Precompile and cache EX31 path tracer variants

31_HLSLPathTracer/CMakeLists.txt

@@ -1,10 +1,37 @@
-include(common RESULT_VARIABLE RES)
-
-if(NOT RES)
-	message(FATAL_ERROR "common.cmake not found. Should be in {repo_root}/cmake directory")
-endif()
+include(common)
+include("${CMAKE_CURRENT_SOURCE_DIR}/pt.cmake")
+include("${CMAKE_CURRENT_SOURCE_DIR}/pt.variant_ids.cmake")
 
 if(NBL_BUILD_IMGUI)
+	# PT_BUILD_MODE controls how EX31 packages triangle polygon-method variants.
+	# WALLTIME_OPTIMIZED keeps triangle polygon-method selection inside one shared shader module.
+	# This preserves the same runtime feature surface while keeping triangle methods on one heavy backend compile path.
+	# SPECIALIZED bakes Area, SolidAngle and ProjectedSolidAngle into separate triangle entrypoints.
+	# That does not add only thin entrypoints. It multiplies the heavy triangle-side path tracing instantiations and pushes more work into the DXC/SPIR-V backend.
+	# On AMD Ryzen 5 5600G with Radeon Graphics (6C/12T),
+	# a Visual Studio Debug x64 full rebuild of the SPIR-V project completed in:
+	#   WALLTIME_OPTIMIZED = 12.785 s
+	#   SPECIALIZED = 18.314 s
+	# SPECIALIZED is +5.529 s slower which is +43.25%.
+	# Equivalently, WALLTIME_OPTIMIZED is 30.19% faster than SPECIALIZED.
+	# This comes from multiplying the heavy triangle-side path tracing instantiations
+	# and pushing more work into the DXC/SPIR-V backend.
+	# Therefore the default stays WALLTIME_OPTIMIZED.
+	set(PT_BUILD_MODE "WALLTIME_OPTIMIZED" CACHE STRING
+		"Choose the EX31 precompiled shader layout. WALLTIME_OPTIMIZED keeps polygon-method selection inside the shared triangle shader module to reduce compile wall time. SPECIALIZED bakes triangle polygon methods into distinct entrypoints, which increases backend work and rebuild time but keeps those variants as separate precompiled entrypoints."
+	)
+	set(PT_CACHE_ROOT "pipeline/cache" CACHE STRING
+		"Relative cache root written to path_tracer.runtime.json in the common bin directory. The runtime resolves this path relative to the JSON file location. Empty disables the generated dev-mode JSON and falls back to --pipeline-cache-dir or LocalAppData."
+	)
+	set_property(CACHE PT_BUILD_MODE PROPERTY STRINGS WALLTIME_OPTIMIZED SPECIALIZED)
+	set(_PT_BUILD_MODE_VALUES WALLTIME_OPTIMIZED SPECIALIZED)
+	if(NOT PT_BUILD_MODE IN_LIST _PT_BUILD_MODE_VALUES)
+		message(FATAL_ERROR "Unsupported PT_BUILD_MODE='${PT_BUILD_MODE}'. Expected one of: ${_PT_BUILD_MODE_VALUES}")
+	endif()
+	if(IS_ABSOLUTE "${PT_CACHE_ROOT}")
+		message(FATAL_ERROR "PT_CACHE_ROOT must stay relative because the runtime resolves it against path_tracer.runtime.json")
+	endif()
+
 	set(NBL_INCLUDE_SERACH_DIRECTORIES
 		"${CMAKE_CURRENT_SOURCE_DIR}/include"
 	)
@@ -16,25 +43,80 @@ if(NBL_BUILD_IMGUI)
 	)
 
