + ok(so_statistics[1].PrimitivesStorageNeeded == 11, "Got unexpected primitives storage needed %u.\n", + (unsigned int)so_statistics[1].PrimitivesStorageNeeded); + release_resource_readback(&rb); + + ID3D12Resource_Release(readback_buffer); + ID3D12Resource_Release(so_buffer); + ID3D12Resource_Release(upload_buffer); + ID3D12Resource_Release(vb); + ID3D12QueryHeap_Release(query_heap); + destroy_test_context(&context); +} + static void test_render_a8(void) { static const float black[] = {0.0f, 0.0f, 0.0f, 0.0f}; @@ -24330,6 +24811,7 @@ START_TEST(d3d12) run_test(test_geometry_shader); run_test(test_layered_rendering); run_test(test_nop_tessellation_shaders); + run_test(test_quad_tessellation); run_test(test_render_a8); run_test(test_cpu_descriptors_lifetime); run_test(test_clip_distance); -- 2.19.2

2 1

[PATCH vkd3d 3/8] vkd3d: Implement SO statistics queries.
by Józef Kucia 20 Feb '19

20 Feb '19

From: Józef Kucia <jkucia(a)codeweavers.com> Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> --- include/vkd3d_d3d12.idl | 6 ++++++ libs/vkd3d/command.c | 35 +++++++++++++++++++++++++++++++---- libs/vkd3d/device.c | 1 + libs/vkd3d/resource.c | 13 ++++++++++--- libs/vkd3d/vkd3d_private.h | 1 + tests/d3d12.c | 11 ++++++++--- 6 files changed, 57 insertions(+), 10 deletions(-) diff --git a/include/vkd3d_d3d12.idl b/include/vkd3d_d3d12.idl index 8b2e70fdadcb..231123bd672f 100644 --- a/include/vkd3d_d3d12.idl +++ b/include/vkd3d_d3d12.idl @@ -1749,6 +1749,12 @@ typedef struct D3D12_QUERY_DATA_PIPELINE_STATISTICS UINT64 CSInvocations; } D3D12_QUERY_DATA_PIPELINE_STATISTICS; +typedef struct D3D12_QUERY_DATA_SO_STATISTICS +{ + UINT64 NumPrimitivesWritten; + UINT64 PrimitivesStorageNeeded; +} D3D12_QUERY_DATA_SO_STATISTICS; + typedef enum D3D12_PREDICATION_OP { D3D12_PREDICATION_OP_EQUAL_ZERO = 0, diff --git a/libs/vkd3d/command.c b/libs/vkd3d/command.c index 7457ae3610fa..5b11c3c43d44 100644 --- a/libs/vkd3d/command.c +++ b/libs/vkd3d/command.c @@ -4080,10 +4080,19 @@ static void STDMETHODCALLTYPE d3d12_command_list_BeginQuery(ID3D12GraphicsComman d3d12_command_list_end_current_render_pass(list); + VK_CALL(vkCmdResetQueryPool(list->vk_command_buffer, query_heap->vk_query_pool, index, 1)); + if (type == D3D12_QUERY_TYPE_OCCLUSION) flags = VK_QUERY_CONTROL_PRECISE_BIT; - VK_CALL(vkCmdResetQueryPool(list->vk_command_buffer, query_heap->vk_query_pool, index, 1)); + if (D3D12_QUERY_TYPE_SO_STATISTICS_STREAM0 <= type && type <= D3D12_QUERY_TYPE_SO_STATISTICS_STREAM3) + { + unsigned int