Reorganize implicit scaling files

Related-To: NEO-6262


Signed-off-by: Zbigniew Zdanowicz <[email protected]>

Author: zzdanowicz

shared/source/command_container/CMakeLists.txt

@@ -25,6 +25,7 @@ if(SUPPORT_XEHP_AND_LATER)
        ${CMAKE_CURRENT_SOURCE_DIR}/command_encoder_raytracing_xehp_and_later.inl
        ${CMAKE_CURRENT_SOURCE_DIR}/command_encoder_xehp_and_later.inl
        ${CMAKE_CURRENT_SOURCE_DIR}/implicit_scaling_xehp_and_later.inl
+       ${CMAKE_CURRENT_SOURCE_DIR}/walker_partition_interface.h
        ${CMAKE_CURRENT_SOURCE_DIR}/walker_partition_xehp_and_later.h
   )
 endif()

shared/source/command_container/walker_partition_interface.h

@@ -0,0 +1,57 @@
+/*
+ * Copyright (C) 2021 Intel Corporation
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ */
+
+#pragma once
+#include <cstddef>
+#include <cstdint>
+
+namespace WalkerPartition {
+
+struct WalkerPartitionArgs {
+    uint64_t workPartitionAllocationGpuVa = 0;
+    uint32_t partitionCount = 0;
+    uint32_t tileCount = 0;
+    bool emitBatchBufferEnd = false;
+    bool secondaryBatchBuffer = false;
+    bool synchronizeBeforeExecution = false;
+    bool crossTileAtomicSynchronization = false;
+    bool semaphoreProgrammingRequired = false;
+    bool staticPartitioning = false;
+    bool emitSelfCleanup = false;
+    bool useAtomicsForSelfCleanup = false;
+    bool initializeWparidRegister = false;
+    bool emitPipeControlStall = false;
+    bool preferredStaticPartitioning = false;
+};
+
+constexpr uint32_t wparidCCSOffset = 0x221C;
+constexpr uint32_t addressOffsetCCSOffset = 0x23B4;
+constexpr uint32_t predicationMaskCCSOffset = 0x21FC;
+
+constexpr uint32_t generalPurposeRegister0 = 0x2600;
+constexpr uint32_t generalPurposeRegister1 = 0x2608;
+constexpr uint32_t generalPurposeRegister2 = 0x2610;
+constexpr uint32_t generalPurposeRegister3 = 0x2618;
+constexpr uint32_t generalPurposeRegister4 = 0x2620;
+constexpr uint32_t generalPurposeRegister5 = 0x2628;
+constexpr uint32_t generalPurposeRegister6 = 0x2630;
+
+struct BatchBufferControlData {
+    uint32_t partitionCount = 0u;
+    uint32_t tileCount = 0u;
+    uint32_t inTileCount = 0u;
+    uint32_t finalSyncTileCount = 0u;
+};
+static constexpr inline size_t dynamicPartitioningFieldsForCleanupCount = sizeof(BatchBufferControlData) / sizeof(uint32_t) - 1;
+
+struct StaticPartitioningControlSection {
+    uint32_t synchronizeBeforeWalkerCounter = 0;
+    uint32_t synchronizeAfterWalkerCounter = 0;
+    uint32_t finalSyncTileCounter = 0;
+};
+static constexpr inline size_t staticPartitioningFieldsForCleanupCount = sizeof(StaticPartitioningControlSection) / sizeof(uint32_t) - 1;
+} // namespace WalkerPartition

shared/source/command_container/walker_partition_xehp_and_later.h

@@ -8,6 +8,7 @@
 #pragma once
 
 #include "shared/source/command_container/command_encoder.h"
+#include "shared/source/command_container/walker_partition_interface.h"
 #include "shared/source/debug_settings/debug_settings_manager.h"
 #include "shared/source/helpers/basic_math.h"
 #include "shared/source/helpers/hw_helper.h"
@@ -18,23 +19,6 @@
 
 namespace WalkerPartition {
 
-struct WalkerPartitionArgs {
-    uint64_t workPartitionAllocationGpuVa = 0;
-    uint32_t partitionCount = 0;
-    uint32_t tileCount = 0;
-    bool emitBatchBufferEnd = false;
-    bool secondaryBatchBuffer = false;
-    bool synchronizeBeforeExecution = false;
-    bool crossTileAtomicSynchronization = false;
-    bool semaphoreProgrammingRequired = false;
-    bool staticPartitioning = false;
-    bool emitSelfCleanup = false;
-    bool useAtomicsForSelfCleanup = false;
-    bool initializeWparidRegister = false;
-    bool emitPipeControlStall = false;
-    bool preferredStaticPartitioning = false;
-};
-
 template <typename GfxFamily>
 using COMPUTE_WALKER = typename GfxFamily::COMPUTE_WALKER;
 template <typename GfxFamily>
@@ -62,26 +46,6 @@ using PIPE_CONTROL = typename GfxFamily::PIPE_CONTROL;
 template <typename GfxFamily>
 using MI_STORE_DATA_IMM = typename GfxFamily::MI_STORE_DATA_IMM;
 
