refactor: reorganize command encode classes 1/n

- remove obsolete file
- move methods from redundant compute mode file into dedicated platform files
- group same implementation into platform specific inl files

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>

shared/source/command_container/CMakeLists.txt

@@ -14,8 +14,6 @@ set(NEO_CORE_COMMAND_CONTAINER
     ${CMAKE_CURRENT_SOURCE_DIR}/command_encoder_enablers.inl
     ${CMAKE_CURRENT_SOURCE_DIR}/command_encoder_tgllp_and_later.inl
     ${CMAKE_CURRENT_SOURCE_DIR}/encode_alu_helper.h
-    ${CMAKE_CURRENT_SOURCE_DIR}/encode_compute_mode_bdw_and_later.inl
-    ${CMAKE_CURRENT_SOURCE_DIR}/encode_compute_mode_tgllp_and_later.inl
     ${CMAKE_CURRENT_SOURCE_DIR}/encode_surface_state.h
     ${CMAKE_CURRENT_SOURCE_DIR}/implicit_scaling.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/implicit_scaling.h
@@ -61,6 +59,12 @@ if(SUPPORT_XE_HPC_AND_BEFORE)
   )
 endif()
 
+if(SUPPORT_XE_HPC_AND_BEFORE OR SUPPORT_XE2_HPG_CORE)
+  list(APPEND NEO_CORE_COMMAND_CONTAINER
+       ${CMAKE_CURRENT_SOURCE_DIR}/command_encoder_from_gen12lp_to_xe2_hpg.inl
+  )
+endif()
+
 set_property(GLOBAL PROPERTY NEO_CORE_COMMAND_CONTAINER ${NEO_CORE_COMMAND_CONTAINER})
 
 add_subdirectories()

shared/source/command_container/command_encoder_from_gen12lp_to_xe2_hpg.inl

@@ -0,0 +1,41 @@
+/*
+ * Copyright (C) 2020-2024 Intel Corporation
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ */
+
+#include "shared/source/command_container/command_encoder.h"
+#include "shared/source/command_stream/linear_stream.h"
+#include "shared/source/helpers/pipe_control_args.h"
+
+namespace NEO {
+
+template <typename Family>
+inline void EncodeComputeMode<Family>::programComputeModeCommandWithSynchronization(
+    LinearStream &csr, StateComputeModeProperties &properties, const PipelineSelectArgs &args,
+    bool hasSharedHandles, const RootDeviceEnvironment &rootDeviceEnvironment, bool isRcs, bool dcFlush) {
+    auto &hwInfo = *rootDeviceEnvironment.getHardwareInfo();
+    NEO::EncodeWA<Family>::encodeAdditionalPipelineSelect(csr, args, true, rootDeviceEnvironment, isRcs);
+    auto *releaseHelper = rootDeviceEnvironment.getReleaseHelper();
+    auto &productHelper = rootDeviceEnvironment.getHelper<ProductHelper>();
+    const auto &[isBasicWARequired, isExtendedWARequired] = productHelper.isPipeControlPriorToNonPipelinedStateCommandsWARequired(hwInfo, isRcs, releaseHelper);
+    std::ignore = isExtendedWARequired;
+
+    if (isBasicWARequired) {
+        PipeControlArgs args;
+        NEO::EncodeWA<Family>::addPipeControlPriorToNonPipelinedStateCommand(csr, args, rootDeviceEnvironment, isRcs);
+    }
+
+    EncodeComputeMode<Family>::programComputeModeCommand(csr, properties, rootDeviceEnvironment);
+
+    if (hasSharedHandles) {
+        PipeControlArgs args;
+        args.csStallOnly = true;
+        MemorySynchronizationCommands<Family>::addSingleBarrier(csr, args);
+    }
+
+    NEO::EncodeWA<Family>::encodeAdditionalPipelineSelect(csr, args, false, rootDeviceEnvironment, isRcs);
+}
+
+} // namespace NEO

shared/source/command_container/command_encoder_pre_xe2_hpg_core.inl

@@ -21,4 +21,10 @@ uint32_t EncodeDispatchKernel<Family>::alignPreferredSlmSize(uint32_t slmSize) {
     return EncodeDispatchKernel<Family>::alignSlmSize(slmSize);
 }
 
