[PATCH v7 0/3] MR777: vkd3d-shader/dxil: Support SV_Depth, SV_DepthGreaterEqual and SV_DepthLessEqual.
Includes !782 to prevent the 32-bit Windows CI crashing. -- v7: vkd3d-shader/dxil: Support SV_Depth, SV_DepthGreaterEqual and SV_DepthLessEqual. tests/hlsl: Add an SV_Depth test. https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/777
From: Conor McCarthy <cmccarthy(a)codeweavers.com> Fixes invalid read in the GL runner. --- tests/hlsl/calculate-lod.shader_test | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/tests/hlsl/calculate-lod.shader_test b/tests/hlsl/calculate-lod.shader_test index 88325af61..74533cfeb 100644 --- a/tests/hlsl/calculate-lod.shader_test +++ b/tests/hlsl/calculate-lod.shader_test @@ -15,6 +15,10 @@ levels 3 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 +0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 +0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 + +0.5 0.5 0.5 0.5 [pixel shader todo] sampler s; -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/777
From: Conor McCarthy <cmccarthy(a)codeweavers.com> --- Makefile.am | 1 + tests/d3d12_test_utils.h | 6 + tests/hlsl/depth-out.shader_test | 102 +++++++++++++++++ tests/shader_runner.c | 110 ++++++++++++------ tests/shader_runner.h | 6 + tests/shader_runner_d3d11.c | 30 ++++- tests/shader_runner_d3d12.c | 38 ++++++- tests/shader_runner_d3d9.c | 7 ++ tests/shader_runner_gl.c | 16 ++- tests/shader_runner_vulkan.c | 185 +++++++++++++++++++++---------- 10 files changed, 410 insertions(+), 91 deletions(-) create mode 100644 tests/hlsl/depth-out.shader_test diff --git a/Makefile.am b/Makefile.am index f823cbc85..ccc4af392 100644 --- a/Makefile.am +++ b/Makefile.am @@ -87,6 +87,7 @@ vkd3d_shader_tests = \ tests/hlsl/cross.shader_test \ tests/hlsl/d3dcolor-to-ubyte4.shader_test \ tests/hlsl/ddxddy.shader_test \ + tests/hlsl/depth-out.shader_test \ tests/hlsl/determinant.shader_test \ tests/hlsl/discard.shader_test \ tests/hlsl/distance.shader_test \ diff --git a/tests/d3d12_test_utils.h b/tests/d3d12_test_utils.h index 4fe6fa716..60512aec6 100644 --- a/tests/d3d12_test_utils.h +++ b/tests/d3d12_test_utils.h @@ -1245,6 +1245,12 @@ static inline D3D12_CPU_DESCRIPTOR_HANDLE get_cpu_rtv_handle(struct test_context return get_cpu_handle(context->device, heap, D3D12_DESCRIPTOR_HEAP_TYPE_RTV, offset); } +static inline D3D12_CPU_DESCRIPTOR_HANDLE get_cpu_dsv_handle(struct test_context *context, + ID3D12DescriptorHeap *heap, unsigned int offset) +{ + return get_cpu_handle(context->device, heap, D3D12_DESCRIPTOR_HEAP_TYPE_DSV, offset); +} + static inline D3D12_GPU_DESCRIPTOR_HANDLE get_gpu_descriptor_handle(struct test_context *context, ID3D12DescriptorHeap *heap, unsigned int offset) { diff --git a/tests/hlsl/depth-out.shader_test b/tests/hlsl/depth-out.shader_test new file mode 100644 index 000000000..b26830fc6 --- /dev/null +++ b/tests/hlsl/depth-out.shader_test @@ -0,0 +1,102 @@ +[require] +shader model >= 4.0 + +[dsv] +size (2d, 640, 480) + +[pixel shader] +float depth; + +float main() : SV_Depth +{ + return depth; +} + +[test] +uniform 0 float 0.0 +dsv clear 1.0 +depth less +todo(sm>=6 | glsl) draw quad +probe dsv all r (0.0) + +uniform 0 float 0.75 +todo(sm>=6 | glsl) draw quad +probe dsv all r (0.75) + +dsv clear 0.5 +depth greater +todo(sm>=6 | glsl) draw quad +probe dsv all r (0.75) + +depth less +todo(sm>=6 | glsl) draw quad +probe dsv all r (0.5) + + +[require] +shader model >= 5.0 + +[rtv 0] +format r32g32b32a32 float +size (2d, 640, 480) + +[vertex shader] +float2 depth; + +void main(float4 in_position : POSITION, out float4 out_position : SV_Position) +{ + out_position = in_position; + out_position.z = depth.x; +} + +[pixel shader todo] +float2 depth; + +float4 main(out float out_depth : SV_DepthLessEqual) : SV_Target +{ + out_depth = depth.y; + return float4(0.0f, 1.0f, 0.0f, 1.0f); +} + +[test] +uniform 0 float4 0.7 0.7 0.0 0.0 +rtv clear 1.0 1.0 1.0 1.0 +dsv clear 0.5 +depth greater equal +todo draw quad +probe all rgba (0.0, 1.0, 0.0, 1.0) +probe dsv all r (0.7) + +uniform 0 float4 0.7 0.4 0.0 0.0 +todo draw quad +probe all rgba (1.0, 1.0, 1.0, 1.0) +probe dsv all r (0.5) + + +[pixel shader todo] +float2 depth; + +float4 main(out float out_depth : SV_DepthGreaterEqual) : SV_Target +{ + out_depth = depth.y; + return float4(0.0f, 1.0f, 0.0f, 1.0f); +} + +[test] +uniform 0 float4 0.7 0.7 0.0 0.0 +rtv clear 1.0 1.0 1.0 1.0 +dsv clear 0.5 +depth greater equal +todo draw quad +probe all rgba (0.0, 1.0, 0.0, 1.0) +probe dsv all r (0.7) + +uniform 0 float4 0.4 0.6 0.0 0.0 +todo draw quad +probe all rgba (0.0, 1.0, 0.0, 1.0) +probe dsv all r (0.6) + +uniform 0 float4 0.4 0.4 0.0 0.0 +todo draw quad +probe all rgba (1.0, 1.0, 1.0, 1.0) +probe dsv all r (0.5) diff --git a/tests/shader_runner.c b/tests/shader_runner.c index f800ded32..416322389 100644 --- a/tests/shader_runner.c +++ b/tests/shader_runner.c @@ -381,6 +381,35 @@ static DXGI_FORMAT parse_format(const char *line, enum texture_data_type *data_t fatal_error("Unknown format '%s'.\n", line); } +static D3D12_COMPARISON_FUNC parse_comparison_func(const char *line) +{ + static const struct + { + const char *string; + D3D12_COMPARISON_FUNC func; + } + funcs[] = + { + {"less equal", D3D12_COMPARISON_FUNC_LESS_EQUAL}, + {"not equal", D3D12_COMPARISON_FUNC_NOT_EQUAL}, + {"greater equal", D3D12_COMPARISON_FUNC_GREATER_EQUAL}, + {"never", D3D12_COMPARISON_FUNC_NEVER}, + {"less", D3D12_COMPARISON_FUNC_LESS}, + {"equal", D3D12_COMPARISON_FUNC_EQUAL}, + {"greater", D3D12_COMPARISON_FUNC_GREATER}, + {"always", D3D12_COMPARISON_FUNC_ALWAYS}, + }; + unsigned int i; + + for (i = 0; i < ARRAY_SIZE(funcs); ++i) + { + if (match_string(line, funcs[i].string, &line)) + return funcs[i].func; + } + + fatal_error("Unknown comparison func '%s'.