Test case 3/4: Disable PS output for ALU and memfetch (#502)

Added knob `disable-ps-output` to disable PS output.
For test case 3, use ` "fullscreen-quads-type": "Solid_Color"` and for
test case 4 use ` "fullscreen-quads-type": "Texture"`
{
    "deterministic": true,
    "enable-skybox": false,
    "enable-spheres": false,
    "enable-metrics": false,
    "fullscreen-quads-count": 1,
    "fullscreen-quads-type": "Texture",
    "fullscreen-quads-single-renderpass": true,
    "vs-alu-instruction-count": 100,
    "texture-count": 2,
    "disable-ps-output": true
}

CMakeLists.txt

@@ -141,6 +141,7 @@ if(MSVC)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${MSVC_DISABLED_WARNINGS} /MP /Zc:__cplusplus /std:c++17")
 else()
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} \
+        -Werror \
         -std=c++17 \
         -fdiagnostics-color=always \
         -Wno-nullability-completeness \
@@ -153,12 +154,6 @@ else()
         -Wno-pointer-bool-conversion \
     ")
 
-    if (NOT PPX_BUILD_XR)
-        # OpenXR has way too many warnings to enable -Werror with it.
-        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} \
-            -Werror \
-        ")
-    endif ()
 endif()
 
 set(CMAKE_CXX_STANDARD 17)

assets/benchmarks/shaders/Benchmark_Texture.hlsl

@@ -15,6 +15,7 @@
 #include "Benchmark_Quad.hlsli"
 
 Texture2D Tex[10]  : register(t1);  // Slot 0 is used by push constant.
+RWStructuredBuffer<float> dataBuffer : register(u11);
 
 float4 psmain(VSOutputPos input) : SV_TARGET
 {
@@ -24,5 +25,7 @@ float4 psmain(VSOutputPos input) : SV_TARGET
     {
         color += Tex[i].Load(uint3(input.position.x, input.position.y, 0))/float(textureCount);
     }
+    if (!any(color))
+        dataBuffer[0] = color.r;
     return color;
 }

benchmarks/graphics_pipeline/GraphicsBenchmarkApp.cpp

@@ -16,6 +16,7 @@
 #include "SphereMesh.h"
 
 #include "ppx/graphics_util.h"
+#include "ppx/grfx/grfx_enums.h"
 #include "ppx/grfx/grfx_format.h"
 #include "ppx/math_config.h"
 #include "ppx/timer.h"
@@ -38,6 +39,7 @@ static constexpr size_t SPHERE_METAL_ROUGHNESS_SAMPLER_REGISTER       = 7;
 
 static constexpr size_t QUADS_CONFIG_UNIFORM_BUFFER_REGISTER = 0;
 static constexpr size_t QUADS_SAMPLED_IMAGE_REGISTER         = 1;
+static constexpr size_t QUADS_DUMMY_BUFFER_REGISTER          = QUADS_SAMPLED_IMAGE_REGISTER + kMaxTextureCount;
 
 #if defined(USE_DX12)
 const grfx::Api kApi = grfx::API_DX_12_0;
@@ -190,6 +192,10 @@ void GraphicsBenchmarkApp::InitKnobs()
     GetKnobManager().InitKnob(&pKnobTextureCount, "texture-count", /* defaultValue = */ 1, /* minValue = */ 1, kMaxTextureCount);
     pKnobTextureCount->SetDisplayName("Number of texture to load in the shader");
     pKnobTextureCount->SetFlagDescription("Select the number of texture to load in the shader.");
+
+    GetKnobManager().InitKnob(&pKnobDisablePsOutput, "disable-ps-output", false);
+    pKnobDisablePsOutput->SetDisplayName("Disable PS output");
+    pKnobDisablePsOutput->SetFlagDescription("Disable PS output.");
 }
 
 void GraphicsBenchmarkApp::Config(ppx::ApplicationSettings& settings)
@@ -240,9 +246,10 @@ void GraphicsBenchmarkApp::Setup()
     // Descriptor Pool
     {
         grfx::DescriptorPoolCreateInfo createInfo = {};
-        createInfo.sampler                        = 5 * GetNumFramesInFlight();  // 1 for skybox, 3 for spheres, 1 for blit
-        createInfo.sampledImage                   = 15 * GetNumFramesInFlight(); // 1 for skybox, 3 for spheres, 10 for quads, 1 for blit
-        createInfo.uniformBuffer                  = 2 * GetNumFramesInFlight();  // 1 for skybox, 1 for spheres
+        createInfo.sampler                        = 5 * GetNumFramesInFlight();                      // 1 for skybox, 3 for spheres, 1 for blit
+        createInfo.sampledImage                   = (5 + kMaxTextureCount) * GetNumFramesInFlight(); // 1 for skybox, 3 for spheres, kMaxTextureCount for quads, 1 for blit
+        createInfo.uniformBuffer                  = 2 * GetNumFramesInFlight();                      // 1 for skybox, 1 for spheres
+        createInfo.structuredBuffer               = 1;                                               // 1 for quads dummy buffer
 