+template <typename Family>
+void EncodeSurfaceState<Family>::disableCompressionFlags(R_SURFACE_STATE *surfaceState) {
+    surfaceState->setAuxiliarySurfaceMode(Family::RENDER_SURFACE_STATE::AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_NONE);
+    surfaceState->setMemoryCompressionEnable(false);
+}
+
 } // namespace NEO

shared/source/command_container/command_encoder_tgllp_and_later.inl

@@ -13,7 +13,58 @@
 
 namespace NEO {
 
-template <>
+template <typename Family>
+size_t EncodeComputeMode<Family>::getCmdSizeForComputeMode(const RootDeviceEnvironment &rootDeviceEnvironment, bool hasSharedHandles, bool isRcs) {
+    size_t size = 0;
+    auto &productHelper = rootDeviceEnvironment.getHelper<ProductHelper>();
+    auto *releaseHelper = rootDeviceEnvironment.getReleaseHelper();
+    auto &hwInfo = *rootDeviceEnvironment.getHardwareInfo();
+    const auto &[isBasicWARequired, isExtendedWARequired] = productHelper.isPipeControlPriorToNonPipelinedStateCommandsWARequired(hwInfo, isRcs, releaseHelper);
+    std::ignore = isExtendedWARequired;
+
+    if (isBasicWARequired) {
+        size += MemorySynchronizationCommands<Family>::getSizeForSingleBarrier(false);
+    }
+    size += sizeof(typename Family::STATE_COMPUTE_MODE);
+    if (hasSharedHandles) {
+        size += MemorySynchronizationCommands<Family>::getSizeForSingleBarrier(false);
+    }
+    if (productHelper.is3DPipelineSelectWARequired() && isRcs) {
+        size += (2 * PreambleHelper<Family>::getCmdSizeForPipelineSelect(rootDeviceEnvironment));
+    }
+    return size;
+}
+
+template <typename GfxFamily>
+inline size_t EncodeComputeMode<GfxFamily>::getSizeForComputeMode() {
+    return sizeof(typename GfxFamily::STATE_COMPUTE_MODE);
+}
+
+template <typename Family>
+void EncodeSetMMIO<Family>::remapOffset(MI_LOAD_REGISTER_MEM *pMiLoadReg) {
+    if (isRemapApplicable(pMiLoadReg->getRegisterAddress())) {
+        pMiLoadReg->setMmioRemapEnable(true);
+    }
+}
+
+template <typename Family>
+void EncodeSetMMIO<Family>::remapOffset(MI_LOAD_REGISTER_REG *pMiLoadReg) {
+    if (isRemapApplicable(pMiLoadReg->getSourceRegisterAddress())) {
+        pMiLoadReg->setMmioRemapEnableSource(true);
+    }
+    if (isRemapApplicable(pMiLoadReg->getDestinationRegisterAddress())) {
+        pMiLoadReg->setMmioRemapEnableDestination(true);
+    }
+}
+
+template <typename Family>
+inline bool EncodeSetMMIO<Family>::isRemapApplicable(uint32_t offset) {
+    return (0x2000 <= offset && offset <= 0x27ff) ||
+           (0x4200 <= offset && offset <= 0x420f) ||
+           (0x4400 <= offset && offset <= 0x441f);
+}
+
+template <typename Family>
 void EncodeWA<Family>::addPipeControlBeforeStateBaseAddress(LinearStream &commandStream,
                                                             const RootDeviceEnvironment &rootDeviceEnvironment, bool isRcs, bool dcFlushRequired) {
     PipeControlArgs args;
@@ -24,7 +75,7 @@ void EncodeWA<Family>::addPipeControlBeforeStateBaseAddress(LinearStream &comman
     NEO::EncodeWA<Family>::addPipeControlPriorToNonPipelinedStateCommand(commandStream, args, rootDeviceEnvironment, isRcs);
 }
 
-template <>
+template <typename Family>
 inline void EncodeMiArbCheck<Family>::adjust(MI_ARB_CHECK &miArbCheck, std::optional<bool> preParserDisable) {
     if (debugManager.flags.ForcePreParserEnabledForMiArbCheck.get() != -1) {
         preParserDisable = !debugManager.flags.ForcePreParserEnabledForMiArbCheck.get();
@@ -34,7 +85,7 @@ inline void EncodeMiArbCheck<Family>::adjust(MI_ARB_CHECK &miArbCheck, std::opti
     }
 }
 