-constexpr uint32_t wparidCCSOffset = 0x221C;
-constexpr uint32_t addressOffsetCCSOffset = 0x23B4;
-constexpr uint32_t predicationMaskCCSOffset = 0x21FC;
-
-constexpr uint32_t generalPurposeRegister0 = 0x2600;
-constexpr uint32_t generalPurposeRegister1 = 0x2608;
-constexpr uint32_t generalPurposeRegister2 = 0x2610;
-constexpr uint32_t generalPurposeRegister3 = 0x2618;
-constexpr uint32_t generalPurposeRegister4 = 0x2620;
-constexpr uint32_t generalPurposeRegister5 = 0x2628;
-constexpr uint32_t generalPurposeRegister6 = 0x2630;
-
-struct BatchBufferControlData {
-    uint32_t partitionCount = 0u;
-    uint32_t tileCount = 0u;
-    uint32_t inTileCount = 0u;
-    uint32_t finalSyncTileCount = 0u;
-};
-static constexpr inline size_t dynamicPartitioningFieldsForCleanupCount = sizeof(BatchBufferControlData) / sizeof(uint32_t) - 1;
-
 template <typename Command>
 Command *putCommand(void *&inputAddress, uint32_t &totalBytesProgrammed) {
     totalBytesProgrammed += sizeof(Command);
@@ -631,13 +595,6 @@ void constructDynamicallyPartitionedCommandBuffer(void *cpuPointer,
     }
 }
 
-struct StaticPartitioningControlSection {
-    uint32_t synchronizeBeforeWalkerCounter = 0;
-    uint32_t synchronizeAfterWalkerCounter = 0;
-    uint32_t finalSyncTileCounter = 0;
-};
-static constexpr inline size_t staticPartitioningFieldsForCleanupCount = sizeof(StaticPartitioningControlSection) / sizeof(uint32_t) - 1;
-
 template <typename GfxFamily>
 uint64_t computeStaticPartitioningControlSectionOffset(WalkerPartitionArgs &args) {
     const auto beforeExecutionSyncAtomicSize = args.synchronizeBeforeExecution

shared/source/enable_gens.cmake

@@ -20,6 +20,7 @@ set(CORE_RUNTIME_SRCS_GENX_CPP_BASE
     create_device_command_stream_receiver
     direct_submission
     experimental_command_buffer
+    implicit_scaling
     hw_helper
     hw_info
     preamble

shared/source/xe_hp_core/command_encoder_xe_hp_core.cpp

@@ -10,8 +10,6 @@
 #include "shared/source/command_container/command_encoder_raytracing_xehp_and_later.inl"
 #include "shared/source/command_container/command_encoder_xehp_and_later.inl"
 #include "shared/source/command_container/encode_compute_mode_tgllp_and_later.inl"
-#include "shared/source/command_container/implicit_scaling.h"
-#include "shared/source/command_container/implicit_scaling_xehp_and_later.inl"
 #include "shared/source/helpers/hw_helper.h"
 #include "shared/source/xe_hp_core/hw_cmds_base.h"
 
@@ -72,6 +70,5 @@ template struct EncodeMiFlushDW<Family>;
 template struct EncodeMemoryPrefetch<Family>;
 template struct EncodeMiArbCheck<Family>;
 template struct EncodeWA<Family>;
-template struct ImplicitScalingDispatch<Family>;
 template struct EncodeEnableRayTracing<Family>;
 } // namespace NEO

shared/source/xe_hp_core/implicit_scaling_xe_hp_core.cpp

@@ -0,0 +1,16 @@
+/*
+ * Copyright (C) 2021 Intel Corporation
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ */
+
+#include "shared/source/command_container/implicit_scaling.h"
+#include "shared/source/command_container/implicit_scaling_xehp_and_later.inl"
+
+namespace NEO {
+
+using Family = XeHpFamily;
+
+template struct ImplicitScalingDispatch<Family>;
+} // namespace NEO

shared/test/common/libult/CMakeLists.txt

@@ -36,7 +36,7 @@ macro(macro_for_each_gen)
       if(EXISTS ${SRC_FILE})
         list(APPEND neo_libult_common_SRCS_ENABLE_TESTED_HW ${SRC_FILE})
       endif()
-      foreach(SRC_IT "enable_" "enable_family_full_core_")
+      foreach(SRC_IT "enable_" "enable_family_full_core_" "implicit_scaling_")
         set(SRC_FILE ${NEO_SHARED_DIRECTORY}${BRANCH}${GEN_TYPE_LOWER}${BRANCH_DIR}${SRC_IT}${GEN_TYPE_LOWER}.cpp)
         if(EXISTS ${SRC_FILE})
           list(APPEND neo_libult_common_SRCS_ENABLE_TESTED_HW ${SRC_FILE})