stream_index = type - D3D12_QUERY_TYPE_SO_STATISTICS_STREAM0; + VK_CALL(vkCmdBeginQueryIndexedEXT(list->vk_command_buffer, + query_heap->vk_query_pool, index, flags, stream_index)); + return; + } + VK_CALL(vkCmdBeginQuery(list->vk_command_buffer, query_heap->vk_query_pool, index, flags)); } @@ -4110,9 +4119,28 @@ static void STDMETHODCALLTYPE d3d12_command_list_EndQuery(ID3D12GraphicsCommandL return; } + if (D3D12_QUERY_TYPE_SO_STATISTICS_STREAM0 <= type && type <= D3D12_QUERY_TYPE_SO_STATISTICS_STREAM3) + { + unsigned int stream_index = type - D3D12_QUERY_TYPE_SO_STATISTICS_STREAM0; + VK_CALL(vkCmdEndQueryIndexedEXT(list->vk_command_buffer, + query_heap->vk_query_pool, index, stream_index)); + return; + } + VK_CALL(vkCmdEndQuery(list->vk_command_buffer, query_heap->vk_query_pool, index)); } +static size_t get_query_stride(D3D12_QUERY_TYPE type) +{ + if (type == D3D12_QUERY_TYPE_PIPELINE_STATISTICS) + return sizeof(D3D12_QUERY_DATA_PIPELINE_STATISTICS); + + if (D3D12_QUERY_TYPE_SO_STATISTICS_STREAM0 <= type && type <= D3D12_QUERY_TYPE_SO_STATISTICS_STREAM3) + return sizeof(D3D12_QUERY_DATA_SO_STATISTICS); + + return sizeof(uint64_t); +} + static void STDMETHODCALLTYPE d3d12_command_list_ResolveQueryData(ID3D12GraphicsCommandList *iface, ID3D12QueryHeap *heap, D3D12_QUERY_TYPE type, UINT start_index, UINT query_count, ID3D12Resource *dst_buffer, UINT64 aligned_dst_buffer_offset) @@ -4120,9 +4148,9 @@ static void STDMETHODCALLTYPE d3d12_command_list_ResolveQueryData(ID3D12Graphics const struct d3d12_query_heap *query_heap = unsafe_impl_from_ID3D12QueryHeap(heap); struct d3d12_command_list *list = impl_from_ID3D12GraphicsCommandList(iface); struct d3d12_resource *buffer = unsafe_impl_from_ID3D12Resource(dst_buffer); - VkDeviceSize offset, stride = sizeof(uint64_t); const struct vkd3d_vk_device_procs *vk_procs; unsigned int i, first, count; + VkDeviceSize offset, stride; TRACE("iface %p, heap %p, type %#x, start_index %u, query_count %u, " "dst_buffer %p, aligned_dst_buffer_offset %#"PRIx64".\n", @@ -4149,8 +4177,7 @@ static void STDMETHODCALLTYPE d3d12_command_list_ResolveQueryData(ID3D12Graphics d3d12_command_list_end_current_render_pass(list); - if (type == D3D12_QUERY_TYPE_PIPELINE_STATISTICS) - stride = sizeof(struct D3D12_QUERY_DATA_PIPELINE_STATISTICS); + stride = get_query_stride(type); count = 0; first = start_index; diff --git a/libs/vkd3d/device.c b/libs/vkd3d/device.c index df59a49dcc17..3d9b8c9aae8d 100644 --- a/libs/vkd3d/device.c +++ b/libs/vkd3d/device.c @@ -945,6 +945,7 @@ static HRESULT vkd3d_init_device_caps(struct d3d12_device *device, vulkan_info->device_limits = device_properties2.properties.limits; vulkan_info->sparse_properties = device_properties2.properties.sparseProperties; vulkan_info->rasterization_stream = xfb_properties.transformFeedbackRasterizationStreamSelect; + vulkan_info->transform_feedback_queries = xfb_properties.transformFeedbackQueries; vulkan_info->max_vertex_attrib_divisor = max(vertex_divisor_properties.maxVertexAttribDivisor, 1); device->feature_options.DoublePrecisionFloatShaderOps = features->shaderFloat64; diff --git a/libs/vkd3d/resource.c b/libs/vkd3d/resource.c index 34f558d6fda3..1176538f5c54 100644 --- a/libs/vkd3d/resource.c +++ b/libs/vkd3d/resource.c @@ -2835,9 +2835,16 @@ HRESULT d3d12_query_heap_create(struct d3d12_device *device, const D3D12_QUERY_H break; case D3D12_QUERY_HEAP_TYPE_SO_STATISTICS: - FIXME("Unsupported query heap type SO_STATISTICS.\n"); - vkd3d_free(object); - return E_NOTIMPL; + if (!device->vk_info.transform_feedback_queries) + { + FIXME("Transform feedback queries are not supported by Vulkan implementation.\n"); + vkd3d_free(object); + return E_NOTIMPL; + } + + pool_info.queryType = VK_QUERY_TYPE_TRANSFORM_FEEDBACK_STREAM_EXT; + pool_info.pipelineStatistics = 0; + break; default: WARN("Invalid query heap type %u.\n", desc->Type); diff --git a/libs/vkd3d/vkd3d_private.h b/libs/vkd3d/vkd3d_private.h index dee129e80e9e..9b2fbaf4abe0 100644 --- a/libs/vkd3d/vkd3d_private.h +++ b/libs/vkd3d/vkd3d_private.h @@ -93,6 +93,7 @@ struct vkd3d_vulkan_info bool EXT_vertex_attribute_divisor; bool rasterization_stream; + bool transform_feedback_queries; bool vertex_attrib_zero_divisor; unsigned int max_vertex_attrib_divisor; diff --git a/tests/d3d12.c b/tests/d3d12.c index 259f0cafa443..e83ea48fdb6a 100644 --- a/tests/d3d12.c +++ b/tests/d3d12.c @@ -19007,10 +19007,15 @@ static void test_create_query_heap(void) heap_desc.NodeMask = 0; hr = ID3D12Device_CreateQueryHeap(device, &heap_desc, &IID_ID3D12QueryHeap, (void **)&query_heap); - todo ok(hr == S_OK, "Failed to create query heap, type %u, hr %#x.\n", heap_desc.Type, hr); - - if (hr == S_OK) + if (hr != E_NOTIMPL) + { + ok(hr == S_OK, "Failed to create query heap, type %u, hr %#x.\n", heap_desc.Type, hr); ID3D12QueryHeap_Release(query_heap); + } + else + { + skip("Stream output is not supported.\n"); + } refcount = ID3D12Device_Release(device); ok(!refcount, "ID3D12Device has %u references left.\n", (unsigned int)refcount); -- 2.19.2