-template <>
+template <typename Family>
 inline void EncodeStoreMemory<Family>::encodeForceCompletionCheck(MI_STORE_DATA_IMM &storeDataImmCmd) {
     storeDataImmCmd.setForceWriteCompletionCheck(true);
 }

shared/source/command_container/command_encoder_xe2_hpg_core_and_later.inl

@@ -36,4 +36,8 @@ void EncodeDispatchKernel<Family>::encodeComputeDispatchAllWalker(WalkerType &wa
     walkerCmd.setComputeDispatchAllWalkerEnable(computeDispatchAllWalkerEnable);
 }
 
+template <typename Family>
+void EncodeSurfaceState<Family>::disableCompressionFlags(R_SURFACE_STATE *surfaceState) {
+}
+
 } // namespace NEO

shared/source/command_container/encode_compute_mode_bdw_and_later.inl

@@ -1,42 +0,0 @@
-/*
- * Copyright (C) 2020-2023 Intel Corporation
- *
- * SPDX-License-Identifier: MIT
- *
- */
-
-#pragma once
-#include "shared/source/command_container/command_encoder.h"
-#include "shared/source/command_stream/linear_stream.h"
-
-namespace NEO {
-
-template <typename Family>
-inline void EncodeComputeMode<Family>::programComputeModeCommandWithSynchronization(
-    LinearStream &csr, StateComputeModeProperties &properties, const PipelineSelectArgs &args,
-    bool hasSharedHandles, const RootDeviceEnvironment &rootDeviceEnvironment, bool isRcs, bool dcFlush) {
-
-    EncodeComputeMode<Family>::programComputeModeCommand(csr, properties, rootDeviceEnvironment);
-}
-
-template <typename Family>
-inline void EncodeStoreMMIO<Family>::appendFlags(MI_STORE_REGISTER_MEM *storeRegMem, bool workloadPartition) {
-}
-
-template <typename Family>
-inline void EncodeSetMMIO<Family>::remapOffset(MI_LOAD_REGISTER_MEM *pMiLoadReg) {
-}
-
-template <typename Family>
-inline void EncodeSetMMIO<Family>::remapOffset(MI_LOAD_REGISTER_REG *pMiLoadReg) {
-}
-
-template <typename Family>
-inline bool EncodeSetMMIO<Family>::isRemapApplicable(uint32_t offset) {
-    return false;
-}
-
-template <typename Family>
-void EncodeSurfaceState<Family>::disableCompressionFlags(R_SURFACE_STATE *surfaceState) {
-}
-} // namespace NEO