 	nbl_create_executable_project("" "" "${NBL_INCLUDE_SERACH_DIRECTORIES}" "${NBL_LIBRARIES}" "${NBL_EXECUTABLE_PROJECT_CREATION_PCH_TARGET}")
+	add_dependencies(${EXECUTABLE_NAME} argparse)
+	target_include_directories(${EXECUTABLE_NAME} PUBLIC $<TARGET_PROPERTY:argparse,INTERFACE_INCLUDE_DIRECTORIES>)
+	target_compile_definitions(${EXECUTABLE_NAME} PRIVATE PT_BUILD_CONFIG_NAME=\"$<CONFIG>\")
+	if(PT_BUILD_MODE STREQUAL "SPECIALIZED")
+		target_compile_definitions(${EXECUTABLE_NAME} PRIVATE PT_BUILD_MODE_SPECIALIZED=1)
+	else()
+		target_compile_definitions(${EXECUTABLE_NAME} PRIVATE PT_BUILD_MODE_WALLTIME_OPTIMIZED=1)
+	endif()
+	if(NOT PT_CACHE_ROOT STREQUAL "")
+		string(REPLACE "\\" "/" PT_CACHE_ROOT_JSON "${PT_CACHE_ROOT}")
+		configure_file(
+			"${CMAKE_CURRENT_SOURCE_DIR}/path_tracer.runtime.json.in"
+			"${CMAKE_CURRENT_BINARY_DIR}/path_tracer.runtime.json"
+			@ONLY
+		)
+		file(GENERATE
+			OUTPUT "$<TARGET_FILE_DIR:${EXECUTABLE_NAME}>/path_tracer.runtime.json"
+			INPUT "${CMAKE_CURRENT_BINARY_DIR}/path_tracer.runtime.json"
+		)
+	endif()
 
-	if(NBL_EMBED_BUILTIN_RESOURCES)
-		set(_BR_TARGET_ ${EXECUTABLE_NAME}_builtinResourceData)
-		set(RESOURCE_DIR "app_resources")
-
-		get_filename_component(_SEARCH_DIRECTORIES_ "${CMAKE_CURRENT_SOURCE_DIR}" ABSOLUTE)
-		get_filename_component(_OUTPUT_DIRECTORY_SOURCE_ "${CMAKE_CURRENT_BINARY_DIR}/src" ABSOLUTE)
-		get_filename_component(_OUTPUT_DIRECTORY_HEADER_ "${CMAKE_CURRENT_BINARY_DIR}/include" ABSOLUTE)
-
-		file(GLOB_RECURSE BUILTIN_RESOURCE_FILES RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}/${RESOURCE_DIR}" "${CMAKE_CURRENT_SOURCE_DIR}/${RESOURCE_DIR}/*")
-
-		foreach(RES_FILE ${BUILTIN_RESOURCE_FILES})
-			LIST_BUILTIN_RESOURCE(RESOURCES_TO_EMBED "${RES_FILE}")
-		endforeach()
-
-		ADD_CUSTOM_BUILTIN_RESOURCES(${_BR_TARGET_} RESOURCES_TO_EMBED "${_SEARCH_DIRECTORIES_}" "${RESOURCE_DIR}" "nbl::this_example::builtin" "${_OUTPUT_DIRECTORY_HEADER_}" "${_OUTPUT_DIRECTORY_SOURCE_}")
+	set(SM 6_8)
+	set(OUTPUT_DIRECTORY "${CMAKE_CURRENT_BINARY_DIR}/auto-gen")
+	set(COMPILE_OPTIONS
+		-I "${CMAKE_CURRENT_SOURCE_DIR}/include"
+		-I "${CMAKE_CURRENT_SOURCE_DIR}/app_resources/hlsl"
+		-I "${NBL_ROOT_PATH}/include" # workaround, the same thing like in IES I will address this issue later
+		-T "lib_${SM}"
+		-Wno-conversion
+		-Wno-sign-conversion
+		-Wno-float-conversion
+		-Wno-shorten-64-to-32
+		-Wno-shadow
+		-Wno-literal-range
+	)
 
