December 2023 - Wine-commits

Zebediah Figura : vkd3d-shader: Do not scan the shader in vkd3d_shader_parser_compile() for assembly targets.
by Alexandre Julliard 14 Dec '23

14 Dec '23

Module: vkd3d Branch: master Commit: 8af47a96ea5f8531e863243cf441d1624083baf3 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/8af47a96ea5f8531e863243cf441d… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Wed Dec 13 15:50:01 2023 -0600 vkd3d-shader: Do not scan the shader in vkd3d_shader_parser_compile() for assembly targets. --- libs/vkd3d-shader/vkd3d_shader_main.c | 12 +++++++----- 1 file changed, 7 insertions(+), 5 deletions(-) diff --git a/libs/vkd3d-shader/vkd3d_shader_main.c b/libs/vkd3d-shader/vkd3d_shader_main.c index 2ee97e6c..0600e722 100644 --- a/libs/vkd3d-shader/vkd3d_shader_main.c +++ b/libs/vkd3d-shader/vkd3d_shader_main.c @@ -1540,9 +1540,6 @@ static int vkd3d_shader_parser_compile(struct vkd3d_shader_parser *parser, scan_info = *compile_info; - if ((ret = scan_with_parser(&scan_info, message_context, &scan_descriptor_info, parser)) < 0) - return ret; - switch (compile_info->target_type) { case VKD3D_SHADER_TARGET_D3D_ASM: @@ -1550,6 +1547,8 @@ static int vkd3d_shader_parser_compile(struct vkd3d_shader_parser *parser, break; case VKD3D_SHADER_TARGET_GLSL: + if ((ret = scan_with_parser(&scan_info, message_context, &scan_descriptor_info, parser)) < 0) + return ret; if (!(glsl_generator = vkd3d_glsl_generator_create(&parser->shader_version, message_context, &parser->location))) { @@ -1560,19 +1559,22 @@ static int vkd3d_shader_parser_compile(struct vkd3d_shader_parser *parser, ret = vkd3d_glsl_generator_generate(glsl_generator, parser, out); vkd3d_glsl_generator_destroy(glsl_generator); + vkd3d_shader_free_scan_descriptor_info1(&scan_descriptor_info); break; case VKD3D_SHADER_TARGET_SPIRV_BINARY: case VKD3D_SHADER_TARGET_SPIRV_TEXT: + if ((ret = scan_with_parser(&scan_info, message_context, &scan_descriptor_info, parser)) < 0) + return ret; ret = spirv_compile(parser, &scan_descriptor_info, compile_info, out, message_context); + vkd3d_shader_free_scan_descriptor_info1(&scan_descriptor_info); break; default: /* Validation should prevent us from reaching this. */ - assert(0); + vkd3d_unreachable(); } - vkd3d_shader_free_scan_descriptor_info1(&scan_descriptor_info); return ret; }

1 0

Zebediah Figura : vkd3d-shader: Do not scan DCL instructions which do not declare resources.
by Alexandre Julliard 14 Dec '23

14 Dec '23

Module: vkd3d Branch: master Commit: 2bc40385d9b4a51a0d66462a6f708fbc888ecfa3 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/2bc40385d9b4a51a0d66462a6f708… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Wed Oct 25 13:58:13 2023 -0500 vkd3d-shader: Do not scan DCL instructions which do not declare resources. --- libs/vkd3d-shader/vkd3d_shader_main.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/libs/vkd3d-shader/vkd3d_shader_main.c b/libs/vkd3d-shader/vkd3d_shader_main.c index 0600e722..49ce886a 100644 --- a/libs/vkd3d-shader/vkd3d_shader_main.c +++ b/libs/vkd3d-shader/vkd3d_shader_main.c @@ -1073,6 +1073,9 @@ static int vkd3d_shader_scan_instruction(struct vkd3d_shader_scan_context *conte vkd3d_shader_scan_sampler_declaration(context, instruction); break; case VKD3DSIH_DCL: + if (instruction->declaration.semantic.resource_type == VKD3D_SHADER_RESOURCE_NONE) + break; + if (instruction->declaration.semantic.resource.reg.reg.type == VKD3DSPR_COMBINED_SAMPLER) { vkd3d_shader_scan_combined_sampler_declaration(context, &instruction->declaration.semantic);

1 0

Zebediah Figura : tests: Use struct vkd3d_shader_scan_signature_info to retrieve the VS input signature.
by Alexandre Julliard 14 Dec '23

