Update pal from commit a33b9bbb

Fix SDMA ODR warnings MRT Use Minimum Resolution Add 2 new TraceSources needed for pull UserMarker into RRA Memory Leak in Execute Indirect Shader Update submodule devdriver Expose the flag 'enablePops' to PAL pipeline user Fix the regression issue for the dynamicDualSourceBlend in fastlinklibrary mode Minor fix for split barrier Bresenham lines with adjacency failing Cleanup gfx query pool code. Don't read over PCIe when using SDMA predication Implement workaround WaCwsrThreadgroupTrap Update spill table register in rpm GenerateCmdDisptachTaskMesh on gfx10/11 Handle Directory Error in Command Buffer Dumping Update submodule address-lib Moves supportSplitReleaseAcquire from gfxipProperrties to queue queueProperties Remove redundant Inits within ElfProcessor and PipelineProcessor Fix PAL_CLIENT_INTERFACE_VERSION Fix SaschaWillems perf regression due to OREO_MODE Workaround Define Release PAL_ANALYSIS_ASSUME to void(0) Update PM4 headers Always log out layout transition info in interface logger (for barrier log) Add alignment asserts for several gfx9 packets Resolve address alignment Set reg for PA_SU_POLY_OFFSET_DB_FMT_CNTL with no depth buffer bound. Fix Vector Erase and define a new FastErase() function Change definition of release token used for CmdRelease and CmdAcquire Optimize split barrier: defer CpDma wait from Release to Acquire time Implement Compute InheritStateFromCmdBuf Add PERF_COUNTER_WINDOW Support Convert enum opcodes for comparison Fix spill threshold so that driver uses minimum spill threshold Add SPM buffer wrap warning Optimize split barrier to not acquire retired release token again Change SubresRange member data types to save size Fix VKD3D Execute Indirect Test Failures Use double when converting float to fixed point number Remove outdated ADDRLIB_VERSION checks RenderOp trace controller Allow mipmaps for T2T SDMA copies Update ASIC Info chunk def Use pairwise tree reduction in msaaFmaskResolve8x Remove hard-coded NullGpuId and AsicRevision numbers Add RGP instrumentation marker to amp/task ACE dispatch Correct the matrixRowOrder in plane 2 of the P208 format Relax blit operations Place an upper limit on internalMemMgr pool size growth Rename PipelineUploader to CodeObjectUploader and move it to its own files Correct the sampleLocX of the packed format Navi2 VRS data cannot be read from miptail Changes to ExecuteIndirectV2 path on Gfx9 Controller config update and active controller Disable use of PERF_COUNTER_WINDOW Add treatSrcAsSrgb flag into ScaledCopyInfo Adding L2 cache values for individual ASICs RenderOp Controller updates Customize Offset in Cmd Generator Creation Fix interface log for ExternalImageOpenInfo to match tip definition Update COVERAGE_TO_SHADER_SELECT bit when underestimation is enabled Remove legacy settings Add new enum for FsrOvr-ApArea Fix nullptr deref in SPM Trace processing
GPUOpen-Drivers · Aug 8, 2024 · d3bfe50 · d3bfe50
1 parent 695337a
commit d3bfe50
Show file tree

Hide file tree

Showing 268 changed files with 67,642 additions and 71,413 deletions.
diff --git a/README.md b/README.md
@@ -7,7 +7,7 @@ PAL client drivers will have no HW-specific code; their responsibility is to tra
 PAL client drivers should have little OS-specific code. PAL and its companion utility collection provide OS abstractions for almost everything a client might need, but there are some cases where this is unavoidable:
 
 * Handling dynamic library infrastructure. I.e., the client has to implement DllMain() on Windows, etc.
-* OS-specific APIs or extensions. DX may have Windows-specific functionality in the core API, and Vulkan/Mantle may export certain OS-specific features as extensions (like for presenting contents to the screen).
+* OS-specific APIs or extensions. DX may have Windows-specific functionality in the core API, and Vulkan may export certain OS-specific features as extensions (like for presenting contents to the screen).
 * Single OS clients (e.g., DX) may choose to make OS-specific calls directly simply out of convenience with no down side.
 