-		LINK_BUILTIN_RESOURCES_TO_TARGET(${EXECUTABLE_NAME} ${_BR_TARGET_})
+	# Keep the payload flat and explicit here. Once Nabla PR #988 lands, these per-rule compile axes should move to first-class packaged-variant support there.
+	PT_APPEND_SPIRV_RULE(VAR JSON INPUT "app_resources/hlsl/spirv/pt.compute.sphere.proxy.hlsl" KEY "pt.compute.sphere" COMPILE_OPTIONS "-DPT_VARIANT_USE_RWMC=0" "-DPT_VARIANT_SCENE_KIND=${PT_SCENE_SPHERE}" "-DPT_VARIANT_ENTRYPOINT_KIND=${PT_ENTRYPOINT_RUNTIME_UNIFORM}" "-DPT_VARIANT_RUNTIME_POLYGON_METHOD=0")
+	PT_APPEND_SPIRV_RULE(VAR JSON INPUT "app_resources/hlsl/spirv/pt.compute.sphere.rwmc.proxy.hlsl" KEY "pt.compute.sphere.rwmc" COMPILE_OPTIONS "-DPT_VARIANT_USE_RWMC=1" "-DPT_VARIANT_SCENE_KIND=${PT_SCENE_SPHERE}" "-DPT_VARIANT_ENTRYPOINT_KIND=${PT_ENTRYPOINT_RUNTIME_UNIFORM}" "-DPT_VARIANT_RUNTIME_POLYGON_METHOD=0")
+	if(PT_BUILD_MODE STREQUAL "SPECIALIZED")
+		PT_APPEND_SPIRV_RULE(VAR JSON INPUT "app_resources/hlsl/spirv/pt.compute.triangle.linear.proxy.hlsl" KEY "pt.compute.triangle.linear" COMPILE_OPTIONS "-DPT_VARIANT_USE_RWMC=0" "-DPT_VARIANT_ENTRYPOINT_KIND=${PT_ENTRYPOINT_LINEAR}")
+		PT_APPEND_SPIRV_RULE(VAR JSON INPUT "app_resources/hlsl/spirv/pt.compute.triangle.persistent.proxy.hlsl" KEY "pt.compute.triangle.persistent" COMPILE_OPTIONS "-DPT_VARIANT_USE_RWMC=0" "-DPT_VARIANT_ENTRYPOINT_KIND=${PT_ENTRYPOINT_PERSISTENT}")
+		PT_APPEND_SPIRV_RULE(VAR JSON INPUT "app_resources/hlsl/spirv/pt.compute.triangle.rwmc.linear.proxy.hlsl" KEY "pt.compute.triangle.rwmc.linear" COMPILE_OPTIONS "-DPT_VARIANT_USE_RWMC=1" "-DPT_VARIANT_ENTRYPOINT_KIND=${PT_ENTRYPOINT_LINEAR}")
+		PT_APPEND_SPIRV_RULE(VAR JSON INPUT "app_resources/hlsl/spirv/pt.compute.triangle.rwmc.persistent.proxy.hlsl" KEY "pt.compute.triangle.rwmc.persistent" COMPILE_OPTIONS "-DPT_VARIANT_USE_RWMC=1" "-DPT_VARIANT_ENTRYPOINT_KIND=${PT_ENTRYPOINT_PERSISTENT}")
+		PT_APPEND_SPIRV_RULE(VAR JSON INPUT "app_resources/hlsl/spirv/pt.compute.rectangle.rwmc.linear.proxy.hlsl" KEY "pt.compute.rectangle.rwmc.linear" COMPILE_OPTIONS "-DPT_VARIANT_USE_RWMC=1" "-DPT_VARIANT_SCENE_KIND=${PT_SCENE_RECTANGLE}" "-DPT_VARIANT_ENTRYPOINT_KIND=${PT_ENTRYPOINT_LINEAR}" "-DPT_VARIANT_RUNTIME_POLYGON_METHOD=1")
+		PT_APPEND_SPIRV_RULE(VAR JSON INPUT "app_resources/hlsl/spirv/pt.compute.rectangle.rwmc.persistent.proxy.hlsl" KEY "pt.compute.rectangle.rwmc.persistent" COMPILE_OPTIONS "-DPT_VARIANT_USE_RWMC=1" "-DPT_VARIANT_SCENE_KIND=${PT_SCENE_RECTANGLE}" "-DPT_VARIANT_ENTRYPOINT_KIND=${PT_ENTRYPOINT_PERSISTENT}" "-DPT_VARIANT_RUNTIME_POLYGON_METHOD=1")
+	else()
+		PT_APPEND_SPIRV_RULE(VAR JSON INPUT "app_resources/hlsl/spirv/pt.compute.triangle.proxy.hlsl" KEY "pt.compute.triangle" COMPILE_OPTIONS "-DPT_VARIANT_USE_RWMC=0" "-DPT_VARIANT_SCENE_KIND=${PT_SCENE_TRIANGLE}" "-DPT_VARIANT_ENTRYPOINT_KIND=${PT_ENTRYPOINT_RUNTIME_UNIFORM}" "-DPT_VARIANT_RUNTIME_POLYGON_METHOD=1")
+		PT_APPEND_SPIRV_RULE(VAR JSON INPUT "app_resources/hlsl/spirv/pt.compute.triangle.rwmc.proxy.hlsl" KEY "pt.compute.triangle.rwmc" COMPILE_OPTIONS "-DPT_VARIANT_USE_RWMC=1" "-DPT_VARIANT_SCENE_KIND=${PT_SCENE_TRIANGLE}" "-DPT_VARIANT_ENTRYPOINT_KIND=${PT_ENTRYPOINT_RUNTIME_UNIFORM}" "-DPT_VARIANT_RUNTIME_POLYGON_METHOD=1")
+		PT_APPEND_SPIRV_RULE(VAR JSON INPUT "app_resources/hlsl/spirv/pt.compute.rectangle.rwmc.proxy.hlsl" KEY "pt.compute.rectangle.rwmc" COMPILE_OPTIONS "-DPT_VARIANT_USE_RWMC=1" "-DPT_VARIANT_SCENE_KIND=${PT_SCENE_RECTANGLE}" "-DPT_VARIANT_ENTRYPOINT_KIND=${PT_ENTRYPOINT_RUNTIME_UNIFORM}" "-DPT_VARIANT_RUNTIME_POLYGON_METHOD=1")
 	endif()
-endif()
+	PT_APPEND_SPIRV_RULE(VAR JSON INPUT "app_resources/hlsl/spirv/pt.compute.rectangle.proxy.hlsl" KEY "pt.compute.rectangle" COMPILE_OPTIONS "-DPT_VARIANT_USE_RWMC=0" "-DPT_VARIANT_SCENE_KIND=${PT_SCENE_RECTANGLE}" "-DPT_VARIANT_ENTRYPOINT_KIND=${PT_ENTRYPOINT_RUNTIME_UNIFORM}" "-DPT_VARIANT_RUNTIME_POLYGON_METHOD=1")
+	PT_APPEND_SPIRV_RULE(VAR JSON INPUT "app_resources/hlsl/resolve.comp.hlsl" KEY "pt.compute.resolve")
+	PT_APPEND_SPIRV_RULE(VAR JSON INPUT "app_resources/hlsl/spirv/pt.misc.proxy.hlsl" KEY "pt.misc")
+	PT_FINALIZE_JSON_PAYLOAD(INOUT JSON)
 