14 Dec '23

Module: vkd3d Branch: master Commit: aa44f9b39029983b5535c92ae3c66622f2940b0b URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/aa44f9b39029983b5535c92ae3c66… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Fri Feb 24 13:50:11 2023 -0600 tests: Use struct vkd3d_shader_scan_signature_info to retrieve the VS input signature. In order to allow retrieving it from d3dbc shaders. --- tests/shader_runner_vulkan.c | 27 +++++++++++++++++---------- 1 file changed, 17 insertions(+), 10 deletions(-) diff --git a/tests/shader_runner_vulkan.c b/tests/shader_runner_vulkan.c index 6da2e86b..bb6b3a30 100644 --- a/tests/shader_runner_vulkan.c +++ b/tests/shader_runner_vulkan.c @@ -64,6 +64,8 @@ struct vulkan_shader_runner VkCommandBuffer cmd_buffer; VkDescriptorPool descriptor_pool; + struct vkd3d_shader_scan_signature_info vs_signatures; + struct vulkan_sampler { VkSampler vk_sampler; @@ -392,7 +394,7 @@ static void vulkan_runner_destroy_resource(struct shader_runner *r, struct resou free(resource); } -static bool compile_shader(const struct vulkan_shader_runner *runner, const char *source, const char *type, +static bool compile_shader(struct vulkan_shader_runner *runner, const char *source, const char *type, struct vkd3d_shader_code *dxbc, struct vkd3d_shader_code *spirv) { struct vkd3d_shader_spirv_target_info spirv_info = {.type = VKD3D_SHADER_STRUCTURE_TYPE_SPIRV_TARGET_INFO}; @@ -528,6 +530,14 @@ static bool compile_shader(const struct vulkan_shader_runner *runner, const char interface_info.push_constant_buffer_count = 1; interface_info.push_constant_buffers = &push_constants; + if (!strcmp(type, "vs")) + { + interface_info.next = &runner->vs_signatures; + + runner->vs_signatures.type = VKD3D_SHADER_STRUCTURE_TYPE_SCAN_SIGNATURE_INFO; + runner->vs_signatures.next = NULL; + } + ret = vkd3d_shader_compile(&info, spirv, &messages); if (messages && vkd3d_test_state.debug_level) trace("%s\n", messages); @@ -538,7 +548,7 @@ static bool compile_shader(const struct vulkan_shader_runner *runner, const char return true; } -static bool create_shader_stage(const struct vulkan_shader_runner *runner, +static bool create_shader_stage(struct vulkan_shader_runner *runner, VkPipelineShaderStageCreateInfo *stage_info, const char *type, enum VkShaderStageFlagBits stage, const char *source, struct vkd3d_shader_code *dxbc_ptr) { @@ -606,7 +616,7 @@ static VkPipelineLayout create_pipeline_layout(const struct vulkan_shader_runner return pipeline_layout; } -static VkPipeline create_graphics_pipeline(const struct vulkan_shader_runner *runner, VkRenderPass render_pass, +static VkPipeline create_graphics_pipeline(struct vulkan_shader_runner *runner, VkRenderPass render_pass, VkPipelineLayout pipeline_layout, D3D_PRIMITIVE_TOPOLOGY primitive_topology) { VkPipelineInputAssemblyStateCreateInfo ia_desc = {.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO}; @@ -622,7 +632,6 @@ static VkPipeline create_graphics_pipeline(const struct vulkan_shader_runner *ru VkPipelineColorBlendAttachmentState attachment_desc[MAX_RESOURCES] = {0}; VkVertexInputAttributeDescription input_attributes[32]; VkVertexInputBindingDescription input_bindings[32]; - struct vkd3d_shader_signature vs_input_signature; VkPipelineShaderStageCreateInfo stage_desc[2]; struct vkd3d_shader_code vs_dxbc; VkDevice device = runner->device; @@ -642,9 +651,6 @@ static VkPipeline create_graphics_pipeline(const struct vulkan_shader_runner *ru return VK_NULL_HANDLE; } - ret = vkd3d_shader_parse_input_signature(&vs_dxbc, &vs_input_signature, NULL); - ok(!ret, "Failed to parse input signature, error %d.\n", ret); - if (runner->r.input_element_count > ARRAY_SIZE(input_attributes)) fatal_error("Input element count %zu is too high.\n", runner->r.input_element_count); @@ -654,7 +660,8 @@ static VkPipeline create_graphics_pipeline(const struct vulkan_shader_runner *ru const struct input_element *element = &runner->r.input_elements[i]; const struct vkd3d_shader_signature_element *signature_element; - signature_element = vkd3d_shader_find_signature_element(&vs_input_signature, element->name, element->index, 0); + signature_element = vkd3d_shader_find_signature_element(&runner->vs_signatures.input, element->name, element->index, 0); + ok(signature_element, "Cannot find signature element %s%u.\n", element->name, element->index); attribute->location = signature_element->register_index; attribute->binding = element->slot; @@ -737,13 +744,13 @@ static VkPipeline create_graphics_pipeline(const struct vulkan_shader_runner *ru VK_CALL(vkDestroyShaderModule(device, stage_desc[0].module, NULL)); VK_CALL(vkDestroyShaderModule(device, stage_desc[1].module, NULL)); - vkd3d_shader_free_shader_signature(&vs_input_signature); + vkd3d_shader_free_scan_signature_info(&runner->vs_signatures); vkd3d_shader_free_shader_code(&vs_dxbc); return pipeline; } -static VkPipeline create_compute_pipeline(const struct vulkan_shader_runner *runner, VkPipelineLayout pipeline_layout) +static VkPipeline create_compute_pipeline(struct vulkan_shader_runner *runner, VkPipelineLayout pipeline_layout) { VkComputePipelineCreateInfo pipeline_desc = {.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO}; VkPipeline pipeline;

1 0

Zebediah Figura : tests: Avoid using "SV_Position" as a name for the vertex shader input.
by Alexandre Julliard 14 Dec '23