\n", line); +} + static D3D12_TEXTURE_ADDRESS_MODE parse_sampler_address_mode(const char *line, const char **rest) { if (match_string(line, "border", rest)) @@ -440,35 +469,9 @@ static void parse_sampler_directive(struct sampler *sampler, const char *line) } else if (match_string(line, "comparison", &line)) { - static const struct - { - const char *string; - D3D12_COMPARISON_FUNC func; - } - funcs[] = - { - {"less equal", D3D12_COMPARISON_FUNC_LESS_EQUAL}, - {"not equal", D3D12_COMPARISON_FUNC_NOT_EQUAL}, - {"greater equal", D3D12_COMPARISON_FUNC_GREATER_EQUAL}, - {"never", D3D12_COMPARISON_FUNC_NEVER}, - {"less", D3D12_COMPARISON_FUNC_LESS}, - {"equal", D3D12_COMPARISON_FUNC_EQUAL}, - {"greater", D3D12_COMPARISON_FUNC_GREATER}, - {"always", D3D12_COMPARISON_FUNC_ALWAYS}, - }; - unsigned int i; - - for (i = 0; i < ARRAY_SIZE(funcs); ++i) - { - if (match_string(line, funcs[i].string, &line)) - { - sampler->filter |= D3D12_FILTER_REDUCTION_TYPE_COMPARISON << D3D12_FILTER_REDUCTION_TYPE_SHIFT; - sampler->func = funcs[i].func; - return; - } - } - - fatal_error("Unknown sampler func '%s'.\n", line); + sampler->filter |= D3D12_FILTER_REDUCTION_TYPE_COMPARISON << D3D12_FILTER_REDUCTION_TYPE_SHIFT; + sampler->func = parse_comparison_func(line); + return; } else { @@ -618,6 +621,12 @@ struct resource *shader_runner_get_resource(struct shader_runner *runner, enum r return NULL; } +static bool shader_runner_has_target(struct shader_runner *runner) +{ + return shader_runner_get_resource(runner, RESOURCE_TYPE_RENDER_TARGET, 0) + || shader_runner_get_resource(runner, RESOURCE_TYPE_DEPTH_STENCIL, 0); +} + static void set_resource(struct shader_runner *runner, struct resource *resource) { size_t i; @@ -798,6 +807,27 @@ static void parse_test_directive(struct shader_runner *runner, const char *line) runner->last_render_failed = !runner->ops->dispatch(runner, x, y, z); } + else if (match_string(line, "rtv clear", &line)) + { + struct vec4 v; + + if (sscanf(line, "%f %f %f %f", &v.x, &v.y, &v.z, &v.w) < 4) + fatal_error("Malformed rtv clear arguments '%s'.\n", line); + memcpy(runner->rtv_clear_value, &v, sizeof(runner->rtv_clear_value)); + runner->clear_rtv = true; + } + else if (match_string(line, "dsv clear", &line)) + { + float x; + + if (!sscanf(line, "%f", &x)) + fatal_error("Malformed dsv clear arguments '%s'.\n", line); + runner->dsv_clear_value = x; + } + else if (match_string(line, "depth", &line)) + { + runner->comparison = parse_comparison_func(line); + } else if (match_string(line, "draw quad", &line)) { struct resource_params params; @@ -818,7 +848,7 @@ static void parse_test_directive(struct shader_runner *runner, const char *line) " return pos;\n" "}"; - if (!shader_runner_get_resource(runner, RESOURCE_TYPE_RENDER_TARGET, 0)) + if (!shader_runner_has_target(runner)) { memset(¶ms, 0, sizeof(params)); params.slot = 0; @@ -868,7 +898,7 @@ static void parse_test_directive(struct shader_runner *runner, const char *line) struct resource_params params; unsigned int vertex_count; - if (!shader_runner_get_resource(runner, RESOURCE_TYPE_RENDER_TARGET, 0)) + if (!shader_runner_has_target(runner)) { memset(¶ms, 0, sizeof(params)); params.slot = 0; @@ -929,6 +959,10 @@ static void parse_test_directive(struct shader_runner *runner, const char *line) resource = shader_runner_get_resource(runner, RESOURCE_TYPE_RENDER_TARGET, slot); } + else if (match_string(line, "dsv", &line)) + { + resource = shader_runner_get_resource(runner, RESOURCE_TYPE_DEPTH_STENCIL, 0); + } else { resource = shader_runner_get_resource(runner, RESOURCE_TYPE_RENDER_TARGET, 0); @@ -1734,6 +1768,20 @@ void run_shader_tests(struct shader_runner *runner, const struct shader_runner_c current_resource.texel_size = 16; current_resource.level_count = 1; } + else if (!strcmp(line, "[dsv]\n")) + { + state = STATE_RESOURCE; + + memset(¤t_resource, 0, sizeof(current_resource)); + + current_resource.slot = 0; + current_resource.type = RESOURCE_TYPE_DEPTH_STENCIL; + current_resource.format = DXGI_FORMAT_D32_FLOAT; + current_resource.is_shadow = true; + current_resource.data_type = TEXTURE_DATA_FLOAT; + current_resource.texel_size = 4; + current_resource.level_count = 1; + } else if (sscanf(line, "[srv %u]\n", &index)) { state = STATE_RESOURCE; diff --git a/tests/shader_runner.h b/tests/shader_runner.h index 2f7e8fb25..ecc59f75e 100644 --- a/tests/shader_runner.h +++ b/tests/shader_runner.h @@ -69,6 +69,7 @@ struct sampler enum resource_type { RESOURCE_TYPE_RENDER_TARGET, + RESOURCE_TYPE_DEPTH_STENCIL, RESOURCE_TYPE_TEXTURE, RESOURCE_TYPE_UAV, RESOURCE_TYPE_VERTEX_BUFFER, @@ -178,6 +179,11 @@ struct shader_runner size_t input_element_count, input_element_capacity; unsigned int compile_options; + + bool clear_rtv; + float rtv_clear_value[4]; + float dsv_clear_value; + D3D12_COMPARISON_FUNC comparison; }; struct shader_runner_ops diff --git a/tests/shader_runner_d3d11.c b/tests/shader_runner_d3d11.c index b26d6a36b..46b8bd5d3 100644 --- a/tests/shader_runner_d3d11.c +++ b/tests/shader_runner_d3d11.c @@ -47,6 +47,7 @@ struct d3d11_resource ID3D11Buffer *buffer; ID3D11Texture2D *texture; ID3D11RenderTargetView *rtv; + ID3D11DepthStencilView *dsv; ID3D11ShaderResourceView *srv; ID3D11UnorderedAccessView *uav; bool is_uav_counter; @@ -377,6 +378,8 @@ static void init_resource_2d(struct d3d11_shader_runner *runner, struct d3d11_re desc.BindFlags = D3D11_BIND_UNORDERED_ACCESS; else if (params->type == RESOURCE_TYPE_RENDER_TARGET) desc.BindFlags = D3D11_BIND_RENDER_TARGET; + else if (params->type == RESOURCE_TYPE_DEPTH_STENCIL) + desc.BindFlags = D3D11_BIND_DEPTH_STENCIL; else desc.BindFlags = D3D11_BIND_SHADER_RESOURCE; @@ -407,6 +410,8 @@ static void init_resource_2d(struct d3d11_shader_runner *runner, struct d3d11_re hr = ID3D11Device_CreateUnorderedAccessView(device, resource->resource, NULL, &resource->uav); else if (params->type == RESOURCE_TYPE_RENDER_TARGET) hr = ID3D11Device_CreateRenderTargetView(device, resource->resource, NULL, &resource->rtv); + else if (params->type == RESOURCE_TYPE_DEPTH_STENCIL) + hr = ID3D11Device_CreateDepthStencilView(device, resource->resource, NULL, &resource->dsv); else hr = ID3D11Device_CreateShaderResourceView(device, resource->resource, NULL, &resource->srv); ok(hr == S_OK, "Failed to create view, hr %#lx.