+	NBL_CREATE_NSC_COMPILE_RULES(
+		TARGET ${EXECUTABLE_NAME}SPIRV
+		LINK_TO ${EXECUTABLE_NAME}
+		DEPENDS ${DEPENDS}
+		BINARY_DIR ${OUTPUT_DIRECTORY}
+		MOUNT_POINT_DEFINE NBL_THIS_EXAMPLE_BUILD_MOUNT_POINT
+		COMMON_OPTIONS ${COMPILE_OPTIONS}
+		OUTPUT_VAR KEYS
+		INCLUDE nbl/this_example/builtin/build/spirv/keys.hpp
+		NAMESPACE nbl::this_example::builtin::build
+		INPUTS ${JSON}
+	)
 
+	NBL_CREATE_RESOURCE_ARCHIVE(
+		NAMESPACE nbl::this_example::builtin::build
+		TARGET ${EXECUTABLE_NAME}_builtinsBuild
+		LINK_TO ${EXECUTABLE_NAME}
+		BIND ${OUTPUT_DIRECTORY}
+		BUILTINS ${KEYS}
+	)
+endif()

31_HLSLPathTracer/app_resources/hlsl/compute.render.common.hlsl

@@ -0,0 +1,62 @@
+#ifndef PATH_TRACER_USE_RWMC
+#error PATH_TRACER_USE_RWMC must be defined before including compute.render.common.hlsl
+#endif
+
+#ifndef PATH_TRACER_ENABLE_LINEAR
+#define PATH_TRACER_ENABLE_LINEAR 1
+#endif
+
+#ifndef PATH_TRACER_ENABLE_PERSISTENT
+#define PATH_TRACER_ENABLE_PERSISTENT 1
+#endif
+
+#if !PATH_TRACER_ENABLE_LINEAR && !PATH_TRACER_ENABLE_PERSISTENT
+#error At least one path tracer entrypoint mode must be enabled
+#endif
+
+#include "nbl/builtin/hlsl/cpp_compat.hlsl"
+#include "nbl/builtin/hlsl/glsl_compat/core.hlsl"
+#include "nbl/builtin/hlsl/random/pcg.hlsl"
+#include "nbl/builtin/hlsl/random/xoroshiro.hlsl"
+#if PATH_TRACER_ENABLE_PERSISTENT
+#include "nbl/builtin/hlsl/morton.hlsl"
+#endif
+#include "nbl/builtin/hlsl/bxdf/reflection.hlsl"
+#include "nbl/builtin/hlsl/bxdf/transmission.hlsl"
+#include "nbl/builtin/hlsl/path_tracing/basic_ray_gen.hlsl"
+#include "nbl/builtin/hlsl/path_tracing/unidirectional.hlsl"
+#include "render_common.hlsl"
+
+#if PATH_TRACER_USE_RWMC
+#include "nbl/builtin/hlsl/rwmc/CascadeAccumulator.hlsl"
+#include "render_rwmc_common.hlsl"
+#else
+#include "nbl/builtin/hlsl/path_tracing/default_accumulator.hlsl"
+#endif
+
+#if PATH_TRACER_USE_RWMC
+[[vk::push_constant]] RenderRWMCPushConstants pc;
+#else
+[[vk::push_constant]] RenderPushConstants pc;
+#endif
+
+[[vk::combinedImageSampler]] [[vk::binding(0, 0)]] Texture2D<float3> envMap;
+[[vk::combinedImageSampler]] [[vk::binding(0, 0)]] SamplerState envSampler;
+
+[[vk::combinedImageSampler]] [[vk::binding(1, 0)]] Texture2D<uint2> scramblebuf;
+[[vk::combinedImageSampler]] [[vk::binding(1, 0)]] SamplerState scrambleSampler;
+
+[[vk::image_format("rgba16f")]] [[vk::binding(2, 0)]] RWTexture2DArray<float32_t4> outImage;
+
+#if PATH_TRACER_USE_RWMC
+[[vk::image_format("rgba16f")]] [[vk::binding(3, 0)]] RWTexture2DArray<float32_t4> cascade;
+#endif
+
+#include "example_common.hlsl"
+#include "rand_gen.hlsl"
+#include "intersector.hlsl"
+#include "material_system.hlsl"
+#include "next_event_estimator.hlsl"
+
+using namespace nbl;
+using namespace hlsl;