14 Dec '23

Module: vkd3d Branch: master Commit: 6a514ebe8ea459983aeee14f4a131d7c87e45e3c URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/6a514ebe8ea459983aeee14f4a131… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Mon Aug 15 19:53:55 2022 -0500 tests: Avoid using "SV_Position" as a name for the vertex shader input. We use vkd3d_shader_find_signature_element() in the Vulkan runner, and vkd3d-shader translates SM1 position to "POSITION". --- tests/hlsl/entry-point-semantics.shader_test | 21 ++++++++++++++------- tests/hlsl/for.shader_test | 3 ++- tests/hlsl/sample-bias.shader_test | 3 ++- tests/hlsl/struct-semantics.shader_test | 17 ++++++++++++++--- tests/hlsl/trigonometry.shader_test | 3 ++- tests/shader_runner.c | 5 +++-- 6 files changed, 37 insertions(+), 15 deletions(-) diff --git a/tests/hlsl/entry-point-semantics.shader_test b/tests/hlsl/entry-point-semantics.shader_test index d177b037..b87dca4d 100644 --- a/tests/hlsl/entry-point-semantics.shader_test +++ b/tests/hlsl/entry-point-semantics.shader_test @@ -2,9 +2,10 @@ % definition and its declarations. [vertex shader] -void main(out float tex : texcoord, inout float4 pos : sv_position) +void main(float4 pos : position, out float tex : texcoord, out float4 out_pos : sv_position) { tex = 0.2; + out_pos = pos; } [pixel shader fail] @@ -53,12 +54,13 @@ probe (0, 0) rgba (0.2, 0.2, 0.2, 0.2) [vertex shader] -void main(out float4 tex[4] : texcoord, inout float4 pos : sv_position) +void main(float4 pos : position, out float4 tex[4] : texcoord, out float4 out_pos : sv_position) { tex[0] = float4(10.0, 11.0, 12.0, 13.0); tex[1] = float4(20.0, 21.0, 22.0, 23.0); tex[2] = float4(30.0, 31.0, 32.0, 33.0); tex[3] = float4(40.0, 41.0, 42.0, 43.0); + out_pos = pos; } @@ -196,10 +198,11 @@ struct apple float4 tex[2] : TEXCOORD0; }; -void main(out apple apl, inout float4 pos : sv_position) +void main(float4 pos : position, out apple apl, out float4 out_pos : sv_position) { apl.tex[0] = float4(1, 2, 3, 4); apl.tex[1] = float4(10, 20, 30, 40); + out_pos = pos; } [pixel shader] @@ -220,10 +223,11 @@ struct apple float2 tex : TEXCOORD0; }; -void main(out apple apls[2], inout float4 pos : sv_position) +void main(float4 pos : position, out apple apls[2], out float4 out_pos : sv_position) { apls[0].tex = float2(1, 2); apls[1].tex = float2(3, 4); + out_pos = pos; } @@ -233,29 +237,32 @@ struct apple float2 f : SEMANTIC; }; -void main(out apple a, out apple b, inout float4 pos : sv_position) +void main(float4 pos : position, out apple a, out apple b, out float4 out_pos : sv_position) { a.f = float2(1, 2); b.f = float2(3, 4); + out_pos = pos; } % Semantic names are case-insensitive. [vertex shader fail] -void main(out float2 a : sem0, out float2 b : SEM, inout float4 pos : sv_position) +void main(float4 pos : position, out float2 a : sem0, out float2 b : SEM, out float4 out_pos : sv_position) { a = float2(1, 2); b = float2(3, 4); + out_pos = pos; } [vertex shader] -void main(out float4 tex[4] : texcoord, inout float4 pos : sv_position) +void main(float4 pos : position, out float4 tex[4] : texcoord, out float4 out_pos : sv_position) { tex[0] = float4(10.0, 11.0, 12.0, 13.0); tex[1] = float4(20.0, 21.0, 22.0, 23.0); tex[2] = float4(30.0, 31.0, 32.0, 33.0); tex[3] = float4(40.0, 41.0, 42.0, 43.0); + out_pos = pos; } diff --git a/tests/hlsl/for.shader_test b/tests/hlsl/for.shader_test index 1392118b..9407c81b 100644 --- a/tests/hlsl/for.shader_test +++ b/tests/hlsl/for.shader_test @@ -1,7 +1,8 @@ [vertex shader] -void main(out float tex : texcoord, inout float4 pos : sv_position) +void main(float4 pos : position, out float tex : texcoord, out float4 out_pos : sv_position) { tex = pos.x; + out_pos = pos; } [pixel shader] diff --git a/tests/hlsl/sample-bias.shader_test b/tests/hlsl/sample-bias.shader_test index e56945d0..5b7067b5 100644 --- a/tests/hlsl/sample-bias.shader_test +++ b/tests/hlsl/sample-bias.shader_test @@ -12,9 +12,10 @@ levels 2 0.0 0.0 1.0 0.0 [vertex shader] -void main(out float2 tex : texcoord, inout float4 pos : sv_position) +void main(float4 pos : position, out float2 tex : texcoord, out float4 out_pos : sv_position) { tex = pos.xy; + out_pos = pos; } [pixel shader] diff --git a/tests/hlsl/struct-semantics.shader_test b/tests/hlsl/struct-semantics.shader_test index 43b07bb8..c88006eb 100644 --- a/tests/hlsl/struct-semantics.shader_test +++ b/tests/hlsl/struct-semantics.shader_test @@ -1,6 +1,6 @@ [input layout] 0 r32g32b32a32 float texcoord -0 r32g32 float sv_position +0 r32g32 float position [vertex buffer 0] 0.0 1.0 0.0 1.0 -2.0 -2.0 @@ -10,7 +10,16 @@ [vertex shader] -struct vertex +struct in_vertex +{ + struct + { + float4 texcoord : texcoord; + float4 pos : position; + } m; +}; + +struct out_vertex { struct { @@ -19,8 +28,10 @@ struct vertex } m; }; -void main(inout struct vertex v) +void main(struct in_vertex i, out struct out_vertex o) { + o.m.pos = i.m.pos; + o.m.texcoord = i.m.texcoord; } [pixel shader] diff --git a/tests/hlsl/trigonometry.shader_test b/tests/hlsl/trigonometry.shader_test index f52d01de..85988119 100644 --- a/tests/hlsl/trigonometry.shader_test +++ b/tests/hlsl/trigonometry.shader_test @@ -1,7 +1,8 @@ [vertex shader] -void main(out float tex : texcoord, inout float4 pos : sv_position) +void main(float4 pos : position, out float tex : texcoord, out float4 out_pos : sv_position) { tex = (pos.x + 1) * 320; + out_pos = pos; } [pixel shader] diff --git a/tests/shader_runner.c b/tests/shader_runner.c index 4847ec75..9b373e01 100644 --- a/tests/shader_runner.c +++ b/tests/shader_runner.c @@ -579,8 +579,9 @@ static void parse_test_directive(struct shader_runner *runner, const char *line) }; static const char vs_source[] = - "void main(inout float4 position : sv_position)\n" + "float4 main(float4 pos : position) : sv_position\n" "{\n" + " return pos;\n" "}"; if (!shader_runner_get_resource(runner, RESOURCE_TYPE_RENDER_TARGET, 0)) @@ -601,7 +602,7 @@ static void parse_test_directive(struct shader_runner *runner, const char *line) vkd3d_array_reserve((void **)&runner->input_elements, &runner->input_element_capacity, 1, sizeof(*runner->input_elements)); element = &runner->input_elements[0]; - element->name = strdup("sv_position"); + element->name = strdup("position"); element->slot = 0; element->format = DXGI_FORMAT_R32G32_FLOAT; element->texel_size = sizeof(*quad);