\n", hr); @@ -462,6 +467,7 @@ static struct resource *d3d11_runner_create_resource(struct shader_runner *r, co switch (params->type) { case RESOURCE_TYPE_RENDER_TARGET: + case RESOURCE_TYPE_DEPTH_STENCIL: case RESOURCE_TYPE_TEXTURE: if (params->dimension == RESOURCE_DIMENSION_BUFFER) init_resource_srv_buffer(runner, resource, params); @@ -492,6 +498,8 @@ static void d3d11_runner_destroy_resource(struct shader_runner *r, struct resour ID3D11Resource_Release(resource->resource); if (resource->rtv) ID3D11RenderTargetView_Release(resource->rtv); + if (resource->dsv) + ID3D11DepthStencilView_Release(resource->dsv); if (resource->srv) ID3D11ShaderResourceView_Release(resource->srv); if (resource->uav) @@ -561,6 +569,7 @@ static bool d3d11_runner_dispatch(struct shader_runner *r, unsigned int x, unsig break; case RESOURCE_TYPE_RENDER_TARGET: + case RESOURCE_TYPE_DEPTH_STENCIL: case RESOURCE_TYPE_VERTEX_BUFFER: break; } @@ -592,7 +601,10 @@ static bool d3d11_runner_draw(struct shader_runner *r, struct d3d11_shader_runner *runner = d3d11_shader_runner(r); ID3D11DeviceContext *context = runner->immediate_context; unsigned int min_uav_slot = ARRAY_SIZE(uavs); + ID3D11DepthStencilState *ds_state = NULL; + D3D11_DEPTH_STENCIL_DESC ds_desc = {0}; ID3D11Device *device = runner->device; + ID3D11DepthStencilView *dsv = NULL; ID3D10Blob *vs_code, *ps_code; unsigned int rtv_count = 0; ID3D11Buffer *cb = NULL; @@ -637,6 +649,19 @@ static bool d3d11_runner_draw(struct shader_runner *r, case RESOURCE_TYPE_RENDER_TARGET: rtvs[resource->r.slot] = resource->rtv; rtv_count = max(rtv_count, resource->r.slot + 1); + if (runner->r.clear_rtv) + ID3D11DeviceContext_ClearRenderTargetView(context, resource->rtv, resource->r.rtv_clear_value); + break; + + case RESOURCE_TYPE_DEPTH_STENCIL: + dsv = resource->dsv; + ID3D11DeviceContext_ClearDepthStencilView(context, dsv, D3D11_CLEAR_DEPTH, runner->r.dsv_clear_value, 0); + ds_desc.DepthEnable = TRUE; + ds_desc.DepthWriteMask = D3D11_DEPTH_WRITE_MASK_ALL; + ds_desc.DepthFunc = runner->r.comparison; + hr = ID3D11Device_CreateDepthStencilState(device, &ds_desc, &ds_state); + ok(hr == S_OK, "Failed to create depth/stencil state, hr %#lx.\n", hr); + ID3D11DeviceContext_OMSetDepthStencilState(context, ds_state, 0); break; case RESOURCE_TYPE_TEXTURE: @@ -655,7 +680,7 @@ static bool d3d11_runner_draw(struct shader_runner *r, } } - ID3D11DeviceContext_OMSetRenderTargetsAndUnorderedAccessViews(context, rtv_count, rtvs, NULL, + ID3D11DeviceContext_OMSetRenderTargetsAndUnorderedAccessViews(context, rtv_count, rtvs, dsv, min_uav_slot, ARRAY_SIZE(uavs) - min_uav_slot, &uavs[min_uav_slot], NULL); for (i = 0; i < runner->r.sampler_count; ++i) @@ -705,6 +730,8 @@ static bool d3d11_runner_draw(struct shader_runner *r, ID3D11VertexShader_Release(vs); if (cb) ID3D11Buffer_Release(cb); + if (ds_state) + ID3D11DepthStencilState_Release(ds_state); return true; } @@ -728,6 +755,7 @@ static struct resource_readback *d3d11_runner_get_resource_readback(struct shade switch (resource->r.type) { case RESOURCE_TYPE_RENDER_TARGET: + case RESOURCE_TYPE_DEPTH_STENCIL: case RESOURCE_TYPE_UAV: if (resource->r.dimension == RESOURCE_DIMENSION_BUFFER) { diff --git a/tests/shader_runner_d3d12.c b/tests/shader_runner_d3d12.c index 56844ae36..c95f7066e 100644 --- a/tests/shader_runner_d3d12.c +++ b/tests/shader_runner_d3d12.c @@ -46,7 +46,7 @@ struct d3d12_shader_runner struct test_context test_context; - ID3D12DescriptorHeap *heap, *rtv_heap; + ID3D12DescriptorHeap *heap, *rtv_heap, *dsv_heap; ID3D12CommandQueue *compute_queue; ID3D12CommandAllocator *compute_allocator; @@ -142,6 +142,16 @@ static struct resource *d3d12_runner_create_resource(struct shader_runner *r, co NULL, get_cpu_rtv_handle(test_context, runner->rtv_heap, resource->r.slot)); break; + case RESOURCE_TYPE_DEPTH_STENCIL: + if (!runner->dsv_heap) + runner->dsv_heap = create_cpu_descriptor_heap(device, D3D12_DESCRIPTOR_HEAP_TYPE_DSV, 1); + + resource->resource = create_default_texture2d(device, params->width, params->height, 1, params->level_count, + params->format, D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL, D3D12_RESOURCE_STATE_DEPTH_WRITE); + ID3D12Device_CreateDepthStencilView(device, resource->resource, + NULL, get_cpu_dsv_handle(test_context, runner->dsv_heap, 0)); + break; + case RESOURCE_TYPE_TEXTURE: if (!runner->heap) runner->heap = create_gpu_descriptor_heap(device, @@ -309,6 +319,7 @@ static ID3D12RootSignature *d3d12_runner_create_root_signature(struct d3d12_shad break; case RESOURCE_TYPE_RENDER_TARGET: + case RESOURCE_TYPE_DEPTH_STENCIL: case RESOURCE_TYPE_VERTEX_BUFFER: break; } @@ -406,6 +417,7 @@ static bool d3d12_runner_dispatch(struct shader_runner *r, unsigned int x, unsig break; case RESOURCE_TYPE_RENDER_TARGET: + case RESOURCE_TYPE_DEPTH_STENCIL: case RESOURCE_TYPE_VERTEX_BUFFER: break; } @@ -439,6 +451,7 @@ static bool d3d12_runner_draw(struct shader_runner *r, ID3D12CommandQueue *queue = test_context->queue; D3D12_INPUT_ELEMENT_DESC *input_element_descs; ID3D12Device *device = test_context->device; + D3D12_CPU_DESCRIPTOR_HANDLE dsv = {0}; ID3D10Blob *vs_code, *ps_code; unsigned int uniform_index; unsigned int rtv_count = 0; @@ -475,6 +488,14 @@ static bool d3d12_runner_draw(struct shader_runner *r, pso_desc.NumRenderTargets = max(pso_desc.NumRenderTargets, resource->r.slot + 1); pso_desc.BlendState.RenderTarget[resource->r.slot].RenderTargetWriteMask = D3D12_COLOR_WRITE_ENABLE_ALL; } + else if (resource->r.type == RESOURCE_TYPE_DEPTH_STENCIL) + { + assert(!resource->r.slot); + pso_desc.DSVFormat = resource->r.format; + pso_desc.DepthStencilState.DepthEnable = true; + pso_desc.DepthStencilState.DepthWriteMask = D3D12_DEPTH_WRITE_MASK_ALL; + pso_desc.DepthStencilState.DepthFunc = runner->r.comparison; + } } pso_desc.VS.pShaderBytecode = ID3D10Blob_GetBufferPointer(vs_code); @@ -533,6 +554,15 @@ static bool d3d12_runner_draw(struct shader_runner *r, case RESOURCE_TYPE_RENDER_TARGET: rtvs[resource->r.slot] = get_cpu_rtv_handle(test_context, runner->rtv_heap, resource->r.slot); rtv_count = max(rtv_count, resource->r.slot + 1); + if (runner->r.clear_rtv) + ID3D12GraphicsCommandList_ClearRenderTargetView(command_list, rtvs[resource->r.slot], + runner->r.rtv_clear_value, 0, NULL); + break; + + case RESOURCE_TYPE_DEPTH_STENCIL: + dsv = get_cpu_dsv_handle(test_context, runner->dsv_heap, 0); + ID3D12GraphicsCommandList_ClearDepthStencilView(command_list, dsv, + D3D12_CLEAR_FLAG_DEPTH, runner->r.dsv_clear_value, 0, 0, NULL); break; case RESOURCE_TYPE_TEXTURE: @@ -557,7 +587,7 @@ static bool d3d12_runner_draw(struct shader_runner *r, } } - ID3D12GraphicsCommandList_OMSetRenderTargets(command_list, rtv_count, rtvs, false, NULL); + ID3D12GraphicsCommandList_OMSetRenderTargets(command_list, rtv_count, rtvs, false, dsv.ptr ? &dsv : NULL); ID3D12GraphicsCommandList_RSSetScissorRects(command_list, 1, &test_context->scissor_rect); ID3D12GraphicsCommandList_RSSetViewports(command_list, 1, &test_context->viewport); @@ -585,6 +615,8 @@ static struct resource_readback *d3d12_runner_get_resource_readback(struct shade if (resource->r.type == RESOURCE_TYPE_RENDER_TARGET) state = D3D12_RESOURCE_STATE_RENDER_TARGET; + else if (resource->r.type == RESOURCE_TYPE_DEPTH_STENCIL) + state = D3D12_RESOURCE_STATE_DEPTH_WRITE; else state = D3D12_RESOURCE_STATE_UNORDERED_ACCESS; @@ -686,5 +718,7 @@ void run_shader_tests_d3d12(void *dxc_compiler) ID3D12DescriptorHeap_Release(runner.heap); if (runner.rtv_heap) ID3D12DescriptorHeap_Release(runner.rtv_heap); + if (runner.dsv_heap) + ID3D12DescriptorHeap_Release(runner.dsv_heap); destroy_test_context(&runner.test_context); } diff --git a/tests/shader_runner_d3d9.c b/tests/shader_runner_d3d9.c index e4e1c6435..70b3d4cdb 100644 --- a/tests/shader_runner_d3d9.c +++ b/tests/shader_runner_d3d9.c @@ -224,6 +224,10 @@ static struct resource *d3d9_runner_create_resource(struct shader_runner *r, con ok(hr == D3D_OK, "Failed to create render target, hr %#lx.\n", hr); break; + case RESOURCE_TYPE_DEPTH_STENCIL: + fatal_error("DSVs are not supported.\n"); + break; + case RESOURCE_TYPE_TEXTURE: { if (params->dimension == RESOURCE_DIMENSION_BUFFER) @@ -382,6 +386,9 @@ static bool d3d9_runner_draw(struct shader_runner *r, ok(hr == D3D_OK, "Failed to set render target, hr %#lx.\n", hr); break; + case RESOURCE_TYPE_DEPTH_STENCIL: + vkd3d_unreachable(); + case RESOURCE_TYPE_TEXTURE: assert(resource->r.dimension != RESOURCE_DIMENSION_BUFFER); diff --git a/tests/shader_runner_gl.c b/tests/shader_runner_gl.c index 2045950e3..5bbbc0769 100644 --- a/tests/shader_runner_gl.c +++ b/tests/shader_runner_gl.c @@ -334,6 +334,7 @@ static const struct format_info *get_format_info(enum DXGI_FORMAT format, bool i {DXGI_FORMAT_R32G32_SINT, 2, true, false, GL_RG32I, GL_RG_INTEGER, GL_INT}, {DXGI_FORMAT_R32_FLOAT, 1, false, false, GL_R32F, GL_RED, GL_FLOAT}, {DXGI_FORMAT_R32_FLOAT, 1, false, true, GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT}, + {DXGI_FORMAT_D32_FLOAT, 1, false, true, GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT}, {DXGI_FORMAT_R32_UINT, 1, true, false, GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_INT}, {DXGI_FORMAT_R32_SINT, 1, true, false, GL_R32I, GL_RED_INTEGER, GL_INT}, }; @@ -394,6 +395,7 @@ static struct resource *gl_runner_create_resource(struct shader_runner *r, const switch (params->type) { case RESOURCE_TYPE_RENDER_TARGET: + case RESOURCE_TYPE_DEPTH_STENCIL: case RESOURCE_TYPE_TEXTURE: case RESOURCE_TYPE_UAV: if (params->dimension == RESOURCE_DIMENSION_BUFFER) @@ -419,6 +421,7 @@ static void gl_runner_destroy_resource(struct shader_runner *r, struct resource switch (resource->r.type) { case RESOURCE_TYPE_RENDER_TARGET: + case RESOURCE_TYPE_DEPTH_STENCIL: case RESOURCE_TYPE_TEXTURE: case RESOURCE_TYPE_UAV: if (res->dimension == RESOURCE_DIMENSION_BUFFER) @@ -705,6 +708,7 @@ static bool gl_runner_dispatch(struct shader_runner *r, unsigned int x, unsigned switch (resource->r.type) { case RESOURCE_TYPE_RENDER_TARGET: + case RESOURCE_TYPE_DEPTH_STENCIL: case RESOURCE_TYPE_VERTEX_BUFFER: case RESOURCE_TYPE_TEXTURE: break; @@ -997,6 +1001,15 @@ static bool gl_runner_draw(struct shader_runner *r, rt_count = resource->r.slot + 1; break; + case RESOURCE_TYPE_DEPTH_STENCIL: + glFramebufferTexture(GL_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, resource->id, 0); + glEnable(GL_DEPTH_TEST); + glDepthMask(GL_TRUE); + glDepthFunc(get_compare_op_gl(runner->r.comparison)); + glClearDepthf(runner->r.dsv_clear_value); + glClear(GL_DEPTH_BUFFER_BIT); + break; + case RESOURCE_TYPE_TEXTURE: break; @@ -1092,7 +1105,8 @@ static struct resource_readback *gl_runner_get_resource_readback(struct shader_r struct gl_resource *resource = gl_resource(res); struct resource_readback *rb; - if (resource->r.type != RESOURCE_TYPE_RENDER_TARGET && resource->r.type != RESOURCE_TYPE_UAV) + if (resource->r.type != RESOURCE_TYPE_RENDER_TARGET && resource->r.type != RESOURCE_TYPE_DEPTH_STENCIL + && resource->r.type != RESOURCE_TYPE_UAV) fatal_error("Unhandled resource type %#x.\n", resource->r.type); rb = malloc(sizeof(*rb)); diff --git a/tests/shader_runner_vulkan.c b/tests/shader_runner_vulkan.c index a3bbf170c..dfc69da1e 100644 --- a/tests/shader_runner_vulkan.c +++ b/tests/shader_runner_vulkan.c @@ -118,7 +118,7 @@ static void end_command_buffer(struct vulkan_shader_runner *runner) } static void transition_image_layout(struct vulkan_shader_runner *runner, - VkImage image, VkImageLayout src_layout, VkImageLayout dst_layout) + VkImage image, VkImageAspectFlags aspect_mask, VkImageLayout src_layout, VkImageLayout dst_layout) { VkImageMemoryBarrier barrier = {.