shared/source/command_container/encode_compute_mode_tgllp_and_later.inl

@@ -1,97 +0,0 @@
-/*
- * Copyright (C) 2020-2024 Intel Corporation
- *
- * SPDX-License-Identifier: MIT
- *
- */
-
-#pragma once
-#include "shared/source/command_container/command_encoder.h"
-#include "shared/source/command_stream/linear_stream.h"
-
-namespace NEO {
-
-template <typename Family>
-size_t EncodeComputeMode<Family>::getCmdSizeForComputeMode(const RootDeviceEnvironment &rootDeviceEnvironment, bool hasSharedHandles, bool isRcs) {
-    size_t size = 0;
-    auto &productHelper = rootDeviceEnvironment.getHelper<ProductHelper>();
-    auto *releaseHelper = rootDeviceEnvironment.getReleaseHelper();
-    auto &hwInfo = *rootDeviceEnvironment.getHardwareInfo();
-    const auto &[isBasicWARequired, isExtendedWARequired] = productHelper.isPipeControlPriorToNonPipelinedStateCommandsWARequired(hwInfo, isRcs, releaseHelper);
-    std::ignore = isExtendedWARequired;
-
-    if (isBasicWARequired) {
-        size += MemorySynchronizationCommands<Family>::getSizeForSingleBarrier(false);
-    }
-    size += sizeof(typename Family::STATE_COMPUTE_MODE);
-    if (hasSharedHandles) {
-        size += MemorySynchronizationCommands<Family>::getSizeForSingleBarrier(false);
-    }
-    if (productHelper.is3DPipelineSelectWARequired() && isRcs) {
-        size += (2 * PreambleHelper<Family>::getCmdSizeForPipelineSelect(rootDeviceEnvironment));
-    }
-    return size;
-}
-
-template <typename GfxFamily>
-inline size_t EncodeComputeMode<GfxFamily>::getSizeForComputeMode() {
-    return sizeof(typename GfxFamily::STATE_COMPUTE_MODE);
-}
-
-template <typename Family>
-inline void EncodeComputeMode<Family>::programComputeModeCommandWithSynchronization(
-    LinearStream &csr, StateComputeModeProperties &properties, const PipelineSelectArgs &args,
-    bool hasSharedHandles, const RootDeviceEnvironment &rootDeviceEnvironment, bool isRcs, bool dcFlush) {
-    auto &hwInfo = *rootDeviceEnvironment.getHardwareInfo();
-    NEO::EncodeWA<Family>::encodeAdditionalPipelineSelect(csr, args, true, rootDeviceEnvironment, isRcs);
-    auto *releaseHelper = rootDeviceEnvironment.getReleaseHelper();
-    auto &productHelper = rootDeviceEnvironment.getHelper<ProductHelper>();
-    const auto &[isBasicWARequired, isExtendedWARequired] = productHelper.isPipeControlPriorToNonPipelinedStateCommandsWARequired(hwInfo, isRcs, releaseHelper);
-    std::ignore = isExtendedWARequired;
-
-    if (isBasicWARequired) {
-        PipeControlArgs args;
-        NEO::EncodeWA<Family>::addPipeControlPriorToNonPipelinedStateCommand(csr, args, rootDeviceEnvironment, isRcs);
-    }
-
-    EncodeComputeMode<Family>::programComputeModeCommand(csr, properties, rootDeviceEnvironment);
-
-    if (hasSharedHandles) {
-        PipeControlArgs args;
-        args.csStallOnly = true;
-        MemorySynchronizationCommands<Family>::addSingleBarrier(csr, args);
-    }
-
-    NEO::EncodeWA<Family>::encodeAdditionalPipelineSelect(csr, args, false, rootDeviceEnvironment, isRcs);
-}
-
-template <typename Family>
-void EncodeSetMMIO<Family>::remapOffset(MI_LOAD_REGISTER_MEM *pMiLoadReg) {
-    if (isRemapApplicable(pMiLoadReg->getRegisterAddress())) {
-        pMiLoadReg->setMmioRemapEnable(true);
-    }
-}
-
-template <typename Family>
-void EncodeSetMMIO<Family>::remapOffset(MI_LOAD_REGISTER_REG *pMiLoadReg) {
-    if (isRemapApplicable(pMiLoadReg->getSourceRegisterAddress())) {
-        pMiLoadReg->setMmioRemapEnableSource(true);
-    }
-    if (isRemapApplicable(pMiLoadReg->getDestinationRegisterAddress())) {
-        pMiLoadReg->setMmioRemapEnableDestination(true);
-    }
-}
-
-template <typename Family>
-inline bool EncodeSetMMIO<Family>::isRemapApplicable(uint32_t offset) {
-    return (0x2000 <= offset && offset <= 0x27ff) ||
-           (0x4200 <= offset && offset <= 0x420f) ||
-           (0x4400 <= offset && offset <= 0x441f);
-}
-
-template <typename Family>
-void EncodeSurfaceState<Family>::disableCompressionFlags(R_SURFACE_STATE *surfaceState) {
-    surfaceState->setAuxiliarySurfaceMode(Family::RENDER_SURFACE_STATE::AUXILIARY_SURFACE_MODE::AUXILIARY_SURFACE_MODE_AUX_NONE);
-    surfaceState->setMemoryCompressionEnable(false);
-}
-} // namespace NEO

shared/source/gen12lp/command_encoder_gen12lp.cpp

@@ -8,7 +8,9 @@
 #include "shared/source/command_container/command_encoder.h"
 #include "shared/source/command_container/command_encoder.inl"
 #include "shared/source/command_container/command_encoder_bdw_and_later.inl"
+#include "shared/source/command_container/command_encoder_from_gen12lp_to_xe2_hpg.inl"
 #include "shared/source/command_container/command_encoder_pre_xe2_hpg_core.inl"
+#include "shared/source/command_container/command_encoder_tgllp_and_later.inl"
 #include "shared/source/command_stream/stream_properties.h"
 #include "shared/source/gen12lp/hw_cmds_base.h"
 #include "shared/source/gen12lp/reg_configs.h"
@@ -21,8 +23,6 @@
 using Family = NEO::Gen12LpFamily;
 