 PAL is a source deliverable. Clients will periodically promote PAL's source into their own tree and build a static pal.lib as part of their build process.

diff --git a/cmake/PalBuildParameters.cmake b/cmake/PalBuildParameters.cmake
@@ -46,12 +46,13 @@ pal_bp(PAL_BUILD_GPUUTIL ON)
 pal_bp(PAL_DEVELOPER_BUILD OFF)
 #endif
 
-# Build PAL with Operating System support
-pal_bp(PAL_BUILD_OSS    ON)
+if (PAL_CLIENT_INTERFACE_MAJOR_VERSION LESS 888)
+    # Build PAL with Operating System support
+    pal_bp(PAL_BUILD_OSS    ON)
 #if PAL_BUILD_OSS2_4
-pal_bp(PAL_BUILD_OSS2_4 ON DEPENDS_ON PAL_BUILD_OSS)
+    pal_bp(PAL_BUILD_OSS2_4 ON DEPENDS_ON PAL_BUILD_OSS)
 #endif
-pal_bp(PAL_BUILD_OSS4   ON DEPENDS_ON PAL_BUILD_OSS)
+endif()
 
 # Clients must define this variable so pal know what API it's facilitating
 pal_bp(PAL_CLIENT "-1" MODE "FATAL_ERROR")
@@ -115,12 +116,12 @@ if (PAL_BUILD_GFX9)
                     PAL_BUILD_GFX11
               )
     endif()
-
-    # We'll remove this soon...
-    pal_bp( PAL_BUILD_PHOENIX2 ON MODE "AUTHOR_WARNING"
-            ASIC_CONFIG
-                PAL_BUILD_GFX11
-          )
+    if (PAL_CLIENT_INTERFACE_MAJOR_VERSION LESS 888)
+        pal_bp( PAL_BUILD_PHOENIX2 ON MODE "AUTHOR_WARNING"
+                ASIC_CONFIG
+                    PAL_BUILD_GFX11
+              )
+    endif()
 #endif
 
 endif() # PAL_BUILD_GFX9

diff --git a/cmake/PalCodegen.cmake b/cmake/PalCodegen.cmake
@@ -58,144 +58,6 @@ function(convert_pal_settings_name SETTINGS_FILE OUT_BASENAME_VAR FOR_FILE)
     set(${OUT_BASENAME_VAR} ${OUT_BASENAME} PARENT_SCOPE)
 endfunction()
 