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER}; @@ -129,7 +129,7 @@ static void transition_image_layout(struct vulkan_shader_runner *runner, barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; barrier.image = image; - barrier.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; + barrier.subresourceRange.aspectMask = aspect_mask; barrier.subresourceRange.baseMipLevel = 0; barrier.subresourceRange.levelCount = VK_REMAINING_MIP_LEVELS; barrier.subresourceRange.baseArrayLayer = 0; @@ -238,7 +238,8 @@ static VkImage create_2d_image(const struct vulkan_shader_runner *runner, uint32 return image; } -static VkImageView create_2d_image_view(const struct vulkan_shader_runner *runner, VkImage image, VkFormat format) +static VkImageView create_2d_image_view(const struct vulkan_shader_runner *runner, VkImage image, VkFormat format, + VkImageAspectFlags aspect_mask) { VkImageViewCreateInfo view_info = {.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO}; VkImageView view; @@ -250,7 +251,7 @@ static VkImageView create_2d_image_view(const struct vulkan_shader_runner *runne view_info.components.g = VK_COMPONENT_SWIZZLE_IDENTITY; view_info.components.b = VK_COMPONENT_SWIZZLE_IDENTITY; view_info.components.a = VK_COMPONENT_SWIZZLE_IDENTITY; - view_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; + view_info.subresourceRange.aspectMask = aspect_mask; view_info.subresourceRange.baseMipLevel = 0; view_info.subresourceRange.levelCount = VK_REMAINING_MIP_LEVELS; view_info.subresourceRange.baseArrayLayer = 0; @@ -280,12 +281,12 @@ static void resource_init_2d(struct vulkan_shader_runner *runner, struct vulkan_ resource->image = create_2d_image(runner, params->width, params->height, params->level_count, usage, format, &resource->memory); - resource->image_view = create_2d_image_view(runner, resource->image, format); + resource->image_view = create_2d_image_view(runner, resource->image, format, VK_IMAGE_ASPECT_COLOR_BIT); if (!params->data) { begin_command_buffer(runner); - transition_image_layout(runner, resource->image, + transition_image_layout(runner, resource->image, VK_IMAGE_ASPECT_COLOR_BIT, VK_IMAGE_LAYOUT_UNDEFINED, layout); end_command_buffer(runner); return; @@ -299,7 +300,7 @@ static void resource_init_2d(struct vulkan_shader_runner *runner, struct vulkan_ begin_command_buffer(runner); - transition_image_layout(runner, resource->image, + transition_image_layout(runner, resource->image, VK_IMAGE_ASPECT_COLOR_BIT, VK_IMAGE_LAYOUT_UNDEFINED, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL); for (unsigned int level = 0; level < params->level_count; ++level) @@ -321,7 +322,7 @@ static void resource_init_2d(struct vulkan_shader_runner *runner, struct vulkan_ buffer_offset += level_width * level_height * params->texel_size; } - transition_image_layout(runner, resource->image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, layout); + transition_image_layout(runner, resource->image, VK_IMAGE_ASPECT_COLOR_BIT, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, layout); end_command_buffer(runner); @@ -373,14 +374,28 @@ static struct resource *vulkan_runner_create_resource(struct shader_runner *r, c resource->image = create_2d_image(runner, params->width, params->height, params->level_count, VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT, format, &resource->memory); - resource->image_view = create_2d_image_view(runner, resource->image, format); + resource->image_view = create_2d_image_view(runner, resource->image, format, VK_IMAGE_ASPECT_COLOR_BIT); begin_command_buffer(runner); - transition_image_layout(runner, resource->image, + transition_image_layout(runner, resource->image, VK_IMAGE_ASPECT_COLOR_BIT, VK_IMAGE_LAYOUT_UNDEFINED, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL); end_command_buffer(runner); break; + case RESOURCE_TYPE_DEPTH_STENCIL: + format = vkd3d_get_vk_format(params->format); + + resource->image = create_2d_image(runner, params->width, params->height, params->level_count, + VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT, format, + &resource->memory); + resource->image_view = create_2d_image_view(runner, resource->image, format, VK_IMAGE_ASPECT_DEPTH_BIT); + + begin_command_buffer(runner); + transition_image_layout(runner, resource->image, VK_IMAGE_ASPECT_DEPTH_BIT, + VK_IMAGE_LAYOUT_UNDEFINED, VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL); + end_command_buffer(runner); + break; + case RESOURCE_TYPE_TEXTURE: case RESOURCE_TYPE_UAV: if (params->dimension == RESOURCE_DIMENSION_BUFFER) @@ -541,6 +556,7 @@ static bool compile_shader(struct vulkan_shader_runner *runner, const char *sour switch (resource->r.type) { case RESOURCE_TYPE_RENDER_TARGET: + case RESOURCE_TYPE_DEPTH_STENCIL: case RESOURCE_TYPE_VERTEX_BUFFER: break; @@ -669,6 +685,31 @@ static VkPipelineLayout create_pipeline_layout(const struct vulkan_shader_runner return pipeline_layout; } +static enum VkCompareOp vk_compare_op_from_d3d12(D3D12_COMPARISON_FUNC op) +{ + switch (op) + { + case D3D12_COMPARISON_FUNC_NEVER: + return VK_COMPARE_OP_NEVER; + case D3D12_COMPARISON_FUNC_LESS: + return VK_COMPARE_OP_LESS; + case D3D12_COMPARISON_FUNC_EQUAL: + return VK_COMPARE_OP_EQUAL; + case D3D12_COMPARISON_FUNC_LESS_EQUAL: + return VK_COMPARE_OP_LESS_OR_EQUAL; + case D3D12_COMPARISON_FUNC_GREATER: + return VK_COMPARE_OP_GREATER; + case D3D12_COMPARISON_FUNC_NOT_EQUAL: + return VK_COMPARE_OP_NOT_EQUAL; + case D3D12_COMPARISON_FUNC_GREATER_EQUAL: + return VK_COMPARE_OP_GREATER_OR_EQUAL; + case D3D12_COMPARISON_FUNC_ALWAYS: + return VK_COMPARE_OP_ALWAYS; + default: + fatal_error("Unhandled compare op %#x.\n", op); + } +} + static VkPipeline create_graphics_pipeline(struct vulkan_shader_runner *runner, VkRenderPass render_pass, VkPipelineLayout pipeline_layout, D3D_PRIMITIVE_TOPOLOGY primitive_topology) { @@ -684,6 +725,7 @@ static VkPipeline create_graphics_pipeline(struct vulkan_shader_runner *runner, VkGraphicsPipelineCreateInfo pipeline_desc = {.