1 0

Zebediah Figura : vkd3d-shader: Allow compiling d3d bytecode to SPIR-V.
by Alexandre Julliard 14 Dec '23

14 Dec '23

Module: vkd3d Branch: master Commit: 4ff389854ce7167f945864002317473bd8517687 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/4ff389854ce7167f9458640023174… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Sat Mar 12 16:57:27 2022 -0600 vkd3d-shader: Allow compiling d3d bytecode to SPIR-V. --- libs/vkd3d-shader/vkd3d_shader_main.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/libs/vkd3d-shader/vkd3d_shader_main.c b/libs/vkd3d-shader/vkd3d_shader_main.c index 49ce886a..de7a66e0 100644 --- a/libs/vkd3d-shader/vkd3d_shader_main.c +++ b/libs/vkd3d-shader/vkd3d_shader_main.c @@ -1626,14 +1626,10 @@ static int compile_d3d_bytecode(const struct vkd3d_shader_compile_info *compile_ return ret; } - if (compile_info->target_type == VKD3D_SHADER_TARGET_D3D_ASM) - { - ret = vkd3d_dxbc_binary_to_text(&parser->instructions, &parser->shader_version, compile_info, out, VSIR_ASM_D3D); - vkd3d_shader_parser_destroy(parser); - return ret; - } + ret = vkd3d_shader_parser_compile(parser, compile_info, out, message_context); - return VKD3D_ERROR; + vkd3d_shader_parser_destroy(parser); + return ret; } static int compile_dxbc_dxil(const struct vkd3d_shader_compile_info *compile_info, @@ -1911,6 +1907,10 @@ const enum vkd3d_shader_target_type *vkd3d_shader_get_supported_target_types( static const enum vkd3d_shader_target_type d3dbc_types[] = { + VKD3D_SHADER_TARGET_SPIRV_BINARY, +#ifdef HAVE_SPIRV_TOOLS + VKD3D_SHADER_TARGET_SPIRV_TEXT, +#endif VKD3D_SHADER_TARGET_D3D_ASM, };

1 0

Henri Verbeet : vkd3d-shader/ir: Pass a uint32_t swizzle to vkd3d_swizzle_get_component().
by Alexandre Julliard 14 Dec '23

14 Dec '23

Module: vkd3d Branch: master Commit: 8a1de71fb1e6e0b3ce59afe260708fdfe67b3bfa URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/8a1de71fb1e6e0b3ce59afe260708… Author: Henri Verbeet <hverbeet(a)codeweavers.com> Date: Tue Dec 5 23:05:40 2023 +0100 vkd3d-shader/ir: Pass a uint32_t swizzle to vkd3d_swizzle_get_component(). --- libs/vkd3d-shader/d3d_asm.c | 8 ++--- libs/vkd3d-shader/d3dbc.c | 10 +++---- libs/vkd3d-shader/ir.c | 4 +-- libs/vkd3d-shader/spirv.c | 50 ++++++++++++++++---------------- libs/vkd3d-shader/tpf.c | 18 ++++++------ libs/vkd3d-shader/vkd3d_shader_private.h | 7 ++--- 6 files changed, 48 insertions(+), 49 deletions(-)