2 1

[PATCH vkd3d 4/8] vkd3d-shader: Avoid using SPIR-V built-ins for shader interface between tessellation stages.
by Józef Kucia 20 Feb '19

20 Feb '19

From: Józef Kucia <jkucia(a)codeweavers.com> We can do this because both tessellation shaders are required in Direct3D. Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> --- libs/vkd3d-shader/spirv.c | 67 ++++++++++++++------------------------- 1 file changed, 24 insertions(+), 43 deletions(-) diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index 6ba169f6c6af..afed8a27050a 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -129,20 +129,6 @@ static const void *vkd3d_find_struct_(const struct vkd3d_struct *chain, return NULL; } -static enum vkd3d_shader_input_sysval_semantic vkd3d_siv_from_sysval(enum vkd3d_sysval_semantic sysval) -{ - switch (sysval) - { - case VKD3D_SV_NONE: - return VKD3D_SIV_NONE; - case VKD3D_SV_POSITION: - return VKD3D_SIV_POSITION; - default: - FIXME("Unhandled sysval %#x.\n", sysval); - return VKD3D_SIV_NONE; - } -} - #define VKD3D_SPIRV_VERSION 0x00010000 #define VKD3D_SPIRV_GENERATOR_ID 18 #define VKD3D_SPIRV_GENERATOR_VERSION 1 @@ -3558,9 +3544,6 @@ static uint32_t vkd3d_dxbc_compiler_emit_input(struct vkd3d_dxbc_compiler *compi return 0; } - if (!sysval && signature_element->sysval_semantic) - sysval = vkd3d_siv_from_sysval(signature_element->sysval_semantic); - builtin = get_spirv_builtin_for_sysval(compiler, sysval); component_idx = vkd3d_write_mask_get_component_idx(dst->write_mask); @@ -5014,11 +4997,12 @@ static const struct vkd3d_shader_phase *vkd3d_dxbc_compiler_get_control_point_ph static void vkd3d_dxbc_compiler_emit_default_control_point_phase(struct vkd3d_dxbc_compiler *compiler) { - const struct vkd3d_shader_signature *signature = compiler->input_signature; + const struct vkd3d_shader_signature *output_signature = compiler->output_signature; + const struct vkd3d_shader_signature *input_signature = compiler->input_signature; struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; + const struct vkd3d_spirv_builtin *builtin, *input_builtin; uint32_t type_id, input_ptr_type_id, output_ptr_type_id; uint32_t input_id, output_id, dst_id, src_id; - const struct vkd3d_spirv_builtin *builtin; enum vkd3d_component_type component_type; unsigned int component_count; uint32_t invocation_id; @@ -5030,48 +5014,45 @@ static void vkd3d_dxbc_compiler_emit_default_control_point_phase(struct vkd3d_dx type_id = vkd3d_spirv_get_type_id(builder, builtin->component_type, builtin->component_count); invocation_id = vkd3d_spirv_build_op_load(builder, type_id, invocation_id, SpvMemoryAccessMaskNone); - for (i = 0; i < signature->element_count; ++i) + assert(input_signature->element_count == output_signature->element_count); + for (i = 0; i < output_signature->element_count; ++i) { - const struct vkd3d_shader_signature_element *e = &signature->elements[i]; + const struct