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO}; VkPipelineColorBlendAttachmentState attachment_desc[MAX_RESOURCES] = {0}; VkVertexInputAttributeDescription input_attributes[32]; + VkPipelineDepthStencilStateCreateInfo ds_desc = {0}; VkVertexInputBindingDescription input_bindings[32]; VkPipelineShaderStageCreateInfo stage_desc[2]; struct vkd3d_shader_code vs_dxbc; @@ -745,6 +787,20 @@ static VkPipeline create_graphics_pipeline(struct vulkan_shader_runner *runner, | VK_COLOR_COMPONENT_B_BIT | VK_COLOR_COMPONENT_A_BIT; break; + case RESOURCE_TYPE_DEPTH_STENCIL: + ds_desc.sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO; + ds_desc.pNext = NULL; + ds_desc.flags = 0; + ds_desc.depthTestEnable = VK_TRUE; + ds_desc.depthWriteEnable = VK_TRUE; + ds_desc.depthCompareOp = vk_compare_op_from_d3d12(runner->r.comparison); + ds_desc.depthBoundsTestEnable = VK_FALSE; + ds_desc.stencilTestEnable = VK_FALSE; + ds_desc.minDepthBounds = 0.0f; + ds_desc.maxDepthBounds = 1.0f; + pipeline_desc.pDepthStencilState = &ds_desc; + break; + case RESOURCE_TYPE_VERTEX_BUFFER: { VkVertexInputBindingDescription *binding = &input_bindings[input_desc.vertexBindingDescriptionCount++]; @@ -841,31 +897,6 @@ static VkSamplerAddressMode vk_address_mode_from_d3d12(D3D12_TEXTURE_ADDRESS_MOD } } -static enum VkCompareOp vk_compare_op_from_d3d12(D3D12_COMPARISON_FUNC op) -{ - switch (op) - { - case D3D12_COMPARISON_FUNC_NEVER: - return VK_COMPARE_OP_NEVER; - case D3D12_COMPARISON_FUNC_LESS: - return VK_COMPARE_OP_LESS; - case D3D12_COMPARISON_FUNC_EQUAL: - return VK_COMPARE_OP_EQUAL; - case D3D12_COMPARISON_FUNC_LESS_EQUAL: - return VK_COMPARE_OP_LESS_OR_EQUAL; - case D3D12_COMPARISON_FUNC_GREATER: - return VK_COMPARE_OP_GREATER; - case D3D12_COMPARISON_FUNC_NOT_EQUAL: - return VK_COMPARE_OP_NOT_EQUAL; - case D3D12_COMPARISON_FUNC_GREATER_EQUAL: - return VK_COMPARE_OP_GREATER_OR_EQUAL; - case D3D12_COMPARISON_FUNC_ALWAYS: - return VK_COMPARE_OP_ALWAYS; - default: - fatal_error("Unhandled compare op %#x.\n", op); - } -} - static VkDescriptorSetLayout create_descriptor_set_layout(struct vulkan_shader_runner *runner) { VkDescriptorSetLayoutCreateInfo set_desc = {.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO}; @@ -887,6 +918,7 @@ static VkDescriptorSetLayout create_descriptor_set_layout(struct vulkan_shader_r switch (resource->r.type) { case RESOURCE_TYPE_RENDER_TARGET: + case RESOURCE_TYPE_DEPTH_STENCIL: case RESOURCE_TYPE_VERTEX_BUFFER: break; @@ -1018,6 +1050,7 @@ static void bind_resources(struct vulkan_shader_runner *runner, VkPipelineBindPo break; case RESOURCE_TYPE_RENDER_TARGET: + case RESOURCE_TYPE_DEPTH_STENCIL: break; } } @@ -1036,44 +1069,58 @@ static void create_render_pass_and_framebuffer(struct vulkan_shader_runner *runn { VkRenderPassCreateInfo render_pass_desc = {.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO}; VkFramebufferCreateInfo fb_desc = {.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO}; + VkAttachmentReference ds_ref = {0}, color_refs[MAX_RESOURCES] = {0}; VkAttachmentDescription attachment_descs[MAX_RESOURCES] = {0}; - VkAttachmentReference color_refs[MAX_RESOURCES] = {0}; VkSubpassDescription subpass_desc = {0}; - VkImageView rtvs[MAX_RESOURCES]; - unsigned int rt_count = 0; + VkImageView views[MAX_RESOURCES]; + unsigned int color_ref_count = 0, view_count = 0; + VkImageLayout layout; unsigned int i; + bool is_ds; for (i = 0; i < runner->r.resource_count; ++i) { const struct vulkan_resource *resource = vulkan_resource(runner->r.resources[i]); - VkAttachmentDescription *attachment_desc = &attachment_descs[rt_count]; - VkAttachmentReference *color_ref = &color_refs[rt_count]; + VkAttachmentDescription *attachment_desc = &attachment_descs[view_count]; + VkAttachmentReference *color_ref = &color_refs[color_ref_count]; - if (resource->r.type != RESOURCE_TYPE_RENDER_TARGET) + if (resource->r.type != RESOURCE_TYPE_RENDER_TARGET && resource->r.type != RESOURCE_TYPE_DEPTH_STENCIL) continue; + is_ds = resource->r.type == RESOURCE_TYPE_DEPTH_STENCIL; + layout = is_ds ? VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; + attachment_desc->format = vkd3d_get_vk_format(resource->r.format); attachment_desc->samples = VK_SAMPLE_COUNT_1_BIT; - attachment_desc->loadOp = VK_ATTACHMENT_LOAD_OP_LOAD; + attachment_desc->loadOp = (is_ds || runner->r.clear_rtv) + ? VK_ATTACHMENT_LOAD_OP_CLEAR : VK_ATTACHMENT_LOAD_OP_LOAD; attachment_desc->storeOp = VK_ATTACHMENT_STORE_OP_STORE; attachment_desc->stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; attachment_desc->stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; - attachment_desc->initialLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; - attachment_desc->finalLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; + attachment_desc->initialLayout = layout; + attachment_desc->finalLayout = layout; - color_ref->attachment = rt_count; - color_ref->layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; - - rtvs[rt_count] = resource->image_view; + if (is_ds) + { + ds_ref.attachment = view_count; + ds_ref.layout = layout; + subpass_desc.pDepthStencilAttachment = &ds_ref; + } + else + { + color_ref->attachment = view_count; + color_ref->layout = layout; + ++color_ref_count; + } - ++rt_count; + views[view_count++] = resource->image_view; } subpass_desc.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS; - subpass_desc.colorAttachmentCount = rt_count; + subpass_desc.colorAttachmentCount = color_ref_count; subpass_desc.pColorAttachments = color_refs; - render_pass_desc.attachmentCount = rt_count; + render_pass_desc.attachmentCount = view_count; render_pass_desc.pAttachments = attachment_descs; render_pass_desc.subpassCount = 1; render_pass_desc.pSubpasses = &subpass_desc; @@ -1081,8 +1128,8 @@ static void create_render_pass_and_framebuffer(struct vulkan_shader_runner *runn VK_CALL(vkCreateRenderPass(runner->device, &render_pass_desc, NULL, render_pass)); fb_desc.renderPass = *render_pass; - fb_desc.attachmentCount = rt_count; - fb_desc.pAttachments = rtvs; + fb_desc.attachmentCount = view_count; + fb_desc.pAttachments = views; fb_desc.width = RENDER_TARGET_WIDTH; fb_desc.height = RENDER_TARGET_HEIGHT; fb_desc.layers = 1; @@ -1140,6 +1187,7 @@ static bool vulkan_runner_draw(struct shader_runner *r, static const VkRect2D rt_rect = {.extent.width = RENDER_TARGET_WIDTH, .extent.height = RENDER_TARGET_HEIGHT}; VkRenderPassBeginInfo pass_begin_desc = {.