-function(target_pal_settings TARGET)
-    set(options NO_REGISTRY)
-    set(singleValArgs MAGIC_BUF CLASS_NAME CODE_TEMPLATE OUT_DIR OUT_BASENAME ROOT_BINARY_DIR)
-    set(multiValArgs SETTINGS ENSURE_DELETED ADDL_NAMESPACES)
-    cmake_parse_arguments(PARSE_ARGV 1 SETGEN "${options}" "${singleValArgs}" "${multiValArgs}")
-
-    # Asserts
-    if (NOT SETGEN_SETTINGS)
-        message(FATAL_ERROR "No settings inputs provided!")
-    endif()
-    if (SETGEN_OUT_BASENAME OR SETGEN_CLASS_NAME)
-        list(LENGTH SETGEN_SETTINGS SETGEN_SETTINGS_LEN)
-        if (${SETGEN_SETTINGS_LEN} GREATER 1)
-            message(FATAL_ERROR "With provided options, only one settings file can be set!")
-        endif()
-    endif()
-    if (SETGEN_ENSURE_DELETED)
-        foreach(DELETED_FILE ${SETGEN_ENSURE_DELETED})
-            get_filename_component(DELETED_FILE "${DELETED_FILE}" ABSOLUTE)
-            if(EXISTS ${DELETED_FILE})
-                message(FATAL_ERROR "Conflicts detected: ${DELETED_FILE} should be deleted!")
-            endif()
-        endforeach()
-    endif()
-
-    # Apply defaults
-    if (NOT SETGEN_OUT_DIR)
-        set(SETGEN_OUT_DIR "${CMAKE_CURRENT_BINARY_DIR}/${TARGET}_settings")
-    endif()
-    if (NOT SETGEN_CODE_TEMPLATE)
-        set(SETGEN_CODE_TEMPLATE "${PAL_GEN_DIR}/settingsCodeTemplates.py")
-    endif()
-    get_filename_component(SETGEN_CODE_TEMPLATE "${SETGEN_CODE_TEMPLATE}" ABSOLUTE)
-
-    # List dependencies which cause a rerun if modified (cmdline changes implicitly rerun).
-    set(ADDL_DEPS
-       "${PAL_GEN_DIR}/genSettingsCode.py"
-       "${SETGEN_CODE_TEMPLATE}"
-    )
-
-    # Build up common args based on the provided settings
-    set(ADDL_ARGS
-        "--codeTemplateFile" "${SETGEN_CODE_TEMPLATE}"
-        "--outDir" "${SETGEN_OUT_DIR}"
-    )
-    if (NOT SETGEN_NO_REGISTRY)
-        list(APPEND ADDL_ARGS "--genRegistryCode")
-    endif()
-    if (SETGEN_CLASS_NAME)
-        list(APPEND ADDL_ARGS "--classNameOverride" "${SETGEN_CLASS_NAME}")
-    endif()
-    if (SETGEN_ADDL_NAMESPACES)
-        list(APPEND ADDL_ARGS "--additionalNamespaces" "${SETGEN_ADDL_NAMESPACES}")
-    endif()
-
-    if (SETGEN_MAGIC_BUF)
-        get_filename_component(SETGEN_MAGIC_BUF "${SETGEN_MAGIC_BUF}" ABSOLUTE)
-        list(APPEND ADDL_DEPS "${SETGEN_MAGIC_BUF}")
-        list(APPEND ADDL_ARGS "--magicBuffer" "${SETGEN_MAGIC_BUF}")
-    endif()
-
-    # Create and include output dir
-    file(MAKE_DIRECTORY ${SETGEN_OUT_DIR})
-    target_include_directories(${TARGET} PRIVATE ${SETGEN_OUT_DIR})
-
-    foreach(SETTINGS_FILE ${SETGEN_SETTINGS})
-        if (SETGEN_OUT_BASENAME)
-            set(OUT_BASENAME ${SETGEN_OUT_BASENAME})
-        else()
-            convert_pal_settings_name(${SETTINGS_FILE} OUT_BASENAME TRUE)
-        endif()
-
-        get_filename_component(SETTINGS_FILE "${SETTINGS_FILE}" ABSOLUTE)
-        target_sources(${TARGET} PRIVATE
-                       ${SETGEN_OUT_DIR}/${OUT_BASENAME}.h
-                       ${SETGEN_OUT_DIR}/${OUT_BASENAME}.cpp
-                       ${SETTINGS_FILE}
-        )
-        set_source_files_properties(
-                       ${SETGEN_OUT_DIR}/${OUT_BASENAME}.h
-                       ${SETGEN_OUT_DIR}/${OUT_BASENAME}.cpp
-                       TARGET_DIRECTORY ${TARGET}
-                       PROPERTIES GENERATED ON
-        )
-
-        if ((NOT EXISTS ${SETGEN_OUT_DIR}/${OUT_BASENAME}.h) OR
-            (NOT EXISTS ${SETGEN_OUT_DIR}/${OUT_BASENAME}.cpp))
-            # Generate these during configuration so that they are guaranteed to exist.
-            execute_process(
-                COMMAND ${Python3_EXECUTABLE} ${PAL_GEN_DIR}/genSettingsCode.py
-                        --settingsFile ${SETTINGS_FILE}
-                        --outFilename ${OUT_BASENAME}
-                        ${ADDL_ARGS}
-            )
-        endif()
-
-        convert_pal_settings_name(${SETTINGS_FILE} SETTING_TGT FALSE)
-        add_custom_command(
-            OUTPUT ${SETGEN_OUT_DIR}/${OUT_BASENAME}.cpp
-                   ${SETGEN_OUT_DIR}/${OUT_BASENAME}.h
-            COMMAND ${Python3_EXECUTABLE} ${PAL_GEN_DIR}/genSettingsCode.py
-                    --settingsFile ${SETTINGS_FILE}
-                    --outFilename ${OUT_BASENAME}
-                    ${ADDL_ARGS}
-            COMMENT "Generating settings from ${SETTINGS_FILE}..."
-            DEPENDS ${SETTINGS_FILE}
-                    ${ADDL_DEPS}
-        )
-        add_custom_target(${SETTING_TGT}
-            DEPENDS ${SETGEN_OUT_DIR}/${OUT_BASENAME}.cpp
-                    ${SETGEN_OUT_DIR}/${OUT_BASENAME}.h
-            SOURCES ${SETGEN_OUT_DIR}/${OUT_BASENAME}.cpp
-                    ${SETGEN_OUT_DIR}/${OUT_BASENAME}.h
-        )
-
-        add_dependencies(${TARGET} ${SETTING_TGT})
-        set_target_properties(${SETTING_TGT}
-            PROPERTIES
-                FOLDER "${CMAKE_FOLDER}/Generate/Settings"
-        )
-
-        if (SETGEN_ROOT_BINARY_DIR)
-            source_group(
-                TREE ${PAL_BINARY_DIR}
-                FILES
-                    ${SETGEN_OUT_DIR}/${OUT_BASENAME}.h
-                    ${SETGEN_OUT_DIR}/${OUT_BASENAME}.cpp
-            )
-        else()
-            source_group("Generated/Settings"
-                FILES
-                    ${SETGEN_OUT_DIR}/${OUT_BASENAME}.h
-                    ${SETGEN_OUT_DIR}/${OUT_BASENAME}.cpp
-            )
-        endif()
-    endforeach()
-endfunction()
-
 function(pal_gen_settings)
     # INPUT_JSON:
     #     Path to a JSON file describing all settings of a component.
@@ -294,6 +156,7 @@ function(pal_gen_settings)
                 ${DEVDRIVER_PATH}/apis/settings/codegen/settings.cpp.jinja2
                 ${DEVDRIVER_PATH}/apis/settings/codegen/settings.h.jinja2
     )
+
     add_custom_target(${SETTING_TGT}
         DEPENDS ${SETTINGS_OUT_DIR}/${GENERATED_HEADER_FILENAME}
                 ${SETTINGS_OUT_DIR}/${GENERATED_SOURCE_FILENAME}