vkd3d_shader_signature_element *output = &output_signature->elements[i]; + const struct vkd3d_shader_signature_element *input = &input_signature->elements[i]; + + assert((input->mask & 0xff) == (output->mask & 0xff)); + assert(input->component_type == output->component_type); - if ((builtin = get_spirv_builtin_for_sysval(compiler, e->sysval_semantic))) + if ((input_builtin = get_spirv_builtin_for_sysval(compiler, input->sysval_semantic))) { - component_type = builtin->component_type; - component_count = builtin->component_count; + component_type = input_builtin->component_type; + component_count = input_builtin->component_count; } else { - component_type = e->component_type; - component_count = vkd3d_write_mask_component_count(e->mask & 0xff); + component_type = input->component_type; + component_count = vkd3d_write_mask_component_count(input->mask & 0xff); } - if (builtin) + if (input_builtin) { input_id = vkd3d_dxbc_compiler_emit_builtin_variable(compiler, - builtin, SpvStorageClassInput, compiler->input_control_point_count); + input_builtin, SpvStorageClassInput, compiler->input_control_point_count); } else { input_id = vkd3d_dxbc_compiler_emit_array_variable(compiler, &builder->global_stream, SpvStorageClassInput, component_type, component_count, compiler->input_control_point_count); vkd3d_spirv_add_iface_variable(builder, input_id); - vkd3d_spirv_build_op_decorate1(builder, input_id, SpvDecorationLocation, e->register_index); + vkd3d_spirv_build_op_decorate1(builder, input_id, SpvDecorationLocation, input->register_index); } - vkd3d_spirv_build_op_name(builder, input_id, "vicp%u", e->register_index); + vkd3d_spirv_build_op_name(builder, input_id, "vicp%u", input->register_index); - if (builtin) - { - output_id = vkd3d_dxbc_compiler_emit_builtin_variable(compiler, - builtin, SpvStorageClassOutput, compiler->output_control_point_count); - } - else - { - output_id = vkd3d_dxbc_compiler_emit_array_variable(compiler, &builder->global_stream, - SpvStorageClassOutput, component_type, component_count, compiler->output_control_point_count); - vkd3d_spirv_add_iface_variable(builder, output_id); - vkd3d_spirv_build_op_decorate1(builder, output_id, SpvDecorationLocation, e->register_index); - } - vkd3d_spirv_build_op_name(builder, output_id, "vocp%u", e->register_index); + output_id = vkd3d_dxbc_compiler_emit_array_variable(compiler, &builder->global_stream, + SpvStorageClassOutput, component_type, component_count, compiler->output_control_point_count); + vkd3d_spirv_add_iface_variable(builder, output_id); + vkd3d_spirv_build_op_decorate1(builder, output_id, SpvDecorationLocation, output->register_index); + vkd3d_spirv_build_op_name(builder, output_id, "vocp%u", output->register_index); type_id = vkd3d_spirv_get_type_id(builder, component_type, component_count); output_ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, SpvStorageClassOutput, type_id); -- 2.19.2