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO}; + union VkClearValue clear_values[D3D12_SIMULTANEOUS_RENDER_TARGET_COUNT]; VkCommandBuffer cmd_buffer = runner->cmd_buffer; VkDescriptorSetLayout set_layout; VkPipelineLayout pipeline_layout; @@ -1164,6 +1212,25 @@ static bool vulkan_runner_draw(struct shader_runner *r, pass_begin_desc.renderPass = render_pass; pass_begin_desc.framebuffer = fb; pass_begin_desc.renderArea = rt_rect; + pass_begin_desc.clearValueCount = 0; + pass_begin_desc.pClearValues = clear_values; + for (i = 0; i < runner->r.resource_count; ++i) + { + const struct resource *resource = runner->r.resources[i]; + + if (resource->type == RESOURCE_TYPE_RENDER_TARGET) + { + memcpy(clear_values[pass_begin_desc.clearValueCount].color.float32, runner->r.rtv_clear_value, + sizeof(clear_values[0].color.float32)); + ++pass_begin_desc.clearValueCount; + } + else if (resource->type == RESOURCE_TYPE_DEPTH_STENCIL) + { + clear_values[pass_begin_desc.clearValueCount].depthStencil.depth = runner->r.dsv_clear_value; + clear_values[pass_begin_desc.clearValueCount].depthStencil.stencil = 0; + ++pass_begin_desc.clearValueCount; + } + } VK_CALL(vkCmdBeginRenderPass(cmd_buffer, &pass_begin_desc, VK_SUBPASS_CONTENTS_INLINE)); @@ -1205,6 +1272,7 @@ static struct resource_readback *vulkan_runner_get_resource_readback(struct shad struct vulkan_resource_readback *rb = malloc(sizeof(*rb)); struct vulkan_resource *resource = vulkan_resource(res); VkDevice device = runner->device; + VkImageAspectFlags aspect_mask; VkBufferImageCopy region = {0}; VkImageLayout layout; @@ -1228,16 +1296,21 @@ static struct resource_readback *vulkan_runner_get_resource_readback(struct shad } else { + aspect_mask = (resource->r.type == RESOURCE_TYPE_DEPTH_STENCIL) + ? VK_IMAGE_ASPECT_DEPTH_BIT : VK_IMAGE_ASPECT_COLOR_BIT; + if (resource->r.type == RESOURCE_TYPE_RENDER_TARGET) layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; + else if (resource->r.type == RESOURCE_TYPE_DEPTH_STENCIL) + layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL; else layout = VK_IMAGE_LAYOUT_GENERAL; begin_command_buffer(runner); - transition_image_layout(runner, resource->image, layout, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL); + transition_image_layout(runner, resource->image, aspect_mask, layout, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL); - region.imageSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; + region.imageSubresource.aspectMask = aspect_mask; region.imageSubresource.layerCount = 1; region.imageExtent.width = resource->r.width; region.imageExtent.height = resource->r.height; @@ -1246,7 +1319,7 @@ static struct resource_readback *vulkan_runner_get_resource_readback(struct shad VK_CALL(vkCmdCopyImageToBuffer(runner->cmd_buffer, resource->image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, rb->buffer, 1, ®ion)); - transition_image_layout(runner, resource->image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, layout); + transition_image_layout(runner, resource->image, aspect_mask, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, layout); end_command_buffer(runner); -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/777
From: Conor McCarthy <cmccarthy(a)codeweavers.com> --- libs/vkd3d-shader/dxil.c | 70 +++++++++++++++++++++++++++----- tests/hlsl/depth-out.shader_test | 18 ++++---- 2 files changed, 68 insertions(+), 20 deletions(-) diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index da8ba662d..0ce91b479 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -794,7 +794,7 @@ struct sm6_parser struct vkd3d_shader_dst_param *output_params; struct vkd3d_shader_dst_param *input_params; - uint32_t input_regs_declared[(VKD3DSPR_COUNT + 0x1f) / 0x20]; + uint32_t io_regs_declared[(VKD3DSPR_COUNT + 0x1f) / 0x20]; struct sm6_function *functions; size_t function_count; @@ -3599,6 +3599,22 @@ static void src_params_init_from_operands(struct vkd3d_shader_src_param *src_par src_param_init_from_value(&src_params[i], operands[i]); } +static enum vkd3d_shader_register_type register_type_from_dxil_semantic_kind( + enum vkd3d_shader_sysval_semantic sysval_semantic) +{ + switch (sysval_semantic) + { + case VKD3D_SHADER_SV_DEPTH: + return VKD3DSPR_DEPTHOUT; + case VKD3D_SHADER_SV_DEPTH_GREATER_EQUAL: + return VKD3DSPR_DEPTHOUTGE; + case VKD3D_SHADER_SV_DEPTH_LESS_EQUAL: + return VKD3DSPR_DEPTHOUTLE; + default: + return VKD3DSPR_INVALID; + } +} + static void sm6_parser_init_signature(struct sm6_parser *sm6, const struct shader_signature *s, enum vkd3d_shader_register_type reg_type, struct vkd3d_shader_dst_param *params) { @@ -3611,6 +3627,13 @@ static void sm6_parser_init_signature(struct sm6_parser *sm6, const struct shade e = &s->elements[i]; param = ¶ms[i]; + + if (e->register_index == UINT_MAX) + { + dst_param_io_init(param, e, register_type_from_dxil_semantic_kind(e->sysval_semantic)); + continue; + } + dst_param_io_init(param, e, reg_type); count = 0; if (e->register_count > 1) @@ -4551,16 +4574,16 @@ static void sm6_parser_emit_dx_cbuffer_load(struct sm6_parser *sm6, enum dx_intr instruction_dst_param_init_ssa_vector(ins, sm6_type_max_vector_size(type), sm6); } -static void sm6_parser_dcl_register_builtin(struct sm6_parser *sm6, +static void sm6_parser_dcl_register_builtin(struct sm6_parser *sm6, enum vkd3d_shader_opcode handler_idx, enum vkd3d_shader_register_type