diff --git a/cmake/PalCompileDefinitions.cmake b/cmake/PalCompileDefinitions.cmake
@@ -73,9 +73,9 @@ function(pal_compile_definitions_gpu TARGET)
             target_compile_definitions(${TARGET} INTERFACE PAL_BUILD_PHOENIX=$<BOOL:${PAL_BUILD_GFX9}>)
             target_compile_definitions(${TARGET} INTERFACE PAL_BUILD_PHOENIX1=$<BOOL:${PAL_BUILD_GFX9}>)
         endif()
-
-        # We'll remove this soon...
-        target_compile_definitions(${TARGET} INTERFACE PAL_BUILD_PHOENIX2=$<BOOL:${PAL_BUILD_GFX9}>)
+        if (PAL_CLIENT_INTERFACE_MAJOR_VERSION LESS 888)
+            target_compile_definitions(${TARGET} INTERFACE PAL_BUILD_PHOENIX2=$<BOOL:${PAL_BUILD_GFX9}>)
+        endif()
 
     endif()
 endfunction()
@@ -167,7 +167,7 @@ function(pal_compile_definitions TARGET)
 
     target_compile_definitions(${TARGET} PRIVATE PAL_BUILD_CORE=$<BOOL:${PAL_BUILD_CORE}>)
 
-    if(PAL_AMDGPU_BUILD)
+    if (PAL_AMDGPU_BUILD)
         message(STATUS "PAL build with amdgpu back-end enabled")
 
         target_compile_definitions(${TARGET} PUBLIC PAL_AMDGPU_BUILD=1)
@@ -177,14 +177,12 @@ function(pal_compile_definitions TARGET)
         target_compile_definitions(${TARGET} PRIVATE PAL_HAVE_WAYLAND_PLATFORM=$<BOOL:${PAL_BUILD_WAYLAND}>)
     endif()
 
-    if (PAL_BUILD_OSS)
-        target_compile_definitions(${TARGET} PRIVATE PAL_BUILD_OSS=1)
-
+    if (PAL_CLIENT_INTERFACE_MAJOR_VERSION LESS 888)
+        if (PAL_BUILD_OSS)
 #if PAL_BUILD_OSS2_4
-        target_compile_definitions(${TARGET} PUBLIC PAL_BUILD_OSS2_4=$<BOOL:${PAL_BUILD_OSS2_4}>)
+            target_compile_definitions(${TARGET} INTERFACE PAL_BUILD_OSS2_4=$<BOOL:${PAL_BUILD_OSS2_4}>)
 #endif
-
-        target_compile_definitions(${TARGET} PRIVATE PAL_BUILD_OSS4=$<BOOL:${PAL_BUILD_OSS4}>)
+        endif()
     endif()
 