1 0

Henri Verbeet : vkd3d-shader/ir: Pass a uint32_t swizzle to vkd3d_swizzle_get_component64().
by Alexandre Julliard 14 Dec '23

14 Dec '23

Module: vkd3d Branch: master Commit: f96a791807701a4583e6c613c2f5bd47c00ea767 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/f96a791807701a4583e6c613c2f5b… Author: Henri Verbeet <hverbeet(a)codeweavers.com> Date: Tue Dec 5 23:08:59 2023 +0100 vkd3d-shader/ir: Pass a uint32_t swizzle to vkd3d_swizzle_get_component64(). --- libs/vkd3d-shader/spirv.c | 4 ++-- libs/vkd3d-shader/vkd3d_shader_private.h | 3 +-- 2 files changed, 3 insertions(+), 4 deletions(-) diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index 9b628a36..45f46a91 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -3702,7 +3702,7 @@ static uint32_t spirv_compiler_emit_load_constant(struct spirv_compiler *compile } static uint32_t spirv_compiler_emit_load_constant64(struct spirv_compiler *compiler, - const struct vkd3d_shader_register *reg, DWORD swizzle, uint32_t write_mask) + const struct vkd3d_shader_register *reg, uint32_t swizzle, uint32_t write_mask) { unsigned int component_count = vsir_write_mask_component_count(write_mask); uint64_t values[VKD3D_DVEC2_SIZE] = {0}; @@ -3720,7 +3720,7 @@ static uint32_t spirv_compiler_emit_load_constant64(struct spirv_compiler *compi for (i = 0, j = 0; i < VKD3D_DVEC2_SIZE; ++i) { if (write_mask & (VKD3DSP_WRITEMASK_0 << i)) - values[j++] = reg->u.immconst_uint64[vkd3d_swizzle_get_component64(swizzle, i)]; + values[j++] = reg->u.immconst_uint64[vsir_swizzle_get_component64(swizzle, i)]; } } diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index 70c5f4fc..f1cc9ad7 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -1562,8 +1562,7 @@ static inline unsigned int vsir_swizzle_get_component(uint32_t swizzle, unsigned return (swizzle >> VKD3D_SHADER_SWIZZLE_SHIFT(idx)) & VKD3D_SHADER_SWIZZLE_MASK; } -static inline unsigned int vkd3d_swizzle_get_component64(DWORD swizzle, - unsigned int idx) +static inline unsigned int vsir_swizzle_get_component64(uint32_t swizzle, unsigned int idx) { return ((swizzle >> VKD3D_SHADER_SWIZZLE_SHIFT(idx * 2)) & VKD3D_SHADER_SWIZZLE_MASK) / 2u; }

1 0

Conor McCarthy : vkd3d: Update the descriptor `next` index before getting a reference for writing.
by Alexandre Julliard 14 Dec '23

14 Dec '23

Module: vkd3d Branch: master Commit: 70962ae7d86297fd4b54cd5b9e2d2fd197f3a4b7 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/70962ae7d86297fd4b54cd5b9e2d2… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Sun Jul 30 13:25:56 2023 +1000 vkd3d: Update the descriptor `next` index before getting a reference for writing. Fixes a race condition where the descriptor is modified between getting its object and resetting the `next` index. The new object would never be written. While it is invalid for the app to write descriptors used by a command list which has been submitted to a queue, unused descriptors may be written. This change also supports writing descriptors in a worker thread. --- libs/vkd3d/resource.c | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-) diff --git a/libs/vkd3d/resource.c b/libs/vkd3d/resource.c index abbdfbe2..0c9c911a 100644 --- a/libs/vkd3d/resource.c +++ b/libs/vkd3d/resource.c @@ -2632,20 +2632,18 @@ void d3d12_desc_flush_vk_heap_updates_locked(struct d3d12_descriptor_heap *descr for (; i != UINT_MAX; i = next) { src = &descriptors[i]; - next = (int)src->next >> 1; + next = vkd3d_atomic_exchange(&src->next, 0); + next = (int)next >> 1; + /* A race exists here between updating src->next and getting the current object. The best + * we can do is get the object last, which may result in a harmless rewrite later. */ u.object = d3d12_desc_get_object_ref(src, device); if (!u.object) - { - vkd3d_atomic_exchange(&src->next, 0); continue; - } writes.held_refs[writes.held_ref_count++] = u.object; d3d12_desc_write_vk_heap(descriptor_heap, i, &writes, u.object, device); - - vkd3d_atomic_exchange(&src->next, 0); } /* Avoid thunk calls wherever possible. */

1 0

Conor McCarthy : vkd3d: Write Vulkan descriptors in a worker thread.
by Alexandre Julliard 14 Dec '23