reg_type, enum vkd3d_data_type data_type, unsigned int component_count) { struct vkd3d_shader_dst_param *dst_param; struct vkd3d_shader_instruction *ins; - if (!bitmap_is_set(sm6->input_regs_declared, reg_type)) + if (!bitmap_is_set(sm6->io_regs_declared, reg_type)) { - bitmap_set(sm6->input_regs_declared, reg_type); - ins = sm6_parser_add_instruction(sm6, VKD3DSIH_DCL_INPUT); + bitmap_set(sm6->io_regs_declared, reg_type); + ins = sm6_parser_add_instruction(sm6, handler_idx); dst_param = &ins->declaration.dst; vsir_register_init(&dst_param->reg, reg_type, data_type, 0); dst_param_init_vector(dst_param, component_count); @@ -4724,7 +4747,7 @@ static void sm6_parser_emit_dx_compute_builtin(struct sm6_parser *sm6, enum dx_i vkd3d_unreachable(); } - sm6_parser_dcl_register_builtin(sm6, reg_type, VKD3D_DATA_UINT, component_count); + sm6_parser_dcl_register_builtin(sm6, VKD3DSIH_DCL_INPUT, reg_type, VKD3D_DATA_UINT, component_count); vsir_instruction_init(ins, &sm6->p.location, VKD3DSIH_MOV); if (!(src_param = instruction_src_params_alloc(ins, 1, sm6))) return; @@ -5349,6 +5372,12 @@ static void sm6_parser_emit_dx_store_output(struct sm6_parser *sm6, enum dx_intr if (e->register_count > 1) register_index_address_init(&dst_param->reg.idx[0], operands[1], sm6); + if (e->register_index == UINT_MAX) + { + sm6_parser_dcl_register_builtin(sm6, VKD3DSIH_DCL_OUTPUT, dst_param->reg.type, + dst_param->reg.data_type, vsir_write_mask_component_count(e->mask)); + } + if ((src_param = instruction_src_params_alloc(ins, 1, sm6))) src_param_init_from_value(src_param, value); } @@ -7683,6 +7712,9 @@ static const enum vkd3d_shader_sysval_semantic sysval_semantic_table[] = [SEMANTIC_KIND_POSITION] = VKD3D_SHADER_SV_POSITION, [SEMANTIC_KIND_ISFRONTFACE] = VKD3D_SHADER_SV_IS_FRONT_FACE, [SEMANTIC_KIND_TARGET] = VKD3D_SHADER_SV_TARGET, + [SEMANTIC_KIND_DEPTH] = VKD3D_SHADER_SV_DEPTH, + [SEMANTIC_KIND_DEPTHLESSEQUAL] = VKD3D_SHADER_SV_DEPTH_LESS_EQUAL, + [SEMANTIC_KIND_DEPTHGREATEREQUAL] = VKD3D_SHADER_SV_DEPTH_GREATER_EQUAL, }; static enum vkd3d_shader_sysval_semantic sysval_semantic_from_dxil_semantic_kind(enum dxil_semantic_kind kind) @@ -8449,6 +8481,7 @@ static enum vkd3d_result sm6_parser_read_signature(struct sm6_parser *sm6, const const struct sm6_metadata_node *node, *element_node; struct signature_element *elements, *e; unsigned int values[10]; + bool is_register; if (!m) return VKD3D_OK; @@ -8557,7 +8590,18 @@ static enum vkd3d_result sm6_parser_read_signature(struct sm6_parser *sm6, const column_count = values[7]; e->register_index = values[8]; e->target_location = e->register_index; - if (e->register_index > MAX_REG_OUTPUT || e->register_count > MAX_REG_OUTPUT - e->register_index) + + if ((is_register = e->register_index == UINT_MAX)) + { + if (register_type_from_dxil_semantic_kind(e->sysval_semantic) == VKD3DSPR_INVALID) + { + WARN("Unhandled I/O register semantic kind %u.\n", j); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_SIGNATURE, + "DXIL semantic kind %u is unhandled for an I/O register.", j); + return VKD3D_ERROR_INVALID_SHADER; + } + } + else if (e->register_index > MAX_REG_OUTPUT || e->register_count > MAX_REG_OUTPUT - e->register_index) { WARN("Invalid row start %u with row count %u.\n", e->register_index, e->register_count); vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_SIGNATURE, @@ -8565,8 +8609,9 @@ static enum vkd3d_result sm6_parser_read_signature(struct sm6_parser *sm6, const e->register_index, e->register_count); return VKD3D_ERROR_INVALID_SHADER; } + index = values[9]; - if (index >= VKD3D_VEC4_SIZE || column_count > VKD3D_VEC4_SIZE - index) + if (index != UINT8_MAX && (index >= VKD3D_VEC4_SIZE || column_count > VKD3D_VEC4_SIZE - index)) { WARN("Invalid column start %u with count %u.\n", index, column_count); vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_SIGNATURE, @@ -8576,10 +8621,13 @@ static enum vkd3d_result sm6_parser_read_signature(struct sm6_parser *sm6, const e->mask = vkd3d_write_mask_from_component_count(column_count); e->used_mask = e->mask; - e->mask <<= index; - signature_element_read_additional_element_values(e, element_node, sm6); - e->used_mask <<= index; + + if (index != UINT8_MAX) + { + e->mask <<= index; + e->used_mask <<= index; + } m = element_node->operands[4]; if (!sm6_metadata_value_is_node(m)) diff --git a/tests/hlsl/depth-out.shader_test b/tests/hlsl/depth-out.shader_test index b26830fc6..41c8033e2 100644 --- a/tests/hlsl/depth-out.shader_test +++ b/tests/hlsl/depth-out.shader_test @@ -16,20 +16,20 @@ float main() : SV_Depth uniform 0 float 0.0 dsv clear 1.0 depth less -todo(sm>=6 | glsl) draw quad +todo(glsl) draw quad probe dsv all r (0.0) uniform 0 float 0.75 -todo(sm>=6 | glsl) draw quad +todo(glsl) draw quad probe dsv all r (0.75) dsv clear 0.5 depth greater -todo(sm>=6 | glsl) draw quad +todo(glsl) draw quad probe dsv all r (0.75) depth less -todo(sm>=6 | glsl) draw quad +todo(glsl) draw quad probe dsv all r (0.5) @@ -63,12 +63,12 @@ uniform 0 float4 0.7 0.7 0.0 0.0 rtv clear 1.0 1.0 1.0 1.0 dsv clear 0.5 depth greater equal -todo draw quad +todo(sm<6 | glsl) draw quad probe all rgba (0.0, 1.0, 0.0, 1.0) probe dsv all r (0.7) uniform 0 float4 0.7 0.4 0.0 0.0 -todo draw quad +todo(sm<6 | glsl) draw quad probe all rgba (1.0, 1.0, 1.0, 1.0) probe dsv all r (0.5) @@ -87,16 +87,16 @@ uniform 0 float4 0.7 0.7 0.0 0.0 rtv clear 1.0 1.0 1.0 1.0 dsv clear 0.5 depth greater equal -todo draw quad +todo(sm<6 | glsl) draw quad probe all rgba (0.0, 1.0, 0.0, 1.0) probe dsv all r (0.7) uniform 0 float4 0.4 0.6 0.0 0.0 -todo draw quad +todo(sm<6 | glsl) draw quad probe all rgba (0.0, 1.0, 0.0, 1.0) probe dsv all r (0.6) uniform 0 float4 0.4 0.4 0.0 0.0 -todo draw quad +todo(sm<6 | glsl) draw quad probe all rgba (1.0, 1.0, 1.0, 1.0) probe dsv all r (0.5) -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/777
participants (2)
-
Conor McCarthy -
Conor McCarthy (＠cmccarthy)