2 1

[PATCH vkd3d 2/8] vkd3d-shader: Set xxxx swizzle for scalar registers.
by Józef Kucia 20 Feb '19

20 Feb '19

From: Józef Kucia <jkucia(a)codeweavers.com> Mainly for vOutputControlPointID. Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> --- libs/vkd3d-shader/dxbc.c | 43 +++++++++++++++++++++------------------- 1 file changed, 23 insertions(+), 20 deletions(-) diff --git a/libs/vkd3d-shader/dxbc.c b/libs/vkd3d-shader/dxbc.c index 5e4babb1e4de..49116681dbab 100644 --- a/libs/vkd3d-shader/dxbc.c +++ b/libs/vkd3d-shader/dxbc.c @@ -1511,6 +1511,25 @@ static bool shader_sm4_read_param(struct vkd3d_sm4_data *priv, const DWORD **ptr return true; } +static bool shader_sm4_is_scalar_register(const struct vkd3d_shader_register *reg) +{ + switch (reg->type) + { + case VKD3DSPR_COVERAGE: + case VKD3DSPR_DEPTHOUT: + case VKD3DSPR_DEPTHOUTGE: + case VKD3DSPR_DEPTHOUTLE: + case VKD3DSPR_GSINSTID: + case VKD3DSPR_LOCALTHREADINDEX: + case VKD3DSPR_OUTPOINTID: + case VKD3DSPR_PRIMID: + case VKD3DSPR_SAMPLEMASK: + return true; + default: + return false; + } +} + static bool shader_sm4_read_src_param(struct vkd3d_sm4_data *priv, const DWORD **ptr, const DWORD *end, enum vkd3d_data_type data_type, struct vkd3d_shader_src_param *src_param) { @@ -1541,7 +1560,10 @@ static bool shader_sm4_read_src_param(struct vkd3d_sm4_data *priv, const DWORD * switch (swizzle_type) { case VKD3D_SM4_SWIZZLE_NONE: - src_param->swizzle = VKD3D_NO_SWIZZLE; + if (shader_sm4_is_scalar_register(&src_param->reg)) + src_param->swizzle = VKD3D_SWIZZLE(VKD3D_SWIZZLE_X, VKD3D_SWIZZLE_X, VKD3D_SWIZZLE_X, VKD3D_SWIZZLE_X); + else + src_param->swizzle = VKD3D_NO_SWIZZLE; break; case VKD3D_SM4_SWIZZLE_SCALAR: @@ -1562,25 +1584,6 @@ static bool shader_sm4_read_src_param(struct vkd3d_sm4_data *priv, const DWORD * return true; } -static bool shader_sm4_is_scalar_register(const struct vkd3d_shader_register *reg) -{ - switch (reg->type) - { - case VKD3DSPR_COVERAGE: - case VKD3DSPR_DEPTHOUT: - case VKD3DSPR_DEPTHOUTGE: - case VKD3DSPR_DEPTHOUTLE: - case VKD3DSPR_GSINSTID: - case VKD3DSPR_LOCALTHREADINDEX: - case VKD3DSPR_OUTPOINTID: - case VKD3DSPR_PRIMID: - case VKD3DSPR_SAMPLEMASK: - return true; - default: - return false; - } -} - static bool shader_sm4_read_dst_param(struct vkd3d_sm4_data *priv, const DWORD **ptr, const DWORD *end, enum vkd3d_data_type data_type, struct vkd3d_shader_dst_param *dst_param) { -- 2.19.2