 #include "shared/source/command_container/command_encoder_heap_addressing.inl"
-#include "shared/source/command_container/command_encoder_tgllp_and_later.inl"
-#include "shared/source/command_container/encode_compute_mode_tgllp_and_later.inl"
 #include "shared/source/command_container/image_surface_state/compression_params_bdw_and_later.inl"
 #include "shared/source/command_container/image_surface_state/compression_params_tgllp_and_later.inl"
 #include "shared/source/command_stream/command_stream_receiver.h"

shared/source/xe2_hpg_core/command_encoder_xe2_hpg_core.cpp

@@ -7,10 +7,11 @@
 
 #include "shared/source/command_container/command_encoder.h"
 #include "shared/source/command_container/command_encoder.inl"
+#include "shared/source/command_container/command_encoder_from_gen12lp_to_xe2_hpg.inl"
 #include "shared/source/command_container/command_encoder_heap_addressing.inl"
+#include "shared/source/command_container/command_encoder_tgllp_and_later.inl"
 #include "shared/source/command_container/command_encoder_xe2_hpg_core_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_stream/stream_properties.h"
 #include "shared/source/helpers/constants.h"
 #include "shared/source/kernel/grf_config.h"
@@ -20,7 +21,6 @@
 
 using Family = NEO::Xe2HpgCoreFamily;
 
-#include "shared/source/command_container/command_encoder_tgllp_and_later.inl"
 #include "shared/source/command_container/command_encoder_xe_hpc_core_and_later.inl"
 #include "shared/source/command_container/command_encoder_xe_hpg_core_and_later.inl"
 #include "shared/source/command_container/image_surface_state/compression_params_xehp_and_later.inl"
@@ -225,10 +225,6 @@ template <>
 inline void EncodeSurfaceState<Family>::setFlagsForMediaCompression(R_SURFACE_STATE *surfaceState, Gmm *gmm) {
 }
 
-template <>
-void EncodeSurfaceState<Family>::disableCompressionFlags(R_SURFACE_STATE *surfaceState) {
-}
-
 template <>
 template <typename InterfaceDescriptorType>
 void EncodeDispatchKernel<Family>::setGrfInfo(InterfaceDescriptorType *pInterfaceDescriptor, uint32_t grfCount,

shared/source/xe_hpc_core/command_encoder_xe_hpc_core.cpp

@@ -7,9 +7,10 @@
 
 #include "shared/source/command_container/command_encoder.h"
 #include "shared/source/command_container/command_encoder.inl"
+#include "shared/source/command_container/command_encoder_from_gen12lp_to_xe2_hpg.inl"
 #include "shared/source/command_container/command_encoder_pre_xe2_hpg_core.inl"
+#include "shared/source/command_container/command_encoder_tgllp_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_stream/stream_properties.h"
 #include "shared/source/helpers/constants.h"
 #include "shared/source/kernel/grf_config.h"
@@ -20,7 +21,6 @@
 using Family = NEO::XeHpcCoreFamily;
 
 #include "shared/source/command_container/command_encoder_heap_addressing.inl"
-#include "shared/source/command_container/command_encoder_tgllp_and_later.inl"
 #include "shared/source/command_container/command_encoder_xe_hpc_core_and_later.inl"
 #include "shared/source/command_container/command_encoder_xe_hpg_core_and_later.inl"
 #include "shared/source/command_container/image_surface_state/compression_params_tgllp_and_later.inl"

shared/source/xe_hpg_core/command_encoder_xe_hpg_core.cpp

@@ -7,9 +7,10 @@
 
 #include "shared/source/command_container/command_encoder.h"
 #include "shared/source/command_container/command_encoder.inl"
+#include "shared/source/command_container/command_encoder_from_gen12lp_to_xe2_hpg.inl"
 #include "shared/source/command_container/command_encoder_pre_xe2_hpg_core.inl"
+#include "shared/source/command_container/command_encoder_tgllp_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_stream/stream_properties.h"
 #include "shared/source/helpers/cache_flush_xehp_and_later.inl"
 #include "shared/source/os_interface/product_helper.h"
@@ -20,7 +21,6 @@
 using Family = NEO::XeHpgCoreFamily;
 
 #include "shared/source/command_container/command_encoder_heap_addressing.inl"
-#include "shared/source/command_container/command_encoder_tgllp_and_later.inl"
 #include "shared/source/command_container/command_encoder_xe_hpg_core_and_later.inl"
 #include "shared/source/command_container/image_surface_state/compression_params_tgllp_and_later.inl"
 #include "shared/source/command_container/image_surface_state/compression_params_xehp_and_later.inl"