14 Dec '23

Module: vkd3d Branch: master Commit: 37e76618ca41b83f4691e9c8afafbc6ebef9228a URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/37e76618ca41b83f4691e9c8afafb… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Sun Jul 30 13:34:09 2023 +1000 vkd3d: Write Vulkan descriptors in a worker thread. Raises framerate by 5-10% in games which write thousands of descriptors per frame, e.g. Horizon Zero Dawn. The worker thread is a generic device worker which can also be used for other purposes if the need arises. --- libs/vkd3d/command.c | 4 ++ libs/vkd3d/device.c | 110 +++++++++++++++++++++++++++++++++++++++++++++ libs/vkd3d/resource.c | 9 ++++ libs/vkd3d/vkd3d_private.h | 10 +++++ 4 files changed, 133 insertions(+) diff --git a/libs/vkd3d/command.c b/libs/vkd3d/command.c index 15c8317b..549f6a45 100644 --- a/libs/vkd3d/command.c +++ b/libs/vkd3d/command.c @@ -2644,6 +2644,8 @@ static bool d3d12_command_list_update_compute_pipeline(struct d3d12_command_list { const struct vkd3d_vk_device_procs *vk_procs = &list->device->vk_procs; + vkd3d_cond_signal(&list->device->worker_cond); + if (list->current_pipeline != VK_NULL_HANDLE) return true; @@ -2665,6 +2667,8 @@ static bool d3d12_command_list_update_graphics_pipeline(struct d3d12_command_lis VkRenderPass vk_render_pass; VkPipeline vk_pipeline; + vkd3d_cond_signal(&list->device->worker_cond); + if (list->current_pipeline != VK_NULL_HANDLE) return true; diff --git a/libs/vkd3d/device.c b/libs/vkd3d/device.c index 5c801ca4..d6e74d01 100644 --- a/libs/vkd3d/device.c +++ b/libs/vkd3d/device.c @@ -2495,6 +2495,28 @@ static ULONG STDMETHODCALLTYPE d3d12_device_AddRef(ID3D12Device5 *iface) return refcount; } +static HRESULT device_worker_stop(struct d3d12_device *device) +{ + HRESULT hr; + + TRACE("device %p.\n", device); + + vkd3d_mutex_lock(&device->worker_mutex); + + device->worker_should_exit = true; + vkd3d_cond_signal(&device->worker_cond); + + vkd3d_mutex_unlock(&device->worker_mutex); + + if (FAILED(hr = vkd3d_join_thread(device->vkd3d_instance, &device->worker_thread))) + return hr; + + vkd3d_mutex_destroy(&device->worker_mutex); + vkd3d_cond_destroy(&device->worker_cond); + + return S_OK; +} + static ULONG STDMETHODCALLTYPE d3d12_device_Release(ID3D12Device5 *iface) { struct d3d12_device *device = impl_from_ID3D12Device5(iface); @@ -2520,6 +2542,9 @@ static ULONG STDMETHODCALLTYPE d3d12_device_Release(ID3D12Device5 *iface) d3d12_device_destroy_vkd3d_queues(device); vkd3d_desc_object_cache_cleanup(&device->view_desc_cache); vkd3d_desc_object_cache_cleanup(&device->cbuffer_desc_cache); + if (device->use_vk_heaps) + device_worker_stop(device); + vkd3d_free(device->heaps); VK_CALL(vkDestroyDevice(device->vk_device, NULL)); if (device->parent) IUnknown_Release(device->parent); @@ -4251,6 +4276,40 @@ struct d3d12_device *unsafe_impl_from_ID3D12Device5(ID3D12Device5 *iface) return impl_from_ID3D12Device5(iface); } +static void *device_worker_main(void *arg) +{ + struct d3d12_descriptor_heap *heap; + struct d3d12_device *device = arg; + size_t i; + + vkd3d_set_thread_name("device_worker"); + + vkd3d_mutex_lock(&device->worker_mutex); + + while (!device->worker_should_exit) + { + for (i = 0; i < device->heap_count; ++i) + { + /* Descriptor updates are not written to Vulkan descriptor sets until a command list + * is submitted to a queue, while the client is free to write d3d12 descriptors earlier, + * from any thread. This causes a delay right before command list execution, so + * handling these updates in a worker thread can speed up execution significantly. */ + heap = device->heaps[i]; + if (heap->dirty_list_head == UINT_MAX) + continue; + vkd3d_mutex_lock(&heap->vk_sets_mutex); + d3d12_desc_flush_vk_heap_updates_locked(heap, device); + vkd3d_mutex_unlock(&heap->vk_sets_mutex); + } + + vkd3d_cond_wait(&device->worker_cond, &device->worker_mutex); + } + + vkd3d_mutex_unlock(&device->worker_mutex); + + return NULL; +} + static HRESULT d3d12_device_init(struct d3d12_device *device, struct vkd3d_instance *instance, const struct vkd3d_device_create_info *create_info) { @@ -4270,6 +4329,14 @@ static HRESULT d3d12_device_init(struct d3d12_device *device, device->vk_device = VK_NULL_HANDLE; + device->heaps = NULL; + device->heap_capacity = 