2 1

[PATCH vkd3d 1/8] vkd3d-shader: Replace VKD3DSIH_TABLE_SIZE with VKD3DSIH_INVALID.
by Józef Kucia 20 Feb '19

20 Feb '19

From: Józef Kucia <jkucia(a)codeweavers.com> Signed-off-by: Józef Kucia <jkucia(a)codeweavers.com> --- libs/vkd3d-shader/dxbc.c | 10 +++++----- libs/vkd3d-shader/trace.c | 2 +- libs/vkd3d-shader/vkd3d_shader_main.c | 4 ++-- libs/vkd3d-shader/vkd3d_shader_private.h | 3 ++- 4 files changed, 10 insertions(+), 9 deletions(-) diff --git a/libs/vkd3d-shader/dxbc.c b/libs/vkd3d-shader/dxbc.c index 0b5f7bc892f6..5e4babb1e4de 100644 --- a/libs/vkd3d-shader/dxbc.c +++ b/libs/vkd3d-shader/dxbc.c @@ -545,7 +545,7 @@ static void shader_sm4_read_shader_data(struct vkd3d_shader_instruction *ins, if (icb_size % 4 || icb_size > MAX_IMMEDIATE_CONSTANT_BUFFER_SIZE) { FIXME("Unexpected immediate constant buffer size %u.\n", icb_size); - ins->handler_idx = VKD3DSIH_TABLE_SIZE; + ins->handler_idx = VKD3DSIH_INVALID; return; } @@ -1689,7 +1689,7 @@ void shader_sm4_read_instruction(void *data, const DWORD **ptr, struct vkd3d_sha if (!(opcode_info = get_opcode_info(opcode))) { FIXME("Unrecognized opcode %#x, opcode_token 0x%08x.\n", opcode, opcode_token); - ins->handler_idx = VKD3DSIH_TABLE_SIZE; + ins->handler_idx = VKD3DSIH_INVALID; *ptr += len; return; } @@ -1733,7 +1733,7 @@ void shader_sm4_read_instruction(void *data, const DWORD **ptr, struct vkd3d_sha if (!(shader_sm4_read_dst_param(priv, &p, *ptr, map_data_type(opcode_info->dst_info[i]), &priv->dst_param[i]))) { - ins->handler_idx = VKD3DSIH_TABLE_SIZE; + ins->handler_idx = VKD3DSIH_INVALID; return; } priv->dst_param[i].modifiers |= instruction_dst_modifier; @@ -1744,7 +1744,7 @@ void shader_sm4_read_instruction(void *data, const DWORD **ptr, struct vkd3d_sha if (!(shader_sm4_read_src_param(priv, &p, *ptr, map_data_type(opcode_info->src_info[i]), &priv->src_param[i]))) { - ins->handler_idx = VKD3DSIH_TABLE_SIZE; + ins->handler_idx = VKD3DSIH_INVALID; return; } } @@ -1754,7 +1754,7 @@ void shader_sm4_read_instruction(void *data, const DWORD **ptr, struct vkd3d_sha fail: *ptr = priv->end; - ins->handler_idx = VKD3DSIH_TABLE_SIZE; + ins->handler_idx = VKD3DSIH_INVALID; return; } diff --git a/libs/vkd3d-shader/trace.c b/libs/vkd3d-shader/trace.c index cb7452bdc2ab..6375e714c1fc 100644 --- a/libs/vkd3d-shader/trace.c +++ b/libs/vkd3d-shader/trace.c @@ -1602,7 +1602,7 @@ void vkd3d_shader_trace(void *data) struct vkd3d_shader_instruction ins; shader_sm4_read_instruction(data, &ptr, &ins); - if (ins.handler_idx == VKD3DSIH_TABLE_SIZE) + if (ins.handler_idx == VKD3DSIH_INVALID) { WARN("Skipping unrecognized instruction.\n"); shader_addline(&buffer, "<unrecognized instruction>\n"); diff --git a/libs/vkd3d-shader/vkd3d_shader_main.c b/libs/vkd3d-shader/vkd3d_shader_main.c index 6e6880c689b4..55cb56c2e191 100644 --- a/libs/vkd3d-shader/vkd3d_shader_main.c +++ b/libs/vkd3d-shader/vkd3d_shader_main.c @@ -175,7 +175,7 @@ int vkd3d_shader_compile_dxbc(const struct vkd3d_shader_code *dxbc, { shader_sm4_read_instruction(parser.data, &parser.ptr, &instruction); - if (instruction.handler_idx == VKD3DSIH_TABLE_SIZE) + if (instruction.handler_idx == VKD3DSIH_INVALID) { WARN("Encountered unrecognized or invalid instruction.\n"); vkd3d_dxbc_compiler_destroy(spirv_compiler); @@ -305,7 +305,7 @@ int vkd3d_shader_scan_dxbc(const struct vkd3d_shader_code *dxbc, { shader_sm4_read_instruction(parser.data, &parser.ptr, &instruction); - if (instruction.handler_idx == VKD3DSIH_TABLE_SIZE) + if (instruction.handler_idx == VKD3DSIH_INVALID) { WARN("Encountered unrecognized or invalid instruction.\n"); vkd3d_shader_parser_destroy(&parser); diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index 40d46b3b86c1..6b095a9cbba6 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -300,7 +300,8 @@ enum VKD3D_SHADER_INSTRUCTION_HANDLER VKD3DSIH_USHR, VKD3DSIH_UTOF, VKD3DSIH_XOR, - VKD3DSIH_TABLE_SIZE + + VKD3DSIH_INVALID, }; enum vkd3d_shader_register_type -- 2.19.2