31_HLSLPathTracer/app_resources/hlsl/compute.render.linear.entrypoints.hlsl

@@ -0,0 +1,25 @@
+#ifndef PATH_TRACER_ENTRYPOINT_NAME
+#define PATH_TRACER_ENTRYPOINT_NAME main
+#endif
+
+#ifndef PATH_TRACER_ENTRYPOINT_POLYGON_METHOD
+#ifdef PATH_TRACER_RUNTIME_POLYGON_METHOD
+#define PATH_TRACER_ENTRYPOINT_POLYGON_METHOD PATH_TRACER_RUNTIME_POLYGON_METHOD
+#else
+#define PATH_TRACER_ENTRYPOINT_POLYGON_METHOD PPM_APPROX_PROJECTED_SOLID_ANGLE
+#endif
+#endif
+
+#if !PATH_TRACER_ENABLE_LINEAR
+#error Linear entrypoint requested while PATH_TRACER_ENABLE_LINEAR is disabled
+#endif
+
+[numthreads(RenderWorkgroupSize, 1, 1)]
+[shader("compute")]
+void PATH_TRACER_ENTRYPOINT_NAME(uint32_t3 threadID : SV_DispatchThreadID)
+{
+	pathtracer_render_variant::runLinear(threadID, PATH_TRACER_ENTRYPOINT_POLYGON_METHOD);
+}
+
+#undef PATH_TRACER_ENTRYPOINT_POLYGON_METHOD
+#undef PATH_TRACER_ENTRYPOINT_NAME

31_HLSLPathTracer/app_resources/hlsl/compute.render.persistent.entrypoints.hlsl

@@ -0,0 +1,25 @@
+#ifndef PATH_TRACER_ENTRYPOINT_NAME
+#define PATH_TRACER_ENTRYPOINT_NAME mainPersistent
+#endif
+
+#ifndef PATH_TRACER_ENTRYPOINT_POLYGON_METHOD
+#ifdef PATH_TRACER_RUNTIME_POLYGON_METHOD
+#define PATH_TRACER_ENTRYPOINT_POLYGON_METHOD PATH_TRACER_RUNTIME_POLYGON_METHOD
+#else
+#define PATH_TRACER_ENTRYPOINT_POLYGON_METHOD PPM_APPROX_PROJECTED_SOLID_ANGLE
+#endif
+#endif
+
+#if !PATH_TRACER_ENABLE_PERSISTENT
+#error Persistent entrypoint requested while PATH_TRACER_ENABLE_PERSISTENT is disabled
+#endif
+
+[numthreads(RenderWorkgroupSize, 1, 1)]
+[shader("compute")]
+void PATH_TRACER_ENTRYPOINT_NAME(uint32_t3 threadID : SV_DispatchThreadID)
+{
+	pathtracer_render_variant::runPersistent(PATH_TRACER_ENTRYPOINT_POLYGON_METHOD);
+}
+
+#undef PATH_TRACER_ENTRYPOINT_POLYGON_METHOD
+#undef PATH_TRACER_ENTRYPOINT_NAME