         PPX_CHECKED_CALL(GetDevice()->CreateDescriptorPool(&createInfo, &mDescriptorPool));
     }
@@ -482,10 +489,22 @@ void GraphicsBenchmarkApp::SetupFullscreenQuadsResources()
         }
     }
 
+    // dummy buffer
+    {
+        grfx::BufferCreateInfo bufferCreateInfo             = {};
+        bufferCreateInfo.size                               = PPX_MINIMUM_STRUCTURED_BUFFER_SIZE;
+        bufferCreateInfo.structuredElementStride            = static_cast<uint32_t>(sizeof(float));
+        bufferCreateInfo.usageFlags.bits.rwStructuredBuffer = true;
+        bufferCreateInfo.memoryUsage                        = grfx::MEMORY_USAGE_GPU_ONLY;
+        bufferCreateInfo.initialState                       = grfx::RESOURCE_STATE_GENERAL;
+        PPX_CHECKED_CALL(GetDevice()->CreateBuffer(&bufferCreateInfo, &mQuadsDummyBuffer));
+    }
+
     // Descriptor set layout for texture shader
     {
         grfx::DescriptorSetLayoutCreateInfo layoutCreateInfo = {};
         layoutCreateInfo.bindings.push_back(grfx::DescriptorBinding(QUADS_SAMPLED_IMAGE_REGISTER, grfx::DESCRIPTOR_TYPE_SAMPLED_IMAGE, kMaxTextureCount));
+        layoutCreateInfo.bindings.push_back(grfx::DescriptorBinding(QUADS_DUMMY_BUFFER_REGISTER, grfx::DESCRIPTOR_TYPE_RW_STRUCTURED_BUFFER));
         PPX_CHECKED_CALL(GetDevice()->CreateDescriptorSetLayout(&layoutCreateInfo, &mFullscreenQuads.descriptorSetLayout));
     }
 
@@ -552,6 +571,15 @@ void GraphicsBenchmarkApp::UpdateFullscreenQuadsDescriptors()
         for (uint32_t j = 0; j < kMaxTextureCount; j++) {
             PPX_CHECKED_CALL(pDescriptorSet->UpdateSampledImage(QUADS_SAMPLED_IMAGE_REGISTER, j, mQuadsTextures[j]));
         }
+        grfx::WriteDescriptor write  = {};
+        write.binding                = QUADS_DUMMY_BUFFER_REGISTER;
+        write.arrayIndex             = 0;
+        write.type                   = grfx::DESCRIPTOR_TYPE_RW_STRUCTURED_BUFFER;
+        write.bufferOffset           = 0;
+        write.bufferRange            = PPX_WHOLE_SIZE;
+        write.structuredElementCount = 1;
+        write.pBuffer                = mQuadsDummyBuffer;
+        PPX_CHECKED_CALL(pDescriptorSet->UpdateDescriptors(1, &write));
     }
 }
 
@@ -751,7 +779,7 @@ Result GraphicsBenchmarkApp::CompilePipeline(const QuadPipelineKey& key)
     gpCreateInfo.frontFace                          = grfx::FRONT_FACE_CW;
     gpCreateInfo.depthReadEnable                    = false;
     gpCreateInfo.depthWriteEnable                   = false;
-    gpCreateInfo.blendModes[0]                      = grfx::BLEND_MODE_NONE;
+    gpCreateInfo.blendModes[0]                      = pKnobDisablePsOutput->GetValue() ? grfx::BLEND_MODE_DISABLE_OUTPUT : grfx::BLEND_MODE_NONE;
     gpCreateInfo.outputState.renderTargetCount      = 1;
     gpCreateInfo.outputState.renderTargetFormats[0] = key.renderFormat;
     gpCreateInfo.outputState.depthStencilFormat     = GetSwapchain()->GetDepthFormat();

benchmarks/graphics_pipeline/GraphicsBenchmarkApp.h

@@ -481,6 +481,7 @@ class GraphicsBenchmarkApp
     Entity2D                                                             mFullscreenQuads;
     grfx::ShaderModulePtr                                                mVSQuads;
     std::array<grfx::TexturePtr, kMaxTextureCount>                       mQuadsTextures;
+    grfx::BufferPtr                                                      mQuadsDummyBuffer;
     QuadPipelineMap                                                      mQuadsPipelines;
     std::array<grfx::PipelineInterfacePtr, kFullscreenQuadsTypes.size()> mQuadsPipelineInterfaces;
     std::array<grfx::ShaderModulePtr, kFullscreenQuadsTypes.size()>      mQuadsPs;
@@ -532,6 +533,7 @@ class GraphicsBenchmarkApp
 
     std::shared_ptr<KnobFlag<int>> pKnobAluCount;
     std::shared_ptr<KnobFlag<int>> pKnobTextureCount;
+    std::shared_ptr<KnobCheckbox>  pKnobDisablePsOutput;
 
 private:
     // =====================================================================