0; + device->heap_count = 0; + memset(&device->worker_thread, 0, sizeof(device->worker_thread)); + device->worker_should_exit = false; + vkd3d_mutex_init(&device->worker_mutex); + vkd3d_cond_init(&device->worker_cond); + if (FAILED(hr = vkd3d_create_vk_device(device, create_info))) goto out_free_instance; @@ -4291,6 +4358,13 @@ static HRESULT d3d12_device_init(struct d3d12_device *device, if (FAILED(hr = vkd3d_vk_descriptor_heap_layouts_init(device))) goto out_cleanup_uav_clear_state; + if (device->use_vk_heaps && FAILED(hr = vkd3d_create_thread(device->vkd3d_instance, + device_worker_main, device, &device->worker_thread))) + { + WARN("Failed to create worker thread, hr %#x.\n", hr); + goto out_cleanup_descriptor_heap_layouts; + } + vkd3d_render_pass_cache_init(&device->render_pass_cache); vkd3d_gpu_va_allocator_init(&device->gpu_va_allocator); vkd3d_time_domains_init(device); @@ -4308,6 +4382,8 @@ static HRESULT d3d12_device_init(struct d3d12_device *device, return S_OK; +out_cleanup_descriptor_heap_layouts: + vkd3d_vk_descriptor_heap_layouts_cleanup(device); out_cleanup_uav_clear_state: vkd3d_uav_clear_state_cleanup(&device->uav_clear_state, device); out_destroy_null_resources: @@ -4361,6 +4437,40 @@ void d3d12_device_mark_as_removed(struct d3d12_device *device, HRESULT reason, device->removed_reason = reason; } +HRESULT d3d12_device_add_descriptor_heap(struct d3d12_device *device, struct d3d12_descriptor_heap *heap) +{ + vkd3d_mutex_lock(&device->worker_mutex); + + if (!vkd3d_array_reserve((void **)&device->heaps, &device->heap_capacity, device->heap_count + 1, + sizeof(*device->heaps))) + { + vkd3d_mutex_unlock(&device->worker_mutex); + return E_OUTOFMEMORY; + } + device->heaps[device->heap_count++] = heap; + + vkd3d_mutex_unlock(&device->worker_mutex); + + return S_OK; +} + +void d3d12_device_remove_descriptor_heap(struct d3d12_device *device, struct d3d12_descriptor_heap *heap) +{ + size_t i; + + vkd3d_mutex_lock(&device->worker_mutex); + + for (i = 0; i < device->heap_count; ++i) + { + if (device->heaps[i] == heap) + { + device->heaps[i] = device->heaps[--device->heap_count]; + break; + } + } + + vkd3d_mutex_unlock(&device->worker_mutex); +} #ifdef _WIN32 struct thread_data diff --git a/libs/vkd3d/resource.c b/libs/vkd3d/resource.c index 0c9c911a..609c6710 100644 --- a/libs/vkd3d/resource.c +++ b/libs/vkd3d/resource.c @@ -3995,6 +3995,9 @@ static ULONG STDMETHODCALLTYPE d3d12_descriptor_heap_Release(ID3D12DescriptorHea { struct d3d12_desc *descriptors = (struct d3d12_desc *)heap->descriptors; + if (heap->use_vk_heaps) + d3d12_device_remove_descriptor_heap(device, heap); + for (i = 0; i < heap->desc.NumDescriptors; ++i) { d3d12_desc_destroy(&descriptors[i], device); @@ -4318,6 +4321,12 @@ HRESULT d3d12_descriptor_heap_create(struct d3d12_device *device, dst[i].next = 0; } object->dirty_list_head = UINT_MAX; + + if (object->use_vk_heaps && FAILED(hr = d3d12_device_add_descriptor_heap(device, object))) + { + vkd3d_free(object); + return hr; + } } else { diff --git a/libs/vkd3d/vkd3d_private.h b/libs/vkd3d/vkd3d_private.h index e0eb9f3d..025ad534 100644 --- a/libs/vkd3d/vkd3d_private.h +++ b/libs/vkd3d/vkd3d_private.h @@ -1804,6 +1804,14 @@ struct d3d12_device unsigned int vk_pool_count; struct vkd3d_vk_descriptor_heap_layout vk_descriptor_heap_layouts[VKD3D_SET_INDEX_COUNT]; bool use_vk_heaps; + + struct d3d12_descriptor_heap **heaps; + size_t heap_capacity; + size_t heap_count; + union vkd3d_thread_handle worker_thread; + struct vkd3d_mutex worker_mutex; + struct vkd3d_cond worker_cond; + bool worker_should_exit; }; HRESULT d3d12_device_create(struct vkd3d_instance *instance, @@ -1813,6 +1821,8 @@ bool d3d12_device_is_uma(struct d3d12_device *device, bool *coherent); void d3d12_device_mark_as_removed(struct d3d12_device *device, HRESULT reason, const char *message, ...) VKD3D_PRINTF_FUNC(3, 4); struct d3d12_device *unsafe_impl_from_ID3D12Device5(ID3D12Device5 *iface); +HRESULT d3d12_device_add_descriptor_heap(struct d3d12_device *device, struct d3d12_descriptor_heap *heap); +void d3d12_device_remove_descriptor_heap(struct d3d12_device *device, struct d3d12_descriptor_heap *heap); static inline HRESULT d3d12_device_query_interface(struct d3d12_device *device, REFIID iid, void **object) {