2 1

[PATCH 4/4] windowscodecs: Avoid redundant checks when reading a TIFF tile.
by Dmitry Timoshkov 20 Feb '19

20 Feb '19

Signed-off-by: Dmitry Timoshkov <dmitry(a)baikal.ru> --- dlls/windowscodecs/tiffformat.c | 40 ++++++++++++--------------------- 1 file changed, 14 insertions(+), 26 deletions(-) diff --git a/dlls/windowscodecs/tiffformat.c b/dlls/windowscodecs/tiffformat.c index 3bc8d696ca..295330aea2 100644 --- a/dlls/windowscodecs/tiffformat.c +++ b/dlls/windowscodecs/tiffformat.c @@ -974,34 +974,25 @@ static HRESULT WINAPI TiffFrameDecode_CopyPalette(IWICBitmapFrameDecode *iface, static HRESULT TiffFrameDecode_ReadTile(TiffFrameDecode *This, UINT tile_x, UINT tile_y) { - HRESULT hr=S_OK; tsize_t ret; int swap_bytes; swap_bytes = pTIFFIsByteSwapped(This->parent->tiff); ret = pTIFFSetDirectory(This->parent->tiff, This->index); - if (ret == -1) - hr = E_FAIL; + return E_FAIL; - if (hr == S_OK) - { - if (This->decode_info.tiled) - { - ret = pTIFFReadEncodedTile(This->parent->tiff, tile_x + tile_y * This->decode_info.tiles_across, This->cached_tile, This->decode_info.tile_size); - } - else - { - ret = pTIFFReadEncodedStrip(This->parent->tiff, tile_y, This->cached_tile, This->decode_info.tile_size); - } + if (This->decode_info.tiled) + ret = pTIFFReadEncodedTile(This->parent->tiff, tile_x + tile_y * This->decode_info.tiles_across, This->cached_tile, This->decode_info.tile_size); + else + ret = pTIFFReadEncodedStrip(This->parent->tiff, tile_y, This->cached_tile, This->decode_info.tile_size); - if (ret == -1) - hr = E_FAIL; - } + if (ret == -1) + return E_FAIL; /* 8bpp grayscale with extra alpha */ - if (hr == S_OK && This->decode_info.source_bpp == 16 && This->decode_info.samples == 2 && This->decode_info.bpp == 32) + if (This->decode_info.source_bpp == 16 && This->decode_info.samples == 2 && This->decode_info.bpp == 32) { BYTE *src; DWORD *dst, count = This->decode_info.tile_width * This->decode_info.tile_height; @@ -1016,7 +1007,7 @@ static HRESULT TiffFrameDecode_ReadTile(TiffFrameDecode *This, UINT tile_x, UINT } } - if (hr == S_OK && This->decode_info.reverse_bgr) + if (This->decode_info.reverse_bgr) { if (This->decode_info.bps == 8) { @@ -1027,7 +1018,7 @@ static HRESULT TiffFrameDecode_ReadTile(TiffFrameDecode *This, UINT tile_x, UINT } } - if (hr == S_OK && swap_bytes && This->decode_info.bps > 8) + if (swap_bytes && This->decode_info.bps > 8) { UINT row, i, samples_per_row; BYTE *sample, temp; @@ -1055,7 +1046,7 @@ static HRESULT TiffFrameDecode_ReadTile(TiffFrameDecode *This, UINT tile_x, UINT } } - if (hr == S_OK && This->decode_info.invert_grayscale) + if (This->decode_info.invert_grayscale) { BYTE *byte, *end; @@ -1071,13 +1062,10 @@ static HRESULT TiffFrameDecode_ReadTile(TiffFrameDecode *This, UINT tile_x, UINT *byte = ~(*byte); } - if (hr == S_OK) - { - This->cached_tile_x = tile_x; - This->cached_tile_y = tile_y; - } + This->cached_tile_x = tile_x; + This->cached_tile_y = tile_y; - return hr; + return S_OK; } static HRESULT WINAPI TiffFrameDecode_CopyPixels(IWICBitmapFrameDecode *iface, -- 2.20.1

2 1

[PATCH 3/4] windowscodecs: Fail earlier in TIFF decoder's Initialize method for unsupported pixel formats.
by Dmitry Timoshkov 20 Feb '19

20 Feb '19

Signed-off-by: Dmitry Timoshkov <dmitry(a)baikal.ru> --- dlls/windowscodecs/tiffformat.c | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-) diff --git a/dlls/windowscodecs/tiffformat.c b/dlls/windowscodecs/tiffformat.c index 58cd04ef17..3bc8d696ca 100644 --- a/dlls/windowscodecs/tiffformat.c +++ b/dlls/windowscodecs/tiffformat.c @@ -638,6 +638,7 @@ static HRESULT WINAPI TiffDecoder_Initialize(IWICBitmapDecoder *iface, IStream * { TiffDecoder *This = impl_from_IWICBitmapDecoder(iface); TIFF *tiff; + tiff_decode_info decode_info; HRESULT hr=S_OK; TRACE("(%p,%p,%x)\n", iface, pIStream, cacheOptions); @@ -651,13 +652,20 @@ static HRESULT WINAPI TiffDecoder_Initialize(IWICBitmapDecoder *iface, IStream * } tiff = tiff_open_stream(pIStream, "r"); - if (!tiff) { hr = E_FAIL; goto exit; } + /* make sure that TIFF format is supported */ + hr = tiff_get_decode_info(tiff, &decode_info); + if (hr != S_OK) + { + pTIFFClose(tiff); + goto exit; + } + This->tiff = tiff; This->stream = pIStream; IStream_AddRef(pIStream); -- 2.20.1

2 1