 #if PAL_BUILD_RPM_GFX_SHADERS

diff --git a/inc/core/g_palPipelineAbiMetadata.h b/inc/core/g_palPipelineAbiMetadata.h
@@ -1917,7 +1917,9 @@ struct PipelineMetadata
             uint8 psSampleMask           : 1;
             /// Indicates whether or not the pipeline uses continuations passing shaders (CPS).
             uint8 usesCps                : 1;
-            uint8 reserved               : 3;
+            /// If using continuations passing shaders (CPS), stack is in global rather than scratch memory.
+            uint8 cpsGlobal              : 1;
+            uint8 reserved               : 2;
         };
         uint8 uAll;
     } flags;
@@ -1952,10 +1954,11 @@ struct PipelineMetadata
             uint32 psDummyExport           : 1;
             uint32 psSampleMask            : 1;
             uint32 usesCps                 : 1;
+            uint32 cpsGlobal               : 1;
             uint32 streamoutVertexStrides  : 1;
             uint32 graphicsRegister        : 1;
             uint32 computeRegister         : 1;
-            uint32 reserved                : 3;
+            uint32 reserved                : 2;
         };
         uint32 uAll;
     } hasEntry;
@@ -2017,6 +2020,7 @@ namespace PipelineMetadataKey
     static constexpr char PsDummyExport[]           = ".ps_dummy_export";
     static constexpr char PsSampleMask[]            = ".ps_sample_mask";
     static constexpr char UsesCps[]                 = ".uses_cps";
+    static constexpr char CpsGlobal[]               = ".cps_global";
     static constexpr char StreamoutVertexStrides[]  = ".streamout_vertex_strides";
     static constexpr char GraphicsRegisters[]       = ".graphics_registers";
     static constexpr char ComputeRegisters[]        = ".compute_registers";

diff --git a/inc/core/g_palPipelineAbiMetadataImpl.h b/inc/core/g_palPipelineAbiMetadataImpl.h
@@ -5700,6 +5700,21 @@ inline Result DeserializePipelineMetadata(
                 break;
             }
 
+            case HashLiteralString(PipelineMetadataKey::CpsGlobal):
+            {
+                PAL_ASSERT(pMetadata->hasEntry.cpsGlobal == 0);
+                bool value = false;
+                result = pReader->UnpackNext(&value);
+
+                if (result == Result::Success)
+                {
+                    pMetadata->flags.cpsGlobal = value;
+                }
+
+                pMetadata->hasEntry.cpsGlobal = (result == Result::Success);
+                break;
+            }
+
             case HashLiteralString(PipelineMetadataKey::StreamoutVertexStrides):
                 PAL_ASSERT(pMetadata->hasEntry.streamoutVertexStrides == 0);
                 result = pReader->UnpackNext(&pMetadata->streamoutVertexStrides);

diff --git a/inc/core/pal.h b/inc/core/pal.h
@@ -626,7 +626,7 @@ enum class TriState : uint8
  * OS abstractions for almost everything a client might need, but there are some cases where this is unavoidable:
  *
  * + Handling dynamic library infrastructure.  I.e., the client has to implement DllMain() on Windows, etc.
- * + OS-specific APIs or extensions.  DX may have Windows-specific functionality in the core API, and Vulkan/Mantle may
+ * + OS-specific APIs or extensions.  DX may have Windows-specific functionality in the core API, and Vulkan may
  *   export certain OS-specific features as extensions (like for presenting contents to the screen).
  * + Single OS clients (e.g., DX) may choose to make OS-specific calls directly simply out of convenience with no down
  *   side.
@@ -666,7 +666,7 @@ enum class TriState : uint8
  *
  * ### Introduction
  * PAL's core interface is defined in the @ref Pal namespace, and defines an object-oriented model for interacting with
- * the GPU and OS.  The interface closely resembles the Mantle, Vulkan, and DX12 APIs.  Some common features of these
+ * the GPU and OS.  The interface closely resembles the Vulkan and DX12 APIs.  Some common features of these
  * APIs that are central to the PAL interface:
  *
  * - All shader stages, and some additional "shader adjacent" state, are glommed together into a monolithic pipeline