1 0

Conor McCarthy : vkd3d: Rename the device mutex to pipeline_cache_mutex.
by Alexandre Julliard 14 Dec '23

14 Dec '23

Module: vkd3d Branch: master Commit: a2741babd95c6f6b908c3f0d117219c432618431 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/a2741babd95c6f6b908c3f0d11721… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Mon Dec 4 15:53:51 2023 +1000 vkd3d: Rename the device mutex to pipeline_cache_mutex. It is used only for pipeline and render pass caching, and renaming it helps prevent inappropriate or erroneous use. --- libs/vkd3d/device.c | 4 ++-- libs/vkd3d/state.c | 12 ++++++------ libs/vkd3d/vkd3d_private.h | 3 ++- 3 files changed, 10 insertions(+), 9 deletions(-) diff --git a/libs/vkd3d/device.c b/libs/vkd3d/device.c index d6e74d01..69a46e91 100644 --- a/libs/vkd3d/device.c +++ b/libs/vkd3d/device.c @@ -2056,7 +2056,7 @@ static HRESULT d3d12_device_init_pipeline_cache(struct d3d12_device *device) VkPipelineCacheCreateInfo cache_info; VkResult vr; - vkd3d_mutex_init(&device->mutex); + vkd3d_mutex_init(&device->pipeline_cache_mutex); cache_info.sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO; cache_info.pNext = NULL; @@ -2080,7 +2080,7 @@ static void d3d12_device_destroy_pipeline_cache(struct d3d12_device *device) if (device->vk_pipeline_cache) VK_CALL(vkDestroyPipelineCache(device->vk_device, device->vk_pipeline_cache, NULL)); - vkd3d_mutex_destroy(&device->mutex); + vkd3d_mutex_destroy(&device->pipeline_cache_mutex); } #define VKD3D_VA_FALLBACK_BASE 0x8000000000000000ull diff --git a/libs/vkd3d/state.c b/libs/vkd3d/state.c index 9a039452..626d6d62 100644 --- a/libs/vkd3d/state.c +++ b/libs/vkd3d/state.c @@ -1691,7 +1691,7 @@ HRESULT vkd3d_render_pass_cache_find(struct vkd3d_render_pass_cache *cache, HRESULT hr = S_OK; unsigned int i; - vkd3d_mutex_lock(&device->mutex); + vkd3d_mutex_lock(&device->pipeline_cache_mutex); for (i = 0; i < cache->render_pass_count; ++i) { @@ -1708,7 +1708,7 @@ HRESULT vkd3d_render_pass_cache_find(struct vkd3d_render_pass_cache *cache, if (!found) hr = vkd3d_render_pass_cache_create_pass_locked(cache, device, key, vk_render_pass); - vkd3d_mutex_unlock(&device->mutex); + vkd3d_mutex_unlock(&device->pipeline_cache_mutex); return hr; } @@ -3615,7 +3615,7 @@ static VkPipeline d3d12_pipeline_state_find_compiled_pipeline(const struct d3d12 *vk_render_pass = VK_NULL_HANDLE; - vkd3d_mutex_lock(&device->mutex); + vkd3d_mutex_lock(&device->pipeline_cache_mutex); LIST_FOR_EACH_ENTRY(current, &graphics->compiled_pipelines, struct vkd3d_compiled_pipeline, entry) { @@ -3627,7 +3627,7 @@ static VkPipeline d3d12_pipeline_state_find_compiled_pipeline(const struct d3d12 } } - vkd3d_mutex_unlock(&device->mutex); + vkd3d_mutex_unlock(&device->pipeline_cache_mutex); return vk_pipeline; } @@ -3646,7 +3646,7 @@ static bool d3d12_pipeline_state_put_pipeline_to_cache(struct d3d12_pipeline_sta compiled_pipeline->vk_pipeline = vk_pipeline; compiled_pipeline->vk_render_pass = vk_render_pass; - vkd3d_mutex_lock(&device->mutex); + vkd3d_mutex_lock(&device->pipeline_cache_mutex); LIST_FOR_EACH_ENTRY(current, &graphics->compiled_pipelines, struct vkd3d_compiled_pipeline, entry) { @@ -3661,7 +3661,7 @@ static bool d3d12_pipeline_state_put_pipeline_to_cache(struct d3d12_pipeline_sta if (compiled_pipeline) list_add_tail(&graphics->compiled_pipelines, &compiled_pipeline->entry); - vkd3d_mutex_unlock(&device->mutex); + vkd3d_mutex_unlock(&device->pipeline_cache_mutex); return compiled_pipeline; } diff --git a/libs/vkd3d/vkd3d_private.h b/libs/vkd3d/vkd3d_private.h index 025ad534..4c5c039a 100644 --- a/libs/vkd3d/vkd3d_private.h +++ b/libs/vkd3d/vkd3d_private.h @@ -1757,9 +1757,10 @@ struct d3d12_device struct vkd3d_gpu_va_allocator gpu_va_allocator; - struct vkd3d_mutex mutex; struct vkd3d_desc_object_cache view_desc_cache; struct vkd3d_desc_object_cache cbuffer_desc_cache; + + struct vkd3d_mutex pipeline_cache_mutex; struct vkd3d_render_pass_cache render_pass_cache; VkPipelineCache vk_pipeline_cache;

1 0