include/ppx/grfx/grfx_enums.h

@@ -69,12 +69,13 @@ enum BlendFactor
 //!
 enum BlendMode
 {
-    BLEND_MODE_NONE          = 0,
-    BLEND_MODE_ADDITIVE      = 1,
-    BLEND_MODE_ALPHA         = 2,
-    BLEND_MODE_OVER          = 3,
-    BLEND_MODE_UNDER         = 4,
-    BLEND_MODE_PREMULT_ALPHA = 5,
+    BLEND_MODE_NONE           = 0,
+    BLEND_MODE_ADDITIVE       = 1,
+    BLEND_MODE_ALPHA          = 2,
+    BLEND_MODE_OVER           = 3,
+    BLEND_MODE_UNDER          = 4,
+    BLEND_MODE_PREMULT_ALPHA  = 5,
+    BLEND_MODE_DISABLE_OUTPUT = 6, // Mode used to disable vs output.
 };
 
 enum BlendOp

include/ppx/grfx/grfx_pipeline.h

@@ -128,13 +128,15 @@ struct BlendAttachmentState
     grfx::ColorComponentFlags colorWriteMask      = grfx::ColorComponentFlags::RGBA();
 
     // These are best guesses based on random formulas off of the internet.
-    // Correct later when authorative literature is found.
+    // Correct later when authoritative literature is found.
     //
+    static grfx::BlendAttachmentState BlendModeNone();
     static grfx::BlendAttachmentState BlendModeAdditive();
     static grfx::BlendAttachmentState BlendModeAlpha();
     static grfx::BlendAttachmentState BlendModeOver();
     static grfx::BlendAttachmentState BlendModeUnder();
     static grfx::BlendAttachmentState BlendModePremultAlpha();
+    static grfx::BlendAttachmentState BlendModeDisableOutput();
 };
 
 struct ColorBlendState

src/ppx/grfx/grfx_pipeline.cpp

@@ -12,16 +12,27 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include "ppx/config.h"
 #include "ppx/grfx/grfx_device.h"
 #include "ppx/grfx/grfx_pipeline.h"
 #include "ppx/grfx/grfx_descriptor.h"
+#include "ppx/grfx/grfx_enums.h"
 
 namespace ppx {
 namespace grfx {
 
 // -------------------------------------------------------------------------------------------------
 // BlendAttachmentState
 // -------------------------------------------------------------------------------------------------
+grfx::BlendAttachmentState BlendAttachmentState::BlendModeNone()
+{
+    grfx::BlendAttachmentState state = {};
+    state.blendEnable                = false;
+    state.colorWriteMask             = grfx::ColorComponentFlags::RGBA();
+
+    return state;
+}
+
 grfx::BlendAttachmentState BlendAttachmentState::BlendModeAdditive()
 {
     grfx::BlendAttachmentState state = {};
@@ -97,6 +108,15 @@ grfx::BlendAttachmentState BlendAttachmentState::BlendModePremultAlpha()
     return state;
 }
 
+grfx::BlendAttachmentState BlendAttachmentState::BlendModeDisableOutput()
+{
+    grfx::BlendAttachmentState state = {};
+    state.blendEnable                = false;
+    state.colorWriteMask             = grfx::ColorComponentFlags(0);
+
+    return state;
+}
+
 namespace internal {
 
 // -------------------------------------------------------------------------------------------------
@@ -123,7 +143,7 @@ void FillOutGraphicsPipelineCreateInfo(
         }
     }
 
-    // Input aasembly
+    // Input assembly
     {
         pDstCreateInfo->inputAssemblyState.topology = pSrcCreateInfo->topology;
     }
@@ -153,7 +173,10 @@ void FillOutGraphicsPipelineCreateInfo(
         pDstCreateInfo->colorBlendState.blendAttachmentCount = pSrcCreateInfo->outputState.renderTargetCount;
         for (uint32_t i = 0; i < pDstCreateInfo->colorBlendState.blendAttachmentCount; ++i) {
             switch (pSrcCreateInfo->blendModes[i]) {
-                default: break;
+                case grfx::BLEND_MODE_NONE: {
+                    pDstCreateInfo->colorBlendState.blendAttachments[i] = grfx::BlendAttachmentState::BlendModeNone();
+                    break;
+                }
 
                 case grfx::BLEND_MODE_ADDITIVE: {
                     pDstCreateInfo->colorBlendState.blendAttachments[i] = grfx::BlendAttachmentState::BlendModeAdditive();
@@ -174,8 +197,17 @@ void FillOutGraphicsPipelineCreateInfo(
                 case grfx::BLEND_MODE_PREMULT_ALPHA: {
                     pDstCreateInfo->colorBlendState.blendAttachments[i] = grfx::BlendAttachmentState::BlendModePremultAlpha();
                 } break;
+
+                case grfx::BLEND_MODE_DISABLE_OUTPUT: {
+                    pDstCreateInfo->colorBlendState.blendAttachments[i] = grfx::BlendAttachmentState::BlendModeDisableOutput();
+                    break;
+                }
+
+                default: {
+                    PPX_ASSERT_MSG(false, "Unknown BlendMode");
+                    break;
+                }
             }
-            pDstCreateInfo->colorBlendState.blendAttachments[i].colorWriteMask = grfx::ColorComponentFlags::RGBA();
         }
     }